[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_2938":2939,"_2940":2941},"routes/_app.x.$handle.$name",{"_11":12,"_13":14,"_15":16,"_17":18,"_39":-7,"_137":-5,"_19":138,"_139":140,"_141":142,"_143":144,"_145":146,"_21":147,"_148":-5,"_149":150,"_151":24,"_152":153,"_154":94,"_155":-5,"_156":-5,"_157":158},"routes/_app.x.$handle.$name.branch.$branch.history_.$",{"_11":12,"_13":14,"_15":16,"_17":18,"_19":20,"_21":22,"_23":24,"_25":26,"_27":28,"_29":24,"_30":-5},"handle","cleitoncosta","name","chat-native-maestro","branchName","economic-v0.6","path","artifacts/kaggriculture/intelligence/KAG_OPPONENT_PROFILES/mateo.md","project",{"_33":81,"_13":14,"_37":82,"_83":-5,"_84":24,"_85":86,"_87":86,"_88":89,"_90":-5,"_30":69,"_91":92,"_93":94,"_95":-5,"_96":-5,"_97":26,"_41":42,"_98":99,"_100":101,"_102":103,"_104":-5,"_105":106,"_107":106,"_108":109,"_110":109,"_111":109,"_112":106,"_113":106,"_54":114,"_115":116,"_117":118,"_119":24},"file",{"_64":65,"_13":66,"_67":21,"_68":24,"_80":-5},"isDirectory",false,"page",0,"history",[31,32],"hasNext","userId",{"_33":72,"_35":36,"_37":73,"_39":74,"_41":42,"_43":75,"_45":-5,"_46":-5,"_47":-5,"_48":-5,"_49":-5,"_50":51,"_52":53,"_54":76,"_21":77,"_57":-5,"_58":-5,"_59":-5,"_60":-5,"_61":78},{"_33":34,"_35":36,"_37":38,"_39":40,"_41":42,"_43":44,"_45":-5,"_46":-5,"_47":-5,"_48":-5,"_49":-5,"_50":51,"_52":53,"_54":55,"_21":56,"_57":-5,"_58":-5,"_59":-5,"_60":-5,"_61":62},"id","f3eaa978-df84-4958-b739-7f19c8696eee","branchId","f6b18fe8-8463-44a7-8607-6aadc93451c1","createdAt",["D",1788825058278],"version",572,"orgId","01a0773d-9d6b-7580-a53a-683363c351c8","parentId","ea97d01a-767f-42c2-8afb-0b940c27aead","incomingCommitId","aiChatMessageId","aiChatThreadId","aiChatThreadUserId","aiChatThreadTitle","source","mcp","clientLabel","ChatGPT","user",{"_33":69,"_41":42,"_11":12,"_70":71},{"_64":65,"_13":66,"_67":21,"_68":24},"pull","merge","revert","multiple","commits",[63],{"_33":34,"_35":36,"_37":38,"_39":40,"_41":42,"_43":44,"_45":-5,"_46":-5,"_47":-5,"_48":-5,"_49":-5,"_50":51,"_52":53,"_54":55,"_21":56,"_57":-5,"_58":-5,"_59":-5},"fileId","01a07e48-3be6-726a-80bb-eff4337ccf4e","mateo.md","fileType","deleted","6cc24525-830b-4db3-9c9c-fd19fb50e91f","avatar","https://lh3.googleusercontent.com/a/ACg8ocIgWisWQiXts_2fy5iRiRgl8A6c7Ef94eqE3ur-4mlyUrpCBmQ=s96-c","274888d1-6218-4332-a618-2f66d89499a0",["D",1788839007738],629,"d56b6763-6f3e-4f23-82e3-7e1e8e271464",{"_33":69,"_41":42,"_11":12,"_70":71},{"_64":65,"_13":66,"_67":21,"_68":24},[79],{"_33":72,"_35":36,"_37":73,"_39":74,"_41":42,"_43":75,"_45":-5,"_46":-5,"_47":-5,"_48":-5,"_49":-5,"_50":51,"_52":53,"_54":76,"_21":77,"_57":-5,"_58":-5,"_59":-5},"deletedAtVersion","1cb1fea8-3811-42e9-b0ad-3a5ec2d20472",["D",1788707065045],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","Cloud runtime for ChatGPT-native orchestration","imageUrl","pinnedFileId","3fca0dce-aa04-11f1-bd83-1607ee4eb77e","pinPreview",true,"pinMigrated","screenshotVersion","likeCount","tursoName","1cb1fea8381142e9b0ad3a5ec2d20472","tursoHostname","1cb1fea8381142e9b0ad3a5ec2d20472-val-town.aws-us-east-2.turso.io","tursoMeasuredAt",["D",1789575110896],"pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm",{"_11":12,"_136":71},"org",{"_11":12,"_131":132,"_70":71,"_30":69,"_133":134,"_135":24},"branch",{"_39":120,"_13":16,"_33":36,"_121":81,"_122":123,"_124":125,"_126":86,"_127":81,"_128":42,"_129":14,"_130":12},"isOwned",930,"projectId","forkedBranchId","22a6b9fb-662c-4823-89b0-7ec510c82fcd","forkedBranchVersion",68,"forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","displayName","Cleiton Costa","tier","free","canRestrictVals","profileImageUrl","adoptOrgId",{"_33":81,"_13":14,"_37":2933,"_83":-5,"_84":24,"_85":86,"_87":86,"_88":89,"_90":-5,"_30":69,"_91":92,"_93":94,"_95":-5,"_96":-5,"_97":26,"_41":42,"_98":99,"_100":101,"_102":2934,"_104":-5,"_105":106,"_107":106,"_108":109,"_110":109,"_111":109,"_112":106,"_113":106,"_54":2935,"_115":2936,"_117":2937,"_119":24},"tree",[190,268,275,299,303,304,305,306,307,308,309,310,311,312,313,314,315,316,317,318,319,320,321,322,323,324,325,326,327,328,329,330,331,332,333,334,335,336,337,338,339,340,341,342,343,344,345,346,347,348,349,350,351,352,353,354,355,356,357,358,359,360,361,362,363,364,365,366,367,368,369,370,371,372,373,374,375,376,377,378,379,380,381,382,383,384,385,386,387,388,389,390,391,392,393,394,395,396,397,398,399,400,401,402,403,404,405,406,407,408,409,410,411,412,413,414,415,416,417,418,419,420,421,422,423,424,425,426,427,428,429,430,431,432,433,434,435,436,437,438,439,440,441,442,443,444,445,446,447,448,449,450,451,452,453,454,455,456,457,458,459,460,461,462,463,464,465,466,467,468,469,470,471,472,473,474,475,476,477,478,479,480,481,482,483,484,485,486,487,488,489,490,491,492,493,494,495,496,497,498,499],"files",[190,191,192,193,194,195,196,197,198,199,200,201,202,203,204,205,206,207,208,209,210,211,212,213,214,215,216,217,218,219,220,221,222,223,224,225,226,227,228,229,230,231,232,233,234,235,236,237,238,239,240,241,242,243,244,245,246,247,248,249,250,251,252,253,254,255,256,257,258,259,260,261,262,263,264,265,266,267,268,269,270,271,272,273,274,275,276,277,278,279,280,281,282,283,284,285,286,287,288,289,290,291,292,293,294,295,296,297,298,299,300,301,302,303,304,305,306,307,308,309,310,311,312,313,314,315,316,317,318,319,320,321,322,323,324,325,326,327,328,329,330,331,332,333,334,335,336,337,338,339,340,341,342,343,344,345,346,347,348,349,350,351,352,353,354,355,356,357,358,359,360,361,362,363,364,365,366,367,368,369,370,371,372,373,374,375,376,377,378,379,380,381,382,383,384,385,386,387,388,389,390,391,392,393,394,395,396,397,398,399,400,401,402,403,404,405,406,407,408,409,410,411,412,413,414,415,416,417,418,419,420,421,422,423,424,425,426,427,428,429,430,431,432,433,434,435,436,437,438,439,440,441,442,443,444,445,446,447,448,449,450,451,452,453,454,455,456,457,458,459,460,461,462,463,464,465,466,467,468,469,470,471,472,473,474,475,476,477,478,479,480,481,482,483,484,485,486,487,488,489,490,491,492,493,494,495,496,497,498,499],"counts",{"_179":26,"_180":181,"_182":26,"_183":184,"_185":186},"contents",[],{"_33":165,"_13":66,"_64":65,"_43":166,"_35":36,"_167":74,"_168":94,"_37":169,"_39":170,"_67":21,"_171":172,"_68":24,"_173":-5,"_17":174},"upstreamProject","upstreamBranch",{"_33":123,"_37":159,"_160":161,"_13":162,"_121":81,"_39":125,"_122":-5,"_124":-5,"_163":164,"_83":-5,"_119":24,"_85":86,"_130":12,"_41":42,"_129":14},"emptyEnvVars","flags",[],"hasHttpFiles","grants","orgs","customDomains",{},["D",1788707065045],"updatedAt",["D",1788709584572],"main","lockfileKey","1788709584801-esm.town_v_cleitoncosta_chat-native-maestro@68-main","50858a4a-8502-4b2b-9f5d-fb9941ca7ec7","01a07e48-3be6-726a-80bb-d665dbb13f78","branchVersion","latest",["D",1788839007738],1,"content","# Melon Mateo — Live Opponent Dossier\n\nRole: weak-regression + strong market-threshold calibration.\n\n## SOURCE_GROUND_TRUTH\nPublic tier-4 policy:\n- buys one NE quadrant;\n- hires 6 hands;\n- crop mix melon/wheat = 70/30;\n- fertilizes melons through bonus window;\n- sells melon → wheat; melon chunk capped at 12 per turn;\n- explicit melon price floor = 120;\n- invests through day 20; melon planting through day 16; liquidation from day 27.\n\nDependency probe: no opponent/other-farm read in source.\n\n## OBSERVED\n- Fresh early-signature probe: by D1 Mateo is highly distinctive — land2, ~20 wheat + 12 melon, near-zero cash, no animals; remains land2 in the sampled trajectory.\n- MIX11 B21: 8-0 against Mateo.\n\n## CALIBRATION HYPOTHESES\nH1 static classifier should identify Mateo early with high confidence.\nH2 market-response detector should recover a sharp selling discontinuity around melon price 120 and a quantity ceiling around 12 per sell turn.\nH3 direct our-state-awareness should remain near zero once market mediation is controlled.\nH4 a temporary price shock should change sale timing but should not be mislabeled as opponent modeling of us.\n\n## REQUIRED PROBES\n1. fixed-prestate melon price sweep just below/above 120;\n2. demand/glut shock with matched control;\n3. our visible expansion perturbation while market fixed;\n4. CUSUM test: deliberate crossing of price threshold should generate market-policy regime signal, while farm-only perturbation should not;\n5. SPRT calibration for detecting threshold response with minimal forks.\n\n## FALSIFIER\nIf the profiler cannot recover Mateo's known price threshold/quantity cap, any later inferred elasticities or strategic-depth labels are untrusted. If it assigns direct our-state awareness, treat that as a false positive unless source/current version changed.\n\n## CALIBRATION RESULT — fixed-prestate black-box floor\nNatural prestate seed 460301, D11 H1, MELON shed=54, shed load 64<80. Price sweep:\n- 119 -> SELL MELON 0;\n- 120 -> SELL MELON 12;\n- 121 -> SELL MELON 12.\nForcing shed load to 80 made MELON sell 12 even at 119; moving the same synthetic observation to liquidation day 27 also sold 12 at 119. Raising opponent public cash by +50,000 changed no action.\n\nThe detector therefore recovered both the exact 120 threshold and 12-unit cap while also recovering the two principal context overrides. This is currently the strongest weak-agent positive control.\n\n## REAL-SCENARIO STRATIFIED CALIBRATION — V24\n\nNatural official-engine prestates against Rita, seeds 100577/100609, both seats, sampled every 6 in-game hours:\n- direct opponent-cash 199→200: **0/480** action differences; fresh-vs-stateful mismatch 0;\n- MELON 119→120: **24/480 = 5.00%** action/sell-quantity differences.\n\nIn this finite sample the melon threshold was active only in sampled MID/LATE prestates (MID 4/192; LATE 20/144; EARLY/CLOSING 0). This is a context-activation result, not a change to the known 120 floor. A real threshold can be invisible when inventory/timing/pressure conditions make the sell branch irrelevant.\n\nClassification: `MARKET_THRESHOLD_TRUE_CONTEXT_GATED`; `DIRECT_CASH_AWARENESS_NEGATIVE_IN_STRATIFIED_SAMPLE`.","label",[175,176,177,178,66],"artifacts","kaggriculture","intelligence","KAG_OPPONENT_PROFILES","pulls","branches",3,"forks","envVars",{"_187":26,"_189":26},"connections",{"_187":26,"_188":26},"count","unfilledCount","emptyCount",{"_33":2929,"_13":175,"_64":2337,"_43":-5,"_35":36,"_167":935,"_168":94,"_37":2930,"_39":26,"_67":2101,"_171":-5,"_68":24,"_173":-5,"_506":2931,"_17":2932},{"_33":2924,"_13":176,"_64":2377,"_43":2337,"_35":36,"_167":2925,"_168":94,"_37":2926,"_39":26,"_67":2101,"_171":-5,"_68":24,"_173":-5,"_506":2927,"_17":2928},{"_33":2919,"_13":2918,"_64":2912,"_43":2377,"_35":36,"_167":2920,"_168":94,"_37":2921,"_39":26,"_67":2101,"_171":-5,"_68":24,"_173":-5,"_506":2922,"_17":2923},{"_33":2910,"_13":2069,"_64":2911,"_43":2912,"_35":36,"_167":2913,"_168":94,"_37":2914,"_39":170,"_67":21,"_171":2915,"_68":24,"_173":-5,"_506":2916,"_17":2917},{"_33":2906,"_13":2897,"_64":2891,"_43":2377,"_35":36,"_167":2901,"_168":94,"_37":2907,"_39":26,"_67":2101,"_171":-5,"_68":24,"_173":-5,"_506":2908,"_17":2909},{"_33":2898,"_13":2899,"_64":2900,"_43":2891,"_35":36,"_167":2901,"_168":94,"_37":2902,"_39":26,"_67":21,"_171":2903,"_68":24,"_173":-5,"_506":2904,"_17":2905},{"_33":2888,"_13":2889,"_64":2890,"_43":2891,"_35":36,"_167":2892,"_168":94,"_37":2893,"_39":26,"_67":21,"_171":2894,"_68":24,"_173":-5,"_506":2895,"_17":2896},{"_33":2881,"_13":2882,"_64":2883,"_43":2377,"_35":36,"_167":2884,"_168":94,"_37":2885,"_39":26,"_67":2101,"_171":-5,"_68":24,"_173":-5,"_506":2886,"_17":2887},{"_33":2876,"_13":2570,"_64":2564,"_43":2377,"_35":36,"_167":2877,"_168":94,"_37":2878,"_39":26,"_67":2101,"_171":-5,"_68":24,"_173":-5,"_506":2879,"_17":2880},{"_33":2867,"_13":2868,"_64":2869,"_43":2564,"_35":36,"_167":2870,"_168":94,"_37":2871,"_39":2872,"_67":21,"_171":2873,"_68":24,"_173":-5,"_506":2874,"_17":2875},{"_33":2859,"_13":2860,"_64":2861,"_43":2564,"_35":36,"_167":2862,"_168":94,"_37":2863,"_39":1035,"_67":21,"_171":2864,"_68":24,"_173":-5,"_506":2865,"_17":2866},{"_33":2851,"_13":2852,"_64":2853,"_43":2564,"_35":36,"_167":2854,"_168":94,"_37":2855,"_39":26,"_67":21,"_171":2856,"_68":24,"_173":-5,"_506":2857,"_17":2858},{"_33":2843,"_13":2844,"_64":2845,"_43":2564,"_35":36,"_167":2846,"_168":94,"_37":2847,"_39":26,"_67":21,"_171":2848,"_68":24,"_173":-5,"_506":2849,"_17":2850},{"_33":2835,"_13":2836,"_64":2837,"_43":2564,"_35":36,"_167":2838,"_168":94,"_37":2839,"_39":26,"_67":21,"_171":2840,"_68":24,"_173":-5,"_506":2841,"_17":2842},{"_33":2827,"_13":2828,"_64":2829,"_43":2564,"_35":36,"_167":2830,"_168":94,"_37":2831,"_39":26,"_67":21,"_171":2832,"_68":24,"_173":-5,"_506":2833,"_17":2834},{"_33":2819,"_13":2820,"_64":2821,"_43":2564,"_35":36,"_167":2822,"_168":94,"_37":2823,"_39":26,"_67":21,"_171":2824,"_68":24,"_173":-5,"_506":2825,"_17":2826},{"_33":2811,"_13":2812,"_64":2813,"_43":2564,"_35":36,"_167":2814,"_168":94,"_37":2815,"_39":26,"_67":21,"_171":2816,"_68":24,"_173":-5,"_506":2817,"_17":2818},{"_33":2803,"_13":2804,"_64":2805,"_43":2564,"_35":36,"_167":2806,"_168":94,"_37":2807,"_39":170,"_67":21,"_171":2808,"_68":24,"_173":-5,"_506":2809,"_17":2810},{"_33":2795,"_13":2796,"_64":2797,"_43":2564,"_35":36,"_167":2798,"_168":94,"_37":2799,"_39":170,"_67":21,"_171":2800,"_68":24,"_173":-5,"_506":2801,"_17":2802},{"_33":2787,"_13":2788,"_64":2789,"_43":2564,"_35":36,"_167":2790,"_168":94,"_37":2791,"_39":26,"_67":21,"_171":2792,"_68":24,"_173":-5,"_506":2793,"_17":2794},{"_33":2779,"_13":2780,"_64":2781,"_43":2564,"_35":36,"_167":2782,"_168":94,"_37":2783,"_39":26,"_67":21,"_171":2784,"_68":24,"_173":-5,"_506":2785,"_17":2786},{"_33":2771,"_13":2772,"_64":2773,"_43":2564,"_35":36,"_167":2774,"_168":94,"_37":2775,"_39":26,"_67":21,"_171":2776,"_68":24,"_173":-5,"_506":2777,"_17":2778},{"_33":2763,"_13":2764,"_64":2765,"_43":2564,"_35":36,"_167":2766,"_168":94,"_37":2767,"_39":26,"_67":21,"_171":2768,"_68":24,"_173":-5,"_506":2769,"_17":2770},{"_33":2755,"_13":2756,"_64":2757,"_43":2564,"_35":36,"_167":2758,"_168":94,"_37":2759,"_39":26,"_67":21,"_171":2760,"_68":24,"_173":-5,"_506":2761,"_17":2762},{"_33":2747,"_13":2748,"_64":2749,"_43":2564,"_35":36,"_167":2750,"_168":94,"_37":2751,"_39":26,"_67":21,"_171":2752,"_68":24,"_173":-5,"_506":2753,"_17":2754},{"_33":2739,"_13":2740,"_64":2741,"_43":2564,"_35":36,"_167":2742,"_168":94,"_37":2743,"_39":26,"_67":21,"_171":2744,"_68":24,"_173":-5,"_506":2745,"_17":2746},{"_33":2731,"_13":2732,"_64":2733,"_43":2564,"_35":36,"_167":2734,"_168":94,"_37":2735,"_39":26,"_67":21,"_171":2736,"_68":24,"_173":-5,"_506":2737,"_17":2738},{"_33":2723,"_13":2724,"_64":2725,"_43":2564,"_35":36,"_167":2726,"_168":94,"_37":2727,"_39":26,"_67":21,"_171":2728,"_68":24,"_173":-5,"_506":2729,"_17":2730},{"_33":2715,"_13":2716,"_64":2717,"_43":2564,"_35":36,"_167":2718,"_168":94,"_37":2719,"_39":26,"_67":21,"_171":2720,"_68":24,"_173":-5,"_506":2721,"_17":2722},{"_33":2707,"_13":2708,"_64":2709,"_43":2564,"_35":36,"_167":2710,"_168":94,"_37":2711,"_39":26,"_67":21,"_171":2712,"_68":24,"_173":-5,"_506":2713,"_17":2714},{"_33":2699,"_13":2700,"_64":2701,"_43":2564,"_35":36,"_167":2702,"_168":94,"_37":2703,"_39":26,"_67":21,"_171":2704,"_68":24,"_173":-5,"_506":2705,"_17":2706},{"_33":2691,"_13":2692,"_64":2693,"_43":2564,"_35":36,"_167":2694,"_168":94,"_37":2695,"_39":26,"_67":21,"_171":2696,"_68":24,"_173":-5,"_506":2697,"_17":2698},{"_33":2683,"_13":2684,"_64":2685,"_43":2564,"_35":36,"_167":2686,"_168":94,"_37":2687,"_39":26,"_67":21,"_171":2688,"_68":24,"_173":-5,"_506":2689,"_17":2690},{"_33":2675,"_13":2676,"_64":2677,"_43":2564,"_35":36,"_167":2678,"_168":94,"_37":2679,"_39":26,"_67":21,"_171":2680,"_68":24,"_173":-5,"_506":2681,"_17":2682},{"_33":2667,"_13":2668,"_64":2669,"_43":2564,"_35":36,"_167":2670,"_168":94,"_37":2671,"_39":170,"_67":21,"_171":2672,"_68":24,"_173":-5,"_506":2673,"_17":2674},{"_33":2659,"_13":2660,"_64":2661,"_43":2564,"_35":36,"_167":2662,"_168":94,"_37":2663,"_39":26,"_67":21,"_171":2664,"_68":24,"_173":-5,"_506":2665,"_17":2666},{"_33":2651,"_13":2652,"_64":2653,"_43":2564,"_35":36,"_167":2654,"_168":94,"_37":2655,"_39":26,"_67":21,"_171":2656,"_68":24,"_173":-5,"_506":2657,"_17":2658},{"_33":2643,"_13":2644,"_64":2645,"_43":2564,"_35":36,"_167":2646,"_168":94,"_37":2647,"_39":26,"_67":21,"_171":2648,"_68":24,"_173":-5,"_506":2649,"_17":2650},{"_33":2635,"_13":2636,"_64":2637,"_43":2564,"_35":36,"_167":2638,"_168":94,"_37":2639,"_39":26,"_67":21,"_171":2640,"_68":24,"_173":-5,"_506":2641,"_17":2642},{"_33":2627,"_13":2628,"_64":2629,"_43":2564,"_35":36,"_167":2630,"_168":94,"_37":2631,"_39":26,"_67":21,"_171":2632,"_68":24,"_173":-5,"_506":2633,"_17":2634},{"_33":2619,"_13":2620,"_64":2621,"_43":2564,"_35":36,"_167":2622,"_168":94,"_37":2623,"_39":26,"_67":21,"_171":2624,"_68":24,"_173":-5,"_506":2625,"_17":2626},{"_33":2611,"_13":2612,"_64":2613,"_43":2564,"_35":36,"_167":2614,"_168":94,"_37":2615,"_39":26,"_67":21,"_171":2616,"_68":24,"_173":-5,"_506":2617,"_17":2618},{"_33":2603,"_13":2604,"_64":2605,"_43":2564,"_35":36,"_167":2606,"_168":94,"_37":2607,"_39":26,"_67":21,"_171":2608,"_68":24,"_173":-5,"_506":2609,"_17":2610},{"_33":2595,"_13":2596,"_64":2597,"_43":2564,"_35":36,"_167":2598,"_168":94,"_37":2599,"_39":26,"_67":21,"_171":2600,"_68":24,"_173":-5,"_506":2601,"_17":2602},{"_33":2587,"_13":2588,"_64":2589,"_43":2564,"_35":36,"_167":2590,"_168":94,"_37":2591,"_39":26,"_67":21,"_171":2592,"_68":24,"_173":-5,"_506":2593,"_17":2594},{"_33":2579,"_13":2580,"_64":2581,"_43":2564,"_35":36,"_167":2582,"_168":94,"_37":2583,"_39":26,"_67":21,"_171":2584,"_68":24,"_173":-5,"_506":2585,"_17":2586},{"_33":2571,"_13":2572,"_64":2573,"_43":2564,"_35":36,"_167":2574,"_168":94,"_37":2575,"_39":26,"_67":21,"_171":2576,"_68":24,"_173":-5,"_506":2577,"_17":2578},{"_33":2561,"_13":2562,"_64":2563,"_43":2564,"_35":36,"_167":2565,"_168":94,"_37":2566,"_39":26,"_67":21,"_171":2567,"_68":24,"_173":-5,"_506":2568,"_17":2569},{"_33":2557,"_13":177,"_64":2402,"_43":2377,"_35":36,"_167":40,"_168":94,"_37":2558,"_39":26,"_67":2101,"_171":-5,"_68":24,"_173":-5,"_506":2559,"_17":2560},{"_33":2553,"_13":178,"_64":166,"_43":2402,"_35":36,"_167":40,"_168":94,"_37":2554,"_39":26,"_67":2101,"_171":-5,"_68":24,"_173":-5,"_506":2555,"_17":2556},{"_33":2545,"_13":2546,"_64":2547,"_43":166,"_35":36,"_167":2548,"_168":94,"_37":2549,"_39":170,"_67":21,"_171":2550,"_68":24,"_173":-5,"_506":2551,"_17":2552},{"_33":2537,"_13":2538,"_64":2539,"_43":166,"_35":36,"_167":2540,"_168":94,"_37":2541,"_39":2184,"_67":21,"_171":2542,"_68":24,"_173":-5,"_506":2543,"_17":2544},{"_33":2529,"_13":2530,"_64":2531,"_43":166,"_35":36,"_167":2532,"_168":94,"_37":2533,"_39":961,"_67":21,"_171":2534,"_68":24,"_173":-5,"_506":2535,"_17":2536},{"_33":2521,"_13":2522,"_64":2523,"_43":166,"_35":36,"_167":2524,"_168":94,"_37":2525,"_39":170,"_67":21,"_171":2526,"_68":24,"_173":-5,"_506":2527,"_17":2528},{"_33":2513,"_13":2514,"_64":2515,"_43":166,"_35":36,"_167":2516,"_168":94,"_37":2517,"_39":970,"_67":21,"_171":2518,"_68":24,"_173":-5,"_506":2519,"_17":2520},{"_33":165,"_13":66,"_64":65,"_43":166,"_35":36,"_167":74,"_168":94,"_37":2510,"_39":170,"_67":21,"_171":172,"_68":24,"_173":-5,"_506":2511,"_17":2512},{"_33":2502,"_13":2503,"_64":2504,"_43":166,"_35":36,"_167":2505,"_168":94,"_37":2506,"_39":970,"_67":21,"_171":2507,"_68":24,"_173":-5,"_506":2508,"_17":2509},{"_33":2494,"_13":2495,"_64":2496,"_43":166,"_35":36,"_167":2497,"_168":94,"_37":2498,"_39":170,"_67":21,"_171":2499,"_68":24,"_173":-5,"_506":2500,"_17":2501},{"_33":2486,"_13":2487,"_64":2488,"_43":166,"_35":36,"_167":2489,"_168":94,"_37":2490,"_39":970,"_67":21,"_171":2491,"_68":24,"_173":-5,"_506":2492,"_17":2493},{"_33":2478,"_13":2479,"_64":2480,"_43":2402,"_35":36,"_167":2481,"_168":94,"_37":2482,"_39":26,"_67":21,"_171":2483,"_68":24,"_173":-5,"_506":2484,"_17":2485},{"_33":2470,"_13":2471,"_64":2472,"_43":2402,"_35":36,"_167":2473,"_168":94,"_37":2474,"_39":2361,"_67":21,"_171":2475,"_68":24,"_173":-5,"_506":2476,"_17":2477},{"_33":2462,"_13":2463,"_64":2464,"_43":2402,"_35":36,"_167":2465,"_168":94,"_37":2466,"_39":170,"_67":21,"_171":2467,"_68":24,"_173":-5,"_506":2468,"_17":2469},{"_33":2454,"_13":2455,"_64":2456,"_43":2402,"_35":36,"_167":2457,"_168":94,"_37":2458,"_39":2042,"_67":21,"_171":2459,"_68":24,"_173":-5,"_506":2460,"_17":2461},{"_33":2447,"_13":2448,"_64":2449,"_43":2402,"_35":36,"_167":40,"_168":94,"_37":2450,"_39":26,"_67":21,"_171":2451,"_68":24,"_173":-5,"_506":2452,"_17":2453},{"_33":2439,"_13":2440,"_64":2441,"_43":2402,"_35":36,"_167":2442,"_168":94,"_37":2443,"_39":961,"_67":21,"_171":2444,"_68":24,"_173":-5,"_506":2445,"_17":2446},{"_33":2432,"_13":2433,"_64":2434,"_43":2402,"_35":36,"_167":40,"_168":94,"_37":2435,"_39":26,"_67":21,"_171":2436,"_68":24,"_173":-5,"_506":2437,"_17":2438},{"_33":2424,"_13":2425,"_64":2426,"_43":2402,"_35":36,"_167":2427,"_168":94,"_37":2428,"_39":181,"_67":21,"_171":2429,"_68":24,"_173":-5,"_506":2430,"_17":2431},{"_33":2417,"_13":2418,"_64":2419,"_43":2402,"_35":36,"_167":40,"_168":94,"_37":2420,"_39":26,"_67":21,"_171":2421,"_68":24,"_173":-5,"_506":2422,"_17":2423},{"_33":2409,"_13":2410,"_64":2411,"_43":2402,"_35":36,"_167":2412,"_168":94,"_37":2413,"_39":170,"_67":21,"_171":2414,"_68":24,"_173":-5,"_506":2415,"_17":2416},{"_33":2399,"_13":2400,"_64":2401,"_43":2402,"_35":36,"_167":2403,"_168":94,"_37":2404,"_39":2405,"_67":21,"_171":2406,"_68":24,"_173":-5,"_506":2407,"_17":2408},{"_33":2391,"_13":2392,"_64":2393,"_43":2377,"_35":36,"_167":2394,"_168":94,"_37":2395,"_39":935,"_67":21,"_171":2396,"_68":24,"_173":-5,"_506":2397,"_17":2398},{"_33":2383,"_13":2384,"_64":2385,"_43":2377,"_35":36,"_167":2386,"_168":94,"_37":2387,"_39":181,"_67":21,"_171":2388,"_68":24,"_173":-5,"_506":2389,"_17":2390},{"_33":2374,"_13":2375,"_64":2376,"_43":2377,"_35":36,"_167":2378,"_168":94,"_37":2379,"_39":26,"_67":21,"_171":2380,"_68":24,"_173":-5,"_506":2381,"_17":2382},{"_33":2366,"_13":2367,"_64":2368,"_43":2337,"_35":36,"_167":2369,"_168":94,"_37":2370,"_39":26,"_67":21,"_171":2371,"_68":24,"_173":-5,"_506":2372,"_17":2373},{"_33":2358,"_13":2359,"_64":2360,"_43":2337,"_35":36,"_167":2361,"_168":94,"_37":2362,"_39":26,"_67":21,"_171":2363,"_68":24,"_173":-5,"_506":2364,"_17":2365},{"_33":2351,"_13":2352,"_64":2353,"_43":2337,"_35":36,"_167":961,"_168":94,"_37":2354,"_39":26,"_67":21,"_171":2355,"_68":24,"_173":-5,"_506":2356,"_17":2357},{"_33":2343,"_13":2344,"_64":2345,"_43":2337,"_35":36,"_167":2346,"_168":94,"_37":2347,"_39":26,"_67":21,"_171":2348,"_68":24,"_173":-5,"_506":2349,"_17":2350},{"_33":2334,"_13":2335,"_64":2336,"_43":2337,"_35":36,"_167":2338,"_168":94,"_37":2339,"_39":26,"_67":21,"_171":2340,"_68":24,"_173":-5,"_506":2341,"_17":2342},{"_33":2330,"_13":2294,"_64":2289,"_43":-5,"_35":36,"_167":170,"_168":94,"_37":2331,"_39":26,"_67":2101,"_171":-5,"_68":24,"_173":-5,"_506":2332,"_17":2333},{"_33":2323,"_13":2324,"_64":2325,"_43":2289,"_35":36,"_167":170,"_168":94,"_37":2326,"_39":26,"_67":21,"_171":2327,"_68":24,"_173":-5,"_506":2328,"_17":2329},{"_33":2316,"_13":2317,"_64":2318,"_43":2289,"_35":36,"_167":170,"_168":94,"_37":2319,"_39":26,"_67":21,"_171":2320,"_68":24,"_173":-5,"_506":2321,"_17":2322},{"_33":2309,"_13":2310,"_64":2311,"_43":2289,"_35":36,"_167":170,"_168":94,"_37":2312,"_39":26,"_67":21,"_171":2313,"_68":24,"_173":-5,"_506":2314,"_17":2315},{"_33":2302,"_13":2303,"_64":2304,"_43":2289,"_35":36,"_167":170,"_168":94,"_37":2305,"_39":26,"_67":21,"_171":2306,"_68":24,"_173":-5,"_506":2307,"_17":2308},{"_33":2295,"_13":2296,"_64":2297,"_43":2289,"_35":36,"_167":170,"_168":94,"_37":2298,"_39":26,"_67":21,"_171":2299,"_68":24,"_173":-5,"_506":2300,"_17":2301},{"_33":2286,"_13":2287,"_64":2288,"_43":2289,"_35":36,"_167":170,"_168":94,"_37":2290,"_39":26,"_67":21,"_171":2291,"_68":24,"_173":-5,"_506":2292,"_17":2293},{"_33":2282,"_13":2112,"_64":2107,"_43":-5,"_35":36,"_167":170,"_168":94,"_37":2283,"_39":26,"_67":2101,"_171":-5,"_68":24,"_173":-5,"_506":2284,"_17":2285},{"_33":2275,"_13":2276,"_64":2277,"_43":2107,"_35":36,"_167":170,"_168":94,"_37":2278,"_39":26,"_67":504,"_171":2279,"_68":24,"_173":-5,"_506":2280,"_17":2281},{"_33":2268,"_13":2269,"_64":2270,"_43":2107,"_35":36,"_167":170,"_168":94,"_37":2271,"_39":26,"_67":504,"_171":2272,"_68":24,"_173":-5,"_506":2273,"_17":2274},{"_33":2261,"_13":2262,"_64":2263,"_43":2107,"_35":36,"_167":1035,"_168":94,"_37":2264,"_39":26,"_67":504,"_171":2265,"_68":24,"_173":-5,"_506":2266,"_17":2267},{"_33":2253,"_13":2254,"_64":2255,"_43":2107,"_35":36,"_167":2256,"_168":94,"_37":2257,"_39":970,"_67":504,"_171":2258,"_68":24,"_173":-5,"_506":2259,"_17":2260},{"_33":2245,"_13":2246,"_64":2247,"_43":2107,"_35":36,"_167":2248,"_168":94,"_37":2249,"_39":26,"_67":504,"_171":2250,"_68":24,"_173":-5,"_506":2251,"_17":2252},{"_33":2237,"_13":2238,"_64":2239,"_43":2107,"_35":36,"_167":2240,"_168":94,"_37":2241,"_39":170,"_67":504,"_171":2242,"_68":24,"_173":-5,"_506":2243,"_17":2244},{"_33":2229,"_13":2230,"_64":2231,"_43":2107,"_35":36,"_167":2232,"_168":94,"_37":2233,"_39":935,"_67":504,"_171":2234,"_68":24,"_173":-5,"_506":2235,"_17":2236},{"_33":2221,"_13":2222,"_64":2223,"_43":2107,"_35":36,"_167":2224,"_168":94,"_37":2225,"_39":170,"_67":504,"_171":2226,"_68":24,"_173":-5,"_506":2227,"_17":2228},{"_33":2213,"_13":2214,"_64":2215,"_43":2107,"_35":36,"_167":2216,"_168":94,"_37":2217,"_39":26,"_67":504,"_171":2218,"_68":24,"_173":-5,"_506":2219,"_17":2220},{"_33":2205,"_13":2206,"_64":2207,"_43":2107,"_35":36,"_167":2208,"_168":94,"_37":2209,"_39":26,"_67":504,"_171":2210,"_68":24,"_173":-5,"_506":2211,"_17":2212},{"_33":2197,"_13":2198,"_64":2199,"_43":2107,"_35":36,"_167":2200,"_168":94,"_37":2201,"_39":26,"_67":504,"_171":2202,"_68":24,"_173":-5,"_506":2203,"_17":2204},{"_33":2189,"_13":2190,"_64":2191,"_43":2107,"_35":36,"_167":2192,"_168":94,"_37":2193,"_39":26,"_67":504,"_171":2194,"_68":24,"_173":-5,"_506":2195,"_17":2196},{"_33":2181,"_13":2182,"_64":2183,"_43":2107,"_35":36,"_167":2184,"_168":94,"_37":2185,"_39":26,"_67":504,"_171":2186,"_68":24,"_173":-5,"_506":2187,"_17":2188},{"_33":2173,"_13":2174,"_64":2175,"_43":2107,"_35":36,"_167":2176,"_168":94,"_37":2177,"_39":26,"_67":504,"_171":2178,"_68":24,"_173":-5,"_506":2179,"_17":2180},{"_33":2165,"_13":2166,"_64":2167,"_43":2107,"_35":36,"_167":2168,"_168":94,"_37":2169,"_39":26,"_67":504,"_171":2170,"_68":24,"_173":-5,"_506":2171,"_17":2172},{"_33":2157,"_13":2158,"_64":2159,"_43":2107,"_35":36,"_167":2160,"_168":94,"_37":2161,"_39":961,"_67":504,"_171":2162,"_68":24,"_173":-5,"_506":2163,"_17":2164},{"_33":2149,"_13":2150,"_64":2151,"_43":2107,"_35":36,"_167":2152,"_168":94,"_37":2153,"_39":26,"_67":504,"_171":2154,"_68":24,"_173":-5,"_506":2155,"_17":2156},{"_33":2141,"_13":2142,"_64":2143,"_43":2107,"_35":36,"_167":2144,"_168":94,"_37":2145,"_39":26,"_67":504,"_171":2146,"_68":24,"_173":-5,"_506":2147,"_17":2148},{"_33":2134,"_13":2135,"_64":2136,"_43":2107,"_35":36,"_167":944,"_168":94,"_37":2137,"_39":26,"_67":504,"_171":2138,"_68":24,"_173":-5,"_506":2139,"_17":2140},{"_33":2127,"_13":2128,"_64":2129,"_43":2107,"_35":36,"_167":170,"_168":94,"_37":2130,"_39":26,"_67":504,"_171":2131,"_68":24,"_173":-5,"_506":2132,"_17":2133},{"_33":2120,"_13":2121,"_64":2122,"_43":2107,"_35":36,"_167":170,"_168":94,"_37":2123,"_39":26,"_67":504,"_171":2124,"_68":24,"_173":-5,"_506":2125,"_17":2126},{"_33":2113,"_13":2114,"_64":2115,"_43":2107,"_35":36,"_167":170,"_168":94,"_37":2116,"_39":26,"_67":504,"_171":2117,"_68":24,"_173":-5,"_506":2118,"_17":2119},{"_33":2104,"_13":2105,"_64":2106,"_43":2107,"_35":36,"_167":170,"_168":94,"_37":2108,"_39":26,"_67":504,"_171":2109,"_68":24,"_173":-5,"_506":2110,"_17":2111},{"_33":2099,"_13":2084,"_64":2079,"_43":-5,"_35":36,"_167":170,"_168":94,"_37":2100,"_39":26,"_67":2101,"_171":-5,"_68":24,"_173":-5,"_506":2102,"_17":2103},{"_33":2092,"_13":2093,"_64":2094,"_43":2079,"_35":36,"_167":170,"_168":94,"_37":2095,"_39":26,"_67":504,"_171":2096,"_68":24,"_173":-5,"_506":2097,"_17":2098},{"_33":2085,"_13":2086,"_64":2087,"_43":2079,"_35":36,"_167":170,"_168":94,"_37":2088,"_39":26,"_67":504,"_171":2089,"_68":24,"_173":-5,"_506":2090,"_17":2091},{"_33":2076,"_13":2077,"_64":2078,"_43":2079,"_35":36,"_167":170,"_168":94,"_37":2080,"_39":26,"_67":504,"_171":2081,"_68":24,"_173":-5,"_506":2082,"_17":2083},{"_33":2068,"_13":2069,"_64":2070,"_43":-5,"_35":36,"_167":2071,"_168":94,"_37":2072,"_39":170,"_67":21,"_171":2073,"_68":24,"_173":-5,"_506":2074,"_17":2075},{"_33":2061,"_13":2062,"_64":2063,"_43":-5,"_35":36,"_167":181,"_168":94,"_37":2064,"_39":26,"_67":504,"_171":2065,"_68":24,"_173":-5,"_506":2066,"_17":2067},{"_33":2054,"_13":2055,"_64":2056,"_43":-5,"_35":36,"_167":970,"_168":94,"_37":2057,"_39":26,"_67":504,"_171":2058,"_68":24,"_173":-5,"_506":2059,"_17":2060},{"_33":2047,"_13":2048,"_64":2049,"_43":-5,"_35":36,"_167":910,"_168":94,"_37":2050,"_39":170,"_67":504,"_171":2051,"_68":24,"_173":-5,"_506":2052,"_17":2053},{"_33":2039,"_13":2040,"_64":2041,"_43":-5,"_35":36,"_167":2042,"_168":94,"_37":2043,"_39":26,"_67":504,"_171":2044,"_68":24,"_173":-5,"_506":2045,"_17":2046},{"_33":2031,"_13":2032,"_64":2033,"_43":-5,"_35":36,"_167":2034,"_168":94,"_37":2035,"_39":1035,"_67":504,"_171":2036,"_68":24,"_173":-5,"_506":2037,"_17":2038},{"_33":2023,"_13":2024,"_64":2025,"_43":-5,"_35":36,"_167":2026,"_168":94,"_37":2027,"_39":26,"_67":504,"_171":2028,"_68":24,"_173":-5,"_506":2029,"_17":2030},{"_33":2015,"_13":2016,"_64":2017,"_43":-5,"_35":36,"_167":2018,"_168":94,"_37":2019,"_39":26,"_67":504,"_171":2020,"_68":24,"_173":-5,"_506":2021,"_17":2022},{"_33":2007,"_13":2008,"_64":2009,"_43":-5,"_35":36,"_167":2010,"_168":94,"_37":2011,"_39":26,"_67":504,"_171":2012,"_68":24,"_173":-5,"_506":2013,"_17":2014},{"_33":1999,"_13":2000,"_64":2001,"_43":-5,"_35":36,"_167":2002,"_168":94,"_37":2003,"_39":26,"_67":504,"_171":2004,"_68":24,"_173":-5,"_506":2005,"_17":2006},{"_33":1991,"_13":1992,"_64":1993,"_43":-5,"_35":36,"_167":1994,"_168":94,"_37":1995,"_39":1035,"_67":504,"_171":1996,"_68":24,"_173":-5,"_506":1997,"_17":1998},{"_33":1982,"_13":1983,"_64":1984,"_43":-5,"_35":36,"_167":1985,"_168":94,"_37":1986,"_39":181,"_67":1987,"_171":1988,"_68":24,"_173":-5,"_506":1989,"_17":1990},{"_33":1974,"_13":1975,"_64":1976,"_43":-5,"_35":36,"_167":1977,"_168":94,"_37":1978,"_39":26,"_67":504,"_171":1979,"_68":24,"_173":-5,"_506":1980,"_17":1981},{"_33":1966,"_13":1967,"_64":1968,"_43":-5,"_35":36,"_167":1969,"_168":94,"_37":1970,"_39":26,"_67":504,"_171":1971,"_68":24,"_173":-5,"_506":1972,"_17":1973},{"_33":1958,"_13":1959,"_64":1960,"_43":-5,"_35":36,"_167":1961,"_168":94,"_37":1962,"_39":26,"_67":504,"_171":1963,"_68":24,"_173":-5,"_506":1964,"_17":1965},{"_33":1950,"_13":1951,"_64":1952,"_43":-5,"_35":36,"_167":1953,"_168":94,"_37":1954,"_39":26,"_67":504,"_171":1955,"_68":24,"_173":-5,"_506":1956,"_17":1957},{"_33":1942,"_13":1943,"_64":1944,"_43":-5,"_35":36,"_167":1945,"_168":94,"_37":1946,"_39":26,"_67":504,"_171":1947,"_68":24,"_173":-5,"_506":1948,"_17":1949},{"_33":1934,"_13":1935,"_64":1936,"_43":-5,"_35":36,"_167":1937,"_168":94,"_37":1938,"_39":26,"_67":504,"_171":1939,"_68":24,"_173":-5,"_506":1940,"_17":1941},{"_33":1926,"_13":1927,"_64":1928,"_43":-5,"_35":36,"_167":1929,"_168":94,"_37":1930,"_39":26,"_67":504,"_171":1931,"_68":24,"_173":-5,"_506":1932,"_17":1933},{"_33":1918,"_13":1919,"_64":1920,"_43":-5,"_35":36,"_167":1921,"_168":94,"_37":1922,"_39":765,"_67":504,"_171":1923,"_68":24,"_173":-5,"_506":1924,"_17":1925},{"_33":1910,"_13":1911,"_64":1912,"_43":-5,"_35":36,"_167":1913,"_168":94,"_37":1914,"_39":1035,"_67":504,"_171":1915,"_68":24,"_173":-5,"_506":1916,"_17":1917},{"_33":1902,"_13":1903,"_64":1904,"_43":-5,"_35":36,"_167":1905,"_168":94,"_37":1906,"_39":765,"_67":504,"_171":1907,"_68":24,"_173":-5,"_506":1908,"_17":1909},{"_33":1894,"_13":1895,"_64":1896,"_43":-5,"_35":36,"_167":1897,"_168":94,"_37":1898,"_39":26,"_67":504,"_171":1899,"_68":24,"_173":-5,"_506":1900,"_17":1901},{"_33":1886,"_13":1887,"_64":1888,"_43":-5,"_35":36,"_167":1889,"_168":94,"_37":1890,"_39":1035,"_67":504,"_171":1891,"_68":24,"_173":-5,"_506":1892,"_17":1893},{"_33":1878,"_13":1879,"_64":1880,"_43":-5,"_35":36,"_167":1881,"_168":94,"_37":1882,"_39":170,"_67":504,"_171":1883,"_68":24,"_173":-5,"_506":1884,"_17":1885},{"_33":1870,"_13":1871,"_64":1872,"_43":-5,"_35":36,"_167":1873,"_168":94,"_37":1874,"_39":970,"_67":504,"_171":1875,"_68":24,"_173":-5,"_506":1876,"_17":1877},{"_33":1862,"_13":1863,"_64":1864,"_43":-5,"_35":36,"_167":1865,"_168":94,"_37":1866,"_39":970,"_67":504,"_171":1867,"_68":24,"_173":-5,"_506":1868,"_17":1869},{"_33":1854,"_13":1855,"_64":1856,"_43":-5,"_35":36,"_167":1857,"_168":94,"_37":1858,"_39":170,"_67":504,"_171":1859,"_68":24,"_173":-5,"_506":1860,"_17":1861},{"_33":1846,"_13":1847,"_64":1848,"_43":-5,"_35":36,"_167":1849,"_168":94,"_37":1850,"_39":935,"_67":504,"_171":1851,"_68":24,"_173":-5,"_506":1852,"_17":1853},{"_33":1838,"_13":1839,"_64":1840,"_43":-5,"_35":36,"_167":1841,"_168":94,"_37":1842,"_39":765,"_67":504,"_171":1843,"_68":24,"_173":-5,"_506":1844,"_17":1845},{"_33":1830,"_13":1831,"_64":1832,"_43":-5,"_35":36,"_167":1833,"_168":94,"_37":1834,"_39":181,"_67":504,"_171":1835,"_68":24,"_173":-5,"_506":1836,"_17":1837},{"_33":1822,"_13":1823,"_64":1824,"_43":-5,"_35":36,"_167":1825,"_168":94,"_37":1826,"_39":181,"_67":504,"_171":1827,"_68":24,"_173":-5,"_506":1828,"_17":1829},{"_33":1814,"_13":1815,"_64":1816,"_43":-5,"_35":36,"_167":1817,"_168":94,"_37":1818,"_39":944,"_67":504,"_171":1819,"_68":24,"_173":-5,"_506":1820,"_17":1821},{"_33":1806,"_13":1807,"_64":1808,"_43":-5,"_35":36,"_167":1809,"_168":94,"_37":1810,"_39":970,"_67":504,"_171":1811,"_68":24,"_173":-5,"_506":1812,"_17":1813},{"_33":1798,"_13":1799,"_64":1800,"_43":-5,"_35":36,"_167":1801,"_168":94,"_37":1802,"_39":1035,"_67":504,"_171":1803,"_68":24,"_173":-5,"_506":1804,"_17":1805},{"_33":1790,"_13":1791,"_64":1792,"_43":-5,"_35":36,"_167":1793,"_168":94,"_37":1794,"_39":26,"_67":504,"_171":1795,"_68":24,"_173":-5,"_506":1796,"_17":1797},{"_33":1782,"_13":1783,"_64":1784,"_43":-5,"_35":36,"_167":1785,"_168":94,"_37":1786,"_39":970,"_67":504,"_171":1787,"_68":24,"_173":-5,"_506":1788,"_17":1789},{"_33":1774,"_13":1775,"_64":1776,"_43":-5,"_35":36,"_167":1777,"_168":94,"_37":1778,"_39":181,"_67":504,"_171":1779,"_68":24,"_173":-5,"_506":1780,"_17":1781},{"_33":1766,"_13":1767,"_64":1768,"_43":-5,"_35":36,"_167":1769,"_168":94,"_37":1770,"_39":970,"_67":504,"_171":1771,"_68":24,"_173":-5,"_506":1772,"_17":1773},{"_33":1758,"_13":1759,"_64":1760,"_43":-5,"_35":36,"_167":1761,"_168":94,"_37":1762,"_39":26,"_67":504,"_171":1763,"_68":24,"_173":-5,"_506":1764,"_17":1765},{"_33":1750,"_13":1751,"_64":1752,"_43":-5,"_35":36,"_167":1753,"_168":94,"_37":1754,"_39":1035,"_67":504,"_171":1755,"_68":24,"_173":-5,"_506":1756,"_17":1757},{"_33":1742,"_13":1743,"_64":1744,"_43":-5,"_35":36,"_167":1745,"_168":94,"_37":1746,"_39":1035,"_67":504,"_171":1747,"_68":24,"_173":-5,"_506":1748,"_17":1749},{"_33":1734,"_13":1735,"_64":1736,"_43":-5,"_35":36,"_167":1737,"_168":94,"_37":1738,"_39":170,"_67":504,"_171":1739,"_68":24,"_173":-5,"_506":1740,"_17":1741},{"_33":1726,"_13":1727,"_64":1728,"_43":-5,"_35":36,"_167":1729,"_168":94,"_37":1730,"_39":170,"_67":504,"_171":1731,"_68":24,"_173":-5,"_506":1732,"_17":1733},{"_33":1718,"_13":1719,"_64":1720,"_43":-5,"_35":36,"_167":1721,"_168":94,"_37":1722,"_39":26,"_67":504,"_171":1723,"_68":24,"_173":-5,"_506":1724,"_17":1725},{"_33":1710,"_13":1711,"_64":1712,"_43":-5,"_35":36,"_167":1713,"_168":94,"_37":1714,"_39":26,"_67":504,"_171":1715,"_68":24,"_173":-5,"_506":1716,"_17":1717},{"_33":1702,"_13":1703,"_64":1704,"_43":-5,"_35":36,"_167":1705,"_168":94,"_37":1706,"_39":170,"_67":504,"_171":1707,"_68":24,"_173":-5,"_506":1708,"_17":1709},{"_33":1694,"_13":1695,"_64":1696,"_43":-5,"_35":36,"_167":1697,"_168":94,"_37":1698,"_39":1035,"_67":504,"_171":1699,"_68":24,"_173":-5,"_506":1700,"_17":1701},{"_33":1686,"_13":1687,"_64":1688,"_43":-5,"_35":36,"_167":1689,"_168":94,"_37":1690,"_39":1035,"_67":504,"_171":1691,"_68":24,"_173":-5,"_506":1692,"_17":1693},{"_33":1678,"_13":1679,"_64":1680,"_43":-5,"_35":36,"_167":1681,"_168":94,"_37":1682,"_39":26,"_67":504,"_171":1683,"_68":24,"_173":-5,"_506":1684,"_17":1685},{"_33":1670,"_13":1671,"_64":1672,"_43":-5,"_35":36,"_167":1673,"_168":94,"_37":1674,"_39":26,"_67":504,"_171":1675,"_68":24,"_173":-5,"_506":1676,"_17":1677},{"_33":1662,"_13":1663,"_64":1664,"_43":-5,"_35":36,"_167":1665,"_168":94,"_37":1666,"_39":26,"_67":504,"_171":1667,"_68":24,"_173":-5,"_506":1668,"_17":1669},{"_33":1654,"_13":1655,"_64":1656,"_43":-5,"_35":36,"_167":1657,"_168":94,"_37":1658,"_39":26,"_67":504,"_171":1659,"_68":24,"_173":-5,"_506":1660,"_17":1661},{"_33":1646,"_13":1647,"_64":1648,"_43":-5,"_35":36,"_167":1649,"_168":94,"_37":1650,"_39":1035,"_67":504,"_171":1651,"_68":24,"_173":-5,"_506":1652,"_17":1653},{"_33":1638,"_13":1639,"_64":1640,"_43":-5,"_35":36,"_167":1641,"_168":94,"_37":1642,"_39":26,"_67":504,"_171":1643,"_68":24,"_173":-5,"_506":1644,"_17":1645},{"_33":1630,"_13":1631,"_64":1632,"_43":-5,"_35":36,"_167":1633,"_168":94,"_37":1634,"_39":26,"_67":504,"_171":1635,"_68":24,"_173":-5,"_506":1636,"_17":1637},{"_33":1622,"_13":1623,"_64":1624,"_43":-5,"_35":36,"_167":1625,"_168":94,"_37":1626,"_39":170,"_67":504,"_171":1627,"_68":24,"_173":-5,"_506":1628,"_17":1629},{"_33":1614,"_13":1615,"_64":1616,"_43":-5,"_35":36,"_167":1617,"_168":94,"_37":1618,"_39":26,"_67":504,"_171":1619,"_68":24,"_173":-5,"_506":1620,"_17":1621},{"_33":1606,"_13":1607,"_64":1608,"_43":-5,"_35":36,"_167":1609,"_168":94,"_37":1610,"_39":26,"_67":504,"_171":1611,"_68":24,"_173":-5,"_506":1612,"_17":1613},{"_33":1598,"_13":1599,"_64":1600,"_43":-5,"_35":36,"_167":1601,"_168":94,"_37":1602,"_39":170,"_67":504,"_171":1603,"_68":24,"_173":-5,"_506":1604,"_17":1605},{"_33":1590,"_13":1591,"_64":1592,"_43":-5,"_35":36,"_167":1593,"_168":94,"_37":1594,"_39":26,"_67":504,"_171":1595,"_68":24,"_173":-5,"_506":1596,"_17":1597},{"_33":1582,"_13":1583,"_64":1584,"_43":-5,"_35":36,"_167":1585,"_168":94,"_37":1586,"_39":1035,"_67":504,"_171":1587,"_68":24,"_173":-5,"_506":1588,"_17":1589},{"_33":1574,"_13":1575,"_64":1576,"_43":-5,"_35":36,"_167":1577,"_168":94,"_37":1578,"_39":170,"_67":504,"_171":1579,"_68":24,"_173":-5,"_506":1580,"_17":1581},{"_33":1566,"_13":1567,"_64":1568,"_43":-5,"_35":36,"_167":1569,"_168":94,"_37":1570,"_39":935,"_67":504,"_171":1571,"_68":24,"_173":-5,"_506":1572,"_17":1573},{"_33":1558,"_13":1559,"_64":1560,"_43":-5,"_35":36,"_167":1561,"_168":94,"_37":1562,"_39":26,"_67":504,"_171":1563,"_68":24,"_173":-5,"_506":1564,"_17":1565},{"_33":1550,"_13":1551,"_64":1552,"_43":-5,"_35":36,"_167":1553,"_168":94,"_37":1554,"_39":26,"_67":504,"_171":1555,"_68":24,"_173":-5,"_506":1556,"_17":1557},{"_33":1542,"_13":1543,"_64":1544,"_43":-5,"_35":36,"_167":1545,"_168":94,"_37":1546,"_39":26,"_67":504,"_171":1547,"_68":24,"_173":-5,"_506":1548,"_17":1549},{"_33":1534,"_13":1535,"_64":1536,"_43":-5,"_35":36,"_167":1537,"_168":94,"_37":1538,"_39":26,"_67":504,"_171":1539,"_68":24,"_173":-5,"_506":1540,"_17":1541},{"_33":1526,"_13":1527,"_64":1528,"_43":-5,"_35":36,"_167":1529,"_168":94,"_37":1530,"_39":26,"_67":504,"_171":1531,"_68":24,"_173":-5,"_506":1532,"_17":1533},{"_33":1518,"_13":1519,"_64":1520,"_43":-5,"_35":36,"_167":1521,"_168":94,"_37":1522,"_39":765,"_67":504,"_171":1523,"_68":24,"_173":-5,"_506":1524,"_17":1525},{"_33":1511,"_13":1512,"_64":1513,"_43":-5,"_35":36,"_167":1514,"_168":94,"_37":1515,"_39":26,"_67":504,"_171":962,"_68":24,"_173":-5,"_506":1516,"_17":1517},{"_33":1503,"_13":1504,"_64":1505,"_43":-5,"_35":36,"_167":1506,"_168":94,"_37":1507,"_39":961,"_67":504,"_171":1508,"_68":24,"_173":-5,"_506":1509,"_17":1510},{"_33":1495,"_13":1496,"_64":1497,"_43":-5,"_35":36,"_167":1498,"_168":94,"_37":1499,"_39":26,"_67":504,"_171":1500,"_68":24,"_173":-5,"_506":1501,"_17":1502},{"_33":1487,"_13":1488,"_64":1489,"_43":-5,"_35":36,"_167":1490,"_168":94,"_37":1491,"_39":26,"_67":504,"_171":1492,"_68":24,"_173":-5,"_506":1493,"_17":1494},{"_33":1479,"_13":1480,"_64":1481,"_43":-5,"_35":36,"_167":1482,"_168":94,"_37":1483,"_39":170,"_67":504,"_171":1484,"_68":24,"_173":-5,"_506":1485,"_17":1486},{"_33":1471,"_13":1472,"_64":1473,"_43":-5,"_35":36,"_167":1474,"_168":94,"_37":1475,"_39":26,"_67":504,"_171":1476,"_68":24,"_173":-5,"_506":1477,"_17":1478},{"_33":1463,"_13":1464,"_64":1465,"_43":-5,"_35":36,"_167":1466,"_168":94,"_37":1467,"_39":970,"_67":504,"_171":1468,"_68":24,"_173":-5,"_506":1469,"_17":1470},{"_33":1455,"_13":1456,"_64":1457,"_43":-5,"_35":36,"_167":1458,"_168":94,"_37":1459,"_39":26,"_67":504,"_171":1460,"_68":24,"_173":-5,"_506":1461,"_17":1462},{"_33":1447,"_13":1448,"_64":1449,"_43":-5,"_35":36,"_167":1450,"_168":94,"_37":1451,"_39":26,"_67":504,"_171":1452,"_68":24,"_173":-5,"_506":1453,"_17":1454},{"_33":1439,"_13":1440,"_64":1441,"_43":-5,"_35":36,"_167":1442,"_168":94,"_37":1443,"_39":26,"_67":504,"_171":1444,"_68":24,"_173":-5,"_506":1445,"_17":1446},{"_33":1431,"_13":1432,"_64":1433,"_43":-5,"_35":36,"_167":1434,"_168":94,"_37":1435,"_39":26,"_67":504,"_171":1436,"_68":24,"_173":-5,"_506":1437,"_17":1438},{"_33":1423,"_13":1424,"_64":1425,"_43":-5,"_35":36,"_167":1426,"_168":94,"_37":1427,"_39":26,"_67":504,"_171":1428,"_68":24,"_173":-5,"_506":1429,"_17":1430},{"_33":1415,"_13":1416,"_64":1417,"_43":-5,"_35":36,"_167":1418,"_168":94,"_37":1419,"_39":26,"_67":504,"_171":1420,"_68":24,"_173":-5,"_506":1421,"_17":1422},{"_33":1407,"_13":1408,"_64":1409,"_43":-5,"_35":36,"_167":1410,"_168":94,"_37":1411,"_39":26,"_67":504,"_171":1412,"_68":24,"_173":-5,"_506":1413,"_17":1414},{"_33":1399,"_13":1400,"_64":1401,"_43":-5,"_35":36,"_167":1402,"_168":94,"_37":1403,"_39":26,"_67":504,"_171":1404,"_68":24,"_173":-5,"_506":1405,"_17":1406},{"_33":1391,"_13":1392,"_64":1393,"_43":-5,"_35":36,"_167":1394,"_168":94,"_37":1395,"_39":26,"_67":504,"_171":1396,"_68":24,"_173":-5,"_506":1397,"_17":1398},{"_33":1383,"_13":1384,"_64":1385,"_43":-5,"_35":36,"_167":1386,"_168":94,"_37":1387,"_39":26,"_67":504,"_171":1388,"_68":24,"_173":-5,"_506":1389,"_17":1390},{"_33":1375,"_13":1376,"_64":1377,"_43":-5,"_35":36,"_167":1378,"_168":94,"_37":1379,"_39":26,"_67":504,"_171":1380,"_68":24,"_173":-5,"_506":1381,"_17":1382},{"_33":1367,"_13":1368,"_64":1369,"_43":-5,"_35":36,"_167":1370,"_168":94,"_37":1371,"_39":170,"_67":504,"_171":1372,"_68":24,"_173":-5,"_506":1373,"_17":1374},{"_33":1359,"_13":1360,"_64":1361,"_43":-5,"_35":36,"_167":1362,"_168":94,"_37":1363,"_39":26,"_67":504,"_171":1364,"_68":24,"_173":-5,"_506":1365,"_17":1366},{"_33":1351,"_13":1352,"_64":1353,"_43":-5,"_35":36,"_167":1354,"_168":94,"_37":1355,"_39":1035,"_67":504,"_171":1356,"_68":24,"_173":-5,"_506":1357,"_17":1358},{"_33":1343,"_13":1344,"_64":1345,"_43":-5,"_35":36,"_167":1346,"_168":94,"_37":1347,"_39":944,"_67":504,"_171":1348,"_68":24,"_173":-5,"_506":1349,"_17":1350},{"_33":1335,"_13":1336,"_64":1337,"_43":-5,"_35":36,"_167":1338,"_168":94,"_37":1339,"_39":1035,"_67":504,"_171":1340,"_68":24,"_173":-5,"_506":1341,"_17":1342},{"_33":1327,"_13":1328,"_64":1329,"_43":-5,"_35":36,"_167":1330,"_168":94,"_37":1331,"_39":26,"_67":504,"_171":1332,"_68":24,"_173":-5,"_506":1333,"_17":1334},{"_33":1319,"_13":1320,"_64":1321,"_43":-5,"_35":36,"_167":1322,"_168":94,"_37":1323,"_39":26,"_67":504,"_171":1324,"_68":24,"_173":-5,"_506":1325,"_17":1326},{"_33":1311,"_13":1312,"_64":1313,"_43":-5,"_35":36,"_167":1314,"_168":94,"_37":1315,"_39":26,"_67":504,"_171":1316,"_68":24,"_173":-5,"_506":1317,"_17":1318},{"_33":1303,"_13":1304,"_64":1305,"_43":-5,"_35":36,"_167":1306,"_168":94,"_37":1307,"_39":1035,"_67":504,"_171":1308,"_68":24,"_173":-5,"_506":1309,"_17":1310},{"_33":1295,"_13":1296,"_64":1297,"_43":-5,"_35":36,"_167":1298,"_168":94,"_37":1299,"_39":935,"_67":504,"_171":1300,"_68":24,"_173":-5,"_506":1301,"_17":1302},{"_33":1287,"_13":1288,"_64":1289,"_43":-5,"_35":36,"_167":1290,"_168":94,"_37":1291,"_39":181,"_67":504,"_171":1292,"_68":24,"_173":-5,"_506":1293,"_17":1294},{"_33":1279,"_13":1280,"_64":1281,"_43":-5,"_35":36,"_167":1282,"_168":94,"_37":1283,"_39":26,"_67":504,"_171":1284,"_68":24,"_173":-5,"_506":1285,"_17":1286},{"_33":1271,"_13":1272,"_64":1273,"_43":-5,"_35":36,"_167":1274,"_168":94,"_37":1275,"_39":26,"_67":504,"_171":1276,"_68":24,"_173":-5,"_506":1277,"_17":1278},{"_33":1263,"_13":1264,"_64":1265,"_43":-5,"_35":36,"_167":1266,"_168":94,"_37":1267,"_39":26,"_67":504,"_171":1268,"_68":24,"_173":-5,"_506":1269,"_17":1270},{"_33":1255,"_13":1256,"_64":1257,"_43":-5,"_35":36,"_167":1258,"_168":94,"_37":1259,"_39":970,"_67":504,"_171":1260,"_68":24,"_173":-5,"_506":1261,"_17":1262},{"_33":1247,"_13":1248,"_64":1249,"_43":-5,"_35":36,"_167":1250,"_168":94,"_37":1251,"_39":970,"_67":504,"_171":1252,"_68":24,"_173":-5,"_506":1253,"_17":1254},{"_33":1239,"_13":1240,"_64":1241,"_43":-5,"_35":36,"_167":1242,"_168":94,"_37":1243,"_39":765,"_67":504,"_171":1244,"_68":24,"_173":-5,"_506":1245,"_17":1246},{"_33":1231,"_13":1232,"_64":1233,"_43":-5,"_35":36,"_167":1234,"_168":94,"_37":1235,"_39":765,"_67":504,"_171":1236,"_68":24,"_173":-5,"_506":1237,"_17":1238},{"_33":1223,"_13":1224,"_64":1225,"_43":-5,"_35":36,"_167":1226,"_168":94,"_37":1227,"_39":935,"_67":504,"_171":1228,"_68":24,"_173":-5,"_506":1229,"_17":1230},{"_33":1215,"_13":1216,"_64":1217,"_43":-5,"_35":36,"_167":1218,"_168":94,"_37":1219,"_39":26,"_67":504,"_171":1220,"_68":24,"_173":-5,"_506":1221,"_17":1222},{"_33":1207,"_13":1208,"_64":1209,"_43":-5,"_35":36,"_167":1210,"_168":94,"_37":1211,"_39":170,"_67":504,"_171":1212,"_68":24,"_173":-5,"_506":1213,"_17":1214},{"_33":1199,"_13":1200,"_64":1201,"_43":-5,"_35":36,"_167":1202,"_168":94,"_37":1203,"_39":26,"_67":504,"_171":1204,"_68":24,"_173":-5,"_506":1205,"_17":1206},{"_33":1191,"_13":1192,"_64":1193,"_43":-5,"_35":36,"_167":1194,"_168":94,"_37":1195,"_39":26,"_67":504,"_171":1196,"_68":24,"_173":-5,"_506":1197,"_17":1198},{"_33":1183,"_13":1184,"_64":1185,"_43":-5,"_35":36,"_167":1186,"_168":94,"_37":1187,"_39":26,"_67":504,"_171":1188,"_68":24,"_173":-5,"_506":1189,"_17":1190},{"_33":1175,"_13":1176,"_64":1177,"_43":-5,"_35":36,"_167":1178,"_168":94,"_37":1179,"_39":26,"_67":504,"_171":1180,"_68":24,"_173":-5,"_506":1181,"_17":1182},{"_33":1167,"_13":1168,"_64":1169,"_43":-5,"_35":36,"_167":1170,"_168":94,"_37":1171,"_39":26,"_67":504,"_171":1172,"_68":24,"_173":-5,"_506":1173,"_17":1174},{"_33":1159,"_13":1160,"_64":1161,"_43":-5,"_35":36,"_167":1162,"_168":94,"_37":1163,"_39":26,"_67":504,"_171":1164,"_68":24,"_173":-5,"_506":1165,"_17":1166},{"_33":1151,"_13":1152,"_64":1153,"_43":-5,"_35":36,"_167":1154,"_168":94,"_37":1155,"_39":26,"_67":504,"_171":1156,"_68":24,"_173":-5,"_506":1157,"_17":1158},{"_33":1143,"_13":1144,"_64":1145,"_43":-5,"_35":36,"_167":1146,"_168":94,"_37":1147,"_39":26,"_67":504,"_171":1148,"_68":24,"_173":-5,"_506":1149,"_17":1150},{"_33":1135,"_13":1136,"_64":1137,"_43":-5,"_35":36,"_167":1138,"_168":94,"_37":1139,"_39":26,"_67":504,"_171":1140,"_68":24,"_173":-5,"_506":1141,"_17":1142},{"_33":1127,"_13":1128,"_64":1129,"_43":-5,"_35":36,"_167":1130,"_168":94,"_37":1131,"_39":26,"_67":504,"_171":1132,"_68":24,"_173":-5,"_506":1133,"_17":1134},{"_33":1119,"_13":1120,"_64":1121,"_43":-5,"_35":36,"_167":1122,"_168":94,"_37":1123,"_39":181,"_67":504,"_171":1124,"_68":24,"_173":-5,"_506":1125,"_17":1126},{"_33":1111,"_13":1112,"_64":1113,"_43":-5,"_35":36,"_167":1114,"_168":94,"_37":1115,"_39":170,"_67":504,"_171":1116,"_68":24,"_173":-5,"_506":1117,"_17":1118},{"_33":1103,"_13":1104,"_64":1105,"_43":-5,"_35":36,"_167":1106,"_168":94,"_37":1107,"_39":26,"_67":504,"_171":1108,"_68":24,"_173":-5,"_506":1109,"_17":1110},{"_33":1095,"_13":1096,"_64":1097,"_43":-5,"_35":36,"_167":1098,"_168":94,"_37":1099,"_39":1035,"_67":504,"_171":1100,"_68":24,"_173":-5,"_506":1101,"_17":1102},{"_33":1087,"_13":1088,"_64":1089,"_43":-5,"_35":36,"_167":1090,"_168":94,"_37":1091,"_39":26,"_67":504,"_171":1092,"_68":24,"_173":-5,"_506":1093,"_17":1094},{"_33":1079,"_13":1080,"_64":1081,"_43":-5,"_35":36,"_167":1082,"_168":94,"_37":1083,"_39":170,"_67":504,"_171":1084,"_68":24,"_173":-5,"_506":1085,"_17":1086},{"_33":1071,"_13":1072,"_64":1073,"_43":-5,"_35":36,"_167":1074,"_168":94,"_37":1075,"_39":26,"_67":504,"_171":1076,"_68":24,"_173":-5,"_506":1077,"_17":1078},{"_33":1063,"_13":1064,"_64":1065,"_43":-5,"_35":36,"_167":1066,"_168":94,"_37":1067,"_39":26,"_67":504,"_171":1068,"_68":24,"_173":-5,"_506":1069,"_17":1070},{"_33":1055,"_13":1056,"_64":1057,"_43":-5,"_35":36,"_167":1058,"_168":94,"_37":1059,"_39":26,"_67":504,"_171":1060,"_68":24,"_173":-5,"_506":1061,"_17":1062},{"_33":1047,"_13":1048,"_64":1049,"_43":-5,"_35":36,"_167":1050,"_168":94,"_37":1051,"_39":170,"_67":504,"_171":1052,"_68":24,"_173":-5,"_506":1053,"_17":1054},{"_33":1039,"_13":1040,"_64":1041,"_43":-5,"_35":36,"_167":1042,"_168":94,"_37":1043,"_39":26,"_67":504,"_171":1044,"_68":24,"_173":-5,"_506":1045,"_17":1046},{"_33":1030,"_13":1031,"_64":1032,"_43":-5,"_35":36,"_167":1033,"_168":94,"_37":1034,"_39":1035,"_67":504,"_171":1036,"_68":24,"_173":-5,"_506":1037,"_17":1038},{"_33":1022,"_13":1023,"_64":1024,"_43":-5,"_35":36,"_167":1025,"_168":94,"_37":1026,"_39":181,"_67":504,"_171":1027,"_68":24,"_173":-5,"_506":1028,"_17":1029},{"_33":1014,"_13":1015,"_64":1016,"_43":-5,"_35":36,"_167":1017,"_168":94,"_37":1018,"_39":935,"_67":504,"_171":1019,"_68":24,"_173":-5,"_506":1020,"_17":1021},{"_33":1006,"_13":1007,"_64":1008,"_43":-5,"_35":36,"_167":1009,"_168":94,"_37":1010,"_39":765,"_67":504,"_171":1011,"_68":24,"_173":-5,"_506":1012,"_17":1013},{"_33":998,"_13":999,"_64":1000,"_43":-5,"_35":36,"_167":1001,"_168":94,"_37":1002,"_39":765,"_67":504,"_171":1003,"_68":24,"_173":-5,"_506":1004,"_17":1005},{"_33":990,"_13":991,"_64":992,"_43":-5,"_35":36,"_167":993,"_168":94,"_37":994,"_39":765,"_67":504,"_171":995,"_68":24,"_173":-5,"_506":996,"_17":997},{"_33":982,"_13":983,"_64":984,"_43":-5,"_35":36,"_167":985,"_168":94,"_37":986,"_39":935,"_67":504,"_171":987,"_68":24,"_173":-5,"_506":988,"_17":989},{"_33":974,"_13":975,"_64":976,"_43":-5,"_35":36,"_167":977,"_168":94,"_37":978,"_39":935,"_67":504,"_171":979,"_68":24,"_173":-5,"_506":980,"_17":981},{"_33":965,"_13":966,"_64":967,"_43":-5,"_35":36,"_167":968,"_168":94,"_37":969,"_39":970,"_67":504,"_171":971,"_68":24,"_173":-5,"_506":972,"_17":973},{"_33":956,"_13":957,"_64":958,"_43":-5,"_35":36,"_167":959,"_168":94,"_37":960,"_39":961,"_67":504,"_171":962,"_68":24,"_173":-5,"_506":963,"_17":964},{"_33":948,"_13":949,"_64":950,"_43":-5,"_35":36,"_167":951,"_168":94,"_37":952,"_39":935,"_67":504,"_171":953,"_68":24,"_173":-5,"_506":954,"_17":955},{"_33":939,"_13":940,"_64":941,"_43":-5,"_35":36,"_167":942,"_168":94,"_37":943,"_39":944,"_67":504,"_171":945,"_68":24,"_173":-5,"_506":946,"_17":947},{"_33":930,"_13":931,"_64":932,"_43":-5,"_35":36,"_167":933,"_168":94,"_37":934,"_39":935,"_67":504,"_171":936,"_68":24,"_173":-5,"_506":937,"_17":938},{"_33":922,"_13":923,"_64":924,"_43":-5,"_35":36,"_167":925,"_168":94,"_37":926,"_39":765,"_67":504,"_171":927,"_68":24,"_173":-5,"_506":928,"_17":929},{"_33":914,"_13":915,"_64":916,"_43":-5,"_35":36,"_167":917,"_168":94,"_37":918,"_39":181,"_67":504,"_171":919,"_68":24,"_173":-5,"_506":920,"_17":921},{"_33":905,"_13":906,"_64":907,"_43":-5,"_35":36,"_167":908,"_168":94,"_37":909,"_39":910,"_67":504,"_171":911,"_68":24,"_173":-5,"_506":912,"_17":913},{"_33":897,"_13":898,"_64":899,"_43":-5,"_35":36,"_167":900,"_168":94,"_37":901,"_39":170,"_67":504,"_171":902,"_68":24,"_173":-5,"_506":903,"_17":904},{"_33":889,"_13":890,"_64":891,"_43":-5,"_35":36,"_167":892,"_168":94,"_37":893,"_39":170,"_67":504,"_171":894,"_68":24,"_173":-5,"_506":895,"_17":896},{"_33":881,"_13":882,"_64":883,"_43":-5,"_35":36,"_167":884,"_168":94,"_37":885,"_39":170,"_67":504,"_171":886,"_68":24,"_173":-5,"_506":887,"_17":888},{"_33":873,"_13":874,"_64":875,"_43":-5,"_35":36,"_167":876,"_168":94,"_37":877,"_39":26,"_67":504,"_171":878,"_68":24,"_173":-5,"_506":879,"_17":880},{"_33":865,"_13":866,"_64":867,"_43":-5,"_35":36,"_167":868,"_168":94,"_37":869,"_39":26,"_67":504,"_171":870,"_68":24,"_173":-5,"_506":871,"_17":872},{"_33":857,"_13":858,"_64":859,"_43":-5,"_35":36,"_167":860,"_168":94,"_37":861,"_39":26,"_67":504,"_171":862,"_68":24,"_173":-5,"_506":863,"_17":864},{"_33":849,"_13":850,"_64":851,"_43":-5,"_35":36,"_167":852,"_168":94,"_37":853,"_39":26,"_67":504,"_171":854,"_68":24,"_173":-5,"_506":855,"_17":856},{"_33":841,"_13":842,"_64":843,"_43":-5,"_35":36,"_167":844,"_168":94,"_37":845,"_39":26,"_67":504,"_171":846,"_68":24,"_173":-5,"_506":847,"_17":848},{"_33":833,"_13":834,"_64":835,"_43":-5,"_35":36,"_167":836,"_168":94,"_37":837,"_39":26,"_67":504,"_171":838,"_68":24,"_173":-5,"_506":839,"_17":840},{"_33":825,"_13":826,"_64":827,"_43":-5,"_35":36,"_167":828,"_168":94,"_37":829,"_39":26,"_67":504,"_171":830,"_68":24,"_173":-5,"_506":831,"_17":832},{"_33":817,"_13":818,"_64":819,"_43":-5,"_35":36,"_167":820,"_168":94,"_37":821,"_39":26,"_67":504,"_171":822,"_68":24,"_173":-5,"_506":823,"_17":824},{"_33":809,"_13":810,"_64":811,"_43":-5,"_35":36,"_167":812,"_168":94,"_37":813,"_39":26,"_67":504,"_171":814,"_68":24,"_173":-5,"_506":815,"_17":816},{"_33":801,"_13":802,"_64":803,"_43":-5,"_35":36,"_167":804,"_168":94,"_37":805,"_39":26,"_67":504,"_171":806,"_68":24,"_173":-5,"_506":807,"_17":808},{"_33":793,"_13":794,"_64":795,"_43":-5,"_35":36,"_167":796,"_168":94,"_37":797,"_39":26,"_67":504,"_171":798,"_68":24,"_173":-5,"_506":799,"_17":800},{"_33":785,"_13":786,"_64":787,"_43":-5,"_35":36,"_167":788,"_168":94,"_37":789,"_39":26,"_67":504,"_171":790,"_68":24,"_173":-5,"_506":791,"_17":792},{"_33":777,"_13":778,"_64":779,"_43":-5,"_35":36,"_167":780,"_168":94,"_37":781,"_39":170,"_67":504,"_171":782,"_68":24,"_173":-5,"_506":783,"_17":784},{"_33":769,"_13":770,"_64":771,"_43":-5,"_35":36,"_167":772,"_168":94,"_37":773,"_39":26,"_67":504,"_171":774,"_68":24,"_173":-5,"_506":775,"_17":776},{"_33":760,"_13":761,"_64":762,"_43":-5,"_35":36,"_167":763,"_168":94,"_37":764,"_39":765,"_67":504,"_171":766,"_68":24,"_173":-5,"_506":767,"_17":768},{"_33":752,"_13":753,"_64":754,"_43":-5,"_35":36,"_167":755,"_168":94,"_37":756,"_39":26,"_67":504,"_171":757,"_68":24,"_173":-5,"_506":758,"_17":759},{"_33":744,"_13":745,"_64":746,"_43":-5,"_35":36,"_167":747,"_168":94,"_37":748,"_39":26,"_67":504,"_171":749,"_68":24,"_173":-5,"_506":750,"_17":751},{"_33":736,"_13":737,"_64":738,"_43":-5,"_35":36,"_167":739,"_168":94,"_37":740,"_39":170,"_67":504,"_171":741,"_68":24,"_173":-5,"_506":742,"_17":743},{"_33":728,"_13":729,"_64":730,"_43":-5,"_35":36,"_167":731,"_168":94,"_37":732,"_39":26,"_67":504,"_171":733,"_68":24,"_173":-5,"_506":734,"_17":735},{"_33":720,"_13":721,"_64":722,"_43":-5,"_35":36,"_167":723,"_168":94,"_37":724,"_39":26,"_67":504,"_171":725,"_68":24,"_173":-5,"_506":726,"_17":727},{"_33":712,"_13":713,"_64":714,"_43":-5,"_35":36,"_167":715,"_168":94,"_37":716,"_39":26,"_67":504,"_171":717,"_68":24,"_173":-5,"_506":718,"_17":719},{"_33":704,"_13":705,"_64":706,"_43":-5,"_35":36,"_167":707,"_168":94,"_37":708,"_39":26,"_67":504,"_171":709,"_68":24,"_173":-5,"_506":710,"_17":711},{"_33":696,"_13":697,"_64":698,"_43":-5,"_35":36,"_167":699,"_168":94,"_37":700,"_39":26,"_67":504,"_171":701,"_68":24,"_173":-5,"_506":702,"_17":703},{"_33":688,"_13":689,"_64":690,"_43":-5,"_35":36,"_167":691,"_168":94,"_37":692,"_39":170,"_67":504,"_171":693,"_68":24,"_173":-5,"_506":694,"_17":695},{"_33":680,"_13":681,"_64":682,"_43":-5,"_35":36,"_167":683,"_168":94,"_37":684,"_39":26,"_67":504,"_171":685,"_68":24,"_173":-5,"_506":686,"_17":687},{"_33":672,"_13":673,"_64":674,"_43":-5,"_35":36,"_167":675,"_168":94,"_37":676,"_39":26,"_67":504,"_171":677,"_68":24,"_173":-5,"_506":678,"_17":679},{"_33":664,"_13":665,"_64":666,"_43":-5,"_35":36,"_167":667,"_168":94,"_37":668,"_39":26,"_67":504,"_171":669,"_68":24,"_173":-5,"_506":670,"_17":671},{"_33":656,"_13":657,"_64":658,"_43":-5,"_35":36,"_167":659,"_168":94,"_37":660,"_39":26,"_67":504,"_171":661,"_68":24,"_173":-5,"_506":662,"_17":663},{"_33":648,"_13":649,"_64":650,"_43":-5,"_35":36,"_167":651,"_168":94,"_37":652,"_39":26,"_67":504,"_171":653,"_68":24,"_173":-5,"_506":654,"_17":655},{"_33":640,"_13":641,"_64":642,"_43":-5,"_35":36,"_167":643,"_168":94,"_37":644,"_39":26,"_67":504,"_171":645,"_68":24,"_173":-5,"_506":646,"_17":647},{"_33":632,"_13":633,"_64":634,"_43":-5,"_35":36,"_167":635,"_168":94,"_37":636,"_39":26,"_67":504,"_171":637,"_68":24,"_173":-5,"_506":638,"_17":639},{"_33":624,"_13":625,"_64":626,"_43":-5,"_35":36,"_167":627,"_168":94,"_37":628,"_39":26,"_67":504,"_171":629,"_68":24,"_173":-5,"_506":630,"_17":631},{"_33":616,"_13":617,"_64":618,"_43":-5,"_35":36,"_167":619,"_168":94,"_37":620,"_39":26,"_67":504,"_171":621,"_68":24,"_173":-5,"_506":622,"_17":623},{"_33":608,"_13":609,"_64":610,"_43":-5,"_35":36,"_167":611,"_168":94,"_37":612,"_39":26,"_67":504,"_171":613,"_68":24,"_173":-5,"_506":614,"_17":615},{"_33":600,"_13":601,"_64":602,"_43":-5,"_35":36,"_167":603,"_168":94,"_37":604,"_39":26,"_67":504,"_171":605,"_68":24,"_173":-5,"_506":606,"_17":607},{"_33":592,"_13":593,"_64":594,"_43":-5,"_35":36,"_167":595,"_168":94,"_37":596,"_39":26,"_67":504,"_171":597,"_68":24,"_173":-5,"_506":598,"_17":599},{"_33":584,"_13":585,"_64":586,"_43":-5,"_35":36,"_167":170,"_168":94,"_37":587,"_39":26,"_67":588,"_171":589,"_68":24,"_173":-5,"_506":590,"_17":591},{"_33":576,"_13":577,"_64":578,"_43":-5,"_35":36,"_167":579,"_168":94,"_37":580,"_39":26,"_67":504,"_171":581,"_68":24,"_173":-5,"_506":582,"_17":583},{"_33":568,"_13":569,"_64":570,"_43":-5,"_35":36,"_167":571,"_168":94,"_37":572,"_39":170,"_67":504,"_171":573,"_68":24,"_173":-5,"_506":574,"_17":575},{"_33":560,"_13":561,"_64":562,"_43":-5,"_35":36,"_167":563,"_168":94,"_37":564,"_39":26,"_67":504,"_171":565,"_68":24,"_173":-5,"_506":566,"_17":567},{"_33":553,"_13":554,"_64":555,"_43":-5,"_35":36,"_167":170,"_168":94,"_37":556,"_39":26,"_67":504,"_171":557,"_68":24,"_173":-5,"_506":558,"_17":559},{"_33":545,"_13":546,"_64":547,"_43":-5,"_35":36,"_167":548,"_168":94,"_37":549,"_39":26,"_67":504,"_171":550,"_68":24,"_173":-5,"_506":551,"_17":552},{"_33":538,"_13":539,"_64":540,"_43":-5,"_35":36,"_167":170,"_168":94,"_37":541,"_39":26,"_67":504,"_171":542,"_68":24,"_173":-5,"_506":543,"_17":544},{"_33":531,"_13":532,"_64":533,"_43":-5,"_35":36,"_167":170,"_168":94,"_37":534,"_39":26,"_67":504,"_171":535,"_68":24,"_173":-5,"_506":536,"_17":537},{"_33":524,"_13":525,"_64":526,"_43":-5,"_35":36,"_167":170,"_168":94,"_37":527,"_39":26,"_67":504,"_171":528,"_68":24,"_173":-5,"_506":529,"_17":530},{"_33":517,"_13":518,"_64":519,"_43":-5,"_35":36,"_167":170,"_168":94,"_37":520,"_39":26,"_67":504,"_171":521,"_68":24,"_173":-5,"_506":522,"_17":523},{"_33":509,"_13":510,"_64":511,"_43":-5,"_35":36,"_167":512,"_168":94,"_37":513,"_39":26,"_67":504,"_171":514,"_68":24,"_173":-5,"_506":515,"_17":516},{"_33":500,"_13":501,"_64":502,"_43":-5,"_35":36,"_167":170,"_168":94,"_37":503,"_39":26,"_67":504,"_171":505,"_68":24,"_173":-5,"_506":507,"_17":508},"64265461-4810-40ad-beab-6f2cdc38131f","url_selftest.ts","01a077bc-2b99-740d-9881-6b7404857766",["D",1788715215771],"script","import { assertSafeUrl, preflightTaskNetwork } from \"./lib/url_security.ts\";\n\nconst cases = [\n \"https://example.com\",\n \"http://127.0.0.1\",\n \"http://169.254.169.254/latest/meta-data\",\n \"http://localhost\",\n \"https://user:pass@example.com\",\n \"https://example.com:8443\"\n];\n\nconst results: Record = {};\nfor (const value of cases) {\n try {\n results[value] = { ok: true, result: await assertSafeUrl(value) };\n } catch (error) {\n results[value] = { ok: false, error: error instanceof Error ? error.message : String(error) };\n }\n}\n\nlet allowedHostBlocked = false;\ntry {\n await preflightTaskNetwork({\n runId: \"url-selftest\",\n targetUrl: \"https://example.com\",\n risk: \"read_only\",\n purpose: \"URL security selftest\",\n allowedHosts: [\"127.0.0.1\"],\n steps: [{ action: \"expect_url\", contains: \"example.com\" }],\n });\n} catch {\n allowedHostBlocked = true;\n}\n\nconsole.log(JSON.stringify({ results, allowedHostBlocked }, null, 2));\n","children",[],[501],"f22c482d-b1ea-423d-a58b-324908fda53d","strategy_variant_migrate.ts","5a26e050-aa1a-11f1-9414-1607ee4eb77e",13,["D",1788716558331],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\n\nconst now = new Date().toISOString();\nawait sqlite.batch([\n { sql: `CREATE TABLE IF NOT EXISTS strategy_variants (\n variant_id TEXT PRIMARY KEY,\n agent_id TEXT NOT NULL,\n parent_variant_id TEXT,\n generation INTEGER NOT NULL DEFAULT 0,\n name TEXT NOT NULL,\n hypothesis TEXT NOT NULL,\n mutation TEXT NOT NULL,\n status TEXT NOT NULL,\n evidence_state TEXT NOT NULL DEFAULT 'UNVERIFIED',\n benchmark_metric TEXT,\n benchmark_value REAL,\n artifact_pointer TEXT,\n notes TEXT,\n created_at TEXT NOT NULL,\n updated_at TEXT NOT NULL,\n FOREIGN KEY(agent_id) REFERENCES economic_agents(agent_id),\n FOREIGN KEY(parent_variant_id) REFERENCES strategy_variants(variant_id)\n )` },\n { sql: `CREATE TABLE IF NOT EXISTS variant_evaluations (\n evaluation_id TEXT PRIMARY KEY,\n variant_id TEXT NOT NULL,\n opponent_variant_id TEXT,\n environment TEXT NOT NULL,\n seed_set TEXT,\n seat_mode TEXT,\n episodes INTEGER NOT NULL DEFAULT 0,\n wins INTEGER NOT NULL DEFAULT 0,\n losses INTEGER NOT NULL DEFAULT 0,\n ties INTEGER NOT NULL DEFAULT 0,\n mean_terminal_bank REAL,\n median_terminal_bank REAL,\n movement_rate REAL,\n pass_rate REAL,\n runtime_errors INTEGER NOT NULL DEFAULT 0,\n evidence_state TEXT NOT NULL,\n evidence_json TEXT,\n created_at TEXT NOT NULL,\n FOREIGN KEY(variant_id) REFERENCES strategy_variants(variant_id)\n )` },\n { sql: `CREATE INDEX IF NOT EXISTS idx_strategy_variants_agent ON strategy_variants(agent_id,generation,status)` },\n { sql: `CREATE INDEX IF NOT EXISTS idx_variant_evals_variant ON variant_evaluations(variant_id,created_at)` },\n { sql: `INSERT INTO schema_meta(key,value,updated_at) VALUES('strategy_variant_schema_version','0.6.1-candidate',?)\n ON CONFLICT(key) DO UPDATE SET value=excluded.value,updated_at=excluded.updated_at`, args:[now] },\n]);\n\nconst variants = [\n ['K0','A4-KAGGRICULTURE',null,0,'Deterministic Control','A compact deterministic executor provides a stable regression oracle and baseline','baseline_control','DESIGNED','UNVERIFIED','artifacts/kaggriculture_neuron_protocol_v1.md'],\n ['K1','A4-KAGGRICULTURE','K0',1,'Labor/Route Optimizer','Reducing walking/thrashing produces larger gains than more complex crop valuation','route_executor','DESIGNED','UNVERIFIED','artifacts/kaggriculture_neuron_protocol_v1.md'],\n ['K2','A4-KAGGRICULTURE','K1',2,'Livestock/Fertilizer Flywheel','Jointly valuing animal product plus fertilizer stream improves asset ROI','livestock_fertilizer','DESIGNED','UNVERIFIED','artifacts/kaggriculture_neuron_protocol_v1.md'],\n ['K3','A4-KAGGRICULTURE','K1',2,'Opponent-Aware Anti-Glut','Visible opponent production can predict shared-market glut early enough to alter profitable production/sale timing','opponent_forecast','DESIGNED','UNVERIFIED','artifacts/kaggriculture_neuron_protocol_v1.md'],\n ['K4','A4-KAGGRICULTURE','K1',2,'Demand/Horizon Portfolio','Explicit remaining-horizon NPV and observed town demand improve allocation and liquidation','horizon_npv','DESIGNED','UNVERIFIED','artifacts/kaggriculture_neuron_protocol_v1.md'],\n ['K5','A4-KAGGRICULTURE','K1',3,'Hybrid Champion Candidate','Only mechanisms that independently survive ablation should be fused into a challenger','evidence_gated_fusion','BLOCKED_BY_EVIDENCE','UNVERIFIED','artifacts/kaggriculture_neuron_protocol_v1.md'],\n] as const;\n\nfor (const [id,agent,parent,generation,name,hypothesis,mutation,status,evidence,artifact] of variants) {\n await sqlite.execute({\n sql:`INSERT INTO strategy_variants(variant_id,agent_id,parent_variant_id,generation,name,hypothesis,mutation,status,evidence_state,artifact_pointer,created_at,updated_at)\n VALUES(?,?,?,?,?,?,?,?,?,?,?,?)\n ON CONFLICT(variant_id) DO UPDATE SET name=excluded.name,hypothesis=excluded.hypothesis,mutation=excluded.mutation,status=excluded.status,evidence_state=excluded.evidence_state,artifact_pointer=excluded.artifact_pointer,updated_at=excluded.updated_at`,\n args:[id,agent,parent,generation,name,hypothesis,mutation,status,evidence,artifact,now,now]\n });\n}\n\nconst rows = await sqlite.execute(`SELECT variant_id,parent_variant_id,generation,name,status,evidence_state FROM strategy_variants WHERE agent_id='A4-KAGGRICULTURE' ORDER BY generation,variant_id`);\nconsole.log(JSON.stringify({status:'PASS',version:'0.6.1-candidate',variants:rows.rows},null,2));\n",[],[510],"d17e92c8-0409-47b8-9ab8-94a9e01ee097","status.ts","01a077bc-2b99-740d-9881-6727e0b68381",["D",1788715215771],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { providerState } from \"./providers/router.ts\";\nimport { getProviderHealth } from \"./lib/provider_health.ts\";\nimport { POLICY } from \"./lib/policy.ts\";\nimport { COST_NOTES } from \"./lib/cost.ts\";\n\nconst schema = await sqlite.execute(\"SELECT key,value,updated_at FROM schema_meta ORDER BY key\");\nconst counts = await sqlite.execute(\"SELECT status, COUNT(*) AS count FROM browser_tasks GROUP BY status ORDER BY status\");\nconst census = await sqlite.execute(\"SELECT provider, priority, state, recurring_free, notes, updated_at FROM provider_census ORDER BY priority\");\nconst recent = await sqlite.execute(\"SELECT run_id,status,provider,attempt_count,error,updated_at FROM browser_tasks ORDER BY id DESC LIMIT 10\");\nconst attemptMetrics = await sqlite.execute(`\n SELECT provider,status,COUNT(*) AS attempts,\n CAST(AVG(duration_ms) AS INTEGER) AS avg_duration_ms,\n COALESCE(SUM(estimated_units),0) AS estimated_units\n FROM browser_attempts\n GROUP BY provider,status\n ORDER BY provider,status\n`);\n\nconst result = {\n service: \"chat-native-maestro\",\n version: POLICY.version,\n status: \"ready\",\n authenticatedDiagnostic: true,\n schema: schema.rows,\n providers: providerState(),\n providerHealth: await getProviderHealth(),\n taskCounts: counts.rows,\n providerCensus: census.rows,\n attemptMetrics: attemptMetrics.rows,\n recentReceipts: recent.rows,\n policy: POLICY,\n costNotes: COST_NOTES,\n environmentRequirements: {\n kernel: \"KERNEL_API_KEY\",\n browserless: \"BROWSERLESS_TOKEN\",\n },\n networkPreflight: {\n dnsOverHttps: \"cloudflare-dns.com\",\n failClosed: true,\n blocksPrivateAndReservedIpRanges: true,\n blocksUrlUserinfo: true,\n blocksNonstandardPorts: true,\n },\n automaticSchedule: \"disabled\",\n timestamp: new Date().toISOString(),\n};\n\nconsole.log(JSON.stringify(result, null, 2));\n",[],[518],"abd98a94-ae35-4165-976e-a3c5acbfba2b","stateful_selftest.ts","01a077bc-2b99-740d-9881-604153bba708",["D",1788715215771],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { decideDisposition, withTimeout } from \"./lib/policy.ts\";\nimport { providerAvailable, recordProviderOutcome } from \"./lib/provider_health.ts\";\nimport type { BrowserReceipt } from \"./lib/types.ts\";\n\nconst now = new Date().toISOString();\nconst syntheticFail = (n: number): BrowserReceipt => ({\n runId: `circuit-selftest-${n}`,\n provider: \"kernel\",\n status: \"FAIL\",\n error: `synthetic failure ${n}`,\n durationMs: 1,\n});\n\nlet timeoutBlocked = false;\ntry {\n await withTimeout(new Promise((resolve) => setTimeout(resolve, 50)), 5, \"synthetic timeout\");\n} catch (error) {\n timeoutBlocked = String(error).includes(\"timed out\");\n}\n\nconst dispositions = {\n readOnlyFirstFail: decideDisposition(\"read_only\", \"FAIL\", 1),\n readOnlyBudgetExhausted: decideDisposition(\"read_only\", \"FAIL\", 2),\n writeFail: decideDisposition(\"consequential\", \"FAIL\", 1),\n};\n\nlet circuitAfterThree: unknown;\nlet availableAfterThree: unknown;\ntry {\n await sqlite.execute({\n sql: `UPDATE provider_health SET circuit_state='CLOSED', consecutive_failures=0, open_until=NULL,\n last_success_at=NULL,last_failure_at=NULL,last_error=NULL,updated_at=? WHERE provider='kernel'`,\n args: [now],\n });\n await recordProviderOutcome(\"kernel\", syntheticFail(1));\n await recordProviderOutcome(\"kernel\", syntheticFail(2));\n await recordProviderOutcome(\"kernel\", syntheticFail(3));\n circuitAfterThree = (await sqlite.execute(\"SELECT provider,circuit_state,consecutive_failures,open_until FROM provider_health WHERE provider='kernel'\")).rows[0];\n availableAfterThree = await providerAvailable(\"kernel\");\n} finally {\n await sqlite.execute({\n sql: `UPDATE provider_health SET circuit_state='CLOSED', consecutive_failures=0, open_until=NULL,\n last_success_at=NULL,last_failure_at=NULL,last_error=NULL,updated_at=? WHERE provider='kernel'`,\n args: [new Date().toISOString()],\n });\n}\n\nconsole.log(JSON.stringify({\n version: \"0.5.0\",\n timeoutBlocked,\n dispositions,\n circuitAfterThree,\n availableAfterThree,\n restoredKernelHealth: (await sqlite.execute(\"SELECT provider,circuit_state,consecutive_failures,open_until FROM provider_health WHERE provider='kernel'\")).rows[0],\n}, null, 2));\n",[],[525],"73e9e358-f044-43f3-8b0c-ee8e3f20e4f8","snapshot_export.ts","01a077bc-2b99-740d-9881-5e611370bab8",["D",1788715215771],"import JSZip from \"npm:jszip\";\nimport { blob } from \"https://esm.town/v/std/blob/main.ts\";\n\nconst release = \"65-main\";\nconst paths = [\n \"main.ts\",\n \"status.ts\",\n \"runner.ts\",\n \"selftest.ts\",\n \"stateful_selftest.ts\",\n \"url_selftest.ts\",\n \"README.md\",\n \"lib/types.ts\",\n \"lib/dsl.ts\",\n \"lib/policy.ts\",\n \"lib/provider_health.ts\",\n \"lib/cost.ts\",\n \"lib/url_security.ts\",\n \"providers/router.ts\",\n \"providers/kernel.ts\",\n \"providers/browserless.ts\",\n \"docs/CURRENT_STATE.md\",\n \"docs/PROVIDER_CENSUS.md\",\n \"docs/ACTIVATION_RUNBOOK.md\",\n \"docs/TASK_SCHEMA.md\",\n \"docs/SMOKE_TASK.json\",\n \"docs/SECURITY_MODEL.md\",\n];\n\nconst zip = new JSZip();\nconst manifest: Array<{ path: string; bytes: number; sha256: string; sourceUrl: string }> = [];\nconst enc = new TextEncoder();\nconst hex = (bytes: ArrayBuffer) => [...new Uint8Array(bytes)].map((x) => x.toString(16).padStart(2, \"0\")).join(\"\");\n\nfor (const path of paths) {\n const sourceUrl = `https://esm.town/v/cleitoncosta/chat-native-maestro@${release}/${path}`;\n const response = await fetch(sourceUrl, { signal: AbortSignal.timeout(10_000) });\n if (!response.ok) throw new Error(`${path}: HTTP ${response.status}`);\n let text = await response.text();\n if (path.endsWith(\".ts\") || path.endsWith(\".tsx\")) {\n text = text.replace(/^\\/\\* Open in Val Town: [^\\n]+ \\*\\/\\n/, \"\");\n }\n const bytes = enc.encode(text);\n const sha256 = hex(await crypto.subtle.digest(\"SHA-256\", bytes));\n manifest.push({ path, bytes: bytes.byteLength, sha256, sourceUrl });\n zip.file(path, text);\n}\n\nconst metadata = {\n name: \"CHATGPT_CHAT_NATIVE_MAESTRO_V0_5_0_CLOUD_SNAPSHOT\",\n sourceVal: \"cleitoncosta/chat-native-maestro\",\n sourceBranchRelease: release,\n appVersion: \"0.5.0\",\n exportedAt: new Date().toISOString(),\n fileCount: paths.length,\n credentialsIncluded: false,\n databaseContentsIncluded: false,\n note: \"Source files fetched from immutable esm.town release URLs; esm.town TypeScript display header removed before packaging.\",\n};\nzip.file(\"LIVE_EXPORT_METADATA.json\", JSON.stringify(metadata, null, 2));\nzip.file(\"SOURCE_MANIFEST.json\", JSON.stringify(manifest, null, 2));\nconst base64 = await zip.generateAsync({ type: \"base64\", compression: \"DEFLATE\", compressionOptions: { level: 9 } });\nconst blobKey = \"release:v0.5.0:source.zip.b64\";\nawait blob.set(blobKey, base64);\nconst base64Sha256 = hex(await crypto.subtle.digest(\"SHA-256\", enc.encode(base64)));\nconsole.log(JSON.stringify({ metadata, manifest, blobKey, base64Length: base64.length, base64Sha256 }));\n",[],[532],"53ccefb1-9a83-4be5-a76f-668e29910d84","selftest.ts","01a077bc-2b99-740d-9881-5a19ff4d14c8",["D",1788715215771],"import { validateTask } from \"./lib/dsl.ts\";\nimport { providerState } from \"./providers/router.ts\";\nimport { POLICY } from \"./lib/policy.ts\";\nimport { getProviderHealth } from \"./lib/provider_health.ts\";\n\nconst validRead = {\n runId: \"selftest-read-v05\",\n targetUrl: \"https://example.com\",\n risk: \"read_only\" as const,\n purpose: \"Validate V0.5 DSL without opening a browser\",\n evidenceRetention: \"minimal\" as const,\n steps: [{ action: \"expect_text\" as const, text: \"Example Domain\" }],\n};\n\nconst blockedWrite = {\n runId: \"selftest-write-v05\",\n targetUrl: \"https://example.com\",\n risk: \"consequential\" as const,\n approved: false,\n purpose: \"Negative control for approval gate\",\n steps: [{ action: \"click\" as const, selector: \"button\" }],\n};\n\nconst blockedSecret = {\n runId: \"selftest-secret-v05\",\n targetUrl: \"https://example.com\",\n risk: \"read_only\" as const,\n approved: false,\n purpose: \"Negative control for secret resolution gate\",\n steps: [{ action: \"fill_secret\" as const, selector: \"#password\", envVar: \"EXAMPLE_PASSWORD\" }],\n};\n\nconst invalidProvider = {\n ...validRead,\n runId: \"selftest-provider-v05\",\n preferredProvider: \"imaginary-provider\" as any,\n};\n\nconst blockedValueExtraction = {\n ...validRead,\n runId: \"selftest-value-v05\",\n steps: [{ action: \"extract_attribute\" as const, selector: \"#password\", attribute: \"value\", key: \"password\" }],\n};\n\nconst result = {\n version: POLICY.version,\n validReadErrors: validateTask(validRead),\n blockedWriteErrors: validateTask(blockedWrite),\n blockedSecretErrors: validateTask(blockedSecret),\n invalidProviderErrors: validateTask(invalidProvider),\n blockedValueExtractionErrors: validateTask(blockedValueExtraction),\n providers: providerState(),\n providerHealth: await getProviderHealth(),\n policy: POLICY,\n expected: {\n validReadErrors: [],\n blockedWriteContains: \"NEEDS_APPROVAL\",\n blockedSecretContains: \"NEEDS_SECRET_APPROVAL\",\n invalidProviderContains: \"invalid preferredProvider\",\n valueExtractionBlocked: true,\n providersConfigured: false,\n },\n};\n\nconsole.log(JSON.stringify(result, null, 2));\n",[],[539],"4c29f586-3db0-4d31-84e4-90a8a3bc8396","selection_engine.ts","ef689386-aa18-11f1-8b71-1607ee4eb77e",11,["D",1788715949727],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { rankGeneration, offspringPlan, type SelectionCandidate, type SelectionResult } from \"./lib/neuron_evolution.ts\";\n\nconst now = new Date().toISOString();\n\nconst agents = await sqlite.execute(`\n SELECT a.agent_id,a.generation,\n COALESCE(SUM(t.gross_revenue),0) AS gross,\n COALESCE(SUM(t.browser_cost+t.llm_cost+t.api_cost+t.hosting_cost+t.platform_fee+t.transaction_cost),0) AS cost,\n COALESCE(SUM(t.net_revenue),0) AS net,\n COALESCE(SUM(CASE WHEN t.operationally_eligible=1 AND t.terminal=1 THEN 1 ELSE 0 END),0) AS terminal_trials,\n COALESCE(SUM(CASE WHEN t.operationally_eligible=1 AND t.terminal=1 AND t.net_revenue>0 THEN 1 ELSE 0 END),0) AS successful_trials,\n COALESCE(SUM(CASE WHEN t.operationally_eligible=1 AND t.terminal=1 AND t.net_revenue<=0 THEN 1 ELSE 0 END),0) AS failed_trials,\n COALESCE(SUM(CASE WHEN t.operationally_eligible=1 AND t.terminal=0 THEN 1 ELSE 0 END),0) AS active_eligible_trials\n FROM economic_agents a\n LEFT JOIN economic_trials t ON t.agent_id=a.agent_id\n GROUP BY a.agent_id,a.generation\n ORDER BY a.generation,a.agent_id\n`);\n\ntype Row = Record;\ntype Derived = { row: Row; candidate: SelectionCandidate };\nconst derived: Derived[] = agents.rows.map((row: Row) => {\n const terminalTrials = Number(row.terminal_trials ?? 0);\n const activeEligibleTrials = Number(row.active_eligible_trials ?? 0);\n return {\n row,\n candidate: {\n agentId: String(row.agent_id),\n netRevenue: Number(row.net ?? 0),\n operationallyEligible: terminalTrials > 0,\n terminal: terminalTrials > 0 && activeEligibleTrials === 0,\n },\n };\n});\n\nconst results = new Map();\nconst generations = [...new Set(derived.map((x) => Number(x.row.generation)))].sort((a,b)=>a-b);\nfor (const generation of generations) {\n const group = derived.filter((x) => Number(x.row.generation) === generation);\n for (const result of rankGeneration(group.map((x) => x.candidate))) results.set(result.agentId, result);\n}\n\nconst statements: Array<{sql:string;args:Array}> = [];\nconst plans: Array> = [];\nfor (const item of derived) {\n const result = results.get(item.candidate.agentId)!;\n const gross = Number(item.row.gross ?? 0);\n const cost = Number(item.row.cost ?? 0);\n const terminalTrials = Number(item.row.terminal_trials ?? 0);\n const successfulTrials = Number(item.row.successful_trials ?? 0);\n const failedTrials = Number(item.row.failed_trials ?? 0);\n statements.push({\n sql: `INSERT INTO agent_fitness(agent_id,selection_state,lifetime_gross,lifetime_cost,lifetime_net,benchmark_net,terminal_trials,successful_trials,failed_trials,rank,offspring_count,updated_at)\n VALUES(?,?,?,?,?,?,?,?,?,?,?,?)\n ON CONFLICT(agent_id) DO UPDATE SET\n selection_state=excluded.selection_state,\n lifetime_gross=excluded.lifetime_gross,\n lifetime_cost=excluded.lifetime_cost,\n lifetime_net=excluded.lifetime_net,\n benchmark_net=CASE WHEN excluded.lifetime_net>agent_fitness.benchmark_net THEN excluded.lifetime_net ELSE agent_fitness.benchmark_net END,\n terminal_trials=excluded.terminal_trials,\n successful_trials=excluded.successful_trials,\n failed_trials=excluded.failed_trials,\n rank=excluded.rank,\n offspring_count=excluded.offspring_count,\n updated_at=excluded.updated_at`,\n args: [result.agentId,result.state,gross,cost,result.netRevenue,Math.max(0,result.benchmark),terminalTrials,successfulTrials,failedTrials,result.rank ?? null,result.offspringCount,now],\n });\n if (result.state === \"DEAD\" || result.state === \"SURVIVOR\") {\n statements.push({\n sql: `UPDATE economic_agents SET status=?,updated_at=? WHERE agent_id=?`,\n args: [result.state,now,result.agentId],\n });\n }\n for (const plan of offspringPlan(result)) plans.push({ generation: Number(item.row.generation), ...plan });\n}\n\nif (statements.length) await sqlite.batch(statements);\nconst fitness = await sqlite.execute(`SELECT agent_id,selection_state,lifetime_gross,lifetime_cost,lifetime_net,benchmark_net,terminal_trials,successful_trials,failed_trials,rank,offspring_count FROM agent_fitness ORDER BY agent_id`);\n\nconsole.log(JSON.stringify({ status: \"PASS\", evaluated: derived.length, fitness: fitness.rows, offspringPlans: plans, timestamp: now }, null, 2));\n",[],[546],"78c1b1fd-24a3-47f8-83ee-334894178982","runner.ts","01a077bc-2b99-740d-9881-57dc4656df82",["D",1788715215771],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport type { BrowserReceipt, BrowserTask } from \"./lib/types.ts\";\nimport { validateTask } from \"./lib/dsl.ts\";\nimport { executeRouted } from \"./providers/router.ts\";\nimport { decideDisposition, POLICY } from \"./lib/policy.ts\";\nimport { estimateProviderUnits } from \"./lib/cost.ts\";\nimport { preflightTaskNetwork } from \"./lib/url_security.ts\";\n\nasync function recoverStaleLeases() {\n const staleBefore = new Date(Date.now() - POLICY.runningLeaseMs).toISOString();\n const now = new Date().toISOString();\n await sqlite.batch([\n {\n sql: `UPDATE browser_tasks SET status='FAIL', claimed_at=NULL, next_attempt_at=NULL,\n error='stale running lease; retry budget exhausted', updated_at=?\n WHERE status='RUNNING' AND claimed_at IS NOT NULL AND claimed_at=?`,\n args: [now, staleBefore, POLICY.maxAttemptsPerTask],\n },\n {\n sql: `UPDATE browser_tasks SET status='PENDING', claimed_at=NULL, next_attempt_at=?,\n error='stale running lease recovered', updated_at=?\n WHERE status='RUNNING' AND claimed_at IS NOT NULL AND claimed_at | undefined;\n}\n\nfunction storageReceipt(receipt: BrowserReceipt, task: BrowserTask) {\n if (task.evidenceRetention === \"full\") return receipt;\n const meta = receipt.providerMeta ?? {};\n return {\n runId: receipt.runId,\n provider: receipt.provider,\n status: receipt.status,\n finalUrl: receipt.finalUrl,\n checks: receipt.checks ?? [],\n error: receipt.error,\n durationMs: receipt.durationMs,\n providerMeta: {\n fallbackFrom: meta.fallbackFrom,\n priorStatus: meta.priorStatus,\n priorError: meta.priorError,\n priorDurationMs: meta.priorDurationMs,\n networkPreflight: meta.networkPreflight,\n },\n outputsPersisted: false,\n };\n}\n\nasync function recordAttempts(taskId: number, attemptNo: number, startedAt: string, receipt: BrowserReceipt, storedReceipt: unknown) {\n const finishedAt = new Date().toISOString();\n const meta = receipt.providerMeta ?? {};\n const statements: Array<{ sql: string; args: Array }> = [];\n\n if (typeof meta.fallbackFrom === \"string\") {\n statements.push({\n sql: `INSERT INTO browser_attempts(task_id,run_id,attempt_no,provider,started_at,finished_at,status,duration_ms,error,receipt_json,estimated_units)\n VALUES(?,?,?,?,?,?,?,?,?,?,?)`,\n args: [taskId, receipt.runId, attemptNo, meta.fallbackFrom, startedAt, finishedAt,\n String(meta.priorStatus ?? \"FAIL\"), Number(meta.priorDurationMs ?? 0),\n meta.priorError == null ? null : String(meta.priorError), null, null],\n });\n }\n\n statements.push({\n sql: `INSERT INTO browser_attempts(task_id,run_id,attempt_no,provider,started_at,finished_at,status,duration_ms,error,receipt_json,estimated_units)\n VALUES(?,?,?,?,?,?,?,?,?,?,?)`,\n args: [taskId, receipt.runId, attemptNo, receipt.provider, startedAt, finishedAt,\n receipt.status, receipt.durationMs ?? null, receipt.error ?? null, JSON.stringify(storedReceipt), estimateProviderUnits(receipt)],\n });\n await sqlite.batch(statements);\n}\n\nasync function finishTask(row: Record, task: BrowserTask, receipt: BrowserReceipt, storedReceipt: unknown) {\n const id = Number(row.id);\n const attemptCount = Number(row.attempt_count);\n const now = new Date().toISOString();\n const disposition = decideDisposition(task.risk, receipt.status, attemptCount);\n const next = disposition.retryable ? new Date(Date.now() + disposition.retryDelayMs).toISOString() : null;\n await sqlite.execute({\n sql: `UPDATE browser_tasks SET status=?, provider=?, receipt_json=?, error=?, updated_at=?, claimed_at=NULL, next_attempt_at=? WHERE id=?`,\n args: [disposition.taskStatus, receipt.provider, JSON.stringify(storedReceipt), receipt.error ?? null, now, next, id],\n });\n}\n\nasync function blockTask(row: Record, task: BrowserTask | undefined, errors: string[]) {\n const id = Number(row.id);\n const receipt = { runId: task?.runId ?? String(row.run_id), status: \"BLOCKED\", errors };\n const now = new Date().toISOString();\n await sqlite.batch([\n {\n sql: `UPDATE browser_tasks SET status='BLOCKED', provider=?, receipt_json=?, error=?, updated_at=?, claimed_at=NULL, next_attempt_at=NULL WHERE id=?`,\n args: [task?.preferredProvider ?? \"validation\", JSON.stringify(receipt), errors.join(\"; \"), now, id],\n },\n {\n sql: `INSERT INTO browser_attempts(task_id,run_id,attempt_no,provider,started_at,finished_at,status,error,receipt_json,estimated_units)\n VALUES(?,?,?,?,?,?,?,?,?,?)`,\n args: [id, String(row.run_id), Number(row.attempt_count), task?.preferredProvider ?? \"validation\",\n String(row.claimed_at ?? now), now, \"BLOCKED\", errors.join(\"; \"), JSON.stringify(receipt), null],\n },\n ]);\n console.log(JSON.stringify(receipt, null, 2));\n return receipt;\n}\n\nasync function main() {\n await recoverStaleLeases();\n const row = await claimNext();\n if (!row) {\n const result = { status: \"NO_PENDING_TASK\" };\n console.log(JSON.stringify(result));\n return result;\n }\n\n let task: BrowserTask;\n try {\n task = JSON.parse(String(row.task_json)) as BrowserTask;\n } catch (error) {\n return blockTask(row, undefined, [`invalid task_json: ${error instanceof Error ? error.message : String(error)}`]);\n }\n\n const errors = validateTask(task);\n if (errors.length) return blockTask(row, task, errors);\n\n let networkPreflight: Array<{ host: string; ips: string[] }>;\n try {\n networkPreflight = await preflightTaskNetwork(task);\n } catch (error) {\n return blockTask(row, task, [`NETWORK_PREFLIGHT_BLOCKED: ${error instanceof Error ? error.message : String(error)}`]);\n }\n\n const receipt = await executeRouted(task);\n receipt.providerMeta = { ...(receipt.providerMeta ?? {}), networkPreflight };\n const storedReceipt = storageReceipt(receipt, task);\n await recordAttempts(Number(row.id), Number(row.attempt_count), String(row.claimed_at), receipt, storedReceipt);\n await finishTask(row, task, receipt, storedReceipt);\n\n console.log(JSON.stringify(receipt, null, 2));\n return receipt;\n}\n\nawait main();\n",[],[554],"783bd826-874b-4a95-bcdc-c472e8ab076c","python_runtime_probe.ts","d0f3b032-aa38-11f1-bbfa-1607ee4eb77e",32,["D",1788729642559],"const cmds=['python3','python'];const out=[];for(const c of cmds){try{const p=new Deno.Command(c,{args:['--version'],stdout:'piped',stderr:'piped'});const r=await p.output();out.push({cmd:c,code:r.code,stdout:new TextDecoder().decode(r.stdout),stderr:new TextDecoder().decode(r.stderr)});}catch(e){out.push({cmd:c,error:String(e)});}}console.log(JSON.stringify({status:out.some((x:any)=>x.code===0)?'PASS':'BLOCKED',out},null,2));",[],[561],"07672df5-5413-4182-b0d5-2b32beaa0081","pyodide_probe.ts","3a9f848e-aa39-11f1-8b0c-1607ee4eb77e",40,["D",1788729833259],"try {\n const mod = await import(\"npm:pyodide\");\n const py = await mod.loadPyodide();\n const result = py.runPython(\"sum(i*i for i in range(10))\");\n console.log(JSON.stringify({status:\"PASS\",version:mod.version,result},null,2));\n} catch (e) {\n console.log(JSON.stringify({status:\"BLOCKED\",error:String(e)},null,2));\n}",[],[569],"d3478695-ab23-4ced-ba33-45fb9cac7566","neuron_selftest.ts","b37a05da-aa18-11f1-a8c4-1607ee4eb77e",9,["D",1788715849210],"import { classifyCandidate, rankGeneration, offspringPlan, MUTATION_ARCHETYPES } from \"./lib/neuron_evolution.ts\";\n\nconst incubation = classifyCandidate({ agentId: \"INC\", netRevenue: 0, operationallyEligible: false, terminal: false });\nconst running = classifyCandidate({ agentId: \"RUN\", netRevenue: 100, operationallyEligible: true, terminal: false });\nconst zero = classifyCandidate({ agentId: \"ZERO\", netRevenue: 0, operationallyEligible: true, terminal: true });\nconst loss = classifyCandidate({ agentId: \"LOSS\", netRevenue: -0.01, operationallyEligible: true, terminal: true });\nconst win = classifyCandidate({ agentId: \"WIN\", netRevenue: 12.5, operationallyEligible: true, terminal: true });\n\nconst ranked = rankGeneration([\n { agentId: \"A\", netRevenue: 100, operationallyEligible: true, terminal: true },\n { agentId: \"B\", netRevenue: 80, operationallyEligible: true, terminal: true },\n { agentId: \"C\", netRevenue: 50, operationallyEligible: true, terminal: true },\n { agentId: \"D\", netRevenue: 10, operationallyEligible: true, terminal: true },\n { agentId: \"E\", netRevenue: 1, operationallyEligible: true, terminal: true },\n { agentId: \"Z\", netRevenue: 0, operationallyEligible: true, terminal: true },\n]);\n\nconst byId = new Map(ranked.map((x) => [x.agentId, x]));\nconst topPlan = offspringPlan(byId.get(\"A\")!);\nconst deadPlan = offspringPlan(byId.get(\"Z\")!);\n\nconst checks = [\n { name: \"incubation is not killed\", pass: incubation.state === \"INCUBATION\" && incubation.offspringCount === 0 },\n { name: \"nonterminal profit is still incubation\", pass: running.state === \"INCUBATION\" && running.offspringCount === 0 },\n { name: \"terminal zero dies\", pass: zero.state === \"DEAD\" && zero.offspringCount === 0 },\n { name: \"terminal loss dies\", pass: loss.state === \"DEAD\" && loss.offspringCount === 0 },\n { name: \"positive terminal survives\", pass: win.state === \"SURVIVOR\" && win.benchmark === 12.5 },\n { name: \"highest net ranks first\", pass: byId.get(\"A\")?.rank === 1 },\n { name: \"top survivor gets more offspring than tail survivor\", pass: (byId.get(\"A\")?.offspringCount ?? 0) > (byId.get(\"E\")?.offspringCount ?? 0) },\n { name: \"dead candidate never spawns\", pass: deadPlan.length === 0 },\n { name: \"top offspring plan contains faithful replica\", pass: topPlan.some((x) => x.mutation === \"faithful_replica\") },\n { name: \"top offspring plan contains opposite hypothesis\", pass: topPlan.some((x) => x.mutation === \"opposite_hypothesis\") },\n { name: \"top offspring plan contains causal ablation\", pass: topPlan.some((x) => x.mutation === \"causal_ablation\") },\n { name: \"offspring inherit parent revenue benchmark\", pass: topPlan.every((x) => x.targetNetRevenue === 100) },\n { name: \"mutation archetypes remain unique\", pass: new Set(MUTATION_ARCHETYPES).size === MUTATION_ARCHETYPES.length },\n];\n\nconst failed = checks.filter((x) => !x.pass);\nconsole.log(JSON.stringify({\n status: failed.length ? \"FAIL\" : \"PASS\",\n checks,\n failed,\n samples: { incubation, running, zero, loss, win },\n ranked,\n topPlan,\n deadPlan,\n}, null, 2));\n\nif (failed.length) throw new Error(`neuron_selftest failed: ${failed.map((x) => x.name).join(\", \")}`);\n",[],[577],"0b8a62cc-6643-43ee-836c-80ffdb940f3c","main.ts","01a077bc-2b99-740d-9881-5062f5c9e3f1",["D",1788715215771],"http","export default async function (req: Request): Promise {\n if (req.method !== \"GET\") {\n return Response.json({ error: \"method_not_allowed\" }, { status: 405 });\n }\n\n return Response.json({\n service: \"chat-native-maestro\",\n version: \"0.5.0\",\n status: \"alive\",\n executionSurface: \"authenticated_val_town_plugin_only\",\n automaticExecution: \"disabled\",\n security: {\n publicEndpointWriteMethods: \"disabled\",\n detailedRuntimeState: \"not_exposed_publicly\",\n arbitraryCodeInput: \"disabled\",\n },\n });\n}\n",[],[585],"faee7910-7236-4be9-a865-f28cdd5603f0","kaggriculture_shop_world_census_v1.ts","7f4d5b02-aa38-11f1-82cf-1607ee4eb77e",31,["D",1788729505529],"import { loadOfficialKaggricultureEngine } from \"./lib/kaggriculture_engine_loader.ts\";\nconst E=await loadOfficialKaggricultureEngine();\nconst N=500;\nconst PASS={farmer:[\"PASS\"],hands:[],market:[]};\nconst products=[\"CARROT\",\"TOMATO\",\"EGG\"];\nconst rows:any[]=[];\nfor(let seed=1;seed<=N;seed++){\n const cfg=E.state.resolveConfig({seed}); let s=E.state.initGameState(2,cfg,seed);\n while(!s.done)s=E.interpreter.step(s,[PASS,PASS],cfg);\n const counts:any={};for(const sh of s.town.unlocked_shops)counts[sh]=(counts[sh]??0)+1;\n const rec:any={seed,shops:[...s.town.unlocked_shops],counts};\n for(const p of products){rec[p]={consumed:10000-s.market.inventory[p],inventory:s.market.inventory[p],price:s.market.prices[p]};}\n rows.push(rec);\n}\nconst quant=(xs:number[],q:number)=>{const a=[...xs].sort((x,y)=>x-y);const i=(a.length-1)*q,lo=Math.floor(i),hi=Math.ceil(i);return lo===hi?a[lo]:a[lo]+(a[hi]-a[lo])*(i-lo);};\nconst summary:any={};\nfor(const p of products){const cons=rows.map(r=>r[p].consumed),price=rows.map(r=>r[p].price);summary[p]={consumed:{min:Math.min(...cons),p10:quant(cons,.1),p25:quant(cons,.25),median:quant(cons,.5),p75:quant(cons,.75),p90:quant(cons,.9),p95:quant(cons,.95),max:Math.max(...cons)},price:{min:Math.min(...price),p50:quant(price,.5),p75:quant(price,.75),p90:quant(price,.9),p95:quant(price,.95),max:Math.max(...price)},gt100:price.filter(x=>x>100).length/N,gt200:price.filter(x=>x>200).length/N,gt500:price.filter(x=>x>500).length/N};}\nconst petHist:any={},fmHist:any={},pizzaHist:any={},eggShopHist:any={};\nfor(const r of rows){const pet=r.counts.PET_CAFE??0,fm=r.counts.FARMERS_MARKET??0,pizza=r.counts.PIZZA_SHOP??0,egg=(r.counts.BAKERY??0)+(r.counts.BRUNCH_SPOT??0);petHist[pet]=(petHist[pet]??0)+1;fmHist[fm]=(fmHist[fm]??0)+1;pizzaHist[pizza]=(pizzaHist[pizza]??0)+1;eggShopHist[egg]=(eggShopHist[egg]??0)+1;}\nconst extremes=(p:string)=>[...rows].sort((a,b)=>b[p].price-a[p].price).slice(0,10).map(r=>({seed:r.seed,price:r[p].price,consumed:r[p].consumed,counts:r.counts,shops:r.shops}));\nconsole.log(JSON.stringify({status:\"PASS\",evidenceState:\"OFFICIAL_TS_PORT_STATISTICAL_SHOP_CENSUS\",N,summary,histograms:{PET_CAFE:petHist,FARMERS_MARKET:fmHist,PIZZA_SHOP:pizzaHist,EGG_DEMAND_SHOPS:eggShopHist},extremes:{CARROT:extremes(\"CARROT\"),TOMATO:extremes(\"TOMATO\"),EGG:extremes(\"EGG\")},caveat:\"TS port RNG is not bit-identical to CPython; this census is for distributional mechanism discovery, not exact seed replay parity.\"},null,2));",[],[593],"413c051e-d17e-4504-9004-92671c8dc075","kaggriculture_seed9403_econ_diagnostic.ts","35ed3a24-aa40-11f1-9086-1607ee4eb77e",55,["D",1788732818434],"import { loadOfficialKaggricultureEngine } from './lib/kaggriculture_engine_loader.ts';\nconst E=await loadOfficialKaggricultureEngine();\nconst seed=9403; const cfg=E.state.resolveConfig({seed}); let s=E.state.initGameState(2,cfg,seed);\nconst PASS={farmer:['PASS'],hands:[],market:[]};\nconst history:any[]=[]; let prev=[] as string[];\nwhile(!s.done){\n const shops=[...s.town.unlocked_shops];\n if(JSON.stringify(shops)!==JSON.stringify(prev)){\n history.push({day:s.day,hour:s.hour,shops:[...shops],prices:{MILK:s.market.prices.MILK,WOOL:s.market.prices.WOOL,WHEAT:s.market.prices.WHEAT},inventory:{MILK:s.market.inventory.MILK,WOOL:s.market.inventory.WOOL,WHEAT:s.market.inventory.WHEAT}});\n prev=shops;\n }\n s=E.interpreter.step(s,[PASS,PASS],cfg);\n}\nconst items=['MILK','WOOL','WHEAT','CARROT','TOMATO']; const final:any={};\nfor(const p of items) final[p]={consumed:10000-s.market.inventory[p],inventory:s.market.inventory[p],price:s.market.prices[p]};\nconst counts:any={}; for(const sh of s.town.unlocked_shops)counts[sh]=(counts[sh]??0)+1;\nconsole.log(JSON.stringify({status:'PASS',seed,counts,history,final},null,2));\n",[],[601],"83965c46-3128-4450-a50f-b64de2454e02","kaggriculture_route_sweep_v1.ts","c0b9a6d2-aa37-11f1-a849-1607ee4eb77e",24,["D",1788729185837],"import { loadOfficialKaggricultureEngine, observationFor } from \"./lib/kaggriculture_engine_loader.ts\";\nimport { buildAgent, K0, STARTER, type AgentFn } from \"./lib/kaggriculture_agents_v1.ts\";\n\nconst E = await loadOfficialKaggricultureEngine();\nconst cfgBase = E.state.resolveConfig();\nconst seeds = [11, 23, 37];\n\nconst variants: Record = {\n K0,\n R1_H1_T5: buildAgent({ crop:\"CARROT\", targetHands:1, maxTiles:5, persistent:true, seedBuffer:8 }),\n R2_H2_T6: buildAgent({ crop:\"CARROT\", targetHands:2, maxTiles:6, persistent:true, seedBuffer:10 }),\n R3_H3_T8: buildAgent({ crop:\"CARROT\", targetHands:3, maxTiles:8, persistent:true, seedBuffer:12 }),\n R4_H4_T10: buildAgent({ crop:\"CARROT\", targetHands:4, maxTiles:10, persistent:true, seedBuffer:15 }),\n R5_H6_T12: buildAgent({ crop:\"CARROT\", targetHands:6, maxTiles:12, persistent:true, seedBuffer:18 }),\n R6_H8_T16: buildAgent({ crop:\"CARROT\", targetHands:8, maxTiles:16, persistent:true, seedBuffer:22 }),\n R7_H10_T25: buildAgent({ crop:\"CARROT\", targetHands:10, maxTiles:25, persistent:true, seedBuffer:30 }),\n R4NP_H4_T10: buildAgent({ crop:\"CARROT\", targetHands:4, maxTiles:10, persistent:false, seedBuffer:15 }),\n};\n\nfunction stat(){return {unit:0,move:0,pass:0,work:0};}\nfunction count(s:any,a:any){for(const u of [a?.farmer??[\"PASS\"],...(a?.hands??[])]){s.unit++;const op=String(u?.[0]??\"PASS\");if([\"NORTH\",\"SOUTH\",\"EAST\",\"WEST\"].includes(op))s.move++;else if(op===\"PASS\")s.pass++;else s.work++;}}\n\nasync function play(a:AgentFn,b:AgentFn,seed:number){\n const cfg={...cfgBase,seed}; let state=E.state.initGameState(2,cfg,seed); const ss=[stat(),stat()];\n while(!state.done){const aa=a(observationFor(state,0)),bb=b(observationFor(state,1));count(ss[0],aa);count(ss[1],bb);state=E.interpreter.step(state,[aa,bb],cfg);}\n return {banks:state.farms.map((f:any)=>Number(f.money)),stats:ss};\n}\n\nconst results:any[]=[];\nfor(const [name,agent] of Object.entries(variants)){\n if(name===\"K0\") continue;\n for(const opponentName of [\"K0\",\"STARTER\"]){\n const opp=opponentName===\"K0\"?K0:STARTER;\n for(const seed of seeds){\n let g=await play(agent,opp,seed); results.push({name,opponentName,seed,seat:0,bank:g.banks[0],oppBank:g.banks[1],won:g.banks[0]>g.banks[1],tied:g.banks[0]===g.banks[1],s:g.stats[0]});\n g=await play(opp,agent,seed); results.push({name,opponentName,seed,seat:1,bank:g.banks[1],oppBank:g.banks[0],won:g.banks[1]>g.banks[0],tied:g.banks[1]===g.banks[0],s:g.stats[1]});\n }\n }\n}\n\nconst avg=(xs:number[])=>xs.reduce((a,b)=>a+b,0)/(xs.length||1);\nconst summary:any[]=[];\nfor(const name of Object.keys(variants).filter(n=>n!==\"K0\")){\n const rows=results.filter(r=>r.name===name);\n const byOpp:any={};\n for(const opp of [\"K0\",\"STARTER\"]){const rr=rows.filter(r=>r.opponentName===opp);byOpp[opp]={episodes:rr.length,wins:rr.filter(r=>r.won).length,ties:rr.filter(r=>r.tied).length,winRate:rr.filter(r=>r.won).length/rr.length,avgBank:avg(rr.map(r=>r.bank)),avgMargin:avg(rr.map(r=>r.bank-r.oppBank))};}\n summary.push({name,episodes:rows.length,wins:rows.filter(r=>r.won).length,winRate:rows.filter(r=>r.won).length/rows.length,avgBank:avg(rows.map(r=>r.bank)),avgMargin:avg(rows.map(r=>r.bank-r.oppBank)),moveRate:avg(rows.map(r=>r.s.move/r.s.unit)),passRate:avg(rows.map(r=>r.s.pass/r.s.unit)),workRate:avg(rows.map(r=>r.s.work/r.s.unit)),byOpp});\n}\nsummary.sort((a,b)=>b.winRate-a.winRate||b.avgMargin-a.avgMargin||b.avgBank-a.avgBank);\nconsole.log(JSON.stringify({status:\"PASS\",evidenceState:\"OFFICIAL_TS_PORT_SCREENING\",seeds,summary},null,2));\n",[],[609],"2d56ab4e-fb23-46ef-86e9-eba4e80d13d9","kaggriculture_rita_neuron_sweep_v1.ts","a8a8a4d8-aa39-11f1-a2b2-1607ee4eb77e",45,["D",1788730004418],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\n\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();\nconst zres=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!zres.ok)throw new Error(String(zres.status));const z=unzipSync(new Uint8Array(await zres.arrayBuffer()));\nconst ritaSrc=strFromU8(z['rancher_rita.py']),mateoSrc=strFromU8(z['melon_mateo.py']);\nconst variants:Record={\n RITA_BASE:{},\n H6:{hands:6},H10:{hands:10},\n FEED12:{feed_float_days:12},FEED20:{feed_float_days:20},\n C8S8:{animal_target:{COW:8,SHEEP:8}},\n C12S4:{animal_target:{COW:12,SHEEP:4}},\n C6S10:{animal_target:{COW:6,SHEEP:10}},\n TOTAL12:{animal_target:{COW:8,SHEEP:4},buildTarget:12},\n TOTAL20:{animal_target:{COW:12,SHEEP:8},buildTarget:20},\n CARE_OFF:{care:false},\n LAND1:{land:1},\n};\nconst py=await loadPyodide();\npy.runPython(`import json, copy\\n_REG={}`);\npy.globals.set('rita_src',ritaSrc);py.globals.set('mateo_src',mateoSrc);\npy.runPython(`\n_m={}\\nexec(mateo_src,_m)\\n_REG['MATEO']=_m['agent']\n`);\nfor(const [name,patch] of Object.entries(variants)){\n py.globals.set('nm',name);py.globals.set('patch_json',JSON.stringify(patch));\n py.runPython(`\n_ns={}\\nexec(rita_src,_ns)\\n_p=json.loads(patch_json)\\n_bt=_p.pop('buildTarget',None)\\nfor _k,_v in _p.items(): _ns['POLICY'][_k]=_v\\nif _bt is not None and _ns['POLICY'].get('build'):\\n _ns['POLICY']['build'][0]['target']=_bt\\n_REG[nm]=_ns['agent']\n`);\n}\npy.runPython(`def call_ref(name,payload):\\n return json.dumps(_REG[name](json.loads(payload)))`);const call=py.globals.get('call_ref');\nconst seeds=[8101,8129,8171];\nasync function play(a:string,b:string,seed:number){const cfg={...cfg0,seed};let s=E.state.initGameState(2,cfg,seed);let errors=0;while(!s.done){const aa=[];for(const [pid,nm] of [[0,a],[1,b]] as any){try{aa[pid]=JSON.parse(String(call(nm,JSON.stringify(observationFor(s,pid)))));}catch(e){errors++;aa[pid]={farmer:['PASS'],hands:[],market:[]};}}s=E.interpreter.step(s,aa,cfg);}return {a,b,seed,ba:s.farms[0].money,bb:s.farms[1].money,errors};}\nconst games:any[]=[];for(const name of Object.keys(variants).filter(n=>n!=='RITA_BASE'))for(const opp of ['RITA_BASE','MATEO'])for(const seed of seeds){games.push(await play(name,opp,seed));games.push(await play(opp,name,seed));}\nconst avg=(xs:number[])=>xs.reduce((a,b)=>a+b,0)/(xs.length||1);const summary:any={};for(const name of Object.keys(variants).filter(n=>n!=='RITA_BASE')){summary[name]={};for(const opp of ['RITA_BASE','MATEO']){const r=games.filter(g=>(g.a===name&&g.b===opp)||(g.a===opp&&g.b===name)).map(g=>g.a===name?{bank:g.ba,opp:g.bb,error:g.errors}:{bank:g.bb,opp:g.ba,error:g.errors});summary[name][opp]={episodes:r.length,wins:r.filter(x=>x.bank>x.opp).length,ties:r.filter(x=>x.bank===x.opp).length,winRate:r.filter(x=>x.bank>x.opp).length/r.length,avgBank:avg(r.map(x=>x.bank)),avgMargin:avg(r.map(x=>x.bank-x.opp)),errors:r.reduce((a,x)=>a+x.error,0)};}}\nconst ranking=Object.entries(summary).map(([name,s]:any)=>({name,parentWin:s.RITA_BASE.winRate,parentMargin:s.RITA_BASE.avgMargin,parentBank:s.RITA_BASE.avgBank,mateoWin:s.MATEO.winRate,mateoMargin:s.MATEO.avgMargin})).sort((a,b)=>b.parentWin-a.parentWin||b.parentMargin-a.parentMargin||b.mateoWin-a.mateoWin);\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'MIT_RITA_POLICY_NEURON_SCREEN_ON_OFFICIAL_TS_PORT',seeds,ranking,summary},null,2));",[],[617],"72afe9f5-9c71-4ff3-a75f-60549a195150","kaggriculture_rita_gen2_sweep_v1.ts","02730b8e-aa3a-11f1-a463-1607ee4eb77e",46,["D",1788730155091],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();\nconst zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!zr.ok)throw new Error(String(zr.status));const z=unzipSync(new Uint8Array(await zr.arrayBuffer()));const src=strFromU8(z['rancher_rita.py']);\nconst patches:Record={RITA_BASE:{},C4S12:{animal_target:{COW:4,SHEEP:12}},C5S11:{animal_target:{COW:5,SHEEP:11}},C6S10:{animal_target:{COW:6,SHEEP:10}},C7S9:{animal_target:{COW:7,SHEEP:9}},C8S8:{animal_target:{COW:8,SHEEP:8}},C6S10_F20:{animal_target:{COW:6,SHEEP:10},feed_float_days:20},C6S10_L1:{animal_target:{COW:6,SHEEP:10},land:1},C6S10_H6:{animal_target:{COW:6,SHEEP:10},hands:6},C6S10_H10:{animal_target:{COW:6,SHEEP:10},hands:10}};\nconst py=await loadPyodide();py.runPython(`import json\\n_REG={}`);py.globals.set('src',src);for(const [name,patch] of Object.entries(patches)){py.globals.set('nm',name);py.globals.set('patch',JSON.stringify(patch));py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n['agent']`);}py.runPython(`def call(name,payload):\\n return json.dumps(_REG[name](json.loads(payload)))`);const call=py.globals.get('call');\nconst seeds=[9101,9137,9187,9221,9277,9311];const challengers=Object.keys(patches).filter(x=>!['RITA_BASE','C6S10'].includes(x));\nasync function play(a:string,b:string,seed:number){const cfg={...cfg0,seed};let s=E.state.initGameState(2,cfg,seed);let errors=0;while(!s.done){const acts=[];for(const [pid,n] of [[0,a],[1,b]] as any){try{acts[pid]=JSON.parse(String(call(n,JSON.stringify(observationFor(s,pid)))));}catch{errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]};}}s=E.interpreter.step(s,acts,cfg);}return{a,b,ba:s.farms[0].money,bb:s.farms[1].money,errors};}\nconst games:any[]=[];for(const c of challengers)for(const opp of ['C6S10','RITA_BASE'])for(const seed of seeds){games.push(await play(c,opp,seed));games.push(await play(opp,c,seed));}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const out:any={};for(const c of challengers){out[c]={};for(const opp of ['C6S10','RITA_BASE']){const r=games.filter(g=>(g.a===c&&g.b===opp)||(g.a===opp&&g.b===c)).map(g=>g.a===c?{b:g.ba,o:g.bb,e:g.errors}:{b:g.bb,o:g.ba,e:g.errors});out[c][opp]={episodes:r.length,wins:r.filter(x=>x.b>x.o).length,ties:r.filter(x=>x.b===x.o).length,winRate:r.filter(x=>x.b>x.o).length/r.length,avgBank:avg(r.map(x=>x.b)),avgMargin:avg(r.map(x=>x.b-x.o)),errors:r.reduce((a,x)=>a+x.e,0)};}}\nconst rank=Object.entries(out).map(([n,s]:any)=>({name:n,vsC6S10:s.C6S10.winRate,marginVsC6S10:s.C6S10.avgMargin,vsRita:s.RITA_BASE.winRate,marginVsRita:s.RITA_BASE.avgMargin,avgBankVsRita:s.RITA_BASE.avgBank})).sort((a,b)=>b.vsC6S10-a.vsC6S10||b.marginVsC6S10-a.marginVsC6S10||b.vsRita-a.vsRita);\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'RITA_GEN2_STATIC_MUTATION_SCREEN_OFFICIAL_TS_PORT',seeds,rank,out},null,2));",[],[625],"75d35ba4-43d8-44f7-96fd-2b8f3405f1c6","kaggriculture_rita_adaptive_race_v2.ts","18e2f87e-aa40-11f1-ad35-1607ee4eb77e",54,["D",1788732769678],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { sprt, wilson } from './lib/kaggriculture_adaptive_racing.ts';\n\nconst baseline='C6S10';\nconst patches:Record={\n RITA_BASE:{},\n C4S12:{animal_target:{COW:4,SHEEP:12}},\n C5S11:{animal_target:{COW:5,SHEEP:11}},\n C6S10:{animal_target:{COW:6,SHEEP:10}},\n C7S9:{animal_target:{COW:7,SHEEP:9}},\n C8S8:{animal_target:{COW:8,SHEEP:8}},\n C6S10_F20:{animal_target:{COW:6,SHEEP:10},feed_float_days:20},\n C6S10_L1:{animal_target:{COW:6,SHEEP:10},land:1},\n C6S10_H6:{animal_target:{COW:6,SHEEP:10},hands:6},\n C6S10_H10:{animal_target:{COW:6,SHEEP:10},hands:10},\n};\nconst candidates=Object.keys(patches).filter(x=>!['RITA_BASE',baseline].includes(x));\nconst seeds=[9403,9419,9431,9461,9491,9511,9533,9551,9587,9613,9643,9677,9701,9721,9749,9769,9791,9811,9833,9857];\nconst now=new Date().toISOString();\n\n// Ensure resumable race rows exist.\nawait sqlite.batch(candidates.map(c=>({\n sql:`INSERT OR IGNORE INTO kag_race_state(candidate_id,baseline_id,updated_at) VALUES(?,?,?)`,\n args:[c,baseline,now]\n})));\nconst prior=await sqlite.execute(`SELECT * FROM kag_race_state WHERE baseline_id=? ORDER BY candidate_id`,[baseline]);\nconst active=(prior.rows as any[]).filter(r=>r.decision==='CONTINUE' && Number(r.next_seed_index)e.opp?'WIN':e.bank=8?test.decision:'CONTINUE';\n statements.push({sql:`UPDATE kag_race_state SET wins=?,losses=?,ties=?,margin_sum=?,episodes=?,llr=?,decision=?,next_seed_index=?,updated_at=? WHERE candidate_id=? AND baseline_id=?`,args:[wins,losses,ties,margin,n,test.llr,decision,idx+1,new Date().toISOString(),candidate,baseline]});\n round.push({candidate,seed,wins,losses,ties,episodes:n,avgMargin:margin/n,llr:test.llr,decision,wilson:wilson(wins,losses)});\n}\nif(statements.length)await sqlite.batch(statements);\nconst after=await sqlite.execute(`SELECT candidate_id,wins,losses,ties,episodes,CASE WHEN episodes>0 THEN margin_sum/episodes ELSE 0 END AS avg_margin,llr,decision,next_seed_index FROM kag_race_state WHERE baseline_id=? ORDER BY decision,candidate_id`,[baseline]);\nconsole.log(JSON.stringify({status:'PASS',method:'RESUMABLE_ADAPTIVE_RACING',baseline,processedCandidates:active.length,round,after:after.rows,caveat:'SPRT tests p=0.50 vs p=0.65 on decisive games; ties remain separate diagnostics. Promotion still requires broader league evaluation.'},null,2));\n",[],[633],"b8decdbc-71ee-4910-a732-3dd159011cbc","kaggriculture_research_fusion_selftest.ts","70ad93f8-aa3f-11f1-a880-1607ee4eb77e",52,["D",1788732487513],"import { EmpiricalLeague, regretMatchingMix, mixEntropy } from \"./lib/kaggriculture_league.ts\";\nimport { BayesianRouter, shouldReplanMacro, opponentFamilyLikelihood } from \"./lib/kaggriculture_bayesian_router.ts\";\nimport { lowerTailCVaR, riskAdjustedProfit, shadowValue, sellHoldDecision, reserveOptionValue, rankBottlenecks, liquidationPressure, returnOnCommittedCapital } from \"./lib/kaggriculture_farm_economics.ts\";\nimport { sprt, wilson, allocateBatch, dominanceDecision } from \"./lib/kaggriculture_adaptive_racing.ts\";\n\nconst checks:{name:string;pass:boolean;detail?:unknown}[]=[];\nconst check=(name:string,pass:boolean,detail?:unknown)=>{checks.push({name,pass,detail}); if(!pass) throw new Error(name)};\n\n// League / PSRO-like controls.\nconst L=new EmpiricalLeague();\nfor(const m of [\n {id:\"MAIN\",role:\"MAIN\" as const},\n {id:\"OLD\",role:\"HISTORICAL\" as const,frozen:true},\n {id:\"EXP\",role:\"EXPLOITER\" as const},\n {id:\"REF\",role:\"REFERENCE\" as const},\n]) L.add(m);\nfor(let i=0;i<8;i++) L.record(\"MAIN\",\"OLD\",\"WIN\");\nfor(let i=0;i<2;i++) L.record(\"MAIN\",\"OLD\",\"LOSS\");\nfor(let i=0;i<2;i++) L.record(\"MAIN\",\"EXP\",\"WIN\");\nfor(let i=0;i<8;i++) L.record(\"MAIN\",\"EXP\",\"LOSS\");\nconst weights=L.pfspWeights(\"MAIN\",[\"OLD\",\"EXP\",\"REF\"]);\nconst wMap=new Map(weights.map(x=>[x.id,x.probability]));\ncheck(\"PFSP prioritizes known weakness over easy historical\",(wMap.get(\"EXP\")??0)>(wMap.get(\"OLD\")??0),weights);\ncheck(\"weakness detector finds exploiter first\",L.weaknesses(\"MAIN\",1)[0].id===\"EXP\",L.weaknesses(\"MAIN\",3));\nL.freezeSnapshot(\"MAIN\",\"MAIN@1\");\ncheck(\"historical snapshot frozen\",L.members.get(\"MAIN@1\")?.frozen===true,L.members.get(\"MAIN@1\"));\n\n// RPS should retain diversity under regret matching, not collapse to one pure strategy.\nconst rps=[[0,-1,1],[1,0,-1],[-1,1,0]];\nconst mix=regretMatchingMix(rps,10000);\ncheck(\"RPS meta-solver preserves three-way mixture\",mix.every(p=>Math.abs(p-1/3)<0.04),mix);\ncheck(\"RPS entropy high\",mixEntropy(mix)>1.05,mixEntropy(mix));\n\n// Bayesian world/opponent routing.\nconst R=new BayesianRouter([{id:\"CROP\",prior:0.25},{id:\"LIVESTOCK\",prior:0.25},{id:\"MARKET_HEAVY\",prior:0.25},{id:\"MIXED\",prior:0.25}]);\nconst like=opponentFamilyLikelihood({cropShare:0.1,livestockCount:12,recentSellIntensity:0.15,landUnlocked:3});\nconst up=R.update({name:\"visible herd\",likelihood:like});\ncheck(\"Bayes raises livestock family\",R.best().id===\"LIVESTOCK\",{like,posterior:R.distribution()});\ncheck(\"macro does not replan every turn\",shouldReplanMacro({previousBest:\"LIVESTOCK\",currentBest:\"LIVESTOCK\",posteriorConfidence:R.confidence(),informationGainKL:up.kl,event:\"NO_EVENT\"})===false);\ncheck(\"horizon phase forces replan\",shouldReplanMacro({previousBest:\"LIVESTOCK\",currentBest:\"LIVESTOCK\",posteriorConfidence:R.confidence(),informationGainKL:0,event:\"HORIZON_PHASE\"})===true);\n\n// Farm economics.\nconst profits=[100,110,120,130,140,150,160,170,180,190];\ncheck(\"lower-tail CVaR uses worst tail\",lowerTailCVaR(profits,0.2)===105,lowerTailCVaR(profits,0.2));\ncheck(\"risk-adjusted profit below mean\",riskAdjustedProfit(profits,0.5,0.2)<145,riskAdjustedProfit(profits,0.5,0.2));\ncheck(\"positive shadow value detected\",shadowValue(1000,1250,5,50)===40,shadowValue(1000,1250,5,50));\nconst hold=sellHoldDecision({spotPrice:60,expectedFuturePrice:90,holdingCost:5,strategicExternality:0,optionFlexibilityValue:3});\ncheck(\"dynamic sell/hold can prefer hold\",hold.action===\"HOLD\",hold);\nconst strategic=sellHoldDecision({spotPrice:60,expectedFuturePrice:90,holdingCost:5,strategicExternality:35,optionFlexibilityValue:3});\ncheck(\"opponent externality can flip hold to sell\",strategic.action===\"SELL\",strategic);\nconst reserve=reserveOptionValue({interruptionProbability:0.25,lossIfInterrupted:1000,reserveCarryingCost:20,reservePurchaseCost:100});\ncheck(\"reserve option valued from avoided interruption\",reserve.shouldHoldReserve===true,reserve);\nconst bottlenecks=rankBottlenecks([{resource:\"LAND\",shadowValue:-5,utilization:1},{resource:\"FEED\",shadowValue:40,utilization:0.95},{resource:\"LABOR\",shadowValue:10,utilization:0.5}]);\ncheck(\"positive binding resource ranks first\",bottlenecks[0].resource===\"FEED\",bottlenecks);\ncheck(\"liquidation pressure rises near horizon\",liquidationPressure(3,3,0,100)>liquidationPressure(100,3,0,100),{near:liquidationPressure(3,3,0,100),far:liquidationPressure(100,3,0,100)});\ncheck(\"return on capital computed\",returnOnCommittedCapital(500,1000)===0.5);\n\n// Adaptive experiment racing.\nconst strong=sprt({wins:40,losses:10,p0:0.5,p1:0.65});\ncheck(\"SPRT can promote strong branch\",strong.decision===\"PROMOTE_H1\",strong);\nconst weak=sprt({wins:10,losses:40,p0:0.5,p1:0.65});\ncheck(\"SPRT can reject weak branch\",weak.decision===\"REJECT_H1\",weak);\nconst ci10=wilson(6,4), ci100=wilson(60,40);\ncheck(\"uncertainty shrinks with more matches\",(ci100.high-ci100.low)<(ci10.high-ci10.low),{ci10,ci100});\nconst batch=allocateBatch([\n {id:\"UNCERTAIN\",wins:6,losses:4,avgMargin:100},\n {id:\"OBVIOUS\",wins:95,losses:5,avgMargin:500},\n],20);\ncheck(\"adaptive racing spends more tests on uncertain branch\",(batch.find(x=>x.id===\"UNCERTAIN\")?.matches??0)>(batch.find(x=>x.id===\"OBVIOUS\")?.matches??0),batch);\nconst dom=dominanceDecision({id:\"A\",wins:95,losses:5,avgMargin:500},{id:\"B\",wins:20,losses:80,avgMargin:-100});\ncheck(\"clear fixed-checkpoint dominance detected\",dom.decision===\"A_DOMINATES\",dom);\n\nconsole.log(JSON.stringify({status:\"PASS\",checks:checks.length,details:checks},null,2));\n",[],[641],"c063ae72-97a0-458b-9ec5-d623aceade02","kaggriculture_remote_probe.ts","ded83b20-aa36-11f1-9d57-1607ee4eb77e",17,["D",1788728806872],"const urls = [\n \"https://raw.githubusercontent.com/Kaggle/kaggle-environments/master/kaggle_environments/envs/kaggriculture/visualizer/playable/src/engine/constants.ts\",\n \"https://raw.githubusercontent.com/Kaggle/kaggle-environments/master/kaggle_environments/envs/kaggriculture/visualizer/playable/src/engine/interpreter.ts\",\n];\n\nconst out = [];\nfor (const url of urls) {\n const res = await fetch(url);\n const text = await res.text();\n out.push({ url, status: res.status, ok: res.ok, length: text.length, head: text.slice(0, 120) });\n}\nconsole.log(JSON.stringify({ status: out.every(x => x.ok) ? \"PASS\" : \"FAIL\", out }, null, 2));\n",[],[649],"8def5161-d26b-4d52-b253-6dc0286e86f3","kaggriculture_reference_ladder_screen_v1.ts","60eb6e46-aa39-11f1-9aaa-1607ee4eb77e",42,["D",1788729884084],"import { unzipSync, strFromU8 } from \"npm:fflate\";\nimport { loadPyodide } from \"npm:pyodide\";\nimport { loadOfficialKaggricultureEngine, observationFor } from \"./lib/kaggriculture_engine_loader.ts\";\nimport { buildAgent, type AgentFn } from \"./lib/kaggriculture_agents_v1.ts\";\n\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();\nconst OUR:Record={\n H3_T8:buildAgent({crop:'CARROT',targetHands:3,maxTiles:8,persistent:true,seedBuffer:12}),\n H4_T10:buildAgent({crop:'CARROT',targetHands:4,maxTiles:10,persistent:true,seedBuffer:15}),\n H6_T12:buildAgent({crop:'CARROT',targetHands:6,maxTiles:12,persistent:true,seedBuffer:18}),\n};\nconst refs=['fallow_finn.py','wheat_walter.py','rotation_rosa.py','homestead_hana.py','melon_mateo.py','rancher_rita.py'];\nconst zipRes=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!zipRes.ok)throw new Error(`dataset ${zipRes.status}`);const files=unzipSync(new Uint8Array(await zipRes.arrayBuffer()));\nconst py=await loadPyodide();py.runPython(`import json\\n_REG={}`);\nfor(const file of refs){py.globals.set('src',strFromU8(files[file]));py.globals.set('nm',file);py.runPython(`_ns={}\\nexec(src,_ns)\\n_REG[nm]=_ns['agent']`);}\npy.runPython(`def call_ref(name,payload):\\n return json.dumps(_REG[name](json.loads(payload)))`);const call=py.globals.get('call_ref');\nconst seeds=[7001,7027];\nasync function play(ourName:string,ref:string,seed:number,ourSeat:0|1){const cfg={...cfg0,seed};let s=E.state.initGameState(2,cfg,seed);let errors=0;while(!s.done){const actions:any[]=[null,null];actions[ourSeat]=OUR[ourName](observationFor(s,ourSeat));const rseat=(1-ourSeat) as 0|1;try{actions[rseat]=JSON.parse(String(call(ref,JSON.stringify(observationFor(s,rseat)))));}catch(e){errors++;actions[rseat]={farmer:['PASS'],hands:[],market:[]};}s=E.interpreter.step(s,actions,cfg);}const ob=s.farms[ourSeat].money,rb=s.farms[1-ourSeat].money;return {ourName,ref,seed,ourSeat,ourBank:ob,refBank:rb,win:ob>rb,tie:ob===rb,errors};}\nconst games:any[]=[];for(const ourName of Object.keys(OUR))for(const ref of refs)for(const seed of seeds){games.push(await play(ourName,ref,seed,0));games.push(await play(ourName,ref,seed,1));}\nconst avg=(xs:number[])=>xs.reduce((a,b)=>a+b,0)/(xs.length||1);const summary:any={};for(const ourName of Object.keys(OUR)){summary[ourName]={};for(const ref of refs){const r=games.filter(g=>g.ourName===ourName&&g.ref===ref);summary[ourName][ref]={episodes:r.length,wins:r.filter(g=>g.win).length,ties:r.filter(g=>g.tie).length,winRate:r.filter(g=>g.win).length/r.length,avgBank:avg(r.map(g=>g.ourBank)),avgRefBank:avg(r.map(g=>g.refBank)),avgMargin:avg(r.map(g=>g.ourBank-g.refBank)),errors:r.reduce((a,g)=>a+g.errors,0)};}}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'PYODIDE_MIT_REFERENCE_AGENTS_ON_OFFICIAL_TS_PORT',seeds,referenceDataset:'raykkretzschmar/kaggriculture-reference-agents v8 MIT',summary},null,2));",[],[657],"db972f17-15a7-4205-b117-04784524bad7","kaggriculture_py_reference_interop_probe.ts","4fb7c73c-aa39-11f1-8d5f-1607ee4eb77e",41,["D",1788729855237],"import { unzipSync, strFromU8 } from \"npm:fflate\";\nimport { loadPyodide } from \"npm:pyodide\";\nimport { loadOfficialKaggricultureEngine, observationFor } from \"./lib/kaggriculture_engine_loader.ts\";\nimport { K0 } from \"./lib/kaggriculture_agents_v1.ts\";\n\nconst zipRes=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');\nif(!zipRes.ok)throw new Error(`dataset ${zipRes.status}`);\nconst files=unzipSync(new Uint8Array(await zipRes.arrayBuffer()));\nconst source=strFromU8(files['rotation_rosa.py']);\nconst py=await loadPyodide();\npy.globals.set('agent_source',source);\npy.runPython(`\nimport json\n_ns={}\nexec(agent_source,_ns)\n_agent=_ns['agent']\ndef call_agent_json(payload):\n return json.dumps(_agent(json.loads(payload)))\n`);\nconst call=py.globals.get('call_agent_json');\nconst E=await loadOfficialKaggricultureEngine();const seed=4242;const cfg=E.state.resolveConfig({seed});let state=E.state.initGameState(2,cfg,seed);let calls=0,errors=0;\nwhile(!state.done){\n const a0=K0(observationFor(state,0));let a1:any;\n try{a1=JSON.parse(String(call(JSON.stringify(observationFor(state,1)))));calls++;}catch(e){errors++;a1={farmer:['PASS'],hands:[],market:[]};}\n state=E.interpreter.step(state,[a0,a1],cfg);\n}\nconsole.log(JSON.stringify({status:errors===0?'PASS':'PARTIAL',reference:'rotation_rosa.py',license:'MIT dataset',seed,calls,errors,finalBanks:state.farms.map((f:any)=>f.money),winner:state.farms[0].money===state.farms[1].money?'TIE':state.farms[0].money>state.farms[1].money?'K0':'ROTATION_ROSA'},null,2));",[],[665],"eb8dc61e-c883-49fb-bc6e-b83ce367c5dd","kaggriculture_official_port_probe.ts","ef9a676c-aa36-11f1-a686-1607ee4eb77e",18,["D",1788728834984],"const BASE = \"https://raw.githubusercontent.com/Kaggle/kaggle-environments/master/kaggle_environments/envs/kaggriculture/visualizer/playable/src/engine/\";\nconst FILES = [\"types.ts\",\"constants.ts\",\"market.ts\",\"rng.ts\",\"state.ts\",\"interpreter.ts\"];\n\nfunction stripImports(src: string): string {\n // Remove multiline/single-line ES imports; concatenation supplies symbols in one module.\n return src.replace(/^import[\\s\\S]*?;\\s*$/gm, \"\");\n}\n\nconst chunks: string[] = [];\nconst fetched: Array> = [];\nfor (const file of FILES) {\n const url = BASE + file;\n const res = await fetch(url);\n const text = await res.text();\n if (!res.ok) throw new Error(`fetch failed ${file}: ${res.status}`);\n chunks.push(`\\n// ===== ${file} =====\\n` + stripImports(text));\n fetched.push({ file, bytes: text.length });\n}\n\nconst source = chunks.join(\"\\n\");\nconst dataUrl = \"data:application/typescript;base64,\" + btoa(unescape(encodeURIComponent(source)));\nconst engine = await import(dataUrl);\n\nconst cfg = engine.resolveConfig({ seed: 12345 });\nlet state = engine.initGameState(2, cfg, 12345);\nconst PASS = { farmer: [\"PASS\"], hands: [], market: [] };\nconst start = { step: state.step, day: state.day, hour: state.hour, money: state.farms.map((f: any)=>f.money) };\nstate = engine.step(state, [PASS, PASS], cfg);\nconst after = { step: state.step, day: state.day, hour: state.hour, money: state.farms.map((f: any)=>f.money), done: state.done };\n\nconsole.log(JSON.stringify({\n status: \"PASS\",\n fetched,\n sourceBytes: source.length,\n exports: Object.keys(engine).sort(),\n config: cfg,\n start,\n after,\n}, null, 2));\n",[],[673],"64d8c272-605c-49ab-bbc0-2124c3f4fc49","kaggriculture_official_module_probe.ts","4a004de8-aa37-11f1-9f0b-1607ee4eb77e",19,["D",1788728986637],"const RAW_BASE = \"https://raw.githubusercontent.com/Kaggle/kaggle-environments/master/kaggle_environments/envs/kaggriculture/visualizer/playable/src/engine/\";\nconst SPEC_URL = \"https://raw.githubusercontent.com/Kaggle/kaggle-environments/master/kaggle_environments/envs/kaggriculture/kaggriculture.json\";\n\nconst names = [\"types\", \"constants\", \"market\", \"rng\", \"state\", \"interpreter\"] as const;\ntype Name = typeof names[number];\nconst raw = new Map();\n\nfor (const name of names) {\n const res = await fetch(`${RAW_BASE}${name}.ts`);\n if (!res.ok) throw new Error(`fetch ${name}.ts failed: ${res.status}`);\n raw.set(name, await res.text());\n}\nconst specRes = await fetch(SPEC_URL);\nif (!specRes.ok) throw new Error(`fetch kaggriculture.json failed: ${specRes.status}`);\nconst specText = await specRes.text();\nJSON.parse(specText); // fail early if source is not valid JSON\n\nconst moduleUrls = new Map();\n\nfunction toDataUrl(source: string) {\n const bytes = new TextEncoder().encode(source);\n let binary = \"\";\n // chunk to avoid argument/stack limits\n for (let i = 0; i < bytes.length; i += 0x8000) {\n binary += String.fromCharCode(...bytes.subarray(i, i + 0x8000));\n }\n return `data:application/typescript;base64,${btoa(binary)}`;\n}\n\nfunction compileModule(name: Name): string {\n let src = raw.get(name)!;\n\n // Type-only imports have no runtime role and are removed.\n src = src.replace(/import\\s+type\\s+\\{[\\s\\S]*?\\}\\s+from\\s+['\"][^'\"]+['\"];?\\s*/g, \"\");\n\n // constants.ts intentionally reads defaults from the official JSON spec.\n if (name === \"constants\") {\n src = src.replace(\n /import\\s+spec\\s+from\\s+['\"]\\.\\.\\/\\.\\.\\/\\.\\.\\/\\.\\.\\/kaggriculture\\.json['\"];?\\s*/,\n `const spec = ${specText};\\n`,\n );\n }\n\n // Rewrite relative runtime imports to already-compiled data modules.\n src = src.replace(\n /import\\s+\\{([\\s\\S]*?)\\}\\s+from\\s+['\"]\\.\\/([A-Za-z0-9_-]+)['\"];?\\s*/g,\n (_all, symbols: string, dep: string) => {\n const depUrl = moduleUrls.get(dep as Name);\n if (!depUrl) throw new Error(`${name}: runtime dependency ${dep} was not compiled first`);\n return `import {${symbols}} from ${JSON.stringify(depUrl)};\\n`;\n },\n );\n\n const unresolved = [...src.matchAll(/from\\s+['\"](\\.\\.?\\/[^'\"]+)['\"]/g)].map((m) => m[1]);\n if (unresolved.length) throw new Error(`${name}: unresolved imports ${unresolved.join(\", \")}`);\n return src;\n}\n\nfor (const name of names) {\n const compiled = compileModule(name);\n moduleUrls.set(name, toDataUrl(compiled));\n}\n\nconst stateMod = await import(moduleUrls.get(\"state\")!);\nconst engine = await import(moduleUrls.get(\"interpreter\")!);\nconst constants = await import(moduleUrls.get(\"constants\")!);\n\nconst cfg = stateMod.resolveConfig({ seed: 12345 });\nlet state = stateMod.initGameState(2, cfg, 12345);\nconst PASS = { farmer: [\"PASS\"], hands: [], market: [] };\nconst initial = {\n step: state.step,\n day: state.day,\n hour: state.hour,\n money: state.farms.map((f: any) => f.money),\n carrot: constants.MARKET_PARAMS.CARROT,\n};\nstate = engine.step(state, [PASS, PASS], cfg);\nconst after1 = {\n step: state.step,\n day: state.day,\n hour: state.hour,\n money: state.farms.map((f: any) => f.money),\n done: state.done,\n};\n\nconsole.log(JSON.stringify({\n status: \"PASS\",\n fetched: names.map((name) => ({ name, bytes: raw.get(name)!.length })),\n officialCommitMode: \"master-at-runtime\",\n config: cfg,\n initial,\n after1,\n}, null, 2));\n",[],[681],"f4ce025a-bc5e-4790-82fd-156e381c5dd8","kaggriculture_market_timing_sweep_v1.ts","f7d5a476-aa38-11f1-9404-1607ee4eb77e",35,["D",1788729723611],"import { loadOfficialKaggricultureEngine, observationFor } from \"./lib/kaggriculture_engine_loader.ts\";\nimport { buildAgent, STARTER, type AgentFn } from \"./lib/kaggriculture_agents_v1.ts\";\nimport { withSellPolicy, AFTER_TOWN_HOURS, BEFORE_TOWN_HOURS } from \"./lib/kaggriculture_market_mutations.ts\";\n\nconst E=await loadOfficialKaggricultureEngine();\nconst base=()=>buildAgent({crop:\"CARROT\",targetHands:4,maxTiles:10,persistent:true,seedBuffer:15});\nconst variants:Record={\n SELL_ALL:base(),\n AFTER_ALL:withSellPolicy(base(),{name:\"AFTER_ALL\",allowedHours:AFTER_TOWN_HOURS}),\n AFTER_L4:withSellPolicy(base(),{name:\"AFTER_L4\",allowedHours:AFTER_TOWN_HOURS,lotSize:4}),\n AFTER_L8:withSellPolicy(base(),{name:\"AFTER_L8\",allowedHours:AFTER_TOWN_HOURS,lotSize:8}),\n AFTER_L12:withSellPolicy(base(),{name:\"AFTER_L12\",allowedHours:AFTER_TOWN_HOURS,lotSize:12}),\n BEFORE_L8:withSellPolicy(base(),{name:\"BEFORE_L8\",allowedHours:BEFORE_TOWN_HOURS,lotSize:8}),\n FLOOR25_L8:withSellPolicy(base(),{name:\"FLOOR25_L8\",allowedHours:AFTER_TOWN_HOURS,lotSize:8,reservePrice:25,forceShedLoad:85}),\n FLOOR40_L8:withSellPolicy(base(),{name:\"FLOOR40_L8\",allowedHours:AFTER_TOWN_HOURS,lotSize:8,reservePrice:40,forceShedLoad:85}),\n FLOOR60_L8:withSellPolicy(base(),{name:\"FLOOR60_L8\",allowedHours:AFTER_TOWN_HOURS,lotSize:8,reservePrice:60,forceShedLoad:85}),\n};\nconst worlds:Record={LOW:[],D7:[\"FARMERS_MARKET\"],D13:[\"PET_CAFE\"],D19:[\"PET_CAFE\",\"FARMERS_MARKET\"],D31:[\"PET_CAFE\",\"PET_CAFE\",\"FARMERS_MARKET\"]};\n\nasync function play(agent:AgentFn,shops:string[]){\n const cfg=E.state.resolveConfig({seed:314159,weedSpawnChance:0,townShopUnlockInterval:9999});\n let s=E.state.initGameState(2,cfg,314159);s.town.unlocked_shops=[...shops];\n let sold=0,sellOrders=0;\n while(!s.done){const a=agent(observationFor(s,0)),b=STARTER(observationFor(s,1));for(const o of a.market??[]){if(o?.[0]===\"SELL\"&&o?.[1]===\"CARROT\"){sold+=Number(o?.[2]??0);sellOrders++;}}s=E.interpreter.step(s,[a,b],cfg);}\n return {bank:s.farms[0].money,starter:s.farms[1].money,margin:s.farms[0].money-s.farms[1].money,price:s.market.prices.CARROT,inventory:s.market.inventory.CARROT,shed:s.privates[0].shed.CARROT,soldRequested:sold,sellOrders};\n}\nconst out:any={};\nfor(const [world,shops] of Object.entries(worlds)){\n const rows=[];\n for(const [name,a] of Object.entries(variants)) rows.push({name,...await play(a,shops)});\n rows.sort((x,y)=>y.margin-x.margin||y.bank-x.bank);\n out[world]=rows;\n}\nconst compact=Object.fromEntries(Object.entries(out).map(([world,rows]:any)=>[world,rows.slice(0,4)]));\nconsole.log(JSON.stringify({status:\"PASS\",evidenceState:\"MARKET_TIMING_ABLATION_OFFICIAL_TS_PORT\",config:{seed:314159,weedSpawnChance:0,townShopUnlockInterval:9999},compact},null,2));",[],[689],"e28a79ad-37b5-4814-af3f-d7c903006e29","kaggriculture_livestock_portfolio_selftest.ts","0ba224fe-aa41-11f1-a1f2-1607ee4eb77e",61,["D",1788733176985],"import { marketPrice,caredProductionUnits,expectedTownDemandRemaining,choosePastureAnimal,scoreAnimalInvestment } from './lib/kaggriculture_livestock_portfolio.ts';\nconst checks:any[]=[];const check=(name:string,pass:boolean,detail?:any)=>{checks.push({name,pass,detail});if(!pass)throw new Error(name)};\nfunction obs(day=0,cash=3000,shops:string[]=[]){const mk={MILK:10000,WOOL:10000,EGG:10000,WHEAT:10000};const prices={MILK:160,WOOL:200,EGG:50,WHEAT:25};const farm={money:cash,tiles:Array.from({length:10},()=>Array(10).fill(null)),hands:[],unlocked_quadrants:['NW']};return{player:0,day,hour:0,market:{inventory:mk,prices},town:{unlocked_shops:shops},farms:[farm,structuredClone(farm)]};}\ncheck('base milk price exact',marketPrice('MILK',10000)===160,marketPrice('MILK',10000));\ncheck('base wool price exact',marketPrice('WOOL',10000)===200,marketPrice('WOOL',10000));\ncheck('cow cared production day0',caredProductionUnits('COW',0)===22,caredProductionUnits('COW',0));\ncheck('sheep cared production day0',caredProductionUnits('SHEEP',0)===16,caredProductionUnits('SHEEP',0));\ncheck('late horizon favors sheep production availability',caredProductionUnits('SHEEP',22)>caredProductionUnits('COW',22),{sheep:caredProductionUnits('SHEEP',22),cow:caredProductionUnits('COW',22)});\nconst d0=obs();check('expected milk demand > wool under unknown future shops',expectedTownDemandRemaining(d0,'MILK')>expectedTownDemandRemaining(d0,'WOOL'),{milk:expectedTownDemandRemaining(d0,'MILK'),wool:expectedTownDemandRemaining(d0,'WOOL')});\nconst early=choosePastureAnimal(obs(0,3000));check('early ranking has finite scores',early.ranking.every(x=>Number.isFinite(x.net)),early);\nconst late=choosePastureAnimal(obs(22,3000));check('late ranking can distinguish horizon',late.ranking[0].net!==late.ranking[1].net,late);\nconst poor=scoreAnimalInvestment({obs:obs(5,450),animal:'SHEEP',cashBuffer:300});check('liquidity penalty applies below buffer',poor.liquidityPenalty>0,poor);\nconsole.log(JSON.stringify({status:'PASS',checks:checks.length,details:checks,early,late},null,2));",[],[697],"419d02aa-12d1-4e8f-bd99-2a67d57c99b0","kaggriculture_knowledge_migrate_v1.ts","05390f3e-aa40-11f1-b79b-1607ee4eb77e",53,["D",1788732736730],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nconst now=new Date().toISOString();\nawait sqlite.batch([\n {sql:`CREATE TABLE IF NOT EXISTS kag_league_members(\n member_id TEXT PRIMARY KEY, role TEXT NOT NULL, lineage TEXT, frozen INTEGER NOT NULL DEFAULT 0,\n status TEXT NOT NULL DEFAULT 'ACTIVE', source TEXT, evidence_state TEXT, created_at TEXT NOT NULL, updated_at TEXT NOT NULL)`},\n {sql:`CREATE TABLE IF NOT EXISTS kag_match_results(\n match_id TEXT PRIMARY KEY, a_id TEXT NOT NULL, b_id TEXT NOT NULL, seed INTEGER, mirrored INTEGER NOT NULL DEFAULT 0,\n bank_a REAL, bank_b REAL, result_a TEXT, runtime_errors INTEGER NOT NULL DEFAULT 0,\n evidence_state TEXT NOT NULL, created_at TEXT NOT NULL)`},\n {sql:`CREATE INDEX IF NOT EXISTS idx_kag_match_pair ON kag_match_results(a_id,b_id,created_at)`},\n {sql:`CREATE TABLE IF NOT EXISTS kag_race_state(\n candidate_id TEXT NOT NULL, baseline_id TEXT NOT NULL, wins INTEGER NOT NULL DEFAULT 0, losses INTEGER NOT NULL DEFAULT 0,\n ties INTEGER NOT NULL DEFAULT 0, margin_sum REAL NOT NULL DEFAULT 0, episodes INTEGER NOT NULL DEFAULT 0,\n llr REAL NOT NULL DEFAULT 0, decision TEXT NOT NULL DEFAULT 'CONTINUE', next_seed_index INTEGER NOT NULL DEFAULT 0,\n updated_at TEXT NOT NULL, PRIMARY KEY(candidate_id,baseline_id))`},\n {sql:`CREATE TABLE IF NOT EXISTS kag_research_claims(\n claim_id TEXT PRIMARY KEY, track TEXT NOT NULL, source_type TEXT NOT NULL, title TEXT NOT NULL, url TEXT,\n mechanism TEXT NOT NULL, kag_mapping TEXT NOT NULL, hypothesis TEXT NOT NULL,\n status TEXT NOT NULL DEFAULT 'CANDIDATE', created_at TEXT NOT NULL, updated_at TEXT NOT NULL)`},\n {sql:`CREATE TABLE IF NOT EXISTS kag_strategy_lessons(\n lesson_id TEXT PRIMARY KEY, lesson_type TEXT NOT NULL, mechanism TEXT NOT NULL, preconditions TEXT,\n counterfactual TEXT, detector TEXT, evidence_state TEXT NOT NULL, created_at TEXT NOT NULL, updated_at TEXT NOT NULL)`},\n {sql:`INSERT OR IGNORE INTO kag_league_members(member_id,role,lineage,frozen,status,source,evidence_state,created_at,updated_at) VALUES\n ('RITA_BASE','REFERENCE','RITA',1,'ACTIVE','Kaggle reference agents','PUBLIC_REFERENCE',?,?),\n ('C6S10','MAIN','RITA→C6S10',1,'CHALLENGER','Neurônio mutation','LOCAL_SCREEN',?,?),\n ('MELON_MATEO','REFERENCE','REFERENCE',1,'ACTIVE','Kaggle reference agents','PUBLIC_REFERENCE',?,?),\n ('ROTATION_ROSA','REFERENCE','REFERENCE',1,'ACTIVE','Kaggle reference agents','PUBLIC_REFERENCE',?,?)`,args:[now,now,now,now,now,now,now,now]},\n {sql:`INSERT OR IGNORE INTO kag_research_claims(claim_id,track,source_type,title,url,mechanism,kag_mapping,hypothesis,status,created_at,updated_at) VALUES\n ('A-LEAGUE-HISTORY','AGENT','RESEARCH','Historical opponent pools','https://arxiv.org/abs/2011.12895','Training against historical opponent pools improves robustness and reduces overfitting','Freeze old champions and sample them during training/evaluation','Historical snapshots prevent forgetting and cyclic collapse','IMPLEMENTED',?,?),\n ('A-PSRO-BR','AGENT','RESEARCH','Policy Space Response Oracles','https://arxiv.org/abs/2408.01072','Iteratively add best responses to a strategy population','Maintain empirical payoff matrix plus specialized exploiters','Best-response offspring expose champion weaknesses faster than blind mutation','IMPLEMENTED',?,?),\n ('B-SELL-CUTOFF','ECONOMICS','RESEARCH','Optimal selling policies for farmer cooperatives','https://journals.sagepub.com/doi/abs/10.1111/poms.13091','Dynamic inventory models produce state-dependent cutoff selling policies','Replace fixed sell floor with expected future value, holding cost, horizon and opponent externality','State-dependent sell/hold beats fixed reservation price across market regimes','IMPLEMENTED',?,?),\n ('B-DUOPOLY-PRICE','ECONOMICS','RESEARCH','Dynamic pricing and inventory control under competition','https://pubsonline.informs.org/doi/abs/10.1287/opre.1090.0718','Competitor strategy changes the value of pricing/inventory decisions','Include opponent exposure as strategic externality in sell/hold threshold','Withholding is only valuable when rival cannot capture the same price lift','IMPLEMENTED',?,?),\n ('M-RACING','METHOD','RESEARCH','Best arm identification / racing','https://link.springer.com/article/10.1007/s10489-021-02719-2','Adaptive allocation eliminates weak candidates and spends samples on uncertain contenders','Replace exhaustive seed sweeps with resumable sequential races','Adaptive racing reduces simulation cost without promoting obvious losers','IMPLEMENTED',?,?)`,args:[now,now,now,now,now,now,now,now,now,now]}\n]);\nconst tables=await sqlite.execute(`SELECT name FROM sqlite_master WHERE type='table' AND name LIKE 'kag_%' ORDER BY name`);\nconst claims=await sqlite.execute(`SELECT claim_id,track,status FROM kag_research_claims ORDER BY claim_id`);\nconsole.log(JSON.stringify({status:'PASS',tables:tables.rows,claims:claims.rows},null,2));\n",[],[705],"9eed7f5f-9afe-4e17-8c0b-b1d15b65747c","kaggriculture_k1_diagnostic.ts","a031b828-aa37-11f1-91ab-1607ee4eb77e",23,["D",1788729131171],"import { loadOfficialKaggricultureEngine, observationFor } from \"./lib/kaggriculture_engine_loader.ts\";\nimport { K1, STARTER } from \"./lib/kaggriculture_agents_v1.ts\";\n\nconst E = await loadOfficialKaggricultureEngine();\nconst seed = 11;\nconst cfg = E.state.resolveConfig({ seed });\nlet state = E.state.initGameState(2, cfg, seed);\nconst ledger = { K1: { hireOrders:0, seedQty:0, sellQty:0, moves:0, work:0, pass:0 }, STARTER: { hireOrders:0, seedQty:0, sellQty:0, moves:0, work:0, pass:0 } };\nconst daily: any[] = [];\n\nfunction scan(name: \"K1\"|\"STARTER\", a: any) {\n const l = ledger[name];\n for (const m of (a.market ?? [])) {\n if (m?.[0] === \"HIRE\") l.hireOrders++;\n if (m?.[0] === \"BUY_SEED\" && m?.[1] === \"CARROT\") l.seedQty += Number(m?.[2] ?? 0);\n if (m?.[0] === \"SELL\" && m?.[1] === \"CARROT\") l.sellQty += Number(m?.[2] ?? 0);\n }\n for (const u of [a.farmer ?? [\"PASS\"], ...(a.hands ?? [])]) {\n const op=String(u?.[0] ?? \"PASS\");\n if ([\"NORTH\",\"SOUTH\",\"EAST\",\"WEST\"].includes(op)) l.moves++;\n else if (op === \"PASS\") l.pass++;\n else l.work++;\n }\n}\n\nfunction tileCounts(farm:any) {\n const out:any={empty:0,weed:0,carrot:0,other:0};\n for(const row of farm.tiles) for(const t of row){\n if(t===null) out.empty++;\n else if(t===\"LOCKED\") continue;\n else if(t?.kind===\"WEED\") out.weed++;\n else if(t?.kind===\"PLANT\"&&t?.crop===\"CARROT\") out.carrot++;\n else out.other++;\n }\n return out;\n}\n\nwhile(!state.done){\n const a0=K1(observationFor(state,0));\n const a1=STARTER(observationFor(state,1));\n scan(\"K1\",a0); scan(\"STARTER\",a1);\n state=E.interpreter.step(state,[a0,a1],cfg);\n if(state.hour===0 || state.done){\n daily.push({\n step:state.step,day:state.day,\n bank:[state.farms[0].money,state.farms[1].money],\n carrotPrice:state.market.prices.CARROT,\n carrotInventory:state.market.inventory.CARROT,\n k1ShedCarrot:state.privates[0].shed.CARROT,\n k1Seeds:state.privates[0].seeds.CARROT,\n k1Tiles:tileCounts(state.farms[0]),\n shops:[...state.town.unlocked_shops],\n });\n }\n}\nconsole.log(JSON.stringify({status:\"PASS\",seed,finalBanks:state.farms.map((f:any)=>f.money),ledger,daily},null,2));\n",[],[713],"442c1080-d81c-4e09-a79b-5451642fa55f","kaggriculture_k10_opponent_gate_benchmark_v1.ts","2705ff5c-aa39-11f1-b1e4-1607ee4eb77e",38,["D",1788729786939],"import { loadOfficialKaggricultureEngine, observationFor } from \"./lib/kaggriculture_engine_loader.ts\";\nimport { buildAgent, STARTER, type AgentFn } from \"./lib/kaggriculture_agents_v1.ts\";\nimport { withSellPolicy, AFTER_TOWN_HOURS } from \"./lib/kaggriculture_market_mutations.ts\";\nimport { opponentAwareCarrotMarket } from \"./lib/kaggriculture_opponent_market.ts\";\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();\nconst factory=()=>buildAgent({crop:\"CARROT\",targetHands:4,maxTiles:10,persistent:true,seedBuffer:15});\nconst BASE=factory();const RESERVE=withSellPolicy(factory(),{name:\"R60\",allowedHours:AFTER_TOWN_HOURS,lotSize:8,reservePrice:60,forceShedLoad:85,liquidationDay:29});\nconst agents:Record={BASE,RESERVE,G2:opponentAwareCarrotMarket(factory,2),G3:opponentAwareCarrotMarket(factory,3),G5:opponentAwareCarrotMarket(factory,5),STARTER};\nconst seeds=Array.from({length:20},(_,i)=>5003+i*53);\nconst challengerNames=['RESERVE','G2','G3','G5'];const opponents=['BASE','STARTER'];\nasync function play(a:string,b:string,seed:number){const cfg={...cfg0,seed};let s=E.state.initGameState(2,cfg,seed);while(!s.done)s=E.interpreter.step(s,[agents[a](observationFor(s,0)),agents[b](observationFor(s,1))],cfg);return {a,b,ba:s.farms[0].money,bb:s.farms[1].money};}\nconst games:any[]=[];for(const c of challengerNames)for(const o of opponents)for(const seed of seeds){games.push(await play(c,o,seed));games.push(await play(o,c,seed));}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const summary:any={};\nfor(const c of challengerNames){summary[c]={};for(const o of opponents){const rows=games.filter(g=>(g.a===c&&g.b===o)||(g.a===o&&g.b===c)).map(g=>g.a===c?{bank:g.ba,opp:g.bb}:{bank:g.bb,opp:g.ba});summary[c][o]={episodes:rows.length,wins:rows.filter(r=>r.bank>r.opp).length,ties:rows.filter(r=>r.bank===r.opp).length,winRate:rows.filter(r=>r.bank>r.opp).length/rows.length,avgBank:avg(rows.map(r=>r.bank)),avgMargin:avg(rows.map(r=>r.bank-r.opp))};}}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'OPPONENT_EXPOSURE_GATE_SCREEN_OFFICIAL_TS_PORT',seeds,summary},null,2));",[],[721],"90523511-0d25-4598-9e86-a65985e604ff","kaggriculture_k10_market_benchmark_v1.ts","0e16571c-aa39-11f1-b2b1-1607ee4eb77e",36,["D",1788729745125],"import { loadOfficialKaggricultureEngine, observationFor } from \"./lib/kaggriculture_engine_loader.ts\";\nimport { buildAgent, STARTER, type AgentFn } from \"./lib/kaggriculture_agents_v1.ts\";\nimport { withSellPolicy, AFTER_TOWN_HOURS } from \"./lib/kaggriculture_market_mutations.ts\";\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();\nconst makeBase=()=>buildAgent({crop:\"CARROT\",targetHands:4,maxTiles:10,persistent:true,seedBuffer:15});\nconst BASE=makeBase();const K10=withSellPolicy(makeBase(),{name:\"K10_FLOOR60_AFTER\",allowedHours:AFTER_TOWN_HOURS,lotSize:8,reservePrice:60,forceShedLoad:85,liquidationDay:29});\nconst agents:Record={BASE,K10,STARTER};\nconst seeds=Array.from({length:30},(_,i)=>1001+i*37);\nconst pairs:Array<[string,string]>=[[\"K10\",\"BASE\"],[\"BASE\",\"K10\"],[\"K10\",\"STARTER\"],[\"STARTER\",\"K10\"],[\"BASE\",\"STARTER\"],[\"STARTER\",\"BASE\"]];\nasync function play(a:string,b:string,seed:number){const cfg={...cfg0,seed};let s=E.state.initGameState(2,cfg,seed);while(!s.done)s=E.interpreter.step(s,[agents[a](observationFor(s,0)),agents[b](observationFor(s,1))],cfg);return {a,b,seed,ba:s.farms[0].money,bb:s.farms[1].money,winner:s.farms[0].money===s.farms[1].money?\"TIE\":s.farms[0].money>s.farms[1].money?a:b,shops:[...s.town.unlocked_shops]};}\nconst games=[];for(const seed of seeds)for(const [a,b] of pairs)games.push(await play(a,b,seed));\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const summary:any={};\nfor(const name of Object.keys(agents)){const r:any[]=[];for(const g of games){if(g.a===name)r.push({bank:g.ba,opp:g.bb,win:g.ba>g.bb,tie:g.ba===g.bb});if(g.b===name)r.push({bank:g.bb,opp:g.ba,win:g.bb>g.ba,tie:g.bb===g.ba});}summary[name]={episodes:r.length,wins:r.filter(x=>x.win).length,ties:r.filter(x=>x.tie).length,losses:r.filter(x=>!x.win&&!x.tie).length,winRate:r.filter(x=>x.win).length/r.length,avgBank:avg(r.map(x=>x.bank)),avgMargin:avg(r.map(x=>x.bank-x.opp))};}\nconst h2h=games.filter(g=>(g.a==='K10'&&g.b==='BASE')||(g.a==='BASE'&&g.b==='K10'));const k10h=h2h.map(g=>g.a==='K10'?{bank:g.ba,opp:g.bb,win:g.ba>g.bb,tie:g.ba===g.bb}:{bank:g.bb,opp:g.ba,win:g.bb>g.ba,tie:g.bb===g.ba});\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'K10_NATURAL_SHOP_SCREEN_OFFICIAL_TS_PORT',seeds,pairCount:games.length,summary,h2hK10vsBase:{episodes:k10h.length,wins:k10h.filter(x=>x.win).length,ties:k10h.filter(x=>x.tie).length,losses:k10h.filter(x=>!x.win&&!x.tie).length,winRate:k10h.filter(x=>x.win).length/k10h.length,avgBank:avg(k10h.map(x=>x.bank)),avgMargin:avg(k10h.map(x=>x.bank-x.opp))}},null,2));",[],[729],"d5944d8b-b305-47c6-a559-0efea817fb65","kaggriculture_herd_mix_telemetry_v1.ts","8b957950-aa40-11f1-9f75-1607ee4eb77e",58,["D",1788732996656],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\n\nconst seed=9403;\nconst variants:Record={\n C4S12:{animal_target:{COW:4,SHEEP:12}},\n C6S10:{animal_target:{COW:6,SHEEP:10}},\n};\nconst E=await loadOfficialKaggricultureEngine(); const cfg0=E.state.resolveConfig();\nconst zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');\nif(!zr.ok)throw new Error(`REF_DATASET_HTTP_${zr.status}`);\nconst z=unzipSync(new Uint8Array(await zr.arrayBuffer())); const src=strFromU8(z['rancher_rita.py']);\nconst py=await loadPyodide(); py.runPython(`import json\\n_REG={}`); py.globals.set('src',src);\nfor(const [name,patch] of Object.entries(variants)){\n py.globals.set('nm',name);py.globals.set('patch',JSON.stringify(patch));\n py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n['agent']`);\n}\npy.runPython(`def call(name,payload):\\n return json.dumps(_REG[name](json.loads(payload)))`); const call=py.globals.get('call');\n\nfunction animalCounts(farm:any){\n const out:any={COW:0,SHEEP:0,GOOSE:0,PASTURE:0,COOP:0};\n for(const row of farm?.tiles??[])for(const t of row??[]){\n if(!t||typeof t!=='object')continue;\n const a=String(t.animal??t.species??t.animal_type??'');\n if(a in out)out[a]++;\n const k=String(t.kind??''); if(k==='PASTURE')out.PASTURE++; if(k==='COOP')out.COOP++;\n }\n return out;\n}\nfunction snapObs(obs:any){\n const me=Number(obs?.player??0),f=obs?.farms?.[me]??{};\n return {\n money:Number(f.money??0), hands:(f.hands??[]).length, land:(f.unlocked_quadrants??[]).length,\n animals:animalCounts(f), shed:{...(obs?.private?.shed??{})}, seeds:{...(obs?.private?.seeds??{})},\n };\n}\nfunction aggOrder(a:any,acc:any){\n for(const o of a?.market??[]){if(!Array.isArray(o)||!o.length)continue;const op=String(o[0]),item=String(o[1]??''),qty=Number(o[2]??1);const k=item?`${op}:${item}`:op;acc[k]=(acc[k]??0)+qty;}\n}\nfunction unitOps(a:any,acc:any){for(const u of [a?.farmer??['PASS'],...(a?.hands??[])]){const op=String(u?.[0]??'PASS');acc[op]=(acc[op]??0)+1;}}\n\nasync function play(aName:string,bName:string){\n const cfg={...cfg0,seed};let s=E.state.initGameState(2,cfg,seed);let errors=0;\n const T:any={}; for(const n of [aName,bName])T[n]={minCash:Infinity,maxCash:-Infinity,market:{},unitOps:{},daily:[],events:[],pricePath:[]};\n let prev:any={};\n while(!s.done){\n const before=s;\n const acts:any[]=[]; const obs:any[]=[];\n for(const [pid,n] of [[0,aName],[1,bName]] as any){\n obs[pid]=observationFor(s,pid); const ss=snapObs(obs[pid]);\n T[n].minCash=Math.min(T[n].minCash,ss.money);T[n].maxCash=Math.max(T[n].maxCash,ss.money);\n try{acts[pid]=JSON.parse(String(call(n,JSON.stringify(obs[pid]))));}catch(e){errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]};}\n aggOrder(acts[pid],T[n].market);unitOps(acts[pid],T[n].unitOps);\n }\n const day=s.day,hour=s.hour;\n if(hour===0){\n for(const [pid,n] of [[0,aName],[1,bName]] as any)T[n].daily.push({day,hour,before:snapObs(obs[pid]),marketPrice:{MILK:s.market.prices.MILK,WOOL:s.market.prices.WOOL,WHEAT:s.market.prices.WHEAT},marketInventory:{MILK:s.market.inventory.MILK,WOOL:s.market.inventory.WOOL,WHEAT:s.market.inventory.WHEAT},shops:[...s.town.unlocked_shops]});\n }\n s=E.interpreter.step(s,acts,cfg);\n if(hour===23){\n for(const [pid,n] of [[0,aName],[1,bName]] as any){\n const oo=observationFor(s,pid); const after=snapObs(oo);\n const rec=T[n].daily.find((x:any)=>x.day===day); if(rec)rec.after=after;\n }\n }\n // Record market trajectory at each shop unlock + every 3 days.\n if((s.hour===0 && s.day%3===0) || s.town.unlocked_shops.length!==(prev.shops??-1)){\n for(const n of [aName,bName])T[n].pricePath.push({day:s.day,hour:s.hour,MILK:s.market.prices.MILK,WOOL:s.market.prices.WOOL,WHEAT:s.market.prices.WHEAT,invMILK:s.market.inventory.MILK,invWOOL:s.market.inventory.WOOL,invWHEAT:s.market.inventory.WHEAT,shops:[...s.town.unlocked_shops]});\n prev.shops=s.town.unlocked_shops.length;\n }\n }\n for(const [pid,n] of [[0,aName],[1,bName]] as any){\n const oo=observationFor(s,pid);T[n].final=snapObs(oo);T[n].bank=s.farms[pid].money;T[n].opponentBank=s.farms[1-pid].money;T[n].margin=T[n].bank-T[n].opponentBank;\n }\n return {a:aName,b:bName,errors,telemetry:T,finalMarket:{prices:s.market.prices,inventory:s.market.inventory},shops:s.town.unlocked_shops};\n}\nconst g1=await play('C4S12','C6S10'); const g2=await play('C6S10','C4S12');\nfunction summarize(g:any){\n const out:any={};\n for(const n of ['C4S12','C6S10']){\n const t=g.telemetry[n];\n const d=t.daily;\n const minDay=[...d].sort((a:any,b:any)=>(a.after?.money??Infinity)-(b.after?.money??Infinity))[0];\n const firstWool=d.find((x:any)=>Number(x.after?.shed?.WOOL??0)>0);\n const firstMilk=d.find((x:any)=>Number(x.after?.shed?.MILK??0)>0);\n const firstCowTarget=d.find((x:any)=>Number(x.after?.animals?.COW??0)>=Number(variants[n].animal_target.COW));\n const firstSheepTarget=d.find((x:any)=>Number(x.after?.animals?.SHEEP??0)>=Number(variants[n].animal_target.SHEEP));\n const checkpoints=[0,4,6,8,10,12,15,18,21,24,27,29].map(day=>{const x=d.find((y:any)=>y.day===day);return x?{day,cash:x.after?.money,cows:x.after?.animals?.COW,sheep:x.after?.animals?.SHEEP,wheat:x.after?.shed?.WHEAT,milk:x.after?.shed?.MILK,wool:x.after?.shed?.WOOL}:null}).filter(Boolean);\n out[n]={bank:t.bank,margin:t.margin,minCash:t.minCash,minCashDay:minDay?.day,maxCash:t.maxCash,marketOrders:t.market,unitOps:t.unitOps,firstWoolInventoryDay:firstWool?.day??null,firstMilkInventoryDay:firstMilk?.day??null,cowTargetDay:firstCowTarget?.day??null,sheepTargetDay:firstSheepTarget?.day??null,checkpoints};\n }\n return out;\n}\nconst A=summarize(g1),B=summarize(g2);\nfunction seatNeutral(name:string){\n const x=[A[name],B[name]];\n return {avgBank:(x[0].bank+x[1].bank)/2,avgMargin:(x[0].margin+x[1].margin)/2,minCash:Math.min(x[0].minCash,x[1].minCash),seatBanks:x.map(y=>y.bank),seatMargins:x.map(y=>y.margin),marketOrders:x.map(y=>y.marketOrders),targetDays:x.map(y=>({cow:y.cowTargetDay,sheep:y.sheepTargetDay})),cashCheckpoints:x.map(y=>y.checkpoints)};\n}\nconsole.log(JSON.stringify({status:'PASS',seed,seatNeutral:{C4S12:seatNeutral('C4S12'),C6S10:seatNeutral('C6S10')},seatA:A,seatB:B,shops:g1.shops,caveat:'Official TS port + Pyodide public MIT Rancher Rita policy. Only animal_target differs.'},null,2));\n",[],[737],"df6ccd57-bff2-4952-8376-34cacfdadace","kaggriculture_herd_mix_telemetry_detail_v1.ts","01a078cb-4c07-74ba-b128-3085fcd62e67",57,["D",1788732984329],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\n\nconst seed=9403;\nconst variants:Record={\n C4S12:{animal_target:{COW:4,SHEEP:12}},\n C6S10:{animal_target:{COW:6,SHEEP:10}},\n};\nconst E=await loadOfficialKaggricultureEngine(); const cfg0=E.state.resolveConfig();\nconst zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');\nif(!zr.ok)throw new Error(`REF_DATASET_HTTP_${zr.status}`);\nconst z=unzipSync(new Uint8Array(await zr.arrayBuffer())); const src=strFromU8(z['rancher_rita.py']);\nconst py=await loadPyodide(); py.runPython(`import json\\n_REG={}`); py.globals.set('src',src);\nfor(const [name,patch] of Object.entries(variants)){\n py.globals.set('nm',name);py.globals.set('patch',JSON.stringify(patch));\n py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n['agent']`);\n}\npy.runPython(`def call(name,payload):\\n return json.dumps(_REG[name](json.loads(payload)))`); const call=py.globals.get('call');\n\nfunction animalCounts(farm:any){\n const out:any={COW:0,SHEEP:0,GOOSE:0,PASTURE:0,COOP:0};\n for(const row of farm?.tiles??[])for(const t of row??[]){\n if(!t||typeof t!=='object')continue;\n const a=String(t.animal??t.species??t.animal_type??'');\n if(a in out)out[a]++;\n const k=String(t.kind??''); if(k==='PASTURE')out.PASTURE++; if(k==='COOP')out.COOP++;\n }\n return out;\n}\nfunction snapObs(obs:any){\n const me=Number(obs?.player??0),f=obs?.farms?.[me]??{};\n return {\n money:Number(f.money??0), hands:(f.hands??[]).length, land:(f.unlocked_quadrants??[]).length,\n animals:animalCounts(f), shed:{...(obs?.private?.shed??{})}, seeds:{...(obs?.private?.seeds??{})},\n };\n}\nfunction aggOrder(a:any,acc:any){\n for(const o of a?.market??[]){if(!Array.isArray(o)||!o.length)continue;const op=String(o[0]),item=String(o[1]??''),qty=Number(o[2]??1);const k=item?`${op}:${item}`:op;acc[k]=(acc[k]??0)+qty;}\n}\nfunction unitOps(a:any,acc:any){for(const u of [a?.farmer??['PASS'],...(a?.hands??[])]){const op=String(u?.[0]??'PASS');acc[op]=(acc[op]??0)+1;}}\n\nasync function play(aName:string,bName:string){\n const cfg={...cfg0,seed};let s=E.state.initGameState(2,cfg,seed);let errors=0;\n const T:any={}; for(const n of [aName,bName])T[n]={minCash:Infinity,maxCash:-Infinity,market:{},unitOps:{},daily:[],events:[],pricePath:[]};\n let prev:any={};\n while(!s.done){\n const before=s;\n const acts:any[]=[]; const obs:any[]=[];\n for(const [pid,n] of [[0,aName],[1,bName]] as any){\n obs[pid]=observationFor(s,pid); const ss=snapObs(obs[pid]);\n T[n].minCash=Math.min(T[n].minCash,ss.money);T[n].maxCash=Math.max(T[n].maxCash,ss.money);\n try{acts[pid]=JSON.parse(String(call(n,JSON.stringify(obs[pid]))));}catch(e){errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]};}\n aggOrder(acts[pid],T[n].market);unitOps(acts[pid],T[n].unitOps);\n }\n const day=s.day,hour=s.hour;\n if(hour===0){\n for(const [pid,n] of [[0,aName],[1,bName]] as any)T[n].daily.push({day,hour,before:snapObs(obs[pid]),marketPrice:{MILK:s.market.prices.MILK,WOOL:s.market.prices.WOOL,WHEAT:s.market.prices.WHEAT},marketInventory:{MILK:s.market.inventory.MILK,WOOL:s.market.inventory.WOOL,WHEAT:s.market.inventory.WHEAT},shops:[...s.town.unlocked_shops]});\n }\n s=E.interpreter.step(s,acts,cfg);\n if(hour===23){\n for(const [pid,n] of [[0,aName],[1,bName]] as any){\n const oo=observationFor(s,pid); const after=snapObs(oo);\n const rec=T[n].daily.find((x:any)=>x.day===day); if(rec)rec.after=after;\n }\n }\n // Record market trajectory at each shop unlock + every 3 days.\n if((s.hour===0 && s.day%3===0) || s.town.unlocked_shops.length!==(prev.shops??-1)){\n for(const n of [aName,bName])T[n].pricePath.push({day:s.day,hour:s.hour,MILK:s.market.prices.MILK,WOOL:s.market.prices.WOOL,WHEAT:s.market.prices.WHEAT,invMILK:s.market.inventory.MILK,invWOOL:s.market.inventory.WOOL,invWHEAT:s.market.inventory.WHEAT,shops:[...s.town.unlocked_shops]});\n prev.shops=s.town.unlocked_shops.length;\n }\n }\n for(const [pid,n] of [[0,aName],[1,bName]] as any){\n const oo=observationFor(s,pid);T[n].final=snapObs(oo);T[n].bank=s.farms[pid].money;T[n].opponentBank=s.farms[1-pid].money;T[n].margin=T[n].bank-T[n].opponentBank;\n }\n return {a:aName,b:bName,errors,telemetry:T,finalMarket:{prices:s.market.prices,inventory:s.market.inventory},shops:s.town.unlocked_shops};\n}\nconst g1=await play('C4S12','C6S10'); const g2=await play('C6S10','C4S12');\nfunction compact(g:any){const out:any={};for(const n of ['C4S12','C6S10']){const t=g.telemetry[n];out[n]={bank:t.bank,margin:t.margin,minCash:t.minCash,maxCash:t.maxCash,market:t.market,unitOps:t.unitOps,daily:t.daily.map((d:any)=>({day:d.day,cash0:d.before.money,cash24:d.after?.money,hands:d.after?.hands,land:d.after?.land,animals:d.after?.animals,shed:d.after?.shed,prices:d.marketPrice,shops:d.shops}))};}return out;}\nconsole.log(JSON.stringify({status:'PASS',seed,seatA:compact(g1),seatB:compact(g2),shops:g1.shops,finalMarket:g1.finalMarket,caveat:'Telemetry derives from official TS port + Pyodide execution of the public MIT Rancher Rita policy with only animal_target mutated.'},null,2));\n",[],[745],"85c0fee9-2170-4b0f-9bbc-5a87da87b689","kaggriculture_dynport_register_v1.ts","47634b7a-aa42-11f1-ae55-1607ee4eb77e",75,["D",1788733706729],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nconst now=new Date().toISOString();\nawait sqlite.batch([\n {sql:`INSERT OR REPLACE INTO kag_league_members(member_id,role,lineage,frozen,status,source,evidence_state,created_at,updated_at) VALUES(?,?,?,?,?,?,?,?,?)`,args:['DYN_PORT_V1','SPECIALIST','RITA→DYN_PORT_V1',1,'CHALLENGER','research-derived closed-loop livestock portfolio','SMALL_MIRRORED_SCREEN_10_OF_12',now,now]},\n {sql:`INSERT OR REPLACE INTO kag_strategy_lessons(lesson_id,lesson_type,mechanism,preconditions,counterfactual,detector,evidence_state,created_at,updated_at) VALUES(?,?,?,?,?,?,?,?,?)`,args:['FAIL-TARGET-RATCHET','FAILURE','TARGET_RATCHET_WITHOUT_EXECUTION','planner changes desired asset count while prior purchase remains unexecuted','rebase desired target on actual committed/owned assets each macro cycle','desired_target - actual_owned grows across cycles without a matching purchase receipt','FALSIFIED_IN_SIM',now,now]},\n {sql:`INSERT OR REPLACE INTO kag_strategy_lessons(lesson_id,lesson_type,mechanism,preconditions,counterfactual,detector,evidence_state,created_at,updated_at) VALUES(?,?,?,?,?,?,?,?,?)`,args:['FAIL-NPV-FINANCE','FAILURE','POSITIVE_NPV_NOT_FINANCIALLY_FEASIBLE','asset scorer ignores feed float, land reserve or working-capital constraint','require scheduler-equivalent financeability gate before ranking as actionable','NPV>0 AND cash={\n DYN_PORT:{animal_target:{COW:0,SHEEP:0}},\n C4S12:{animal_target:{COW:4,SHEEP:12}},\n C6S10:{animal_target:{COW:6,SHEEP:10}},\n RITA_BASE:{},\n};\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();\nconst zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!zr.ok)throw new Error(String(zr.status));\nconst z=unzipSync(new Uint8Array(await zr.arrayBuffer()));const src=strFromU8(z['rancher_rita.py']);\nconst py=await loadPyodide();py.runPython(`import json,copy\\n_REG={}\\n_BASE={}`);py.globals.set('src',src);\nfor(const [name,patch] of Object.entries(patches)){py.globals.set('nm',name);py.globals.set('patch',JSON.stringify(patch));py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n\\n_BASE[nm]=copy.deepcopy(_n['POLICY'])`);}\npy.runPython(`\ndef call(name,payload):\n return json.dumps(_REG[name]['agent'](json.loads(payload)))\ndef set_herd(name,cows,sheep,order):\n _REG[name]['POLICY']['animal_target']={'COW':int(cows),'SHEEP':int(sheep)}\n _REG[name]['POLICY']['animals']=json.loads(order)\ndef reset(name):\n _REG[name]['POLICY'].clear();_REG[name]['POLICY'].update(copy.deepcopy(_BASE[name]))\n`);\nconst call=py.globals.get('call'),setHerd=py.globals.get('set_herd'),reset=py.globals.get('reset');\n\nfunction countOwned(obs:any,animal:string){const me=Number(obs?.player??0),f=obs?.farms?.[me]??{};let n=Number(obs?.private?.shed?.[animal]??0);for(const inv of obs?.private?.inventories??[])n+=Number(inv?.[animal]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===animal)n++;return n;}\nfunction enforceAnimalTargets(obs:any,action:any,targets:{COW:number;SHEEP:number}){const actual:any={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};const out={...action,market:[] as any[]};for(const order of action?.market??[]){if(Array.isArray(order)&&order[0]==='BUY_ANIMAL'&&(order[1]==='COW'||order[1]==='SHEEP')){const a=order[1] as 'COW'|'SHEEP';const allowed=Math.max(0,Math.min(Number(order[2]??1),targets[a]-actual[a]));if(allowed>0){out.market.push(['BUY_ANIMAL',a,allowed]);actual[a]+=allowed;}}else out.market.push(order);}return out;}\nasync function play(a:string,b:string,seed:number){\n reset(a);reset(b);let s=E.state.initGameState(2,{...cfg0,seed},seed);let errors=0;let targets={COW:0,SHEEP:0};const decisions:any[]=[];\n while(!s.done){const acts:any[]=[];\n for(const [pid,n] of [[0,a],[1,b]] as any){const obs=observationFor(s,pid);\n if(n==='DYN_PORT'&&Number(obs.hour)===0){\n const owned={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};\n // An unexecuted target is a revocable option, not committed capital. Rebase intent on actual scheduler-visible ownership every macro cycle.\n targets={COW:owned.COW,SHEEP:owned.SHEEP};\n const actualTotal=owned.COW+owned.SHEEP;const choice=choosePastureAnimal(obs,{cashBuffer:400,strategicWeight:1.0});\n if(actualTotal<16 && choice.choice){const room=16-actualTotal,batch=Math.min(2,room);targets[choice.choice]+=batch;}\n const order=[...choice.ranking].map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP');\n setHerd('DYN_PORT',targets.COW,targets.SHEEP,JSON.stringify(order));\n decisions.push({day:obs.day,player:pid,money:obs.farms[pid].money,owned,targets:{...targets},choice:choice.choice,ranking:choice.ranking.map(x=>({animal:x.animal,net:Math.round(x.net),roi:Number(x.roi.toFixed(2)),currentPrice:x.currentPrice,projectedPrice:x.projectedPrice,townDemand:Number(x.townDemand.toFixed(1)),ownSupply:x.ownCommittedSupply,oppSupply:x.opponentSupply,liquidityPenalty:x.liquidityPenalty}))});\n }\n try{acts[pid]=JSON.parse(String(call(n,JSON.stringify(obs))));if(n==='DYN_PORT')acts[pid]=enforceAnimalTargets(obs,acts[pid],targets);}catch{errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]};}\n }\n s=E.interpreter.step(s,acts,{...cfg0,seed});\n }\n return{a,b,seed,bankA:s.farms[0].money,bankB:s.farms[1].money,errors,decisions};\n}\nconst games:any[]=[];for(const opp of opponents)for(const seed of seeds){games.push(await play('DYN_PORT',opp,seed));games.push(await play(opp,'DYN_PORT',seed));}\nconst summary:any={};for(const opp of opponents){const r=games.filter(g=>g.a===opp||g.b===opp).map(g=>g.a==='DYN_PORT'?{ours:g.bankA,opp:g.bankB,e:g.errors}:{ours:g.bankB,opp:g.bankA,e:g.errors});summary[opp]={episodes:r.length,wins:r.filter(x=>x.ours>x.opp).length,losses:r.filter(x=>x.oursx.ours===x.opp).length,avgBank:r.reduce((s,x)=>s+x.ours,0)/r.length,avgMargin:r.reduce((s,x)=>s+x.ours-x.opp,0)/r.length,errors:r.reduce((s,x)=>s+x.e,0)};}\nconst decisionDigest=games.filter(g=>g.a==='DYN_PORT').map(g=>({opponent:g.b,seed:g.seed,first12:g.decisions.slice(0,12),finalDecision:g.decisions[g.decisions.length-1]}));\nconsole.log(JSON.stringify({status:'PASS',agent:'DYN_PORT',mechanism:'GREEDY_MARGINAL_LIVESTOCK_PORTFOLIO_WITH_EVENT_DAILY_REALLOCATION',seeds,summary,decisionDigest,caveat:'Research prototype. Uses an NPV-like heuristic, not a proven optimal policy; requires ablation and broader league evaluation.'},null,2));",[],[761],"0525c8ee-d719-4cdd-b18a-4d1bd5c33d95","kaggriculture_demand_thresholds_v1.ts","080e77ba-aa38-11f1-89da-1607ee4eb77e",27,["D",1788729305458],"import { loadOfficialKaggricultureEngine, observationFor } from \"./lib/kaggriculture_engine_loader.ts\";\nimport { buildAgent, STARTER, type AgentFn } from \"./lib/kaggriculture_agents_v1.ts\";\nconst E=await loadOfficialKaggricultureEngine();\nconst variants:Record={\n H4_T10:buildAgent({crop:\"CARROT\",targetHands:4,maxTiles:10,persistent:true,seedBuffer:15}),\n H6_T12:buildAgent({crop:\"CARROT\",targetHands:6,maxTiles:12,persistent:true,seedBuffer:18}),\n H8_T16:buildAgent({crop:\"CARROT\",targetHands:8,maxTiles:16,persistent:true,seedBuffer:22}),\n H10_T25:buildAgent({crop:\"CARROT\",targetHands:10,maxTiles:25,persistent:true,seedBuffer:30}),\n};\nconst worlds:Record={D1:[],D7:[\"FARMERS_MARKET\"],D13:[\"PET_CAFE\"],D19:[\"PET_CAFE\",\"FARMERS_MARKET\"],D25:[\"PET_CAFE\",\"PET_CAFE\"],D31:[\"PET_CAFE\",\"PET_CAFE\",\"FARMERS_MARKET\"],D37:[\"PET_CAFE\",\"PET_CAFE\",\"PET_CAFE\"]};\nasync function play(agent:AgentFn,shops:string[]){const cfg=E.state.resolveConfig({seed:999,weedSpawnChance:0,townShopUnlockInterval:9999});let s=E.state.initGameState(2,cfg,999);s.town.unlocked_shops=[...shops];while(!s.done){s=E.interpreter.step(s,[agent(observationFor(s,0)),STARTER(observationFor(s,1))],cfg);}return {bank:s.farms[0].money,starter:s.farms[1].money,margin:s.farms[0].money-s.farms[1].money,price:s.market.prices.CARROT,inventory:s.market.inventory.CARROT};}\nconst out:any={};for(const [world,shops] of Object.entries(worlds)){const rows=[];for(const [variant,a] of Object.entries(variants))rows.push({variant,...await play(a,shops)});rows.sort((x,y)=>y.margin-x.margin);out[world]={shops,rank:rows};}\nconsole.log(JSON.stringify({status:\"PASS\",evidenceState:\"CONTROLLED_CARROT_DEMAND_THRESHOLD_MAP\",sinkFormula:\"1 + 12*PET_CAFE + 6*FARMERS_MARKET units/day\",out},null,2));\n",[],[770],"f784c391-fa5d-4231-939d-1ba6e0e1b30e","kaggriculture_demand_control_v1.ts","dab53ba0-aa37-11f1-b30e-1607ee4eb77e",26,["D",1788729252852],"import { loadOfficialKaggricultureEngine, observationFor } from \"./lib/kaggriculture_engine_loader.ts\";\nimport { buildAgent, K0, STARTER, type AgentFn } from \"./lib/kaggriculture_agents_v1.ts\";\n\nconst E=await loadOfficialKaggricultureEngine();\nconst variants:Record={\n K0,\n R2_H2_T6:buildAgent({crop:\"CARROT\",targetHands:2,maxTiles:6,persistent:true,seedBuffer:10}),\n R3_H3_T8:buildAgent({crop:\"CARROT\",targetHands:3,maxTiles:8,persistent:true,seedBuffer:12}),\n R4_H4_T10:buildAgent({crop:\"CARROT\",targetHands:4,maxTiles:10,persistent:true,seedBuffer:15}),\n R4NP_H4_T10:buildAgent({crop:\"CARROT\",targetHands:4,maxTiles:10,persistent:false,seedBuffer:15}),\n R5_H6_T12:buildAgent({crop:\"CARROT\",targetHands:6,maxTiles:12,persistent:true,seedBuffer:18}),\n R7_H10_T25:buildAgent({crop:\"CARROT\",targetHands:10,maxTiles:25,persistent:true,seedBuffer:30}),\n};\nconst worlds:Record={\n LOW:[],\n MID_FARMERS:[\"FARMERS_MARKET\"],\n HIGH_PET:[\"PET_CAFE\"],\n VERY_HIGH:[\"PET_CAFE\",\"PET_CAFE\",\"FARMERS_MARKET\"],\n};\n\nfunction stats(){return {unit:0,move:0,pass:0,work:0};}\nfunction count(s:any,a:any){for(const u of [a?.farmer??[\"PASS\"],...(a?.hands??[])]){s.unit++;const op=String(u?.[0]??\"PASS\");if([\"NORTH\",\"SOUTH\",\"EAST\",\"WEST\"].includes(op))s.move++;else if(op===\"PASS\")s.pass++;else s.work++;}}\nasync function play(agent:AgentFn,shops:string[]){\n const cfg=E.state.resolveConfig({seed:424242,weedSpawnChance:0,townShopUnlockInterval:9999});\n let state=E.state.initGameState(2,cfg,424242); state.town.unlocked_shops=[...shops]; const s=stats();\n while(!state.done){const a=agent(observationFor(state,0)),b=STARTER(observationFor(state,1));count(s,a);state=E.interpreter.step(state,[a,b],cfg);}\n return {bank:state.farms[0].money,starterBank:state.farms[1].money,margin:state.farms[0].money-state.farms[1].money,finalPrice:state.market.prices.CARROT,finalInventory:state.market.inventory.CARROT,moveRate:s.move/s.unit,passRate:s.pass/s.unit,workRate:s.work/s.unit};\n}\nconst rows:any[]=[];\nfor(const [world,shops] of Object.entries(worlds))for(const [variant,agent] of Object.entries(variants))rows.push({world,shops,variant,...await play(agent,shops)});\nconst byWorld:any={};\nfor(const world of Object.keys(worlds))byWorld[world]=rows.filter(r=>r.world===world).sort((a,b)=>b.margin-a.margin||b.bank-a.bank);\nconst compact=Object.fromEntries(Object.entries(byWorld).map(([world,rows]:any)=>[world,{top3:rows.slice(0,3).map((r:any)=>({variant:r.variant,bank:r.bank,margin:r.margin,finalPrice:r.finalPrice,finalInventory:r.finalInventory,moveRate:r.moveRate})),R7:rows.find((r:any)=>r.variant===\"R7_H10_T25\"),K0:rows.find((r:any)=>r.variant===\"K0\")} ]));\nconsole.log(JSON.stringify({status:\"PASS\",evidenceState:\"CONTROLLED_DEMAND_ABLATION_OFFICIAL_TS_PORT\",config:{seed:424242,weedSpawnChance:0,townShopUnlockInterval:9999},compact},null,2));\n",[],[778],"565c3d43-035c-4f4f-b624-8261d09ef0e8","kaggriculture_config_probe_v1.ts","e3916fce-aa40-11f1-bb9f-1607ee4eb77e",59,["D",1788733109760],"import { loadOfficialKaggricultureEngine } from './lib/kaggriculture_engine_loader.ts';\nconst E=await loadOfficialKaggricultureEngine(); const c=E.state.resolveConfig();\nconsole.log(JSON.stringify({status:'PASS',config:c},null,2));",[],[786],"568fe673-8bee-4941-b72d-6e04ec879d9b","kaggriculture_benchmark_g0_v1.ts","83c5465a-aa37-11f1-8f9e-1607ee4eb77e",22,["D",1788729083534],"import { loadOfficialKaggricultureEngine, observationFor } from \"./lib/kaggriculture_engine_loader.ts\";\nimport { K0, K1, STARTER, type AgentFn } from \"./lib/kaggriculture_agents_v1.ts\";\n\nconst engine = await loadOfficialKaggricultureEngine();\nconst cfgBase = engine.state.resolveConfig();\n\nconst AGENTS: Record = { K0, K1, STARTER };\nconst SEEDS = [11, 23, 37, 53, 71];\nconst PAIRS: Array<[string,string]> = [\n [\"K0\", \"STARTER\"],\n [\"STARTER\", \"K0\"],\n [\"K1\", \"STARTER\"],\n [\"STARTER\", \"K1\"],\n [\"K1\", \"K0\"],\n [\"K0\", \"K1\"],\n];\n\nfunction freshStats() {\n return { unit: 0, move: 0, pass: 0, work: 0, marketOrders: 0, runtimeErrors: 0 };\n}\nfunction countAction(stats: any, action: any) {\n const units = [action?.farmer ?? [\"PASS\"], ...(Array.isArray(action?.hands) ? action.hands : [])];\n for (const u of units) {\n stats.unit++;\n const op = Array.isArray(u) && u.length ? String(u[0]) : \"PASS\";\n if ([\"NORTH\",\"SOUTH\",\"EAST\",\"WEST\"].includes(op)) stats.move++;\n else if (op === \"PASS\") stats.pass++;\n else stats.work++;\n }\n stats.marketOrders += Array.isArray(action?.market) ? action.market.length : 0;\n}\n\nasync function play(aName: string, bName: string, seed: number) {\n const cfg = { ...cfgBase, seed };\n let state = engine.state.initGameState(2, cfg, seed);\n const stats = [freshStats(), freshStats()];\n let turns = 0;\n while (!state.done) {\n const actions: any[] = [];\n for (let pid = 0; pid < 2; pid++) {\n const name = pid === 0 ? aName : bName;\n try {\n const action = AGENTS[name](observationFor(state, pid));\n actions.push(action);\n countAction(stats[pid], action);\n } catch (err) {\n stats[pid].runtimeErrors++;\n const pass = { farmer: [\"PASS\"], hands: [], market: [] };\n actions.push(pass);\n countAction(stats[pid], pass);\n }\n }\n state = engine.interpreter.step(state, actions, cfg);\n turns++;\n if (turns > 1000) throw new Error(`episode did not terminate seed=${seed}`);\n }\n const banks = state.farms.map((f: any) => Number(f.money));\n return {\n seed, aName, bName, turns, banks,\n winner: banks[0] === banks[1] ? \"TIE\" : banks[0] > banks[1] ? aName : bName,\n marginP0: banks[0] - banks[1],\n stats: stats.map((s: any) => ({\n ...s,\n moveRate: s.unit ? s.move / s.unit : 0,\n passRate: s.unit ? s.pass / s.unit : 0,\n workRate: s.unit ? s.work / s.unit : 0,\n })),\n };\n}\n\nconst games = [];\nfor (const seed of SEEDS) {\n for (const [a,b] of PAIRS) games.push(await play(a,b,seed));\n}\n\nconst summary: Record = {};\nfor (const name of Object.keys(AGENTS)) {\n const rows: any[] = [];\n for (const g of games as any[]) {\n if (g.aName === name) rows.push({ bank: g.banks[0], won: g.banks[0] > g.banks[1], tied: g.banks[0] === g.banks[1], stat: g.stats[0] });\n if (g.bName === name) rows.push({ bank: g.banks[1], won: g.banks[1] > g.banks[0], tied: g.banks[1] === g.banks[0], stat: g.stats[1] });\n }\n rows.sort((a,b)=>a.bank-b.bank);\n const n = rows.length;\n const avg = (xs: number[]) => xs.reduce((a,b)=>a+b,0)/(xs.length || 1);\n summary[name] = {\n games: n,\n wins: rows.filter(r=>r.won).length,\n ties: rows.filter(r=>r.tied).length,\n losses: rows.filter(r=>!r.won && !r.tied).length,\n winRate: rows.filter(r=>r.won).length/n,\n avgBank: avg(rows.map(r=>r.bank)),\n medianBank: n % 2 ? rows[Math.floor(n/2)].bank : (rows[n/2-1].bank+rows[n/2].bank)/2,\n minBank: rows[0]?.bank ?? null,\n maxBank: rows[n-1]?.bank ?? null,\n avgMoveRate: avg(rows.map(r=>r.stat.moveRate)),\n avgPassRate: avg(rows.map(r=>r.stat.passRate)),\n avgWorkRate: avg(rows.map(r=>r.stat.workRate)),\n runtimeErrors: rows.reduce((a,r)=>a+r.stat.runtimeErrors,0),\n };\n}\n\nconsole.log(JSON.stringify({\n status: \"PASS\",\n evidenceState: \"OFFICIAL_TYPESCRIPT_PORT_TESTED_RNG_NOT_PYTHON_BIT_PARITY\",\n seeds: SEEDS,\n pairingsPerSeed: PAIRS,\n gameCount: games.length,\n summary,\n games,\n caveat: \"Official Kaggle playable TypeScript interpreter is a port of kaggriculture.py, but its RNG seeding documents non-bit-identical stochastic parity with CPython. Treat these as strong mechanics/control tests, not exact Python replay parity.\",\n}, null, 2));\n",[],[794],"7c65030b-2ecd-4fb2-a11c-d4404cbe0b9b","kaggle_rita_capital_logic_exact.ts","3eefcc44-aa41-11f1-9650-1607ee4eb77e",66,["D",1788733263058],"import { unzipSync,strFromU8 } from 'npm:fflate';const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const s=strFromU8(z['rancher_rita.py']).split('\\n');console.log(s.slice(365,441).map((x,i)=>`${366+i}|${x}`).join('\\n'));",[],[802],"0dd746a7-6361-487a-925b-93a535daf6bf","kaggle_rita_animal_logic_probe.ts","2b81efa2-aa41-11f1-9bce-1607ee4eb77e",64,["D",1788733230456],"import { unzipSync,strFromU8 } from 'npm:fflate';\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!r.ok)throw new Error(String(r.status));const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const s=strFromU8(z['rancher_rita.py']);const lines=s.split('\\n');const hits:number[]=[];for(let i=0;i({line:i+1,text:lines.slice(Math.max(0,i-15),Math.min(lines.length,i+25)).map((x,j)=>`${Math.max(0,i-15)+j+1}|${x}`).join('\\n')}));console.log(JSON.stringify({status:'PASS',hits:hits.map(x=>x+1),ranges},null,2));",[],[810],"a1a50443-17e8-47e6-a3d4-e81d6f04fc68","kaggle_rita_animal_logic_exact.ts","3429dfde-aa41-11f1-ab5f-1607ee4eb77e",65,["D",1788733244974],"import { unzipSync,strFromU8 } from 'npm:fflate';const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const s=strFromU8(z['rancher_rita.py']).split('\\n');console.log(s.slice(440,485).map((x,i)=>`${441+i}|${x}`).join('\\n'));",[],[818],"66ba8c1f-293b-40e4-8438-0756b2c94ff9","kaggle_reference_policy_only.ts","922827d8-aa39-11f1-9c43-1607ee4eb77e",44,["D",1788729966698],"import { unzipSync,strFromU8 } from 'npm:fflate';\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!r.ok)throw new Error(String(r.status));const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const out:any={};for(const n of ['rotation_rosa.py','homestead_hana.py','melon_mateo.py','rancher_rita.py']){const s=strFromU8(z[n]);const a=s.indexOf('POLICY = '),b=s.indexOf('\\n\\n\\n#',a);out[n]=s.slice(a,b>0?b:a+5000);}console.log(JSON.stringify({status:'PASS',out},null,2));",[],[826],"f0e4e78d-ebca-42cc-9346-d227f7bd39c4","kaggle_reference_policy_extract.ts","80e39002-aa39-11f1-82f3-1607ee4eb77e",43,["D",1788729937723],"import { unzipSync, strFromU8 } from \"npm:fflate\";\nconst res=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!res.ok)throw new Error(String(res.status));const files=unzipSync(new Uint8Array(await res.arrayBuffer()));\nconst out:any={};for(const n of ['rotation_rosa.py','homestead_hana.py','melon_mateo.py','rancher_rita.py']){const src=strFromU8(files[n]);const lines=src.split(/\\r?\\n/);out[n]={head:lines.slice(0,180).join('\\n'),market:lines.slice(Math.max(0,lines.findIndex(l=>l.includes('def _plan_market'))),Math.max(0,lines.findIndex(l=>l.includes('def _plan_market')))+150).join('\\n')};}\nconsole.log(JSON.stringify({status:'PASS',out},null,2));",[],[834],"59fd3792-a66e-4c9b-8854-480075891c46","kaggle_reference_agents_audit.ts","4bf767de-aa38-11f1-a4ac-1607ee4eb77e",29,["D",1788729419450],"import { unzipSync, strFromU8 } from \"npm:fflate\";\n\nconst slug=\"raykkretzschmar/kaggriculture-reference-agents\";\nconst [zipRes,metaRes]=await Promise.all([\n fetch(`https://www.kaggle.com/api/v1/datasets/download/${slug}`),\n fetch(`https://www.kaggle.com/api/v1/datasets/view/${slug}`),\n]);\nif(!zipRes.ok) throw new Error(`zip ${zipRes.status}`);\nif(!metaRes.ok) throw new Error(`meta ${metaRes.status}`);\nconst meta=await metaRes.json();\nconst files=unzipSync(new Uint8Array(await zipRes.arrayBuffer()));\nconst names=Object.keys(files).sort();\nconst text=(name:string,max=6000)=>files[name]?strFromU8(files[name]).slice(0,max):null;\nconst py=names.filter(n=>n.endsWith('.py'));\nconst hashes:any={};\nfor(const n of py){const d=await crypto.subtle.digest('SHA-256',files[n]);hashes[n]=Array.from(new Uint8Array(d)).map(x=>x.toString(16).padStart(2,'0')).join('');}\nconsole.log(JSON.stringify({\n status:\"PASS\",\n metadata:{title:meta.title,subtitle:meta.subtitle,licenseName:meta.licenseName,ownerName:meta.ownerName,lastUpdated:meta.lastUpdated,datasetId:meta.id,versionNumber:meta.currentVersionNumber},\n zipBytes:Number(zipRes.headers.get('content-length')??0),\n fileCount:names.length,\n names,\n pythonHashes:hashes,\n manifest:text('agents_manifest.csv',10000),\n baselineLeague:text('baseline_league.csv',5000),\n readme:names.find(n=>/readme/i.test(n))?text(names.find(n=>/readme/i.test(n))!,10000):null,\n sampleHeaders:Object.fromEntries(py.map(n=>[n,text(n,1200)])),\n},null,2));",[],[842],"4546adee-4395-4e0d-9c83-bb1fbafd1aef","kaggle_reference_agent_structure.ts","5a9de1d2-aa38-11f1-a4ef-1607ee4eb77e",30,["D",1788729444028],"import { unzipSync, strFromU8 } from \"npm:fflate\";\nconst res=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!res.ok)throw new Error(String(res.status));const files=unzipSync(new Uint8Array(await res.arrayBuffer()));\nconst targets=['melon_mateo.py','rancher_rita.py','broker_bea.py','ledger_lena.py','slotter_silas.py','closer_cleo.py'];\nconst out:any={};\nfor(const name of targets){const src=strFromU8(files[name]);const lines=src.split(/\\r?\\n/);out[name]={bytes:files[name].length,lines:lines.length,defs:lines.filter(l=>/^def\\s+/.test(l)).map(l=>l.trim()),classes:lines.filter(l=>/^class\\s+/.test(l)).map(l=>l.trim()),constants:lines.filter(l=>/^[A-Z][A-Z0-9_]{2,}\\s*=/.test(l)).slice(0,50),comments:lines.filter(l=>/^#/.test(l)).slice(0,80)};}\nconsole.log(JSON.stringify({status:'PASS',out},null,2));",[],[850],"b586b96b-9e11-4e44-b5bf-043c3f4e453f","kaggle_public_dataset_probe.ts","3cb189d0-aa38-11f1-a244-1607ee4eb77e",28,["D",1788729393822],"const urls = [\n \"https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents\",\n \"https://www.kaggle.com/api/v1/datasets/download/dariushafshar/kaggriculture-engine-reference-tables\",\n];\nconst out=[];\nfor(const url of urls){\n const res=await fetch(url,{redirect:\"follow\"});\n const buf=await res.arrayBuffer();\n out.push({url,status:res.status,ok:res.ok,contentType:res.headers.get(\"content-type\"),contentLength:res.headers.get(\"content-length\"),finalUrl:res.url,bytes:buf.byteLength,head:Array.from(new Uint8Array(buf.slice(0,8)))});\n}\nconsole.log(JSON.stringify({status:out.every(x=>x.ok)?\"PASS\":\"PARTIAL\",out},null,2));",[],[858],"abac1049-43ae-43c3-b266-a6bc36ba6803","kag_world_regime_diag_v1.ts","9dc18f04-aa42-11f1-894c-1607ee4eb77e",81,["D",1788733851634],"import { loadOfficialKaggricultureEngine } from './lib/kaggriculture_engine_loader.ts';\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();\nconst seeds=[9403,9419,9431,9461,9491];const PASS={farmer:['PASS'],hands:[],market:[]};\nconst items=['WHEAT','CARROT','TOMATO','STRAWBERRY','MELON','EGG','MILK','WOOL'];const out:any[]=[];\nfor(const seed of seeds){let s=E.state.initGameState(2,{...cfg0,seed},seed);const unlocks:any[]=[];let last=0;while(!s.done){if(s.town.unlocked_shops.length!==last){unlocks.push({day:s.day,hour:s.hour,new:s.town.unlocked_shops.slice(last),all:[...s.town.unlocked_shops]});last=s.town.unlocked_shops.length;}s=E.interpreter.step(s,[PASS,PASS],{...cfg0,seed});}const counts:any={};for(const sh of s.town.unlocked_shops)counts[sh]=(counts[sh]??0)+1;const final:any={};for(const p of items)final[p]={demand:10000-s.market.inventory[p],price:s.market.prices[p]};out.push({seed,counts,unlocks,final});}\nconsole.log(JSON.stringify({status:'PASS',worlds:out},null,2));",[],[866],"5f7f768c-e8c6-435b-8825-8cd6c7e2f9f9","kag_wheat_market_curve_probe_v1.ts","6717a64a-aa5c-11f1-9e9a-1607ee4eb77e",132,["D",1788744926837],"import { unzipSync,strFromU8 } from 'npm:fflate';\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!r.ok)throw new Error(String(r.status));const z=unzipSync(new Uint8Array(await r.arrayBuffer()));for(const f of ['price_curves.csv','crop_economics.csv'])console.log(`---${f}---\\n`+strFromU8(z[f]).split('\\n').filter(x=>/WHEAT|product|item|name/i.test(x)).slice(0,20).join('\\n'));",[],[874],"3e488cd7-7742-499d-a2e2-5688c77b8853","kag_wait3_heldout_v1.ts","01a07933-a3c2-71ca-b74d-29cb9897dcd3",93,["D",1788739831777],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['BASE','WAIT3'];const seeds=[12007,12113,12227,12347,12457,12569,12671,12781,12893,13001,13109,13217,13327,13441,13553,13669,13781,13883,13997,14107];const opp='RITA_BASE';\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();const zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await zr.arrayBuffer()));const src=strFromU8(z['rancher_rita.py']);const py=await loadPyodide();py.runPython(`import json,copy\\n_REG={}\\n_BASE={}`);py.globals.set('src',src);for(const [n,p] of Object.entries({DYN:{animal_target:{COW:0,SHEEP:0}},RITA_BASE:{}})){py.globals.set('nm',n);py.globals.set('patch',JSON.stringify(p));py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n\\n_BASE[nm]=copy.deepcopy(_n['POLICY'])`);}py.runPython(`\\ndef call(name,payload): return json.dumps(_REG[name]['agent'](json.loads(payload)))\\ndef set_herd(name,cows,sheep,order):\\n _REG[name]['POLICY']['animal_target']={'COW':int(cows),'SHEEP':int(sheep)}\\n _REG[name]['POLICY']['animals']=json.loads(order)\\ndef reset(name): _REG[name]['POLICY'].clear();_REG[name]['POLICY'].update(copy.deepcopy(_BASE[name]))\\n`);const call=py.globals.get('call'),setHerd=py.globals.get('set_herd'),reset=py.globals.get('reset');\nfunction owned(obs:any,a:string){const me=+obs.player,f=obs.farms[me];let n=+(obs.private?.shed?.[a]??0);for(const inv of obs.private?.inventories??[])n+=+(inv?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n;}\nfunction enforce(obs:any,act:any,t:any){const x:any={COW:owned(obs,'COW'),SHEEP:owned(obs,'SHEEP')};const o={...act,market:[] as any[]};for(const m of act?.market??[]){if(Array.isArray(m)&&m[0]==='BUY_ANIMAL'&&(m[1]==='COW'||m[1]==='SHEEP')){const a=m[1],q=Math.max(0,Math.min(+m[2]||1,t[a]-x[a]));if(q>0){o.market.push(['BUY_ANIMAL',a,q]);x[a]+=q;}}else o.market.push(m);}return o;}\nfunction initialGoal(v:string){if(v==='COMMIT_COW2')return{COW:2,SHEEP:0};if(v==='COMMIT_SHEEP2')return{COW:0,SHEEP:2};if(v==='COMMIT_MIX11')return{COW:1,SHEEP:1};return null;}\nasync function play(v:string,seed:number,seat:0|1){reset('DYN');reset(opp);let s=E.state.initGameState(2,{...cfg0,seed},seed),targets:any={COW:0,SHEEP:0},errors=0;const goal=initialGoal(v);let commitmentDone=!goal;const diary:any[]=[];while(!s.done){const acts:any[]=[];for(const pid of [0,1] as const){const n=pid===seat?'DYN':opp,obs=observationFor(s,pid);if(n==='DYN'&&+obs.hour===0){const own={COW:owned(obs,'COW'),SHEEP:owned(obs,'SHEEP')};targets={...own};if(goal&&!commitmentDone){commitmentDone=own.COW>=goal.COW&&own.SHEEP>=goal.SHEEP;if(!commitmentDone){targets={COW:Math.max(own.COW,goal.COW),SHEEP:Math.max(own.SHEEP,goal.SHEEP)};setHerd('DYN',targets.COW,targets.SHEEP,JSON.stringify(goal.SHEEP>0?['SHEEP','COW']:['COW','SHEEP']));}}\nif(commitmentDone){if(v==='WAIT3'&&+obs.day<3){setHerd('DYN',own.COW,own.SHEEP,JSON.stringify(['COW','SHEEP']));targets={...own};}else{const total=own.COW+own.SHEEP,c=choosePastureAnimal(obs,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)targets[c.choice]+=Math.min(2,16-total);setHerd('DYN',targets.COW,targets.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')));}}\nif([0,1,2,3,6,9,12,18].includes(+obs.day))diary.push({day:+obs.day,cash:obs.farms[pid].money,own:{...own},targets:{...targets},commitmentDone,shops:[...(obs.town?.unlocked_shops??[])]});}\ntry{acts[pid]=JSON.parse(String(call(n,JSON.stringify(obs))));if(n==='DYN')acts[pid]=enforce(obs,acts[pid],targets);}catch{errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]};}}s=E.interpreter.step(s,acts,{...cfg0,seed});}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,errors,diary};}\nconst games:any[]=[];for(const v of variants)for(const seed of seeds){games.push(await play(v,seed,0));games.push(await play(v,seed,1));}const summary:any={};for(const v of variants){const rs=games.filter(g=>g.v===v);summary[v]={episodes:rs.length,wins:rs.filter(r=>r.win).length,winRate:rs.filter(r=>r.win).length/rs.length,avgBank:rs.reduce((a,r)=>a+r.ours,0)/rs.length,avgMargin:rs.reduce((a,r)=>a+r.margin,0)/rs.length,bySeed:Object.fromEntries(seeds.map(seed=>{const x=rs.filter(r=>r.seed===seed);return[seed,{wins:x.filter(r=>r.win).length,avgMargin:x.reduce((a,r)=>a+r.margin,0)/x.length,avgBank:x.reduce((a,r)=>a+r.ours,0)/x.length}]}))};}const b=games.filter(g=>g.v==='BASE');const paired=variants.slice(1).map(v=>{const r=games.filter(g=>g.v===v);return{variant:v,avgMarginDelta:r.reduce((s,g,i)=>s+(g.margin-b[i].margin),0)/r.length,better:r.filter((g,i)=>g.margin>b[i].margin).length,worse:r.filter((g,i)=>g.marging.seed===9491&&g.seat===0).map(g=>({v:g.v,diary:g.diary}))},null,2));",[],[882],"9ebf1b0f-94fc-436f-9919-6f69f4108bb3","kag_voc_gate_conditional_v1.ts","01a07a2d-4545-7101-af5d-38f2f34b084b",275,["D",1788756199933],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES];\nconst cfg=E.state.resolveConfig({seed:949494,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst decisionDay=9,modes=['N','H','W','B'],MAXN=16;\nconst actual=['PET_CAFE','FARMERS_MARKET','SMOOTHIE_SHOP','BAKERY','PIZZA_SHOP','SMOOTHIE_SHOP','PET_CAFE','PIZZA_SHOP'];\nconst prefix=actual.slice(0,3);\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}\nconst scenarios:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction compact(o:any){const me=+o.player,f=o.farms[me],g=o.farms[1-me];return{day:+o.day,cash:+f.money,oppCash:+g.money,shops:o.town.unlocked_shops}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,949494),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=compact(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst OUTER=8,CASES=[{id:'YBY',prefix:['YARN_STORE','BRUNCH_SPOT','YARN_STORE'],action:'W'},{id:'BIB',prefix:['BAKERY','ICE_CREAM_SHOP','BAKERY'],action:'H'}];\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))};const result:any={};for(let ci=0;cir.delta),mu=avg(ds),se=sd(ds)/Math.sqrt(ds.length);result[c.id]={prefix:c.prefix,action:c.action,meanDelta:mu,se,lo95:mu-1.96*se,hi95:mu+1.96*se,better:ds.filter(x=>x>0).length,worse:ds.filter(x=>x<0).length,min:Math.min(...ds),max:Math.max(...ds),rows}}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_VOC_GATE_CONDITIONAL_FALSIFICATION',outerPerPrefix:OUTER,result,interpretation:'independent suffixes conditioned on each prefix that produced a confidence-gated intervention; tests whether realized losses were residual risk or systematically wrong expected-value decisions'},null,2));",[],[890],"ee4a90bc-b63a-40c2-a701-8f4a8aa14378","kag_voc_bib_stability_v1.ts","01a07a2e-577f-709d-acb8-8c7cddd36a31",277,["D",1788756260730],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES];\nconst cfg=E.state.resolveConfig({seed:949494,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst decisionDay=9,modes=['N','H','W','B'],MAXN=16;\nconst actual=['BAKERY','ICE_CREAM_SHOP','BAKERY','PET_CAFE','BRUNCH_SPOT','PET_CAFE','FARMERS_MARKET','PIZZA_SHOP'];\nconst prefix=actual.slice(0,3);\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}\nconst scenarios:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction compact(o:any){const me=+o.player,f=o.farms[me],g=o.farms[1-me];return{day:+o.day,cash:+f.money,oppCash:+g.money,shops:o.town.unlocked_shops}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,949494),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=compact(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst vals:any={};const preSet=new Set();for(const m of modes){vals[m]=[];for(const sc of scenarios){const r=await play(sc,m);vals[m].push(r.margin);preSet.add(JSON.stringify(r.pre))}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},sizes=[2,4,6,8,12,16];\nconst stages:any[]=[];for(const n of sizes){const stats:any={};for(const m of modes){const a=vals[m].slice(0,n);stats[m]={mean:avg(a)}}for(const m of modes.filter(m=>m!=='N')){const d=vals[m].slice(0,n).map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=sd(d)/Math.sqrt(n);stats[m].deltaVsN={mean:mu,se,z:se?mu/se:null,lo95:mu-1.96*se,hi95:mu+1.96*se}}const pick=[...modes].sort((a,b)=>stats[b].mean-stats[a].mean)[0],ordered=[...modes].sort((a,b)=>stats[b].mean-stats[a].mean),gap=stats[ordered[0]].mean-stats[ordered[1]].mean;stages.push({n,pick,gap,stats})}\nconst actualRuns:any={};for(const m of modes)actualRuns[m]=await play(actual,m);const oracle=[...modes].sort((a,b)=>actualRuns[b].margin-actualRuns[a].margin)[0],base=actualRuns.N.margin;\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_VALUE_OF_COMPUTATION_FALSE_POSITIVE_AUDIT',prefix,hiddenActualSuffix:actual.slice(3),preStateIdentical:preSet.size===1,scenarioCount:MAXN,stages,actual:{oracle,deltas:Object.fromEntries(modes.map(m=>[m,actualRuns[m].margin-base]))},interpretation:'paired scenario estimates; continue compute when top-action advantage remains uncertain relative to paired standard error'},null,2));",[],[898],"3775471c-f040-4af8-a977-3d52a5477fa4","kag_top_meta_dossier_v1.ts","fd08b188-aa54-11f1-8ec9-1607ee4eb77e",128,["D",1788744653638],14,"import { sqlite } from 'https://esm.town/v/std/sqlite/main.ts';\nimport { buildBattleDossier,mirrorTimeline } from './lib/kaggriculture_battle_dossier_v1.ts';\nimport { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst ref='broker_bea.py',seeds=[17011,17123,17239,17351];\nconst E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));\nconst py=await loadPyodide();py.runPython(`import json\\n_REG={}`);py.globals.set('src',strFromU8(z[ref]));py.runPython(`_r={}\\nexec(src,_r)\\n_REG['REF']=_r['agent']`);py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\\ndef cr(p):return json.dumps(_REG['REF'](json.loads(p)))\\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\\ndef reset():\\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0}\\ndef seth(c,s,o):\\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)}\\n _D['POLICY']['animals']=json.loads(o)\\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),reset=py.globals.get('reset'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}\nfunction enforce(o:any,act:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},y={...act,market:[] as any[]};for(const m of act?.market??[]){if(Array.isArray(m)&&m[0]==='BUY_ANIMAL'&&(m[1]==='COW'||m[1]==='SHEEP')){const a=m[1],q=Math.max(0,Math.min(+m[2]||1,t[a]-x[a]));if(q){y.market.push(['BUY_ANIMAL',a,q]);x[a]+=q}}else y.market.push(m)}return y}\nfunction profile(o:any,pid:number){const f=o.farms[pid],animals:any={COW:0,SHEEP:0,GOOSE:0},crops:any={};for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.animal)animals[t.animal]=(animals[t.animal]??0)+1;if(t.crop)crops[t.crop]=(crops[t.crop]??0)+1;if(t.plant)crops[t.plant]=(crops[t.plant]??0)+1;if(t.kind==='PLANT'&&t.type)crops[t.type]=(crops[t.type]??0)+1}return{money:+f.money,hands:(f.hands??[]).length,animals,crops,shedFertilizer:+(o.private?.shed?.FERTILIZER??0),shedMilk:+(o.private?.shed?.MILK??0),shedWool:+(o.private?.shed?.WOOL??0),shedWheat:+(o.private?.shed?.WHEAT??0),carriedWheat:(o.private?.inventories??[]).reduce((a:number,i:any)=>a+Number(i?.WHEAT??0),0)}}\nfunction emptyTel(){return{hire:0,land:0,buyAnimal:{} as any,buySeed:{} as any,buyProduct:{} as any,sell:{} as any,sellExecutable:{} as any,first:{} as any,ops:{} as any}}\nfunction add(t:any,act:any,day:number,obs:any){for(const u of [act?.farmer,...(act?.hands??[])]){const op=Array.isArray(u)?u[0]:null;if(op)t.ops[op]=(t.ops[op]??0)+1}const available:any={...(obs?.private?.shed??{})};for(const m of act?.market??[]){const k=m?.[0];if(k==='HIRE')t.hire++;else if(k==='BUY_LAND'){t.land++;t.first.land??=day}else if(k==='BUY_ANIMAL'){t.buyAnimal[m[1]]=(t.buyAnimal[m[1]]??0)+Number(m[2]??1);t.first[`animal_${m[1]}`]??=day}else if(k==='BUY_SEED'){t.buySeed[m[1]]=(t.buySeed[m[1]]??0)+Number(m[2]??0);t.first[`seed_${m[1]}`]??=day}else if(k==='BUY'||k==='BUY_PRODUCT'){t.buyProduct[m[1]]=(t.buyProduct[m[1]]??0)+Number(m[2]??0)}else if(k==='SELL'){const item=m[1],q=Math.max(0,Number(m[2]??0));t.sell[item]=(t.sell[item]??0)+q;t.first[`sell_${item}`]??=day;const exe=Math.max(0,Math.min(q,Number(available[item]??0)));t.sellExecutable[item]=(t.sellExecutable[item]??0)+exe;available[item]=Math.max(0,Number(available[item]??0)-exe)}}}\nasync function play(seed:number,ourSeat:0|1){reset();let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},done=false,errors=0;const tel=[emptyTel(),emptyTel()],timeline:any[]=[];while(!s.done){if(s.hour===0){const o0=observationFor(s,0),o1=observationFor(s,1);timeline.push({day:s.day,ours:profile(ourSeat===0?o0:o1,ourSeat),ref:profile(ourSeat===0?o1:o0,1-ourSeat),market:{MILK:{p:s.market.prices.MILK,i:s.market.inventory.MILK},WOOL:{p:s.market.prices.WOOL,i:s.market.inventory.WOOL},WHEAT:{p:s.market.prices.WHEAT,i:s.market.inventory.WHEAT},STRAWBERRY:{p:s.market.prices.STRAWBERRY,i:s.market.inventory.STRAWBERRY}},shops:[...(s.town?.unlocked_shops??[])]})}const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===ourSeat){if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!done){done=ow.COW>=1&&ow.SHEEP>=1;if(!done){tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)};seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}}if(done){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')))}}aa[pid]=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg)}else aa[pid]=JSON.parse(String(cr(JSON.stringify(o))))}catch{errors++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}add(tel[pid],aa[pid],s.day,o)}s=E.interpreter.step(s,aa,{...cfg0,seed})}const ours=s.farms[ourSeat].money,theirs=s.farms[1-ourSeat].money;return{seed,ourSeat,ours,theirs,margin:ours-theirs,win:ours>theirs,errors,ourTel:tel[ourSeat],refTel:tel[1-ourSeat],timeline}}\nconst games:any[]=[];for(const seed of seeds){games.push(await play(seed,0));games.push(await play(seed,1))}\nconst avg=(xs:number[])=>xs.reduce((a,b)=>a+b,0)/(xs.length||1);const days=[0,3,6,9,12,15,18,21,24,27,29];const avgTimeline=days.map(day=>{const rows=games.map(g=>g.timeline.find((x:any)=>x.day===day)).filter(Boolean);return{day,ourBank:avg(rows.map((x:any)=>x.ours.money)),refBank:avg(rows.map((x:any)=>x.ref.money)),ourHands:avg(rows.map((x:any)=>x.ours.hands)),refHands:avg(rows.map((x:any)=>x.ref.hands)),ourCow:avg(rows.map((x:any)=>x.ours.animals.COW??0)),ourSheep:avg(rows.map((x:any)=>x.ours.animals.SHEEP??0)),refCow:avg(rows.map((x:any)=>x.ref.animals.COW??0)),refSheep:avg(rows.map((x:any)=>x.ref.animals.SHEEP??0)),milkPrice:avg(rows.map((x:any)=>x.market.MILK.p)),woolPrice:avg(rows.map((x:any)=>x.market.WOOL.p)),strawberryPrice:avg(rows.map((x:any)=>x.market.STRAWBERRY.p))}});\nfunction agg(side:'ourTel'|'refTel'){const r:any={hire:0,land:0,buyAnimal:{},buySeed:{},buyProduct:{},sell:{},sellExecutable:{},ops:{}};for(const g of games){const t=g[side];r.hire+=t.hire;r.land+=t.land;for(const k of ['buyAnimal','buySeed','buyProduct','sell','sellExecutable','ops'])for(const [p,q] of Object.entries(t[k]??{}))r[k][p]=(r[k][p]??0)+Number(q)}r.hire/=games.length;r.land/=games.length;for(const k of ['buyAnimal','buySeed','buyProduct','sell','sellExecutable','ops'])for(const p of Object.keys(r[k]))r[k][p]/=games.length;return r}\nconst now=new Date().toISOString();\nconst dossiers:any[]=[];\nfor(const g of games){const matchId=`MIX11-v-BROKER-${g.seed}-S${g.ourSeat}`;const ourTel={...g.ourTel,sell:g.ourTel.sellExecutable??{}};const refTel={...g.refTel,sell:g.refTel.sellExecutable??{}};const ours=buildBattleDossier({matchId,subjectId:'MIX11',opponentId:'BROKER_BEA',seed:g.seed,seat:g.ourSeat,finalBank:g.ours,opponentBank:g.theirs,timeline:mirrorTimeline(g.timeline,true),telemetry:ourTel,opponentTelemetry:refTel});const theirs=buildBattleDossier({matchId,subjectId:'BROKER_BEA',opponentId:'MIX11',seed:g.seed,seat:1-g.ourSeat,finalBank:g.theirs,opponentBank:g.ours,timeline:mirrorTimeline(g.timeline,false),telemetry:refTel,opponentTelemetry:ourTel});dossiers.push(ours,theirs)}\nfor(const d of dossiers){await sqlite.execute({sql:`INSERT OR REPLACE INTO kag_battle_dossiers(dossier_id,match_id,subject_id,opponent_id,subject_role,seed,seat,final_bank,opponent_bank,margin,min_cash,min_cash_day,first_lead_day,first_trail_day,max_lead,max_lead_day,max_deficit,max_deficit_day,investment_fingerprint,revenue_mix,market_exposure,observed_strengths,observed_failures,causal_hypotheses,counterfactuals,evidence_state,created_at) VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)`,args:[d.dossier_id,d.match_id,d.subject_id,d.opponent_id,d.subject_role,d.seed,d.seat,d.final_bank,d.opponent_bank,d.margin,d.min_cash,d.min_cash_day,d.first_lead_day,d.first_trail_day,d.max_lead,d.max_lead_day,d.max_deficit,d.max_deficit_day,JSON.stringify(d.investment_fingerprint),JSON.stringify(d.revenue_mix),JSON.stringify(d.market_exposure),JSON.stringify(d.observed_strengths),JSON.stringify(d.observed_failures),JSON.stringify(d.causal_hypotheses),JSON.stringify(d.counterfactuals),d.evidence_state,now]})}\nconst dossierCounts=await sqlite.execute(`SELECT subject_id,subject_role,COUNT(*) n FROM kag_battle_dossiers WHERE match_id LIKE 'MIX11-v-BROKER-%' GROUP BY subject_id,subject_role ORDER BY subject_id,subject_role`);\n\nconst feedTrail=games.find(g=>g.seed===17011&&g.ourSeat===0)?.timeline.map((x:any)=>{const oc=x.ours.animals.COW+x.ours.animals.SHEEP,rc=x.ref.animals.COW+x.ref.animals.SHEEP,wp=x.market.WHEAT.p;const ow=x.ours.shedWheat+x.ours.carriedWheat,rw=x.ref.shedWheat+x.ref.carriedWheat;return{day:x.day,wheatPrice:wp,ours:{flock:oc,wheatOwned:ow,daysOnHand:oc?ow/oc:null,nominal16dCashReserve:oc*wp*16,stockValueAtSpot:ow*wp,netNominalReserveAfterOwnedStock:Math.max(0,oc*wp*16-ow*wp)},ref:{flock:rc,wheatOwned:rw,daysOnHand:rc?rw/rc:null,nominal16dCashReserve:rc*wp*16,stockValueAtSpot:rw*wp}}});\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'BLACK_BOX_TOP_REFERENCE_DOSSIER_NO_CODE_REUSE',opponent:ref,seeds,summary:{episodes:games.length,wins:games.filter(g=>g.win).length,winRate:games.filter(g=>g.win).length/games.length,ourBank:avg(games.map(g=>g.ours)),refBank:avg(games.map(g=>g.theirs)),avgMargin:avg(games.map(g=>g.margin)),errors:games.reduce((a,g)=>a+g.errors,0)},avgTimeline,ourEconomy:agg('ourTel'),winnerEconomy:agg('refTel'),fertilizerTrail:games.find(g=>g.seed===17011&&g.ourSeat===0)?.timeline.map((x:any)=>({day:x.day,ours:x.ours.shedFertilizer,ref:x.ref.shedFertilizer})),feedTrail,sampleLoss:games.sort((a,b)=>a.margin-b.margin)[0]},null,2));",[],[906],"fdb1efd9-9023-4293-ad52-7b416feaf9ca","kag_teacher_q_racing_positive_control_v1.ts","01a07a68-080a-7078-9d7e-f7b54e444216",314,["D",1788760055904],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=1;const cfg=E.state.resolveConfig({seed:959595,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(_ci:number){return ['ICE_CREAM_SHOP','SMOOTHIE_SHOP','PIZZA_SHOP']}function innerWorlds(prefix:string[],ci:number){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,959595),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??1.96;rows.push({action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0);contexts.push({contextId:`BROKER-D9-RACE-${prefix.join('_')}-${ci}`,opponent:'broker_bea.py',decisionDay,prefix,preStateIdentical:preSame,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes:INNER*modes.length,savingsPct:100*(1-episodesUsed/(INNER*modes.length)),rows})}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'TEACHER_Q_RACING_BATCH1',maxSampleN:INNER,contexts},null,2));",[],[915],"bf005146-05d2-4e3b-a583-622070f0eabe","kag_teacher_q_racing_batch1_v1.ts","01a07a4d-0051-738a-b2d4-feb74e51b624",306,["D",1788759869919],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=1;const cfg=E.state.resolveConfig({seed:969696,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number){const r=rng(1717000000+ci*104729);return Array.from({length:3},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,969696),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??1.96;rows.push({action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0);contexts.push({contextId:`BROKER-D9-RACE-${prefix.join('_')}-${ci}`,opponent:'broker_bea.py',decisionDay,prefix,preStateIdentical:preSame,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes:INNER*modes.length,savingsPct:100*(1-episodesUsed/(INNER*modes.length)),rows})}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'TEACHER_Q_RACING_BATCH1',maxSampleN:INNER,contexts},null,2));",[],[923],"ed1af90e-37a5-4dc1-a549-7456d36540ac","kag_teacher_q_persistent_silas_d9_v1.ts","01a07aa3-5bae-738a-87db-17c40a7916e8",451,["D",1788763984919],6,"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=1;const cfg=E.state.resolveConfig({seed:999299,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number,attempt=0){const r=rng(4117000000+ci*104729+attempt*49979687);return Array.from({length:3},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number,attempt=0){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,999299),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst existingQ=await sqlite.execute(\"SELECT DISTINCT prefix_json FROM kag_teacher_q WHERE opponent='slotter_silas.py' AND decision_day=9\"),seen=new Set((existingQ.rows??[]).map((r:any)=>String(r.prefix_json)));const contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??1.96;rows.push({action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0);contexts.push({contextId:`SILAS-D9-RACE-${prefix.join('_')}`,opponent:'slotter_silas.py',decisionDay,prefix,prefixAttempt:attempt-1,preStateIdentical:preSame,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes:INNER*modes.length,savingsPct:100*(1-episodesUsed/(INNER*modes.length)),rows})}\nconst sourceArtifact='kag_teacher_q_persistent_silas_d9_v1.ts',evidenceState='TEACHER_Q_RACING_PERSISTENT_SILAS_D9_V1';\nconst before=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconst statements:any[]=[];for(const c of contexts){statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_contexts (context_id,opponent,decision_day,prefix_json,feature_json,teacher_pick,teacher_scenarios_used,pre_state_identical,race_audit_json,episodes_used,full_episodes,savings_pct,evidence_state,source_artifact) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),c.teacherPick,c.teacherScenariosUsed,c.preStateIdentical?1:0,JSON.stringify(c.raceAudit),c.episodesUsed,c.fullEpisodes,c.savingsPct,evidenceState,sourceArtifact]});for(const r of c.rows)statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_q (context_id,opponent,decision_day,prefix_json,feature_json,action,q_mean,delta_vs_n,q_sd,lcb95,cvar50,positive_rate,sample_n,teacher_pick,evidence_state,source_artifact,ucb95,pruned) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),r.action,r.qMean,r.deltaVsN,r.qSd,r.lcb95,r.cvar50,r.positiveRate,r.sampleN,c.teacherPick,evidenceState,sourceArtifact,r.ucb95,r.pruned?1:0]})}\nawait sqlite.batch(statements);\nconst after=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconsole.log(JSON.stringify({status:'PASS',evidenceState,maxSampleN:INNER,persistence:{sourceArtifact,before:before.rows[0],attemptedRows:statements.length,after:after.rows[0]},contexts},null,2));",[],[931],"59e17f0b-043b-405a-b4ee-bfa574fb2687","kag_teacher_q_persistent_microbatch_v1.ts","01a07a78-e761-76ad-b095-5fd906282b24",354,["D",1788761921145],8,"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=1;const cfg=E.state.resolveConfig({seed:994949,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number,attempt=0){const r=rng(2917000000+ci*104729+attempt*49979687);return Array.from({length:3},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number,attempt=0){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,994949),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst existingQ=await sqlite.execute(\"SELECT DISTINCT prefix_json FROM kag_teacher_q WHERE opponent='broker_bea.py' AND decision_day=9\"),seen=new Set((existingQ.rows??[]).map((r:any)=>String(r.prefix_json)));const contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??1.96;rows.push({action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0);contexts.push({contextId:`BROKER-D9-RACE-${prefix.join('_')}`,opponent:'broker_bea.py',decisionDay,prefix,prefixAttempt:attempt-1,preStateIdentical:preSame,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes:INNER*modes.length,savingsPct:100*(1-episodesUsed/(INNER*modes.length)),rows})}\nconst sourceArtifact='kag_teacher_q_persistent_microbatch_v1.ts',evidenceState='TEACHER_Q_RACING_PERSISTENT_MICROBATCH_V1';\nconst before=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconst statements:any[]=[];for(const c of contexts){statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_contexts (context_id,opponent,decision_day,prefix_json,feature_json,teacher_pick,teacher_scenarios_used,pre_state_identical,race_audit_json,episodes_used,full_episodes,savings_pct,evidence_state,source_artifact) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),c.teacherPick,c.teacherScenariosUsed,c.preStateIdentical?1:0,JSON.stringify(c.raceAudit),c.episodesUsed,c.fullEpisodes,c.savingsPct,evidenceState,sourceArtifact]});for(const r of c.rows)statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_q (context_id,opponent,decision_day,prefix_json,feature_json,action,q_mean,delta_vs_n,q_sd,lcb95,cvar50,positive_rate,sample_n,teacher_pick,evidence_state,source_artifact,ucb95,pruned) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),r.action,r.qMean,r.deltaVsN,r.qSd,r.lcb95,r.cvar50,r.positiveRate,r.sampleN,c.teacherPick,evidenceState,sourceArtifact,r.ucb95,r.pruned?1:0]})}\nawait sqlite.batch(statements);\nconst after=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconsole.log(JSON.stringify({status:'PASS',evidenceState,maxSampleN:INNER,persistence:{sourceArtifact,before:before.rows[0],attemptedRows:statements.length,after:after.rows[0]},contexts},null,2));",[],[940],"524b8417-8a72-4f59-b33e-f0a2ba000955","kag_teacher_q_persistent_lena_d9_v4.ts","01a07aa1-865a-75bc-9c1a-4a57509b56cd",444,["D",1788763860463],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=1;const cfg=E.state.resolveConfig({seed:999199,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number,attempt=0){const r=rng(3917000000+ci*104729+attempt*49979687);return Array.from({length:3},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number,attempt=0){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,999199),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst existingQ=await sqlite.execute(\"SELECT DISTINCT prefix_json FROM kag_teacher_q WHERE opponent='ledger_lena.py' AND decision_day=9\"),seen=new Set((existingQ.rows??[]).map((r:any)=>String(r.prefix_json)));const contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??1.96;rows.push({action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0);contexts.push({contextId:`LENA-D9-RACE-${prefix.join('_')}`,opponent:'ledger_lena.py',decisionDay,prefix,prefixAttempt:attempt-1,preStateIdentical:preSame,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes:INNER*modes.length,savingsPct:100*(1-episodesUsed/(INNER*modes.length)),rows})}\nconst sourceArtifact='kag_teacher_q_persistent_lena_d9_v4.ts',evidenceState='TEACHER_Q_RACING_PERSISTENT_LENA_D9_V4';\nconst before=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconst statements:any[]=[];for(const c of contexts){statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_contexts (context_id,opponent,decision_day,prefix_json,feature_json,teacher_pick,teacher_scenarios_used,pre_state_identical,race_audit_json,episodes_used,full_episodes,savings_pct,evidence_state,source_artifact) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),c.teacherPick,c.teacherScenariosUsed,c.preStateIdentical?1:0,JSON.stringify(c.raceAudit),c.episodesUsed,c.fullEpisodes,c.savingsPct,evidenceState,sourceArtifact]});for(const r of c.rows)statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_q (context_id,opponent,decision_day,prefix_json,feature_json,action,q_mean,delta_vs_n,q_sd,lcb95,cvar50,positive_rate,sample_n,teacher_pick,evidence_state,source_artifact,ucb95,pruned) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),r.action,r.qMean,r.deltaVsN,r.qSd,r.lcb95,r.cvar50,r.positiveRate,r.sampleN,c.teacherPick,evidenceState,sourceArtifact,r.ucb95,r.pruned?1:0]})}\nawait sqlite.batch(statements);\nconst after=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconsole.log(JSON.stringify({status:'PASS',evidenceState,maxSampleN:INNER,persistence:{sourceArtifact,before:before.rows[0],attemptedRows:statements.length,after:after.rows[0]},contexts},null,2));",[],[949],"228612b9-2b3f-4493-9aca-61a6ab653754","kag_teacher_q_persistent_lena_d9_v2.ts","01a07a8a-61e4-77f9-80a8-25a858ed9c8f",381,["D",1788762358818],7,"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';import { extractTeacherFeaturesV2,KAG_TEACHER_FEATURE_SCHEMA } from './lib/kag_teacher_features_v2.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=1,OPPONENT='ledger_lena.py';const cfg=E.state.resolveConfig({seed:997979,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number,attempt=0){const r=rng(2917000000+ci*104729+attempt*49979687);return Array.from({length:3},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number,attempt=0){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){return extractTeacherFeaturesV2(o)}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,997979),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst existingQ=await sqlite.execute(`SELECT DISTINCT prefix_json FROM kag_teacher_q WHERE opponent='${OPPONENT}' AND decision_day=${decisionDay}`),seen=new Set((existingQ.rows??[]).map((r:any)=>String(r.prefix_json)));const contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??1.96;rows.push({action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0);contexts.push({contextId:`LENA-D9-RACE-${prefix.join('_')}`,opponent:OPPONENT,decisionDay,prefix,prefixAttempt:attempt-1,preStateIdentical:preSame,featureSchema:KAG_TEACHER_FEATURE_SCHEMA,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes:INNER*modes.length,savingsPct:100*(1-episodesUsed/(INNER*modes.length)),rows})}\nconst sourceArtifact='kag_teacher_q_persistent_lena_d9_v2.ts',evidenceState='TEACHER_Q_RACING_PERSISTENT_LENA_D9_V2';\nconst before=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconst statements:any[]=[];for(const c of contexts){statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_contexts (context_id,opponent,decision_day,prefix_json,feature_json,teacher_pick,teacher_scenarios_used,pre_state_identical,race_audit_json,episodes_used,full_episodes,savings_pct,evidence_state,source_artifact,feature_schema) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),c.teacherPick,c.teacherScenariosUsed,c.preStateIdentical?1:0,JSON.stringify(c.raceAudit),c.episodesUsed,c.fullEpisodes,c.savingsPct,evidenceState,sourceArtifact,c.featureSchema]});for(const r of c.rows)statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_q (context_id,opponent,decision_day,prefix_json,feature_json,action,q_mean,delta_vs_n,q_sd,lcb95,cvar50,positive_rate,sample_n,teacher_pick,evidence_state,source_artifact,ucb95,pruned,feature_schema) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),r.action,r.qMean,r.deltaVsN,r.qSd,r.lcb95,r.cvar50,r.positiveRate,r.sampleN,c.teacherPick,evidenceState,sourceArtifact,r.ucb95,r.pruned?1:0,c.featureSchema]})}\nawait sqlite.batch(statements);\nconst after=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconsole.log(JSON.stringify({status:'PASS',evidenceState,maxSampleN:INNER,persistence:{sourceArtifact,before:before.rows[0],attemptedRows:statements.length,after:after.rows[0]},contexts},null,2));",[],[957],"94b24cdc-92bb-46ea-bb5c-bbb01a8f510c","kag_teacher_q_persistent_generic_v2.ts","01a07a8d-5355-77fc-ba50-457e1f6cacdf",388,["D",1788762526330],4,"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';import { extractTeacherFeaturesV2,KAG_TEACHER_FEATURE_SCHEMA } from './lib/kag_teacher_features_v2.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=1,OPPONENT='slotter_silas.py',OPP_TAG='SILAS',ENGINE_SEED=998989;const cfg=E.state.resolveConfig({seed:ENGINE_SEED,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number,attempt=0){const r=rng(2917000000+ci*104729+attempt*49979687);return Array.from({length:Math.floor(decisionDay/3)},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number,attempt=0){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){return extractTeacherFeaturesV2(o)}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,ENGINE_SEED),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst existingQ=await sqlite.execute(`SELECT DISTINCT prefix_json FROM kag_teacher_q WHERE opponent='${OPPONENT}' AND decision_day=${decisionDay}`),seen=new Set((existingQ.rows??[]).map((r:any)=>String(r.prefix_json)));const contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??1.96;rows.push({action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0);contexts.push({contextId:`${OPP_TAG}-D${decisionDay}-RACE-${prefix.join('_')}`,opponent:OPPONENT,decisionDay,prefix,prefixAttempt:attempt-1,preStateIdentical:preSame,featureSchema:KAG_TEACHER_FEATURE_SCHEMA,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes:INNER*modes.length,savingsPct:100*(1-episodesUsed/(INNER*modes.length)),rows})}\nconst sourceArtifact='kag_teacher_q_persistent_lena_d9_v2.ts',evidenceState='TEACHER_Q_RACING_PERSISTENT_LENA_D9_V2';\nconst before=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconst statements:any[]=[];for(const c of contexts){statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_contexts (context_id,opponent,decision_day,prefix_json,feature_json,teacher_pick,teacher_scenarios_used,pre_state_identical,race_audit_json,episodes_used,full_episodes,savings_pct,evidence_state,source_artifact,feature_schema) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),c.teacherPick,c.teacherScenariosUsed,c.preStateIdentical?1:0,JSON.stringify(c.raceAudit),c.episodesUsed,c.fullEpisodes,c.savingsPct,evidenceState,sourceArtifact,c.featureSchema]});for(const r of c.rows)statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_q (context_id,opponent,decision_day,prefix_json,feature_json,action,q_mean,delta_vs_n,q_sd,lcb95,cvar50,positive_rate,sample_n,teacher_pick,evidence_state,source_artifact,ucb95,pruned,feature_schema) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),r.action,r.qMean,r.deltaVsN,r.qSd,r.lcb95,r.cvar50,r.positiveRate,r.sampleN,c.teacherPick,evidenceState,sourceArtifact,r.ucb95,r.pruned?1:0,c.featureSchema]})}\nawait sqlite.batch(statements);\nconst after=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconsole.log(JSON.stringify({status:'PASS',evidenceState,maxSampleN:INNER,persistence:{sourceArtifact,before:before.rows[0],attemptedRows:statements.length,after:after.rows[0]},contexts},null,2));",[],[966],"2d51132a-f177-41a4-b8c1-b426e8029ff3","kag_teacher_q_persistent_cleo_d9_v2.ts","01a07a9f-cd25-70df-8952-a8fae7e4fd52",437,["D",1788763747025],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=1;const cfg=E.state.resolveConfig({seed:998989,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number,attempt=0){const r=rng(3717000000+ci*104729+attempt*49979687);return Array.from({length:3},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number,attempt=0){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,998989),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst existingQ=await sqlite.execute(\"SELECT DISTINCT prefix_json FROM kag_teacher_q WHERE opponent='closer_cleo.py' AND decision_day=9\"),seen=new Set((existingQ.rows??[]).map((r:any)=>String(r.prefix_json)));const contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??1.96;rows.push({action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0);contexts.push({contextId:`CLEO-D9-RACE-${prefix.join('_')}`,opponent:'closer_cleo.py',decisionDay,prefix,prefixAttempt:attempt-1,preStateIdentical:preSame,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes:INNER*modes.length,savingsPct:100*(1-episodesUsed/(INNER*modes.length)),rows})}\nconst sourceArtifact='kag_teacher_q_persistent_cleo_d9_v2.ts',evidenceState='TEACHER_Q_RACING_PERSISTENT_CLEO_D9_V2';\nconst before=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconst statements:any[]=[];for(const c of contexts){statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_contexts (context_id,opponent,decision_day,prefix_json,feature_json,teacher_pick,teacher_scenarios_used,pre_state_identical,race_audit_json,episodes_used,full_episodes,savings_pct,evidence_state,source_artifact) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),c.teacherPick,c.teacherScenariosUsed,c.preStateIdentical?1:0,JSON.stringify(c.raceAudit),c.episodesUsed,c.fullEpisodes,c.savingsPct,evidenceState,sourceArtifact]});for(const r of c.rows)statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_q (context_id,opponent,decision_day,prefix_json,feature_json,action,q_mean,delta_vs_n,q_sd,lcb95,cvar50,positive_rate,sample_n,teacher_pick,evidence_state,source_artifact,ucb95,pruned) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),r.action,r.qMean,r.deltaVsN,r.qSd,r.lcb95,r.cvar50,r.positiveRate,r.sampleN,c.teacherPick,evidenceState,sourceArtifact,r.ucb95,r.pruned?1:0]})}\nawait sqlite.batch(statements);\nconst after=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconsole.log(JSON.stringify({status:'PASS',evidenceState,maxSampleN:INNER,persistence:{sourceArtifact,before:before.rows[0],attemptedRows:statements.length,after:after.rows[0]},contexts},null,2));",[],[975],"968fb1ea-9937-461c-926c-0cbf2ee96bdd","kag_teacher_q_persistent_cleo_d9_v1.ts","01a07a86-414d-719d-b023-34d8d3bff620",365,["D",1788762065718],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=1,OPPONENT='closer_cleo.py';const cfg=E.state.resolveConfig({seed:996969,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number,attempt=0){const r=rng(2917000000+ci*104729+attempt*49979687);return Array.from({length:3},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number,attempt=0){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,996969),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst existingQ=await sqlite.execute(`SELECT DISTINCT prefix_json FROM kag_teacher_q WHERE opponent='${OPPONENT}' AND decision_day=${decisionDay}`),seen=new Set((existingQ.rows??[]).map((r:any)=>String(r.prefix_json)));const contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??1.96;rows.push({action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0);contexts.push({contextId:`CLEO-D9-RACE-${prefix.join('_')}`,opponent:OPPONENT,decisionDay,prefix,prefixAttempt:attempt-1,preStateIdentical:preSame,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes:INNER*modes.length,savingsPct:100*(1-episodesUsed/(INNER*modes.length)),rows})}\nconst sourceArtifact='kag_teacher_q_persistent_cleo_d9_v1.ts',evidenceState='TEACHER_Q_RACING_PERSISTENT_CLEO_D9_V1';\nconst before=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconst statements:any[]=[];for(const c of contexts){statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_contexts (context_id,opponent,decision_day,prefix_json,feature_json,teacher_pick,teacher_scenarios_used,pre_state_identical,race_audit_json,episodes_used,full_episodes,savings_pct,evidence_state,source_artifact) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),c.teacherPick,c.teacherScenariosUsed,c.preStateIdentical?1:0,JSON.stringify(c.raceAudit),c.episodesUsed,c.fullEpisodes,c.savingsPct,evidenceState,sourceArtifact]});for(const r of c.rows)statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_q (context_id,opponent,decision_day,prefix_json,feature_json,action,q_mean,delta_vs_n,q_sd,lcb95,cvar50,positive_rate,sample_n,teacher_pick,evidence_state,source_artifact,ucb95,pruned) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),r.action,r.qMean,r.deltaVsN,r.qSd,r.lcb95,r.cvar50,r.positiveRate,r.sampleN,c.teacherPick,evidenceState,sourceArtifact,r.ucb95,r.pruned?1:0]})}\nawait sqlite.batch(statements);\nconst after=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconsole.log(JSON.stringify({status:'PASS',evidenceState,maxSampleN:INNER,persistence:{sourceArtifact,before:before.rows[0],attemptedRows:statements.length,after:after.rows[0]},contexts},null,2));",[],[983],"2b5fd901-3523-42ae-ad63-f03c91649ee8","kag_teacher_q_persistent_broker_d6_v2.ts","01a07a9c-40d4-76ae-adb5-098ed697023b",424,["D",1788763506938],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=6,modes=['N','H','W','B'],INNER=16,CONTEXTS=1;const cfg=E.state.resolveConfig({seed:996969,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number,attempt=0){const r=rng(3317000000+ci*104729+attempt*49979687);return Array.from({length:2},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number,attempt=0){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,996969),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst existingQ=await sqlite.execute(\"SELECT DISTINCT prefix_json FROM kag_teacher_q WHERE opponent='broker_bea.py' AND decision_day=6\"),seen=new Set((existingQ.rows??[]).map((r:any)=>String(r.prefix_json)));const contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??1.96;rows.push({action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0);contexts.push({contextId:`BROKER-D6-RACE-${prefix.join('_')}`,opponent:'broker_bea.py',decisionDay,prefix,prefixAttempt:attempt-1,preStateIdentical:preSame,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes:INNER*modes.length,savingsPct:100*(1-episodesUsed/(INNER*modes.length)),rows})}\nconst sourceArtifact='kag_teacher_q_persistent_broker_d6_v2.ts',evidenceState='TEACHER_Q_RACING_PERSISTENT_BROKER_D6_V2';\nconst before=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconst statements:any[]=[];for(const c of contexts){statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_contexts (context_id,opponent,decision_day,prefix_json,feature_json,teacher_pick,teacher_scenarios_used,pre_state_identical,race_audit_json,episodes_used,full_episodes,savings_pct,evidence_state,source_artifact) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),c.teacherPick,c.teacherScenariosUsed,c.preStateIdentical?1:0,JSON.stringify(c.raceAudit),c.episodesUsed,c.fullEpisodes,c.savingsPct,evidenceState,sourceArtifact]});for(const r of c.rows)statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_q (context_id,opponent,decision_day,prefix_json,feature_json,action,q_mean,delta_vs_n,q_sd,lcb95,cvar50,positive_rate,sample_n,teacher_pick,evidence_state,source_artifact,ucb95,pruned) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),r.action,r.qMean,r.deltaVsN,r.qSd,r.lcb95,r.cvar50,r.positiveRate,r.sampleN,c.teacherPick,evidenceState,sourceArtifact,r.ucb95,r.pruned?1:0]})}\nawait sqlite.batch(statements);\nconst after=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconsole.log(JSON.stringify({status:'PASS',evidenceState,maxSampleN:INNER,persistence:{sourceArtifact,before:before.rows[0],attemptedRows:statements.length,after:after.rows[0]},contexts},null,2));",[],[991],"fe9d4ec6-8940-4bee-8eb4-2fe9431e57a2","kag_teacher_q_persistent_broker_d6_v1.ts","01a07a84-6e15-74aa-8715-801599773262",358,["D",1788761953683],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=6,modes=['N','H','W','B'],INNER=16,CONTEXTS=1;const cfg=E.state.resolveConfig({seed:995959,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number,attempt=0){const r=rng(3117000000+ci*104729+attempt*49979687);return Array.from({length:2},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number,attempt=0){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,995959),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??1.96;rows.push({action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0);contexts.push({contextId:`BROKER-D6-RACE-${prefix.join('_')}-${ci}`,opponent:'broker_bea.py',decisionDay,prefix,preStateIdentical:preSame,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes:INNER*modes.length,savingsPct:100*(1-episodesUsed/(INNER*modes.length)),rows})}\nconst sourceArtifact='kag_teacher_q_persistent_broker_d6_v1.ts',evidenceState='TEACHER_Q_RACING_PERSISTENT_BROKER_D6_V1';\nconst before=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconst statements:any[]=[];for(const c of contexts)for(const r of c.rows)statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_q (context_id,opponent,decision_day,prefix_json,feature_json,action,q_mean,delta_vs_n,q_sd,lcb95,cvar50,positive_rate,sample_n,teacher_pick,evidence_state,source_artifact) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),r.action,r.qMean,r.deltaVsN,r.qSd,r.lcb95,r.cvar50,r.positiveRate,r.sampleN,c.teacherPick,evidenceState,sourceArtifact]});\nawait sqlite.batch(statements);\nconst after=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconsole.log(JSON.stringify({status:'PASS',evidenceState,maxSampleN:INNER,persistence:{sourceArtifact,before:before.rows[0],attemptedRows:statements.length,after:after.rows[0]},contexts},null,2));",[],[999],"609eb4dd-222a-4bb5-b88c-dfccf25986dd","kag_teacher_q_persistent_broker_d12_v2.ts","01a07a9e-1e82-7336-a590-ca6d61a60766",430,["D",1788763627228],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=12,modes=['N','H','W','B'],INNER=16,CONTEXTS=1;const cfg=E.state.resolveConfig({seed:997979,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number,attempt=0){const r=rng(3517000000+ci*104729+attempt*49979687);return Array.from({length:4},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number,attempt=0){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,997979),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst existingQ=await sqlite.execute(\"SELECT DISTINCT prefix_json FROM kag_teacher_q WHERE opponent='broker_bea.py' AND decision_day=12\"),seen=new Set((existingQ.rows??[]).map((r:any)=>String(r.prefix_json)));const contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??1.96;rows.push({action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0);contexts.push({contextId:`BROKER-D12-RACE-${prefix.join('_')}`,opponent:'broker_bea.py',decisionDay,prefix,prefixAttempt:attempt-1,preStateIdentical:preSame,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes:INNER*modes.length,savingsPct:100*(1-episodesUsed/(INNER*modes.length)),rows})}\nconst sourceArtifact='kag_teacher_q_persistent_broker_d12_v2.ts',evidenceState='TEACHER_Q_RACING_PERSISTENT_BROKER_D12_V2';\nconst before=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconst statements:any[]=[];for(const c of contexts){statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_contexts (context_id,opponent,decision_day,prefix_json,feature_json,teacher_pick,teacher_scenarios_used,pre_state_identical,race_audit_json,episodes_used,full_episodes,savings_pct,evidence_state,source_artifact) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),c.teacherPick,c.teacherScenariosUsed,c.preStateIdentical?1:0,JSON.stringify(c.raceAudit),c.episodesUsed,c.fullEpisodes,c.savingsPct,evidenceState,sourceArtifact]});for(const r of c.rows)statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_q (context_id,opponent,decision_day,prefix_json,feature_json,action,q_mean,delta_vs_n,q_sd,lcb95,cvar50,positive_rate,sample_n,teacher_pick,evidence_state,source_artifact,ucb95,pruned) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),r.action,r.qMean,r.deltaVsN,r.qSd,r.lcb95,r.cvar50,r.positiveRate,r.sampleN,c.teacherPick,evidenceState,sourceArtifact,r.ucb95,r.pruned?1:0]})}\nawait sqlite.batch(statements);\nconst after=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconsole.log(JSON.stringify({status:'PASS',evidenceState,maxSampleN:INNER,persistence:{sourceArtifact,before:before.rows[0],attemptedRows:statements.length,after:after.rows[0]},contexts},null,2));",[],[1007],"76bfe044-d46a-4ee8-ad22-01de5288446a","kag_teacher_q_persistent_broker_d12_v1.ts","01a07a87-2185-70bc-b945-cf685ffa4a95",372,["D",1788762128153],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=12,modes=['N','H','W','B'],INNER=16,CONTEXTS=1;const cfg=E.state.resolveConfig({seed:996969,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number,attempt=0){const r=rng(3317000000+ci*104729+attempt*49979687);return Array.from({length:4},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number,attempt=0){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,996969),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst existingQ=await sqlite.execute(\"SELECT DISTINCT prefix_json FROM kag_teacher_q WHERE opponent='broker_bea.py' AND decision_day=12\"),seen=new Set((existingQ.rows??[]).map((r:any)=>String(r.prefix_json)));const contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??1.96;rows.push({action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0);contexts.push({contextId:`BROKER-D12-RACE-${prefix.join('_')}`,opponent:'broker_bea.py',decisionDay,prefix,prefixAttempt:attempt-1,preStateIdentical:preSame,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes:INNER*modes.length,savingsPct:100*(1-episodesUsed/(INNER*modes.length)),rows})}\nconst sourceArtifact='kag_teacher_q_persistent_broker_d12_v1.ts',evidenceState='TEACHER_Q_RACING_PERSISTENT_BROKER_D12_V1';\nconst before=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconst statements:any[]=[];for(const c of contexts){statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_contexts (context_id,opponent,decision_day,prefix_json,feature_json,teacher_pick,teacher_scenarios_used,pre_state_identical,race_audit_json,episodes_used,full_episodes,savings_pct,evidence_state,source_artifact) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),c.teacherPick,c.teacherScenariosUsed,c.preStateIdentical?1:0,JSON.stringify(c.raceAudit),c.episodesUsed,c.fullEpisodes,c.savingsPct,evidenceState,sourceArtifact]});for(const r of c.rows)statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_q (context_id,opponent,decision_day,prefix_json,feature_json,action,q_mean,delta_vs_n,q_sd,lcb95,cvar50,positive_rate,sample_n,teacher_pick,evidence_state,source_artifact,ucb95,pruned) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),r.action,r.qMean,r.deltaVsN,r.qSd,r.lcb95,r.cvar50,r.positiveRate,r.sampleN,c.teacherPick,evidenceState,sourceArtifact,r.ucb95,r.pruned?1:0]})}\nawait sqlite.batch(statements);\nconst after=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconsole.log(JSON.stringify({status:'PASS',evidenceState,maxSampleN:INNER,persistence:{sourceArtifact,before:before.rows[0],attemptedRows:statements.length,after:after.rows[0]},contexts},null,2));",[],[1015],"4241ebdd-5a44-45fc-8faf-617a212073f0","kag_teacher_q_full_compare_race1_v1.ts","01a07a66-84f4-73eb-9b3c-cb7dfc2fddb6",310,["D",1788759958851],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=1;const cfg=E.state.resolveConfig({seed:969696,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(_ci:number){return ['SMOOTHIE_SHOP','PIZZA_SHOP','PET_CAFE']}function innerWorlds(prefix:string[],ci:number){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,969696),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={8:2.365,12:2.201,16:2.131};function teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=sd(d)/Math.sqrt(n);st[m]={delta:mu,lcb:mu-tcrit[n]*se}}const pos=modes.filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=sd(d)/Math.sqrt(INNER);rows.push({action:m,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tcrit[16]*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/INNER})}contexts.push({contextId:`BROKER-D9-${prefix.join('_')}-${ci}`,opponent:'broker_bea.py',decisionDay,prefix,preStateIdentical:preSame,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,rows})}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'TEACHER_Q_DATASET_BATCH2',sampleN:INNER,contexts},null,2));",[],[1023],"302becdf-7759-4fa3-b0e4-b047072416c1","kag_teacher_q_batch2_v1.ts","01a07a49-eae8-70e9-bf31-e0593d64882d",300,["D",1788758075819],2,"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=2;const cfg=E.state.resolveConfig({seed:959595,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number){const r=rng(1517000000+ci*104729);return Array.from({length:3},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,959595),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={8:2.365,12:2.201,16:2.131};function teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=sd(d)/Math.sqrt(n);st[m]={delta:mu,lcb:mu-tcrit[n]*se}}const pos=modes.filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=sd(d)/Math.sqrt(INNER);rows.push({action:m,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tcrit[16]*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/INNER})}contexts.push({contextId:`BROKER-D9-${prefix.join('_')}-${ci}`,opponent:'broker_bea.py',decisionDay,prefix,preStateIdentical:preSame,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,rows})}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'TEACHER_Q_DATASET_BATCH2',sampleN:INNER,contexts},null,2));",[],[1031],"3c9d8f7d-1682-4440-a33c-4e39f70920a1","kag_teacher_q_batch1_v1.ts","98f30f9c-aa7a-11f1-88cf-1607ee4eb77e",297,["D",1788757895346],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=2;const cfg=E.state.resolveConfig({seed:959595,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number){const r=rng(1317000000+ci*104729);return Array.from({length:3},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,959595),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={8:2.365,12:2.201,16:2.131};function teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=sd(d)/Math.sqrt(n);st[m]={delta:mu,lcb:mu-tcrit[n]*se}}const pos=modes.filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=sd(d)/Math.sqrt(INNER);rows.push({action:m,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tcrit[16]*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/INNER})}contexts.push({contextId:`BROKER-D9-${prefix.join('_')}-${ci}`,opponent:'broker_bea.py',decisionDay,prefix,preStateIdentical:preSame,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,rows})}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'TEACHER_Q_DATASET_BATCH1',sampleN:INNER,contexts},null,2));",[],[1040],"83ed77cf-e4c5-4d67-942c-efe1a168f160","kag_teacher_features_v3_selftest.ts","dc34aa1c-aa85-11f1-8d04-1607ee4eb77e",395,["D",1788762816593],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { extractTeacherFeaturesV3,KAG_TEACHER_FEATURE_SCHEMA_V3 } from './lib/kag_teacher_features_v3.ts';\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:556677,weedSpawnChance:0});const rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`import json\\n_ns={}\\nexec(src,_ns)\\n_A=_ns['agent']\\ndef call(p):return json.dumps(_A(json.loads(p)))`);const call=py.globals.get('call');let s=E.state.initGameState(2,cfg,556677),prev:any=null,o:any=null;while(!s.done){o=observationFor(s,0);if(+o.day===9&&+o.hour===0)break;prev=o;const a=JSON.parse(String(call(JSON.stringify(o))));s=E.interpreter.step(s,[a,{farmer:[],hands:[],market:[]}],cfg)}const obsBefore=JSON.stringify(o),prevBefore=JSON.stringify(prev),a=extractTeacherFeaturesV3(o,prev),obsAfter=JSON.stringify(o),prevAfter=JSON.stringify(prev),b=extractTeacherFeaturesV3(o,prev),raw=JSON.stringify(a);const bad=/NaN|Infinity|undefined/.test(raw),nonMutating=obsBefore===obsAfter&&prevBefore===prevAfter;console.log(JSON.stringify({status:!bad&&nonMutating&&JSON.stringify(a)===JSON.stringify(b)&&a.schema===KAG_TEACHER_FEATURE_SCHEMA_V3?'PASS':'FAIL',schema:a.schema,deterministic:JSON.stringify(a)===JSON.stringify(b),finite:!bad,nonMutating,keySignals:{currentHands:a.me.hands,prevHands:a.temporal?.me?.hands,prevHires:a.temporal?.me?.hiresToday,prevCarriedWheat:a.temporal?.me?.carried?.WHEAT,currentCarriedTotal:a.private.carriedTotal,wheat:a.me.crops.WHEAT,cows:a.me.animals.COW,workerSpatial:a.workerSpatial}},null,2));",[],[1048],"e9d958bc-6908-4acf-a288-04375a2f959a","kag_strawberry_wheat_attack_dossier_v1.ts","5353aefa-aa62-11f1-b97f-1607ee4eb77e",169,["D",1788747470648],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={M60:[.4,.6,0],M70:[.3,.7,0],S10W:[.3,.6,.1]};const seedList=[43003,43117,43223,43331];const E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['broker_bea.py']));py.runPython(`import json\\n_r={}\\nexec(src,_r)\\n_REF=_r['agent']`);py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(p):return json.dumps(_REF(json.loads(p)))\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep(w,m,s):\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON','STRAWBERRY'];_D['POLICY']['crop_share']={'WHEAT':w,'MELON':m,'STRAWBERRY':s};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16,'STRAWBERRY':16};_D['POLICY']['sell_order']=['STRAWBERRY','MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(v:string,seed:number,seat:0|1){prep(...variants[v]);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,maxFlock=0,finalFlock=0;const our:any={buy:{},sell:{},seed:{}},opp:any={buy:{},sell:{},seed:{}},days:any[]=[];while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){if(+o.hour===0){days.push({day:+o.day,ourCash:+o.farms[pid].money,oppCash:+o.farms[1-pid].money,wheatP:+o.market.prices.WHEAT,strawP:+o.market.prices.STRAWBERRY,melonP:+o.market.prices.MELON,wheatI:+o.market.inventory.WHEAT,strawI:+o.market.inventory.STRAWBERRY,melonI:+o.market.inventory.MELON});const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}const a=addFert(o,enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg));for(const q of a.market??[]){const [k,it,n]=q;if(k==='BUY_PRODUCT')our.buy[it]=(our.buy[it]??0)+Number(n??0);if(k==='SELL')our.sell[it]=(our.sell[it]??0)+Number(n??0);if(k==='BUY_SEED')our.seed[it]=(our.seed[it]??0)+Number(n??0)}aa[pid]=a;maxFlock=Math.max(maxFlock,own(o,'COW')+own(o,'SHEEP'))}else{const a=JSON.parse(String(cr(JSON.stringify(o))));for(const q of a.market??[]){const [k,it,n]=q;if(k==='BUY_PRODUCT')opp.buy[it]=(opp.buy[it]??0)+Number(n??0);if(k==='SELL')opp.sell[it]=(opp.sell[it]??0)+Number(n??0);if(k==='BUY_SEED')opp.seed[it]=(opp.seed[it]??0)+Number(n??0)}aa[pid]=a}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{v,seed,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,animalLoss:maxFlock-finalFlock,our,opp,days}}\nconst games:any[]=[];for(const v of Object.keys(variants))for(const seed of seedList){games.push(await play(v,seed,0));games.push(await play(v,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);function agg(v:string){const x=games.filter(g=>g.v===v),dayset=[6,9,12,15,18,21,24,27,29];const d=dayset.map(day=>{const r=x.map(g=>g.days.find((q:any)=>q.day===day)).filter(Boolean);return{day,ourCash:avg(r.map((q:any)=>q.ourCash)),oppCash:avg(r.map((q:any)=>q.oppCash)),wheatP:avg(r.map((q:any)=>q.wheatP)),strawP:avg(r.map((q:any)=>q.strawP)),melonP:avg(r.map((q:any)=>q.melonP)),wheatI:avg(r.map((q:any)=>q.wheatI)),strawI:avg(r.map((q:any)=>q.strawI)),melonI:avg(r.map((q:any)=>q.melonI))}});const get=(who:'our'|'opp',type:'buy'|'sell'|'seed',it:string)=>avg(x.map(g=>g[who][type][it]??0));return{episodes:x.length,ourBank:avg(x.map(g=>g.ours)),oppBank:avg(x.map(g=>g.theirs)),margin:avg(x.map(g=>g.margin)),animalLoss:avg(x.map(g=>g.animalLoss)),our:{buyWheat:get('our','buy','WHEAT'),sellWheat:get('our','sell','WHEAT'),sellStraw:get('our','sell','STRAWBERRY'),sellMelon:get('our','sell','MELON'),seedStraw:get('our','seed','STRAWBERRY')},opp:{buyWheat:get('opp','buy','WHEAT'),sellWheat:get('opp','sell','WHEAT'),sellStraw:get('opp','sell','STRAWBERRY'),sellMelon:get('opp','sell','MELON'),seedStraw:get('opp','seed','STRAWBERRY')},days:d}}\nconst out:any={};for(const v of Object.keys(variants))out[v]=agg(v);for(const v of ['M70','S10W'])out[v].deltaVsM60={ourBank:out[v].ourBank-out.M60.ourBank,oppBank:out[v].oppBank-out.M60.oppBank,margin:out[v].margin-out.M60.margin};console.log(JSON.stringify({status:'PASS',question:'DOES_S10W_ATTACK_TOP_TIER_THROUGH_STRAWBERRY_REVENUE_COMPETITION_WHEAT_INPUT_INFLATION_OR_BOTH',officialCurves:{WHEAT:'slow log glut; base25',STRAWBERRY:'base120; excess +50 =>24; +100=>1'},out},null,2));",[],[1056],"c5dd305d-8af2-4058-848f-a2bb67ab8f73","kag_strawberry_signal_hybrid_v1.ts","e05fc67c-aa57-11f1-9819-1607ee4eb77e",126,["D",1788742982831],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['MIX11','STRAW_SIGNAL'],refs=['melon_mateo.py','rancher_rita.py','broker_bea.py'],seeds=[22003,22111,22229,22343,22453,22567,22679,22787];const strawSinks=new Set(['BRUNCH_SPOT','ICE_CREAM_SHOP','SMOOTHIE_SHOP','FARMERS_MARKET']);const E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();const rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\\ndef reset():\\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0}\\ndef seth(c,s,o):\\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)}\\n _D['POLICY']['animals']=json.loads(o)\\ndef cropmode(on):\\n if on:\\n _D['POLICY']['crops']=['WHEAT','STRAWBERRY'];_D['POLICY']['crop_share']={'WHEAT':0.65,'STRAWBERRY':0.35};_D['POLICY']['plant_until']={'WHEAT':25,'STRAWBERRY':16};_D['POLICY']['sell_order']=['STRAWBERRY','MILK','WOOL','WHEAT']\\n else:\\n _D['POLICY']['crops']=['WHEAT'];_D['POLICY']['crop_share']={'WHEAT':1.0};_D['POLICY'].pop('plant_until',None);_D['POLICY']['sell_order']=['MILK','WOOL','WHEAT']\\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),reset=py.globals.get('reset'),seth=py.globals.get('seth'),cropmode=py.globals.get('cropmode');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}\nasync function play(v:string,ref:string,seed:number,seat:0|1){reset();let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},done=false,err=0,signalDay:null|number=null;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){if(+o.hour===0){const signal=(o.town?.unlocked_shops??[]).some((x:string)=>strawSinks.has(x))&&+o.day<=16;if(v==='STRAW_SIGNAL'&&signal&&signalDay===null)signalDay=+o.day;cropmode(v==='STRAW_SIGNAL'&&signal);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!done){done=ow.COW>=1&&ow.SHEEP>=1;if(!done){tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)};seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}}if(done){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')))}}aa[pid]=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,signalDay}}\nconst games:any[]=[];for(const v of variants)for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const summary:any={};for(const v of variants){const x=games.filter(g=>g.v===v);summary[v]={episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOppBank:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),signals:x.filter(g=>g.signalDay!==null).length,avgSignalDay:avg(x.filter(g=>g.signalDay!==null).map(g=>g.signalDay)),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);summary[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOppBank:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin))}}}const b=games.filter(g=>g.v==='MIX11'),h=games.filter(g=>g.v==='STRAW_SIGNAL');console.log(JSON.stringify({status:'PASS',hypothesis:'USE_VISIBLE_STRAWBERRY_DEMAND_AS_A_RECOURSE_SIGNAL_FOR_A_SMALL_SECONDARY_ENTERPRISE_WHILE_PRESERVING_WHEAT_FEED_CAPACITY',independentDesign:{share:{WHEAT:0.65,STRAWBERRY:0.35},deadline:16,sinks:[...strawSinks]},summary,paired:{bankDelta:avg(h.map((g,i)=>g.ours-b[i].ours)),oppDelta:avg(h.map((g,i)=>g.theirs-b[i].theirs)),marginDelta:avg(h.map((g,i)=>g.margin-b[i].margin)),better:h.filter((g,i)=>g.margin>b[i].margin).length,worse:h.filter((g,i)=>g.margin=2)return[.3,.3,.4];if(k>=1)return[.35,.35,.3];return d<9?[.4,.5,.1]:[.4,.6,0]}const s=Math.min(.4,.08+.08*k),w=Math.max(.30,.40-.02*k),m=Math.max(.25,1-w-s);return[w,m,s]}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(v:string,ref:string,seed:number,seat:0|1){prep();let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0,strawSell=0,strawSeed=0,shopDays:any={};while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const [w,m,st]=shares(v,o);setcrop(w,m,st);if(+o.hour===0)shopDays[o.day]=shops(o);const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}const a=addFert(o,enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg));for(const q of a.market??[]){if(q?.[0]==='SELL'&&q?.[1]==='STRAWBERRY')strawSell+=Number(q[2]??0);if(q?.[0]==='BUY_SEED'&&q?.[1]==='STRAWBERRY')strawSeed+=Number(q[2]??0)}aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,animalLoss:maxFlock-finalFlock,strawSell,strawSeed,finalStrawShops:Object.values(shopDays).at(-1)??0}}\nconst games:any[]=[];for(const v of variants)for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const summary:any={};for(const v of variants){const x=games.filter(g=>g.v===v);summary[v]={episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),avgStrawSeed:avg(x.map(g=>g.strawSeed)),avgStrawSellAttempt:avg(x.map(g=>g.strawSell)),avgFinalStrawShops:avg(x.map(g=>g.finalStrawShops)),errors:x.reduce((a,g)=>a+g.err,0),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);summary[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin)),avgStrawSeed:avg(y.map(g=>g.strawSeed)),avgStrawShops:avg(y.map(g=>g.finalStrawShops))}}}console.log(JSON.stringify({status:'PASS',question:'CAN_SMALL_EARLY_STRAWBERRY_OPTIONALITY_PLUS_DEMAND_RECOURSE_OUTPERFORM_THE_PREVIOUS_LATE_35_PERCENT_STRAW_SIGNAL_FAILURE',officialFacts:{strawberryDemandShopTypes:4,totalShopTypes:8,firstYieldDay:10,ongoingIntervalDays:2},summary},null,2));",[],[1072],"270d8b52-366a-472c-a090-089799dd35a5","kag_strawberry_curve_probe_v1.ts","6ae6d48a-aa61-11f1-adf1-1607ee4eb77e",166,["D",1788747121355],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadOfficialKaggricultureEngine } from './lib/kaggriculture_engine_loader.ts';const E=await loadOfficialKaggricultureEngine();const p=E.constants.MARKET_PARAMS.STRAWBERRY;const inv=[9800,9900,9950,10000,10050,10100,10150,10200].map(x=>({inv:x,price:E.market.marketPrice('STRAWBERRY',x)}));const shops=Object.entries(E.constants.SHOPS).filter(([_,v]:any)=>JSON.stringify(v).includes('STRAWBERRY'));const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const econ=strFromU8(z['crop_economics.csv']).split('\\n').filter(x=>/STRAWBERRY|item/i.test(x));console.log(JSON.stringify({params:p,samples:inv,shops,econ,officialCrop:E.constants.CROPS.STRAWBERRY,officialMelon:E.constants.CROPS.MELON},null,2));",[],[1080],"9f0b208c-b78e-41bc-9ef4-e3f364aa129c","kag_strawberry_attack_ablation_v1.ts","ebbde396-aa61-11f1-ac92-1607ee4eb77e",168,["D",1788747296861],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={M60:[.4,.6,0],S05_SPLIT:[.375,.575,.05],S10_MELON:[.4,.5,.1],S10_WHEAT:[.3,.6,.1],S10_SPLIT:[.35,.55,.1],S15_SPLIT:[.325,.525,.15]};const refs=['broker_bea.py','ledger_lena.py','slotter_silas.py','closer_cleo.py'],seeds=[42013,42119,42223];const E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep(w,m,s):\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON','STRAWBERRY'];_D['POLICY']['crop_share']={'WHEAT':w,'MELON':m,'STRAWBERRY':s};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16,'STRAWBERRY':16};_D['POLICY']['sell_order']=['STRAWBERRY','MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(v:string,ref:string,seed:number,seat:0|1){prep(...variants[v]);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0,strawSell=0;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}const a=addFert(o,enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg));for(const q of a.market??[])if(q?.[0]==='SELL'&&q?.[1]==='STRAWBERRY')strawSell+=Number(q[2]??0);aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,animalLoss:maxFlock-finalFlock,strawSell}}\nconst games:any[]=[];for(const v of Object.keys(variants))for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const summary:any={};for(const v of Object.keys(variants)){const x=games.filter(g=>g.v===v);summary[v]={shares:variants[v],episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),avgStrawSellAttempt:avg(x.map(g=>g.strawSell)),errors:x.reduce((a,g)=>a+g.err,0),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);summary[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin))}}}const b=games.filter(g=>g.v==='M60');const paired=Object.keys(variants).filter(v=>v!=='M60').map(v=>{const x=games.filter(g=>g.v===v);return{v,bankDelta:avg(x.map((g,i)=>g.ours-b[i].ours)),oppDelta:avg(x.map((g,i)=>g.theirs-b[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-b[i].margin)),better:x.filter((g,i)=>g.margin>b[i].margin).length,worse:x.filter((g,i)=>g.marging.win&&!b[i].win).length}});console.log(JSON.stringify({status:'PASS',question:'IS_THE_TOP_TIER_DAMAGE_CAUSED_BY_STRAWBERRY_ENTRY_ITSELF_OR_BY_WHICH_BASE_CROP_IT_DISPLACES',summary,paired},null,2));",[],[1088],"74d00517-6da4-4bde-9a1c-011a697d8712","kag_stability_gate_false_negative_audit_v1.ts","01a07a42-cc50-7488-9bf1-fe24efd1e3a8",293,["D",1788757611573],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES];\nconst cfg=E.state.resolveConfig({seed:949494,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst decisionDay=9,modes=['N','H','W','B'],MAXN=16;\nconst actual=['PET_CAFE','FARMERS_MARKET','SMOOTHIE_SHOP','BAKERY','PIZZA_SHOP','SMOOTHIE_SHOP','PET_CAFE','PIZZA_SHOP'];\nconst prefix=actual.slice(0,3);\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}\nconst scenarios:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction compact(o:any){const me=+o.player,f=o.farms[me],g=o.farms[1-me];return{day:+o.day,cash:+f.money,oppCash:+g.money,shops:o.town.unlocked_shops}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,949494),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=compact(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst OUTER=8,CASES=[{id:'PFB',prefix:['PET_CAFE','FARMERS_MARKET','BAKERY'],action:'H'},{id:'BBP',prefix:['BAKERY','BRUNCH_SPOT','PIZZA_SHOP'],action:'H'}];\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))};const result:any={};for(let ci=0;cir.delta),mu=avg(ds),se=sd(ds)/Math.sqrt(ds.length);result[c.id]={prefix:c.prefix,action:c.action,meanDelta:mu,se,lo95:mu-1.96*se,hi95:mu+1.96*se,better:ds.filter(x=>x>0).length,worse:ds.filter(x=>x<0).length,min:Math.min(...ds),max:Math.max(...ds),rows}}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_STABILITY_GATE_FALSE_NEGATIVE_AUDIT',outerPerPrefix:OUTER,result,interpretation:'independent suffixes conditioned on prefixes where Stability Gate stayed at N; tests whether conservative non-intervention missed systematic positive H value or correctly avoided uncertain intervention'},null,2));",[],[1096],"f1e1a626-2b1d-4b6a-9fad-0b5198ce00ab","kag_shop_consumption_probe_v1.ts","7ee67f14-aa6e-11f1-a8b1-1607ee4eb77e",232,["D",1788752697719],"import { loadOfficialKaggricultureEngine } from './lib/kaggriculture_engine_loader.ts';const E=await loadOfficialKaggricultureEngine();const c:any=E.constants;const keys=Object.keys(c).filter(k=>/SHOP|CONSUM|DEMAND/i.test(k));console.log(JSON.stringify({keys,values:Object.fromEntries(keys.map(k=>[k,c[k]]))},null,2));",[],[1104],"68d1730d-3f08-407c-a9c3-c97560060ac8","kag_rng_state_probe_v1.ts","ce30a8ae-aa89-11f1-8632-1607ee4eb77e",459,["D",1788764453437],"const base='https://raw.githubusercontent.com/Kaggle/kaggle-environments/master/kaggle_environments/envs/kaggriculture/visualizer/playable/src/engine/';const files=['state.ts','interpreter.ts','rng.ts','types.ts'];const out:any={};for(const f of files){const r=await fetch(base+f);const t=await r.text();const lines=t.split('\\n');out[f]=lines.map((x,i)=>({line:i+1,text:x})).filter(x=>/rng|seed|random|weed/i.test(x.text)).slice(0,120)}console.log(JSON.stringify({status:'PASS',source:base,findings:out},null,2));",[],[1112],"b9d55ad4-2a83-43f5-8306-4f26d247ab4e","kag_rng_policy_invariance_probe_v1.ts","01a079a9-1e8c-7737-8b88-ebc2e6e2d4f1",173,["D",1788747552156],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={M60:[.4,.6,0],M70:[.3,.7,0],S10W:[.3,.6,.1]};const seedList=[43003];const E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['broker_bea.py']));py.runPython(`import json\\n_r={}\\nexec(src,_r)\\n_REF=_r['agent']`);py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(p):return json.dumps(_REF(json.loads(p)))\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep(w,m,s):\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON','STRAWBERRY'];_D['POLICY']['crop_share']={'WHEAT':w,'MELON':m,'STRAWBERRY':s};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16,'STRAWBERRY':16};_D['POLICY']['sell_order']=['STRAWBERRY','MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(v:string,seed:number,seat:0|1){prep(...variants[v]);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,maxFlock=0,finalFlock=0;const our:any={buy:{},sell:{},seed:{}},opp:any={buy:{},sell:{},seed:{}},days:any[]=[];while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){if(+o.hour===0){days.push({day:+o.day,ourCash:+o.farms[pid].money,oppCash:+o.farms[1-pid].money,wheatP:+o.market.prices.WHEAT,strawP:+o.market.prices.STRAWBERRY,melonP:+o.market.prices.MELON,wheatI:+o.market.inventory.WHEAT,strawI:+o.market.inventory.STRAWBERRY,melonI:+o.market.inventory.MELON,shops:[...(o.town?.unlocked_shops??[])]});const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}const a=addFert(o,enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg));for(const q of a.market??[]){const [k,it,n]=q;if(k==='BUY_PRODUCT')our.buy[it]=(our.buy[it]??0)+Number(n??0);if(k==='SELL')our.sell[it]=(our.sell[it]??0)+Number(n??0);if(k==='BUY_SEED')our.seed[it]=(our.seed[it]??0)+Number(n??0)}aa[pid]=a;maxFlock=Math.max(maxFlock,own(o,'COW')+own(o,'SHEEP'))}else{const a=JSON.parse(String(cr(JSON.stringify(o))));for(const q of a.market??[]){const [k,it,n]=q;if(k==='BUY_PRODUCT')opp.buy[it]=(opp.buy[it]??0)+Number(n??0);if(k==='SELL')opp.sell[it]=(opp.sell[it]??0)+Number(n??0);if(k==='BUY_SEED')opp.seed[it]=(opp.seed[it]??0)+Number(n??0)}aa[pid]=a}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{v,seed,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,animalLoss:maxFlock-finalFlock,our,opp,days}}\nconst games:any[]=[];for(const v of Object.keys(variants))for(const seed of seedList){games.push(await play(v,seed,0));games.push(await play(v,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);function agg(v:string){const x=games.filter(g=>g.v===v),dayset=[6,9,12,15,18,21,24,27,29];const d=dayset.map(day=>{const r=x.map(g=>g.days.find((q:any)=>q.day===day)).filter(Boolean);return{day,ourCash:avg(r.map((q:any)=>q.ourCash)),oppCash:avg(r.map((q:any)=>q.oppCash)),wheatP:avg(r.map((q:any)=>q.wheatP)),strawP:avg(r.map((q:any)=>q.strawP)),melonP:avg(r.map((q:any)=>q.melonP)),wheatI:avg(r.map((q:any)=>q.wheatI)),strawI:avg(r.map((q:any)=>q.strawI)),melonI:avg(r.map((q:any)=>q.melonI))}});const get=(who:'our'|'opp',type:'buy'|'sell'|'seed',it:string)=>avg(x.map(g=>g[who][type][it]??0));return{episodes:x.length,ourBank:avg(x.map(g=>g.ours)),oppBank:avg(x.map(g=>g.theirs)),margin:avg(x.map(g=>g.margin)),animalLoss:avg(x.map(g=>g.animalLoss)),our:{buyWheat:get('our','buy','WHEAT'),sellWheat:get('our','sell','WHEAT'),sellStraw:get('our','sell','STRAWBERRY'),sellMelon:get('our','sell','MELON'),seedStraw:get('our','seed','STRAWBERRY')},opp:{buyWheat:get('opp','buy','WHEAT'),sellWheat:get('opp','sell','WHEAT'),sellStraw:get('opp','sell','STRAWBERRY'),sellMelon:get('opp','sell','MELON'),seedStraw:get('opp','seed','STRAWBERRY')},days:d}}\nconsole.log(JSON.stringify({status:'PASS',question:'ARE_SHOP_DRAWS_POLICY_INVARIANT_UNDER_SAME_SEED',games:games.map(g=>({v:g.v,seat:g.seat,final:[g.ours,g.theirs],shops:g.days.filter((d:any)=>[3,6,9,12,15,18,21,24,29].includes(d.day)).map((d:any)=>({day:d.day,shops:d.shops}))}))},null,2));",[],[1120],"7d0fd93f-114b-47d6-bc8a-5fa4b6e1d5db","kag_rita_wheat_logic_probe_v1.ts","504e3b02-aa63-11f1-a489-1607ee4eb77e",177,["D",1788747895046],"import { unzipSync,strFromU8 } from 'npm:fflate';const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const s=strFromU8(z['rancher_rita.py']).split('\\n');for(let i=0;i`${540+i}|${x}`).join('\\n'));",[],[1144],"4f3dab8f-4e82-468f-a02c-3b45a41dc86b","kag_rita_policy_probe_v1.ts","3ed49cae-aa63-11f1-b65c-1607ee4eb77e",176,["D",1788747865719],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`import json\\n_d={}\\nexec(src,_d)\\nout=json.dumps(_d['POLICY'])`);console.log(String(py.globals.get('out')));",[],[1152],"da6a2190-6e98-45e0-9069-af5b52183213","kag_rita_plan_market_head_v1.ts","eb6a9a8a-aa55-11f1-a2e8-1607ee4eb77e",113,["D",1788742142368],"import { unzipSync,strFromU8 } from 'npm:fflate';const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const s=strFromU8(z['rancher_rita.py']).split('\\n');const i=s.findIndex(x=>x.startsWith('def _plan_market'));console.log(s.slice(i,i+35).map((x,j)=>`${i+j+1}|${x}`).join('\\n'));",[],[1160],"47bc57bc-4651-4b1a-8d20-1870c453483f","kag_rita_market_context_v1.ts","cb65e62c-aa55-11f1-9303-1607ee4eb77e",110,["D",1788742088644],"import { unzipSync,strFromU8 } from 'npm:fflate';\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const s=strFromU8(z['rancher_rita.py']).split('\\n');console.log(s.slice(385,545).map((x,i)=>`${386+i}|${x}`).join('\\n'));",[],[1168],"f54c171a-b739-442b-8377-136787d0c31c","kag_rita_job_tail_v1.ts","6f4de608-aa57-11f1-8f81-1607ee4eb77e",124,["D",1788742793125],"import { unzipSync,strFromU8 } from 'npm:fflate';const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const s=strFromU8(z['rancher_rita.py']).split('\\n');console.log(s.slice(275,349).map((x,i)=>`${276+i}|${x}`).join('\\n'));",[],[1176],"b90b7006-a3cb-49c4-9a5f-542275260071","kag_rita_fertilizer_probe_v1.ts","46203f26-aa55-11f1-8c43-1607ee4eb77e",103,["D",1788741865052],"import { unzipSync,strFromU8 } from 'npm:fflate';\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const s=strFromU8(z['rancher_rita.py']).split('\\n');const terms=['COLLECT_FERTILIZER','FERTILIZER','CARE','HARVEST'];for(const term of terms){const hits=s.map((x,i)=>({i:i+1,x})).filter(h=>h.x.includes(term));console.log(JSON.stringify({term,hits},null,2));}",[],[1184],"0ccfd662-8f8f-425f-8f1a-5531214c9c6a","kag_rita_collect_context_v1.ts","bae7afc4-aa55-11f1-a2b4-1607ee4eb77e",109,["D",1788742060971],"import { unzipSync,strFromU8 } from 'npm:fflate';\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const s=strFromU8(z['rancher_rita.py']).split('\\n');console.log(s.slice(215,275).map((x,i)=>`${216+i}|${x}`).join('\\n'));console.log('\\nPOLICY FLAGS\\n');console.log(s.slice(115,175).map((x,i)=>`${116+i}|${x}`).join('\\n'));",[],[1192],"4695de12-630d-4938-98e7-b0f1e2537844","kag_rita_agent_tail_v1.ts","e109c494-aa55-11f1-98f0-1607ee4eb77e",112,["D",1788742124943],"import { unzipSync,strFromU8 } from 'npm:fflate';const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const s=strFromU8(z['rancher_rita.py']).split('\\n');console.log(s.slice(640,735).map((x,i)=>`${641+i}|${x}`).join('\\n'));",[],[1200],"8b72e686-3f7d-45e9-99b4-9428469c79a4","kag_reference_interepisode_state_probe_v1.ts","4fe49d8e-aa7a-11f1-af08-1607ee4eb77e",296,["D",1788757796106],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nconst E=await loadOfficialKaggricultureEngine();const rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();\npy.runPython(`import json,hashlib,types\\n_N={}\\n_A=None\\ndef load(src):\\n global _N,_A\\n _N={}\\n exec(src,_N)\\n _A=_N['agent']\\ndef call(p): return json.dumps(_A(json.loads(p)),sort_keys=True)\\ndef snap():\\n out={}\\n for k,v in _N.items():\\n if k.startswith('__') or isinstance(v,(types.ModuleType,types.FunctionType,type)): continue\\n try:\\n r=repr(v); out[k]={'type':type(v).__name__,'len':len(v) if hasattr(v,'__len__') else None,'sha':hashlib.sha256(r.encode()).hexdigest()}\\n except Exception: pass\\n return json.dumps(out,sort_keys=True)`);\nconst load=py.globals.get('load'),call=py.globals.get('call'),snap=py.globals.get('snap');\nasync function episode(){const cfg=E.state.resolveConfig({seed:424242});let s=E.state.initGameState(2,cfg,424242),actions:string[]=[];while(!s.done){const o=observationFor(s,1),raw=String(call(JSON.stringify(o)));actions.push(raw);const a=JSON.parse(raw);s=E.interpreter.step(s,[{farmer:[],hands:[],market:[]},a],cfg)}return{money:s.farms[1].money,opp:s.farms[0].money,actionHash:await crypto.subtle.digest('SHA-256',new TextEncoder().encode(actions.join('\\n'))).then(b=>Array.from(new Uint8Array(b)).map(x=>x.toString(16).padStart(2,'0')).join('')),calls:actions.length}}\nconst files=['ledger_lena.py','slotter_silas.py'];const result:any={};for(const f of files){const src=strFromU8(z[f]);load(src);const pre=JSON.parse(String(snap())),e1=await episode(),post1=JSON.parse(String(snap())),e2=await episode(),post2=JSON.parse(String(snap()));load(src);const fresh=JSON.parse(String(snap())),e3=await episode();const keys=[...new Set([...Object.keys(pre),...Object.keys(post1)])],changed=keys.filter(k=>JSON.stringify(pre[k])!==JSON.stringify(post1[k]));result[f]={changedGlobalsAfterEpisode:changed,preVsFreshSame:JSON.stringify(pre)===JSON.stringify(fresh),reuseEpisodeMatchesFirst:e2.actionHash===e1.actionHash&&e2.money===e1.money,freshEpisodeMatchesFirst:e3.actionHash===e1.actionHash&&e3.money===e1.money,episodes:{first:e1,reuse:e2,fresh:e3},changedDetails:Object.fromEntries(changed.slice(0,20).map(k=>[k,{before:pre[k],after:post1[k],afterSecond:post2[k]}]))}}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'REFERENCE_INTEREPISODE_STATEFULNESS_PROBE',seed:424242,result},null,2));",[],[1208],"3016c188-82ae-448e-8b25-f71be72dc9a5","kag_recourse_vs_preemption_probe_v1.ts","8fa86e72-aa4e-11f1-b2f6-1607ee4eb77e",83,["D",1788738981936],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['PREEMPT','RECOURSE4','RECOURSE6'];\nconst seeds=[9403,9419,9431,9461,9491]; const opponents=['C4S12','C6S10','RITA_BASE'];\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();\nconst zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!zr.ok)throw new Error(String(zr.status));const z=unzipSync(new Uint8Array(await zr.arrayBuffer()));const src=strFromU8(z['rancher_rita.py']);\nconst py=await loadPyodide();py.runPython(`import json,copy\\n_REG={}\\n_BASE={}`);py.globals.set('src',src);const patches:Record={DYN:{animal_target:{COW:0,SHEEP:0}},C4S12:{animal_target:{COW:4,SHEEP:12}},C6S10:{animal_target:{COW:6,SHEEP:10}},RITA_BASE:{}};for(const [n,p] of Object.entries(patches)){py.globals.set('nm',n);py.globals.set('patch',JSON.stringify(p));py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n\\n_BASE[nm]=copy.deepcopy(_n['POLICY'])`);}py.runPython(`\\ndef call(name,payload): return json.dumps(_REG[name]['agent'](json.loads(payload)))\\ndef set_herd(name,cows,sheep,order):\\n _REG[name]['POLICY']['animal_target']={'COW':int(cows),'SHEEP':int(sheep)}\\n _REG[name]['POLICY']['animals']=json.loads(order)\\ndef reset(name):\\n _REG[name]['POLICY'].clear();_REG[name]['POLICY'].update(copy.deepcopy(_BASE[name]))\\n`);const call=py.globals.get('call'),setHerd=py.globals.get('set_herd'),reset=py.globals.get('reset');\nfunction countOwned(obs:any,a:string){const me=+obs.player,f=obs.farms[me];let n=+(obs.private?.shed?.[a]??0);for(const inv of obs.private?.inventories??[])n+=+(inv?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n;}\nfunction enforce(obs:any,act:any,t:{COW:number;SHEEP:number}){const x:any={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};const o={...act,market:[] as any[]};for(const m of act?.market??[]){if(Array.isArray(m)&&m[0]==='BUY_ANIMAL'&&(m[1]==='COW'||m[1]==='SHEEP')){const a=m[1] as 'COW'|'SHEEP',q=Math.max(0,Math.min(+m[2]||1,t[a]-x[a]));if(q>0){o.market.push(['BUY_ANIMAL',a,q]);x[a]+=q;}}else o.market.push(m);}return o;}\nfunction capFor(v:string,obs:any){if(v==='PREEMPT')return 16;const k=(obs.town?.unlocked_shops??[]).length;return Math.min(16,(v==='RECOURSE4'?4:6)+2*k);}\nasync function play(v:string,opp:string,seed:number,seat:0|1){reset('DYN');reset(opp);let s=E.state.initGameState(2,{...cfg0,seed},seed),targets={COW:0,SHEEP:0},errors=0;while(!s.done){const acts:any[]=[];for(const pid of [0,1] as const){const n=pid===seat?'DYN':opp,obs=observationFor(s,pid);if(n==='DYN'&&+obs.hour===0){const owned={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};targets={...owned};const total=owned.COW+owned.SHEEP,cap=capFor(v,obs);const c=choosePastureAnimal(obs,{cashBuffer:400,strategicWeight:0});if(totalx.animal).filter(x=>x==='COW'||x==='SHEEP')));}try{acts[pid]=JSON.parse(String(call(n,JSON.stringify(obs))));if(n==='DYN')acts[pid]=enforce(obs,acts[pid],targets);}catch{errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]};}}s=E.interpreter.step(s,acts,{...cfg0,seed});}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,opp,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,errors};}\nconst games:any[]=[];for(const v of variants)for(const o of opponents)for(const seed of seeds){games.push(await play(v,o,seed,0));games.push(await play(v,o,seed,1));}\nconst summary:any={};for(const v of variants){const rs=games.filter(g=>g.v===v);summary[v]={episodes:rs.length,wins:rs.filter(r=>r.win).length,winRate:rs.filter(r=>r.win).length/rs.length,avgBank:rs.reduce((a,r)=>a+r.ours,0)/rs.length,avgMargin:rs.reduce((a,r)=>a+r.margin,0)/rs.length,byOpponent:{}};for(const o of opponents){const xs=rs.filter(r=>r.opp===o);summary[v].byOpponent[o]={wins:xs.filter(r=>r.win).length,winRate:xs.filter(r=>r.win).length/xs.length,avgMargin:xs.reduce((a,r)=>a+r.margin,0)/xs.length};}}\nconst p=games.filter(g=>g.v==='PREEMPT');const paired=variants.slice(1).map(v=>{const r=games.filter(g=>g.v===v);return{variant:v,avgPairedMarginDelta:r.reduce((s,g,i)=>s+(g.margin-p[i].margin),0)/r.length,betterEpisodes:r.filter((g,i)=>g.margin>p[i].margin).length,worseEpisodes:r.filter((g,i)=>g.margin{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number){const r=rng(2317000000+ci*104729);return Array.from({length:3},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,991919),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nfunction simulateRace(full:any){const active=new Set(['H','W','B']),pruneAt:any={},audit:any[]=[];for(const n of [4,8,12]){const stage:any={n,activeBefore:[...active],pruned:[],stats:{}};for(const m of [...active]){const s=stat(full,m,n);stage.stats[m]=s;if(s.ucb<0){active.delete(m);pruneAt[m]=n;stage.pruned.push(m)}}stage.activeAfter=[...active];audit.push(stage)}const raced:any={N:full.N};for(const m of ['H','W','B'])raced[m]=full[m].slice(0,pruneAt[m]??16);const t=teacher(raced),episodesUsed=16+['H','W','B'].reduce((s,m)=>s+(pruneAt[m]??16),0);return{pick:t.pick,pruneAt,audit,episodesUsed,savingsPct:100*(1-episodesUsed/64),teacherAudit:t.audit}}\nconst contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,fullPick:full.pick,racePick:race.pick,equivalent:full.pick===race.pick,falsePrune:full.pick!=='N'&&race.pruneAt[full.pick]!=null,fullTeacherAudit:full.audit,race})}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'RACING_TEACHER_EQUIVALENCE_CLEO_COMPACT',contexts:contexts.map(x=>({contextId:x.contextId,prefix:x.prefix,fullPick:x.fullPick,racePick:x.racePick,equivalent:x.equivalent,falsePrune:x.falsePrune,pruneAt:x.race.pruneAt,episodesUsed:x.race.episodesUsed,savingsPct:x.race.savingsPct})),aggregate:{equivalent:contexts.filter(x=>x.equivalent).length,total:contexts.length,falsePrunes:contexts.filter(x=>x.falsePrune).length,avgSavingsPct:avg(contexts.map(x=>x.race.savingsPct))}},null,2));",[],[1224],"a9aac483-39e4-43bf-9e72-6af272ec6a2a","kag_racing_teacher_equivalence_broker_d6_v1.ts","01a07a73-562b-77ac-9331-87c25e9b5f9a",337,["D",1788760813229],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=6,modes=['N','H','W','B'],INNER=16,CONTEXTS=2;const cfg=E.state.resolveConfig({seed:992929,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number){const r=rng(2517000000+ci*104729);return Array.from({length:2},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,992929),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nfunction simulateRace(full:any){const active=new Set(['H','W','B']),pruneAt:any={},audit:any[]=[];for(const n of [4,8,12]){const stage:any={n,activeBefore:[...active],pruned:[],stats:{}};for(const m of [...active]){const s=stat(full,m,n);stage.stats[m]=s;if(s.ucb<0){active.delete(m);pruneAt[m]=n;stage.pruned.push(m)}}stage.activeAfter=[...active];audit.push(stage)}const raced:any={N:full.N};for(const m of ['H','W','B'])raced[m]=full[m].slice(0,pruneAt[m]??16);const t=teacher(raced),episodesUsed=16+['H','W','B'].reduce((s,m)=>s+(pruneAt[m]??16),0);return{pick:t.pick,pruneAt,audit,episodesUsed,savingsPct:100*(1-episodesUsed/64),teacherAudit:t.audit}}\nconst contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,fullPick:full.pick,racePick:race.pick,equivalent:full.pick===race.pick,falsePrune:full.pick!=='N'&&race.pruneAt[full.pick]!=null,fullTeacherAudit:full.audit,race})}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'RACING_TEACHER_EQUIVALENCE_BROKER_D6_COMPACT',contexts:contexts.map(x=>({contextId:x.contextId,prefix:x.prefix,fullPick:x.fullPick,racePick:x.racePick,equivalent:x.equivalent,falsePrune:x.falsePrune,pruneAt:x.race.pruneAt,episodesUsed:x.race.episodesUsed,savingsPct:x.race.savingsPct})),aggregate:{equivalent:contexts.filter(x=>x.equivalent).length,total:contexts.length,falsePrunes:contexts.filter(x=>x.falsePrune).length,avgSavingsPct:avg(contexts.map(x=>x.race.savingsPct))}},null,2));",[],[1232],"3d150d92-6d57-4a6c-a4e7-4d528717c9fe","kag_racing_teacher_equivalence_broker_d12_v1.ts","01a07a75-ec0c-71de-b6ef-563137e80265",343,["D",1788760985945],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=12,modes=['N','H','W','B'],INNER=16,CONTEXTS=2;const cfg=E.state.resolveConfig({seed:993939,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number){const r=rng(2717000000+ci*104729);return Array.from({length:4},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,993939),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nfunction simulateRace(full:any){const active=new Set(['H','W','B']),pruneAt:any={},audit:any[]=[];for(const n of [4,8,12]){const stage:any={n,activeBefore:[...active],pruned:[],stats:{}};for(const m of [...active]){const s=stat(full,m,n);stage.stats[m]=s;if(s.ucb<0){active.delete(m);pruneAt[m]=n;stage.pruned.push(m)}}stage.activeAfter=[...active];audit.push(stage)}const raced:any={N:full.N};for(const m of ['H','W','B'])raced[m]=full[m].slice(0,pruneAt[m]??16);const t=teacher(raced),episodesUsed=16+['H','W','B'].reduce((s,m)=>s+(pruneAt[m]??16),0);return{pick:t.pick,pruneAt,audit,episodesUsed,savingsPct:100*(1-episodesUsed/64),teacherAudit:t.audit}}\nconst contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,fullPick:full.pick,racePick:race.pick,equivalent:full.pick===race.pick,falsePrune:full.pick!=='N'&&race.pruneAt[full.pick]!=null,fullTeacherAudit:full.audit,race})}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'RACING_TEACHER_EQUIVALENCE_BROKER_D12_COMPACT',contexts:contexts.map(x=>({contextId:x.contextId,prefix:x.prefix,fullPick:x.fullPick,racePick:x.racePick,equivalent:x.equivalent,falsePrune:x.falsePrune,pruneAt:x.race.pruneAt,episodesUsed:x.race.episodesUsed,savingsPct:x.race.savingsPct})),aggregate:{equivalent:contexts.filter(x=>x.equivalent).length,total:contexts.length,falsePrunes:contexts.filter(x=>x.falsePrune).length,avgSavingsPct:avg(contexts.map(x=>x.race.savingsPct))}},null,2));",[],[1240],"afa79267-a8ec-41d2-941d-4970af858e31","kag_racing_teacher_equivalence_batch2_v1.ts","01a07a6b-fb05-742b-9352-809f600a23d8",324,["D",1788760456135],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=2;const cfg=E.state.resolveConfig({seed:989898,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number){const r=rng(2117000000+ci*104729);return Array.from({length:3},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,989898),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nfunction simulateRace(full:any){const active=new Set(['H','W','B']),pruneAt:any={},audit:any[]=[];for(const n of [4,8,12]){const stage:any={n,activeBefore:[...active],pruned:[],stats:{}};for(const m of [...active]){const s=stat(full,m,n);stage.stats[m]=s;if(s.ucb<0){active.delete(m);pruneAt[m]=n;stage.pruned.push(m)}}stage.activeAfter=[...active];audit.push(stage)}const raced:any={N:full.N};for(const m of ['H','W','B'])raced[m]=full[m].slice(0,pruneAt[m]??16);const t=teacher(raced),episodesUsed=16+['H','W','B'].reduce((s,m)=>s+(pruneAt[m]??16),0);return{pick:t.pick,pruneAt,audit,episodesUsed,savingsPct:100*(1-episodesUsed/64),teacherAudit:t.audit}}\nconst contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,fullPick:full.pick,racePick:race.pick,equivalent:full.pick===race.pick,falsePrune:full.pick!=='N'&&race.pruneAt[full.pick]!=null,fullTeacherAudit:full.audit,race})}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'RACING_TEACHER_EQUIVALENCE_BATCH2_COMPACT',contexts:contexts.map(x=>({contextId:x.contextId,prefix:x.prefix,fullPick:x.fullPick,racePick:x.racePick,equivalent:x.equivalent,falsePrune:x.falsePrune,pruneAt:x.race.pruneAt,episodesUsed:x.race.episodesUsed,savingsPct:x.race.savingsPct})),aggregate:{equivalent:contexts.filter(x=>x.equivalent).length,total:contexts.length,falsePrunes:contexts.filter(x=>x.falsePrune).length,avgSavingsPct:avg(contexts.map(x=>x.race.savingsPct))}},null,2));",[],[1248],"3e1b3795-3569-4830-b41c-98a6691b9708","kag_racing_teacher_equivalence_batch1_v1.ts","01a07a69-5950-751a-97af-d846b76637bf",319,["D",1788760172552],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=2;const cfg=E.state.resolveConfig({seed:979797,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number){const r=rng(1917000000+ci*104729);return Array.from({length:3},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,979797),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nfunction simulateRace(full:any){const active=new Set(['H','W','B']),pruneAt:any={},audit:any[]=[];for(const n of [4,8,12]){const stage:any={n,activeBefore:[...active],pruned:[],stats:{}};for(const m of [...active]){const s=stat(full,m,n);stage.stats[m]=s;if(s.ucb<0){active.delete(m);pruneAt[m]=n;stage.pruned.push(m)}}stage.activeAfter=[...active];audit.push(stage)}const raced:any={N:full.N};for(const m of ['H','W','B'])raced[m]=full[m].slice(0,pruneAt[m]??16);const t=teacher(raced),episodesUsed=16+['H','W','B'].reduce((s,m)=>s+(pruneAt[m]??16),0);return{pick:t.pick,pruneAt,audit,episodesUsed,savingsPct:100*(1-episodesUsed/64),teacherAudit:t.audit}}\nconst contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,fullPick:full.pick,racePick:race.pick,equivalent:full.pick===race.pick,falsePrune:full.pick!=='N'&&race.pruneAt[full.pick]!=null,fullTeacherAudit:full.audit,race})}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'RACING_TEACHER_EQUIVALENCE_BATCH1',contexts,aggregate:{equivalent:contexts.filter(x=>x.equivalent).length,total:contexts.length,falsePrunes:contexts.filter(x=>x.falsePrune).length,avgSavingsPct:avg(contexts.map(x=>x.race.savingsPct))}},null,2));",[],[1256],"3e83d048-ae92-46e0-8ece-7c25dabeb24e","kag_public_top_round_robin_v1.ts","81cd5340-aa66-11f1-a771-1607ee4eb77e",184,["D",1788749266616],"import { unzipSync, strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine, observationFor } from './lib/kaggriculture_engine_loader.ts';\n\nconst refs = ['broker_bea.py','ledger_lena.py','slotter_silas.py','closer_cleo.py'];\nconst seeds = [31013,31121,31231,31337,31469,31583,31699,31817];\nconst E = await loadOfficialKaggricultureEngine();\nconst cfg0 = E.state.resolveConfig();\nconst zr = await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');\nif (!zr.ok) throw new Error(`reference dataset ${zr.status}`);\nconst z = unzipSync(new Uint8Array(await zr.arrayBuffer()));\nconst py = await loadPyodide();\npy.runPython(`import json\\n_SRC={}\\n_REG={}`);\nfor (const f of refs) {\n py.globals.set('nm', f);\n py.globals.set('src', strFromU8(z[f]));\n py.runPython(`_SRC[nm]=src`);\n}\npy.runPython(`\ndef reset_pair(a,b):\n global _REG\n _REG={}\n for n in (a,b):\n ns={}\n exec(_SRC[n],ns)\n _REG[n]=ns['agent']\n\ndef call_ref(name,payload):\n return json.dumps(_REG[name](json.loads(payload)))\n`);\nconst resetPair = py.globals.get('reset_pair');\nconst callRef = py.globals.get('call_ref');\n\ntype G={a:string,b:string,seed:number,aSeat:0|1,aMoney:number,bMoney:number,margin:number,winner:string,tie:boolean,errors:number};\nasync function play(a:string,b:string,seed:number,aSeat:0|1):Promise{\n resetPair(a,b);\n let s=E.state.initGameState(2,{...cfg0,seed},seed), errors=0;\n while(!s.done){\n const acts:any[]=[];\n for(const pid of [0,1] as const){\n const name=pid===aSeat?a:b;\n const o=observationFor(s,pid);\n try{acts[pid]=JSON.parse(String(callRef(name,JSON.stringify(o))))}\n catch{errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]}}\n }\n s=E.interpreter.step(s,acts,{...cfg0,seed});\n }\n const aMoney=s.farms[aSeat].money,bMoney=s.farms[1-aSeat].money;\n return {a,b,seed,aSeat,aMoney,bMoney,margin:aMoney-bMoney,winner:aMoney===bMoney?'TIE':aMoney>bMoney?a:b,tie:aMoney===bMoney,errors};\n}\n\nconst games:G[]=[];\nfor(let i=0;ixs.reduce((a,b)=>a+b,0)/(xs.length||1);\nconst score:any={};\nfor(const r of refs) score[r]={games:0,wins:0,losses:0,ties:0,avgBank:0,avgMargin:0,errors:0,bankSamples:[] as number[],marginSamples:[] as number[]};\nfor(const g of games){\n for(const r of [g.a,g.b]){\n const isA=r===g.a,money=isA?g.aMoney:g.bMoney,margin=isA?g.margin:-g.margin;\n const x=score[r];x.games++;x.bankSamples.push(money);x.marginSamples.push(margin);x.errors+=g.errors;\n if(g.tie)x.ties++; else if(g.winner===r)x.wins++; else x.losses++;\n }\n}\nfor(const r of refs){score[r].avgBank=avg(score[r].bankSamples);score[r].avgMargin=avg(score[r].marginSamples);score[r].winRate=score[r].wins/score[r].games;delete score[r].bankSamples;delete score[r].marginSamples;}\nconst pairwise:any={};\nfor(let i=0;ig.a===a&&g.b===b);\n pairwise[`${a}__vs__${b}`]={episodes:x.length,aWins:x.filter(g=>g.winner===a).length,bWins:x.filter(g=>g.winner===b).length,ties:x.filter(g=>g.tie).length,aAvgBank:avg(x.map(g=>g.aMoney)),bAvgBank:avg(x.map(g=>g.bMoney)),aAvgMargin:avg(x.map(g=>g.margin)),errors:x.reduce((s,g)=>s+g.errors,0)};\n}\nconst ranking=refs.map(r=>({ref:r,...score[r]})).sort((a,b)=>b.winRate-a.winRate||b.avgMargin-a.avgMargin||b.avgBank-a.avgBank);\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'PUBLIC_REFERENCE_COMPETITIVE_ROUND_ROBIN_NOT_CAUSAL_ATTRIBUTION',design:{refs,seeds,mirroredSeats:true,episodes:games.length,caveat:'Competitive ranking only. Policy-dependent RNG can alter exogenous trajectories; do not interpret pairwise deltas as isolated causal effects.'},ranking,pairwise},null,2));",[],[1264],"e61392d9-c464-4d0b-8d83-3f8b5713539b","kag_public_melon_tile_probe_v1.ts","14a6f01a-aa60-11f1-9c4b-1607ee4eb77e",157,["D",1788746506491],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig(),seed=35003;const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['melon_mateo.py']));py.runPython(`import json\\n_n={}\\nexec(src,_n)\\ndef a(p):return json.dumps(_n['agent'](json.loads(p)))`);const a=py.globals.get('a');let s=E.state.initGameState(2,{...cfg,seed},seed);let found:any=null;while(!s.done&&!found){const acts:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);acts[pid]=JSON.parse(String(a(JSON.stringify(o))));if(pid===0){const opp=o.farms[1];for(const row of opp.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.kind==='PLANT'&&t.crop==='MELON'){found={day:o.day,hour:o.hour,tile:t};break}}}s=E.interpreter.step(s,acts,{...cfg,seed})}console.log(JSON.stringify({found},null,2));",[],[1272],"1fe2335f-3102-467b-802d-3eacf600975c","kag_prefeed_recourse_probe_v1.ts","a156e33e-aa5c-11f1-8b6e-1607ee4eb77e",134,["D",1788745024534],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={BASE16:{float:16,prefeed:0},FLOAT4_ONLY:{float:4,prefeed:0},PREFEED4_FLOAT4:{float:4,prefeed:4},PREFEED4_FLOAT2:{float:2,prefeed:4},PREFEED6_FLOAT2:{float:2,prefeed:6}};const refs=['rancher_rita.py','broker_bea.py'],seeds=[24007,24109,24223];\nconst E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();const rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\\ndef prep(ff):\\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['feed_float_days']=int(ff)\\ndef seth(c,s,o,pf):\\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o);_D['POLICY']['prefeed_stock']=int(pf)\\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}\nasync function play(v:string,ref:string,seed:number,seat:0|1){const cfg=variants[v];prep(cfg.float);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0,th:any={},feedBuy=0;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){for(const k of [2,4,8,12,14,16])if(th[k]===undefined&&flock>=k)th[k]=+o.day;const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);const order=c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP');seth(tg.COW,tg.SHEEP,JSON.stringify(order),cfg.prefeed*(tg.COW+tg.SHEEP))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']),cfg.prefeed*(tg.COW+tg.SHEEP))}const a=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg);for(const q of a.market??[])if(q?.[0]==='BUY_PRODUCT'&&q?.[1]==='WHEAT')feedBuy+=Number(q[2]??0);aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,maxFlock,finalFlock,animalLoss:maxFlock-finalFlock,th,feedBuy}}\nconst games:any[]=[];for(const v of Object.keys(variants))for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const sum:any={};for(const v of Object.keys(variants)){const x=games.filter(g=>g.v===v);sum[v]={episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),avgFeedBuy:avg(x.map(g=>g.feedBuy)),reach8Day:avg(x.filter(g=>g.th[8]!=null).map(g=>g.th[8])),reach14Share:x.filter(g=>g.th[14]!=null).length/x.length,reach14Day:avg(x.filter(g=>g.th[14]!=null).map(g=>g.th[14])),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);sum[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin)),avgAnimalLoss:avg(y.map(g=>g.animalLoss))}}}const b=games.filter(g=>g.v==='BASE16');const paired=Object.keys(variants).slice(1).map(v=>{const x=games.filter(g=>g.v===v);return{v,bankDelta:avg(x.map((g,i)=>g.ours-b[i].ours)),oppDelta:avg(x.map((g,i)=>g.theirs-b[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-b[i].margin)),better:x.filter((g,i)=>g.margin>b[i].margin).length,worse:x.filter((g,i)=>g.marging.win&&!b[i].win).length,lossFlips:x.filter((g,i)=>!g.win&&b[i].win).length}});console.log(JSON.stringify({status:'PASS',hypothesis:'PHYSICALIZE_NEAR_TERM_FEED_BEFORE_ACCELERATING_HERD_INSTEAD_OF_HOLDING_A_LARGE_STATIC_CASH_FLOAT',independentDesign:variants,summary:sum,paired},null,2));",[],[1280],"fe0f2317-a882-4163-9ba6-29d967cf2950","kag_pasture_labor_frontier_v1.ts","01a079d6-978e-710e-9573-7c685e0e4e84",204,["D",1788750551948],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['BASE','J2B0_H8','J2B1_H8','J4B0_H8','J4B1_H8','J4B2_H8','J6B2_H8','J4B2_H6','J4B2_H10'];const E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:969696,weedSpawnChance:0,townShopUnlockInterval:9999});const shops=['SMOOTHIE_SHOP','FARMERS_MARKET','ICE_CREAM_SHOP','YARN_STORE','FARMERS_MARKET','BAKERY','SMOOTHIE_SHOP','BAKERY'];\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['broker_bea.py']));py.runPython(`import json\\n_b={}\\nexec(src,_b)\\n_B=_b['agent']`);py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cb(p):return json.dumps(_B(json.loads(p)))\ndef cm(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setops(h,b,l):\n _D['POLICY']['hands']=int(h);_D['POLICY']['build'][0]['target']=int(b);_D['POLICY']['land_after_animals']=int(l)\n`);const cb=py.globals.get('cb'),cm=py.globals.get('cm'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setops=py.globals.get('setops');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number){return shops.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}function pub(f:any){let crops=0,pastures=0,animals=0;for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f?.unlocked_quadrants??[]).length,crops,pastures,animals}}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction ops(v:string,day:number,ownedAnimals:number){if(v==='BASE')return{hands:8,build:16,landAfter:10};const m=v.match(/^J(\\d)B(\\d)_H(\\d+)$/);if(!m)throw new Error('BAD_VARIANT_'+v);const floor=Number(m[1]),buffer=Number(m[2]),hands=Number(m[3]);return{hands,build:Math.max(floor,Math.min(16,ownedAnimals+buffer)),landAfter:10}}\nasync function play(v:string,seat:0|1){prep();let s=E.state.initGameState(2,cfg,969696),tg:any={COW:0,SHEEP:0},comm=false,minCash=Infinity,maxFlock=0,finalFlock=0;const days:any[]=[];while(!s.done){s.town.unlocked_shops=fixed(+s.day);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){const owned=own(o,'COW')+own(o,'SHEEP'),op=ops(v,+o.day,owned);setops(op.hands,op.build,op.landAfter);if(+o.hour===0){days.push({day:+o.day,...pub(o.farms[pid]),handsTarget:op.hands,buildTarget:op.build});const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[pid]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg));minCash=Math.min(minCash,+o.farms[pid].money);maxFlock=Math.max(maxFlock,owned)}else aa[pid]=JSON.parse(String(cb(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{v,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,win:s.farms[seat].money>s.farms[1-seat].money,minCash,animalLoss:maxFlock-finalFlock,days}}\nconst games:any[]=[];for(const v of variants){games.push(await play(v,0));games.push(await play(v,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const b=games.filter(g=>g.v==='BASE'),summary:any={};for(const v of variants){const x=games.filter(g=>g.v===v),d=x.map((g,i)=>g.margin-b[i].margin);summary[v]={bank:avg(x.map(g=>g.ours)),opp:avg(x.map(g=>g.theirs)),margin:avg(x.map(g=>g.margin)),wins:x.filter(g=>g.win).length,minCash:Math.min(...x.map(g=>g.minCash)),animalLoss:avg(x.map(g=>g.animalLoss)),delta:avg(d),snapshots:[1,3,6,9,12,15,18].map(day=>{const r=x.map(g=>g.days.find((q:any)=>q.day===day)).filter(Boolean);return{day,cash:avg(r.map((q:any)=>q.cash)),land:avg(r.map((q:any)=>q.land)),crops:avg(r.map((q:any)=>q.crops)),pastures:avg(r.map((q:any)=>q.pastures)),animals:avg(r.map((q:any)=>q.animals)),handsTarget:avg(r.map((q:any)=>q.handsTarget)),buildTarget:avg(r.map((q:any)=>q.buildTarget))}})}}const ranking=Object.entries(summary).map(([v,x]:any)=>({v,bank:x.bank,opp:x.opp,margin:x.margin,delta:x.delta,minCash:x.minCash,animalLoss:x.animalLoss,d1:x.snapshots.find((q:any)=>q.day===1),d12:x.snapshots.find((q:any)=>q.day===12),d18:x.snapshots.find((q:any)=>q.day===18)})).sort((a,b)=>b.delta-a.delta);console.log(JSON.stringify({status:'PASS',evidenceState:'FIXED_WORLD_PASTURE_LABOR_FRONTIER',ranking},null,2));",[],[1288],"41829b10-3851-4a53-813a-1714e60d2668","kag_outcome_boundary_census_v1.ts","01a07ad1-410f-765c-b59e-280851e31584",495,["D",1788767030813],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';import { summarizeAction,chooseDualObjective } from './lib/kag_teacher_stats_v2.ts';\nconst E=await loadOfficialKaggricultureEngine(),decisionDay=9,INNER=6;const OPP_FILES=['rotation_rosa.py','homestead_hana.py','melon_mateo.py','rancher_rita.py','broker_bea.py'];const PREFIXES=[['PIZZA_SHOP','PET_CAFE','ICE_CREAM_SHOP'],['YARN_STORE','FARMERS_MARKET','BRUNCH_SPOT'],['BAKERY','BAKERY','SMOOTHIE_SHOP'],['ICE_CREAM_SHOP','YARN_STORE','FARMERS_MARKET']];\nconst MODES=['N'] as const;\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`_OPPS={}`);for(const f of OPP_FILES){py.globals.set('src',strFromU8(z[f]));py.globals.set('oppkey',f);py.runPython(`_o={}\\nexec(src,_o)\\n_OPPS[oppkey]=_o['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`import json\\n_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(k,p):return json.dumps(_OPPS[k](json.loads(p)))\ndef cm(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setops(h,b):\n _D['POLICY']['hands']=int(h);_D['POLICY']['build'][0]['target']=int(b);_D['POLICY']['land_after_animals']=10\n`);const cr=py.globals.get('cr'),cm=py.globals.get('cm'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setops=py.globals.get('setops');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number,shops:string[]){return shops.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(prefix:string[],opponent:string,futureSeed:number){prep();const mode='N',baseSeed=994949,cfg=E.state.resolveConfig({seed:baseSeed,weedSpawnChance:0.005,townShopUnlockInterval:3});let s=E.state.initGameState(2,cfg,baseSeed),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null,forked=false;while(!s.done){if(+s.day<=decisionDay)s.town.unlocked_shops=fixed(+s.day,prefix);if(!forked&&+s.day===decisionDay&&+s.hour===0){s.seed=futureSeed;forked=true}const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(opponent,JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,finalShops:s.town.unlocked_shops}}\n// Fixed-N confirmatory statistics use sample SD + paired t interval and an exact one-sided sign gate from kag_teacher_stats_v2.ts.\nconst results:any[]=[];for(const opponent of OPP_FILES){for(let pi=0;pi();for(let i=0;ix>0).length,ties=margins.filter(x=>x===0).length,losses=margins.filter(x=>x<0).length,marginMean=margins.reduce((a,b)=>a+b,0)/margins.length,marginMin=Math.min(...margins),marginMax=Math.max(...margins);results.push({opponent,prefix,preStateHashes:preHashes.size,preStateIdentical:preHashes.size===1,n:INNER,wins,ties,losses,winRate:wins/INNER,scoreMean:(wins+0.5*ties)/INNER,marginMean,marginMin,marginMax,boundary:wins>0&&losses>0})}}const boundaryCandidates=results.filter(r=>r.boundary).sort((a,b)=>Math.abs(a.marginMean)-Math.abs(b.marginMean));\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'OUTCOME_BOUNDARY_CENSUS_CONDITIONAL_FORK_V1',decisionDay,innerScenarios:INNER,opponents:OPP_FILES,prefixCount:PREFIXES.length,totalCells:results.length,allPreStatesIdentical:results.every(r=>r.preStateIdentical),boundaryCandidates,results},null,2));",[],[1296],"4102a2da-d411-452d-bf63-9c365f81a4ed","kag_opponent_state_probe_v1.ts","01a07ad0-43e1-764a-bb9a-d60e697320fe",488,["D",1788766886055],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';import { summarizeAction,chooseDualObjective } from './lib/kag_teacher_stats_v2.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=32;\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const PREFIX=['ICE_CREAM_SHOP','YARN_STORE','FARMERS_MARKET'];function innerWorlds(){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(prefix:string[],mode:string,futureSeed:number){prep();const baseSeed=994949,cfg=E.state.resolveConfig({seed:baseSeed,weedSpawnChance:0.005,townShopUnlockInterval:3});let s=E.state.initGameState(2,cfg,baseSeed),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null,forked=false;while(!s.done){if(+s.day<=decisionDay)s.town.unlocked_shops=fixed(+s.day,prefix);if(!forked&&+s.day===decisionDay&&+s.hour===0){s.seed=futureSeed;forked=true}const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,finalShops:s.town.unlocked_shops}}\n// Fixed-N confirmatory statistics use sample SD + paired t interval and an exact one-sided sign gate from kag_teacher_stats_v2.ts.\nconst tests:any[]=[];for(const mode of ['N','H','W','B']){for(const futureSeed of [3333333,4444444]){const a=await play(PREFIX,mode,futureSeed),b=await play(PREFIX,mode,futureSeed);tests.push({mode,futureSeed,marginA:a.margin,marginB:b.margin,preEqual:JSON.stringify(a.pre)===JSON.stringify(b.pre),shopsEqual:JSON.stringify(a.finalShops)===JSON.stringify(b.finalShops),marginEqual:a.margin===b.margin})}}const deterministic=tests.every(x=>x.preEqual&&x.shopsEqual&&x.marginEqual);\nconsole.log(JSON.stringify({status:deterministic?'PASS':'FAIL',evidenceState:'BROKER_OPPONENT_STATE_REPEATABILITY_PROBE_V1',opponent:'broker_bea.py',sameProcessRepeatedEpisodes:true,deterministic,tests},null,2));",[],[1304],"ed0bd52d-45d3-4b8a-85ba-d23e27f7bc7d","kag_opponent_early_signature_probe_v1.ts","f2283804-aa60-11f1-ab00-1607ee4eb77e",162,["D",1788746878124],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nconst refs=['melon_mateo.py','rancher_rita.py','broker_bea.py','ledger_lena.py','slotter_silas.py','closer_cleo.py'],seeds=[39007,39113,39229];const E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();const rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.runPython(`\\ndef call(n,p):return json.dumps(_REG[n](json.loads(p)))`);const call=py.globals.get('call');\nfunction sig(f:any){const crops:any={},animals:any={};for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops[t.crop]=(crops[t.crop]??0)+1;if(t.animal)animals[t.animal]=(animals[t.animal]??0)+1}return{money:+f.money,land:(f.unlocked_quadrants??[]).length,crops,animals}}\nconst rows:any[]=[];for(const ref of refs)for(const seed of seeds){let s=E.state.initGameState(2,{...cfg0,seed},seed),seen=new Set();while(!s.done&&s.day<=9){const acts:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);acts[pid]=JSON.parse(String(call(ref,JSON.stringify(o))));if(pid===0&&+o.hour===0&&[1,3,6,9].includes(+o.day)&&!seen.has(+o.day)){seen.add(+o.day);rows.push({ref,seed,day:+o.day,...sig(o.farms[1])})}}s=E.interpreter.step(s,acts,{...cfg0,seed})}}\nfunction avg(xs:number[]){return xs.reduce((a,b)=>a+b,0)/(xs.length||1)}const summary:any={};for(const ref of refs){summary[ref]={};for(const day of [1,3,6,9]){const x=rows.filter(r=>r.ref===ref&&r.day===day),cropKeys=['WHEAT','MELON','STRAWBERRY','CARROT','TOMATO'],animalKeys=['COW','SHEEP','GOOSE'];const crops:any={},animals:any={};for(const k of cropKeys)crops[k]=avg(x.map(r=>r.crops[k]??0));for(const k of animalKeys)animals[k]=avg(x.map(r=>r.animals[k]??0));summary[ref][day]={money:avg(x.map(r=>r.money)),land:avg(x.map(r=>r.land)),crops,animals,totalCrops:Object.values(crops).reduce((a:any,b:any)=>a+b,0),totalAnimals:Object.values(animals).reduce((a:any,b:any)=>a+b,0)}}}console.log(JSON.stringify({status:'PASS',question:'HOW_EARLY_CAN_PUBLIC_FARM_STATE_DISTINGUISH_OPPONENT_ECONOMIC_FAMILIES_WITHOUT_PRIVATE_INFORMATION',summary},null,2));",[],[1312],"c5def592-04a1-468b-aebf-b74ce2d17e1e","kag_official_runtime_budget_probe_v1.ts","1719fc42-aa7a-11f1-b9aa-1607ee4eb77e",294,["D",1788757677534],"const u='https://raw.githubusercontent.com/Kaggle/kaggle-environments/master/kaggle_environments/envs/kaggriculture/kaggriculture.json';const r=await fetch(u);if(!r.ok)throw new Error(`HTTP ${r.status}`);const j=await r.json();console.log(JSON.stringify({status:'PASS',source:u,version:j.version,name:j.name,configuration:j.configuration,agents:j.agents,timeoutFields:Object.fromEntries(Object.entries(j.configuration??{}).filter(([k])=>/timeout|episode|step|act|run/i.test(k)))},null,2));",[],[1320],"05db558f-a095-402e-bf9e-01dfd1e3874d","kag_observation_schema_probe_v1.ts","4911d3ea-aa85-11f1-be51-1607ee4eb77e",384,["D",1788762485837],"import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nconst E=await loadOfficialKaggricultureEngine();const cfg=E.state.resolveConfig({seed:123456});let s=E.state.initGameState(2,cfg,123456);const o=observationFor(s,0);\nfunction paths(x:any,p='',out:any[]=[]){if(x==null||typeof x!=='object')return out;if(Array.isArray(x)){for(let i=0;it&&typeof t==='object');console.log(JSON.stringify({status:'PASS',topKeys:Object.keys(o),farmKeys:Object.keys(o.farms?.[0]??{}),privateKeys:Object.keys(o.private??{}),private:o.private,sampleTile,sampleTileKeys:Object.keys(sampleTile??{}),interestingPaths:paths(o)},null,2));",[],[1328],"99190cf6-c717-47ae-b4b5-2ab1de79023c","kag_noise_broker_strongh_v1.ts","01a07aad-eddf-7425-b391-abd96ca6392a",465,["D",1788764631653],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=12;\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const PREFIX=['PIZZA_SHOP','PET_CAFE','ICE_CREAM_SHOP'];function innerWorlds(){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(prefix:string[],mode:string,futureSeed:number){prep();const baseSeed=994949,cfg=E.state.resolveConfig({seed:baseSeed,weedSpawnChance:0.005,townShopUnlockInterval:3});let s=E.state.initGameState(2,cfg,baseSeed),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null,forked=false;while(!s.done){if(+s.day<=decisionDay)s.town.unlocked_shops=fixed(+s.day,prefix);if(!forked&&+s.day===decisionDay&&+s.hour===0){s.seed=futureSeed;forked=true}const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,finalShops:s.town.unlocked_shops}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst vals:any={N:[],H:[],W:[],B:[]},preHashes=new Set(),scenarioShops:any[]=[];for(let i=0;i{const q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(INNER),tc=tcrit[INNER];return{action:m,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/INNER}});const t=teacher(vals);\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'BROKER_STRONG_H_CONDITIONAL_FUTURE_FORK_V1',prefix:PREFIX,basePastSeed:994949,officialWeedSpawnChance:0.005,officialFutureShopUnlock:true,futureSeedForkAt:{day:9,hour:0},pairedAcrossActions:true,innerScenarios:INNER,preStateHashes:preHashes.size,preStateIdenticalAcrossAllForks:preHashes.size===1,teacherPick:t.pick,teacherAudit:t.audit,rows,scenarioShops},null,2));",[],[1336],"c03251ad-d695-446e-9ad4-d406ebdd894b","kag_noise_broker_boundary_v1.ts","01a07aa8-92f9-753a-9191-5388803895fb",462,["D",1788764530446],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=12;\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const PREFIX=['ICE_CREAM_SHOP','YARN_STORE','FARMERS_MARKET'];function innerWorlds(){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(prefix:string[],mode:string,futureSeed:number){prep();const baseSeed=994949,cfg=E.state.resolveConfig({seed:baseSeed,weedSpawnChance:0.005,townShopUnlockInterval:3});let s=E.state.initGameState(2,cfg,baseSeed),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null,forked=false;while(!s.done){if(+s.day<=decisionDay)s.town.unlocked_shops=fixed(+s.day,prefix);if(!forked&&+s.day===decisionDay&&+s.hour===0){s.seed=futureSeed;forked=true}const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,finalShops:s.town.unlocked_shops}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst vals:any={N:[],H:[],W:[],B:[]},preHashes=new Set(),scenarioShops:any[]=[];for(let i=0;i{const q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(INNER),tc=tcrit[INNER];return{action:m,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/INNER}});const t=teacher(vals);\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'BROKER_BOUNDARY_CONDITIONAL_FUTURE_FORK_V2',prefix:PREFIX,basePastSeed:994949,officialWeedSpawnChance:0.005,officialFutureShopUnlock:true,futureSeedForkAt:{day:9,hour:0},pairedAcrossActions:true,innerScenarios:INNER,preStateHashes:preHashes.size,preStateIdenticalAcrossAllForks:preHashes.size===1,teacherPick:t.pick,teacherAudit:t.audit,rows,scenarioShops},null,2));",[],[1344],"c6f85872-8110-420c-a26c-08b71b4640d0","kag_noise_broker_boundary_b2.ts","01a07aaf-50a4-75d7-8905-541cd382f2fb",468,["D",1788764722062],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=12;\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const PREFIX=['ICE_CREAM_SHOP','YARN_STORE','FARMERS_MARKET'];function innerWorlds(){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(prefix:string[],mode:string,futureSeed:number){prep();const baseSeed=994949,cfg=E.state.resolveConfig({seed:baseSeed,weedSpawnChance:0.005,townShopUnlockInterval:3});let s=E.state.initGameState(2,cfg,baseSeed),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null,forked=false;while(!s.done){if(+s.day<=decisionDay)s.town.unlocked_shops=fixed(+s.day,prefix);if(!forked&&+s.day===decisionDay&&+s.hour===0){s.seed=futureSeed;forked=true}const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,finalShops:s.town.unlocked_shops}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst vals:any={N:[],H:[],W:[],B:[]},preHashes=new Set(),scenarioShops:any[]=[];for(let i=0;i{const q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(INNER),tc=tcrit[INNER];return{action:m,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/INNER}});const t=teacher(vals);\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'BROKER_BOUNDARY_CONDITIONAL_FORK_BATCH2',prefix:PREFIX,basePastSeed:994949,officialWeedSpawnChance:0.005,officialFutureShopUnlock:true,futureSeedForkAt:{day:9,hour:0},pairedAcrossActions:true,innerScenarios:INNER,preStateHashes:preHashes.size,preStateIdenticalAcrossAllForks:preHashes.size===1,teacherPick:t.pick,teacherAudit:t.audit,rows,scenarioShops},null,2));",[],[1352],"6a9b5cef-9884-40f2-b76e-b948dba70d3b","kag_mix11_ops_ablation_v1.ts","c5ad48fa-aa56-11f1-8596-1607ee4eb77e",122,["D",1788742508545],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['BASE','ROUTE','DROP','FERT','OPS_COMBO'];const refs=['rancher_rita.py','broker_bea.py'];const seeds=[19001,19121,19237,19373,19483,19597];\nconst E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();const rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\\ndef reset(v):\\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0}\\n _D['POLICY']['persistent_routing']=v in ('ROUTE','OPS_COMBO')\\n _D['POLICY']['drop_threshold']=4 if v in ('DROP','OPS_COMBO') else 0\\n _D['POLICY']['drop_after_hour']=12\\n _D['POLICY']['collect_fertilizer']=v in ('FERT','OPS_COMBO')\\ndef seth(c,s,o):\\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)}\\n _D['POLICY']['animals']=json.loads(o)\\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),reset=py.globals.get('reset'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}\nfunction enforce(o:any,act:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of act?.market??[]){if(Array.isArray(q)&&q[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const a=q[1],n=Math.max(0,Math.min(+q[2]||1,t[a]-x[a]));if(n){m.push(['BUY_ANIMAL',a,n]);x[a]+=n}}else m.push(q)}return{...act,market:m}}\nfunction addFertSell(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);if(q>0&&p>=60&&(a.market?.length??0)<10)return{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]};return a}\nfunction countOps(t:any,a:any){for(const u of [a?.farmer,...(a?.hands??[])]){const op=Array.isArray(u)?u[0]:'NONE';t[op]=(t[op]??0)+1}}\nasync function play(v:string,ref:string,seed:number,seat:0|1){reset(v);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},done=false,err=0,ops:any={};while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!done){done=ow.COW>=1&&ow.SHEEP>=1;if(!done){tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)};seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}}if(done){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')))}}let a=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg);if(v==='FERT'||v==='OPS_COMBO')a=addFertSell(o,a);aa[pid]=a;countOps(ops,a)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,ops}}\nconst games:any[]=[];for(const v of variants)for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(xs:number[])=>xs.reduce((a,b)=>a+b,0)/(xs.length||1);const summary:any={};for(const v of variants){const rs=games.filter(g=>g.v===v);summary[v]={episodes:rs.length,wins:rs.filter(g=>g.win).length,winRate:rs.filter(g=>g.win).length/rs.length,avgBank:avg(rs.map(g=>g.ours)),avgMargin:avg(rs.map(g=>g.margin)),avgPass:avg(rs.map(g=>g.ops.PASS??0)),avgCollect:avg(rs.map(g=>g.ops.COLLECT_FERTILIZER??0)),avgDrop:avg(rs.map(g=>g.ops.DROP??0)),byOpponent:{}};for(const ref of refs){const x=rs.filter(g=>g.ref===ref);summary[v].byOpponent[ref]={wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgRefBank:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),avgPass:avg(x.map(g=>g.ops.PASS??0)),avgCollect:avg(x.map(g=>g.ops.COLLECT_FERTILIZER??0)),avgDrop:avg(x.map(g=>g.ops.DROP??0))}}}\nconst b=games.filter(g=>g.v==='BASE');const paired=variants.slice(1).map(v=>{const x=games.filter(g=>g.v===v);return{variant:v,marginDelta:avg(x.map((g,i)=>g.margin-b[i].margin)),bankDelta:avg(x.map((g,i)=>g.ours-b[i].ours)),better:x.filter((g,i)=>g.margin>b[i].margin).length,worse:x.filter((g,i)=>g.margin=60'},summary,paired},null,2));",[],[1360],"dab9490a-2719-4b9b-82fe-9b1b20c82edf","kag_mix11_heldout_v1.ts","01a07935-023a-70fc-9222-ccbd433656e1",95,["D",1788739919464],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['BASE','COMMIT_MIX11'];const seeds=[12007,12113,12227,12347,12457,12569,12671,12781,12893,13001,13109,13217,13327,13441,13553,13669,13781,13883,13997,14107];const opp='RITA_BASE';\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();const zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await zr.arrayBuffer()));const src=strFromU8(z['rancher_rita.py']);const py=await loadPyodide();py.runPython(`import json,copy\\n_REG={}\\n_BASE={}`);py.globals.set('src',src);for(const [n,p] of Object.entries({DYN:{animal_target:{COW:0,SHEEP:0}},RITA_BASE:{}})){py.globals.set('nm',n);py.globals.set('patch',JSON.stringify(p));py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n\\n_BASE[nm]=copy.deepcopy(_n['POLICY'])`);}py.runPython(`\\ndef call(name,payload): return json.dumps(_REG[name]['agent'](json.loads(payload)))\\ndef set_herd(name,cows,sheep,order):\\n _REG[name]['POLICY']['animal_target']={'COW':int(cows),'SHEEP':int(sheep)}\\n _REG[name]['POLICY']['animals']=json.loads(order)\\ndef reset(name): _REG[name]['POLICY'].clear();_REG[name]['POLICY'].update(copy.deepcopy(_BASE[name]))\\n`);const call=py.globals.get('call'),setHerd=py.globals.get('set_herd'),reset=py.globals.get('reset');\nfunction owned(obs:any,a:string){const me=+obs.player,f=obs.farms[me];let n=+(obs.private?.shed?.[a]??0);for(const inv of obs.private?.inventories??[])n+=+(inv?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n;}\nfunction enforce(obs:any,act:any,t:any){const x:any={COW:owned(obs,'COW'),SHEEP:owned(obs,'SHEEP')};const o={...act,market:[] as any[]};for(const m of act?.market??[]){if(Array.isArray(m)&&m[0]==='BUY_ANIMAL'&&(m[1]==='COW'||m[1]==='SHEEP')){const a=m[1],q=Math.max(0,Math.min(+m[2]||1,t[a]-x[a]));if(q>0){o.market.push(['BUY_ANIMAL',a,q]);x[a]+=q;}}else o.market.push(m);}return o;}\nfunction initialGoal(v:string){if(v==='COMMIT_COW2')return{COW:2,SHEEP:0};if(v==='COMMIT_SHEEP2')return{COW:0,SHEEP:2};if(v==='COMMIT_MIX11')return{COW:1,SHEEP:1};return null;}\nasync function play(v:string,seed:number,seat:0|1){reset('DYN');reset(opp);let s=E.state.initGameState(2,{...cfg0,seed},seed),targets:any={COW:0,SHEEP:0},errors=0;const goal=initialGoal(v);let commitmentDone=!goal;const diary:any[]=[];while(!s.done){const acts:any[]=[];for(const pid of [0,1] as const){const n=pid===seat?'DYN':opp,obs=observationFor(s,pid);if(n==='DYN'&&+obs.hour===0){const own={COW:owned(obs,'COW'),SHEEP:owned(obs,'SHEEP')};targets={...own};if(goal&&!commitmentDone){commitmentDone=own.COW>=goal.COW&&own.SHEEP>=goal.SHEEP;if(!commitmentDone){targets={COW:Math.max(own.COW,goal.COW),SHEEP:Math.max(own.SHEEP,goal.SHEEP)};setHerd('DYN',targets.COW,targets.SHEEP,JSON.stringify(goal.SHEEP>0?['SHEEP','COW']:['COW','SHEEP']));}}\nif(commitmentDone){if(v==='WAIT3'&&+obs.day<3){setHerd('DYN',own.COW,own.SHEEP,JSON.stringify(['COW','SHEEP']));targets={...own};}else{const total=own.COW+own.SHEEP,c=choosePastureAnimal(obs,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)targets[c.choice]+=Math.min(2,16-total);setHerd('DYN',targets.COW,targets.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')));}}\nif([0,1,2,3,6,9,12,18].includes(+obs.day))diary.push({day:+obs.day,cash:obs.farms[pid].money,own:{...own},targets:{...targets},commitmentDone,shops:[...(obs.town?.unlocked_shops??[])]});}\ntry{acts[pid]=JSON.parse(String(call(n,JSON.stringify(obs))));if(n==='DYN')acts[pid]=enforce(obs,acts[pid],targets);}catch{errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]};}}s=E.interpreter.step(s,acts,{...cfg0,seed});}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,errors,diary};}\nconst games:any[]=[];for(const v of variants)for(const seed of seeds){games.push(await play(v,seed,0));games.push(await play(v,seed,1));}const summary:any={};for(const v of variants){const rs=games.filter(g=>g.v===v);summary[v]={episodes:rs.length,wins:rs.filter(r=>r.win).length,winRate:rs.filter(r=>r.win).length/rs.length,avgBank:rs.reduce((a,r)=>a+r.ours,0)/rs.length,avgMargin:rs.reduce((a,r)=>a+r.margin,0)/rs.length,bySeed:Object.fromEntries(seeds.map(seed=>{const x=rs.filter(r=>r.seed===seed);return[seed,{wins:x.filter(r=>r.win).length,avgMargin:x.reduce((a,r)=>a+r.margin,0)/x.length,avgBank:x.reduce((a,r)=>a+r.ours,0)/x.length}]}))};}const b=games.filter(g=>g.v==='BASE');const paired=variants.slice(1).map(v=>{const r=games.filter(g=>g.v===v);return{variant:v,avgMarginDelta:r.reduce((s,g,i)=>s+(g.margin-b[i].margin),0)/r.length,better:r.filter((g,i)=>g.margin>b[i].margin).length,worse:r.filter((g,i)=>g.marging.seed===9491&&g.seat===0).map(g=>({v:g.v,diary:g.diary}))},null,2));",[],[1368],"66fcb11b-b307-4698-b494-3be82041d00d","kag_mix11_full_reference_league_v1.ts","9cca496c-aa54-11f1-a6ea-1607ee4eb77e",98,["D",1788741580951],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst variants=['BASE','MIX11'];\nconst refs=['rotation_rosa.py','homestead_hana.py','melon_mateo.py','rancher_rita.py','broker_bea.py','ledger_lena.py','slotter_silas.py','closer_cleo.py'];\nconst tier:any={'rotation_rosa.py':2,'homestead_hana.py':3,'melon_mateo.py':4,'rancher_rita.py':5,'broker_bea.py':6,'ledger_lena.py':7,'slotter_silas.py':8,'closer_cleo.py':9};\nconst seeds=[17011,17123,17239,17351];\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();\nconst zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!zr.ok)throw new Error(`dataset ${zr.status}`);const z=unzipSync(new Uint8Array(await zr.arrayBuffer()));\nconst py=await loadPyodide();py.runPython(`import json\\n_REG={}`);\nfor(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}\npy.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_n={}\\nexec(src,_n)\\n_N=_n\\n_BASE_POLICY=json.loads(json.dumps(_n['POLICY']))`);\npy.runPython(`\\ndef call_ref(name,payload):return json.dumps(_REG[name](json.loads(payload)))\\ndef call_dyn(payload):return json.dumps(_N['agent'](json.loads(payload)))\\ndef reset_dyn():\\n _N['POLICY'].clear();_N['POLICY'].update(json.loads(json.dumps(_BASE_POLICY)));_N['POLICY']['animal_target']={'COW':0,'SHEEP':0}\\ndef set_herd(c,s,order):\\n _N['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)}\\n _N['POLICY']['animals']=json.loads(order)\\n`);\nconst callRef=py.globals.get('call_ref'),callDyn=py.globals.get('call_dyn'),resetDyn=py.globals.get('reset_dyn'),setHerd=py.globals.get('set_herd');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}\nfunction enforce(o:any,act:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},y={...act,market:[] as any[]};for(const m of act?.market??[]){if(Array.isArray(m)&&m[0]==='BUY_ANIMAL'&&(m[1]==='COW'||m[1]==='SHEEP')){const a=m[1],q=Math.max(0,Math.min(+m[2]||1,t[a]-x[a]));if(q){y.market.push(['BUY_ANIMAL',a,q]);x[a]+=q}}else y.market.push(m)}return y}\nasync function play(v:string,ref:string,seed:number,seat:0|1){resetDyn();let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},commitDone=v==='BASE',errors=0,minCash=Infinity;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!commitDone){commitDone=ow.COW>=1&&ow.SHEEP>=1;if(!commitDone){tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)};setHerd(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}}if(commitDone){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);setHerd(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')))}}try{aa[pid]=enforce(o,JSON.parse(String(callDyn(JSON.stringify(o)))),tg)}catch{errors++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}minCash=Math.min(minCash,+o.farms[pid].money)}else{try{aa[pid]=JSON.parse(String(callRef(ref,JSON.stringify(o))))}catch{errors++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,tier:tier[ref],seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,tie:ours===theirs,errors,minCash}}\nconst games:any[]=[];for(const v of variants)for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}\nconst avg=(xs:number[])=>xs.reduce((a,b)=>a+b,0)/(xs.length||1);const summary:any={};for(const v of variants){summary[v]={overall:{},byTier:{}};const all=games.filter(g=>g.v===v);summary[v].overall={episodes:all.length,wins:all.filter(g=>g.win).length,ties:all.filter(g=>g.tie).length,winRate:all.filter(g=>g.win).length/all.length,avgBank:avg(all.map(g=>g.ours)),avgMargin:avg(all.map(g=>g.margin)),errors:all.reduce((a,g)=>a+g.errors,0)};for(const ref of refs){const x=all.filter(g=>g.ref===ref);summary[v].byTier[ref]={tier:tier[ref],episodes:x.length,wins:x.filter(g=>g.win).length,ties:x.filter(g=>g.tie).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgRefBank:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),errors:x.reduce((a,g)=>a+g.errors,0)}}}\nconst b=games.filter(g=>g.v==='BASE'),m=games.filter(g=>g.v==='MIX11');const paired={avgMarginDelta:avg(m.map((g,i)=>g.margin-b[i].margin)),better:m.filter((g,i)=>g.margin>b[i].margin).length,worse:m.filter((g,i)=>g.marging.margin===b[i].margin).length};\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'BLACK_BOX_PUBLIC_REFERENCE_AGENTS_ON_OFFICIAL_TS_PORT_HIGH_TIERS_NOT_REUSED_AS_CODE',seeds,summary,paired},null,2));",[],[1376],"085c812d-ffa3-4c83-959c-2c55b7c3c496","kag_mix11_fert_byproduct_probe_v1.ts","047b6fe0-aa56-11f1-b4d6-1607ee4eb77e",114,["D",1788742184402],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['MIX11','MIX11_FERT'],refs=['rancher_rita.py','broker_bea.py'],seeds=[18013,18121,18233,18347,18461,18583];\nconst E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\\ndef reset(fert):\\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['collect_fertilizer']=bool(fert)\\ndef seth(c,s,o):\\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)}\\n _D['POLICY']['animals']=json.loads(o)\\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),reset=py.globals.get('reset'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}\nfunction enforce(o:any,act:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},y={...act,market:[...(act?.market??[])] as any[]};const mm:any[]=[];for(const m of y.market){if(Array.isArray(m)&&m[0]==='BUY_ANIMAL'&&(m[1]==='COW'||m[1]==='SHEEP')){const a=m[1],q=Math.max(0,Math.min(+m[2]||1,t[a]-x[a]));if(q){mm.push(['BUY_ANIMAL',a,q]);x[a]+=q}}else mm.push(m)}y.market=mm;return y}\nfunction addFertSell(o:any,act:any){const q=+(o.private?.shed?.FERTILIZER??0),price=+(o.market?.prices?.FERTILIZER??0);if(q>0&&price>=60&&(act.market?.length??0)<10){act={...act,market:[...(act.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}}return act}\nfunction opCount(act:any,name:string){let n=0;for(const u of [act?.farmer,...(act?.hands??[])])if(Array.isArray(u)&&u[0]===name)n++;return n}\nasync function play(v:string,ref:string,seed:number,seat:0|1){reset(v==='MIX11_FERT');let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},done=false,errors=0,collect=0,fertSold=0;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!done){done=ow.COW>=1&&ow.SHEEP>=1;if(!done){tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)};seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}}if(done){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')))}}let a=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg);if(v==='MIX11_FERT')a=addFertSell(o,a);collect+=opCount(a,'COLLECT_FERTILIZER');for(const m of a.market??[])if(m?.[0]==='SELL'&&m?.[1]==='FERTILIZER')fertSold+=Number(m?.[2]??0);aa[pid]=a}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{errors++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,errors,collect,fertSold}}\nconst games:any[]=[];for(const v of variants)for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const summary:any={};for(const v of variants){const rs=games.filter(g=>g.v===v);summary[v]={episodes:rs.length,wins:rs.filter(g=>g.win).length,winRate:rs.filter(g=>g.win).length/rs.length,avgBank:avg(rs.map(g=>g.ours)),avgMargin:avg(rs.map(g=>g.margin)),avgCollect:avg(rs.map(g=>g.collect)),avgFertSold:avg(rs.map(g=>g.fertSold)),byOpponent:{}};for(const ref of refs){const x=rs.filter(g=>g.ref===ref);summary[v].byOpponent[ref]={wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgRefBank:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),avgCollect:avg(x.map(g=>g.collect)),avgFertSold:avg(x.map(g=>g.fertSold))}}}const b=games.filter(g=>g.v==='MIX11'),f=games.filter(g=>g.v==='MIX11_FERT');console.log(JSON.stringify({status:'PASS',hypothesis:'MONETIZE_UNUSED_LIVESTOCK_BYPRODUCT_WITHOUT_DISPLACING_EXISTING_MARKET_ORDERS',guard:'collect enabled; append SELL FERTILIZER only if current market queue has spare slot and price>=60',summary,paired:{avgMarginDelta:avg(f.map((g,i)=>g.margin-b[i].margin)),better:f.filter((g,i)=>g.margin>b[i].margin).length,worse:f.filter((g,i)=>g.marging.margin===b[i].margin).length}},null,2));",[],[1384],"a903bea9-1c3e-4a7e-a545-9e62825dfacd","kag_mix11_crossopponent_screen_v1.ts","011a456a-aa51-11f1-adc5-1607ee4eb77e",96,["D",1788740031212],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['BASE','MIX11'],opponents=['C4S12','C6S10','RITA_BASE'],seeds=[15013,15121,15233,15349,15461,15569,15679,15791,15901,16007];const E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer())),src=strFromU8(z['rancher_rita.py']),py=await loadPyodide();py.runPython(`import json,copy\\n_REG={}\\n_BASE={}`);py.globals.set('src',src);const patches:any={DYN:{animal_target:{COW:0,SHEEP:0}},C4S12:{animal_target:{COW:4,SHEEP:12}},C6S10:{animal_target:{COW:6,SHEEP:10}},RITA_BASE:{}};for(const [n,p] of Object.entries(patches)){py.globals.set('nm',n);py.globals.set('patch',JSON.stringify(p));py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items():_n['POLICY'][_k]=_v\\n_REG[nm]=_n\\n_BASE[nm]=copy.deepcopy(_n['POLICY'])`)}py.runPython(`\\ndef call(n,p):return json.dumps(_REG[n]['agent'](json.loads(p)))\\ndef seth(n,c,s,o):\\n _REG[n]['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)}\\n _REG[n]['POLICY']['animals']=json.loads(o)\\ndef reset(n):_REG[n]['POLICY'].clear();_REG[n]['POLICY'].update(copy.deepcopy(_BASE[n]))\\n`);const call=py.globals.get('call'),seth=py.globals.get('seth'),reset=py.globals.get('reset');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,act:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},y={...act,market:[] as any[]};for(const m of act?.market??[]){if(Array.isArray(m)&&m[0]==='BUY_ANIMAL'&&(m[1]==='COW'||m[1]==='SHEEP')){const a=m[1],q=Math.max(0,Math.min(+m[2]||1,t[a]-x[a]));if(q){y.market.push(['BUY_ANIMAL',a,q]);x[a]+=q}}else y.market.push(m)}return y}\nasync function play(v:string,opp:string,seed:number,seat:0|1){reset('DYN');reset(opp);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},done=v==='BASE';while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const n=pid===seat?'DYN':opp,o=observationFor(s,pid);if(n==='DYN'&&+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!done){done=ow.COW>=1&&ow.SHEEP>=1;if(!done){tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)};seth('DYN',tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}}if(done){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth('DYN',tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')))}}aa[pid]=JSON.parse(String(call(n,JSON.stringify(o))));if(n==='DYN')aa[pid]=enforce(o,aa[pid],tg)}s=E.interpreter.step(s,aa,{...cfg0,seed})}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,opp,seed,ours,theirs,margin:ours-theirs,win:ours>theirs}}\nconst games:any[]=[];for(const v of variants)for(const o of opponents)for(const seed of seeds){games.push(await play(v,o,seed,0));games.push(await play(v,o,seed,1))}const summary:any={};for(const v of variants){const rs=games.filter(g=>g.v===v);summary[v]={wins:rs.filter(r=>r.win).length,episodes:rs.length,winRate:rs.filter(r=>r.win).length/rs.length,avgBank:rs.reduce((a,r)=>a+r.ours,0)/rs.length,avgMargin:rs.reduce((a,r)=>a+r.margin,0)/rs.length,byOpponent:{}};for(const o of opponents){const x=rs.filter(r=>r.opp===o);summary[v].byOpponent[o]={wins:x.filter(r=>r.win).length,winRate:x.filter(r=>r.win).length/x.length,avgBank:x.reduce((a,r)=>a+r.ours,0)/x.length,avgOppBank:x.reduce((a,r)=>a+r.theirs,0)/x.length,avgMargin:x.reduce((a,r)=>a+r.margin,0)/x.length}}}const b=games.filter(g=>g.v==='BASE'),m=games.filter(g=>g.v==='MIX11');console.log(JSON.stringify({status:'PASS',seeds,summary,paired:{avgMarginDelta:m.reduce((s,g,i)=>s+g.margin-b[i].margin,0)/m.length,better:m.filter((g,i)=>g.margin>b[i].margin).length,worse:m.filter((g,i)=>g.margin=1&&ow.SHEEP>=1;if(!done){tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)};seth('DYN',tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}}if(done){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth('DYN',tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')))}}aa[pid]=JSON.parse(String(call(n,JSON.stringify(o))));if(n==='DYN')aa[pid]=enforce(o,aa[pid],tg);add(tel[pid],aa[pid])}s=E.interpreter.step(s,aa,{...cfg0,seed})}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,ourTel:tel[seat],oppTel:tel[1-seat],ourAnimals:countTiles(s.farms[seat]),oppAnimals:countTiles(s.farms[1-seat]),finalMarket:{MILK:{p:s.market.prices.MILK,i:s.market.inventory.MILK},WOOL:{p:s.market.prices.WOOL,i:s.market.inventory.WOOL},WHEAT:{p:s.market.prices.WHEAT,i:s.market.inventory.WHEAT}}}}\nconst games:any[]=[];for(const v of variants)for(const seed of seeds){games.push(await play(v,seed,0));games.push(await play(v,seed,1))}const pairs=seeds.map(seed=>{const row:any={seed};for(const v of variants){const x=games.filter(g=>g.v===v&&g.seed===seed);row[v]={margin:x.reduce((a,g)=>a+g.margin,0)/2,ours:x.reduce((a,g)=>a+g.ours,0)/2,theirs:x.reduce((a,g)=>a+g.theirs,0)/2,ourSell:Object.fromEntries(['MILK','WOOL','WHEAT'].map(k=>[k,x.reduce((a,g)=>a+(g.ourTel.sell[k]??0),0)/2])),oppSell:Object.fromEntries(['MILK','WOOL','WHEAT'].map(k=>[k,x.reduce((a,g)=>a+(g.oppTel.sell[k]??0),0)/2])),ourAnimal:x[0].ourAnimals,oppAnimal:x[0].oppAnimals};}row.delta=row.MIX11.margin-row.BASE.margin;return row}).sort((a,b)=>b.delta-a.delta);console.log(JSON.stringify({status:'PASS',question:'WHY_DOES_MIX11_REDUCE_C6_CONTINUATION_VALUE_WITHOUT_REDUCING_OURS',pairs,topImprovement:pairs.slice(0,3),worst:pairs.slice(-3)},null,2));",[],[1400],"a35b2689-c0a9-47f5-bd87-9349955bf2af","kag_melon_synergy_ablation_v1.ts","3550ddaa-aa5e-11f1-96c9-1607ee4eb77e",145,["D",1788745702313],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['BASE','M40_PLAIN','M40_COLLECT_ONLY','M40_CLOSED','M40_BUYFERT'];const refs=['rancher_rita.py','broker_bea.py'],seeds=[30011,30119,30223];\nconst E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\\ndef prep(v):\\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0}\\n if v!='BASE':\\n _D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.6,'MELON':0.4};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':120};_D['POLICY']['sell_chunk']=20\\n if v=='M40_COLLECT_ONLY': _D['POLICY']['collect_fertilizer']=True\\n if v=='M40_CLOSED': _D['POLICY']['collect_fertilizer']=True;_D['POLICY']['fertilize_crops']=['MELON'];_D['POLICY']['fert_stock']=0\\n if v=='M40_BUYFERT': _D['POLICY']['fertilize_crops']=['MELON'];_D['POLICY']['fert_stock']=6\\ndef seth(c,s,o):\\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}\nasync function play(v:string,ref:string,seed:number,seat:0|1){prep(v);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0,ops:any={},buyFert=0;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}const a=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg);for(const u of [a?.farmer,...(a?.hands??[])]){const k=u?.[0]??'NONE';ops[k]=(ops[k]??0)+1}for(const q of a.market??[])if(q?.[0]==='BUY_PRODUCT'&&q?.[1]==='FERTILIZER')buyFert+=Number(q[2]??0);aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,animalLoss:maxFlock-finalFlock,ops,buyFert}}\nconst games:any[]=[];for(const v of variants)for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const sum:any={};for(const v of variants){const x=games.filter(g=>g.v===v);sum[v]={episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),avgCollect:avg(x.map(g=>g.ops.COLLECT_FERTILIZER??0)),avgFertilize:avg(x.map(g=>g.ops.FERTILIZE??0)),avgPass:avg(x.map(g=>g.ops.PASS??0)),avgBuyFert:avg(x.map(g=>g.buyFert)),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);sum[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin))}}}const plain=games.filter(g=>g.v==='M40_PLAIN');const paired=variants.filter(v=>v.startsWith('M40_')&&v!=='M40_PLAIN').map(v=>{const x=games.filter(g=>g.v===v);return{v,bankDeltaVsPlain:avg(x.map((g,i)=>g.ours-plain[i].ours)),oppDeltaVsPlain:avg(x.map((g,i)=>g.theirs-plain[i].theirs)),marginDeltaVsPlain:avg(x.map((g,i)=>g.margin-plain[i].margin)),better:x.filter((g,i)=>g.margin>plain[i].margin).length,worse:x.filter((g,i)=>g.margin0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(v:string,ref:string,seed:number,seat:0|1){prep(variants[v]);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[pid]=addFert(o,enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg));minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,tie:ours===theirs,err,minCash,animalLoss:maxFlock-finalFlock}}\nconst games:any[]=[];for(const v of Object.keys(variants))for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const summary:any={};for(const v of Object.keys(variants)){const x=games.filter(g=>g.v===v),w=x.filter(g=>g.win).length,l=x.filter(g=>!g.win&&!g.tie).length;summary[v]={share:variants[v],episodes:x.length,wins:w,losses:l,ties:x.length-w-l,winRate:w/x.length,wilson:wilson(w,l),avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),errors:x.reduce((a,g)=>a+g.err,0),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);summary[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,losses:y.filter(g=>!g.win&&!g.tie).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin))}}}console.log(JSON.stringify({status:'PASS',evidence:'BLACK_BOX_TOP_REFERENCE_NO_TRACE_REUSE',seeds,summary},null,2));",[],[1416],"0c45a6a3-8ed2-4378-81d5-2984f135dc10","kag_melon_share_frontier_v1.ts","68a68c40-aa5e-11f1-a70f-1607ee4eb77e",146,["D",1788745788446],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={M30:.30,M40:.40,M50:.50,M60:.60,M70:.70};const refs=['rancher_rita.py','broker_bea.py'],seeds=[31013,31121,31231];const E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\\ndef prep(m):\\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':1.0-m,'MELON':m};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':120};_D['POLICY']['sell_chunk']=20\\ndef seth(c,s,o):\\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}\nasync function play(v:string,ref:string,seed:number,seat:0|1){prep(variants[v]);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0,wheatBuy=0,melonSell=0,melonSeed=0;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}const a=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg);for(const q of a.market??[]){if(q?.[0]==='BUY_PRODUCT'&&q?.[1]==='WHEAT')wheatBuy+=Number(q[2]??0);if(q?.[0]==='BUY_SEED'&&q?.[1]==='MELON')melonSeed+=Number(q[2]??0);if(q?.[0]==='SELL'&&q?.[1]==='MELON')melonSell+=Number(q[2]??0)}aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,animalLoss:maxFlock-finalFlock,wheatBuy,melonSell,melonSeed}}\nconst games:any[]=[];for(const v of Object.keys(variants))for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const sum:any={};for(const v of Object.keys(variants)){const x=games.filter(g=>g.v===v);sum[v]={share:variants[v],episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),avgWheatBuy:avg(x.map(g=>g.wheatBuy)),avgMelonSeed:avg(x.map(g=>g.melonSeed)),avgMelonSellAttempt:avg(x.map(g=>g.melonSell)),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);sum[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin))}}}console.log(JSON.stringify({status:'PASS',question:'WHERE_IS_THE_MELON_SHARE_FRONTIER_BEFORE_FEED_SUBSTITUTION_AND_GLUT_DESTROY_THE_GAIN',summary:sum},null,2));",[],[1424],"08175b14-9ee5-4afc-bc3a-544189cd6512","kag_melon_fertsell_probe_v1.ts","a1862084-aa5e-11f1-b80f-1607ee4eb77e",147,["D",1788745883854],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={M40:{m:.4,f:false},M40_FERTSELL:{m:.4,f:true},M60:{m:.6,f:false},M60_FERTSELL:{m:.6,f:true}};const refs=['rancher_rita.py','broker_bea.py'],seeds=[32003,32117,32233,32341];const E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\\ndef prep(m,f):\\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':1.0-m,'MELON':m};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':120};_D['POLICY']['sell_chunk']=20;_D['POLICY']['collect_fertilizer']=bool(f)\\ndef seth(c,s,o):\\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(v:string,ref:string,seed:number,seat:0|1){const vc=variants[v];prep(vc.m,vc.f);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0,collect=0,pass=0,fertSell=0;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}let a=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg);if(vc.f)a=addFert(o,a);for(const u of [a?.farmer,...(a?.hands??[])]){if(u?.[0]==='COLLECT_FERTILIZER')collect++;if(u?.[0]==='PASS')pass++}for(const q of a.market??[])if(q?.[0]==='SELL'&&q?.[1]==='FERTILIZER')fertSell+=Number(q[2]??0);aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,animalLoss:maxFlock-finalFlock,collect,pass,fertSell}}\nconst games:any[]=[];for(const v of Object.keys(variants))for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const sum:any={};for(const v of Object.keys(variants)){const x=games.filter(g=>g.v===v);sum[v]={episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),avgCollect:avg(x.map(g=>g.collect)),avgPass:avg(x.map(g=>g.pass)),avgFertSellAttempt:avg(x.map(g=>g.fertSell)),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);sum[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin))}}}const paired=['M40_FERTSELL','M60_FERTSELL'].map(v=>{const base=games.filter(g=>g.v===v.replace('_FERTSELL','')),x=games.filter(g=>g.v===v);return{v,bankDelta:avg(x.map((g,i)=>g.ours-base[i].ours)),oppDelta:avg(x.map((g,i)=>g.theirs-base[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-base[i].margin)),better:x.filter((g,i)=>g.margin>base[i].margin).length,worse:x.filter((g,i)=>g.margin/MELON|product|item/i.test(x)).slice(0,10).join('\\n'));",[],[1440],"5d1ea951-3ce2-4efc-ba05-bd9579524deb","kag_market_slot_ablation_v1.ts","18648e7c-aa67-11f1-8590-1607ee4eb77e",188,["D",1788749519268],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst variants=['UNCHANGED','INPLACE_SELL_SORT','HOIST_SELLS'];\nconst refs=['rancher_rita.py','broker_bea.py','closer_cleo.py'];\nconst seeds=[32003,32117,32233,32341,32467,32573,32687,32797];\nconst E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();\nconst zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!zr.ok)throw new Error(`dataset ${zr.status}`);const z=unzipSync(new Uint8Array(await zr.arrayBuffer()));\nconst py=await loadPyodide();py.runPython(`import json,copy\\n_SRC={}\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_SRC[nm]=src`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_D={}\\nexec(src,_D)\\n_BASE=copy.deepcopy(_D['POLICY'])`);py.runPython(`\ndef reset_all(ref):\n _D['POLICY'].clear();_D['POLICY'].update(copy.deepcopy(_BASE));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0}\n ns={};exec(_SRC[ref],ns);_REG.clear();_REG[ref]=ns['agent']\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)}\n _D['POLICY']['animals']=json.loads(o)\n`);const resetAll=py.globals.get('reset_all'),cd=py.globals.get('cd'),cr=py.globals.get('cr'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}\nfunction sellScore(o:any,q:any){const item=String(q?.[1]??''),qty=Number(q?.[2]??0),p=Number(o.market?.prices?.[item]??0);return p*qty}\nfunction mutateMarket(v:string,o:any,a:any){const m=[...(a.market??[])];if(v==='UNCHANGED')return a;const sells=m.filter(q=>q?.[0]==='SELL').sort((x,y)=>sellScore(o,y)-sellScore(o,x));if(v==='INPLACE_SELL_SORT'){let k=0;const out=m.map(q=>q?.[0]==='SELL'?sells[k++]:q);return{...a,market:out}}const nons=m.filter(q=>q?.[0]!=='SELL');return{...a,market:[...sells,...nons].slice(0,10)}}\nfunction queueStats(acc:any,before:any[],after:any[]){acc.steps++;const sig=(x:any[])=>x.map(q=>q?.[0]).join('>');if(sig(before)!==sig(after))acc.changed++;for(let i=0;iq?.[0]==='BUY_PRODUCT'&&after.slice(0,i).some(z=>z?.[0]==='SELL'))?1:0)}\nasync function play(v:string,ref:string,seed:number,seat:0|1){resetAll(ref);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},commit=false,errors=0,minCash=Infinity;const qs={steps:0,changed:0,slotTypeChanges:0,buyProductAfterSell:0};while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!commit){commit=ow.COW>=1&&ow.SHEEP>=1;if(!commit){tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)};seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}}if(commit){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')))}}const raw=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg),mut=mutateMarket(v,o,raw);queueStats(qs,raw.market??[],mut.market??[]);aa[pid]=mut;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{errors++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,errors,minCash,qs}}\nconst games:any[]=[];for(const v of variants)for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),summary:any={};for(const v of variants){const x=games.filter(g=>g.v===v);summary[v]={episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),errors:x.reduce((a,g)=>a+g.errors,0),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);summary[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin)),changedQueues:avg(y.map(g=>g.qs.changed)),slotTypeChanges:avg(y.map(g=>g.qs.slotTypeChanges)),buyProductAfterSellSteps:avg(y.map(g=>g.qs.buyProductAfterSell))}}}\nconst b=games.filter(g=>g.v==='UNCHANGED'),paired:any={};for(const v of variants.slice(1)){const x=games.filter(g=>g.v===v);paired[v]={bankDelta:avg(x.map((g,i)=>g.ours-b[i].ours)),oppDelta:avg(x.map((g,i)=>g.theirs-b[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-b[i].margin)),better:x.filter((g,i)=>g.margin>b[i].margin).length,worse:x.filter((g,i)=>g.marging.margin===b[i].margin).length}}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'OWN_FIELD_PLAN_MARKET_SLOT_ABLATION',hypothesis:'Reordering SELLs only inside pre-existing SELL slots may preserve same-turn purchase funding better than globally hoisting SELLs.',design:{variants,refs,seeds,score:'current unit price * quantity'},summary,paired},null,2));",[],[1448],"9cc7adb8-4b36-4496-9548-f452cb1f2cbc","kag_m60_opponent_melon_market_screen_v1.ts","2edc80f8-aa60-11f1-8dba-1607ee4eb77e",158,["D",1788746550447],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['BASE80','ANY1','ANY25','MATURE1','DOM1','BEHIND1','AHEAD1'];\nconst refs=['melon_mateo.py','rancher_rita.py','broker_bea.py'],seeds=[36007,36113];\nconst E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));\nconst py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}\npy.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(n,p): return json.dumps(_REG[n](json.loads(p)))\ndef cd(p): return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setfloor(x): _D['POLICY']['price_floor']={'MELON':int(x)}\n`);\nconst cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setfloor=py.globals.get('setfloor');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}\nfunction cropStats(f:any,crop:string,day:number){let tiles=0,matureSoon=0;for(const row of f?.tiles??[])for(const t of row??[]){if(t&&typeof t==='object'&&t.kind==='PLANT'&&t.crop===crop){tiles++;const age=day-Number(t.planted_day??day);if(age>=8)matureSoon++}}return{tiles,matureSoon}}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}\nfunction addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction floorFor(v:string,o:any){if(v==='BASE80')return 80;const me=+o.player,opp=o.farms[1-me],mine=o.farms[me],d=+o.day,os=cropStats(opp,'MELON',d),ms=cropStats(mine,'MELON',d),behind=+mine.money<+opp.money;let trig=false,floor=80;if(v==='ANY1'){trig=os.tiles>0;floor=1}else if(v==='ANY25'){trig=os.tiles>0;floor=25}else if(v==='MATURE1'){trig=os.matureSoon>0;floor=1}else if(v==='DOM1'){trig=os.tiles>ms.tiles;floor=1}else if(v==='BEHIND1'){trig=os.tiles>0&&behind;floor=1}else if(v==='AHEAD1'){trig=os.tiles>0&&!behind;floor=1}return trig?floor:80}\nasync function play(v:string,ref:string,seed:number,seat:0|1){prep();let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0,triggerTurns=0,totalTurns=0,lowFloorSell=0,melonSell=0;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}const f=floorFor(v,o);setfloor(f);totalTurns++;if(f<80)triggerTurns++;let a=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg);a=addFert(o,a);for(const q of a.market??[])if(q?.[0]==='SELL'&&q?.[1]==='MELON'){melonSell+=Number(q[2]??0);if(f<80)lowFloorSell+=Number(q[2]??0)}aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,animalLoss:maxFlock-finalFlock,triggerShare:triggerTurns/Math.max(1,totalTurns),lowFloorSell,melonSell}}\nconst games:any[]=[];for(const v of variants)for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const sum:any={};for(const v of variants){const x=games.filter(g=>g.v===v);sum[v]={episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),avgTriggerShare:avg(x.map(g=>g.triggerShare)),avgLowFloorMelonSell:avg(x.map(g=>g.lowFloorSell)),avgTotalMelonSell:avg(x.map(g=>g.melonSell)),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);sum[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin)),triggerShare:avg(y.map(g=>g.triggerShare))}}}\nconst b=games.filter(g=>g.v==='BASE80');const paired=variants.slice(1).map(v=>{const x=games.filter(g=>g.v===v);return{v,bankDelta:avg(x.map((g,i)=>g.ours-b[i].ours)),oppDelta:avg(x.map((g,i)=>g.theirs-b[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-b[i].margin)),better:x.filter((g,i)=>g.margin>b[i].margin).length,worse:x.filter((g,i)=>g.marging.win&&!b[i].win).length,lossFlips:x.filter((g,i)=>!g.win&&b[i].win).length}});\nconsole.log(JSON.stringify({status:'PASS',question:'CAN_PUBLIC_OPPONENT_MELON_EXPOSURE_GUIDE_INTENTIONAL_GLUT_WITHOUT_SACRIFICING_OUR_CONTINUATION_VALUE',officialCurve:{base:250,priceAfter50ExcessInventory:225,after100:150,after150:25,floorAt158:1},sum,paired},null,2));",[],[1456],"b94d968a-7256-452a-a718-be4229c908ee","kag_m60_market_floor_screen_v1.ts","01a07993-c0f9-70dd-9066-3cf4482f82b4",153,["D",1788746155185],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={F80:80,F100:100,F120:120,F140:140,F160:160};const refs=['rancher_rita.py','broker_bea.py'],seeds=[34007,34123,34231];const E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep(floor):\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':int(floor)};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(v:string,ref:string,seed:number,seat:0|1){prep(variants[v]);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0,melonAttempt=0,melonOrderCount=0,fertAttempt=0;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}let a=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg);a=addFert(o,a);for(const q of a.market??[]){if(q?.[0]==='SELL'&&q?.[1]==='MELON'){melonAttempt+=Number(q[2]??0);melonOrderCount++}if(q?.[0]==='SELL'&&q?.[1]==='FERTILIZER')fertAttempt+=Number(q[2]??0)}aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,animalLoss:maxFlock-finalFlock,melonAttempt,melonOrderCount,fertAttempt}}\nconst games:any[]=[];for(const v of Object.keys(variants))for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const sum:any={};for(const v of Object.keys(variants)){const x=games.filter(g=>g.v===v);sum[v]={floor:variants[v],episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),avgMelonAttempt:avg(x.map(g=>g.melonAttempt)),avgMelonOrders:avg(x.map(g=>g.melonOrderCount)),avgFertAttempt:avg(x.map(g=>g.fertAttempt)),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);sum[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin))}}}console.log(JSON.stringify({status:'PASS',question:'WHAT_MELON_PRICE_FLOOR_BEST_BALANCES_CASHFLOW_SCARCITY_RENT_AND_SHARED_MARKET_PRESSURE_FOR_M60_FERTSELL',summary:sum},null,2));",[],[1464],"e95829ac-2fef-4530-bef9-e76c09051067","kag_m60_market_chunk_screen_v1.ts","0cd86ae0-aa5f-11f1-aaad-1607ee4eb77e",148,["D",1788746063912],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={C8:8,C12:12,C16:16,C20:20};const refs=['rancher_rita.py','broker_bea.py'],seeds=[33013,33119,33223];const E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep(chunk):\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':120};_D['POLICY']['sell_chunk']=int(chunk);_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(v:string,ref:string,seed:number,seat:0|1){prep(variants[v]);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0,melonAttempt=0,melonOrderCount=0,fertAttempt=0;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}let a=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg);a=addFert(o,a);for(const q of a.market??[]){if(q?.[0]==='SELL'&&q?.[1]==='MELON'){melonAttempt+=Number(q[2]??0);melonOrderCount++}if(q?.[0]==='SELL'&&q?.[1]==='FERTILIZER')fertAttempt+=Number(q[2]??0)}aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,animalLoss:maxFlock-finalFlock,melonAttempt,melonOrderCount,fertAttempt}}\nconst games:any[]=[];for(const v of Object.keys(variants))for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const sum:any={};for(const v of Object.keys(variants)){const x=games.filter(g=>g.v===v);sum[v]={chunk:variants[v],episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),avgMelonAttempt:avg(x.map(g=>g.melonAttempt)),avgMelonOrders:avg(x.map(g=>g.melonOrderCount)),avgFertAttempt:avg(x.map(g=>g.fertAttempt)),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);sum[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin))}}}console.log(JSON.stringify({status:'PASS',question:'WHAT_MELON_SELL_LOT_SIZE_BEST_BALANCES_PRICE_IMPACT_CASHFLOW_AND_SHARED_MARKET_PRESSURE_FOR_M60_FERTSELL',summary:sum},null,2));",[],[1472],"5041131f-f476-4a9e-89ba-8f4d08cf8c2c","kag_m60_m70_crossmarket_dossier_v1.ts","7ad84bc2-aa60-11f1-a84f-1607ee4eb77e",160,["D",1788746694480],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={M60:.60,M70:.70};const refs=['broker_bea.py'];const seeds=[37003,37117,37223,37337];const E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);py.globals.set('src',strFromU8(z['broker_bea.py']));py.runPython(`_n={}\\nexec(src,_n)\\n_REG['broker_bea.py']=_n['agent']`);py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(p):return json.dumps(_REG['broker_bea.py'](json.loads(p)))\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep(m):\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':1.0-m,'MELON':m};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(v:string,seed:number,seat:0|1){prep(variants[v]);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,maxFlock=0,finalFlock=0;const oursOrd:any={buy:{},sell:{},seed:{}},oppOrd:any={buy:{},sell:{},seed:{}},timeline:any[]=[];while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){if(+o.hour===0){timeline.push({day:+o.day,ourCash:+o.farms[pid].money,oppCash:+o.farms[1-pid].money,wheatP:+o.market.prices.WHEAT,melonP:+o.market.prices.MELON,wheatI:+o.market.inventory.WHEAT,melonI:+o.market.inventory.MELON,ourFlock:own(o,'COW')+own(o,'SHEEP')});const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}let a=addFert(o,enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg));for(const q of a.market??[]){const [k,item,n]=q;if(k==='BUY_PRODUCT')oursOrd.buy[item]=(oursOrd.buy[item]??0)+Number(n??0);if(k==='SELL')oursOrd.sell[item]=(oursOrd.sell[item]??0)+Number(n??0);if(k==='BUY_SEED')oursOrd.seed[item]=(oursOrd.seed[item]??0)+Number(n??0)}aa[pid]=a;maxFlock=Math.max(maxFlock,own(o,'COW')+own(o,'SHEEP'));}else{const a=JSON.parse(String(cr(JSON.stringify(o))));for(const q of a.market??[]){const [k,item,n]=q;if(k==='BUY_PRODUCT')oppOrd.buy[item]=(oppOrd.buy[item]??0)+Number(n??0);if(k==='SELL')oppOrd.sell[item]=(oppOrd.sell[item]??0)+Number(n??0);if(k==='BUY_SEED')oppOrd.seed[item]=(oppOrd.seed[item]??0)+Number(n??0)}aa[pid]=a}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{v,seed,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,win:s.farms[seat].money>s.farms[1-seat].money,animalLoss:maxFlock-finalFlock,oursOrd,oppOrd,timeline}}\nconst games:any[]=[];for(const v of Object.keys(variants))for(const seed of seeds){games.push(await play(v,seed,0));games.push(await play(v,seed,1))}const avg=(xs:number[])=>xs.reduce((a,b)=>a+b,0)/(xs.length||1);function agg(v:string){const x=games.filter(g=>g.v===v);const days=[3,6,9,12,15,18,21,24,27,29].map(day=>{const rows=x.map(g=>g.timeline.find((t:any)=>t.day===day)).filter(Boolean);return{day,ourCash:avg(rows.map((r:any)=>r.ourCash)),oppCash:avg(rows.map((r:any)=>r.oppCash)),wheatP:avg(rows.map((r:any)=>r.wheatP)),melonP:avg(rows.map((r:any)=>r.melonP)),wheatI:avg(rows.map((r:any)=>r.wheatI)),melonI:avg(rows.map((r:any)=>r.melonI)),ourFlock:avg(rows.map((r:any)=>r.ourFlock))}});const get=(path:'buy'|'sell'|'seed',item:string,who:'oursOrd'|'oppOrd')=>avg(x.map(g=>g[who][path][item]??0));return{episodes:x.length,wins:x.filter(g=>g.win).length,ourBank:avg(x.map(g=>g.ours)),oppBank:avg(x.map(g=>g.theirs)),margin:avg(x.map(g=>g.margin)),animalLoss:avg(x.map(g=>g.animalLoss)),ourOrders:{buyWheat:get('buy','WHEAT','oursOrd'),sellWheat:get('sell','WHEAT','oursOrd'),seedWheat:get('seed','WHEAT','oursOrd'),seedMelon:get('seed','MELON','oursOrd'),sellMelon:get('sell','MELON','oursOrd')},oppOrders:{buyWheat:get('buy','WHEAT','oppOrd'),sellWheat:get('sell','WHEAT','oppOrd'),seedMelon:get('seed','MELON','oppOrd'),sellMelon:get('sell','MELON','oppOrd')},days}}\nconst a=agg('M60'),b=agg('M70');console.log(JSON.stringify({status:'PASS',question:'WHY_DOES_M70_HURT_BROKER_MORE_THAN_US_WHICH_SHARED_MARKET_CHANNEL_MOVES',M60:a,M70:b,delta:{ourBank:b.ourBank-a.ourBank,oppBank:b.oppBank-a.oppBank,margin:b.margin-a.margin,ourBuyWheat:b.ourOrders.buyWheat-a.ourOrders.buyWheat,ourSellWheat:b.ourOrders.sellWheat-a.ourOrders.sellWheat,ourSellMelon:b.ourOrders.sellMelon-a.ourOrders.sellMelon,oppBuyWheat:b.oppOrders.buyWheat-a.oppOrders.buyWheat,oppSellMelon:b.oppOrders.sellMelon-a.oppOrders.sellMelon}},null,2));",[],[1480],"38476107-f12f-48de-9f75-8bbf345cf6ca","kag_liquidity_labor_coupling_probe_v1.ts","b376b200-aa5d-11f1-95b8-1607ee4eb77e",143,["D",1788745484455],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={BASE16_H8:{d:16,labor:'H8'},D12_H8:{d:12,labor:'H8'},D10_H8:{d:10,labor:'H8'},D10_RAMP:{d:10,labor:'RAMP'},D8_RAMP:{d:8,labor:'RAMP'},D8_H12:{d:8,labor:'H12'}};const refs=['rancher_rita.py','broker_bea.py'],seeds=[28001,28111,28219];\nconst E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\\ndef prep(d):\\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['feed_float_days']=int(d)\\ndef setstate(c,s,o,h):\\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o);_D['POLICY']['hands']=int(h)\\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),setstate=py.globals.get('setstate');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function hands(mode:string,flock:number){if(mode==='H12')return 12;if(mode==='RAMP')return flock<6?8:flock<10?10:12;return 8}\nasync function play(v:string,ref:string,seed:number,seat:0|1){const vc=variants[v];prep(vc.d);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0,th:any={},care=0,feed=0,pass=0,hires=0;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){for(const k of [2,4,8,12,14,16])if(th[k]===undefined&&flock>=k)th[k]=+o.day;const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0,liquidityFeedFloatDays:vc.d});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);setstate(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')),hands(vc.labor,flock))}else setstate(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']),hands(vc.labor,flock))}const a=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg);for(const u of [a?.farmer,...(a?.hands??[])]){if(u?.[0]==='CARE')care++;else if(u?.[0]==='FEED')feed++;else if(u?.[0]==='PASS')pass++}for(const q of a.market??[])if(q?.[0]==='HIRE')hires++;aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,maxFlock,finalFlock,animalLoss:maxFlock-finalFlock,th,care,feed,pass,hires}}\nconst games:any[]=[];for(const v of Object.keys(variants))for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const sum:any={};for(const v of Object.keys(variants)){const x=games.filter(g=>g.v===v);sum[v]={episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),avgCare:avg(x.map(g=>g.care)),avgFeed:avg(x.map(g=>g.feed)),avgPass:avg(x.map(g=>g.pass)),avgHires:avg(x.map(g=>g.hires)),reach8Day:avg(x.filter(g=>g.th[8]!=null).map(g=>g.th[8])),reach12Share:x.filter(g=>g.th[12]!=null).length/x.length,reach12Day:avg(x.filter(g=>g.th[12]!=null).map(g=>g.th[12])),reach14Share:x.filter(g=>g.th[14]!=null).length/x.length,reach14Day:avg(x.filter(g=>g.th[14]!=null).map(g=>g.th[14])),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);sum[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin)),avgAnimalLoss:avg(y.map(g=>g.animalLoss))}}}const b=games.filter(g=>g.v==='BASE16_H8');const paired=Object.keys(variants).slice(1).map(v=>{const x=games.filter(g=>g.v===v);return{v,bankDelta:avg(x.map((g,i)=>g.ours-b[i].ours)),oppDelta:avg(x.map((g,i)=>g.theirs-b[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-b[i].margin)),better:x.filter((g,i)=>g.margin>b[i].margin).length,worse:x.filter((g,i)=>g.marging.win&&!b[i].win).length,lossFlips:x.filter((g,i)=>!g.win&&b[i].win).length}});console.log(JSON.stringify({status:'PASS',hypothesis:'FASTER_CAPITAL_DEPLOYMENT_ONLY_WORKS_IF_LABOR_CAPACITY_EXPANDS_WITH_THE_LIVESTOCK_SERVICE_LOAD',independentRamp:'8 hands below 6 animals; 10 at 6-9; 12 at 10+',summary:sum,paired},null,2));",[],[1488],"c9c6fd17-f711-435e-8d5b-32cdf8397e71","kag_lexicographic_adversarial_tiebreak_v1.ts","92a5bd36-aa4f-11f1-bf63-1607ee4eb77e",88,["D",1788739416447],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { rankAnimalInvestments } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['PURE','LEX5','LEX10'];const eps:any={PURE:0,LEX5:.05,LEX10:.10};const seeds=[9403,9419,9431,9461,9491],opponents=['C4S12','C6S10','RITA_BASE'];\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();const zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await zr.arrayBuffer()));const src=strFromU8(z['rancher_rita.py']);const py=await loadPyodide();py.runPython(`import json,copy\\n_REG={}\\n_BASE={}`);py.globals.set('src',src);const patches:any={DYN:{animal_target:{COW:0,SHEEP:0}},C4S12:{animal_target:{COW:4,SHEEP:12}},C6S10:{animal_target:{COW:6,SHEEP:10}},RITA_BASE:{}};for(const [n,p] of Object.entries(patches)){py.globals.set('nm',n);py.globals.set('patch',JSON.stringify(p));py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n\\n_BASE[nm]=copy.deepcopy(_n['POLICY'])`);}py.runPython(`\\ndef call(name,payload): return json.dumps(_REG[name]['agent'](json.loads(payload)))\\ndef set_herd(name,cows,sheep,order):\\n _REG[name]['POLICY']['animal_target']={'COW':int(cows),'SHEEP':int(sheep)}\\n _REG[name]['POLICY']['animals']=json.loads(order)\\ndef reset(name):\\n _REG[name]['POLICY'].clear();_REG[name]['POLICY'].update(copy.deepcopy(_BASE[name]))\\n`);const call=py.globals.get('call'),setHerd=py.globals.get('set_herd'),reset=py.globals.get('reset');\nfunction countOwned(obs:any,a:string){const me=+obs.player,f=obs.farms[me];let n=+(obs.private?.shed?.[a]??0);for(const inv of obs.private?.inventories??[])n+=+(inv?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n;}\nfunction enforce(obs:any,act:any,t:any){const x:any={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};const o={...act,market:[] as any[]};for(const m of act?.market??[]){if(Array.isArray(m)&&m[0]==='BUY_ANIMAL'&&(m[1]==='COW'||m[1]==='SHEEP')){const a=m[1],q=Math.max(0,Math.min(+m[2]||1,t[a]-x[a]));if(q>0){o.market.push(['BUY_ANIMAL',a,q]);x[a]+=q;}}else o.market.push(m);}return o;}\nfunction lexChoice(obs:any,v:string){const r=rankAnimalInvestments(obs,{cashBuffer:400,strategicWeight:0}).filter(x=>x.animal==='COW'||x.animal==='SHEEP');const positive=r.filter(x=>x.financeable&&x.net>0);if(!positive.length)return{choice:null,ranking:r,switched:false};const best=positive[0];if(v==='PURE')return{choice:best.animal,ranking:r,switched:false};const band=best.net*(1-eps[v]);const eligible=positive.filter(x=>x.net>=band);eligible.sort((a,b)=>(a.rivalPriceImpact-b.rivalPriceImpact)||(b.net-a.net));return{choice:eligible[0].animal,ranking:r,switched:eligible[0].animal!==best.animal};}\nasync function play(v:string,opp:string,seed:number,seat:0|1){reset('DYN');reset(opp);let s=E.state.initGameState(2,{...cfg0,seed},seed),targets:any={COW:0,SHEEP:0},errors=0,switches=0;while(!s.done){const acts:any[]=[];for(const pid of [0,1] as const){const n=pid===seat?'DYN':opp,obs=observationFor(s,pid);if(n==='DYN'&&+obs.hour===0){const owned={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};targets={...owned};const total=owned.COW+owned.SHEEP,c=lexChoice(obs,v);if(total<16&&c.choice){targets[c.choice]+=Math.min(2,16-total);if(c.switched)switches++;}const order=[c.choice,...c.ranking.map(x=>x.animal)].filter((x,i,a)=>x&&(x==='COW'||x==='SHEEP')&&a.indexOf(x)===i);setHerd('DYN',targets.COW,targets.SHEEP,JSON.stringify(order));}try{acts[pid]=JSON.parse(String(call(n,JSON.stringify(obs))));if(n==='DYN')acts[pid]=enforce(obs,acts[pid],targets);}catch{errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]};}}s=E.interpreter.step(s,acts,{...cfg0,seed});}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,opp,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,errors,switches};}\nconst games:any[]=[];for(const v of variants)for(const o of opponents)for(const seed of seeds){games.push(await play(v,o,seed,0));games.push(await play(v,o,seed,1));}const summary:any={};for(const v of variants){const rs=games.filter(g=>g.v===v);summary[v]={episodes:rs.length,wins:rs.filter(r=>r.win).length,winRate:rs.filter(r=>r.win).length/rs.length,avgBank:rs.reduce((a,r)=>a+r.ours,0)/rs.length,avgMargin:rs.reduce((a,r)=>a+r.margin,0)/rs.length,switches:rs.reduce((a,r)=>a+r.switches,0),byOpponent:{}};for(const o of opponents){const xs=rs.filter(r=>r.opp===o);summary[v].byOpponent[o]={wins:xs.filter(r=>r.win).length,winRate:xs.filter(r=>r.win).length/xs.length,avgMargin:xs.reduce((a,r)=>a+r.margin,0)/xs.length,switches:xs.reduce((a,r)=>a+r.switches,0)};}}const b=games.filter(g=>g.v==='PURE');const paired=variants.slice(1).map(v=>{const r=games.filter(g=>g.v===v);return{variant:v,avgMarginDelta:r.reduce((s,g,i)=>s+(g.margin-b[i].margin),0)/r.length,better:r.filter((g,i)=>g.margin>b[i].margin).length,worse:r.filter((g,i)=>g.margina+g.switches,0)};});console.log(JSON.stringify({status:'PASS',hypothesis:'HARM_RIVAL_ONLY_AS_SECONDARY_LEXICOGRAPHIC_TIEBREAKER_SHOULD_AVOID_SELF_DESTRUCTIVE_STRATEGIC_WEIGHTING',rule:'maximize own positive financeable NPV first; only within epsilon-band choose animal causing larger rival price damage',summary,paired},null,2));",[],[1496],"8b6332b6-c155-423d-82c8-851a55857d31","kag_lena_d9_v3_teacher.ts","01a07a92-e81b-76d4-9d53-d6de52b3f8fe",403,["D",1788762928975],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';import { extractTeacherFeaturesV3,KAG_TEACHER_FEATURE_SCHEMA_V3 } from './lib/kag_teacher_features_v3.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=1,OPPONENT='ledger_lena.py';const cfg=E.state.resolveConfig({seed:997979,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number,attempt=0){const r=rng(2917000000+ci*104729+attempt*49979687);return Array.from({length:3},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number,attempt=0){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any,prev:any){return extractTeacherFeaturesV3(o,prev)}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,997979),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null,prevObs:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];let ownObs:any=null;for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){ownObs=o;const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o,prevObs);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}const nextPrev=ownObs?JSON.parse(JSON.stringify(ownObs)):null;s=E.interpreter.step(s,aa,cfg);if(nextPrev)prevObs=nextPrev}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst existingQ=await sqlite.execute(`SELECT DISTINCT prefix_json FROM kag_teacher_q WHERE opponent='${OPPONENT}' AND decision_day=${decisionDay} AND feature_schema='${KAG_TEACHER_FEATURE_SCHEMA_V3}'`),seen=new Set((existingQ.rows??[]).map((r:any)=>String(r.prefix_json)));const contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??1.96;rows.push({action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0);contexts.push({contextId:`LENA-D9-V3-RACE-${prefix.join('_')}`,opponent:OPPONENT,decisionDay,prefix,prefixAttempt:attempt-1,preStateIdentical:preSame,featureSchema:KAG_TEACHER_FEATURE_SCHEMA_V3,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes:INNER*modes.length,savingsPct:100*(1-episodesUsed/(INNER*modes.length)),rows})}\nconst sourceArtifact='kag_lena_d9_v3_teacher.ts',evidenceState='TEACHER_Q_RACING_PERSISTENT_LENA_D9_V3';\nconst before=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconst statements:any[]=[];for(const c of contexts){statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_contexts (context_id,opponent,decision_day,prefix_json,feature_json,teacher_pick,teacher_scenarios_used,pre_state_identical,race_audit_json,episodes_used,full_episodes,savings_pct,evidence_state,source_artifact,feature_schema) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),c.teacherPick,c.teacherScenariosUsed,c.preStateIdentical?1:0,JSON.stringify(c.raceAudit),c.episodesUsed,c.fullEpisodes,c.savingsPct,evidenceState,sourceArtifact,c.featureSchema]});for(const r of c.rows)statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_q (context_id,opponent,decision_day,prefix_json,feature_json,action,q_mean,delta_vs_n,q_sd,lcb95,cvar50,positive_rate,sample_n,teacher_pick,evidence_state,source_artifact,ucb95,pruned,feature_schema) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),r.action,r.qMean,r.deltaVsN,r.qSd,r.lcb95,r.cvar50,r.positiveRate,r.sampleN,c.teacherPick,evidenceState,sourceArtifact,r.ucb95,r.pruned?1:0,c.featureSchema]})}\nawait sqlite.batch(statements);\nconst after=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconsole.log(JSON.stringify({status:'PASS',evidenceState,maxSampleN:INNER,persistence:{sourceArtifact,before:before.rows[0],attemptedRows:statements.length,after:after.rows[0]},contexts},null,2));",[],[1504],"bb247b2b-f44b-4c2d-adfd-7c87faf514de","kag_lena_d9_v3_control.ts","01a07a91-f01b-7694-9b4e-350f628832eb",394,["D",1788762779679],[],[1512],"af2c69e2-dfc5-4b8e-84e2-e17e363c04e5","kag_lena_d9_v31_teacher.ts","01a07a98-148c-756f-ad92-1eab7786b0cc",418,["D",1788763231674],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';import { extractTeacherFeaturesV31,KAG_TEACHER_FEATURE_SCHEMA_V31 } from './lib/kag_teacher_features_v31.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=16,CONTEXTS=1,OPPONENT='ledger_lena.py';const cfg=E.state.resolveConfig({seed:997979,weedSpawnChance:0,townShopUnlockInterval:9999});\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}function mkPrefix(ci:number,attempt=0){const r=rng(2917000000+ci*104729+attempt*49979687);return Array.from({length:3},()=>NAMES[Math.floor(r()*NAMES.length)])}function innerWorlds(prefix:string[],ci:number,attempt=0){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any,prev:any){return extractTeacherFeaturesV31(o,prev)}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,997979),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null,prevObs:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];let ownObs:any=null;for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){ownObs=o;const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o,prevObs);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}const nextPrev=ownObs?JSON.parse(JSON.stringify(ownObs)):null;s=E.interpreter.step(s,aa,cfg);if(nextPrev)prevObs=nextPrev}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:16,audit}}\nconst existingQ=await sqlite.execute(`SELECT DISTINCT prefix_json FROM kag_teacher_q WHERE opponent='${OPPONENT}' AND decision_day=${decisionDay} AND feature_schema='${KAG_TEACHER_FEATURE_SCHEMA_V31}'`),seen=new Set((existingQ.rows??[]).map((r:any)=>String(r.prefix_json)));const contexts:any[]=[];for(let ci=0;ciJSON.stringify(x))).size===1,t=teacher(vals),rows:any[]=[];for(const m of modes){const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??1.96;rows.push({action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0);contexts.push({contextId:`LENA-D9-V31-RACE-${prefix.join('_')}`,opponent:OPPONENT,decisionDay,prefix,prefixAttempt:attempt-1,preStateIdentical:preSame,featureSchema:KAG_TEACHER_FEATURE_SCHEMA_V31,features:pres[0],teacherPick:t.pick,teacherScenariosUsed:t.used,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes:INNER*modes.length,savingsPct:100*(1-episodesUsed/(INNER*modes.length)),rows})}\nconst sourceArtifact='kag_lena_d9_v3_teacher.ts',evidenceState='TEACHER_Q_RACING_PERSISTENT_LENA_D9_V3';\nconst before=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconst statements:any[]=[];for(const c of contexts){statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_contexts (context_id,opponent,decision_day,prefix_json,feature_json,teacher_pick,teacher_scenarios_used,pre_state_identical,race_audit_json,episodes_used,full_episodes,savings_pct,evidence_state,source_artifact,feature_schema) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),c.teacherPick,c.teacherScenariosUsed,c.preStateIdentical?1:0,JSON.stringify(c.raceAudit),c.episodesUsed,c.fullEpisodes,c.savingsPct,evidenceState,sourceArtifact,c.featureSchema]});for(const r of c.rows)statements.push({sql:'INSERT OR REPLACE INTO kag_teacher_q (context_id,opponent,decision_day,prefix_json,feature_json,action,q_mean,delta_vs_n,q_sd,lcb95,cvar50,positive_rate,sample_n,teacher_pick,evidence_state,source_artifact,ucb95,pruned,feature_schema) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)',args:[c.contextId,c.opponent,c.decisionDay,JSON.stringify(c.prefix),JSON.stringify(c.features),r.action,r.qMean,r.deltaVsN,r.qSd,r.lcb95,r.cvar50,r.positiveRate,r.sampleN,c.teacherPick,evidenceState,sourceArtifact,r.ucb95,r.pruned?1:0,c.featureSchema]})}\nawait sqlite.batch(statements);\nconst after=await sqlite.execute('SELECT COUNT(*) AS rows, COUNT(DISTINCT context_id) AS contexts FROM kag_teacher_q');\nconsole.log(JSON.stringify({status:'PASS',evidenceState,maxSampleN:INNER,persistence:{sourceArtifact,before:before.rows[0],attemptedRows:statements.length,after:after.rows[0]},contexts},null,2));",[],[1519],"cc122abf-ce35-4b78-a952-a65e99bbd61d","kag_initial_commitment_counterfactual_v1.ts","dbd2a276-aa4f-11f1-86a2-1607ee4eb77e",89,["D",1788739539181],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['BASE','COW2','SHEEP2','MIX11','WAIT3'];const seeds=[9403,9461,9491];const opp='RITA_BASE';const E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();const zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await zr.arrayBuffer()));const src=strFromU8(z['rancher_rita.py']);const py=await loadPyodide();py.runPython(`import json,copy\\n_REG={}\\n_BASE={}`);py.globals.set('src',src);for(const [n,p] of Object.entries({DYN:{animal_target:{COW:0,SHEEP:0}},RITA_BASE:{}})){py.globals.set('nm',n);py.globals.set('patch',JSON.stringify(p));py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n\\n_BASE[nm]=copy.deepcopy(_n['POLICY'])`);}py.runPython(`\\ndef call(name,payload): return json.dumps(_REG[name]['agent'](json.loads(payload)))\\ndef set_herd(name,cows,sheep,order):\\n _REG[name]['POLICY']['animal_target']={'COW':int(cows),'SHEEP':int(sheep)}\\n _REG[name]['POLICY']['animals']=json.loads(order)\\ndef reset(name):\\n _REG[name]['POLICY'].clear();_REG[name]['POLICY'].update(copy.deepcopy(_BASE[name]))\\n`);const call=py.globals.get('call'),setHerd=py.globals.get('set_herd'),reset=py.globals.get('reset');\nfunction countOwned(obs:any,a:string){const me=+obs.player,f=obs.farms[me];let n=+(obs.private?.shed?.[a]??0);for(const inv of obs.private?.inventories??[])n+=+(inv?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n;}\nfunction enforce(obs:any,act:any,t:any){const x:any={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};const o={...act,market:[] as any[]};for(const m of act?.market??[]){if(Array.isArray(m)&&m[0]==='BUY_ANIMAL'&&(m[1]==='COW'||m[1]==='SHEEP')){const a=m[1],q=Math.max(0,Math.min(+m[2]||1,t[a]-x[a]));if(q>0){o.market.push(['BUY_ANIMAL',a,q]);x[a]+=q;}}else o.market.push(m);}return o;}\nasync function play(v:string,seed:number,seat:0|1){reset('DYN');reset(opp);let s=E.state.initGameState(2,{...cfg0,seed},seed),targets:any={COW:0,SHEEP:0},errors=0;const diary:any[]=[];while(!s.done){const acts:any[]=[];for(const pid of [0,1] as const){const n=pid===seat?'DYN':opp,obs=observationFor(s,pid);if(n==='DYN'&&+obs.hour===0){const owned={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};targets={...owned};let order:string[]=[];if(+obs.day===0&&v!=='BASE'&&v!=='WAIT3'){if(v==='COW2'){targets.COW=2;order=['COW','SHEEP'];}if(v==='SHEEP2'){targets.SHEEP=2;order=['SHEEP','COW'];}if(v==='MIX11'){targets={COW:1,SHEEP:1};order=['SHEEP','COW'];}}else if(v==='WAIT3'&&+obs.day<3){order=['COW','SHEEP'];}else{const total=owned.COW+owned.SHEEP,c=choosePastureAnimal(obs,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)targets[c.choice]+=Math.min(2,16-total);order=c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP') as string[];}setHerd('DYN',targets.COW,targets.SHEEP,JSON.stringify(order));if([0,3,6,9,12,15,18].includes(+obs.day))diary.push({day:+obs.day,cash:obs.farms[pid].money,owned:{...owned},targets:{...targets},shops:[...(obs.town?.unlocked_shops??[])]});}try{acts[pid]=JSON.parse(String(call(n,JSON.stringify(obs))));if(n==='DYN')acts[pid]=enforce(obs,acts[pid],targets);}catch{errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]};}}s=E.interpreter.step(s,acts,{...cfg0,seed});}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,errors,diary};}\nconst games:any[]=[];for(const v of variants)for(const seed of seeds){games.push(await play(v,seed,0));games.push(await play(v,seed,1));}const summary:any={};for(const v of variants){const rs=games.filter(g=>g.v===v);summary[v]={episodes:rs.length,wins:rs.filter(r=>r.win).length,avgBank:rs.reduce((a,r)=>a+r.ours,0)/rs.length,avgMargin:rs.reduce((a,r)=>a+r.margin,0)/rs.length,bySeed:Object.fromEntries(seeds.map(seed=>{const x=rs.filter(r=>r.seed===seed);return[seed,{wins:x.filter(r=>r.win).length,avgMargin:x.reduce((a,r)=>a+r.margin,0)/x.length,avgBank:x.reduce((a,r)=>a+r.ours,0)/x.length}]}))};}const b=games.filter(g=>g.v==='BASE');const paired=variants.slice(1).map(v=>{const r=games.filter(g=>g.v===v);return{variant:v,avgMarginDelta:r.reduce((s,g,i)=>s+(g.margin-b[i].margin),0)/r.length,better:r.filter((g,i)=>g.margin>b[i].margin).length,worse:r.filter((g,i)=>g.marging.seed===9491&&g.seat===0).map(g=>({v:g.v,diary:g.diary}))},null,2));",[],[1527],"f71c2a35-9d1c-4a6f-8cee-86616e05dd0d","kag_initial_action_shadow_probe_v1.ts","402a660a-aa50-11f1-b41f-1607ee4eb77e",90,["D",1788739707571],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const src=strFromU8(z['rancher_rita.py']);const py=await loadPyodide();py.runPython(`import json,copy\\n_REG={}\\n_BASE={}`);py.globals.set('src',src);for(const [n,p] of Object.entries({BASE:{animal_target:{COW:0,SHEEP:0}},COW2:{animal_target:{COW:2,SHEEP:0},animals:['COW','SHEEP']},SHEEP2:{animal_target:{COW:0,SHEEP:2},animals:['SHEEP','COW']},MIX11:{animal_target:{COW:1,SHEEP:1},animals:['SHEEP','COW']}})){py.globals.set('nm',n);py.globals.set('patch',JSON.stringify(p));py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n\\n_BASE[nm]=copy.deepcopy(_n['POLICY'])`);}py.runPython(`\\ndef call(name,payload): return json.dumps(_REG[name]['agent'](json.loads(payload)))\\ndef reset(name): _REG[name]['POLICY'].clear();_REG[name]['POLICY'].update(copy.deepcopy(_BASE[name]))\\n`);const call=py.globals.get('call');\nfunction owned(s:any,pid:number){const f=s.farms[pid];const c:any={COW:0,SHEEP:0};for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&(t.animal==='COW'||t.animal==='SHEEP'))c[t.animal]++;const pr=observationFor(s,pid)?.private;for(const a of ['COW','SHEEP']){c[a]+=Number(pr?.shed?.[a]??0);for(const inv of pr?.inventories??[])c[a]+=Number(inv?.[a]??0);}return c;}\nconst out:any={};for(const name of ['BASE','COW2','SHEEP2','MIX11']){let s=E.state.initGameState(2,{...cfg0,seed:9491},9491);const steps:any[]=[];for(let t=0;t<72;t++){const obs=observationFor(s,0);const act=JSON.parse(String(call(name,JSON.stringify(obs))));if(t<8||act.market?.some((x:any)=>x?.[0]==='BUY_ANIMAL'))steps.push({turn:t,day:s.day,hour:s.hour,money:s.farms[0].money,owned:owned(s,0),market:act.market});s=E.interpreter.step(s,[act,{farmer:['PASS'],hands:[],market:[]}],{...cfg0,seed:9491});}out[name]=steps;}console.log(JSON.stringify({status:'PASS',seed:9491,out},null,2));",[],[1535],"cc9af65d-e0bf-40c2-b9aa-82664834f86e","kag_hold_router_heldout_v1.ts","2f1147e6-aa67-11f1-9dee-1607ee4eb77e",189,["D",1788749557292],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';import { lowerTailCVaR } from './lib/kaggriculture_farm_economics.ts';\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:929292,weedSpawnChance:0,townShopUnlockInterval:9999});const NAMES=[...E.constants.SHOP_NAMES];function rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const worlds:any[]=[];for(let i=0;i<24;i++){const r=rng(4200000+i*130363);worlds.push({id:`H${i}`,shops:Array.from({length:8},()=>NAMES[Math.floor(r()*NAMES.length)])})}const shopHas=(s:string,p:string)=>(E.constants.SHOPS[s]??[]).includes(p);\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['broker_bea.py']));py.runPython(`import json\\n_r={}\\nexec(src,_r)\\n_REF=_r['agent']`);py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(p):return json.dumps(_REF(json.loads(p)))\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number,s:string[]){return s.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction holdDecision(strategy:string,o:any){const d=+o.day;if(d<3||d>6)return false;if(strategy==='STATIC')return true;if(strategy!=='ROUTER')return false;const shops=[...(o.town?.unlocked_shops??[])];if(!shops.length)return false;const first=shops[0];if(first!=='PET_CAFE')return true;return shops.some((s:string)=>shopHas(s,'MILK'))}\nasync function play(strategy:string,w:any,seat:0|1){prep();let s=E.state.initGameState(2,cfg,929292),tg:any={COW:0,SHEEP:0},comm=false,minCash=Infinity,maxFlock=0,finalFlock=0,heldTurns=0;while(!s.done){s.town.unlocked_shops=fixed(+s.day,w.shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}let a=addFert(o,enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg));if(holdDecision(strategy,o)){heldTurns++;a={...a,market:(a.market??[]).filter((x:any)=>!(x?.[0]==='SELL'&&x?.[1]==='WHEAT'))}}aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money);maxFlock=Math.max(maxFlock,own(o,'COW')+own(o,'SHEEP'))}else aa[pid]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{strategy,world:w.id,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,win:s.farms[seat].money>s.farms[1-seat].money,minCash,animalLoss:maxFlock-finalFlock,heldTurns}}\nconst games:any[]=[];for(const w of worlds)for(const strategy of ['BASE','STATIC','ROUTER']){games.push(await play(strategy,w,0));games.push(await play(strategy,w,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),q=(xs:number[],p:number)=>{const a=[...xs].sort((x,y)=>x-y);return a[Math.min(a.length-1,Math.max(0,Math.floor((a.length-1)*p)))]};const base=games.filter(g=>g.strategy==='BASE'),summary:any={},paired:any={};for(const strategy of ['BASE','STATIC','ROUTER']){const x=games.filter(g=>g.strategy===strategy),m=x.map(g=>g.margin);summary[strategy]={episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,bank:avg(x.map(g=>g.ours)),opp:avg(x.map(g=>g.theirs)),margin:avg(m),medianMargin:q(m,.5),p10Margin:q(m,.1),cvar10:lowerTailCVaR(m,.1),minCash:Math.min(...x.map(g=>g.minCash)),animalLoss:avg(x.map(g=>g.animalLoss)),avgHeldTurns:avg(x.map(g=>g.heldTurns))};if(strategy!=='BASE'){const d=x.map((g,i)=>g.margin-base[i].margin);paired[strategy]={avgDelta:avg(d),medianDelta:q(d,.5),p10Delta:q(d,.1),p90Delta:q(d,.9),better:d.filter(z=>z>0).length,worse:d.filter(z=>z<0).length,winFlips:x.filter((g,i)=>g.win&&!base[i].win).length,lossFlips:x.filter((g,i)=>!g.win&&base[i].win).length}}}console.log(JSON.stringify({status:'PASS',evidenceState:'HELDOUT_EXOGENOUS_WORLDS_RULE_NOT_TUNED_HERE',rule:'day3-6 hold Wheat unless first shop is PET_CAFE; if first PET_CAFE, begin holding only after any observed MILK-demanding shop appears',worldCount:worlds.length,summary,paired},null,2));",[],[1543],"15216830-ec51-41d7-8933-be9855a9b3fe","kag_hold_d3_discovery_v1.ts","de1603fe-aa66-11f1-938b-1607ee4eb77e",186,["D",1788749421441],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:919191,weedSpawnChance:0,townShopUnlockInterval:9999});const NAMES=[...E.constants.SHOP_NAMES];function rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const worlds:any[]=[];for(let i=0;i<24;i++){const r=rng(1200000+i*104729);worlds.push({id:`D${i}`,shops:Array.from({length:8},()=>NAMES[Math.floor(r()*NAMES.length)])})}const shopHas=(s:string,p:string)=>(E.constants.SHOPS[s]??[]).includes(p);\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['broker_bea.py']));py.runPython(`import json\\n_r={}\\nexec(src,_r)\\n_REF=_r['agent']`);py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(p):return json.dumps(_REF(json.loads(p)))\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number,s:string[]){return s.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(hold:boolean,w:any,seat:0|1){prep();let s=E.state.initGameState(2,cfg,919191),tg:any={COW:0,SHEEP:0},comm=false;while(!s.done){s.town.unlocked_shops=fixed(+s.day,w.shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}let a=addFert(o,enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg));if(hold&&+o.day>=3&&+o.day<=6)a={...a,market:(a.market??[]).filter((x:any)=>!(x?.[0]==='SELL'&&x?.[1]==='WHEAT'))};aa[pid]=a}else aa[pid]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{s:hold?'HOLD_D3_6':'BASE',world:w.id,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money}}\nconst games:any[]=[];for(const w of worlds)for(const hold of [false,true]){games.push(await play(hold,w,0));games.push(await play(hold,w,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const rows=worlds.map(w=>{const b=games.filter(g=>g.world===w.id&&g.s==='BASE'),h=games.filter(g=>g.world===w.id&&g.s==='HOLD_D3_6');const prefix=w.shops.slice(0,2);return{world:w.id,first:w.shops[0],second:w.shops[1],prefix,earlyWheat:prefix.filter((s:string)=>shopHas(s,'WHEAT')).length,earlyMilk:prefix.filter((s:string)=>shopHas(s,'MILK')).length,earlyWool:prefix.filter((s:string)=>shopHas(s,'WOOL')).length,earlyStraw:prefix.filter((s:string)=>shopHas(s,'STRAWBERRY')).length,totalWheat:w.shops.filter((s:string)=>shopHas(s,'WHEAT')).length,totalMilk:w.shops.filter((s:string)=>shopHas(s,'MILK')).length,totalWool:w.shops.filter((s:string)=>shopHas(s,'WOOL')).length,totalStraw:w.shops.filter((s:string)=>shopHas(s,'STRAWBERRY')).length,delta:avg(h.map((g,i)=>g.margin-b[i].margin))}});function group(key:(r:any)=>string){const m:any={};for(const r of rows){const k=key(r);(m[k]??=[]).push(r.delta)}return Object.fromEntries(Object.entries(m).map(([k,x]:any)=>[k,{n:x.length,mean:avg(x),positive:x.filter((z:number)=>z>0).length,min:Math.min(...x),max:Math.max(...x)}]))}console.log(JSON.stringify({status:'PASS',evidenceState:'DISCOVERY_ONLY_EXOGENOUS_WORLDS',overall:{n:rows.length,meanDelta:avg(rows.map(r=>r.delta)),positive:rows.filter(r=>r.delta>0).length},byFirstShop:group(r=>r.first),byEarlyWheat:group(r=>String(r.earlyWheat)),byEarlyMilk:group(r=>String(r.earlyMilk)),byEarlyWool:group(r=>String(r.earlyWool)),byEarlyStraw:group(r=>String(r.earlyStraw)),diagnosticFuture:{byTotalWheat:group(r=>String(r.totalWheat)),byTotalMilk:group(r=>String(r.totalMilk)),byTotalWool:group(r=>String(r.totalWool)),byTotalStraw:group(r=>String(r.totalStraw))},rows},null,2));",[],[1551],"2d837a53-a70b-4eaa-823a-87a8e51103a6","kag_hands_lifecycle_probe_v1.ts","8b607918-aa85-11f1-be7b-1607ee4eb77e",389,["D",1788762597082],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nconst E=await loadOfficialKaggricultureEngine();const cfg=E.state.resolveConfig({seed:778899,weedSpawnChance:0});const rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`import json\\n_ns={}\\nexec(src,_ns)\\n_A=_ns['agent']\\ndef call(p):return json.dumps(_A(json.loads(p)))`);const call=py.globals.get('call');\nfunction snap(o:any){const f=o.farms[+o.player];return{day:o.day,hour:o.hour,step:o.step,handsLen:(f.hands??[]).length,hiresToday:f.hires_today,farmer:f.farmer,hand0:(f.hands??[])[0]??null,handKeys:(f.hands??[])[0]?Object.keys((f.hands??[])[0]):[],privateInventories:o.private?.inventories,seeds:o.private?.seeds}}\nconst targets=new Set(['8:23','9:0','9:1','9:2','9:23','10:0','10:1']);const rows:any[]=[];let s=E.state.initGameState(2,cfg,778899);while(!s.done){const o=observationFor(s,0),k=`${o.day}:${o.hour}`,a=JSON.parse(String(call(JSON.stringify(o))));if(targets.has(k))rows.push({phase:'BEFORE',...snap(o),market:a.market,handsAction:a.hands,farmerAction:a.farmer,hireOrders:(a.market??[]).filter((x:any)=>JSON.stringify(x).toUpperCase().includes('HIRE'))});s=E.interpreter.step(s,[a,{farmer:[],hands:[],market:[]}],cfg);if(targets.has(k)&&!s.done){const p=observationFor(s,0);rows.push({phase:'AFTER_STEP',from:k,...snap(p)})}}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'HANDS_DAILY_LIFECYCLE_PROBE',rows},null,2));",[],[1559],"e9334685-99c2-43af-8f8d-a624835c66b1","kag_flightplan_router_v2_top.ts","01a079ce-fabe-7548-b735-6515023fb699",198,["D",1788750096118],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:949494,weedSpawnChance:0,townShopUnlockInterval:9999});const NAMES=[...E.constants.SHOP_NAMES];function rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const worlds:any[]=[];for(let i=0;i<3;i++){const r=rng(8200000+i*15485863);worlds.push({id:`V2${i}`,shops:Array.from({length:8},()=>NAMES[Math.floor(r()*NAMES.length)])})}const refs=['broker_bea.py','ledger_lena.py','slotter_silas.py','closer_cleo.py'];const shopHas=(s:string,p:string)=>(E.constants.SHOPS[s]??[]).includes(p);\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setcrop(m): _D['POLICY']['crop_share']={'WHEAT':1.0-float(m),'MELON':float(m)}\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setcrop=py.globals.get('setcrop');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number,s:string[]){return s.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}function pub(f:any){let melon=0,wheat=0,animals=0;for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT'&&t.crop==='MELON')melon++;if(t.kind==='PLANT'&&t.crop==='WHEAT')wheat++;if(t.animal)animals++}return{melon,wheat,animals,money:+f.money,land:(f?.unlocked_quadrants??[]).length}}\nfunction rawClass(o:any){const me=+o.player,f=pub(o.farms[1-me]);if((f.melon>=5&&f.animals>=2)||(f.melon>=5&&f.wheat>=5&&f.money<150&&f.land<=1))return'TOP_MIXED';if(f.melon>=8&&f.animals===0&&f.land>=2)return'CROP';if(f.melon===0&&f.wheat>=8&&(f.animals>=2||f.money>1000))return'LIVESTOCK';return'UNKNOWN'}function advanceClass(prev:string,raw:string){if(prev==='TOP_MIXED')return prev;if(raw==='TOP_MIXED')return raw;if(prev==='CROP'||prev==='LIVESTOCK')return prev;return raw}function shareFor(cls:string){return cls==='CROP'?.5:cls==='LIVESTOCK'?.7:.6}function holdTop(o:any,cls:string){const d=+o.day;if(cls!=='TOP_MIXED'||d<3||d>6)return false;const shops=[...(o.town?.unlocked_shops??[])];if(!shops.length)return false;if(shops[0]!=='PET_CAFE')return true;return shops.some((s:string)=>shopHas(s,'MILK'))}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(strategy:string,ref:string,w:any,seat:0|1){prep();let s=E.state.initGameState(2,cfg,949494),tg:any={COW:0,SHEEP:0},comm=false,minCash=Infinity,maxFlock=0,finalFlock=0,latched='UNKNOWN';const classTurns:any={};while(!s.done){s.town.unlocked_shops=fixed(+s.day,w.shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){const raw=rawClass(o);if(strategy==='V2')latched=advanceClass(latched,raw);const cls=strategy==='BASE'?'UNKNOWN':strategy==='V1'?raw:latched;const share=strategy==='BASE'?.6:shareFor(cls);setcrop(share);classTurns[cls]=(classTurns[cls]??0)+1;if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}let a=addFert(o,enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg));if(strategy!=='BASE'&&holdTop(o,cls))a={...a,market:(a.market??[]).filter((x:any)=>!(x?.[0]==='SELL'&&x?.[1]==='WHEAT'))};aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money);maxFlock=Math.max(maxFlock,own(o,'COW')+own(o,'SHEEP'))}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{strategy,ref,world:w.id,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,win:s.farms[seat].money>s.farms[1-seat].money,minCash,animalLoss:maxFlock-finalFlock,classTurns}}\nconst games:any[]=[];for(const w of worlds)for(const ref of refs)for(const strategy of ['BASE','V1','V2']){games.push(await play(strategy,ref,w,0));games.push(await play(strategy,ref,w,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const summary:any={};for(const ref of refs){summary[ref]={};const b=games.filter(g=>g.ref===ref&&g.strategy==='BASE');for(const strategy of ['BASE','V1','V2']){const a=games.filter(g=>g.ref===ref&&g.strategy===strategy);summary[ref][strategy]={episodes:a.length,wins:a.filter((g:any)=>g.win).length,winRate:a.filter((g:any)=>g.win).length/a.length,bank:avg(a.map((g:any)=>g.ours)),opp:avg(a.map((g:any)=>g.theirs)),margin:avg(a.map((g:any)=>g.margin)),minCash:Math.min(...a.map((g:any)=>g.minCash)),animalLoss:avg(a.map((g:any)=>g.animalLoss)),classExample:a[0]?.classTurns};if(strategy!=='BASE'){const d=a.map((g:any,i:number)=>g.margin-b[i].margin);summary[ref][`${strategy}_delta`]={avg:avg(d),better:d.filter((z:number)=>z>0).length,worse:d.filter((z:number)=>z<0).length}}}}\nconst allB=games.filter(g=>g.strategy==='BASE');const overall:any={};for(const strategy of ['V1','V2']){const a=games.filter(g=>g.strategy===strategy),d=a.map((g,i)=>g.margin-allB[i].margin);overall[strategy]={wins:a.filter(g=>g.win).length,avgDelta:avg(d),better:d.filter(z=>z>0).length,worse:d.filter(z=>z<0).length}}console.log(JSON.stringify({status:'PASS',evidenceState:'HELDOUT_EXOGENOUS_TOP_ROUTER_V1_V2',worldCount:worlds.length,policyV2:'early top-mixed signature + hysteretic capability memory',overall,summary},null,2));",[],[1567],"7b6e37bd-7f83-43d0-82f8-e686272891e5","kag_flightplan_router_v1.ts","9d0778a6-aa67-11f1-b9ad-1607ee4eb77e",191,["D",1788749885222],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:939393,weedSpawnChance:0,townShopUnlockInterval:9999});const NAMES=[...E.constants.SHOP_NAMES];function rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const worlds:any[]=[];for(let i=0;i<3;i++){const r=rng(7200000+i*15485863);worlds.push({id:`F${i}`,shops:Array.from({length:8},()=>NAMES[Math.floor(r()*NAMES.length)])})}const refs=['melon_mateo.py','rancher_rita.py','broker_bea.py','ledger_lena.py','slotter_silas.py','closer_cleo.py'];const shopHas=(s:string,p:string)=>(E.constants.SHOPS[s]??[]).includes(p);\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setcrop(m): _D['POLICY']['crop_share']={'WHEAT':1.0-float(m),'MELON':float(m)}\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setcrop=py.globals.get('setcrop');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number,s:string[]){return s.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}function pub(f:any){let melon=0,wheat=0,animals=0;for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT'&&t.crop==='MELON')melon++;if(t.kind==='PLANT'&&t.crop==='WHEAT')wheat++;if(t.animal)animals++}return{melon,wheat,animals,money:+f.money}}\nfunction classify(o:any){const me=+o.player,f=pub(o.farms[1-me]);if(f.melon>=5&&f.animals>=2)return'TOP_MIXED';if(f.melon>=8&&f.animals===0)return'CROP';if(f.melon===0&&f.wheat>=8&&(f.animals>=2||f.money>1000))return'LIVESTOCK';return'UNKNOWN'}function shareFor(cls:string){return cls==='CROP'?.5:cls==='LIVESTOCK'?.7:.6}function holdTop(o:any,cls:string){const d=+o.day;if(cls!=='TOP_MIXED'||d<3||d>6)return false;const shops=[...(o.town?.unlocked_shops??[])];if(!shops.length)return false;if(shops[0]!=='PET_CAFE')return true;return shops.some((s:string)=>shopHas(s,'MILK'))}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(strategy:string,ref:string,w:any,seat:0|1){prep();let s=E.state.initGameState(2,cfg,939393),tg:any={COW:0,SHEEP:0},comm=false,minCash=Infinity,maxFlock=0,finalFlock=0;const classTurns:any={};while(!s.done){s.town.unlocked_shops=fixed(+s.day,w.shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){const cls=strategy==='FLIGHT'?classify(o):'UNKNOWN';const share=strategy==='FLIGHT'?shareFor(cls):.6;setcrop(share);classTurns[cls]=(classTurns[cls]??0)+1;if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}let a=addFert(o,enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg));if(strategy==='FLIGHT'&&holdTop(o,cls))a={...a,market:(a.market??[]).filter((x:any)=>!(x?.[0]==='SELL'&&x?.[1]==='WHEAT'))};aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money);maxFlock=Math.max(maxFlock,own(o,'COW')+own(o,'SHEEP'))}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{strategy,ref,world:w.id,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,win:s.farms[seat].money>s.farms[1-seat].money,minCash,animalLoss:maxFlock-finalFlock,classTurns}}\nconst games:any[]=[];for(const w of worlds)for(const ref of refs)for(const strategy of ['BASE','FLIGHT']){games.push(await play(strategy,ref,w,0));games.push(await play(strategy,ref,w,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const summary:any={};for(const ref of refs){summary[ref]={};const b=games.filter(g=>g.ref===ref&&g.strategy==='BASE'),x=games.filter(g=>g.ref===ref&&g.strategy==='FLIGHT');for(const [name,a] of [['BASE',b],['FLIGHT',x]] as any){summary[ref][name]={episodes:a.length,wins:a.filter((g:any)=>g.win).length,winRate:a.filter((g:any)=>g.win).length/a.length,bank:avg(a.map((g:any)=>g.ours)),opp:avg(a.map((g:any)=>g.theirs)),margin:avg(a.map((g:any)=>g.margin)),minCash:Math.min(...a.map((g:any)=>g.minCash)),animalLoss:avg(a.map((g:any)=>g.animalLoss))}}const d=x.map((g,i)=>g.margin-b[i].margin);summary[ref].delta={avg:avg(d),better:d.filter(z=>z>0).length,worse:d.filter(z=>z<0).length,winFlips:x.filter((g,i)=>g.win&&!b[i].win).length,lossFlips:x.filter((g,i)=>!g.win&&b[i].win).length,classExample:x[0]?.classTurns}}\nconst allB=games.filter(g=>g.strategy==='BASE'),allF=games.filter(g=>g.strategy==='FLIGHT'),dd=allF.map((g,i)=>g.margin-allB[i].margin);console.log(JSON.stringify({status:'PASS',evidenceState:'HELDOUT_EXOGENOUS_OPPONENT_ROUTER_SCREEN',worldCount:worlds.length,policy:{CROP:.5,LIVESTOCK:.7,TOP_MIXED:.6,UNKNOWN:.6,TOP_MIXED_WHEAT_HOLD:'learned heldout rule'},overall:{baseWins:allB.filter(g=>g.win).length,flightWins:allF.filter(g=>g.win).length,avgDelta:avg(dd),better:dd.filter(z=>z>0).length,worse:dd.filter(z=>z<0).length},summary},null,2));",[],[1575],"73a7cb9c-1353-4b4e-8154-69a5947f7e7a","kag_fixed_world_wheat_timing_v1.ts","24aabf54-aa66-11f1-9f42-1607ee4eb77e",182,["D",1788749172242],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={E0:0,E8:8,E16:16,E24:24};const worlds:Record={NON_STRAW:['BAKERY','YARN_STORE','PIZZA_SHOP','PET_CAFE','BAKERY','YARN_STORE','PIZZA_SHOP','PET_CAFE'],STRAW_MAX:['SMOOTHIE_SHOP','FARMERS_MARKET','ICE_CREAM_SHOP','BRUNCH_SPOT','SMOOTHIE_SHOP','FARMERS_MARKET','ICE_CREAM_SHOP','BRUNCH_SPOT'],MIXED:['SMOOTHIE_SHOP','FARMERS_MARKET','ICE_CREAM_SHOP','YARN_STORE','FARMERS_MARKET','BAKERY','SMOOTHIE_SHOP','BAKERY']};\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:717171,weedSpawnChance:0,townShopUnlockInterval:9999});const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['broker_bea.py']));py.runPython(`import json\\n_r={}\\nexec(src,_r)\\n_REF=_r['agent']`);py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(p):return json.dumps(_REF(json.loads(p)))\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number,s:string[]){return s.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(v:string,world:string,seat:0|1){prep();let s=E.state.initGameState(2,cfg,717171),tg:any={COW:0,SHEEP:0},comm=false,minCash=Infinity,maxFlock=0,finalFlock=0,forced=false,forcedDay=-1,forcedPrice=0;const days:any[]=[];while(!s.done){s.town.unlocked_shops=fixed(+s.day,worlds[world]);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){if(+o.hour===0){const wheat=+(o.private?.shed?.WHEAT??0)+(o.private?.inventories??[]).reduce((a:number,i:any)=>a+Number(i?.WHEAT??0),0);days.push({day:+o.day,ourCash:+o.farms[pid].money,oppCash:+o.farms[1-pid].money,wheatP:+o.market.prices.WHEAT,wheatI:+o.market.inventory.WHEAT,ownedWheat:wheat});const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}let a=addFert(o,enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg));const q=variants[v];if(q>0&&+o.day<=6)a={...a,market:(a.market??[]).filter((x:any)=>!(x?.[0]==='SELL'&&x?.[1]==='WHEAT'))};if(q>0&&!forced&&+o.day>=2&&+o.day<=4&&+o.market.prices.WHEAT<=35&&(a.market?.length??0)<10&&+o.farms[pid].money>=q*(+o.market.prices.WHEAT)+800){a={...a,market:[['BUY_PRODUCT','WHEAT',q],...(a.market??[])]};forced=true;forcedDay=+o.day;forcedPrice=+o.market.prices.WHEAT}aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money);maxFlock=Math.max(maxFlock,own(o,'COW')+own(o,'SHEEP'))}else aa[pid]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{v,world,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,win:s.farms[seat].money>s.farms[1-seat].money,minCash,animalLoss:maxFlock-finalFlock,forced,forcedDay,forcedPrice,days}}\nconst games:any[]=[];for(const world of Object.keys(worlds))for(const v of Object.keys(variants)){games.push(await play(v,world,0));games.push(await play(v,world,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const summary:any={},causal:any[]=[];for(const world of Object.keys(worlds)){summary[world]={};const base=games.filter(g=>g.world===world&&g.v==='E0');for(const v of Object.keys(variants)){const x=games.filter(g=>g.world===world&&g.v===v);summary[world][v]={wins:x.filter(g=>g.win).length,bank:avg(x.map(g=>g.ours)),opp:avg(x.map(g=>g.theirs)),margin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),animalLoss:avg(x.map(g=>g.animalLoss)),forcedRate:x.filter(g=>g.forced).length/x.length,forcedDay:avg(x.filter(g=>g.forced).map(g=>g.forcedDay)),forcedPrice:avg(x.filter(g=>g.forced).map(g=>g.forcedPrice)),days:[2,3,4,6,9,12].map(day=>{const rr=x.map(g=>g.days.find((d:any)=>d.day===day)).filter(Boolean);return{day,wheatP:avg(rr.map((d:any)=>d.wheatP)),wheatI:avg(rr.map((d:any)=>d.wheatI)),ourCash:avg(rr.map((d:any)=>d.ourCash)),oppCash:avg(rr.map((d:any)=>d.oppCash)),ownedWheat:avg(rr.map((d:any)=>d.ownedWheat))}})};if(v!=='E0')causal.push({world,v,bankDelta:avg(x.map((g,i)=>g.ours-base[i].ours)),oppDelta:avg(x.map((g,i)=>g.theirs-base[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-base[i].margin)),better:x.filter((g,i)=>g.margin>base[i].margin).length,worse:x.filter((g,i)=>g.margin[w,Object.fromEntries(Object.entries(summary[w]).map(([v,x]:any)=>[v,{bank:x.bank,opp:x.opp,margin:x.margin,minCash:x.minCash,animalLoss:x.animalLoss,forcedRate:x.forcedRate,forcedDay:x.forcedDay,forcedPrice:x.forcedPrice}]))]))},null,2));",[],[1583],"be7dba96-b7de-4478-87bc-a26b91256b17","kag_fixed_world_wheat_hedge_v1.ts","720b19cc-aa63-11f1-8b61-1607ee4eb77e",179,["D",1788747951673],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={BASE0:0,H16:16,H24:24,H32:32,H48:48};const worlds:Record={NON_STRAW:['BAKERY','YARN_STORE','PIZZA_SHOP','PET_CAFE','BAKERY','YARN_STORE','PIZZA_SHOP','PET_CAFE'],STRAW_MAX:['SMOOTHIE_SHOP','FARMERS_MARKET','ICE_CREAM_SHOP','BRUNCH_SPOT','SMOOTHIE_SHOP','FARMERS_MARKET','ICE_CREAM_SHOP','BRUNCH_SPOT'],MIXED:['SMOOTHIE_SHOP','FARMERS_MARKET','ICE_CREAM_SHOP','YARN_STORE','FARMERS_MARKET','BAKERY','SMOOTHIE_SHOP','BAKERY']};\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:616161,weedSpawnChance:0,townShopUnlockInterval:9999});const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['broker_bea.py']));py.runPython(`import json\\n_r={}\\nexec(src,_r)\\n_REF=_r['agent']`);py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(p):return json.dumps(_REF(json.loads(p)))\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setpf(x): _D['POLICY']['prefeed_stock']=int(x)\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setpf=py.globals.get('setpf');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number,s:string[]){return s.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(v:string,world:string,seat:0|1){prep();let s=E.state.initGameState(2,cfg,616161),tg:any={COW:0,SHEEP:0},comm=false,minCash=Infinity,maxFlock=0,finalFlock=0;const days:any[]=[],ord:any={ourBuyWheat:0,ourSellWheat:0,oppBuyWheat:0,oppSellWheat:0};while(!s.done){s.town.unlocked_shops=fixed(+s.day,worlds[world]);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){const target=(+o.day<=12)?variants[v]:0;setpf(target);if(+o.hour===0){const wheat=+(o.private?.shed?.WHEAT??0)+(o.private?.inventories??[]).reduce((a:number,i:any)=>a+Number(i?.WHEAT??0),0);days.push({day:+o.day,ourCash:+o.farms[pid].money,oppCash:+o.farms[1-pid].money,wheatP:+o.market.prices.WHEAT,wheatI:+o.market.inventory.WHEAT,ownedWheat:wheat});const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}const a=addFert(o,enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg));for(const q of a.market??[]){if(q?.[0]==='BUY_PRODUCT'&&q?.[1]==='WHEAT')ord.ourBuyWheat+=Number(q[2]??0);if(q?.[0]==='SELL'&&q?.[1]==='WHEAT')ord.ourSellWheat+=Number(q[2]??0)}aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money);maxFlock=Math.max(maxFlock,own(o,'COW')+own(o,'SHEEP'))}else{const a=JSON.parse(String(cr(JSON.stringify(o))));for(const q of a.market??[]){if(q?.[0]==='BUY_PRODUCT'&&q?.[1]==='WHEAT')ord.oppBuyWheat+=Number(q[2]??0);if(q?.[0]==='SELL'&&q?.[1]==='WHEAT')ord.oppSellWheat+=Number(q[2]??0)}aa[pid]=a}}s=E.interpreter.step(s,aa,cfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{v,world,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,win:s.farms[seat].money>s.farms[1-seat].money,minCash,animalLoss:maxFlock-finalFlock,ord,days}}\nconst games:any[]=[];for(const world of Object.keys(worlds))for(const v of Object.keys(variants)){games.push(await play(v,world,0));games.push(await play(v,world,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const summary:any={};const causal:any[]=[];for(const world of Object.keys(worlds)){summary[world]={};const base=games.filter(g=>g.world===world&&g.v==='BASE0');for(const v of Object.keys(variants)){const x=games.filter(g=>g.world===world&&g.v===v);summary[world][v]={wins:x.filter(g=>g.win).length,bank:avg(x.map(g=>g.ours)),opp:avg(x.map(g=>g.theirs)),margin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),animalLoss:avg(x.map(g=>g.animalLoss)),ourBuyWheat:avg(x.map(g=>g.ord.ourBuyWheat)),ourSellWheat:avg(x.map(g=>g.ord.ourSellWheat)),oppBuyWheat:avg(x.map(g=>g.ord.oppBuyWheat)),oppSellWheat:avg(x.map(g=>g.ord.oppSellWheat))};if(v!=='BASE0')causal.push({world,v,bankDelta:avg(x.map((g,i)=>g.ours-base[i].ours)),oppDelta:avg(x.map((g,i)=>g.theirs-base[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-base[i].margin)),better:x.filter((g,i)=>g.margin>base[i].margin).length,worse:x.filter((g,i)=>g.margin={\n NO_SHOPS:[],\n NON_STRAW:['BAKERY','YARN_STORE','PIZZA_SHOP','PET_CAFE','BAKERY','YARN_STORE','PIZZA_SHOP','PET_CAFE'],\n STRAW_MAX:['SMOOTHIE_SHOP','FARMERS_MARKET','ICE_CREAM_SHOP','BRUNCH_SPOT','SMOOTHIE_SHOP','FARMERS_MARKET','ICE_CREAM_SHOP','BRUNCH_SPOT'],\n MIXED:['SMOOTHIE_SHOP','FARMERS_MARKET','ICE_CREAM_SHOP','YARN_STORE','FARMERS_MARKET','BAKERY','SMOOTHIE_SHOP','BAKERY'],\n};\nconst refs=['rancher_rita.py','broker_bea.py'];\nconst E=await loadOfficialKaggricultureEngine();\nconst baseCfg=E.state.resolveConfig({seed:515151,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!rr.ok)throw new Error(String(rr.status));const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));\nconst py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}\npy.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep(w,m,s):\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON','STRAWBERRY'];_D['POLICY']['crop_share']={'WHEAT':w,'MELON':m,'STRAWBERRY':s};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16,'STRAWBERRY':16};_D['POLICY']['sell_order']=['STRAWBERRY','MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\n`);\nconst cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}\nfunction addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction fixedShops(day:number,sched:string[]){const n=Math.max(0,Math.min(8,Math.floor(day/3)));return sched.slice(0,n)}\nasync function play(v:string,world:string,ref:string,seat:0|1){prep(...variants[v]);let s=E.state.initGameState(2,baseCfg,515151),tg:any={COW:0,SHEEP:0},comm=false,err=0,maxFlock=0,finalFlock=0,minCash=Infinity;const ord:any={our:{buy:{},sell:{},seed:{}},opp:{buy:{},sell:{},seed:{}}};const timeline:any[]=[];while(!s.done){s.town.unlocked_shops=fixedShops(+s.day,worlds[world]);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){timeline.push({day:+o.day,ourCash:+o.farms[pid].money,oppCash:+o.farms[1-pid].money,wheatP:+o.market.prices.WHEAT,melonP:+o.market.prices.MELON,strawP:+o.market.prices.STRAWBERRY,wheatI:+o.market.inventory.WHEAT,melonI:+o.market.inventory.MELON,strawI:+o.market.inventory.STRAWBERRY,shops:[...(o.town?.unlocked_shops??[])]});const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}const a=addFert(o,enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg));for(const q of a.market??[]){const [k,it,n]=q;if(k==='BUY_PRODUCT')ord.our.buy[it]=(ord.our.buy[it]??0)+Number(n??0);if(k==='SELL')ord.our.sell[it]=(ord.our.sell[it]??0)+Number(n??0);if(k==='BUY_SEED')ord.our.seed[it]=(ord.our.seed[it]??0)+Number(n??0)}aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else{const a=JSON.parse(String(cr(ref,JSON.stringify(o))));for(const q of a.market??[]){const [k,it,n]=q;if(k==='BUY_PRODUCT')ord.opp.buy[it]=(ord.opp.buy[it]??0)+Number(n??0);if(k==='SELL')ord.opp.sell[it]=(ord.opp.sell[it]??0)+Number(n??0);if(k==='BUY_SEED')ord.opp.seed[it]=(ord.opp.seed[it]??0)+Number(n??0)}aa[pid]=a}}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,baseCfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{v,world,ref,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,win:s.farms[seat].money>s.farms[1-seat].money,err,minCash,animalLoss:maxFlock-finalFlock,ord,timeline}}\nconst games:any[]=[];for(const world of Object.keys(worlds))for(const ref of refs)for(const v of Object.keys(variants)){games.push(await play(v,world,ref,0));games.push(await play(v,world,ref,1))}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const summary:any={};for(const world of Object.keys(worlds)){summary[world]={};for(const ref of refs){summary[world][ref]={};for(const v of Object.keys(variants)){const x=games.filter(g=>g.world===world&&g.ref===ref&&g.v===v);summary[world][ref][v]={wins:x.filter(g=>g.win).length,episodes:x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),sell:{melon:avg(x.map(g=>g.ord.our.sell.MELON??0)),straw:avg(x.map(g=>g.ord.our.sell.STRAWBERRY??0)),wheat:avg(x.map(g=>g.ord.our.sell.WHEAT??0))},buyWheat:avg(x.map(g=>g.ord.our.buy.WHEAT??0)),oppSell:{melon:avg(x.map(g=>g.ord.opp.sell.MELON??0)),straw:avg(x.map(g=>g.ord.opp.sell.STRAWBERRY??0)),wheat:avg(x.map(g=>g.ord.opp.sell.WHEAT??0))}}}}}\nconst causal:any=[];for(const world of Object.keys(worlds))for(const ref of refs){const base=games.filter(g=>g.world===world&&g.ref===ref&&g.v==='M60');for(const v of ['M70','S10W']){const x=games.filter(g=>g.world===world&&g.ref===ref&&g.v===v);causal.push({world,ref,v,bankDelta:avg(x.map((g,i)=>g.ours-base[i].ours)),oppDelta:avg(x.map((g,i)=>g.theirs-base[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-base[i].margin)),better:x.filter((g,i)=>g.margin>base[i].margin).length,worse:x.filter((g,i)=>g.margin[w,summary[w]['broker_bea.py']]))},null,2));",[],[1599],"e44842fa-fa97-46bb-8b6f-ac25e1677432","kag_first_commitment_persistent_v2.ts","57b8d1d0-aa50-11f1-b615-1607ee4eb77e",91,["D",1788739747088],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['BASE','COMMIT_COW2','COMMIT_SHEEP2','COMMIT_MIX11','WAIT3'];const seeds=[9403,9419,9431,9461,9491];const opp='RITA_BASE';\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();const zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await zr.arrayBuffer()));const src=strFromU8(z['rancher_rita.py']);const py=await loadPyodide();py.runPython(`import json,copy\\n_REG={}\\n_BASE={}`);py.globals.set('src',src);for(const [n,p] of Object.entries({DYN:{animal_target:{COW:0,SHEEP:0}},RITA_BASE:{}})){py.globals.set('nm',n);py.globals.set('patch',JSON.stringify(p));py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n\\n_BASE[nm]=copy.deepcopy(_n['POLICY'])`);}py.runPython(`\\ndef call(name,payload): return json.dumps(_REG[name]['agent'](json.loads(payload)))\\ndef set_herd(name,cows,sheep,order):\\n _REG[name]['POLICY']['animal_target']={'COW':int(cows),'SHEEP':int(sheep)}\\n _REG[name]['POLICY']['animals']=json.loads(order)\\ndef reset(name): _REG[name]['POLICY'].clear();_REG[name]['POLICY'].update(copy.deepcopy(_BASE[name]))\\n`);const call=py.globals.get('call'),setHerd=py.globals.get('set_herd'),reset=py.globals.get('reset');\nfunction owned(obs:any,a:string){const me=+obs.player,f=obs.farms[me];let n=+(obs.private?.shed?.[a]??0);for(const inv of obs.private?.inventories??[])n+=+(inv?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n;}\nfunction enforce(obs:any,act:any,t:any){const x:any={COW:owned(obs,'COW'),SHEEP:owned(obs,'SHEEP')};const o={...act,market:[] as any[]};for(const m of act?.market??[]){if(Array.isArray(m)&&m[0]==='BUY_ANIMAL'&&(m[1]==='COW'||m[1]==='SHEEP')){const a=m[1],q=Math.max(0,Math.min(+m[2]||1,t[a]-x[a]));if(q>0){o.market.push(['BUY_ANIMAL',a,q]);x[a]+=q;}}else o.market.push(m);}return o;}\nfunction initialGoal(v:string){if(v==='COMMIT_COW2')return{COW:2,SHEEP:0};if(v==='COMMIT_SHEEP2')return{COW:0,SHEEP:2};if(v==='COMMIT_MIX11')return{COW:1,SHEEP:1};return null;}\nasync function play(v:string,seed:number,seat:0|1){reset('DYN');reset(opp);let s=E.state.initGameState(2,{...cfg0,seed},seed),targets:any={COW:0,SHEEP:0},errors=0;const goal=initialGoal(v);let commitmentDone=!goal;const diary:any[]=[];while(!s.done){const acts:any[]=[];for(const pid of [0,1] as const){const n=pid===seat?'DYN':opp,obs=observationFor(s,pid);if(n==='DYN'&&+obs.hour===0){const own={COW:owned(obs,'COW'),SHEEP:owned(obs,'SHEEP')};targets={...own};if(goal&&!commitmentDone){commitmentDone=own.COW>=goal.COW&&own.SHEEP>=goal.SHEEP;if(!commitmentDone){targets={COW:Math.max(own.COW,goal.COW),SHEEP:Math.max(own.SHEEP,goal.SHEEP)};setHerd('DYN',targets.COW,targets.SHEEP,JSON.stringify(goal.SHEEP>0?['SHEEP','COW']:['COW','SHEEP']));}}\nif(commitmentDone){if(v==='WAIT3'&&+obs.day<3){setHerd('DYN',own.COW,own.SHEEP,JSON.stringify(['COW','SHEEP']));targets={...own};}else{const total=own.COW+own.SHEEP,c=choosePastureAnimal(obs,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)targets[c.choice]+=Math.min(2,16-total);setHerd('DYN',targets.COW,targets.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')));}}\nif([0,1,2,3,6,9,12,18].includes(+obs.day))diary.push({day:+obs.day,cash:obs.farms[pid].money,own:{...own},targets:{...targets},commitmentDone,shops:[...(obs.town?.unlocked_shops??[])]});}\ntry{acts[pid]=JSON.parse(String(call(n,JSON.stringify(obs))));if(n==='DYN')acts[pid]=enforce(obs,acts[pid],targets);}catch{errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]};}}s=E.interpreter.step(s,acts,{...cfg0,seed});}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,errors,diary};}\nconst games:any[]=[];for(const v of variants)for(const seed of seeds){games.push(await play(v,seed,0));games.push(await play(v,seed,1));}const summary:any={};for(const v of variants){const rs=games.filter(g=>g.v===v);summary[v]={episodes:rs.length,wins:rs.filter(r=>r.win).length,winRate:rs.filter(r=>r.win).length/rs.length,avgBank:rs.reduce((a,r)=>a+r.ours,0)/rs.length,avgMargin:rs.reduce((a,r)=>a+r.margin,0)/rs.length,bySeed:Object.fromEntries(seeds.map(seed=>{const x=rs.filter(r=>r.seed===seed);return[seed,{wins:x.filter(r=>r.win).length,avgMargin:x.reduce((a,r)=>a+r.margin,0)/x.length,avgBank:x.reduce((a,r)=>a+r.ours,0)/x.length}]}))};}const b=games.filter(g=>g.v==='BASE');const paired=variants.slice(1).map(v=>{const r=games.filter(g=>g.v===v);return{variant:v,avgMarginDelta:r.reduce((s,g,i)=>s+(g.margin-b[i].margin),0)/r.length,better:r.filter((g,i)=>g.margin>b[i].margin).length,worse:r.filter((g,i)=>g.marging.seed===9491&&g.seat===0).map(g=>({v:g.v,diary:g.diary}))},null,2));",[],[1607],"a3e623f8-18d9-45ea-a31c-4a569c8edf57","kag_financial_flexibility_probe_v1.ts","f2e2e918-aa4e-11f1-8789-1607ee4eb77e",85,["D",1788739148386],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['BASE400','ACCEL0','FLEX100'];const seeds=[9403,9419,9431,9461,9491],opponents=['C4S12','C6S10','RITA_BASE'];\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();const zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await zr.arrayBuffer()));const src=strFromU8(z['rancher_rita.py']);const py=await loadPyodide();py.runPython(`import json,copy\\n_REG={}\\n_BASE={}`);py.globals.set('src',src);const patches:any={DYN:{animal_target:{COW:0,SHEEP:0}},C4S12:{animal_target:{COW:4,SHEEP:12}},C6S10:{animal_target:{COW:6,SHEEP:10}},RITA_BASE:{}};for(const [n,p] of Object.entries(patches)){py.globals.set('nm',n);py.globals.set('patch',JSON.stringify(p));py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n\\n_BASE[nm]=copy.deepcopy(_n['POLICY'])`);}py.runPython(`\\ndef call(name,payload): return json.dumps(_REG[name]['agent'](json.loads(payload)))\\ndef set_herd(name,cows,sheep,order):\\n _REG[name]['POLICY']['animal_target']={'COW':int(cows),'SHEEP':int(sheep)}\\n _REG[name]['POLICY']['animals']=json.loads(order)\\ndef reset(name):\\n _REG[name]['POLICY'].clear();_REG[name]['POLICY'].update(copy.deepcopy(_BASE[name]))\\n`);const call=py.globals.get('call'),setHerd=py.globals.get('set_herd'),reset=py.globals.get('reset');\nfunction countOwned(obs:any,a:string){const me=+obs.player,f=obs.farms[me];let n=+(obs.private?.shed?.[a]??0);for(const inv of obs.private?.inventories??[])n+=+(inv?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n;}\nfunction enforce(obs:any,act:any,t:any){const x:any={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};const o={...act,market:[] as any[]};for(const m of act?.market??[]){if(Array.isArray(m)&&m[0]==='BUY_ANIMAL'&&(m[1]==='COW'||m[1]==='SHEEP')){const a=m[1],q=Math.max(0,Math.min(+m[2]||1,t[a]-x[a]));if(q>0){o.market.push(['BUY_ANIMAL',a,q]);x[a]+=q;}}else o.market.push(m);}return o;}\nfunction bufferFor(v:string,obs:any){if(v==='ACCEL0')return +obs.day<8?0:400;if(v==='FLEX100'){const unseen=Math.max(0,8-(obs.town?.unlocked_shops??[]).length);return 400+100*unseen;}return 400;}\nasync function play(v:string,opp:string,seed:number,seat:0|1){reset('DYN');reset(opp);let s=E.state.initGameState(2,{...cfg0,seed},seed),targets:any={COW:0,SHEEP:0},errors=0;while(!s.done){const acts:any[]=[];for(const pid of [0,1] as const){const n=pid===seat?'DYN':opp,obs=observationFor(s,pid);if(n==='DYN'&&+obs.hour===0){const owned={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};targets={...owned};const total=owned.COW+owned.SHEEP;const c=choosePastureAnimal(obs,{cashBuffer:bufferFor(v,obs),strategicWeight:0});if(total<16&&c.choice)targets[c.choice]+=Math.min(2,16-total);setHerd('DYN',targets.COW,targets.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')));}try{acts[pid]=JSON.parse(String(call(n,JSON.stringify(obs))));if(n==='DYN')acts[pid]=enforce(obs,acts[pid],targets);}catch{errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]};}}s=E.interpreter.step(s,acts,{...cfg0,seed});}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,opp,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,errors};}\nconst games:any[]=[];for(const v of variants)for(const o of opponents)for(const seed of seeds){games.push(await play(v,o,seed,0));games.push(await play(v,o,seed,1));}const summary:any={};for(const v of variants){const rs=games.filter(g=>g.v===v);summary[v]={episodes:rs.length,wins:rs.filter(r=>r.win).length,winRate:rs.filter(r=>r.win).length/rs.length,avgBank:rs.reduce((a,r)=>a+r.ours,0)/rs.length,avgMargin:rs.reduce((a,r)=>a+r.margin,0)/rs.length,byOpponent:{}};for(const o of opponents){const xs=rs.filter(r=>r.opp===o);summary[v].byOpponent[o]={wins:xs.filter(r=>r.win).length,winRate:xs.filter(r=>r.win).length/xs.length,avgBank:xs.reduce((a,r)=>a+r.ours,0)/xs.length,avgMargin:xs.reduce((a,r)=>a+r.margin,0)/xs.length};}}const b=games.filter(g=>g.v==='BASE400');const paired=variants.slice(1).map(v=>{const r=games.filter(g=>g.v===v);return{variant:v,avgMarginDelta:r.reduce((s,g,i)=>s+(g.margin-b[i].margin),0)/r.length,better:r.filter((g,i)=>g.margin>b[i].margin).length,worse:r.filter((g,i)=>g.margin0){o.market.push(['BUY_ANIMAL',a,q]);x[a]+=q;}}else o.market.push(m);}return o;}\nfunction bufferFor(v:string,obs:any){if(v==='ACCEL0')return +obs.day<8?0:400;if(v==='FLEX100'){const unseen=Math.max(0,8-(obs.town?.unlocked_shops??[]).length);return 400+100*unseen;}return 400;}\nasync function play(v:string,opp:string,seed:number,seat:0|1){reset('DYN');reset(opp);let s=E.state.initGameState(2,{...cfg0,seed},seed),targets:any={COW:0,SHEEP:0},errors=0;while(!s.done){const acts:any[]=[];for(const pid of [0,1] as const){const n=pid===seat?'DYN':opp,obs=observationFor(s,pid);if(n==='DYN'&&+obs.hour===0){const owned={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};targets={...owned};const total=owned.COW+owned.SHEEP;const c=choosePastureAnimal(obs,{cashBuffer:bufferFor(v,obs),strategicWeight:0});if(total<16&&c.choice)targets[c.choice]+=Math.min(2,16-total);setHerd('DYN',targets.COW,targets.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')));}try{acts[pid]=JSON.parse(String(call(n,JSON.stringify(obs))));if(n==='DYN')acts[pid]=enforce(obs,acts[pid],targets);}catch{errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]};}}s=E.interpreter.step(s,acts,{...cfg0,seed});}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,opp,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,errors};}\nconst games:any[]=[];for(const v of variants)for(const o of opponents)for(const seed of seeds){games.push(await play(v,o,seed,0));games.push(await play(v,o,seed,1));}const summary:any={};for(const v of variants){const rs=games.filter(g=>g.v===v);summary[v]={episodes:rs.length,wins:rs.filter(r=>r.win).length,winRate:rs.filter(r=>r.win).length/rs.length,avgBank:rs.reduce((a,r)=>a+r.ours,0)/rs.length,avgMargin:rs.reduce((a,r)=>a+r.margin,0)/rs.length,byOpponent:{}};for(const o of opponents){const xs=rs.filter(r=>r.opp===o);summary[v].byOpponent[o]={wins:xs.filter(r=>r.win).length,winRate:xs.filter(r=>r.win).length/xs.length,avgBank:xs.reduce((a,r)=>a+r.ours,0)/xs.length,avgMargin:xs.reduce((a,r)=>a+r.margin,0)/xs.length};}}const b=games.filter(g=>g.v==='BASE400');const paired=variants.slice(1).map(v=>{const r=games.filter(g=>g.v===v);return{variant:v,avgMarginDelta:r.reduce((s,g,i)=>s+(g.margin-b[i].margin),0)/r.length,better:r.filter((g,i)=>g.margin>b[i].margin).length,worse:r.filter((g,i)=>g.margin0&&p>=60&&(a.market?.length??0)<10)return{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]};return a}\nfunction collectOps(a:any){let n=0;for(const u of [a?.farmer,...(a?.hands??[])])if(Array.isArray(u)&&u[0]==='COLLECT_FERTILIZER')n++;return n}\nasync function play(v:string,ref:string,seed:number,seat:0|1){reset(v);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},done=false,err=0,collect=0,fertAttempt=0,priceTrail:number[]=[];while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!done){done=ow.COW>=1&&ow.SHEEP>=1;if(!done){tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)};seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}}if(done){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')))}}let a=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg);if(v==='COLLECT_SELL')a=addFertSell(o,a);collect+=collectOps(a);for(const q of a.market??[])if(q?.[0]==='SELL'&&q?.[1]==='FERTILIZER')fertAttempt+=Number(q?.[2]??0);aa[pid]=a;if(+o.hour===0)priceTrail.push(+(o.market?.prices?.FERTILIZER??0))}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,collect,fertAttempt,avgFertPrice:priceTrail.reduce((a,b)=>a+b,0)/(priceTrail.length||1),finalFertPrice:s.market.prices.FERTILIZER}}\nconst games:any[]=[];for(const v of variants)for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const summary:any={};for(const v of variants){const rs=games.filter(g=>g.v===v);summary[v]={episodes:rs.length,wins:rs.filter(g=>g.win).length,winRate:rs.filter(g=>g.win).length/rs.length,avgBank:avg(rs.map(g=>g.ours)),avgOppBank:avg(rs.map(g=>g.theirs)),avgMargin:avg(rs.map(g=>g.margin)),avgCollect:avg(rs.map(g=>g.collect)),avgFertAttempt:avg(rs.map(g=>g.fertAttempt)),avgFertPrice:avg(rs.map(g=>g.avgFertPrice)),avgFinalFertPrice:avg(rs.map(g=>g.finalFertPrice)),byOpponent:{}};for(const ref of refs){const x=rs.filter(g=>g.ref===ref);summary[v].byOpponent[ref]={wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOppBank:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),avgFertPrice:avg(x.map(g=>g.avgFertPrice)),avgFinalFertPrice:avg(x.map(g=>g.finalFertPrice))}}}const b=games.filter(g=>g.v==='BASE');const cmp=variants.slice(1).map(v=>{const x=games.filter(g=>g.v===v);return{v,ownBankDelta:avg(x.map((g,i)=>g.ours-b[i].ours)),oppBankDelta:avg(x.map((g,i)=>g.theirs-b[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-b[i].margin)),better:x.filter((g,i)=>g.margin>b[i].margin).length,worse:x.filter((g,i)=>g.margin0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(v:string,ref:string,seed:number,seat:0|1){reset(v);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},done=false,err=0;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!done){done=ow.COW>=1&&ow.SHEEP>=1;if(!done){tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)};seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}}if(done){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')))}}let a=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg);if(v==='FERT'||v==='FERT_DROP')a=sellF(o,a);aa[pid]=a}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err}}\nconst games:any[]=[];for(const v of variants)for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const summary:any={};for(const v of variants){const x=games.filter(g=>g.v===v);summary[v]={wins:x.filter(g=>g.win).length,episodes:x.length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOppBank:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);summary[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOppBank:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin))}}}const b=games.filter(g=>g.v==='BASE');const paired=variants.slice(1).map(v=>{const x=games.filter(g=>g.v===v);return{v,bankDelta:avg(x.map((g,i)=>g.ours-b[i].ours)),oppDelta:avg(x.map((g,i)=>g.theirs-b[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-b[i].margin)),better:x.filter((g,i)=>g.margin>b[i].margin).length,worse:x.filter((g,i)=>g.margina+Number(i?.WHEAT??0),0),cash=+o.farms[pid].money,wp=+o.market.prices.WHEAT;morning.push({day:+o.day,cash,flock,wheat,daysOnHand:flock?wheat/flock:null,wheatPrice:wp,unlocked:(o.farms[pid].unlocked_quadrants??[]).length});for(const k of [2,4,8,12,14,16])if(threshold[k]===undefined&&flock>=k)threshold[k]=+o.day;const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!committed){committed=ow.COW>=1&&ow.SHEEP>=1;if(!committed){tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)};seth(v,tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}}if(committed){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(v,tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}}let a=enforce(o,JSON.parse(String(cd(v,JSON.stringify(o)))),tg);aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const finalObs=observationFor(s,seat);finalFlock=own(finalObs,'COW')+own(finalObs,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,maxFlock,finalFlock,animalLoss:maxFlock-finalFlock,threshold,morning}}\nconst games:any[]=[];for(const v of variants)for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const med=(x:number[])=>{const y=[...x].sort((a,b)=>a-b);return y.length?(y[(y.length-1)>>1]+y[y.length>>1])/2:0};\nconst summary:any={};for(const v of variants){const rs=games.filter(g=>g.v===v);summary[v]={episodes:rs.length,wins:rs.filter(g=>g.win).length,winRate:rs.filter(g=>g.win).length/rs.length,avgBank:avg(rs.map(g=>g.ours)),avgOppBank:avg(rs.map(g=>g.theirs)),avgMargin:avg(rs.map(g=>g.margin)),medianMargin:med(rs.map(g=>g.margin)),minCash:Math.min(...rs.map(g=>g.minCash)),avgAnimalLoss:avg(rs.map(g=>g.animalLoss)),reach8Day:avg(rs.filter(g=>g.threshold[8]!==undefined).map(g=>g.threshold[8])),reach14Day:avg(rs.filter(g=>g.threshold[14]!==undefined).map(g=>g.threshold[14])),reach14Share:rs.filter(g=>g.threshold[14]!==undefined).length/rs.length,byOpponent:{}};for(const ref of refs){const x=rs.filter(g=>g.ref===ref);summary[v].byOpponent[ref]={wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOppBank:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss))}}}\nconst base=games.filter(g=>g.v==='BASE16');const paired=variants.slice(1).map(v=>{const x=games.filter(g=>g.v===v);return{v,bankDelta:avg(x.map((g,i)=>g.ours-base[i].ours)),oppDelta:avg(x.map((g,i)=>g.theirs-base[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-base[i].margin)),better:x.filter((g,i)=>g.margin>base[i].margin).length,worse:x.filter((g,i)=>g.marging.win&&!base[i].win).length,lossFlips:x.filter((g,i)=>!g.win&&base[i].win).length}});\nconst one=games.find(g=>g.v==='BASE16'&&g.ref==='broker_bea.py'&&g.seed===23003&&g.seat===0),credit=games.find(g=>g.v==='STOCK_CREDIT16'&&g.ref==='broker_bea.py'&&g.seed===23003&&g.seat===0);\nconsole.log(JSON.stringify({status:'PASS',question:'IS_16_DAY_FEED_FLOAT_OVERCONSERVATIVE_BECAUSE_IT_IGNORES_FEED_ALREADY_OWNED_OR_SEASON_HORIZON',mutationProvenance:'independent local counterfactuals on MIT-authored Rita feed-float formula; high-tier TRACE not reused',summary,paired,example:{base:one?.morning.filter((x:any)=>[1,2,3,4,6,8,10,12,16,20].includes(x.day)),stockCredit:credit?.morning.filter((x:any)=>[1,2,3,4,6,8,10,12,16,20].includes(x.day))}},null,2));",[],[1647],"3dcb0a6e-d4b4-4261-ad97-06f292f1e971","kag_features_v31_semantics_test.ts","dbe2a19e-aa86-11f1-9710-1607ee4eb77e",412,["D",1788763161638],"import { extractTeacherFeaturesV31,KAG_TEACHER_FEATURE_SCHEMA_V31 } from './lib/kag_teacher_features_v31.ts';\nconst empty=()=>Array.from({length:10},()=>Array.from({length:10},()=>({})));const a=empty(),b=empty();a[0][0]={kind:'PLANT',crop:'STRAWBERRY',planted_day:7,watered_today:false,consecutive_unwatered:1,yield_units:0,max_lifespan_step:-1,fertilized_until_day:-1};a[0][1]={kind:'PLANT',crop:'WHEAT',planted_day:5,watered_today:true,consecutive_unwatered:0,yield_units:3,max_lifespan_step:240,fertilized_until_day:9};const products=['WHEAT','CARROT','TOMATO','STRAWBERRY','MELON','EGG','MILK','WOOL','FERTILIZER'],prices:any={},inventory:any={},shed:any={};for(const p of products){prices[p]=10;inventory[p]=10000;shed[p]=0}const o:any={player:0,step:216,day:9,hour:0,farms:[{money:1000,tiles:a,farmer:[4,4],hands:[],unlocked_quadrants:[0],hires_today:0},{money:1000,tiles:b,farmer:[4,4],hands:[],unlocked_quadrants:[0],hires_today:0}],private:{shed,seeds:{WHEAT:0,CARROT:0,TOMATO:0,STRAWBERRY:0,MELON:0},inventories:[{}]},market:{prices,inventory},town:{unlocked_shops:[]}};const f=extractTeacherFeaturesV31(o,null),s=f.me.crops.STRAWBERRY,w=f.me.crops.WHEAT;const ok=f.schema===KAG_TEACHER_FEATURE_SCHEMA_V31&&s.finiteLifespanCount===0&&s.minLifeRemaining===-1&&w.finiteLifespanCount===1&&w.minLifeRemaining===24&&w.fertilized===1;console.log(JSON.stringify({status:ok?'PASS':'FAIL',schema:f.schema,strawberry:s,wheat:w},null,2));",[],[1655],"c2d49c53-0231-43e7-9e65-e0b27aa57986","kag_feature_negative_space_probe_v1.ts","a4978e58-aa85-11f1-bf00-1607ee4eb77e",390,["D",1788762639377],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nconst E=await loadOfficialKaggricultureEngine();const cfg=E.state.resolveConfig({seed:445566,weedSpawnChance:0});const rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`import json\\n_ns={}\\nexec(src,_ns)\\n_A=_ns['agent']\\ndef call(p):return json.dumps(_A(json.loads(p)))`);const call=py.globals.get('call');\nlet s=E.state.initGameState(2,cfg,445566),target:any=null,prev:any=null;while(!s.done){const o=observationFor(s,0);if(+o.day===9&&+o.hour===0){target={o,prev};break}prev=o;const a=JSON.parse(String(call(JSON.stringify(o))));s=E.interpreter.step(s,[a,{farmer:[],hands:[],market:[]}],cfg)}if(!target)throw new Error('TARGET_NOT_REACHED');const o=target.o,p=target.prev;function tileCensus(f:any){const seen=new Map();for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object'||Object.keys(t).length===0)continue;const sig=Object.keys(t).sort().join(',');if(!seen.has(sig))seen.set(sig,t)}return[...seen.entries()].map(([keys,sample])=>({keys,sample}))}function invTotals(inv:any[]){const z:any={};for(const bag of inv??[])for(const [k,v] of Object.entries(bag??{}))z[k]=(z[k]??0)+(+v||0);return z}console.log(JSON.stringify({status:'PASS',evidenceState:'FEATURE_NEGATIVE_SPACE_D9H0',current:{day:o.day,hour:o.hour,farmer:o.farms[0].farmer,hands:o.farms[0].hands,hires:o.farms[0].hires_today,seeds:o.private.seeds,inventoryBags:o.private.inventories,inventoryTotals:invTotals(o.private.inventories),tileCensus:tileCensus(o.farms[0])},previous:{day:p?.day,hour:p?.hour,farmer:p?.farms?.[0]?.farmer,handsLen:(p?.farms?.[0]?.hands??[]).length,handPositions:p?.farms?.[0]?.hands,hires:p?.farms?.[0]?.hires_today,inventoryTotals:invTotals(p?.private?.inventories),tileCensus:tileCensus(p?.farms?.[0])}},null,2));",[],[1663],"363b3a36-df85-40d9-b3d5-d28e5f35e938","kag_farm_schema_probe_v1.ts","285e7e32-aa85-11f1-8b64-1607ee4eb77e",382,["D",1788762430965],"import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';const E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:123});let s=E.state.initGameState(2,cfg,123);const o=observationFor(s,0),f=o.farms[0];console.log(JSON.stringify({farmKeys:Object.keys(f),farm:f,privateKeys:Object.keys(o.private??{}),obsKeys:Object.keys(o)},null,2));",[],[1671],"b10283d2-ffe0-43b1-8799-ea53172387f9","kag_exogenous_world_crn_wheat_v1.ts","80771fd0-aa66-11f1-93a4-1607ee4eb77e",183,["D",1788749264371],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';import { lowerTailCVaR } from './lib/kaggriculture_farm_economics.ts';\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:818181,weedSpawnChance:0,townShopUnlockInterval:9999});const SHOP_NAMES=[...E.constants.SHOP_NAMES];function mulberry32(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const worlds:any[]=[];for(let i=0;i<12;i++){const rr=mulberry32(900000+i*7919);worlds.push({id:`W${String(i).padStart(2,'0')}`,shops:Array.from({length:8},()=>SHOP_NAMES[Math.floor(rr()*SHOP_NAMES.length)])})}\nconst variants:any={BASE:{buy:0,hold:false},HOLD6:{buy:0,hold:true},BUY8:{buy:8,hold:false},BUY8_HOLD6:{buy:8,hold:true}};\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['broker_bea.py']));py.runPython(`import json\\n_r={}\\nexec(src,_r)\\n_REF=_r['agent']`);py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(p):return json.dumps(_REF(json.loads(p)))\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number,s:string[]){return s.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(v:string,w:any,seat:0|1){prep();let s=E.state.initGameState(2,cfg,818181),tg:any={COW:0,SHEEP:0},comm=false,minCash=Infinity,maxFlock=0,finalFlock=0,forced=false,forcedDay=-1,forcedPrice=0;while(!s.done){s.town.unlocked_shops=fixed(+s.day,w.shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}let a=addFert(o,enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg));const spec=variants[v];if(spec.hold&&+o.day<=6)a={...a,market:(a.market??[]).filter((x:any)=>!(x?.[0]==='SELL'&&x?.[1]==='WHEAT'))};if(spec.buy>0&&!forced&&+o.day>=2&&+o.day<=4&&+o.market.prices.WHEAT<=35&&(a.market?.length??0)<10&&+o.farms[pid].money>=spec.buy*(+o.market.prices.WHEAT)+800){a={...a,market:[['BUY_PRODUCT','WHEAT',spec.buy],...(a.market??[])]};forced=true;forcedDay=+o.day;forcedPrice=+o.market.prices.WHEAT}aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money);maxFlock=Math.max(maxFlock,own(o,'COW')+own(o,'SHEEP'))}else aa[pid]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{v,world:w.id,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,win:s.farms[seat].money>s.farms[1-seat].money,minCash,animalLoss:maxFlock-finalFlock,forced,forcedDay,forcedPrice}}\nconst games:any[]=[];for(const w of worlds)for(const v of Object.keys(variants)){games.push(await play(v,w,0));games.push(await play(v,w,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),q=(xs:number[],p:number)=>{const a=[...xs].sort((x,y)=>x-y);return a[Math.min(a.length-1,Math.max(0,Math.floor((a.length-1)*p)))]};const base=games.filter(g=>g.v==='BASE'),summary:any={},paired:any[]=[];for(const v of Object.keys(variants)){const x=games.filter(g=>g.v===v),m=x.map(g=>g.margin);summary[v]={episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(m),p10Margin:q(m,.1),medianMargin:q(m,.5),p90Margin:q(m,.9),cvar10:lowerTailCVaR(m,.1),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),forcedRate:x.filter(g=>g.forced).length/x.length};if(v!=='BASE'){const d=x.map((g,i)=>g.margin-base[i].margin);paired.push({v,avgMarginDelta:avg(d),medianDelta:q(d,.5),p10Delta:q(d,.1),p90Delta:q(d,.9),better:d.filter(z=>z>0).length,worse:d.filter(z=>z<0).length,tied:d.filter(z=>z===0).length,winFlips:x.filter((g,i)=>g.win&&!base[i].win).length,lossFlips:x.filter((g,i)=>!g.win&&base[i].win).length,byWorld:worlds.map(w=>{const y=x.filter(g=>g.world===w.id),b=base.filter(g=>g.world===w.id);return{world:w.id,shops:w.shops,delta:avg(y.map((g,i)=>g.margin-b[i].margin))}})})}}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'EXOGENOUS_COMMON_RANDOM_WORLDS_POLICY_INVARIANT',worldCount:worlds.length,shopUniverse:SHOP_NAMES,summary,paired},null,2));",[],[1679],"186140ef-af46-42a3-b70e-71e4b489790e","kag_engine_market_introspection_v1.ts","000b3f44-aa60-11f1-b675-1607ee4eb77e",164,["D",1788747062667],"import { loadOfficialKaggricultureEngine } from './lib/kaggriculture_engine_loader.ts';\nconst E=await loadOfficialKaggricultureEngine();\nconst mp=E.constants.MARKET_PARAMS?.MELON; const samples=[10000,10050,10100,10150,10158,10200].map(inv=>({inv,price:E.market.marketPrice('MELON',inv)})); const melonShops=Object.entries(E.constants.SHOPS??{}).filter(([_,v]:any)=>JSON.stringify(v).includes('MELON')); console.log(JSON.stringify({marketKeys:Object.keys(E.market??{}),constantKeys:Object.keys(E.constants??{}),config:E.state.resolveConfig(),melonParams:mp,melonSamples:samples,melonShops,townCenterProducts:E.constants.TOWN_CENTER_PRODUCTS},null,2));",[],[1687],"81a15bd4-132f-4da6-bf56-7347f47a0fc1","kag_dynport_screen_v2.ts","01a078d6-c2c7-70c3-9b36-c48e42613635",78,["D",1788733750649],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst seeds=[9403,9419,9431,9461,9491];\nconst opponents=['C4S12','C6S10','RITA_BASE'];\nconst patches:Record={\n DYN_PORT:{animal_target:{COW:0,SHEEP:0}},\n C4S12:{animal_target:{COW:4,SHEEP:12}},\n C6S10:{animal_target:{COW:6,SHEEP:10}},\n RITA_BASE:{},\n};\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();\nconst zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!zr.ok)throw new Error(String(zr.status));\nconst z=unzipSync(new Uint8Array(await zr.arrayBuffer()));const src=strFromU8(z['rancher_rita.py']);\nconst py=await loadPyodide();py.runPython(`import json,copy\\n_REG={}\\n_BASE={}`);py.globals.set('src',src);\nfor(const [name,patch] of Object.entries(patches)){py.globals.set('nm',name);py.globals.set('patch',JSON.stringify(patch));py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n\\n_BASE[nm]=copy.deepcopy(_n['POLICY'])`);}\npy.runPython(`\ndef call(name,payload):\n return json.dumps(_REG[name]['agent'](json.loads(payload)))\ndef set_herd(name,cows,sheep,order):\n _REG[name]['POLICY']['animal_target']={'COW':int(cows),'SHEEP':int(sheep)}\n _REG[name]['POLICY']['animals']=json.loads(order)\ndef reset(name):\n _REG[name]['POLICY'].clear();_REG[name]['POLICY'].update(copy.deepcopy(_BASE[name]))\n`);\nconst call=py.globals.get('call'),setHerd=py.globals.get('set_herd'),reset=py.globals.get('reset');\n\nfunction countOwned(obs:any,animal:string){const me=Number(obs?.player??0),f=obs?.farms?.[me]??{};let n=Number(obs?.private?.shed?.[animal]??0);for(const inv of obs?.private?.inventories??[])n+=Number(inv?.[animal]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===animal)n++;return n;}\nfunction enforceAnimalTargets(obs:any,action:any,targets:{COW:number;SHEEP:number}){const actual:any={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};const out={...action,market:[] as any[]};for(const order of action?.market??[]){if(Array.isArray(order)&&order[0]==='BUY_ANIMAL'&&(order[1]==='COW'||order[1]==='SHEEP')){const a=order[1] as 'COW'|'SHEEP';const allowed=Math.max(0,Math.min(Number(order[2]??1),targets[a]-actual[a]));if(allowed>0){out.market.push(['BUY_ANIMAL',a,allowed]);actual[a]+=allowed;}}else out.market.push(order);}return out;}\nasync function play(a:string,b:string,seed:number){\n reset(a);reset(b);let s=E.state.initGameState(2,{...cfg0,seed},seed);let errors=0;let targets={COW:0,SHEEP:0};const decisions:any[]=[];\n while(!s.done){const acts:any[]=[];\n for(const [pid,n] of [[0,a],[1,b]] as any){const obs=observationFor(s,pid);\n if(n==='DYN_PORT'&&Number(obs.hour)===0){\n const owned={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};\n // An unexecuted target is a revocable option, not committed capital. Rebase intent on actual scheduler-visible ownership every macro cycle.\n targets={COW:owned.COW,SHEEP:owned.SHEEP};\n const actualTotal=owned.COW+owned.SHEEP;const choice=choosePastureAnimal(obs,{cashBuffer:400,strategicWeight:1.0});\n if(actualTotal<16 && choice.choice){const room=16-actualTotal,batch=Math.min(2,room);targets[choice.choice]+=batch;}\n const order=[...choice.ranking].map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP');\n setHerd('DYN_PORT',targets.COW,targets.SHEEP,JSON.stringify(order));\n decisions.push({day:obs.day,player:pid,money:obs.farms[pid].money,owned,targets:{...targets},choice:choice.choice,ranking:choice.ranking.map(x=>({animal:x.animal,net:Math.round(x.net),roi:Number(x.roi.toFixed(2)),currentPrice:x.currentPrice,projectedPrice:x.projectedPrice,townDemand:Number(x.townDemand.toFixed(1)),ownSupply:x.ownCommittedSupply,oppSupply:x.opponentSupply,liquidityPenalty:x.liquidityPenalty}))});\n }\n try{acts[pid]=JSON.parse(String(call(n,JSON.stringify(obs))));if(n==='DYN_PORT')acts[pid]=enforceAnimalTargets(obs,acts[pid],targets);}catch{errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]};}\n }\n s=E.interpreter.step(s,acts,{...cfg0,seed});\n }\n return{a,b,seed,bankA:s.farms[0].money,bankB:s.farms[1].money,errors,decisions};\n}\nconst games:any[]=[];for(const opp of opponents)for(const seed of seeds){games.push(await play('DYN_PORT',opp,seed));games.push(await play(opp,'DYN_PORT',seed));}\nconst summary:any={};for(const opp of opponents){const r=games.filter(g=>g.a===opp||g.b===opp).map(g=>g.a==='DYN_PORT'?{ours:g.bankA,opp:g.bankB,e:g.errors}:{ours:g.bankB,opp:g.bankA,e:g.errors});summary[opp]={episodes:r.length,wins:r.filter(x=>x.ours>x.opp).length,losses:r.filter(x=>x.oursx.ours===x.opp).length,avgBank:r.reduce((s,x)=>s+x.ours,0)/r.length,avgMargin:r.reduce((s,x)=>s+x.ours-x.opp,0)/r.length,errors:r.reduce((s,x)=>s+x.e,0)};}\nconst decisionDigest=games.filter(g=>g.a==='DYN_PORT'&&g.seed===seeds[0]).map(g=>({opponent:g.b,seed:g.seed,first8:g.decisions.slice(0,8),finalDecision:g.decisions[g.decisions.length-1]}));\nconsole.log(JSON.stringify({status:'PASS',agent:'DYN_PORT',mechanism:'GREEDY_MARGINAL_LIVESTOCK_PORTFOLIO_WITH_EVENT_DAILY_REALLOCATION',seeds,summary,decisionDigest,caveat:'Research prototype. Uses an NPV-like heuristic, not a proven optimal policy; requires ablation and broader league evaluation.'},null,2));",[],[1695],"4f472327-6146-4ba4-bebd-26acc1823c9e","kag_dynport_diag_v3.ts","01a078d7-bb34-770f-a990-c7be4a90e8d6",80,["D",1788733808160],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst seeds=[9403,9419,9431,9461,9491];\nconst opponents=['C4S12','C6S10','RITA_BASE'];\nconst patches:Record={\n DYN_PORT:{animal_target:{COW:0,SHEEP:0}},\n C4S12:{animal_target:{COW:4,SHEEP:12}},\n C6S10:{animal_target:{COW:6,SHEEP:10}},\n RITA_BASE:{},\n};\nconst E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();\nconst zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!zr.ok)throw new Error(String(zr.status));\nconst z=unzipSync(new Uint8Array(await zr.arrayBuffer()));const src=strFromU8(z['rancher_rita.py']);\nconst py=await loadPyodide();py.runPython(`import json,copy\\n_REG={}\\n_BASE={}`);py.globals.set('src',src);\nfor(const [name,patch] of Object.entries(patches)){py.globals.set('nm',name);py.globals.set('patch',JSON.stringify(patch));py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n\\n_BASE[nm]=copy.deepcopy(_n['POLICY'])`);}\npy.runPython(`\ndef call(name,payload):\n return json.dumps(_REG[name]['agent'](json.loads(payload)))\ndef set_herd(name,cows,sheep,order):\n _REG[name]['POLICY']['animal_target']={'COW':int(cows),'SHEEP':int(sheep)}\n _REG[name]['POLICY']['animals']=json.loads(order)\ndef reset(name):\n _REG[name]['POLICY'].clear();_REG[name]['POLICY'].update(copy.deepcopy(_BASE[name]))\n`);\nconst call=py.globals.get('call'),setHerd=py.globals.get('set_herd'),reset=py.globals.get('reset');\n\nfunction countOwned(obs:any,animal:string){const me=Number(obs?.player??0),f=obs?.farms?.[me]??{};let n=Number(obs?.private?.shed?.[animal]??0);for(const inv of obs?.private?.inventories??[])n+=Number(inv?.[animal]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===animal)n++;return n;}\nfunction enforceAnimalTargets(obs:any,action:any,targets:{COW:number;SHEEP:number}){const actual:any={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};const out={...action,market:[] as any[]};for(const order of action?.market??[]){if(Array.isArray(order)&&order[0]==='BUY_ANIMAL'&&(order[1]==='COW'||order[1]==='SHEEP')){const a=order[1] as 'COW'|'SHEEP';const allowed=Math.max(0,Math.min(Number(order[2]??1),targets[a]-actual[a]));if(allowed>0){out.market.push(['BUY_ANIMAL',a,allowed]);actual[a]+=allowed;}}else out.market.push(order);}return out;}\nasync function play(a:string,b:string,seed:number){\n reset(a);reset(b);let s=E.state.initGameState(2,{...cfg0,seed},seed);let errors=0;let targets={COW:0,SHEEP:0};const decisions:any[]=[];\n while(!s.done){const acts:any[]=[];\n for(const [pid,n] of [[0,a],[1,b]] as any){const obs=observationFor(s,pid);\n if(n==='DYN_PORT'&&Number(obs.hour)===0){\n const owned={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};\n // An unexecuted target is a revocable option, not committed capital. Rebase intent on actual scheduler-visible ownership every macro cycle.\n targets={COW:owned.COW,SHEEP:owned.SHEEP};\n const actualTotal=owned.COW+owned.SHEEP;const choice=choosePastureAnimal(obs,{cashBuffer:400,strategicWeight:1.0});\n if(actualTotal<16 && choice.choice){const room=16-actualTotal,batch=Math.min(2,room);targets[choice.choice]+=batch;}\n const order=[...choice.ranking].map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP');\n setHerd('DYN_PORT',targets.COW,targets.SHEEP,JSON.stringify(order));\n decisions.push({day:obs.day,player:pid,money:obs.farms[pid].money,owned,targets:{...targets},choice:choice.choice,ranking:choice.ranking.map(x=>({animal:x.animal,net:Math.round(x.net),roi:Number(x.roi.toFixed(2)),currentPrice:x.currentPrice,projectedPrice:x.projectedPrice,townDemand:Number(x.townDemand.toFixed(1)),ownSupply:x.ownCommittedSupply,oppSupply:x.opponentSupply,liquidityPenalty:x.liquidityPenalty}))});\n }\n try{acts[pid]=JSON.parse(String(call(n,JSON.stringify(obs))));if(n==='DYN_PORT')acts[pid]=enforceAnimalTargets(obs,acts[pid],targets);}catch{errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]};}\n }\n s=E.interpreter.step(s,acts,{...cfg0,seed});\n }\n return{a,b,seed,bankA:s.farms[0].money,bankB:s.farms[1].money,errors,decisions};\n}\nconst games:any[]=[];for(const opp of opponents)for(const seed of seeds){games.push(await play('DYN_PORT',opp,seed));games.push(await play(opp,'DYN_PORT',seed));}\nconst summary:any={};for(const opp of opponents){const r=games.filter(g=>g.a===opp||g.b===opp).map(g=>g.a==='DYN_PORT'?{ours:g.bankA,opp:g.bankB,e:g.errors}:{ours:g.bankB,opp:g.bankA,e:g.errors});summary[opp]={episodes:r.length,wins:r.filter(x=>x.ours>x.opp).length,losses:r.filter(x=>x.oursx.ours===x.opp).length,avgBank:r.reduce((s,x)=>s+x.ours,0)/r.length,avgMargin:r.reduce((s,x)=>s+x.ours-x.opp,0)/r.length,errors:r.reduce((s,x)=>s+x.e,0)};}\nconst decisionDigest=games.filter(g=>g.a==='DYN_PORT'&&g.seed===seeds[0]).map(g=>({opponent:g.b,seed:g.seed,first8:g.decisions.slice(0,8),finalDecision:g.decisions[g.decisions.length-1]}));\nconst gameResults=games.map(g=>({seed:g.seed,seat:g.a==='DYN_PORT'?'A':'B',opponent:g.a==='DYN_PORT'?g.b:g.a,ours:g.a==='DYN_PORT'?g.bankA:g.bankB,theirs:g.a==='DYN_PORT'?g.bankB:g.bankA,margin:(g.a==='DYN_PORT'?g.bankA-g.bankB:g.bankB-g.bankA),result:(g.a==='DYN_PORT'?g.bankA:g.bankB)>(g.a==='DYN_PORT'?g.bankB:g.bankA)?'WIN':(g.a==='DYN_PORT'?g.bankA:g.bankB)<(g.a==='DYN_PORT'?g.bankB:g.bankA)?'LOSS':'TIE'}));\nconsole.log(JSON.stringify({status:'PASS',agent:'DYN_PORT',mechanism:'GREEDY_MARGINAL_LIVESTOCK_PORTFOLIO_WITH_EVENT_DAILY_REALLOCATION',seeds,summary,gameResults,decisionDigest,caveat:'Research prototype. Uses an NPV-like heuristic, not a proven optimal policy; requires ablation and broader league evaluation.'},null,2));",[],[1703],"2c476cf7-641f-49e6-8a1e-1f6312f100c6","kag_dossier_status_v1.ts","af135b7a-aa56-11f1-b9ab-1607ee4eb77e",121,["D",1788742470557],"import { sqlite } from 'https://esm.town/v/std/sqlite/main.ts';\nconst counts=await sqlite.execute(`SELECT subject_id,subject_role,COUNT(*) n,ROUND(AVG(margin),1) avg_margin FROM kag_battle_dossiers GROUP BY subject_id,subject_role ORDER BY subject_id,subject_role`);const sample=await sqlite.execute(`SELECT dossier_id,subject_id,subject_role,seed,seat,final_bank,opponent_bank,margin,observed_strengths,observed_failures,causal_hypotheses,counterfactuals,evidence_state FROM kag_battle_dossiers ORDER BY ABS(margin) DESC LIMIT 4`);console.log(JSON.stringify({status:'PASS',counts:counts.rows,sample:sample.rows},null,2));",[],[1711],"c07bd5f7-ecd7-4d8e-bc04-f5ebcd1880a5","kag_crop_lifespan_semantics_probe_v1.ts","89976f3c-aa86-11f1-b9ee-1607ee4eb77e",404,["D",1788763023576],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:997979,weedSpawnChance:0});const rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();function load(name:string,key:string){py.globals.set('src',strFromU8(z[name]));py.runPython(`${key}={}\\nexec(src,${key})`)}load('rancher_rita.py','_a');load('ledger_lena.py','_b');py.runPython(`import json\\ndef ca(p):return json.dumps(_a['agent'](json.loads(p)))\\ndef cb(p):return json.dumps(_b['agent'](json.loads(p)))`);const ca=py.globals.get('ca'),cb=py.globals.get('cb');let s=E.state.initGameState(2,cfg,997979),o:any=null;while(!s.done){o=observationFor(s,0);if(+o.day===9&&+o.hour===0)break;const a=JSON.parse(String(ca(JSON.stringify(o)))),ob=observationFor(s,1),b=JSON.parse(String(cb(JSON.stringify(ob))));s=E.interpreter.step(s,[a,b],cfg)}const tiles:any[]=[];for(let r=0;r<(o.farms[1].tiles??[]).length;r++)for(let c=0;c<(o.farms[1].tiles[r]??[]).length;c++){const t=o.farms[1].tiles[r][c];if(t?.kind==='PLANT'&&t?.crop==='STRAWBERRY')tiles.push({r,c,...t,computedAbsRemaining:(+t.max_lifespan_step)-(+o.step)})}console.log(JSON.stringify({status:'PASS',day:o.day,hour:o.hour,step:o.step,count:tiles.length,maxLifespanValues:[...new Set(tiles.map(t=>t.max_lifespan_step))],plantedDays:[...new Set(tiles.map(t=>t.planted_day))],tiles},null,2));",[],[1719],"625755de-f24b-4b33-bccc-16b934b65f28","kag_craf_voc_falsepositive_v1.ts","01a07a26-85ad-71be-9775-1163ccc9111d",267,["D",1788755765802],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES];\nconst cfg=E.state.resolveConfig({seed:949494,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst decisionDay=9,modes=['N','H','W','B'],MAXN=16;\nconst actual=['PET_CAFE','FARMERS_MARKET','SMOOTHIE_SHOP','BAKERY','PIZZA_SHOP','SMOOTHIE_SHOP','PET_CAFE','PIZZA_SHOP'];\nconst prefix=actual.slice(0,3);\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}\nconst scenarios:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction compact(o:any){const me=+o.player,f=o.farms[me],g=o.farms[1-me];return{day:+o.day,cash:+f.money,oppCash:+g.money,shops:o.town.unlocked_shops}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,949494),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=compact(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst vals:any={};const preSet=new Set();for(const m of modes){vals[m]=[];for(const sc of scenarios){const r=await play(sc,m);vals[m].push(r.margin);preSet.add(JSON.stringify(r.pre))}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},sizes=[2,4,6,8,12,16];\nconst stages:any[]=[];for(const n of sizes){const stats:any={};for(const m of modes){const a=vals[m].slice(0,n);stats[m]={mean:avg(a)}}for(const m of modes.filter(m=>m!=='N')){const d=vals[m].slice(0,n).map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=sd(d)/Math.sqrt(n);stats[m].deltaVsN={mean:mu,se,z:se?mu/se:null,lo95:mu-1.96*se,hi95:mu+1.96*se}}const pick=[...modes].sort((a,b)=>stats[b].mean-stats[a].mean)[0],ordered=[...modes].sort((a,b)=>stats[b].mean-stats[a].mean),gap=stats[ordered[0]].mean-stats[ordered[1]].mean;stages.push({n,pick,gap,stats})}\nconst actualRuns:any={};for(const m of modes)actualRuns[m]=await play(actual,m);const oracle=[...modes].sort((a,b)=>actualRuns[b].margin-actualRuns[a].margin)[0],base=actualRuns.N.margin;\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_VALUE_OF_COMPUTATION_FALSE_POSITIVE_AUDIT',prefix,hiddenActualSuffix:actual.slice(3),preStateIdentical:preSet.size===1,scenarioCount:MAXN,stages,actual:{oracle,deltas:Object.fromEntries(modes.map(m=>[m,actualRuns[m].margin-base]))},interpretation:'paired scenario estimates; continue compute when top-action advantage remains uncertain relative to paired standard error'},null,2));",[],[1727],"055a5518-e62e-47a6-8981-cce866019761","kag_craf_voc_conditional_outer_validation_v1.ts","01a07a29-ab32-716c-95f6-f681f3998041",269,["D",1788755963984],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES];\nconst cfg=E.state.resolveConfig({seed:949494,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst decisionDay=9,modes=['N','H','W','B'],MAXN=16;\nconst actual=['PET_CAFE','FARMERS_MARKET','SMOOTHIE_SHOP','BAKERY','PIZZA_SHOP','SMOOTHIE_SHOP','PET_CAFE','PIZZA_SHOP'];\nconst prefix=actual.slice(0,3);\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}\nconst scenarios:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction compact(o:any){const me=+o.player,f=o.farms[me],g=o.farms[1-me];return{day:+o.day,cash:+f.money,oppCash:+g.money,shops:o.town.unlocked_shops}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,949494),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=compact(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money}}\nconst OUTER=8;const outerWorlds:string[][]=[];for(let i=0;ix.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))};const rows:any[]=[];for(let i=0;i();for(const m of modes){runs[m]=await play(w,m);preSet.add(JSON.stringify(runs[m].pre))}const base=runs.N.margin,oracle=[...modes].sort((a,b)=>runs[b].margin-runs[a].margin)[0];rows.push({i,suffix:w.slice(prefix.length),preStateIdentical:preSet.size===1,oracle,deltas:Object.fromEntries(modes.map(m=>[m,runs[m].margin-base]))})}\nconst summary:any={};for(const m of modes){const d=rows.map(r=>r.deltas[m] as number),mu=avg(d),se=sd(d)/Math.sqrt(d.length);summary[m]={meanDelta:mu,se,lo95:mu-1.96*se,hi95:mu+1.96*se,better:d.filter(x=>x>0).length,worse:d.filter(x=>x<0).length,min:Math.min(...d),max:Math.max(...d)}}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_VOC_CONDITIONAL_OUTER_VALIDATION',prefix,outerWorlds:OUTER,summary,rows,interpretation:'independent outer suffixes conditioned on the same observed prefix; tests whether W/B positive shadow expectation was genuine conditional value or inner-scenario bias'},null,2));",[],[1735],"02488887-2f0f-4b79-9f75-9866a1b29d72","kag_craf_shadow_lockin_v1.ts","01a07a1c-3acf-7373-8082-01e167af54af",248,["D",1788755105681],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst E=await loadOfficialKaggricultureEngine();\nconst cfg=E.state.resolveConfig({seed:989898,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst world={id:'HV1',shops:['BRUNCH_SPOT','ICE_CREAM_SHOP','BAKERY','PIZZA_SHOP','BRUNCH_SPOT','PIZZA_SHOP','PIZZA_SHOP','BAKERY']};\nconst refs=['rancher_rita.py','broker_bea.py','closer_cleo.py'];\nconst switchDays=[6,9,12];\nconst modes=['N','H','W','B']; // N=keep J8, H=JIT+9 hands, W=widen pasture to16, B=widen+9\n\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');\nconst z=unzipSync(new Uint8Array(await rr.arrayBuffer()));\nconst py=await loadPyodide();py.runPython(`import json\\n_REG={}`);\nfor(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}\npy.globals.set('src',strFromU8(z['rancher_rita.py']));\npy.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);\npy.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cm(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setops(h,b):\n _D['POLICY']['hands']=int(h);_D['POLICY']['build'][0]['target']=int(b);_D['POLICY']['land_after_animals']=10\n`);\nconst cr=py.globals.get('cr'),cm=py.globals.get('cm'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setops=py.globals.get('setops');\n\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}\nfunction fixed(day:number){return world.shops.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}\nfunction addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction compact(o:any){const me=+o.player;const tile=(f:any)=>{let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}};return{day:+o.day,hour:+o.hour,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON},shed:o.private?.shed,shops:o.town?.unlocked_shops}}\n\nasync function play(ref:string,switchDay:number,mode:string){prep();let s=E.state.initGameState(2,cfg,989898),tg:any={COW:0,SHEEP:0},comm=false;let pre:any=null;const marks:any={};while(!s.done){s.town.unlocked_shops=fixed(+s.day);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP');const post=+o.day>=switchDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===switchDay)pre=compact(o);const rel=+o.farms[0].money-+o.farms[1].money;if(+o.day>=switchDay)marks[+o.day]={rel,...compact(o)};const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(ref,JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{ref,switchDay,mode,pre,marks,terminal:{ours:s.farms[0].money,opp:s.farms[1].money,margin:s.farms[0].money-s.farms[1].money}}}\n\nconst runs:any[]=[];for(const ref of refs)for(const d of switchDays)for(const m of modes)runs.push(await play(ref,d,m));\nconst audit:any={};const allLocks:number[]=[];for(const ref of refs){audit[ref]={};for(const d of switchDays){const x=runs.filter(r=>r.ref===ref&&r.switchDay===d),n=x.find(r=>r.mode==='N'),days=Array.from({length:29-d},(_,i)=>d+1+i);const delta=(r:any,day:number)=>(r.marks[day]?.rel??r.terminal.margin)-(n.marks[day]?.rel??n.terminal.margin),td=(r:any)=>r.terminal.margin-n.terminal.margin;const terminalBest=[...x].sort((a,b)=>td(b)-td(a))[0].mode;const dailyBest=days.map(day=>({day,best:[...x].sort((a,b)=>delta(b,day)-delta(a,day))[0].mode}));let bestLock:number|null=null;for(const q of days){if(dailyBest.filter(z=>z.day>=q).every(z=>z.best===terminalBest)){bestLock=q;break}}const modeLocks:any={};for(const r of x.filter(r=>r.mode!=='N')){const t=td(r),sgn=Math.sign(t),series=days.map(day=>({day,v:delta(r,day),s:Math.sign(delta(r,day))}));let lock:number|null=null;for(const q of days){const tail=series.filter(z=>z.day>=q);if(tail.length&&tail.every(z=>z.s===sgn)){lock=q;break}}let changes=0;for(let i=1;id+h<=29).map(h=>[h,delta(r,d+h)])};if(lock!=null)allLocks.push(lock-d)}audit[ref][d]={preStateIdentical:new Set(x.map(r=>JSON.stringify(r.pre))).size===1,terminalBest,bestActionLockDay:bestLock,bestActionLockHorizon:bestLock==null?null:bestLock-d,bestActionChanges:dailyBest.reduce((c,z,i)=>i&&z.best!==dailyBest[i-1].best?c+1:c,0),modeLocks}}}\nconst sorted=[...allLocks].sort((a,b)=>a-b),quant=(p:number)=>sorted.length?sorted[Math.floor((sorted.length-1)*p)]:null;console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_SHADOW_LOCKIN_AUDIT',world:world.shops,summary:{modeSignLocks:allLocks.length,medianLockHorizon:quant(.5),p75LockHorizon:quant(.75),p90LockHorizon:quant(.9),maxLockHorizon:sorted.at(-1)??null},audit,interpretation:'earliest day after intervention from which sign/best-action no longer reverses through day29; null means no stable lock before terminal'},null,2));",[],[1743],"f6b32be0-e38a-4bed-8a14-e1bc1e45c986","kag_craf_landscape_stageB_v1.ts","01a079e7-bec2-76e2-9acc-a8d9216b8121",208,["D",1788751645913],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['BASE','J4B2_H8','J4B2_H9','J4B2_H10','J3B1_H9','J3B2_H9','J3B3_H9','J4B1_H9','J4B3_H9','J5B1_H9','J5B2_H9','J5B3_H9'];\nconst worlds=[{id:'MIXED',shops:['SMOOTHIE_SHOP','FARMERS_MARKET','ICE_CREAM_SHOP','YARN_STORE','FARMERS_MARKET','BAKERY','SMOOTHIE_SHOP','BAKERY']},{id:'NON_STRAW',shops:['PIZZA_SHOP','BAKERY','YARN_STORE','PET_CAFE','PIZZA_SHOP','BAKERY','YARN_STORE','PET_CAFE']}];\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:979797,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['broker_bea.py']));py.runPython(`import json\\n_b={}\\nexec(src,_b)\\n_B=_b['agent']`);py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cb(p):return json.dumps(_B(json.loads(p)))\ndef cm(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setops(h,b,l):\n _D['POLICY']['hands']=int(h);_D['POLICY']['build'][0]['target']=int(b);_D['POLICY']['land_after_animals']=int(l)\n`);const cb=py.globals.get('cb'),cm=py.globals.get('cm'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setops=py.globals.get('setops');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}\nfunction fixed(day:number,s:string[]){return s.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}\nfunction pub(f:any){let crops=0,pastures=0,animals=0;for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f?.unlocked_quadrants??[]).length,crops,pastures,animals}}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}\nfunction addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction ops(v:string,owned:number){if(v==='BASE')return{hands:8,build:16,landAfter:10};const m=v.match(/^J(\\d)B(\\d)_H(\\d+)$/);if(!m)throw new Error('BAD_'+v);const floor=+m[1],buffer=+m[2],hands=+m[3];return{hands,build:Math.max(floor,Math.min(16,owned+buffer)),landAfter:10}}\nasync function play(v:string,w:any,seat:0|1){prep();let s=E.state.initGameState(2,cfg,979797),tg:any={COW:0,SHEEP:0},comm=false,minCash=Infinity,maxFlock=0,finalFlock=0;const snaps:any[]=[];while(!s.done){s.town.unlocked_shops=fixed(+s.day,w.shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){const owned=own(o,'COW')+own(o,'SHEEP'),op=ops(v,owned);setops(op.hands,op.build,op.landAfter);if(+o.hour===0){if([1,3,6,9,12,15,18].includes(+o.day))snaps.push({day:+o.day,...pub(o.farms[pid]),hands:op.hands,build:op.build});const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[pid]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg));minCash=Math.min(minCash,+o.farms[pid].money);maxFlock=Math.max(maxFlock,owned)}else aa[pid]=JSON.parse(String(cb(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{v,world:w.id,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,minCash,animalLoss:maxFlock-finalFlock,snaps}}\nconst games:any[]=[];for(const w of worlds)for(const v of variants){games.push(await play(v,w,0));games.push(await play(v,w,1))}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const base=games.filter(g=>g.v==='BASE'),summary:any={};for(const v of variants){const x=games.filter(g=>g.v===v),d=x.map((g,i)=>g.margin-base[i].margin);summary[v]={bank:avg(x.map(g=>g.ours)),opp:avg(x.map(g=>g.theirs)),margin:avg(x.map(g=>g.margin)),delta:avg(d),minCash:Math.min(...x.map(g=>g.minCash)),animalLoss:avg(x.map(g=>g.animalLoss)),better:d.filter(z=>z>0).length,worse:d.filter(z=>z<0).length,byWorld:Object.fromEntries(worlds.map(w=>{const xx=x.filter(g=>g.world===w.id),bb=base.filter(g=>g.world===w.id);return[w.id,{delta:avg(xx.map((g,i)=>g.margin-bb[i].margin)),bank:avg(xx.map(g=>g.ours)),opp:avg(xx.map(g=>g.theirs))}]}))}}\nconst S=(f:number,b:number,h=9)=>summary[`J${f}B${b}_H${h}`]?.delta;const center=S(4,2,9),curvature={dF:(S(5,2,9)-S(3,2,9))/2,dB:(S(4,3,9)-S(4,1,9))/2,dH:(summary['J4B2_H10'].delta-summary['J4B2_H8'].delta)/2,d2F:S(5,2,9)-2*center+S(3,2,9),d2B:S(4,3,9)-2*center+S(4,1,9),d2H:summary['J4B2_H10'].delta-2*center+summary['J4B2_H8'].delta,crossFB:(S(5,3,9)-S(5,1,9)-S(3,3,9)+S(3,1,9))/4};const ranking=Object.entries(summary).map(([v,x]:any)=>({v,...x})).sort((a,b)=>b.delta-a.delta);console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_STYLE_CAUSAL_LANDSCAPE_STAGE_B',worlds:worlds.map(w=>w.id),ranking,curvature,interpretation:'selective mutation after Stage A identified hands as steepest local axis; discrete local response surface around J4B2_H9'},null,2));",[],[1751],"711a0bd6-f691-4475-ac3c-e61dbec56fe3","kag_craf_landscape_stageA_v1.ts","d71eee1c-aa6b-11f1-ace2-1607ee4eb77e",205,["D",1788751557240],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['BASE','J3B1_H8','J3B2_H8','J3B3_H8','J4B1_H8','J4B2_H8','J4B3_H8','J5B1_H8','J5B2_H8','J5B3_H8','J4B2_H7','J4B2_H9'];\nconst worlds=[{id:'MIXED',shops:['SMOOTHIE_SHOP','FARMERS_MARKET','ICE_CREAM_SHOP','YARN_STORE','FARMERS_MARKET','BAKERY','SMOOTHIE_SHOP','BAKERY']},{id:'NON_STRAW',shops:['PIZZA_SHOP','BAKERY','YARN_STORE','PET_CAFE','PIZZA_SHOP','BAKERY','YARN_STORE','PET_CAFE']}];\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:979797,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['broker_bea.py']));py.runPython(`import json\\n_b={}\\nexec(src,_b)\\n_B=_b['agent']`);py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cb(p):return json.dumps(_B(json.loads(p)))\ndef cm(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setops(h,b,l):\n _D['POLICY']['hands']=int(h);_D['POLICY']['build'][0]['target']=int(b);_D['POLICY']['land_after_animals']=int(l)\n`);const cb=py.globals.get('cb'),cm=py.globals.get('cm'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setops=py.globals.get('setops');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}\nfunction fixed(day:number,s:string[]){return s.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}\nfunction pub(f:any){let crops=0,pastures=0,animals=0;for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f?.unlocked_quadrants??[]).length,crops,pastures,animals}}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}\nfunction addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction ops(v:string,owned:number){if(v==='BASE')return{hands:8,build:16,landAfter:10};const m=v.match(/^J(\\d)B(\\d)_H(\\d+)$/);if(!m)throw new Error('BAD_'+v);const floor=+m[1],buffer=+m[2],hands=+m[3];return{hands,build:Math.max(floor,Math.min(16,owned+buffer)),landAfter:10}}\nasync function play(v:string,w:any,seat:0|1){prep();let s=E.state.initGameState(2,cfg,979797),tg:any={COW:0,SHEEP:0},comm=false,minCash=Infinity,maxFlock=0,finalFlock=0;const snaps:any[]=[];while(!s.done){s.town.unlocked_shops=fixed(+s.day,w.shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){const owned=own(o,'COW')+own(o,'SHEEP'),op=ops(v,owned);setops(op.hands,op.build,op.landAfter);if(+o.hour===0){if([1,3,6,9,12,15,18].includes(+o.day))snaps.push({day:+o.day,...pub(o.farms[pid]),hands:op.hands,build:op.build});const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[pid]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg));minCash=Math.min(minCash,+o.farms[pid].money);maxFlock=Math.max(maxFlock,owned)}else aa[pid]=JSON.parse(String(cb(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{v,world:w.id,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,minCash,animalLoss:maxFlock-finalFlock,snaps}}\nconst games:any[]=[];for(const w of worlds)for(const v of variants){games.push(await play(v,w,0));games.push(await play(v,w,1))}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const base=games.filter(g=>g.v==='BASE'),summary:any={};for(const v of variants){const x=games.filter(g=>g.v===v),d=x.map((g,i)=>g.margin-base[i].margin);summary[v]={bank:avg(x.map(g=>g.ours)),opp:avg(x.map(g=>g.theirs)),margin:avg(x.map(g=>g.margin)),delta:avg(d),minCash:Math.min(...x.map(g=>g.minCash)),animalLoss:avg(x.map(g=>g.animalLoss)),better:d.filter(z=>z>0).length,worse:d.filter(z=>z<0).length,byWorld:Object.fromEntries(worlds.map(w=>{const xx=x.filter(g=>g.world===w.id),bb=base.filter(g=>g.world===w.id);return[w.id,{delta:avg(xx.map((g,i)=>g.margin-bb[i].margin)),bank:avg(xx.map(g=>g.ours)),opp:avg(xx.map(g=>g.theirs))}]}))}}\nconst S=(f:number,b:number,h=8)=>summary[`J${f}B${b}_H${h}`]?.delta;const center=S(4,2,8),curvature={dF:(S(5,2,8)-S(3,2,8))/2,dB:(S(4,3,8)-S(4,1,8))/2,dH:(summary['J4B2_H9'].delta-summary['J4B2_H7'].delta)/2,d2F:S(5,2,8)-2*center+S(3,2,8),d2B:S(4,3,8)-2*center+S(4,1,8),d2H:summary['J4B2_H9'].delta-2*center+summary['J4B2_H7'].delta,crossFB:(S(5,3,8)-S(5,1,8)-S(3,3,8)+S(3,1,8))/4};const ranking=Object.entries(summary).map(([v,x]:any)=>({v,...x})).sort((a,b)=>b.delta-a.delta);console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_STYLE_CAUSAL_LANDSCAPE_STAGE_A',worlds:worlds.map(w=>w.id),ranking,curvature,interpretation:'discrete local response surface around J4B2_H8; not a Hessian proof, used to locate slopes/curvature/interactions before robust validation'},null,2));",[],[1759],"0ddf6431-a099-4b67-b2aa-b9bc296ce751","kag_craf_jit_hand_deconfound_v1.ts","01a079f4-5ae0-74d7-b1a3-c10bb7dd398a",231,["D",1788752587086],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:989898,weedSpawnChance:0,townShopUnlockInterval:9999});const NAMES=[...E.constants.SHOP_NAMES];function rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const worlds:any[]=[];for(const i of [1,2]){const r=rng(9300000+i*32452843);worlds.push({id:`HV${i}`,shops:Array.from({length:8},()=>NAMES[Math.floor(r()*NAMES.length)])})}const refs=['rancher_rita.py','broker_bea.py','ledger_lena.py','slotter_silas.py','closer_cleo.py'];\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cm(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setops(h,b,l):\n _D['POLICY']['hands']=int(h);_D['POLICY']['build'][0]['target']=int(b);_D['POLICY']['land_after_animals']=int(l)\n`);const cr=py.globals.get('cr'),cm=py.globals.get('cm'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setops=py.globals.get('setops');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number,s:string[]){return s.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(strategy:string,ref:string,w:any,seat:0|1){prep();let s=E.state.initGameState(2,cfg,989898),tg:any={COW:0,SHEEP:0},comm=false,minCash=Infinity,maxFlock=0,finalFlock=0;while(!s.done){s.town.unlocked_shops=fixed(+s.day,w.shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){const owned=own(o,'COW')+own(o,'SHEEP'),jit=strategy.startsWith('J'),hands=strategy.endsWith('9')?9:8;const build=jit?Math.max(4,Math.min(16,owned+2)):16;setops(hands,build,10);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[pid]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg));minCash=Math.min(minCash,+o.farms[pid].money);maxFlock=Math.max(maxFlock,owned)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{strategy,ref,world:w.id,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,win:s.farms[seat].money>s.farms[1-seat].money,minCash,animalLoss:maxFlock-finalFlock}}\nconst STR=['S8','S9','J8','J9'];const games:any[]=[];for(const w of worlds)for(const ref of refs)for(const strategy of STR){games.push(await play(strategy,ref,w,0));games.push(await play(strategy,ref,w,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);function M(strategy:string,ref?:string,world?:string){return games.filter(g=>g.strategy===strategy&&(!ref||g.ref===ref)&&(!world||g.world===world))}function fx(ref?:string,world?:string){const s8=M('S8',ref,world),s9=M('S9',ref,world),j8=M('J8',ref,world),j9=M('J9',ref,world),m=(x:any[])=>avg(x.map(g=>g.margin));return{S8:m(s8),S9:m(s9),J8:m(j8),J9:m(j9),jitAt8:m(j8)-m(s8),hand9Static:m(s9)-m(s8),hand9Jit:m(j9)-m(j8),interaction:(m(j9)-m(j8))-(m(s9)-m(s8)),j9VsS8:m(j9)-m(s8),wins:Object.fromEntries(STR.map(s=>[s,M(s,ref,world).filter(g=>g.win).length]))}}const byRef=Object.fromEntries(refs.map(r=>[r,fx(r)])),byWorld=Object.fromEntries(worlds.map(w=>[w.id,fx(undefined,w.id)]));const byRefWorld:any={};for(const r of refs)byRefWorld[r]=Object.fromEntries(worlds.map(w=>[w.id,fx(r,w.id)]));console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_DECONFOUND_2X2_JIT_X_HANDS',worlds:Object.fromEntries(worlds.map(w=>[w.id,w.shops])),overall:fx(),byWorld,byRef,byRefWorld,interpretation:'full 2x2 factorial: static-vs-JIT pasture capacity and 8-vs-9 hands; interaction separates component effects before adaptive routing'},null,2));",[],[1767],"25aa8e98-54d9-4170-a4b6-159d391374d5","kag_craf_h9_hv1_late_basin_v1.ts","01a079f2-dfaa-73e7-b170-37a1ca5f8b20",226,["D",1788752400541],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst shops=['BRUNCH_SPOT','ICE_CREAM_SHOP','BAKERY','PIZZA_SHOP','BRUNCH_SPOT','PIZZA_SHOP','PIZZA_SHOP','BAKERY'];const refs=['broker_bea.py','closer_cleo.py'];const E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:989898,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cm(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setops(h,b,l):\n _D['POLICY']['hands']=int(h);_D['POLICY']['build'][0]['target']=int(b);_D['POLICY']['land_after_animals']=int(l)\n`);const cr=py.globals.get('cr'),cm=py.globals.get('cm'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setops=py.globals.get('setops');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number){return shops.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}function pub(f:any){const crops:any={},animals:any={},structures:any={};for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops[t.crop]=(crops[t.crop]??0)+1;if(t.animal)animals[t.animal]=(animals[t.animal]??0)+1;if(t.kind&&t.kind!=='PLANT')structures[t.kind]=(structures[t.kind]??0)+1}return{cash:+f.money,land:(f?.unlocked_quadrants??[]).length,crops,animals,structures}}\nfunction ast(a:any){const units=[a?.farmer??['PASS'],...(a?.hands??[])];let pass=0,move=0,work=0;for(const u of units){const op=String(u?.[0]??'PASS');if(op==='PASS')pass++;else if(['NORTH','SOUTH','EAST','WEST'].includes(op))move++;else work++}return{pass,move,work,market:(a?.market??[]).length}}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(strategy:string,ref:string){prep();let s=E.state.initGameState(2,cfg,989898),tg:any={COW:0,SHEEP:0},comm=false;const days:any={};while(!s.done){s.town.unlocked_shops=fixed(+s.day);const o0=observationFor(s,0),o1=observationFor(s,1),owned=own(o0,'COW')+own(o0,'SHEEP');const build=strategy==='H9'?Math.max(4,Math.min(16,owned+2)):16;setops(strategy==='H9'?9:8,build,10);if(+o0.hour===0){const ow={COW:own(o0,'COW'),SHEEP:own(o0,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o0,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}let a0=addFert(o0,enforce(o0,JSON.parse(String(cm(JSON.stringify(o0)))),tg)),a1=JSON.parse(String(cr(ref,JSON.stringify(o1))));const d=+s.day,A=days[d]??={our:{work:0,pass:0,move:0,market:0},opp:{work:0,pass:0,move:0,market:0}};for(const [k,a] of [['our',a0],['opp',a1]] as any){const x=ast(a);A[k].work+=x.work;A[k].pass+=x.pass;A[k].move+=x.move;A[k].market+=x.market}if(+s.hour===0)A.snap={our:pub(s.farms[0]),opp:pub(s.farms[1]),prices:{WHEAT:s.market.prices.WHEAT,MELON:s.market.prices.MELON,MILK:s.market.prices.MILK,WOOL:s.market.prices.WOOL},build,hands:strategy==='H9'?9:8};s=E.interpreter.step(s,[a0,a1],cfg)}return{strategy,ref,final:{our:s.farms[0].money,opp:s.farms[1].money,margin:s.farms[0].money-s.farms[1].money},days}}\nconst sum=(o:any)=>({crops:Object.values(o.crops??{}).reduce((a:any,b:any)=>a+b,0),pastures:+(o.structures?.PASTURE??0),animals:Object.values(o.animals??{}).reduce((a:any,b:any)=>a+b,0),land:o.land,cash:o.cash});const out:any={};for(const ref of refs){const b=await play('BASE',ref),h=await play('H9',ref),timeline:any[]=[];for(const d of [15,18,21,24,27,29]){const B=b.days[d],H=h.days[d],bo=sum(B.snap.our),ho=sum(H.snap.our),bp=sum(B.snap.opp),hp=sum(H.snap.opp);timeline.push({day:d,relCashDelta:(ho.cash-hp.cash)-(bo.cash-bp.cash),our:{cash:ho.cash-bo.cash,crops:ho.crops-bo.crops,pastures:ho.pastures-bo.pastures,animals:ho.animals-bo.animals,land:ho.land-bo.land,work:H.our.work-B.our.work,pass:H.our.pass-B.our.pass,move:H.our.move-B.our.move,market:H.our.market-B.our.market},opp:{cash:hp.cash-bp.cash,crops:hp.crops-bp.crops,pastures:hp.pastures-bp.pastures,animals:hp.animals-bp.animals,land:hp.land-bp.land},priceDelta:{WHEAT:H.snap.prices.WHEAT-B.snap.prices.WHEAT,MELON:H.snap.prices.MELON-B.snap.prices.MELON,MILK:H.snap.prices.MILK-B.snap.prices.MILK,WOOL:H.snap.prices.WOOL-B.snap.prices.WOOL}})}out[ref]={finalDelta:h.final.margin-b.final.margin,base:b.final,h9:h.final,timeline}}\nconst compact:any={};for(const [ref,v] of Object.entries(out) as any){compact[ref]={finalDelta:v.finalDelta,base:v.base,h9:v.h9,relPath:v.timeline.map((x:any)=>[x.day,x.relCashDelta]),key:v.timeline.filter((x:any)=>[18,21,24].includes(x.day)).map((x:any)=>({day:x.day,our:x.our,opp:x.opp,priceDelta:x.priceDelta}))}}console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_FAILURE_ATLAS_HV1_COMPACT',shops,compact},null,2));",[],[1775],"82534c98-e45e-4ee3-86a6-f241e43662aa","kag_craf_h9_hv1_failure_atlas_v1.ts","70b0b410-aa6d-11f1-8876-1607ee4eb77e",222,["D",1788752323104],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst shops=['BRUNCH_SPOT','ICE_CREAM_SHOP','BAKERY','PIZZA_SHOP','BRUNCH_SPOT','PIZZA_SHOP','PIZZA_SHOP','BAKERY'];const refs=['broker_bea.py','closer_cleo.py','rancher_rita.py'];const E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:989898,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cm(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setops(h,b,l):\n _D['POLICY']['hands']=int(h);_D['POLICY']['build'][0]['target']=int(b);_D['POLICY']['land_after_animals']=int(l)\n`);const cr=py.globals.get('cr'),cm=py.globals.get('cm'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setops=py.globals.get('setops');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number){return shops.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}function pub(f:any){const crops:any={},animals:any={},structures:any={};for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops[t.crop]=(crops[t.crop]??0)+1;if(t.animal)animals[t.animal]=(animals[t.animal]??0)+1;if(t.kind&&t.kind!=='PLANT')structures[t.kind]=(structures[t.kind]??0)+1}return{cash:+f.money,land:(f?.unlocked_quadrants??[]).length,crops,animals,structures}}\nfunction ast(a:any){const units=[a?.farmer??['PASS'],...(a?.hands??[])];let pass=0,move=0,work=0;for(const u of units){const op=String(u?.[0]??'PASS');if(op==='PASS')pass++;else if(['NORTH','SOUTH','EAST','WEST'].includes(op))move++;else work++}return{pass,move,work,market:(a?.market??[]).length}}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(strategy:string,ref:string){prep();let s=E.state.initGameState(2,cfg,989898),tg:any={COW:0,SHEEP:0},comm=false;const days:any={};while(!s.done){s.town.unlocked_shops=fixed(+s.day);const o0=observationFor(s,0),o1=observationFor(s,1),owned=own(o0,'COW')+own(o0,'SHEEP');const build=strategy==='H9'?Math.max(4,Math.min(16,owned+2)):16;setops(strategy==='H9'?9:8,build,10);if(+o0.hour===0){const ow={COW:own(o0,'COW'),SHEEP:own(o0,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o0,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}let a0=addFert(o0,enforce(o0,JSON.parse(String(cm(JSON.stringify(o0)))),tg)),a1=JSON.parse(String(cr(ref,JSON.stringify(o1))));const d=+s.day,A=days[d]??={our:{work:0,pass:0,move:0,market:0},opp:{work:0,pass:0,move:0,market:0}};for(const [k,a] of [['our',a0],['opp',a1]] as any){const x=ast(a);A[k].work+=x.work;A[k].pass+=x.pass;A[k].move+=x.move;A[k].market+=x.market}if(+s.hour===0)A.snap={our:pub(s.farms[0]),opp:pub(s.farms[1]),prices:{WHEAT:s.market.prices.WHEAT,MELON:s.market.prices.MELON,MILK:s.market.prices.MILK,WOOL:s.market.prices.WOOL},build,hands:strategy==='H9'?9:8};s=E.interpreter.step(s,[a0,a1],cfg)}return{strategy,ref,final:{our:s.farms[0].money,opp:s.farms[1].money,margin:s.farms[0].money-s.farms[1].money},days}}\nconst sum=(o:any)=>({crops:Object.values(o.crops??{}).reduce((a:any,b:any)=>a+b,0),pastures:+(o.structures?.PASTURE??0),animals:Object.values(o.animals??{}).reduce((a:any,b:any)=>a+b,0),land:o.land,cash:o.cash});const out:any={};for(const ref of refs){const b=await play('BASE',ref),h=await play('H9',ref),timeline:any[]=[];for(const d of [1,3,6,9,12,15,18]){const B=b.days[d],H=h.days[d],bo=sum(B.snap.our),ho=sum(H.snap.our),bp=sum(B.snap.opp),hp=sum(H.snap.opp);timeline.push({day:d,relCashDelta:(ho.cash-hp.cash)-(bo.cash-bp.cash),our:{cash:ho.cash-bo.cash,crops:ho.crops-bo.crops,pastures:ho.pastures-bo.pastures,animals:ho.animals-bo.animals,land:ho.land-bo.land,work:H.our.work-B.our.work,pass:H.our.pass-B.our.pass,move:H.our.move-B.our.move,market:H.our.market-B.our.market},opp:{cash:hp.cash-bp.cash,crops:hp.crops-bp.crops,pastures:hp.pastures-bp.pastures,animals:hp.animals-bp.animals,land:hp.land-bp.land},priceDelta:{WHEAT:H.snap.prices.WHEAT-B.snap.prices.WHEAT,MELON:H.snap.prices.MELON-B.snap.prices.MELON,MILK:H.snap.prices.MILK-B.snap.prices.MILK,WOOL:H.snap.prices.WOOL-B.snap.prices.WOOL}})}out[ref]={finalDelta:h.final.margin-b.final.margin,base:b.final,h9:h.final,timeline}}\nconst compact:any={};for(const [ref,v] of Object.entries(out) as any){compact[ref]={finalDelta:v.finalDelta,base:v.base,h9:v.h9,relPath:v.timeline.map((x:any)=>[x.day,x.relCashDelta]),key:v.timeline.filter((x:any)=>[9,12,15,18].includes(x.day)).map((x:any)=>({day:x.day,our:x.our,opp:x.opp,priceDelta:x.priceDelta}))}}console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_FAILURE_ATLAS_HV1_COMPACT',shops,compact},null,2));",[],[1783],"cb25f486-c00f-4f37-99b5-e5b8e6cb8c1f","kag_craf_h9_heldout_v1.ts","3ba1ab18-aa6c-11f1-942f-1607ee4eb77e",209,["D",1788751725829],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:989898,weedSpawnChance:0,townShopUnlockInterval:9999});const NAMES=[...E.constants.SHOP_NAMES];function rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const worlds:any[]=[];for(let i=0;i<3;i++){const r=rng(9300000+i*32452843);worlds.push({id:`HV${i}`,shops:Array.from({length:8},()=>NAMES[Math.floor(r()*NAMES.length)])})}const refs=['melon_mateo.py','rancher_rita.py','broker_bea.py','ledger_lena.py','slotter_silas.py','closer_cleo.py'];\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cm(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setops(h,b,l):\n _D['POLICY']['hands']=int(h);_D['POLICY']['build'][0]['target']=int(b);_D['POLICY']['land_after_animals']=int(l)\n`);const cr=py.globals.get('cr'),cm=py.globals.get('cm'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setops=py.globals.get('setops');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number,s:string[]){return s.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(strategy:string,ref:string,w:any,seat:0|1){prep();let s=E.state.initGameState(2,cfg,989898),tg:any={COW:0,SHEEP:0},comm=false,minCash=Infinity,maxFlock=0,finalFlock=0;while(!s.done){s.town.unlocked_shops=fixed(+s.day,w.shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){const owned=own(o,'COW')+own(o,'SHEEP');const build=strategy==='H9'?Math.max(4,Math.min(16,owned+2)):16;setops(strategy==='H9'?9:8,build,10);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[pid]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg));minCash=Math.min(minCash,+o.farms[pid].money);maxFlock=Math.max(maxFlock,owned)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{strategy,ref,world:w.id,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,win:s.farms[seat].money>s.farms[1-seat].money,minCash,animalLoss:maxFlock-finalFlock}}\nconst games:any[]=[];for(const w of worlds)for(const ref of refs)for(const strategy of ['BASE','H9']){games.push(await play(strategy,ref,w,0));games.push(await play(strategy,ref,w,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),q=(x:number[],p:number)=>{const a=[...x].sort((u,v)=>u-v);return a[Math.floor((a.length-1)*p)]};const allB=games.filter(g=>g.strategy==='BASE'),allH=games.filter(g=>g.strategy==='H9'),d=allH.map((g,i)=>g.margin-allB[i].margin);const byRef:any={};for(const ref of refs){const b=allB.filter(g=>g.ref===ref),h=allH.filter(g=>g.ref===ref),dd=h.map((g,i)=>g.margin-b[i].margin);byRef[ref]={baseWins:b.filter(g=>g.win).length,h9Wins:h.filter(g=>g.win).length,baseBank:avg(b.map(g=>g.ours)),h9Bank:avg(h.map(g=>g.ours)),baseMargin:avg(b.map(g=>g.margin)),h9Margin:avg(h.map(g=>g.margin)),avgDelta:avg(dd),better:dd.filter(z=>z>0).length,worse:dd.filter(z=>z<0).length,minDelta:Math.min(...dd),maxDelta:Math.max(...dd)}}console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_STYLE_HELDOUT_EXOGENOUS_MULTI_OPPONENT',worlds:worlds.map(w=>w.shops),overall:{episodesPerStrategy:allH.length,baseWins:allB.filter(g=>g.win).length,h9Wins:allH.filter(g=>g.win).length,baseBank:avg(allB.map(g=>g.ours)),h9Bank:avg(allH.map(g=>g.ours)),baseMargin:avg(allB.map(g=>g.margin)),h9Margin:avg(allH.map(g=>g.margin)),avgDelta:avg(d),medianDelta:q(d,.5),p10Delta:q(d,.1),p90Delta:q(d,.9),better:d.filter(z=>z>0).length,worse:d.filter(z=>z<0).length,minDelta:Math.min(...d),maxDelta:Math.max(...d)},byRef},null,2));",[],[1791],"c98ea368-e458-49d3-bfcb-dca0696e87fd","kag_craf_h9_case_deltas_v1.ts","01a079ef-91f4-7056-a4eb-3dfbb5673d1a",217,["D",1788752159531],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:989898,weedSpawnChance:0,townShopUnlockInterval:9999});const NAMES=[...E.constants.SHOP_NAMES];function rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const worlds:any[]=[];for(let i=0;i<3;i++){const r=rng(9300000+i*32452843);worlds.push({id:`HV${i}`,shops:Array.from({length:8},()=>NAMES[Math.floor(r()*NAMES.length)])})}const refs=['rancher_rita.py','broker_bea.py','ledger_lena.py','slotter_silas.py','closer_cleo.py'];\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cm(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setops(h,b,l):\n _D['POLICY']['hands']=int(h);_D['POLICY']['build'][0]['target']=int(b);_D['POLICY']['land_after_animals']=int(l)\n`);const cr=py.globals.get('cr'),cm=py.globals.get('cm'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setops=py.globals.get('setops');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number,s:string[]){return s.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(strategy:string,ref:string,w:any,seat:0|1){prep();let s=E.state.initGameState(2,cfg,989898),tg:any={COW:0,SHEEP:0},comm=false,minCash=Infinity,maxFlock=0,finalFlock=0;while(!s.done){s.town.unlocked_shops=fixed(+s.day,w.shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){const owned=own(o,'COW')+own(o,'SHEEP');const build=strategy==='H9'?Math.max(4,Math.min(16,owned+2)):16;setops(strategy==='H9'?9:8,build,10);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[pid]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg));minCash=Math.min(minCash,+o.farms[pid].money);maxFlock=Math.max(maxFlock,owned)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{strategy,ref,world:w.id,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,win:s.farms[seat].money>s.farms[1-seat].money,minCash,animalLoss:maxFlock-finalFlock}}\nconst games:any[]=[];for(const w of worlds)for(const ref of refs)for(const strategy of ['BASE','H9']){games.push(await play(strategy,ref,w,0));games.push(await play(strategy,ref,w,1))}const cases:any[]=[];for(const w of worlds)for(const ref of refs)for(const seat of [0,1]){const b=games.find(g=>g.strategy==='BASE'&&g.world===w.id&&g.ref===ref&&g.seat===seat),h=games.find(g=>g.strategy==='H9'&&g.world===w.id&&g.ref===ref&&g.seat===seat);cases.push({world:w.id,ref,seat,delta:h.margin-b.margin,baseMargin:b.margin,h9Margin:h.margin,baseBank:b.ours,h9Bank:h.ours})}cases.sort((a,b)=>a.delta-b.delta);console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_STYLE_CASE_DELTA_ATLAS',worlds:Object.fromEntries(worlds.map(w=>[w.id,w.shops])),cases},null,2));",[],[1799],"5f33e8bb-9e84-4918-9b13-5041b84ff368","kag_craf_commission_v1.ts","01a079eb-11bd-77ab-b60b-6319f7dfa4a0",214,["D",1788751907508],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:999001,weedSpawnChance:0,townShopUnlockInterval:9999});const NAMES=[...E.constants.SHOP_NAMES];function rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const worlds:any[]=[];for(let i=0;i<2;i++){const r=rng(12300000+i*49979687);worlds.push({id:`CV${i}`,shops:Array.from({length:8},()=>NAMES[Math.floor(r()*NAMES.length)])})}const refs=['melon_mateo.py','rancher_rita.py','broker_bea.py','ledger_lena.py','slotter_silas.py','closer_cleo.py'];\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cm(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setops(h,b,l):\n _D['POLICY']['hands']=int(h);_D['POLICY']['build'][0]['target']=int(b);_D['POLICY']['land_after_animals']=int(l)\n`);const cr=py.globals.get('cr'),cm=py.globals.get('cm'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setops=py.globals.get('setops');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number,s:string[]){return s.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}function pub(f:any){let melon=0,wheat=0,animals=0;for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT'&&t.crop==='MELON')melon++;if(t.kind==='PLANT'&&t.crop==='WHEAT')wheat++;if(t.animal)animals++}return{melon,wheat,animals,money:+f.money,land:(f?.unlocked_quadrants??[]).length}}function classify(o:any){const me=+o.player,f=pub(o.farms[1-me]);if(f.melon>=5&&f.animals>=2&&f.money<30)return'TOP_LOW_CASH';if(f.melon>=5&&f.animals>=2)return'TOP_MIXED';if(f.melon>=8&&f.animals===0&&f.land>=2)return'CROP';if(f.melon===0&&f.wheat>=8)return'LIVESTOCK';return'UNKNOWN'}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(strategy:string,ref:string,w:any,seat:0|1){prep();let s=E.state.initGameState(2,cfg,999001),tg:any={COW:0,SHEEP:0},comm=false,minCash=Infinity,maxFlock=0,finalFlock=0,latched='UNKNOWN',decisionDay:number|null=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,w.shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){const owned=own(o,'COW')+own(o,'SHEEP');if(strategy==='COMMISSION'&&latched==='UNKNOWN'){const c=classify(o);if(c!=='UNKNOWN'){latched=c;decisionDay=+o.day}}let hands=8,build=16;if(strategy==='H9'){hands=9;build=Math.max(4,Math.min(16,owned+2))}else if(strategy==='COMMISSION'){if(latched==='UNKNOWN'){hands=8;build=Math.max(4,Math.min(16,owned+2))}else if(latched==='LIVESTOCK'||latched==='TOP_LOW_CASH'){hands=9;build=Math.max(4,Math.min(16,owned+2))}else{hands=8;build=16}}setops(hands,build,10);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[pid]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg));minCash=Math.min(minCash,+o.farms[pid].money);maxFlock=Math.max(maxFlock,owned)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{strategy,ref,world:w.id,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,win:s.farms[seat].money>s.farms[1-seat].money,minCash,animalLoss:maxFlock-finalFlock,latched,decisionDay}}\nconst games:any[]=[];for(const w of worlds)for(const ref of refs)for(const strategy of ['BASE','H9','COMMISSION']){games.push(await play(strategy,ref,w,0));games.push(await play(strategy,ref,w,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),q=(x:number[],p:number)=>{const a=[...x].sort((u,v)=>u-v);return a[Math.floor((a.length-1)*p)]};const base=games.filter(g=>g.strategy==='BASE'),summary:any={};for(const strategy of ['BASE','H9','COMMISSION']){const x=games.filter(g=>g.strategy===strategy),d=strategy==='BASE'?[]:x.map((g,i)=>g.margin-base[i].margin);summary[strategy]={episodes:x.length,wins:x.filter(g=>g.win).length,bank:avg(x.map(g=>g.ours)),opp:avg(x.map(g=>g.theirs)),margin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),animalLoss:avg(x.map(g=>g.animalLoss)),avgDelta:strategy==='BASE'?0:avg(d),medianDelta:strategy==='BASE'?0:q(d,.5),p10Delta:strategy==='BASE'?0:q(d,.1),better:strategy==='BASE'?0:d.filter(z=>z>0).length,worse:strategy==='BASE'?0:d.filter(z=>z<0).length}}const byRef:any={};for(const ref of refs){const b=base.filter(g=>g.ref===ref);byRef[ref]={};for(const strategy of ['H9','COMMISSION']){const x=games.filter(g=>g.ref===ref&&g.strategy===strategy),d=x.map((g,i)=>g.margin-b[i].margin);byRef[ref][strategy]={wins:x.filter(g=>g.win).length,avgDelta:avg(d),better:d.filter(z=>z>0).length,worse:d.filter(z=>z<0).length,latched:[...new Set(x.map(g=>g.latched))],decisionDays:[...new Set(x.map(g=>g.decisionDay))]}}}console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_STYLE_COMMISSIONING_SCREEN_HELDOUT_EXOGENOUS',worlds:worlds.map(w=>w.shops),doctrine:'start minimum-regret JIT H8; latch opponent type; select H9+JIT only for LIVESTOCK or TOP_LOW_CASH, otherwise revert to H8/static capacity',summary,byRef},null,2));",[],[1807],"2702b103-7f80-43e3-95c0-2a63eab28446","kag_craf_capacity_shadow_stage1_v1.ts","01a07a17-9d9f-7380-9ab4-0b56299622b2",245,["D",1788755008007],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst E=await loadOfficialKaggricultureEngine();\nconst cfg=E.state.resolveConfig({seed:989898,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst world={id:'HV1',shops:['BRUNCH_SPOT','ICE_CREAM_SHOP','BAKERY','PIZZA_SHOP','BRUNCH_SPOT','PIZZA_SHOP','PIZZA_SHOP','BAKERY']};\nconst refs=['rancher_rita.py','broker_bea.py','closer_cleo.py'];\nconst switchDays=[6,9,12];\nconst modes=['N','H','W','B']; // N=keep J8, H=JIT+9 hands, W=widen pasture to16, B=widen+9\n\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');\nconst z=unzipSync(new Uint8Array(await rr.arrayBuffer()));\nconst py=await loadPyodide();py.runPython(`import json\\n_REG={}`);\nfor(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}\npy.globals.set('src',strFromU8(z['rancher_rita.py']));\npy.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);\npy.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cm(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setops(h,b):\n _D['POLICY']['hands']=int(h);_D['POLICY']['build'][0]['target']=int(b);_D['POLICY']['land_after_animals']=10\n`);\nconst cr=py.globals.get('cr'),cm=py.globals.get('cm'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setops=py.globals.get('setops');\n\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}\nfunction fixed(day:number){return world.shops.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}\nfunction addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction compact(o:any){const me=+o.player;const tile=(f:any)=>{let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}};return{day:+o.day,hour:+o.hour,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON},shed:o.private?.shed,shops:o.town?.unlocked_shops}}\n\nasync function play(ref:string,switchDay:number,mode:string){prep();let s=E.state.initGameState(2,cfg,989898),tg:any={COW:0,SHEEP:0},comm=false;let pre:any=null;const marks:any={};while(!s.done){s.town.unlocked_shops=fixed(+s.day);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP');const post=+o.day>=switchDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===switchDay)pre=compact(o);const rel=+o.farms[0].money-+o.farms[1].money;if([switchDay,switchDay+3,switchDay+6,switchDay+9,switchDay+12].includes(+o.day))marks[+o.day]={rel,...compact(o)};const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(ref,JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{ref,switchDay,mode,pre,marks,terminal:{ours:s.farms[0].money,opp:s.farms[1].money,margin:s.farms[0].money-s.farms[1].money}}}\n\nconst runs:any[]=[];for(const ref of refs)for(const d of switchDays)for(const m of modes)runs.push(await play(ref,d,m));\nconst out:any={};for(const ref of refs){out[ref]={};for(const d of switchDays){const x=runs.filter(r=>r.ref===ref&&r.switchDay===d),n=x.find(r=>r.mode==='N');const preHashes=[...new Set(x.map(r=>JSON.stringify(r.pre)))];const deltaAt=(r:any,h:number)=>(r.marks[d+h]?.rel??r.terminal.margin)-(n.marks[d+h]?.rel??n.terminal.margin);const rows=x.map(r=>({mode:r.mode,d3:deltaAt(r,3),d6:deltaAt(r,6),d9:deltaAt(r,9),d12:deltaAt(r,12),terminalDelta:r.terminal.margin-n.terminal.margin,terminal:r.terminal}));out[ref][d]={preStateIdentical:preHashes.length===1,pre:n.pre,rows,bestTerminal:[...rows].sort((a,b)=>b.terminalDelta-a.terminalDelta)[0].mode}}}\nconst compactOut:any={};for(const [ref,byDay] of Object.entries(out) as any){compactOut[ref]={};for(const [d,v] of Object.entries(byDay) as any)compactOut[ref][d]={preStateIdentical:v.preStateIdentical,pre:{cash:v.pre.me.cash,oppCash:v.pre.opp.cash,crops:v.pre.me.crops,pastures:v.pre.me.pastures,animals:v.pre.me.animals,land:v.pre.me.land,melon:v.pre.market.MELON,milk:v.pre.market.MILK,wool:v.pre.market.WOOL},rows:v.rows.map((r:any)=>({mode:r.mode,d3:r.d3,d6:r.d6,terminalDelta:r.terminalDelta})),bestTerminal:v.bestTerminal,bestD3:v.bestD3}}\n// stage1 forecast-horizon audit\nconst audits:any[]=[];for(const [ref,byDay] of Object.entries(out) as any)for(const [d,v] of Object.entries(byDay) as any){const non=v.rows.filter((r:any)=>r.mode!=='N'),pick=(k:string)=>[...v.rows].sort((a:any,b:any)=>b[k]-a[k])[0].mode;const best3=pick('d3'),best6=pick('d6'),best9=pick('d9'),best12=pick('d12'),bestT=pick('terminalDelta');for(const r of non)audits.push({ref,day:+d,mode:r.mode,d3:r.d3,d6:r.d6,d9:r.d9,d12:r.d12,t:r.terminalDelta,sign3:Math.sign(r.d3)===Math.sign(r.terminalDelta),sign6:Math.sign(r.d6)===Math.sign(r.terminalDelta),sign9:Math.sign(r.d9)===Math.sign(r.terminalDelta),sign12:Math.sign(r.d12)===Math.sign(r.terminalDelta)});audits.push({ref,day:+d,mode:'__BEST__',best3,best6,best9,best12,bestT,hit3:best3===bestT,hit6:best6===bestT,hit9:best9===bestT,hit12:best12===bestT})}\nconst bests=audits.filter(x=>x.mode==='__BEST__'),acts=audits.filter(x=>x.mode!=='__BEST__');const reversals=acts.filter(x=>[x.d3,x.d6,x.d9,x.d12].some((v:number)=>Math.sign(v)!==Math.sign(x.t))).sort((a,b)=>Math.abs(b.t)-Math.abs(a.t));console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_SHADOW_HORIZON_AUDIT_EXTENDED',decisionContexts:bests.length,bestActionAccuracy:{d3:bests.filter(x=>x.hit3).length/bests.length,d6:bests.filter(x=>x.hit6).length/bests.length,d9:bests.filter(x=>x.hit9).length/bests.length,d12:bests.filter(x=>x.hit12).length/bests.length},signAccuracy:{d3:acts.filter(x=>x.sign3).length/acts.length,d6:acts.filter(x=>x.sign6).length/acts.length,d9:acts.filter(x=>x.sign9).length/acts.length,d12:acts.filter(x=>x.sign12).length/acts.length},bests,reversals:reversals.slice(0,12)},null,2));",[],[1815],"3cbfc787-4cae-40d3-8b5f-98bc7358fb8f","kag_craf_capacity_regime_discovery_v1.ts","01a079f8-6433-775d-8db7-554f16a5d93b",236,["D",1788752751521],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:919191,weedSpawnChance:0,townShopUnlockInterval:9999});const NAMES=[...E.constants.SHOP_NAMES];function rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const worlds:any[]=[];for(let i=0;i<8;i++){const r=rng(19000000+i*67867967);worlds.push({id:`D${i}`,shops:Array.from({length:8},()=>NAMES[Math.floor(r()*NAMES.length)])})}const refs=['broker_bea.py','ledger_lena.py','slotter_silas.py'];\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cm(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setops(h,b,l):\n _D['POLICY']['hands']=int(h);_D['POLICY']['build'][0]['target']=int(b);_D['POLICY']['land_after_animals']=int(l)\n`);const cr=py.globals.get('cr'),cm=py.globals.get('cm'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setops=py.globals.get('setops');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number,s:string[]){return s.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(strategy:string,ref:string,w:any,seat:0|1){prep();let s=E.state.initGameState(2,cfg,919191),tg:any={COW:0,SHEEP:0},comm=false,minCash=Infinity,maxFlock=0,finalFlock=0;while(!s.done){s.town.unlocked_shops=fixed(+s.day,w.shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){const owned=own(o,'COW')+own(o,'SHEEP'),jit=strategy.startsWith('J'),hands=strategy.endsWith('9')?9:8;const build=jit?Math.max(4,Math.min(16,owned+2)):16;setops(hands,build,10);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[pid]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg));minCash=Math.min(minCash,+o.farms[pid].money);maxFlock=Math.max(maxFlock,owned)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{strategy,ref,world:w.id,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,win:s.farms[seat].money>s.farms[1-seat].money,minCash,animalLoss:maxFlock-finalFlock}}\nconst STR=['S8','S9','J8','J9'];const games:any[]=[];for(const w of worlds)for(const ref of refs)for(const strategy of STR)games.push(await play(strategy,ref,w,0));const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);function M(strategy:string,world:string){return games.filter(g=>g.strategy===strategy&&g.world===world)}function fx(world:string){const m=(s:string)=>avg(M(s,world).map(g=>g.margin)),S8=m('S8'),S9=m('S9'),J8=m('J8'),J9=m('J9');return{S8,S9,J8,J9,jitAt8:J8-S8,hand9Static:S9-S8,hand9Jit:J9-J8,interaction:(J9-J8)-(S9-S8),j9VsS8:J9-S8,best:[['S8',S8],['S9',S9],['J8',J8],['J9',J9]].sort((a:any,b:any)=>b[1]-a[1])[0][0]}}const SH:any=E.constants.SHOPS;function feat(w:any){const first=w.shops.slice(0,2),products=first.flatMap((s:string)=>SH[s]??[]);const cnt=(p:string)=>products.filter((x:string)=>x===p).length;return{first2:first,milk:cnt('MILK'),wheat:cnt('WHEAT'),strawberry:cnt('STRAWBERRY'),egg:cnt('EGG'),carrot:cnt('CARROT'),tomato:cnt('TOMATO'),wool:cnt('WOOL')}}const rows=worlds.map(w=>({world:w.id,...feat(w),...fx(w.id)})).sort((a,b)=>b.j9VsS8-a.j9VsS8);console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_CAPACITY_REGIME_DISCOVERY_TOP3_FIXED_WORLDS',rows,warning:'discovery set only; any rule derived here requires heldout worlds'},null,2));",[],[1823],"76c14b44-aff2-46a2-86d5-509e06da029c","kag_craf_bm_mpc_voc_gate_batch1_v1.ts","01a07a2a-d676-72b8-ae5f-3ecfb23466bb",273,["D",1788756063353],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES];\nconst cfg=E.state.resolveConfig({seed:929292,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst modes=['N','H','W','B'],decisionDay=9,OUTER=2,INNER=12;\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}\nfunction fullWorld(seed:number){const r=rng(seed);return Array.from({length:8},()=>NAMES[Math.floor(r()*NAMES.length)])}\nfunction innerWorlds(prefix:string[],outerIdx:number){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction compact(o:any){const me=+o.player,f=o.farms[me],g=o.farms[1-me];return{day:+o.day,cash:+f.money,oppCash:+g.money,shops:o.town.unlocked_shops,market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,929292),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=compact(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,ours:s.farms[0].money,opp:s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},stages=[4,8,12];\nfunction chooseGate(vals:any){const audit:any[]=[];let finalPick='N',used=INNER;for(const n of stages){const stats:any={N:{mean:avg(vals.N.slice(0,n)),lo95:0,delta:0}};for(const m of modes.filter(x=>x!=='N')){const d=vals[m].slice(0,n).map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=sd(d)/Math.sqrt(n),lo=mu-1.96*se;stats[m]={mean:avg(vals[m].slice(0,n)),delta:mu,se,lo95:lo}}const eligible=modes.filter(m=>m!=='N'&&stats[m].lo95>0).sort((a,b)=>stats[b].lo95-stats[a].lo95);const pick=eligible[0]??'N';audit.push({n,pick,eligible,stats});if(eligible.length){finalPick=pick;used=n;break}}return{pick:finalPick,used,audit}}\nconst rows:any[]=[];for(let oi=0;oi();for(const m of modes){vals[m]=[];for(const iw of inners){const r=await play(iw,m);vals[m].push(r.margin);preSet.add(JSON.stringify(r.pre))}}const gate=chooseGate(vals),actualRuns:any={};for(const m of modes)actualRuns[m]=await play(actual,m);const oracle=[...modes].sort((a,b)=>actualRuns[b].margin-actualRuns[a].margin)[0],base=actualRuns.N.margin;rows.push({outer:oi,prefix,hiddenSuffix:actual.slice(prefix.length),preStateIdentical:preSet.size===1,gatePick:gate.pick,scenariosUsed:gate.used,gateAudit:gate.audit,oracle,actualDeltas:Object.fromEntries(modes.map(m=>[m,actualRuns[m].margin-base])),realizedDelta:actualRuns[gate.pick].margin-base,oracleDelta:actualRuns[oracle].margin-base,regret:actualRuns[oracle].margin-actualRuns[gate.pick].margin})}\nconst s=(k:string)=>avg(rows.map(r=>r[k]));console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_BM_MPC_VOC_CONFIDENCE_GATE_BATCH1',opponent:'broker_bea.py',decisionDay,outerWorlds:OUTER,maxInnerScenarios:INNER,gate:'sequential n=4/8/12; intervene only if paired 95% lower bound vs N > 0; choose highest lower bound',aggregate:{realizedDelta:s('realizedDelta'),regret:s('regret'),oracleDelta:s('oracleDelta'),avgScenariosUsed:s('scenariosUsed'),positive:rows.filter(r=>r.realizedDelta>0).length,negative:rows.filter(r=>r.realizedDelta<0).length,neutral:rows.filter(r=>r.realizedDelta===0).length},rows},null,2));",[],[1831],"2b1fe468-fe23-4761-8ffb-1e5af32aefd0","kag_craf_bm_mpc_stability_gate_batch2_v1.ts","01a07a3f-e2af-72f3-ae18-808a5b843210",290,["D",1788757453239],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES];\nconst cfg=E.state.resolveConfig({seed:929292,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst modes=['N','H','W','B'],decisionDay=9,OUTER=2,INNER=16;\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}\nfunction fullWorld(seed:number){const r=rng(seed);return Array.from({length:8},()=>NAMES[Math.floor(r()*NAMES.length)])}\nfunction innerWorlds(prefix:string[],outerIdx:number){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction compact(o:any){const me=+o.player,f=o.farms[me],g=o.farms[1-me];return{day:+o.day,cash:+f.money,oppCash:+g.money,shops:o.town.unlocked_shops,market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,929292),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=compact(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,ours:s.farms[0].money,opp:s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},stages=[8,12,16],tcrit:any={8:2.365,12:2.201,16:2.131};\nfunction chooseGate(vals:any){const audit:any[]=[];let finalPick='N',used=INNER,prev:any=null;for(const n of stages){const stats:any={N:{mean:avg(vals.N.slice(0,n)),lo95:0,delta:0}};for(const m of modes.filter(x=>x!=='N')){const d=vals[m].slice(0,n).map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=sd(d)/Math.sqrt(n),lo=mu-tcrit[n]*se;stats[m]={mean:avg(vals[m].slice(0,n)),delta:mu,se,lo95:lo}}const positive=modes.filter(m=>m!=='N'&&stats[m].lo95>0);const persistent=prev?positive.filter(m=>prev[m]?.lo95>0).map(m=>({m,robustLo:Math.min(prev[m].lo95,stats[m].lo95)})).sort((a,b)=>b.robustLo-a.robustLo):[];const pick=persistent[0]?.m??'N';audit.push({n,pick,positive,persistent,stats});if(persistent.length){finalPick=pick;used=n;break}prev=stats}return{pick:finalPick,used,audit}}\nfunction chooseOld(vals:any){for(const n of [4,8,12]){const cand:any[]=[];for(const m of modes.filter(x=>x!=='N')){const d=vals[m].slice(0,n).map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=sd(d)/Math.sqrt(n),lo=mu-1.96*se;if(lo>0)cand.push({m,lo})}cand.sort((a,b)=>b.lo-a.lo);if(cand.length)return{pick:cand[0].m,used:n}}return{pick:'N',used:12}}\nconst rows:any[]=[];for(let oi=0;oi();for(const m of modes){vals[m]=[];for(const iw of inners){const r=await play(iw,m);vals[m].push(r.margin);preSet.add(JSON.stringify(r.pre))}}const gate=chooseGate(vals),old=chooseOld(vals),actualRuns:any={};for(const m of modes)actualRuns[m]=await play(actual,m);const oracle=[...modes].sort((a,b)=>actualRuns[b].margin-actualRuns[a].margin)[0],base=actualRuns.N.margin;rows.push({outer:oi,prefix,hiddenSuffix:actual.slice(prefix.length),preStateIdentical:preSet.size===1,gatePick:gate.pick,scenariosUsed:gate.used,oldPick:old.pick,oldScenariosUsed:old.used,gateAudit:gate.audit,oracle,actualDeltas:Object.fromEntries(modes.map(m=>[m,actualRuns[m].margin-base])),realizedDelta:actualRuns[gate.pick].margin-base,oldRealizedDelta:actualRuns[old.pick].margin-base,oracleDelta:actualRuns[oracle].margin-base,regret:actualRuns[oracle].margin-actualRuns[gate.pick].margin,oldRegret:actualRuns[oracle].margin-actualRuns[old.pick].margin})}\nconst s=(k:string)=>avg(rows.map(r=>r[k]));console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_BM_MPC_STABILITY_GATE_BATCH2_RULE_COMPARISON',opponent:'broker_bea.py',decisionDay,outerWorlds:OUTER,maxInnerScenarios:INNER,newGate:'paired Student-t LCB + persistence across 2 consecutive stages',oldGate:'first positive normal-approx LCB at n=4/8/12',aggregate:{stability:{realizedDelta:s('realizedDelta'),regret:s('regret'),avgScenariosUsed:s('scenariosUsed'),positive:rows.filter(r=>r.realizedDelta>0).length,negative:rows.filter(r=>r.realizedDelta<0).length,neutral:rows.filter(r=>r.realizedDelta===0).length},old:{realizedDelta:s('oldRealizedDelta'),regret:s('oldRegret'),avgScenariosUsed:s('oldScenariosUsed'),positive:rows.filter(r=>r.oldRealizedDelta>0).length,negative:rows.filter(r=>r.oldRealizedDelta<0).length,neutral:rows.filter(r=>r.oldRealizedDelta===0).length},oracleDelta:s('oracleDelta')},rows},null,2));",[],[1839],"10956ddb-22cb-42df-82bc-8611a2f957cd","kag_craf_bm_mpc_stability_gate_batch1_v1.ts","01a07a3a-acfa-772a-81e3-19c92cd11d74",284,["D",1788757265181],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES];\nconst cfg=E.state.resolveConfig({seed:929292,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst modes=['N','H','W','B'],decisionDay=9,OUTER=2,INNER=16;\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}\nfunction fullWorld(seed:number){const r=rng(seed);return Array.from({length:8},()=>NAMES[Math.floor(r()*NAMES.length)])}\nfunction innerWorlds(prefix:string[],outerIdx:number){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction compact(o:any){const me=+o.player,f=o.farms[me],g=o.farms[1-me];return{day:+o.day,cash:+f.money,oppCash:+g.money,shops:o.town.unlocked_shops,market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,929292),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=compact(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,ours:s.farms[0].money,opp:s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},stages=[8,12,16],tcrit:any={8:2.365,12:2.201,16:2.131};\nfunction chooseGate(vals:any){const audit:any[]=[];let finalPick='N',used=INNER,prev:any=null;for(const n of stages){const stats:any={N:{mean:avg(vals.N.slice(0,n)),lo95:0,delta:0}};for(const m of modes.filter(x=>x!=='N')){const d=vals[m].slice(0,n).map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=sd(d)/Math.sqrt(n),lo=mu-tcrit[n]*se;stats[m]={mean:avg(vals[m].slice(0,n)),delta:mu,se,lo95:lo}}const positive=modes.filter(m=>m!=='N'&&stats[m].lo95>0);const persistent=prev?positive.filter(m=>prev[m]?.lo95>0).map(m=>({m,robustLo:Math.min(prev[m].lo95,stats[m].lo95)})).sort((a,b)=>b.robustLo-a.robustLo):[];const pick=persistent[0]?.m??'N';audit.push({n,pick,positive,persistent,stats});if(persistent.length){finalPick=pick;used=n;break}prev=stats}return{pick:finalPick,used,audit}}\nconst rows:any[]=[];for(let oi=0;oi();for(const m of modes){vals[m]=[];for(const iw of inners){const r=await play(iw,m);vals[m].push(r.margin);preSet.add(JSON.stringify(r.pre))}}const gate=chooseGate(vals),actualRuns:any={};for(const m of modes)actualRuns[m]=await play(actual,m);const oracle=[...modes].sort((a,b)=>actualRuns[b].margin-actualRuns[a].margin)[0],base=actualRuns.N.margin;rows.push({outer:oi,prefix,hiddenSuffix:actual.slice(prefix.length),preStateIdentical:preSet.size===1,gatePick:gate.pick,scenariosUsed:gate.used,gateAudit:gate.audit,oracle,actualDeltas:Object.fromEntries(modes.map(m=>[m,actualRuns[m].margin-base])),realizedDelta:actualRuns[gate.pick].margin-base,oracleDelta:actualRuns[oracle].margin-base,regret:actualRuns[oracle].margin-actualRuns[gate.pick].margin})}\nconst s=(k:string)=>avg(rows.map(r=>r[k]));console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_BM_MPC_STABILITY_GATE_BATCH1',opponent:'broker_bea.py',decisionDay,outerWorlds:OUTER,maxInnerScenarios:INNER,gate:'paired Student-t LCB; observe n=8; intervene only after same action has LCB>0 in two consecutive stages (8->12 or 12->16); otherwise N',aggregate:{realizedDelta:s('realizedDelta'),regret:s('regret'),oracleDelta:s('oracleDelta'),avgScenariosUsed:s('scenariosUsed'),positive:rows.filter(r=>r.realizedDelta>0).length,negative:rows.filter(r=>r.realizedDelta<0).length,neutral:rows.filter(r=>r.realizedDelta===0).length},rows},null,2));",[],[1847],"7993ec29-e6d9-4dc0-bbcf-3c0b28fc9c70","kag_craf_bm_mpc_scenario_stage2_v1.ts","01a07a1d-d4b6-70b7-bc4a-d15335c799d4",250,["D",1788755201517],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES];\nconst cfg=E.state.resolveConfig({seed:989898,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst actual=['BRUNCH_SPOT','ICE_CREAM_SHOP','BAKERY','PIZZA_SHOP','BRUNCH_SPOT','PIZZA_SHOP','PIZZA_SHOP','BAKERY'];\nconst switchDays=[6,9],modes=['N','H','W','B'],SCEN=6;\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}\nfunction scenarios(day:number){const k=Math.floor(day/3),prefix=actual.slice(0,k),out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction compact(o:any){const me=+o.player,f=o.farms[me],g=o.farms[1-me];return{day:+o.day,cash:+f.money,oppCash:+g.money,shops:o.town.unlocked_shops,market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL}}}\nasync function play(shops:string[],switchDay:number,mode:string){prep();let s=E.state.initGameState(2,cfg,989898),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=switchDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===switchDay)pre=compact(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,ours:s.farms[0].money,opp:s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),worst=(x:number[])=>[...x].sort((a,b)=>a-b),cvar33=(x:number[])=>avg(worst(x).slice(0,Math.max(1,Math.ceil(x.length/3))));\nconst result:any={};for(const d of switchDays){const sc=scenarios(d),shadow:any={};const preHashes=new Set();for(const m of modes){const vals:any[]=[];for(const s of sc){const r=await play(s,d,m);vals.push(r);preHashes.add(JSON.stringify(r.pre))}shadow[m]={mean:avg(vals.map(v=>v.margin)),cvar33:cvar33(vals.map(v=>v.margin)),min:Math.min(...vals.map(v=>v.margin)),max:Math.max(...vals.map(v=>v.margin))}}const actualRuns:any={};for(const m of modes)actualRuns[m]=await play(actual,d,m);const oracle=[...modes].sort((a,b)=>actualRuns[b].margin-actualRuns[a].margin)[0],meanPick=[...modes].sort((a,b)=>shadow[b].mean-shadow[a].mean)[0],riskPick=[...modes].sort((a,b)=>(shadow[b].mean+.35*shadow[b].cvar33)-(shadow[a].mean+.35*shadow[a].cvar33))[0];result[d]={observedPrefix:actual.slice(0,Math.floor(d/3)),scenarioSuffixes:sc.map(s=>s.slice(Math.floor(d/3))),preStateIdenticalAcrossShadow:preHashes.size===1,shadow,meanPick,riskPick,actual:Object.fromEntries(modes.map(m=>[m,{margin:actualRuns[m].margin,delta:actualRuns[m].margin-actualRuns.N.margin}])),oracle,meanHit:meanPick===oracle,riskHit:riskPick===oracle}}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_BM_MPC_SCENARIO_TERMINAL_STAGE2',opponent:'broker_bea.py',actualWorld:actual,scenarioCount:SCEN,decision:'terminal shadow over sampled future shops conditioned on observed prefix; no access to actual suffix',result},null,2));",[],[1855],"6a1f923d-7881-46ce-af7b-bb51e3f22556","kag_craf_bm_mpc_scenario_heldout_v1.ts","01a07a1f-6b47-74f8-9de9-520f07e40a93",255,["D",1788755359501],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES];\nconst cfg=E.state.resolveConfig({seed:989899,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst actual=['SMOOTHIE_SHOP','PET_CAFE','ICE_CREAM_SHOP','BAKERY','FARMERS_MARKET','PIZZA_SHOP','BAKERY','BRUNCH_SPOT'];\nconst switchDays=[6,9],modes=['N','H','W','B'],SCEN=6;\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}\nfunction scenarios(day:number){const k=Math.floor(day/3),prefix=actual.slice(0,k),out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction compact(o:any){const me=+o.player,f=o.farms[me],g=o.farms[1-me];return{day:+o.day,cash:+f.money,oppCash:+g.money,shops:o.town.unlocked_shops,market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL}}}\nasync function play(shops:string[],switchDay:number,mode:string){prep();let s=E.state.initGameState(2,cfg,989899),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=switchDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===switchDay)pre=compact(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,ours:s.farms[0].money,opp:s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),worst=(x:number[])=>[...x].sort((a,b)=>a-b),cvar33=(x:number[])=>avg(worst(x).slice(0,Math.max(1,Math.ceil(x.length/3))));\nconst result:any={};for(const d of switchDays){const sc=scenarios(d),shadow:any={};const preHashes=new Set();for(const m of modes){const vals:any[]=[];for(const s of sc){const r=await play(s,d,m);vals.push(r);preHashes.add(JSON.stringify(r.pre))}shadow[m]={mean:avg(vals.map(v=>v.margin)),cvar33:cvar33(vals.map(v=>v.margin)),min:Math.min(...vals.map(v=>v.margin)),max:Math.max(...vals.map(v=>v.margin))}}const actualRuns:any={};for(const m of modes)actualRuns[m]=await play(actual,d,m);const oracle=[...modes].sort((a,b)=>actualRuns[b].margin-actualRuns[a].margin)[0],meanPick=[...modes].sort((a,b)=>shadow[b].mean-shadow[a].mean)[0],riskPick=[...modes].sort((a,b)=>(shadow[b].mean+.35*shadow[b].cvar33)-(shadow[a].mean+.35*shadow[a].cvar33))[0];result[d]={observedPrefix:actual.slice(0,Math.floor(d/3)),scenarioSuffixes:sc.map(s=>s.slice(Math.floor(d/3))),preStateIdenticalAcrossShadow:preHashes.size===1,shadow,meanPick,riskPick,actual:Object.fromEntries(modes.map(m=>[m,{margin:actualRuns[m].margin,delta:actualRuns[m].margin-actualRuns.N.margin}])),oracle,meanHit:meanPick===oracle,riskHit:riskPick===oracle}}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_BM_MPC_SCENARIO_TERMINAL_STAGE2',opponent:'broker_bea.py',actualWorld:actual,scenarioCount:SCEN,decision:'terminal shadow over sampled future shops conditioned on observed prefix; no access to actual suffix',result},null,2));",[],[1863],"6d911b68-322e-45f1-a78a-a76b81282399","kag_craf_bm_mpc_nested_batch2_v1.ts","01a07a24-fd20-7695-9e9b-a3448eb1cab6",265,["D",1788755665612],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES];\nconst cfg=E.state.resolveConfig({seed:939393,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst modes=['N','H','W','B'],decisionDay=9,OUTER=3,INNER=4;\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}\nfunction fullWorld(seed:number){const r=rng(seed);return Array.from({length:8},()=>NAMES[Math.floor(r()*NAMES.length)])}\nfunction innerWorlds(prefix:string[],outerIdx:number){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction compact(o:any){const me=+o.player,f=o.farms[me],g=o.farms[1-me];return{day:+o.day,cash:+f.money,oppCash:+g.money,shops:o.town.unlocked_shops,market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,939393),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=compact(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,ours:s.farms[0].money,opp:s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.max(1,Math.ceil(a.length/2))))};\nconst rows:any[]=[];for(let oi=0;oi();for(const m of modes){const vals=[];for(const iw of inners){const r=await play(iw,m);vals.push(r.margin);preSet.add(JSON.stringify(r.pre))}shadow[m]={mean:avg(vals),cvar50:cvar50(vals)}}const meanPick=[...modes].sort((a,b)=>shadow[b].mean-shadow[a].mean)[0],riskPick=[...modes].sort((a,b)=>(shadow[b].mean+.25*shadow[b].cvar50)-(shadow[a].mean+.25*shadow[a].cvar50))[0];const actualRuns:any={};for(const m of modes)actualRuns[m]=await play(actual,m);const oracle=[...modes].sort((a,b)=>actualRuns[b].margin-actualRuns[a].margin)[0],base=actualRuns.N.margin;rows.push({outer:oi,prefix,hiddenSuffix:actual.slice(prefix.length),preStateIdentical:preSet.size===1,shadow,meanPick,riskPick,oracle,actualDeltas:Object.fromEntries(modes.map(m=>[m,actualRuns[m].margin-base])),meanRealizedDelta:actualRuns[meanPick].margin-base,riskRealizedDelta:actualRuns[riskPick].margin-base,oracleDelta:actualRuns[oracle].margin-base,meanRegret:actualRuns[oracle].margin-actualRuns[meanPick].margin,riskRegret:actualRuns[oracle].margin-actualRuns[riskPick].margin})}\nconst s=(k:string)=>avg(rows.map(r=>r[k]));console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_BM_MPC_NESTED_MONTE_CARLO_BATCH2',opponent:'broker_bea.py',decisionDay,outerWorlds:OUTER,innerScenarios:INNER,aggregate:{meanSelectorRealizedDelta:s('meanRealizedDelta'),riskSelectorRealizedDelta:s('riskRealizedDelta'),meanRegret:s('meanRegret'),riskRegret:s('riskRegret'),meanOracleDelta:s('oracleDelta'),meanPickBetterThanN:rows.filter(r=>r.meanRealizedDelta>0).length,riskPickBetterThanN:rows.filter(r=>r.riskRealizedDelta>0).length},rows},null,2));",[],[1871],"3ead06df-37ae-48d8-ac62-7e29c5a2385a","kag_craf_bm_mpc_nested_batch1_v1.ts","01a07a23-8224-7596-823e-95c4567c68db",260,["D",1788755567053],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES];\nconst cfg=E.state.resolveConfig({seed:929292,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst modes=['N','H','W','B'],decisionDay=9,OUTER=3,INNER=4;\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}\nfunction fullWorld(seed:number){const r=rng(seed);return Array.from({length:8},()=>NAMES[Math.floor(r()*NAMES.length)])}\nfunction innerWorlds(prefix:string[],outerIdx:number){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction compact(o:any){const me=+o.player,f=o.farms[me],g=o.farms[1-me];return{day:+o.day,cash:+f.money,oppCash:+g.money,shops:o.town.unlocked_shops,market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL}}}\nasync function play(shops:string[],mode:string){prep();let s=E.state.initGameState(2,cfg,929292),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=compact(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,ours:s.farms[0].money,opp:s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.max(1,Math.ceil(a.length/2))))};\nconst rows:any[]=[];for(let oi=0;oi();for(const m of modes){const vals=[];for(const iw of inners){const r=await play(iw,m);vals.push(r.margin);preSet.add(JSON.stringify(r.pre))}shadow[m]={mean:avg(vals),cvar50:cvar50(vals)}}const meanPick=[...modes].sort((a,b)=>shadow[b].mean-shadow[a].mean)[0],riskPick=[...modes].sort((a,b)=>(shadow[b].mean+.25*shadow[b].cvar50)-(shadow[a].mean+.25*shadow[a].cvar50))[0];const actualRuns:any={};for(const m of modes)actualRuns[m]=await play(actual,m);const oracle=[...modes].sort((a,b)=>actualRuns[b].margin-actualRuns[a].margin)[0],base=actualRuns.N.margin;rows.push({outer:oi,prefix,hiddenSuffix:actual.slice(prefix.length),preStateIdentical:preSet.size===1,shadow,meanPick,riskPick,oracle,actualDeltas:Object.fromEntries(modes.map(m=>[m,actualRuns[m].margin-base])),meanRealizedDelta:actualRuns[meanPick].margin-base,riskRealizedDelta:actualRuns[riskPick].margin-base,oracleDelta:actualRuns[oracle].margin-base,meanRegret:actualRuns[oracle].margin-actualRuns[meanPick].margin,riskRegret:actualRuns[oracle].margin-actualRuns[riskPick].margin})}\nconst s=(k:string)=>avg(rows.map(r=>r[k]));console.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_BM_MPC_NESTED_MONTE_CARLO_BATCH1',opponent:'broker_bea.py',decisionDay,outerWorlds:OUTER,innerScenarios:INNER,aggregate:{meanSelectorRealizedDelta:s('meanRealizedDelta'),riskSelectorRealizedDelta:s('riskRealizedDelta'),meanRegret:s('meanRegret'),riskRegret:s('riskRegret'),meanOracleDelta:s('oracleDelta'),meanPickBetterThanN:rows.filter(r=>r.meanRealizedDelta>0).length,riskPickBetterThanN:rows.filter(r=>r.riskRealizedDelta>0).length},rows},null,2));",[],[1879],"6c40c1d3-5949-44aa-935c-6a12712bc3d7","kag_craf_bm_mpc_d9_diagnostic_v1.ts","01a07a21-cdcb-769f-b51f-34058e2d2352",258,["D",1788755458361],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\n\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES];\nconst cfg=E.state.resolveConfig({seed:989899,weedSpawnChance:0,townShopUnlockInterval:9999});\nconst actual=['SMOOTHIE_SHOP','PET_CAFE','ICE_CREAM_SHOP','BAKERY','FARMERS_MARKET','PIZZA_SHOP','BAKERY','BRUNCH_SPOT'];\nconst switchDays=[9],modes=['N','H','W','B'],SCEN=12;\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}\nfunction scenarios(day:number){const k=Math.floor(day/3),prefix=actual.slice(0,k),out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction compact(o:any){const me=+o.player,f=o.farms[me],g=o.farms[1-me];return{day:+o.day,cash:+f.money,oppCash:+g.money,shops:o.town.unlocked_shops,market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL}}}\nasync function play(shops:string[],switchDay:number,mode:string){prep();let s=E.state.initGameState(2,cfg,989899),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null;while(!s.done){s.town.unlocked_shops=fixed(+s.day,shops);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=switchDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===switchDay)pre=compact(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,ours:s.farms[0].money,opp:s.farms[1].money}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),worst=(x:number[])=>[...x].sort((a,b)=>a-b),cvar33=(x:number[])=>avg(worst(x).slice(0,Math.max(1,Math.ceil(x.length/3))));\nconst result:any={};for(const d of switchDays){const sc=scenarios(d),shadow:any={},valsByMode:any={};const preHashes=new Set();for(const m of modes){const vals:any[]=[];for(const s of sc){const r=await play(s,d,m);vals.push(r);preHashes.add(JSON.stringify(r.pre))}valsByMode[m]=vals;shadow[m]={mean:avg(vals.map(v=>v.margin)),cvar33:cvar33(vals.map(v=>v.margin)),min:Math.min(...vals.map(v=>v.margin)),max:Math.max(...vals.map(v=>v.margin))}}const perScenario=sc.map((s,i)=>{const ms=Object.fromEntries(modes.map(m=>[m,valsByMode[m][i].margin])),n=ms.N as number,best=[...modes].sort((a,b)=>(ms[b] as number)-(ms[a] as number))[0];return{suffix:s.slice(Math.floor(d/3)),best,deltas:Object.fromEntries(modes.map(m=>[m,(ms[m] as number)-n]))}});const actualRuns:any={};for(const m of modes)actualRuns[m]=await play(actual,d,m);const oracle=[...modes].sort((a,b)=>actualRuns[b].margin-actualRuns[a].margin)[0],meanPick=[...modes].sort((a,b)=>shadow[b].mean-shadow[a].mean)[0],riskPick=[...modes].sort((a,b)=>(shadow[b].mean+.35*shadow[b].cvar33)-(shadow[a].mean+.35*shadow[a].cvar33))[0];result[d]={observedPrefix:actual.slice(0,Math.floor(d/3)),preStateIdenticalAcrossShadow:preHashes.size===1,shadow,perScenario,meanPick,riskPick,actual:{suffix:actual.slice(Math.floor(d/3)),...Object.fromEntries(modes.map(m=>[m,{margin:actualRuns[m].margin,delta:actualRuns[m].margin-actualRuns.N.margin}]))},oracle,meanHit:meanPick===oracle,riskHit:riskPick===oracle}}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'CRAF_BM_MPC_SCENARIO_TERMINAL_STAGE2',opponent:'broker_bea.py',actualWorld:actual,scenarioCount:SCEN,decision:'terminal shadow over sampled future shops conditioned on observed prefix; no access to actual suffix',result},null,2));",[],[1887],"e80b12d3-5962-4493-aacc-8cea6778b9a1","kag_coupled_liquidity_selector_probe_v1.ts","6f61b38a-aa5d-11f1-a6d7-1607ee4eb77e",142,["D",1788745370239],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={BASE16:{d:16,credit:false},D12:{d:12,credit:false},D8:{d:8,credit:false},D6:{d:6,credit:false},D4:{d:4,credit:false},D6_STOCK:{d:6,credit:true}};const refs=['rancher_rita.py','broker_bea.py'],seeds=[27011,27127,27239];\nconst E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\\ndef prep(d):\\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['feed_float_days']=int(d)\\ndef seth(c,s,o):\\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}\nasync function play(v:string,ref:string,seed:number,seat:0|1){const vc=variants[v];prep(vc.d);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0,th:any={},animalBuy=0,wheatBuy=0;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){for(const k of [2,4,8,12,14,16])if(th[k]===undefined&&flock>=k)th[k]=+o.day;const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0,liquidityFeedFloatDays:vc.d,liquidityStockCredit:vc.credit});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}const a=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg);for(const q of a.market??[]){if(q?.[0]==='BUY_ANIMAL')animalBuy+=Number(q[2]??0);if(q?.[0]==='BUY_PRODUCT'&&q?.[1]==='WHEAT')wheatBuy+=Number(q[2]??0)}aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,maxFlock,finalFlock,animalLoss:maxFlock-finalFlock,th,animalBuy,wheatBuy}}\nconst games:any[]=[];for(const v of Object.keys(variants))for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const sum:any={};for(const v of Object.keys(variants)){const x=games.filter(g=>g.v===v);sum[v]={episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),avgAnimalBuy:avg(x.map(g=>g.animalBuy)),avgWheatBuy:avg(x.map(g=>g.wheatBuy)),reach8Day:avg(x.filter(g=>g.th[8]!=null).map(g=>g.th[8])),reach12Share:x.filter(g=>g.th[12]!=null).length/x.length,reach12Day:avg(x.filter(g=>g.th[12]!=null).map(g=>g.th[12])),reach14Share:x.filter(g=>g.th[14]!=null).length/x.length,reach14Day:avg(x.filter(g=>g.th[14]!=null).map(g=>g.th[14])),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);sum[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin)),avgAnimalLoss:avg(y.map(g=>g.animalLoss))}}}const b=games.filter(g=>g.v==='BASE16');const paired=Object.keys(variants).slice(1).map(v=>{const x=games.filter(g=>g.v===v);return{v,bankDelta:avg(x.map((g,i)=>g.ours-b[i].ours)),oppDelta:avg(x.map((g,i)=>g.theirs-b[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-b[i].margin)),better:x.filter((g,i)=>g.margin>b[i].margin).length,worse:x.filter((g,i)=>g.marging.win&&!b[i].win).length,lossFlips:x.filter((g,i)=>!g.win&&b[i].win).length}});console.log(JSON.stringify({status:'PASS',rootCause:'selector previously hardcoded 16-day liquidity gate, so scheduler-only reserve mutations could not materially change herd targets',hypothesis:'COUPLE_INVESTMENT_SELECTOR_AND_EXECUTION_SCHEDULER_TO_THE_SAME_LIQUIDITY_HORIZON',design:variants,summary:sum,paired},null,2));",[],[1895],"7512b87a-1371-49e6-ad77-1a5cc4e72ff0","kag_condfork_voc_boundary_v1.ts","01a07ab0-cd83-7517-bbfc-75b9f158b520",474,["D",1788766273434],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=24;\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const PREFIX=['ICE_CREAM_SHOP','YARN_STORE','FARMERS_MARKET'];function innerWorlds(){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(prefix:string[],mode:string,futureSeed:number){prep();const baseSeed=994949,cfg=E.state.resolveConfig({seed:baseSeed,weedSpawnChance:0.005,townShopUnlockInterval:3});let s=E.state.initGameState(2,cfg,baseSeed),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null,forked=false;while(!s.done){if(+s.day<=decisionDay)s.town.unlocked_shops=fixed(+s.day,prefix);if(!forked&&+s.day===decisionDay&&+s.hour===0){s.seed=futureSeed;forked=true}const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,finalShops:s.town.unlocked_shops}}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1),sd=(x:number[])=>{const m=avg(x);return Math.sqrt(avg(x.map(v=>(v-m)**2)))},cvar50=(x:number[])=>{const a=[...x].sort((u,v)=>u-v);return avg(a.slice(0,Math.ceil(a.length/2)))},tcrit:any={4:3.182,8:2.365,12:2.201,16:2.131,24:2.069};function stat(vals:any,m:string,n:number){const q=vals[m].slice(0,n),d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n);return{delta:mu,se,lcb:mu-tcrit[n]*se,ucb:mu+tcrit[n]*se}}\nfunction teacher(vals:any){let prev:any=null;const audit:any[]=[];for(const n of [8,12,16,24]){const st:any={};for(const m of modes)if(vals[m].length>=n)st[m]=stat(vals,m,n);const pos=Object.keys(st).filter(m=>m!=='N'&&st[m].lcb>0),pers=prev?pos.filter(m=>prev[m]?.lcb>0).map(m=>({m,r:Math.min(prev[m].lcb,st[m].lcb)})).sort((a,b)=>b.r-a.r):[];audit.push({n,pos,pers,available:Object.keys(st)});if(pers.length)return{pick:pers[0].m,used:n,audit};prev=st}return{pick:'N',used:24,audit}}\nconst vals:any={N:[],H:[],W:[],B:[]},preHashes=new Set(),scenarioShops:any[]=[],active=new Set(['H','W','B']),raceAudit:any[]=[];for(let i=0;i{const n=vals[m].length,q=vals[m],d=m==='N'?q.map(()=>0):q.map((v:number,i:number)=>v-vals.N[i]),mu=avg(d),se=m==='N'?0:sd(d)/Math.sqrt(n),tc=tcrit[n]??2.069;return{action:m,sampleN:n,qMean:avg(q),qSd:sd(q),deltaVsN:mu,lcb95:m==='N'?0:mu-tc*se,ucb95:m==='N'?0:mu+tc*se,cvar50:cvar50(q),positiveRate:m==='N'?0:d.filter((x:number)=>x>0).length/n,pruned:ns+r.sampleN,0),fullEpisodes=INNER*modes.length;\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'BROKER_BOUNDARY_CONDITIONAL_VOC24_V1',prefix:PREFIX,basePastSeed:994949,officialWeedSpawnChance:0.005,officialFutureShopUnlock:true,futureSeedForkAt:{day:9,hour:0},pairedAcrossActions:true,innerScenarios:INNER,preStateHashes:preHashes.size,preStateIdenticalAcrossAllForks:preHashes.size===1,teacherPick:t.pick,teacherAudit:t.audit,raceAudit,episodesUsed,fullEpisodes,savingsPct:100*(1-episodesUsed/fullEpisodes),rows,scenarioShops},null,2));",[],[1903],"8507a6e8-1622-4eed-aa5d-94acc851125e","kag_condfork_dual_strongh_v2.ts","01a07acd-e9e8-71da-a652-43809d04dbd5",485,["D",1788766728947],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';import { summarizeAction,chooseDualObjective } from './lib/kag_teacher_stats_v2.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=32;\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const PREFIX=['PIZZA_SHOP','PET_CAFE','ICE_CREAM_SHOP'];function innerWorlds(){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(prefix:string[],mode:string,futureSeed:number){prep();const baseSeed=994949,cfg=E.state.resolveConfig({seed:baseSeed,weedSpawnChance:0.005,townShopUnlockInterval:3});let s=E.state.initGameState(2,cfg,baseSeed),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null,forked=false;while(!s.done){if(+s.day<=decisionDay)s.town.unlocked_shops=fixed(+s.day,prefix);if(!forked&&+s.day===decisionDay&&+s.hour===0){s.seed=futureSeed;forked=true}const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,finalShops:s.town.unlocked_shops}}\n// Fixed-N confirmatory statistics use sample SD + paired t interval and an exact one-sided sign gate from kag_teacher_stats_v2.ts.\nconst vals:any={N:[],H:[],W:[],B:[]},preHashes=new Set(),scenarioShops:any[]=[];for(let i=0;isummarizeAction(m,vals[m],vals.N));const teacher=chooseDualObjective(rows);\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'BROKER_STRONGH_DUAL_OBJECTIVE_FIXED32_HOLDOUT_V2',prefix:PREFIX,basePastSeed:994949,officialWeedSpawnChance:0.005,officialFutureShopUnlock:true,futureSeedForkAt:{day:9,hour:0},pairedAcrossActions:true,holdoutFutureSeeds:true,innerScenarios:INNER,preStateHashes:preHashes.size,preStateIdenticalAcrossAllForks:preHashes.size===1,statMethod:'SAMPLE_SD_FIXED_N_PAIRED_T_PLUS_EXACT_SIGN',teacher,rows,scenarioShops},null,2));",[],[1911],"7688a2ff-5e1f-43a1-affd-2059f9e62cc0","kag_condfork_dual_boundary_v2.ts","01a07acb-1a1b-75d5-96c3-072c6326a77d",482,["D",1788766585328],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';import { summarizeAction,chooseDualObjective } from './lib/kag_teacher_stats_v2.ts';\nconst E=await loadOfficialKaggricultureEngine(),NAMES=[...E.constants.SHOP_NAMES],decisionDay=9,modes=['N','H','W','B'],INNER=32;\nfunction rng(a:number){return()=>{let t=a+=0x6D2B79F5;t=Math.imul(t^t>>>15,t|1);t^=t+Math.imul(t^t>>>7,t|61);return((t^t>>>14)>>>0)/4294967296}}const PREFIX=['ICE_CREAM_SHOP','YARN_STORE','FARMERS_MARKET'];function innerWorlds(){const out:string[][]=[];for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction features(o:any){const me=+o.player;function tile(f:any){let crops=0,pastures=0,animals=0;for(const row of f.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f.unlocked_quadrants??[]).length,crops,pastures,animals}}return{day:+o.day,hour:+o.hour,shops:o.town.unlocked_shops,me:tile(o.farms[me]),opp:tile(o.farms[1-me]),market:{WHEAT:o.market.prices.WHEAT,MELON:o.market.prices.MELON,MILK:o.market.prices.MILK,WOOL:o.market.prices.WOOL,invW:o.market.inventory.WHEAT,invM:o.market.inventory.MELON,invMilk:o.market.inventory.MILK,invWool:o.market.inventory.WOOL},shed:{WHEAT:+(o.private?.shed?.WHEAT??0),MELON:+(o.private?.shed?.MELON??0),MILK:+(o.private?.shed?.MILK??0),WOOL:+(o.private?.shed?.WOOL??0),FERTILIZER:+(o.private?.shed?.FERTILIZER??0)}}}\nasync function play(prefix:string[],mode:string,futureSeed:number){prep();const baseSeed=994949,cfg=E.state.resolveConfig({seed:baseSeed,weedSpawnChance:0.005,townShopUnlockInterval:3});let s=E.state.initGameState(2,cfg,baseSeed),tg:any={COW:0,SHEEP:0},comm=false,pre:any=null,forked=false;while(!s.done){if(+s.day<=decisionDay)s.town.unlocked_shops=fixed(+s.day,prefix);if(!forked&&+s.day===decisionDay&&+s.hour===0){s.seed=futureSeed;forked=true}const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===0){const owned=own(o,'COW')+own(o,'SHEEP'),post=+o.day>=decisionDay;let hands=8,build=Math.max(4,Math.min(16,owned+2));if(post&&(mode==='H'||mode==='B'))hands=9;if(post&&(mode==='W'||mode==='B'))build=16;setops(hands,build);if(+o.hour===0){if(+o.day===decisionDay)pre=features(o);const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[0]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg))}else aa[1]=JSON.parse(String(cr(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}return{pre,margin:s.farms[0].money-s.farms[1].money,finalShops:s.town.unlocked_shops}}\n// Fixed-N confirmatory statistics use sample SD + paired t interval and an exact one-sided sign gate from kag_teacher_stats_v2.ts.\nconst vals:any={N:[],H:[],W:[],B:[]},preHashes=new Set(),scenarioShops:any[]=[];for(let i=0;isummarizeAction(m,vals[m],vals.N));const teacher=chooseDualObjective(rows);\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'BROKER_BOUNDARY_DUAL_OBJECTIVE_FIXED32_HOLDOUT_V2',prefix:PREFIX,basePastSeed:994949,officialWeedSpawnChance:0.005,officialFutureShopUnlock:true,futureSeedForkAt:{day:9,hour:0},pairedAcrossActions:true,holdoutFutureSeeds:true,innerScenarios:INNER,preStateHashes:preHashes.size,preStateIdenticalAcrossAllForks:preHashes.size===1,statMethod:'SAMPLE_SD_FIXED_N_PAIRED_T_PLUS_EXACT_SIGN',teacher,rows,scenarioShops},null,2));",[],[1919],"3063b2c8-c894-42ea-bc8c-ee2461bfffcc","kag_closed_loop_melon_screen_v1.ts","fc32b232-aa5d-11f1-89d8-1607ee4eb77e",144,["D",1788745606468],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={BASE:{melon:0,labor:'H8'},M20:{melon:.20,labor:'H8'},M30:{melon:.30,labor:'H8'},M40:{melon:.40,labor:'H8'},M30_RAMP:{melon:.30,labor:'RAMP'}};const refs=['melon_mateo.py','rancher_rita.py','broker_bea.py'],seeds=[29009,29123];\nconst E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();const r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\\ndef prep(m):\\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0}\\n if m>0:\\n _D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':1.0-m,'MELON':m};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['fertilize_crops']=['MELON'];_D['POLICY']['collect_fertilizer']=True;_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':120};_D['POLICY']['sell_chunk']=20\\ndef setstate(c,s,o,h):\\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o);_D['POLICY']['hands']=int(h)\\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),setstate=py.globals.get('setstate');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function hands(mode:string,flock:number){return mode==='RAMP'?(flock<6?8:flock<10?10:12):8}\nasync function play(v:string,ref:string,seed:number,seat:0|1){const vc=variants[v];prep(vc.melon);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0,ops:any={},sales:any={};while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);setstate(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')),hands(vc.labor,flock))}else setstate(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']),hands(vc.labor,flock))}const a=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg);for(const u of [a?.farmer,...(a?.hands??[])]){const k=u?.[0]??'NONE';ops[k]=(ops[k]??0)+1}for(const q of a.market??[])if(q?.[0]==='SELL')sales[q[1]]=(sales[q[1]]??0)+Number(q[2]??0);aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,animalLoss:maxFlock-finalFlock,ops,sales}}\nconst games:any[]=[];for(const v of Object.keys(variants))for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const sum:any={};for(const v of Object.keys(variants)){const x=games.filter(g=>g.v===v);sum[v]={episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),avgCollect:avg(x.map(g=>g.ops.COLLECT_FERTILIZER??0)),avgFertilize:avg(x.map(g=>g.ops.FERTILIZE??0)),avgPass:avg(x.map(g=>g.ops.PASS??0)),avgMelonSellAttempt:avg(x.map(g=>g.sales.MELON??0)),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);sum[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin))}}}const b=games.filter(g=>g.v==='BASE');const paired=Object.keys(variants).slice(1).map(v=>{const x=games.filter(g=>g.v===v);return{v,bankDelta:avg(x.map((g,i)=>g.ours-b[i].ours)),oppDelta:avg(x.map((g,i)=>g.theirs-b[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-b[i].margin)),better:x.filter((g,i)=>g.margin>b[i].margin).length,worse:x.filter((g,i)=>g.marging.win&&!b[i].win).length,lossFlips:x.filter((g,i)=>!g.win&&b[i].win).length}});console.log(JSON.stringify({status:'PASS',hypothesis:'CLOSED_LOOP_LIVESTOCK_TO_FERTILIZER_TO_PREMIUM_CROP_CAN_CREATE_A_SECOND_REVENUE_ENGINE_WITHOUT_COPYING_HIGH_TIER_TRACE',provenance:'Rita and Mateo generic policy features are MIT-authored; portfolio shares are independent experimental values',summary:sum,paired},null,2));",[],[1927],"71f148ec-cbaf-4cd8-969f-6050d35ada76","kag_cleo_policy_probe_v1.ts","dbeded8a-aa66-11f1-836e-1607ee4eb77e",185,["D",1788749417798],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nconst url='https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents';\nconst r=await fetch(url);if(!r.ok)throw new Error(`dataset ${r.status}`);const z=unzipSync(new Uint8Array(await r.arrayBuffer()));\nconst names=['broker_bea.py','ledger_lena.py','slotter_silas.py','closer_cleo.py'];\nconst py=await loadPyodide();\nconst out:any={};\nfor(const name of names){\n const src=strFromU8(z[name]);\n py.globals.set('src',src);\n const meta=JSON.parse(String(py.runPython(`\nimport ast,json,re\nm=ast.parse(src)\nfuncs=[n.name for n in m.body if isinstance(n,(ast.FunctionDef,ast.AsyncFunctionDef))]\nassigns=[]\nfor n in m.body:\n if isinstance(n,(ast.Assign,ast.AnnAssign)):\n try:\n target=ast.unparse(n.targets[0] if isinstance(n,ast.Assign) else n.target)\n value=ast.unparse(n.value)\n if any(k in target.upper() for k in ['POLICY','TARGET','PRICE','CROP','ANIMAL','SELL','BUY','FERT','LAND','HAND','FEED']): assigns.append({'target':target,'value':value[:1200]})\n except: pass\nns={}\nexec(src,ns)\npol=ns.get('POLICY',None)\njson.dumps({'functions':funcs,'assignments':assigns,'policy':pol},default=str)\n`)));\n const lines=src.split('\\n').map((text,i)=>({line:i+1,text})).filter(x=>/POLICY|crop_share|animal_target|sell_order|price_floor|FERTIL|BUY_PRODUCT|BUY_SEED|BUY_ANIMAL|UNLOCK|land|hands|hire|feed|strawberry|melon|wheat/i.test(x.text)).slice(0,140);\n out[name]={bytes:src.length,lines:src.split('\\n').length,...meta,keywordLines:lines};\n}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'PUBLIC_MIT_REFERENCE_SOURCE_POLICY_AUDIT',out},null,2));",[],[1935],"4148ba57-1cb7-4fd4-b967-b9487f782026","kag_cleo_exact_excerpt_v1.ts","f1ed26b4-aa66-11f1-81b0-1607ee4eb77e",187,["D",1788749454719],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');if(!r.ok)throw new Error(String(r.status));const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const src=strFromU8(z['closer_cleo.py']);\nconst py=await loadPyodide();py.globals.set('src',src);const ast=JSON.parse(String(py.runPython(`\nimport ast,json\nm=ast.parse(src)\nout=[]\nfor n in m.body:\n if isinstance(n,(ast.Assign,ast.AnnAssign)):\n try:\n t=ast.unparse(n.targets[0] if isinstance(n,ast.Assign) else n.target)\n v=ast.unparse(n.value)\n if len(v)<=1800: out.append({'target':t,'value':v})\n except: pass\njson.dumps({'doc':ast.get_docstring(m),'functions':[n.name for n in m.body if isinstance(n,(ast.FunctionDef,ast.AsyncFunctionDef))],'assignments':out})\n`)));\nconst L=src.split('\\n');const hit=/front.run|terminal|clone|signature|sell|wheat|strawberry|melon|fertiliz|price|cash|opponent|reserve|glut|demand|animal|cow|sheep|land|hire|feed/i;const picked:any[]=[];const seen=new Set();for(let i=0;i0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nlet s=E.state.initGameState(2,cfg,959595),tg:any={COW:0,SHEEP:0},comm=false;const acc:any={};while(!s.done){s.town.unlocked_shops=fixed(+s.day);const o0=observationFor(s,0),o1=observationFor(s,1);if(+o0.hour===0){const ow={COW:own(o0,'COW'),SHEEP:own(o0,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o0,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}let a0=addFert(o0,enforce(o0,JSON.parse(String(cm(JSON.stringify(o0)))),tg)),a1=JSON.parse(String(cb(JSON.stringify(o1))));const d=+s.day;const A=acc[d]??={our:{hands:[],work:0,pass:0,move:0,market:0},opp:{hands:[],work:0,pass:0,move:0,market:0}};for(const [k,a] of [['our',a0],['opp',a1]] as any){const x=actionStats(a);A[k].hands.push(x.hands);A[k].work+=x.work;A[k].pass+=x.pass;A[k].move+=x.move;A[k].market+=x.market}if(+s.hour===0){A.public={our:pub(s.farms[0]),opp:pub(s.farms[1]),prices:{WHEAT:s.market.prices.WHEAT,MELON:s.market.prices.MELON,STRAWBERRY:s.market.prices.STRAWBERRY,MILK:s.market.prices.MILK,WOOL:s.market.prices.WOOL}}}s=E.interpreter.step(s,[a0,a1],cfg)}\nconst avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const days=[0,1,3,6,9,12,15,18,21,24,27,29].map(d=>{const A=acc[d];return{day:d,our:A.public.our,opp:A.public.opp,prices:A.public.prices,ops:{our:{avgHands:avg(A.our.hands),work:A.our.work,pass:A.our.pass,move:A.our.move,marketOrders:A.our.market},opp:{avgHands:avg(A.opp.hands),work:A.opp.work,pass:A.opp.pass,move:A.opp.move,marketOrders:A.opp.market}}}});console.log(JSON.stringify({status:'PASS',evidenceState:'BLACK_BOX_CAPACITY_AUTOPSY_FIXED_WORLD_NO_TRACE',final:{our:s.farms[0].money,broker:s.farms[1].money,margin:s.farms[0].money-s.farms[1].money},days},null,2));",[],[1951],"4b50a8aa-9e4d-4fe3-8a65-552b9d9d5edf","kag_capacity_ablation_v1.ts","028e066c-aa69-11f1-8628-1607ee4eb77e",200,["D",1788750341620],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['BASE','H_RAMP','PASTURE_JIT','LAND4','COMBO'];const E=await loadOfficialKaggricultureEngine(),cfg=E.state.resolveConfig({seed:969696,weedSpawnChance:0,townShopUnlockInterval:9999});const shops=['SMOOTHIE_SHOP','FARMERS_MARKET','ICE_CREAM_SHOP','YARN_STORE','FARMERS_MARKET','BAKERY','SMOOTHIE_SHOP','BAKERY'];\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.globals.set('src',strFromU8(z['broker_bea.py']));py.runPython(`import json\\n_b={}\\nexec(src,_b)\\n_B=_b['agent']`);py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cb(p):return json.dumps(_B(json.loads(p)))\ndef cm(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep():\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setops(h,b,l):\n _D['POLICY']['hands']=int(h);_D['POLICY']['build'][0]['target']=int(b);_D['POLICY']['land_after_animals']=int(l)\n`);const cb=py.globals.get('cb'),cm=py.globals.get('cm'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setops=py.globals.get('setops');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function fixed(day:number){return shops.slice(0,Math.max(0,Math.min(8,Math.floor(day/3))))}function pub(f:any){let crops=0,pastures=0,animals=0;for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT')crops++;if(t.kind==='PASTURE')pastures++;if(t.animal)animals++}return{cash:+f.money,land:(f?.unlocked_quadrants??[]).length,crops,pastures,animals}}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nfunction ops(v:string,day:number,ownedAnimals:number){let hands=8,build=16,landAfter=10;if(v==='H_RAMP'||v==='COMBO')hands=day<3?4:day<6?6:day<9?8:10;if(v==='PASTURE_JIT'||v==='COMBO')build=Math.max(4,Math.min(16,ownedAnimals+2));if(v==='LAND4'||v==='COMBO')landAfter=4;return{hands,build,landAfter}}\nasync function play(v:string,seat:0|1){prep();let s=E.state.initGameState(2,cfg,969696),tg:any={COW:0,SHEEP:0},comm=false,minCash=Infinity,maxFlock=0,finalFlock=0;const days:any[]=[];while(!s.done){s.town.unlocked_shops=fixed(+s.day);const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===seat){const owned=own(o,'COW')+own(o,'SHEEP'),op=ops(v,+o.day,owned);setops(op.hands,op.build,op.landAfter);if(+o.hour===0){days.push({day:+o.day,...pub(o.farms[pid]),handsTarget:op.hands,buildTarget:op.build});const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[pid]=addFert(o,enforce(o,JSON.parse(String(cm(JSON.stringify(o)))),tg));minCash=Math.min(minCash,+o.farms[pid].money);maxFlock=Math.max(maxFlock,owned)}else aa[pid]=JSON.parse(String(cb(JSON.stringify(o))))}s=E.interpreter.step(s,aa,cfg)}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');return{v,seat,ours:s.farms[seat].money,theirs:s.farms[1-seat].money,margin:s.farms[seat].money-s.farms[1-seat].money,win:s.farms[seat].money>s.farms[1-seat].money,minCash,animalLoss:maxFlock-finalFlock,days}}\nconst games:any[]=[];for(const v of variants){games.push(await play(v,0));games.push(await play(v,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const b=games.filter(g=>g.v==='BASE'),summary:any={};for(const v of variants){const x=games.filter(g=>g.v===v),d=x.map((g,i)=>g.margin-b[i].margin);summary[v]={bank:avg(x.map(g=>g.ours)),opp:avg(x.map(g=>g.theirs)),margin:avg(x.map(g=>g.margin)),wins:x.filter(g=>g.win).length,minCash:Math.min(...x.map(g=>g.minCash)),animalLoss:avg(x.map(g=>g.animalLoss)),delta:avg(d),snapshots:[1,3,6,9,12,15,18].map(day=>{const r=x.map(g=>g.days.find((q:any)=>q.day===day)).filter(Boolean);return{day,cash:avg(r.map((q:any)=>q.cash)),land:avg(r.map((q:any)=>q.land)),crops:avg(r.map((q:any)=>q.crops)),pastures:avg(r.map((q:any)=>q.pastures)),animals:avg(r.map((q:any)=>q.animals)),handsTarget:avg(r.map((q:any)=>q.handsTarget)),buildTarget:avg(r.map((q:any)=>q.buildTarget))}})}}console.log(JSON.stringify({status:'PASS',evidenceState:'FIXED_WORLD_CAPACITY_ABLATION_V1',summary},null,2));",[],[1959],"074b06aa-d831-44b8-b6e0-0875b862c63b","kag_broker_feed_tactics_trace_v1.ts","7b610e48-aa5c-11f1-9786-1607ee4eb77e",133,["D",1788744960863],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig(),seed=17011,ourSeat:0|1=0;\nconst r=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await r.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);py.globals.set('src',strFromU8(z['broker_bea.py']));py.runPython(`_r={}\\nexec(src,_r)\\n_REG['B']=_r['agent']`);py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\\ndef cb(p):return json.dumps(_REG['B'](json.loads(p)))\\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\\ndef reset():\\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0}\\ndef seth(c,s,o):\\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\\n`);const cb=py.globals.get('cb'),cd=py.globals.get('cd'),reset=py.globals.get('reset'),seth=py.globals.get('seth');reset();\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}\nlet s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false;const days:any[]=[];let cur:any=null;while(!s.done){if(s.day>12)break;const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);if(pid===ourSeat){if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm){tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)};seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}}aa[pid]=enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg)}else{const a=JSON.parse(String(cb(JSON.stringify(o))));aa[pid]=a;if(+o.hour===0){const flock=own(o,'COW')+own(o,'SHEEP'),wheat=+(o.private?.shed?.WHEAT??0)+(o.private?.inventories??[]).reduce((a:number,i:any)=>a+Number(i?.WHEAT??0),0);cur={day:+o.day,morningCash:+o.farms[pid].money,flock,wheatOwned:wheat,daysOnHand:flock?wheat/flock:null,wheatPrice:+o.market.prices.WHEAT,marketOrders:0,buyWheat:0,buyAnimals:{},sell:{},hire:0,buyLand:0};days.push(cur)}for(const q of a.market??[]){cur.marketOrders++;if(q?.[0]==='BUY_PRODUCT'&&q?.[1]==='WHEAT')cur.buyWheat+=Number(q[2]??0);else if(q?.[0]==='BUY_ANIMAL')cur.buyAnimals[q[1]]=(cur.buyAnimals[q[1]]??0)+Number(q[2]??0);else if(q?.[0]==='SELL')cur.sell[q[1]]=(cur.sell[q[1]]??0)+Number(q[2]??0);else if(q?.[0]==='HIRE')cur.hire++;else if(q?.[0]==='BUY_LAND')cur.buyLand++}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}\nconsole.log(JSON.stringify({status:'PASS',evidenceState:'BLACK_BOX_ACTION_TRACE_NO_HIGH_TIER_CODE_REUSE',seed,days},null,2));",[],[1967],"8873835f-d30f-44d7-9c3f-cc4a58113b7a","kag_battle_dossier_migrate_v1.ts","221d72b0-aa55-11f1-bee3-1607ee4eb77e",101,["D",1788741804633],"import { sqlite } from 'https://esm.town/v/std/sqlite/main.ts';\nconst now=new Date().toISOString();\nawait sqlite.batch([\n {sql:`CREATE TABLE IF NOT EXISTS kag_battle_dossiers(\n dossier_id TEXT PRIMARY KEY,\n match_id TEXT NOT NULL,\n subject_id TEXT NOT NULL,\n opponent_id TEXT NOT NULL,\n subject_role TEXT NOT NULL,\n seed INTEGER NOT NULL,\n seat INTEGER NOT NULL,\n final_bank REAL NOT NULL,\n opponent_bank REAL NOT NULL,\n margin REAL NOT NULL,\n min_cash REAL,\n min_cash_day INTEGER,\n first_lead_day INTEGER,\n first_trail_day INTEGER,\n max_lead REAL,\n max_lead_day INTEGER,\n max_deficit REAL,\n max_deficit_day INTEGER,\n investment_fingerprint TEXT,\n revenue_mix TEXT,\n market_exposure TEXT,\n observed_strengths TEXT,\n observed_failures TEXT,\n causal_hypotheses TEXT,\n counterfactuals TEXT,\n evidence_state TEXT NOT NULL,\n created_at TEXT NOT NULL)`},\n {sql:`CREATE INDEX IF NOT EXISTS idx_kag_dossier_subject ON kag_battle_dossiers(subject_id,subject_role,created_at)`},\n {sql:`CREATE INDEX IF NOT EXISTS idx_kag_dossier_match ON kag_battle_dossiers(match_id,subject_role)`},\n {sql:`CREATE TABLE IF NOT EXISTS kag_battle_events(\n event_id TEXT PRIMARY KEY,\n match_id TEXT NOT NULL,\n turn INTEGER,\n day INTEGER,\n event_type TEXT NOT NULL,\n actor_id TEXT,\n payload TEXT,\n evidence_state TEXT NOT NULL,\n created_at TEXT NOT NULL)`},\n {sql:`CREATE INDEX IF NOT EXISTS idx_kag_battle_events_match ON kag_battle_events(match_id,day,turn)`},\n {sql:`CREATE TABLE IF NOT EXISTS kag_counterfactual_queue(\n hypothesis_id TEXT PRIMARY KEY,\n source_match_id TEXT NOT NULL,\n target_agent TEXT NOT NULL,\n hypothesis TEXT NOT NULL,\n mechanism TEXT,\n minimal_intervention TEXT,\n falsifier TEXT,\n expected_signal TEXT,\n priority REAL NOT NULL DEFAULT 0,\n status TEXT NOT NULL DEFAULT 'QUEUED',\n evidence_state TEXT NOT NULL DEFAULT 'UNVERIFIED',\n created_at TEXT NOT NULL,\n updated_at TEXT NOT NULL)`},\n]);\nconst tables=await sqlite.execute(`SELECT name FROM sqlite_master WHERE type='table' AND name IN ('kag_battle_dossiers','kag_battle_events','kag_counterfactual_queue') ORDER BY name`);\nconsole.log(JSON.stringify({status:'PASS',now,tables:tables.rows},null,2));",[],[1975],"6d4a2e56-2485-4674-803a-2736f308a8b1","kag_autolab_worker.ts","99b19090-aa96-11f1-9abf-1607ee4eb77e",502,["D",1788770060053],"interval","import { sqlite } from 'https://esm.town/v/std/sqlite/main.ts';\nimport { blob } from 'https://esm.town/v/std/blob/main.ts';\nimport { runAutolabJob, renderArtifactMarkdown, AUTOLAB_CORE_VERSION } from './lib/kag_autolab_core.ts';\n\nconst WORKER_VERSION='KAG_AUTOLAB_WORKER_V1';\nconst MAX_JOBS_PER_TICK=2;\nconst SOFT_BUDGET_MS=35000;\n\nasync function sha256(s:string){const b=await crypto.subtle.digest('SHA-256',new TextEncoder().encode(s));return Array.from(new Uint8Array(b)).map(x=>x.toString(16).padStart(2,'0')).join('')}\nfunction stamp(d=new Date()){return d.toISOString().replace(/[-:]/g,'').replace(/\\.\\d{3}Z$/,'Z')}\nasync function nextJob(){const q=await sqlite.execute(`SELECT * FROM kag_autolab_queue WHERE status='READY' AND (run_after IS NULL OR run_after<=CURRENT_TIMESTAMP) ORDER BY priority ASC, created_at ASC LIMIT 1`);return q.rows?.[0]??null}\nasync function claim(job:any){const r=await sqlite.execute({sql:`UPDATE kag_autolab_queue SET status='RUNNING',attempts=attempts+1,started_at=CURRENT_TIMESTAMP,updated_at=CURRENT_TIMESTAMP,last_error=NULL WHERE job_id=? AND status='READY'`,args:[job.job_id]});return Number(r.rowsAffected??0)===1}\nasync function persistSuccess(job:any,runId:string,started:number,result:any){const finished=new Date(),prefix=`artifacts/kaggriculture/autolab/${stamp(finished)}_${job.job_id}`;const json=JSON.stringify({worker:WORKER_VERSION,core:AUTOLAB_CORE_VERSION,job:{job_id:job.job_id,job_type:job.job_type,payload:JSON.parse(job.payload_json)},result},null,2);const md=renderArtifactMarkdown(job,result);const jhash=await sha256(json),mhash=await sha256(md);await blob.set(`${prefix}.json`,json);await blob.set(`${prefix}.md`,md);await sqlite.batch([\n {sql:`UPDATE kag_autolab_runs SET status='DONE',finished_at=CURRENT_TIMESTAMP,duration_ms=?,result_json=?,artifact_prefix=?,evidence_state='AUTOLAB_COMPUTE_RESULT_NOT_CHAMPION_PROOF' WHERE run_id=?`,args:[Date.now()-started,json,prefix,runId]},\n {sql:`UPDATE kag_autolab_queue SET status='DONE',finished_at=CURRENT_TIMESTAMP,updated_at=CURRENT_TIMESTAMP,evidence_state='AUTOLAB_COMPUTE_RESULT_NOT_CHAMPION_PROOF' WHERE job_id=?`,args:[job.job_id]},\n {sql:`INSERT OR REPLACE INTO kag_autolab_artifacts(artifact_id,run_id,job_id,artifact_type,artifact_key,content_type,content_text,sha256,evidence_state) VALUES(?,?,?,?,?,?,?,?,?)`,args:[crypto.randomUUID(),runId,job.job_id,'JSON',`${prefix}.json`,'application/json',json,jhash,'AUTOLAB_COMPUTE_RESULT_NOT_CHAMPION_PROOF']},\n {sql:`INSERT OR REPLACE INTO kag_autolab_artifacts(artifact_id,run_id,job_id,artifact_type,artifact_key,content_type,content_text,sha256,evidence_state) VALUES(?,?,?,?,?,?,?,?,?)`,args:[crypto.randomUUID(),runId,job.job_id,'MARKDOWN',`${prefix}.md`,'text/markdown',md,mhash,'AUTOLAB_COMPUTE_RESULT_NOT_CHAMPION_PROOF']},\n {sql:`INSERT OR REPLACE INTO kag_autolab_state(state_key,state_json,updated_at) VALUES('last_success',?,CURRENT_TIMESTAMP)`,args:[JSON.stringify({job_id:job.job_id,run_id:runId,artifact_prefix:prefix,finished_at:finished.toISOString()})]}\n ]);return{prefix,result}}\nasync function persistFailure(job:any,runId:string,started:number,error:any){const msg=String(error?.stack??error?.message??error).slice(0,12000);const nextStatus=(Number(job.attempts??0)+1)>=Number(job.max_attempts??3)?'FAILED':'READY';await sqlite.batch([\n {sql:`UPDATE kag_autolab_runs SET status='FAILED',finished_at=CURRENT_TIMESTAMP,duration_ms=?,error_text=?,evidence_state='EXECUTION_FAILED' WHERE run_id=?`,args:[Date.now()-started,msg,runId]},\n {sql:`UPDATE kag_autolab_queue SET status=?,run_after=CASE WHEN ?='READY' THEN datetime('now','+15 minutes') ELSE run_after END,updated_at=CURRENT_TIMESTAMP,finished_at=CASE WHEN ?='FAILED' THEN CURRENT_TIMESTAMP ELSE finished_at END,last_error=?,evidence_state=CASE WHEN ?='FAILED' THEN 'EXECUTION_FAILED_RETRY_EXHAUSTED' ELSE 'EXECUTION_FAILED_RETRY_PENDING' END WHERE job_id=?`,args:[nextStatus,nextStatus,nextStatus,msg,nextStatus,job.job_id]},\n {sql:`INSERT OR REPLACE INTO kag_autolab_state(state_key,state_json,updated_at) VALUES('last_failure',?,CURRENT_TIMESTAMP)`,args:[JSON.stringify({job_id:job.job_id,run_id:runId,error:msg.slice(0,2000),next_status:nextStatus})]}\n ]);return{error:msg,nextStatus}}\nasync function processOne(job:any){if(!(await claim(job)))return{status:'SKIPPED_CLAIM_RACE',job_id:job.job_id};const runId=crypto.randomUUID(),started=Date.now();await sqlite.execute({sql:`INSERT INTO kag_autolab_runs(run_id,job_id,job_type,status,started_at,worker_version,evidence_state) VALUES(?,?,?,'RUNNING',CURRENT_TIMESTAMP,?,'RUNNING_UNVERIFIED')`,args:[runId,job.job_id,job.job_type,WORKER_VERSION]});try{const result=await runAutolabJob(job.job_type,JSON.parse(job.payload_json));const saved=await persistSuccess(job,runId,started,result);return{status:'DONE',job_id:job.job_id,run_id:runId,artifact_prefix:saved.prefix,summary:result.kind==='PUBLIC_CROSSPLAY'?{a:result.a,b:result.b,episodes:result.episodes,winsA:result.winsA,winsB:result.winsB,ties:result.ties,avgMarginA:result.avgMarginA}:{variant:result.variant,ref:result.ref,episodes:result.episodes,wins:result.wins,losses:result.losses,ties:result.ties,avgMargin:result.avgMargin}}}catch(e){const f=await persistFailure(job,runId,started,e);return{status:'FAILED_ATTEMPT',job_id:job.job_id,run_id:runId,...f}}}\n\nexport default async function(interval:any){const tickStart=Date.now(),processed:any[]=[];while(processed.length={DYN:{animal_target:{COW:0,SHEEP:0}},C4S12:{animal_target:{COW:4,SHEEP:12}},C6S10:{animal_target:{COW:6,SHEEP:10}},RITA_BASE:{}};\nfor(const [name,patch] of Object.entries(patches)){py.globals.set('nm',name);py.globals.set('patch',JSON.stringify(patch));py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n\\n_BASE[nm]=copy.deepcopy(_n['POLICY'])`);}\npy.runPython(`\\ndef call(name,payload): return json.dumps(_REG[name]['agent'](json.loads(payload)))\\ndef set_herd(name,cows,sheep,order):\\n _REG[name]['POLICY']['animal_target']={'COW':int(cows),'SHEEP':int(sheep)}\\n _REG[name]['POLICY']['animals']=json.loads(order)\\ndef reset(name):\\n _REG[name]['POLICY'].clear();_REG[name]['POLICY'].update(copy.deepcopy(_BASE[name]))\\n`);\nconst call=py.globals.get('call'),setHerd=py.globals.get('set_herd'),reset=py.globals.get('reset');\nfunction countOwned(obs:any,animal:string){const me=Number(obs?.player??0),f=obs?.farms?.[me]??{};let n=Number(obs?.private?.shed?.[animal]??0);for(const inv of obs?.private?.inventories??[])n+=Number(inv?.[animal]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===animal)n++;return n;}\nfunction enforce(obs:any,action:any,targets:{COW:number;SHEEP:number}){const actual:any={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};const out={...action,market:[] as any[]};for(const order of action?.market??[]){if(Array.isArray(order)&&order[0]==='BUY_ANIMAL'&&(order[1]==='COW'||order[1]==='SHEEP')){const a=order[1] as 'COW'|'SHEEP';const allowed=Math.max(0,Math.min(Number(order[2]??1),targets[a]-actual[a]));if(allowed>0){out.market.push(['BUY_ANIMAL',a,allowed]);actual[a]+=allowed;}} else out.market.push(order);}return out;}\nasync function play(sw:number,opp:string,seed:number,dynSeat:0|1){reset('DYN');reset(opp);let s=E.state.initGameState(2,{...cfg0,seed},seed);let targets={COW:0,SHEEP:0},errors=0;const choices:any[]=[];while(!s.done){const acts:any[]=[];for(const pid of [0,1] as const){const n=pid===dynSeat?'DYN':opp;const obs=observationFor(s,pid);if(n==='DYN'&&Number(obs.hour)===0){const owned={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};targets={...owned};const total=owned.COW+owned.SHEEP;const choice=choosePastureAnimal(obs,{cashBuffer:400,strategicWeight:sw});if(total<16&&choice.choice){const batch=Math.min(2,16-total);targets[choice.choice]+=batch;}const order=[...choice.ranking].map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP');setHerd('DYN',targets.COW,targets.SHEEP,JSON.stringify(order));choices.push({day:obs.day,choice:choice.choice,top:choice.ranking[0]?.animal,topNet:Math.round(choice.ranking[0]?.net??0),rivalImpact:Math.round(choice.ranking[0]?.rivalPriceImpact??0)});}try{acts[pid]=JSON.parse(String(call(n,JSON.stringify(obs))));if(n==='DYN')acts[pid]=enforce(obs,acts[pid],targets);}catch{errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]};}}s=E.interpreter.step(s,acts,{...cfg0,seed});}const ours=s.farms[dynSeat].money,theirs=s.farms[1-dynSeat].money;return{sw,opp,seed,dynSeat,ours,theirs,margin:ours-theirs,win:ours>theirs,errors,firstChoices:choices.slice(0,6)};}\nconst games:any[]=[];for(const sw of weights)for(const opp of opponents)for(const seed of seeds){games.push(await play(sw,opp,seed,0));games.push(await play(sw,opp,seed,1));}\nconst summary:any={};for(const sw of weights){const rows=games.filter(g=>g.sw===sw);summary[String(sw)]={overall:{episodes:rows.length,wins:rows.filter(r=>r.win).length,winRate:rows.filter(r=>r.win).length/rows.length,avgBank:rows.reduce((a,r)=>a+r.ours,0)/rows.length,avgMargin:rows.reduce((a,r)=>a+r.margin,0)/rows.length,errors:rows.reduce((a,r)=>a+r.errors,0)},byOpponent:{}};for(const opp of opponents){const rs=rows.filter(r=>r.opp===opp);summary[String(sw)].byOpponent[opp]={episodes:rs.length,wins:rs.filter(r=>r.win).length,winRate:rs.filter(r=>r.win).length/rs.length,avgBank:rs.reduce((a,r)=>a+r.ours,0)/rs.length,avgMargin:rs.reduce((a,r)=>a+r.margin,0)/rs.length};}}\nconsole.log(JSON.stringify({status:'PASS',hypothesis:'VALUING_RIVAL_PRICE_DAMAGE_CAN_IMPROVE_RELATIVE_OUTCOME_BUT_TOO_MUCH_WEIGHT_MAY_DESTROY_OWN_VALUE',weights,seeds,summary,pairedDeltas:weights.slice(1).map(sw=>({sw,vs0:games.filter(g=>g.sw===sw).reduce((s,g,i)=>s+(g.margin-games.filter(x=>x.sw===0)[i].margin),0)/games.filter(g=>g.sw===sw).length}))},null,2));",[],[2000],"501fa9ba-d087-4d6d-801d-0742d22bb623","kag_adaptive_crop_router_screen_v1.ts","1619f1d0-aa61-11f1-bf71-1607ee4eb77e",163,["D",1788746938423],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants=['M50','M60','M70','ROUTER_SIMPLE','ROUTER_CASH'];const refs=['melon_mateo.py','rancher_rita.py','broker_bea.py','ledger_lena.py','slotter_silas.py','closer_cleo.py'],seeds=[40009,40123];const E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();\nconst rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const py=await loadPyodide();py.runPython(`import json\\n_REG={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}py.globals.set('src',strFromU8(z['rancher_rita.py']));py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=json.loads(json.dumps(_d['POLICY']))`);py.runPython(`\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\ndef cd(p):return json.dumps(_D['agent'](json.loads(p)))\ndef prep(m):\n _D['POLICY'].clear();_D['POLICY'].update(json.loads(json.dumps(_BASE)));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0};_D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':1.0-m,'MELON':m};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12;_D['POLICY']['collect_fertilizer']=True\ndef seth(c,s,o):\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)\ndef setcrop(m): _D['POLICY']['crop_share']={'WHEAT':1.0-m,'MELON':m}\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth'),setcrop=py.globals.get('setcrop');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function farmFeatures(f:any){let melon=0,wheat=0,animals=0;for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT'&&t.crop==='MELON')melon++;if(t.kind==='PLANT'&&t.crop==='WHEAT')wheat++;if(t.animal)animals++}return{melon,wheat,animals,money:+f.money}}\nfunction route(v:string,o:any){if(v==='M50')return .5;if(v==='M60')return .6;if(v==='M70')return .7;const me=+o.player,f=farmFeatures(o.farms[1-me]);if(f.melon>0&&f.animals===0)return .5;if(f.melon===0&&f.wheat>0)return .7;if(f.melon>0&&f.animals>0){if(v==='ROUTER_CASH'&&f.money<30)return .7;return .6}return .5}\nfunction enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}function addFert(o:any,a:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);return q>0&&p>=60&&(a.market?.length??0)<10?{...a,market:[...(a.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]}:a}\nasync function play(v:string,ref:string,seed:number,seat:0|1){prep(.5);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0,routeCounts:any={};while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const share=route(v,o);setcrop(share);routeCounts[String(share)]=(routeCounts[String(share)]??0)+1;const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}aa[pid]=addFert(o,enforce(o,JSON.parse(String(cd(JSON.stringify(o)))),tg));minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,animalLoss:maxFlock-finalFlock,routeCounts}}\nconst games:any[]=[];for(const v of variants)for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const summary:any={};for(const v of variants){const x=games.filter(g=>g.v===v);summary[v]={episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),errors:x.reduce((a,g)=>a+g.err,0),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);summary[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin)),routeCounts:y[0]?.routeCounts}}}console.log(JSON.stringify({status:'PASS',question:'CAN_EARLY_PUBLIC_OPPONENT_SIGNATURES_ROUTE_CROP_CAPITAL_BETTER_THAN_ONE_STATIC_MELON_SHARE',routerRules:{CROP_SPECIALIST:.5,LIVESTOCK_SPECIALIST:.7,MIXED_TOP:.6,DEEP_CASH_COMMIT_MIXED:.7},summary},null,2));",[],[2008],"6ea9880f-6b1c-4116-b408-0728001f9122","kag_active_constraint_probe_v1.ts","d005d360-aa4e-11f1-bec9-1607ee4eb77e",84,["D",1788739089926],"import { unzipSync,strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst seeds=[9403,9419,9431,9461,9491],opponents=['C4S12','C6S10','RITA_BASE'];const E=await loadOfficialKaggricultureEngine();const cfg0=E.state.resolveConfig();\nconst zr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await zr.arrayBuffer()));const src=strFromU8(z['rancher_rita.py']);const py=await loadPyodide();py.runPython(`import json,copy\\n_REG={}\\n_BASE={}`);py.globals.set('src',src);const patches:any={DYN:{animal_target:{COW:0,SHEEP:0}},C4S12:{animal_target:{COW:4,SHEEP:12}},C6S10:{animal_target:{COW:6,SHEEP:10}},RITA_BASE:{}};for(const [n,p] of Object.entries(patches)){py.globals.set('nm',n);py.globals.set('patch',JSON.stringify(p));py.runPython(`_n={}\\nexec(src,_n)\\nfor _k,_v in json.loads(patch).items(): _n['POLICY'][_k]=_v\\n_REG[nm]=_n\\n_BASE[nm]=copy.deepcopy(_n['POLICY'])`);}py.runPython(`\\ndef call(name,payload): return json.dumps(_REG[name]['agent'](json.loads(payload)))\\ndef set_herd(name,cows,sheep,order):\\n _REG[name]['POLICY']['animal_target']={'COW':int(cows),'SHEEP':int(sheep)}\\n _REG[name]['POLICY']['animals']=json.loads(order)\\ndef reset(name):\\n _REG[name]['POLICY'].clear();_REG[name]['POLICY'].update(copy.deepcopy(_BASE[name]))\\n`);const call=py.globals.get('call'),setHerd=py.globals.get('set_herd'),reset=py.globals.get('reset');\nfunction countOwned(obs:any,a:string){const me=+obs.player,f=obs.farms[me];let n=+(obs.private?.shed?.[a]??0);for(const inv of obs.private?.inventories??[])n+=+(inv?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n;}\nfunction enforce(obs:any,act:any,t:any){const x:any={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};const o={...act,market:[] as any[]};for(const m of act?.market??[]){if(Array.isArray(m)&&m[0]==='BUY_ANIMAL'&&(m[1]==='COW'||m[1]==='SHEEP')){const a=m[1],q=Math.max(0,Math.min(+m[2]||1,t[a]-x[a]));if(q>0){o.market.push(['BUY_ANIMAL',a,q]);x[a]+=q;}}else o.market.push(m);}return o;}\nasync function play(opp:string,seed:number){reset('DYN');reset(opp);let s=E.state.initGameState(2,{...cfg0,seed},seed),targets:any={COW:0,SHEEP:0};const d:any[]=[];while(!s.done){const acts:any[]=[];for(const pid of [0,1] as const){const n=pid===0?'DYN':opp,obs=observationFor(s,pid);if(n==='DYN'&&+obs.hour===0){const owned={COW:countOwned(obs,'COW'),SHEEP:countOwned(obs,'SHEEP')};targets={...owned};const c=choosePastureAnimal(obs,{cashBuffer:400,strategicWeight:0});const top=c.ranking[0];const total=owned.COW+owned.SHEEP;let reason='CHOICE';if(total>=16)reason='HERD_CAP';else if(!top.financeable)reason='LIQUIDITY';else if(top.net<=0)reason='NONPOSITIVE_NET';else if(!c.choice)reason='NO_CHOICE';if(total<16&&c.choice){targets[c.choice]+=Math.min(2,16-total);}setHerd('DYN',targets.COW,targets.SHEEP,JSON.stringify(c.ranking.map(x=>x.animal).filter(x=>x==='COW'||x==='SHEEP')));d.push({day:obs.day,reason,cash:obs.farms[0].money,knownShops:(obs.town?.unlocked_shops??[]).length,owned,total,choice:c.choice,top:top.animal,net:Math.round(top.net),requiredLiquidity:Math.round(top.requiredLiquidity),financeable:top.financeable});}acts[pid]=JSON.parse(String(call(n,JSON.stringify(obs))));if(n==='DYN')acts[pid]=enforce(obs,acts[pid],targets);}s=E.interpreter.step(s,acts,{...cfg0,seed});}return{opp,seed,decisions:d};}\nconst runs:any[]=[];for(const o of opponents)for(const seed of seeds)runs.push(await play(o,seed));const all=runs.flatMap(r=>r.decisions.map((d:any)=>({...d,opp:r.opp,seed:r.seed})));const counts:any={};for(const d of all)counts[d.reason]=(counts[d.reason]??0)+1;const byDay:any={};for(const d of all){const k=String(d.day);byDay[k]=byDay[k]??{};byDay[k][d.reason]=(byDay[k][d.reason]??0)+1;}console.log(JSON.stringify({status:'PASS',dailyDecisions:all.length,constraintCounts:counts,byDay,examples:all.filter(d=>d.reason!=='CHOICE').slice(0,20)},null,2));",[],[2016],"8a59e971-c8bf-47b1-b9f6-c9904e2bb2c5","kag_acquisition_prefeed_probe_v1.ts","db63acce-aa5c-11f1-b329-1607ee4eb77e",135,["D",1788745121935],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={BASE16:{float:16,pf:0},ACQPF4_F4:{float:4,pf:4},ACQPF4_F2:{float:2,pf:4},ACQPF3_F3:{float:3,pf:3},ACQPF2_F4:{float:4,pf:2}};const refs=['rancher_rita.py','broker_bea.py'],seeds=[25013,25117,25229];\nconst E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();const rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const baseSrc=strFromU8(z['rancher_rita.py']);const old=` float_days = policy.get(\"feed_float_days\", 0)\\n wheat_price = prices.get(\"WHEAT\", 25)\\n n = 0\\n for candidate in range(min(gap, policy.get(\"animal_batch\", 2)), 0, -1):\\n feed_float = (flock + candidate) * wheat_price * float_days\\n if money >= candidate * info[\"cost\"] + buffer + feed_float:\\n n = candidate\\n break\\n if n > 0 and spend(n * info[\"cost\"]):\\n orders.append([\"BUY_ANIMAL\", animal, n])`;\nconst repl=` float_days = policy.get(\"feed_float_days\", 0)\\n prefeed_days = policy.get(\"acquisition_prefeed_days\", 0)\\n wheat_price = prices.get(\"WHEAT\", 25)\\n n = 0\\n planned_feed = 0\\n for candidate in range(min(gap, policy.get(\"animal_batch\", 2)), 0, -1):\\n projected_flock = flock + candidate\\n prefeed_need = max(0, projected_flock * prefeed_days - shed.get(\"WHEAT\", 0) - carried_wheat)\\n prefeed_cost = prefeed_need * wheat_price\\n feed_float = projected_flock * wheat_price * float_days\\n if wheat_price <= policy.get(\"max_wheat_price\", 60) and money >= candidate * info[\"cost\"] + buffer + feed_float + prefeed_cost:\\n n = candidate\\n planned_feed = prefeed_need\\n break\\n if n > 0:\\n if planned_feed > 0 and spend(planned_feed * wheat_price):\\n orders.append([\"BUY_PRODUCT\", \"WHEAT\", planned_feed])\\n if spend(n * info[\"cost\"]):\\n orders.append([\"BUY_ANIMAL\", animal, n])`;\nif(!baseSrc.includes(old))throw new Error('mutation target missing');const mutated=baseSrc.replace(old,repl);\nconst py=await loadPyodide();py.runPython(`import json\\n_REG={}\\n_D={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}for(const v of Object.keys(variants)){py.globals.set('nm',v);py.globals.set('src',v==='BASE16'?baseSrc:mutated);py.runPython(`_n={}\\nexec(src,_n)\\n_D[nm]=_n`)}py.runPython(`\\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\\ndef cd(n,p):return json.dumps(_D[n]['agent'](json.loads(p)))\\ndef init():\\n for n,d in _D.items(): d['_BASE_POLICY']=json.loads(json.dumps(d['POLICY']))\\ndef prep(n,ff,pf):\\n d=_D[n];d['POLICY'].clear();d['POLICY'].update(json.loads(json.dumps(d['_BASE_POLICY'])));d['POLICY']['animal_target']={'COW':0,'SHEEP':0};d['POLICY']['feed_float_days']=int(ff);d['POLICY']['acquisition_prefeed_days']=int(pf)\\ndef seth(n,c,s,o):\\n d=_D[n];d['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};d['POLICY']['animals']=json.loads(o)\\ninit()\\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}\nasync function play(v:string,ref:string,seed:number,seat:0|1){const vc=variants[v];prep(v,vc.float,vc.pf);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0,th:any={},wheatBuy=0,animalBuy=0;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){for(const k of [2,4,8,12,14,16])if(th[k]===undefined&&flock>=k)th[k]=+o.day;const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(v,tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(v,tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}const a=enforce(o,JSON.parse(String(cd(v,JSON.stringify(o)))),tg);for(const q of a.market??[]){if(q?.[0]==='BUY_PRODUCT'&&q?.[1]==='WHEAT')wheatBuy+=Number(q[2]??0);if(q?.[0]==='BUY_ANIMAL')animalBuy+=Number(q[2]??0)}aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,maxFlock,finalFlock,animalLoss:maxFlock-finalFlock,th,wheatBuy,animalBuy}}\nconst games:any[]=[];for(const v of Object.keys(variants))for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const sum:any={};for(const v of Object.keys(variants)){const x=games.filter(g=>g.v===v);sum[v]={episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),avgWheatBuy:avg(x.map(g=>g.wheatBuy)),avgAnimalBuy:avg(x.map(g=>g.animalBuy)),reach8Day:avg(x.filter(g=>g.th[8]!=null).map(g=>g.th[8])),reach14Share:x.filter(g=>g.th[14]!=null).length/x.length,reach14Day:avg(x.filter(g=>g.th[14]!=null).map(g=>g.th[14])),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);sum[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin)),avgAnimalLoss:avg(y.map(g=>g.animalLoss))}}}const b=games.filter(g=>g.v==='BASE16');const paired=Object.keys(variants).slice(1).map(v=>{const x=games.filter(g=>g.v===v);return{v,bankDelta:avg(x.map((g,i)=>g.ours-b[i].ours)),oppDelta:avg(x.map((g,i)=>g.theirs-b[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-b[i].margin)),better:x.filter((g,i)=>g.margin>b[i].margin).length,worse:x.filter((g,i)=>g.marging.win&&!b[i].win).length,lossFlips:x.filter((g,i)=>!g.win&&b[i].win).length}});console.log(JSON.stringify({status:'PASS',hypothesis:'BUY_ONLY_THE_PROSPECTIVE_HERDS_NEAR_TERM_FEED_AT_ACQUISITION_TIME_THEN_KEEP_SMALL_CASH_FLOAT',provenance:'independent mutation of MIT-authored Rita; high-tier black-box behavior used only as hypothesis evidence; no TRACE code or parameters copied',design:variants,summary:sum,paired},null,2));",[],[2024],"53907683-fca5-47fe-a182-84c5dae5aa64","kag_acquisition_prefeed_atomic_probe_v1.ts","01a07985-b5fd-765a-ba5c-5ef413238ad2",138,["D",1788745223171],"import { unzipSync,strFromU8 } from 'npm:fflate';import { loadPyodide } from 'npm:pyodide';import { loadOfficialKaggricultureEngine,observationFor } from './lib/kaggriculture_engine_loader.ts';import { choosePastureAnimal } from './lib/kaggriculture_livestock_portfolio.ts';\nconst variants:any={BASE16:{float:16,pf:0},ACQPF4_F4_ATOMIC:{float:4,pf:4},ACQPF3_F3_ATOMIC:{float:3,pf:3},ACQPF2_F4_ATOMIC:{float:4,pf:2}};const refs=['rancher_rita.py','broker_bea.py'],seeds=[26003,26111,26227,26339];\nconst E=await loadOfficialKaggricultureEngine(),cfg0=E.state.resolveConfig();const rr=await fetch('https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents');const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));const baseSrc=strFromU8(z['rancher_rita.py']);const old=` float_days = policy.get(\"feed_float_days\", 0)\\n wheat_price = prices.get(\"WHEAT\", 25)\\n n = 0\\n for candidate in range(min(gap, policy.get(\"animal_batch\", 2)), 0, -1):\\n feed_float = (flock + candidate) * wheat_price * float_days\\n if money >= candidate * info[\"cost\"] + buffer + feed_float:\\n n = candidate\\n break\\n if n > 0 and spend(n * info[\"cost\"]):\\n orders.append([\"BUY_ANIMAL\", animal, n])`;\nconst repl=` float_days = policy.get(\"feed_float_days\", 0)\\n prefeed_days = policy.get(\"acquisition_prefeed_days\", 0)\\n wheat_price = prices.get(\"WHEAT\", 25)\\n n = 0\\n planned_feed = 0\\n for candidate in range(min(gap, policy.get(\"animal_batch\", 2)), 0, -1):\\n projected_flock = flock + candidate\\n prefeed_need = max(0, projected_flock * prefeed_days - shed.get(\"WHEAT\", 0) - carried_wheat)\\n prefeed_cost = prefeed_need * wheat_price\\n feed_float = projected_flock * wheat_price * float_days\\n needed_slots = 1 + (1 if prefeed_need > 0 else 0)\\n slots_ok = len(orders) + needed_slots <= policy.get(\"max_orders\", 10)\\n if slots_ok and wheat_price <= policy.get(\"max_wheat_price\", 60) and money >= candidate * info[\"cost\"] + buffer + feed_float + prefeed_cost:\\n n = candidate\\n planned_feed = prefeed_need\\n break\\n if n > 0:\\n if planned_feed > 0 and spend(planned_feed * wheat_price):\\n orders.append([\"BUY_PRODUCT\", \"WHEAT\", planned_feed])\\n if spend(n * info[\"cost\"]):\\n orders.append([\"BUY_ANIMAL\", animal, n])`;\nif(!baseSrc.includes(old))throw new Error('mutation target missing');const mutated=baseSrc.replace(old,repl);\nconst py=await loadPyodide();py.runPython(`import json\\n_REG={}\\n_D={}`);for(const f of refs){py.globals.set('nm',f);py.globals.set('src',strFromU8(z[f]));py.runPython(`_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`)}for(const v of Object.keys(variants)){py.globals.set('nm',v);py.globals.set('src',v==='BASE16'?baseSrc:mutated);py.runPython(`_n={}\\nexec(src,_n)\\n_D[nm]=_n`)}py.runPython(`\\ndef cr(n,p):return json.dumps(_REG[n](json.loads(p)))\\ndef cd(n,p):return json.dumps(_D[n]['agent'](json.loads(p)))\\ndef init():\\n for n,d in _D.items(): d['_BASE_POLICY']=json.loads(json.dumps(d['POLICY']))\\ndef prep(n,ff,pf):\\n d=_D[n];d['POLICY'].clear();d['POLICY'].update(json.loads(json.dumps(d['_BASE_POLICY'])));d['POLICY']['animal_target']={'COW':0,'SHEEP':0};d['POLICY']['feed_float_days']=int(ff);d['POLICY']['acquisition_prefeed_days']=int(pf)\\ndef seth(n,c,s,o):\\n d=_D[n];d['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};d['POLICY']['animals']=json.loads(o)\\ninit()\\n`);const cr=py.globals.get('cr'),cd=py.globals.get('cd'),prep=py.globals.get('prep'),seth=py.globals.get('seth');\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}function enforce(o:any,a:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},m:any[]=[];for(const q of a?.market??[]){if(q?.[0]==='BUY_ANIMAL'&&(q[1]==='COW'||q[1]==='SHEEP')){const k=q[1],n=Math.max(0,Math.min(+q[2]||1,t[k]-x[k]));if(n){m.push(['BUY_ANIMAL',k,n]);x[k]+=n}}else m.push(q)}return{...a,market:m}}\nasync function play(v:string,ref:string,seed:number,seat:0|1){const vc=variants[v];prep(v,vc.float,vc.pf);let s=E.state.initGameState(2,{...cfg0,seed},seed),tg:any={COW:0,SHEEP:0},comm=false,err=0,minCash=Infinity,maxFlock=0,finalFlock=0,th:any={},wheatBuy=0,animalBuy=0;while(!s.done){const aa:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid);try{if(pid===seat){const flock=own(o,'COW')+own(o,'SHEEP');maxFlock=Math.max(maxFlock,flock);if(+o.hour===0){for(const k of [2,4,8,12,14,16])if(th[k]===undefined&&flock>=k)th[k]=+o.day;const ow={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')};tg={...ow};if(!comm){comm=ow.COW>=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);seth(v,tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else seth(v,tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}const a=enforce(o,JSON.parse(String(cd(v,JSON.stringify(o)))),tg);for(const q of a.market??[]){if(q?.[0]==='BUY_PRODUCT'&&q?.[1]==='WHEAT')wheatBuy+=Number(q[2]??0);if(q?.[0]==='BUY_ANIMAL')animalBuy+=Number(q[2]??0)}aa[pid]=a;minCash=Math.min(minCash,+o.farms[pid].money)}else aa[pid]=JSON.parse(String(cr(ref,JSON.stringify(o))))}catch{err++;aa[pid]={farmer:['PASS'],hands:[],market:[]}}}s=E.interpreter.step(s,aa,{...cfg0,seed})}const fo=observationFor(s,seat);finalFlock=own(fo,'COW')+own(fo,'SHEEP');const ours=s.farms[seat].money,theirs=s.farms[1-seat].money;return{v,ref,seed,seat,ours,theirs,margin:ours-theirs,win:ours>theirs,err,minCash,maxFlock,finalFlock,animalLoss:maxFlock-finalFlock,th,wheatBuy,animalBuy}}\nconst games:any[]=[];for(const v of Object.keys(variants))for(const ref of refs)for(const seed of seeds){games.push(await play(v,ref,seed,0));games.push(await play(v,ref,seed,1))}const avg=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);const sum:any={};for(const v of Object.keys(variants)){const x=games.filter(g=>g.v===v);sum[v]={episodes:x.length,wins:x.filter(g=>g.win).length,winRate:x.filter(g=>g.win).length/x.length,avgBank:avg(x.map(g=>g.ours)),avgOpp:avg(x.map(g=>g.theirs)),avgMargin:avg(x.map(g=>g.margin)),minCash:Math.min(...x.map(g=>g.minCash)),avgAnimalLoss:avg(x.map(g=>g.animalLoss)),avgWheatBuy:avg(x.map(g=>g.wheatBuy)),avgAnimalBuy:avg(x.map(g=>g.animalBuy)),reach8Day:avg(x.filter(g=>g.th[8]!=null).map(g=>g.th[8])),reach14Share:x.filter(g=>g.th[14]!=null).length/x.length,reach14Day:avg(x.filter(g=>g.th[14]!=null).map(g=>g.th[14])),byOpponent:{}};for(const ref of refs){const y=x.filter(g=>g.ref===ref);sum[v].byOpponent[ref]={wins:y.filter(g=>g.win).length,winRate:y.filter(g=>g.win).length/y.length,avgBank:avg(y.map(g=>g.ours)),avgOpp:avg(y.map(g=>g.theirs)),avgMargin:avg(y.map(g=>g.margin)),avgAnimalLoss:avg(y.map(g=>g.animalLoss))}}}const b=games.filter(g=>g.v==='BASE16');const paired=Object.keys(variants).slice(1).map(v=>{const x=games.filter(g=>g.v===v);return{v,bankDelta:avg(x.map((g,i)=>g.ours-b[i].ours)),oppDelta:avg(x.map((g,i)=>g.theirs-b[i].theirs)),marginDelta:avg(x.map((g,i)=>g.margin-b[i].margin)),better:x.filter((g,i)=>g.margin>b[i].margin).length,worse:x.filter((g,i)=>g.marging.win&&!b[i].win).length,lossFlips:x.filter((g,i)=>!g.win&&b[i].win).length}});console.log(JSON.stringify({status:'PASS',hypothesis:'BUY_ONLY_THE_PROSPECTIVE_HERDS_NEAR_TERM_FEED_AT_ACQUISITION_TIME_THEN_KEEP_SMALL_CASH_FLOAT',provenance:'independent mutation of MIT-authored Rita; high-tier black-box behavior used only as hypothesis evidence; no TRACE code or parameters copied',design:variants,summary:sum,paired},null,2));",[],[2032],"cc6d6013-d074-40d9-86d1-e032efb5f9ed","economic_trial_migrate.ts","c6c01576-aa18-11f1-9bd9-1607ee4eb77e",10,["D",1788715881407],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\n\nconst now = new Date().toISOString();\n\nawait sqlite.batch([\n {\n sql: `CREATE TABLE IF NOT EXISTS economic_trials (\n trial_id TEXT PRIMARY KEY,\n agent_id TEXT NOT NULL,\n opportunity_id TEXT,\n lifecycle_state TEXT NOT NULL,\n operationally_eligible INTEGER NOT NULL DEFAULT 0,\n terminal INTEGER NOT NULL DEFAULT 0,\n gross_revenue REAL NOT NULL DEFAULT 0,\n browser_cost REAL NOT NULL DEFAULT 0,\n llm_cost REAL NOT NULL DEFAULT 0,\n api_cost REAL NOT NULL DEFAULT 0,\n hosting_cost REAL NOT NULL DEFAULT 0,\n platform_fee REAL NOT NULL DEFAULT 0,\n transaction_cost REAL NOT NULL DEFAULT 0,\n net_revenue REAL NOT NULL DEFAULT 0,\n currency TEXT NOT NULL DEFAULT 'BRL',\n evidence_json TEXT,\n started_at TEXT NOT NULL,\n updated_at TEXT NOT NULL,\n finished_at TEXT,\n FOREIGN KEY(agent_id) REFERENCES economic_agents(agent_id),\n FOREIGN KEY(opportunity_id) REFERENCES opportunities(opportunity_id)\n )`,\n },\n {\n sql: `CREATE TABLE IF NOT EXISTS agent_fitness (\n agent_id TEXT PRIMARY KEY,\n selection_state TEXT NOT NULL DEFAULT 'INCUBATION',\n lifetime_gross REAL NOT NULL DEFAULT 0,\n lifetime_cost REAL NOT NULL DEFAULT 0,\n lifetime_net REAL NOT NULL DEFAULT 0,\n benchmark_net REAL NOT NULL DEFAULT 0,\n terminal_trials INTEGER NOT NULL DEFAULT 0,\n successful_trials INTEGER NOT NULL DEFAULT 0,\n failed_trials INTEGER NOT NULL DEFAULT 0,\n rank INTEGER,\n offspring_count INTEGER NOT NULL DEFAULT 0,\n updated_at TEXT NOT NULL,\n FOREIGN KEY(agent_id) REFERENCES economic_agents(agent_id)\n )`,\n },\n { sql: `CREATE INDEX IF NOT EXISTS idx_trials_agent_state ON economic_trials(agent_id,lifecycle_state)` },\n { sql: `CREATE INDEX IF NOT EXISTS idx_trials_opportunity ON economic_trials(opportunity_id,lifecycle_state)` },\n {\n sql: `INSERT OR IGNORE INTO agent_fitness(agent_id,selection_state,updated_at)\n SELECT agent_id,'INCUBATION',? FROM economic_agents`,\n args: [now],\n },\n {\n sql: `INSERT INTO schema_meta(key,value,updated_at) VALUES('economic_trial_schema_version','0.6.1-candidate',?)\n ON CONFLICT(key) DO UPDATE SET value=excluded.value,updated_at=excluded.updated_at`,\n args: [now],\n },\n]);\n\nconst tables = await sqlite.execute(`SELECT name FROM sqlite_master WHERE type='table' AND name IN ('economic_trials','agent_fitness') ORDER BY name`);\nconst fitness = await sqlite.execute(`SELECT agent_id,selection_state,lifetime_net,terminal_trials FROM agent_fitness ORDER BY agent_id`);\nconsole.log(JSON.stringify({ status: 'PASS', version: '0.6.1-candidate', tables: tables.rows, fitness: fitness.rows }, null, 2));\n",[],[2040],"3746465a-cdd1-4c28-a3bd-e17821cba719","economic_status.ts","af4f16f4-aa17-11f1-a15c-1607ee4eb77e",["D",1788716592876],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { ECONOMIC_POLICY } from \"./lib/economic_policy.ts\";\n\nconst agents = await sqlite.execute(`SELECT status, COUNT(*) AS count FROM economic_agents GROUP BY status ORDER BY status`);\nconst opportunities = await sqlite.execute(`SELECT status, COUNT(*) AS count FROM opportunities GROUP BY status ORDER BY status`);\nconst runs = await sqlite.execute(`SELECT state, COUNT(*) AS count, COALESCE(SUM(net_revenue),0) AS net_revenue FROM economic_runs GROUP BY state ORDER BY state`);\nconst trials = await sqlite.execute(`SELECT lifecycle_state,COUNT(*) AS count,COALESCE(SUM(net_revenue),0) AS net_revenue FROM economic_trials GROUP BY lifecycle_state ORDER BY lifecycle_state`);\nconst fitness = await sqlite.execute(`SELECT agent_id,selection_state,lifetime_gross,lifetime_cost,lifetime_net,benchmark_net,terminal_trials,successful_trials,failed_trials,rank,offspring_count,updated_at FROM agent_fitness ORDER BY generation IS NULL, rank IS NULL, rank, agent_id`).catch(async () =>\n await sqlite.execute(`SELECT agent_id,selection_state,lifetime_gross,lifetime_cost,lifetime_net,benchmark_net,terminal_trials,successful_trials,failed_trials,rank,offspring_count,updated_at FROM agent_fitness ORDER BY rank IS NULL,rank,agent_id`)\n);\nconst variants = await sqlite.execute(`SELECT variant_id,agent_id,parent_variant_id,generation,name,mutation,status,evidence_state,benchmark_metric,benchmark_value,artifact_pointer,updated_at FROM strategy_variants ORDER BY agent_id,generation,variant_id`);\nconst variantEvals = await sqlite.execute(`SELECT variant_id,COUNT(*) AS evaluations,COALESCE(SUM(episodes),0) AS episodes,COALESCE(SUM(wins),0) AS wins,COALESCE(SUM(losses),0) AS losses,COALESCE(SUM(ties),0) AS ties,COALESCE(SUM(runtime_errors),0) AS runtime_errors FROM variant_evaluations GROUP BY variant_id ORDER BY variant_id`);\nconst champions = await sqlite.execute(`SELECT champion_class,agent_id,run_id,metric_value,currency,updated_at FROM economic_champions ORDER BY champion_class`);\nconst latestAgents = await sqlite.execute(`SELECT agent_id,parent_id,generation,lineage,status,hypothesis,strategy,updated_at FROM economic_agents ORDER BY generation DESC,agent_id LIMIT 20`);\nconst latestOpportunities = await sqlite.execute(`SELECT opportunity_id,title,source,url,category,deadline,payout_currency,payout_max,estimated_cost,human_dependency,evidence_state,status,score,updated_at FROM opportunities ORDER BY score DESC,updated_at DESC LIMIT 30`);\nconst latestTrials = await sqlite.execute(`SELECT trial_id,agent_id,opportunity_id,lifecycle_state,operationally_eligible,terminal,gross_revenue,net_revenue,currency,updated_at,finished_at FROM economic_trials ORDER BY updated_at DESC LIMIT 30`);\nconst failures = await sqlite.execute(`SELECT failure_class,COUNT(*) AS count,MAX(created_at) AS latest FROM negative_knowledge GROUP BY failure_class ORDER BY latest DESC,failure_class`);\nconst success = await sqlite.execute(`SELECT COUNT(*) AS count, COALESCE(MAX(replication_count),0) AS max_replications FROM success_patterns`);\nconst totals = await sqlite.execute(`SELECT\n COALESCE(SUM(gross_revenue),0) AS gross_revenue,\n COALESCE(SUM(browser_cost+llm_cost+api_cost+hosting_cost+platform_fee+transaction_cost),0) AS total_cost,\n COALESCE(SUM(net_revenue),0) AS net_revenue\n FROM economic_trials`);\nconst schema = await sqlite.execute(`SELECT key,value,updated_at FROM schema_meta WHERE key LIKE '%version%' ORDER BY key`);\n\nconst totalRow = totals.rows[0] ?? {};\nconst realRevenue = Number(totalRow.gross_revenue ?? 0);\nconst realNet = Number(totalRow.net_revenue ?? 0);\nconst terminalTrialCount = latestTrials.rows.filter((x: Record) => Number(x.terminal ?? 0) === 1).length;\n\nconsole.log(JSON.stringify({\n service: \"chat-native-maestro/economic\",\n version: ECONOMIC_POLICY.version,\n schema: schema.rows,\n agents: agents.rows,\n opportunities: opportunities.rows,\n runs: runs.rows,\n trials: trials.rows,\n fitness: fitness.rows,\n variants: variants.rows,\n variantEvaluations: variantEvals.rows,\n champions: champions.rows,\n latestAgents: latestAgents.rows,\n rankedOpportunities: latestOpportunities.rows,\n latestTrials: latestTrials.rows,\n failureClasses: failures.rows,\n successPatterns: success.rows,\n totals: totalRow,\n terminalTrialCount,\n realRevenueState: realRevenue > 0 ? \"NONZERO\" : \"ZERO\",\n netRevenueState: realNet > 0 ? \"POSITIVE\" : realNet < 0 ? \"NEGATIVE\" : \"ZERO\",\n selfSustaining: realNet > 0 && terminalTrialCount > 0,\n timestamp: new Date().toISOString(),\n}, null, 2));\n",[],[2048],"aa858d97-bbd1-487f-951e-c22790b9a453","economic_selftest.ts","66d41424-aa17-11f1-9a39-1607ee4eb77e",["D",1788715291111],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport { ECONOMIC_POLICY, evaluateEconomicAction, netRevenue } from \"./lib/economic_policy.ts\";\n\nconst cases = {\n r0: evaluateEconomicAction({ actionClass: \"R0_OBSERVE\" }),\n r1: evaluateEconomicAction({ actionClass: \"R1_PREPARE\" }),\n r2: evaluateEconomicAction({ actionClass: \"R2_REVERSIBLE_WRITE\" }),\n r3Blocked: evaluateEconomicAction({ actionClass: \"R3_COMMIT_SUBMIT\", approved: false }),\n r3Approved: evaluateEconomicAction({ actionClass: \"R3_COMMIT_SUBMIT\", approved: true }),\n r4BlockedNoApproval: evaluateEconomicAction({ actionClass: \"R4_ECONOMIC_SPEND\", amount: 1, maxBudget: 5, approved: false }),\n r4BlockedBudget: evaluateEconomicAction({ actionClass: \"R4_ECONOMIC_SPEND\", amount: 6, maxBudget: 5, approved: true }),\n r4Approved: evaluateEconomicAction({ actionClass: \"R4_ECONOMIC_SPEND\", amount: 4, maxBudget: 5, approved: true }),\n r5: evaluateEconomicAction({ actionClass: \"R5_IDENTITY_LEGAL\", approved: true }),\n};\n\nconst profitCase = netRevenue({ grossRevenue: 10, browserCost: 1, llmCost: 2, apiCost: 0.5, platformFee: 1 });\nconst lossCase = netRevenue({ grossRevenue: 5, browserCost: 4, llmCost: 3 });\nconst tables = await sqlite.execute(`SELECT name FROM sqlite_master WHERE type='table' AND name IN (\n 'economic_agents','opportunities','economic_runs','negative_knowledge','success_patterns','economic_champions'\n) ORDER BY name`);\nconst root = await sqlite.execute(`SELECT agent_id,parent_id,generation,lineage,status FROM economic_agents WHERE agent_id='A0'`);\n\nconst checks = [\n { name: \"R0 auto\", pass: cases.r0.allowed && !cases.r0.humanGate },\n { name: \"R1 auto\", pass: cases.r1.allowed && !cases.r1.humanGate },\n { name: \"R2 auto\", pass: cases.r2.allowed && !cases.r2.humanGate },\n { name: \"R3 blocks without approval\", pass: !cases.r3Blocked.allowed && cases.r3Blocked.reason === \"NEEDS_COMMIT_APPROVAL\" },\n { name: \"R3 allows approved\", pass: cases.r3Approved.allowed },\n { name: \"R4 blocks no approval\", pass: !cases.r4BlockedNoApproval.allowed },\n { name: \"R4 enforces budget\", pass: !cases.r4BlockedBudget.allowed && cases.r4BlockedBudget.reason === \"BUDGET_EXCEEDED\" },\n { name: \"R4 permits approved in-budget\", pass: cases.r4Approved.allowed },\n { name: \"R5 always human gate\", pass: !cases.r5.allowed && cases.r5.humanGate },\n { name: \"net revenue profit\", pass: profitCase === 5.5 },\n { name: \"net revenue loss\", pass: lossCase === -2 },\n { name: \"all economic tables present\", pass: tables.rows.length === 6 },\n { name: \"A0 root present\", pass: root.rows.length === 1 && root.rows[0].generation === 0 },\n];\n\nconst failed = checks.filter((x) => !x.pass);\nconsole.log(JSON.stringify({\n status: failed.length ? \"FAIL\" : \"PASS\",\n version: ECONOMIC_POLICY.version,\n checks,\n failed,\n cases,\n profitCase,\n lossCase,\n tables: tables.rows,\n root: root.rows,\n}, null, 2));\nif (failed.length) throw new Error(`economic_selftest failed: ${failed.map((x) => x.name).join(', ')}`);\n",[],[2055],"c2e35226-ae2c-4a39-bfb8-3af34f7addf3","economic_migrate.ts","5a85f9ee-aa17-11f1-8f39-1607ee4eb77e",["D",1788715270460],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\n\nconst now = new Date().toISOString();\n\nawait sqlite.batch([\n {\n sql: `CREATE TABLE IF NOT EXISTS economic_agents (\n agent_id TEXT PRIMARY KEY,\n parent_id TEXT,\n generation INTEGER NOT NULL DEFAULT 0,\n lineage TEXT NOT NULL,\n hypothesis TEXT NOT NULL,\n strategy TEXT NOT NULL,\n status TEXT NOT NULL,\n created_at TEXT NOT NULL,\n updated_at TEXT NOT NULL,\n FOREIGN KEY(parent_id) REFERENCES economic_agents(agent_id)\n )`,\n },\n {\n sql: `CREATE TABLE IF NOT EXISTS opportunities (\n opportunity_id TEXT PRIMARY KEY,\n title TEXT NOT NULL,\n source TEXT NOT NULL,\n url TEXT,\n category TEXT NOT NULL,\n deadline TEXT,\n payout_currency TEXT,\n payout_max REAL,\n estimated_cost REAL NOT NULL DEFAULT 0,\n human_dependency REAL NOT NULL DEFAULT 1,\n evidence_state TEXT NOT NULL,\n status TEXT NOT NULL,\n score REAL,\n created_at TEXT NOT NULL,\n updated_at TEXT NOT NULL\n )`,\n },\n {\n sql: `CREATE TABLE IF NOT EXISTS economic_runs (\n run_id TEXT PRIMARY KEY,\n agent_id TEXT NOT NULL,\n opportunity_id TEXT,\n browser_run_id TEXT,\n action_class TEXT NOT NULL,\n strategy_variant TEXT,\n provider TEXT,\n state TEXT NOT NULL,\n gross_revenue REAL NOT NULL DEFAULT 0,\n currency TEXT NOT NULL DEFAULT 'BRL',\n browser_cost REAL NOT NULL DEFAULT 0,\n llm_cost REAL NOT NULL DEFAULT 0,\n api_cost REAL NOT NULL DEFAULT 0,\n hosting_cost REAL NOT NULL DEFAULT 0,\n platform_fee REAL NOT NULL DEFAULT 0,\n transaction_cost REAL NOT NULL DEFAULT 0,\n net_revenue REAL NOT NULL DEFAULT 0,\n evidence_json TEXT,\n started_at TEXT NOT NULL,\n finished_at TEXT,\n FOREIGN KEY(agent_id) REFERENCES economic_agents(agent_id),\n FOREIGN KEY(opportunity_id) REFERENCES opportunities(opportunity_id)\n )`,\n },\n {\n sql: `CREATE TABLE IF NOT EXISTS negative_knowledge (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n agent_id TEXT,\n opportunity_id TEXT,\n failure_class TEXT NOT NULL,\n cause TEXT NOT NULL,\n context TEXT,\n detector TEXT,\n avoid_rule TEXT,\n evidence_json TEXT,\n created_at TEXT NOT NULL\n )`,\n },\n {\n sql: `CREATE TABLE IF NOT EXISTS success_patterns (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n agent_id TEXT,\n opportunity_id TEXT,\n mechanism TEXT NOT NULL,\n preconditions TEXT,\n counterfactual TEXT,\n boundaries TEXT,\n replication_count INTEGER NOT NULL DEFAULT 0,\n confidence REAL NOT NULL DEFAULT 0,\n evidence_json TEXT,\n created_at TEXT NOT NULL,\n updated_at TEXT NOT NULL\n )`,\n },\n {\n sql: `CREATE TABLE IF NOT EXISTS economic_champions (\n champion_class TEXT PRIMARY KEY,\n agent_id TEXT NOT NULL,\n run_id TEXT,\n metric_value REAL NOT NULL DEFAULT 0,\n currency TEXT NOT NULL DEFAULT 'BRL',\n updated_at TEXT NOT NULL\n )`,\n },\n { sql: `CREATE INDEX IF NOT EXISTS idx_economic_agents_parent ON economic_agents(parent_id)` },\n { sql: `CREATE INDEX IF NOT EXISTS idx_opportunities_status_score ON opportunities(status, score)` },\n { sql: `CREATE INDEX IF NOT EXISTS idx_economic_runs_agent ON economic_runs(agent_id, started_at)` },\n { sql: `CREATE INDEX IF NOT EXISTS idx_economic_runs_opportunity ON economic_runs(opportunity_id, started_at)` },\n {\n sql: `INSERT OR IGNORE INTO economic_agents(agent_id,parent_id,generation,lineage,hypothesis,strategy,status,created_at,updated_at)\n VALUES('A0',NULL,0,'A0','Find economically viable opportunities with minimal human dependency','Neurônio root: branch, measure, falsify, replicate winners','INCUBATION',?,?)`,\n args: [now, now],\n },\n {\n sql: `INSERT INTO schema_meta(key,value,updated_at) VALUES('economic_schema_version','0.6.0-candidate',?)\n ON CONFLICT(key) DO UPDATE SET value=excluded.value, updated_at=excluded.updated_at`,\n args: [now],\n },\n]);\n\nconst tables = await sqlite.execute(`SELECT name FROM sqlite_master WHERE type='table' AND name IN (\n 'economic_agents','opportunities','economic_runs','negative_knowledge','success_patterns','economic_champions'\n) ORDER BY name`);\nconst root = await sqlite.execute(`SELECT agent_id,generation,lineage,status FROM economic_agents WHERE agent_id='A0'`);\nconsole.log(JSON.stringify({ status: 'PASS', economicSchemaVersion: '0.6.0-candidate', tables: tables.rows, root: root.rows }, null, 2));\n",[],[2062],"c0c8d74e-751f-4373-be52-e3516ba9a92b","README.md","01a077bc-2b99-740d-9881-4e48ec93504b",587,["D",1788834981335],"# Chat-Native Maestro V0.5.0\n\n> **KAGGRICULTURE HARD BOOTSTRAP:** antes de qualquer hipótese, mutação, interpretação de resultado ou enqueue relacionado ao Kaggriculture, **LER** `artifacts/kaggriculture/00_READ_FIRST.md` e cumprir o `KAG_CONTEXT_PREFLIGHT`. `FOUND != READ`; chat/memória não substituem estado persistido.\n\nCloud runtime controlled from normal ChatGPT mobile/web.\n\n## Canonical constraint\n\n- No Codex.\n- No user PC.\n- No local browser runtime.\n- Normal ChatGPT is the control plane.\n- Val Town is the persistent cloud compute/state plane.\n- No automatic browser schedule is enabled.\n\n## Runtime path\n\n`ChatGPT -> authenticated Val Town plugin -> val-scoped SQLite -> runner.ts -> provider router -> cloud browser -> verified receipt`\n\nThe public HTTP endpoint is liveness-only. Provider configuration, task state, receipts, health and cost metrics are available only through authenticated `status.ts` execution.\n\n## Browser routing\n\n1. Kernel — primary/champion.\n2. Browserless — recurring-free fallback.\n3. Browserbase — researched fallback, not implemented.\n4. Steel — one-time-credit fallback, not preferred for steady state.\n\nCross-provider fallback is allowed only for `read_only`. A write failure is never replayed automatically on another provider because failure is not proof of zero side effects.\n\n## Current provider economics (research snapshot 2026-09-06)\n\n### Kernel Free\n\n- $0/month plus usage costs.\n- $5/month free credits.\n- 5 concurrent browsers.\n- live view.\n- stealth mode with CAPTCHA solver/proxy.\n- managed auth.\n- no card required to get started.\n- usage model: GB-second.\n\n### Browserless Free\n\n- 1,000 units/month.\n- 2 concurrent browsers.\n- 2-minute max session.\n- no card required.\n- browser time: 1 unit per 30-second block.\n- successful CAPTCHA solve: additional units.\n- built-in proxies: additional units.\n- provider supports an overage cap that should be set to $0 for this project's zero-surprise-cost policy.\n\n## Browser DSL\n\nSupported:\n- `goto`\n- `wait_for`\n- `click`\n- `fill`\n- `fill_secret`\n- `type`\n- `select`\n- `extract_text`\n- `extract_attribute`\n- `expect_text`\n- `expect_url`\n\nArbitrary JavaScript and shell input are intentionally absent from the task schema.\n\nThe `value` attribute is blocked from `extract_attribute` to reduce accidental credential exfiltration after form fill.\n\n## Risk gates\n\n- `read_only`: normal execution.\n- `reversible_write`: requires explicit `approved=true`.\n- `consequential`: requires explicit `approved=true`.\n- `fill_secret`: separately requires explicit approval, even during a read-oriented flow.\n\n## Secret handling\n\nPasswords, cookies and provider tokens must never be stored in `task_json`.\n\nUse:\n\n```json\n{\"action\":\"fill_secret\",\"selector\":\"#password\",\"envVar\":\"PORTAL_PASSWORD\"}\n```\n\nOnly the environment-variable name is stored. The value is resolved in memory immediately before execution.\n\nProvider credentials belong only in Val Town env vars:\n- `KERNEL_API_KEY`\n- `BROWSERLESS_TOKEN`\n- optional `BROWSERLESS_BASE_URL`\n\n## Network / SSRF boundary\n\nBefore any provider call, `runner.ts` runs a fail-closed network preflight:\n\n- only HTTP/HTTPS;\n- URL userinfo is blocked;\n- nonstandard ports are blocked;\n- localhost/private/link-local/reserved/documentation IP ranges are blocked;\n- known metadata/internal hostnames are blocked;\n- `allowedHosts` are validated;\n- hostnames are resolved with DNS-over-HTTPS and all A/AAAA results must be public.\n\nThe browser program itself also restricts main-frame navigation to the original target host plus explicit `allowedHosts`.\n\nDNS preflight reduces SSRF/DNS-rebinding risk but is not a mathematical proof against DNS rebinding after the preflight. Provider isolation and post-action validation remain required.\n\n## Read-only browser firewall\n\nWhen `risk=read_only`, browser network requests using methods other than GET/HEAD/OPTIONS are aborted. Popups are closed.\n\nThis prevents common POST/PUT/PATCH/DELETE side effects. It cannot guarantee zero side effects on badly designed GET endpoints.\n\n## Reliability controls\n\n- task `run_id` is unique/idempotent;\n- atomic queue claim;\n- 90-second RUNNING lease;\n- stale lease recovery;\n- max 2 task attempts;\n- read-only FAIL retry delay: 10 seconds;\n- writes are not auto-retried;\n- browser timeout: 25 seconds;\n- Kernel cleanup timeout: 5 seconds;\n- provider circuit opens after 3 consecutive provider FAILs;\n- circuit cooldown: 15 minutes;\n- BLOCKED configuration errors do not degrade provider health;\n- no cron/automatic runner, so eligible retries still require an explicit ChatGPT runner invocation.\n\n## Evidence retention\n\nDefault is `minimal`:\n- status;\n- final URL;\n- expectations/checks;\n- error;\n- duration;\n- fallback metadata;\n- DNS preflight evidence.\n\nExtracted outputs and provider session IDs are not persisted by default.\n\nUse `evidenceRetention: \"full\"` only when the extracted output is intentionally safe to retain.\n\n## Cost telemetry\n\n`browser_attempts` records per-attempt provider/status/duration and an estimated Browserless browser-time unit count.\n\nBrowserless estimates exclude CAPTCHA/proxy add-ons. Kernel usage is not estimated locally because its billing basis is GB-second; actual account usage must be observed from Kernel.\n\n## PASS rule\n\n`TOOL_SUCCESS_RESPONSE != BEHAVIOR_PROOF`.\n\nA browser execution with no positive expectation remains `PARTIAL`. `PASS` requires all explicit expectations to succeed.\n\n## Current live evidence\n\nPASS:\n- Val Town plugin authenticated.\n- Free workspace.\n- create/edit/run/deploy.\n- public liveness endpoint.\n- authenticated diagnostics.\n- val-scoped SQLite.\n- schema versioning.\n- atomic task claim.\n- stale lease recovery.\n- persisted task/attempt receipts.\n- retry decision logic.\n- timeout control.\n- provider circuit breaker and restore test.\n- risk approval gate.\n- secret approval gate.\n- invalid provider rejection.\n- `value` attribute extraction rejection.\n- DNS-over-HTTPS preflight.\n- private/metadata URL blocking in both self-test and runner pipeline.\n- Kernel SDK module loading.\n- Browserless adapter module loading.\n- safe missing-provider-credential classification.\n\nUNVERIFIED:\n- Kernel API authentication for this account.\n- live Kernel browser creation.\n- live Kernel Playwright execution.\n- Kernel actual credit consumption.\n- Browserless API authentication for this account.\n- live Browserless `/function` execution.\n- Browserless actual unit consumption.\n\nNo user-created environment variables are currently configured, so no live browser provider has been promoted to OPERATIONAL.\n\n## Next milestone\n\n`docs/SMOKE_TASK.json` is deliberately a DRAFT, not a queued task. After exactly one provider credential is configured, ChatGPT can enqueue that read-only smoke task, run `runner.ts`, persist the attempt and promote the provider only if the explicit expectations PASS.\n\nSee:\n- `docs/CURRENT_STATE.md`\n- `docs/PROVIDER_CENSUS.md`\n- `docs/ACTIVATION_RUNBOOK.md`\n- `docs/TASK_SCHEMA.md`\n- `docs/SMOKE_TASK.json`\n",[],[2069],"55194c44-26a6-467a-b820-b6fead64159d","router.ts","01a077bc-2b99-740d-9881-4721aef6fa58","01a077bc-2b99-740d-9881-4952f29c18f3",["D",1788715215771],"import type { BrowserReceipt, BrowserTask, ProviderName } from \"../lib/types.ts\";\nimport { executeKernel, kernelConfigured } from \"./kernel.ts\";\nimport { browserlessConfigured, executeBrowserless } from \"./browserless.ts\";\nimport { providerAvailable, recordProviderOutcome } from \"../lib/provider_health.ts\";\n\nexport function providerState() {\n const kernel = kernelConfigured();\n const browserless = browserlessConfigured();\n return {\n kernel: { priority: 1, configured: kernel, state: kernel ? \"READY\" : \"ACCOUNT_KEY_REQUIRED\" },\n browserless: { priority: 2, configured: browserless, state: browserless ? \"READY\" : \"ACCOUNT_TOKEN_REQUIRED\" },\n };\n}\n\nasync function runProvider(provider: ProviderName, task: BrowserTask): Promise {\n const health = await providerAvailable(provider);\n if (!health.available) {\n return { runId: task.runId, provider, status: \"BLOCKED\", error: health.reason ?? \"provider circuit open\" };\n }\n\n const receipt = provider === \"kernel\" ? await executeKernel(task) : await executeBrowserless(task);\n await recordProviderOutcome(provider, receipt);\n return receipt;\n}\n\nfunction withFallbackMeta(receipt: BrowserReceipt, prior: BrowserReceipt): BrowserReceipt {\n return {\n ...receipt,\n providerMeta: {\n ...(receipt.providerMeta ?? {}),\n fallbackFrom: prior.provider,\n priorStatus: prior.status,\n priorError: prior.error ?? null,\n priorDurationMs: prior.durationMs ?? null,\n },\n };\n}\n\nexport async function executeRouted(task: BrowserTask): Promise {\n if (task.preferredProvider === \"browserless\") return runProvider(\"browserless\", task);\n if (task.preferredProvider === \"kernel\") return runProvider(\"kernel\", task);\n\n if (kernelConfigured()) {\n const primary = await runProvider(\"kernel\", task);\n if (primary.status === \"PASS\" || primary.status === \"PARTIAL\") return primary;\n if (task.risk === \"read_only\" && browserlessConfigured()) {\n const fallback = await runProvider(\"browserless\", task);\n return withFallbackMeta(fallback, primary);\n }\n return primary;\n }\n\n if (browserlessConfigured()) return runProvider(\"browserless\", task);\n return runProvider(\"kernel\", task);\n}\n",[],[2084,2077],"providers","fc83999d-4ff9-4e7c-abc1-af6110dc1d07","kernel.ts","01a077bc-2b99-740d-9881-42c7df42306f",["D",1788715215771],"import Kernel from \"npm:@onkernel/sdk\";\nimport type { BrowserReceipt, BrowserTask } from \"../lib/types.ts\";\nimport { buildKernelPlaywrightCode, resolveSecrets } from \"../lib/dsl.ts\";\nimport { POLICY, withTimeout } from \"../lib/policy.ts\";\n\nexport function kernelConfigured(): boolean {\n return Boolean(Deno.env.get(\"KERNEL_API_KEY\"));\n}\n\nexport async function executeKernel(task: BrowserTask): Promise {\n const apiKey = Deno.env.get(\"KERNEL_API_KEY\");\n if (!apiKey) {\n return { runId: task.runId, provider: \"kernel\", status: \"BLOCKED\", error: \"KERNEL_API_KEY missing\" };\n }\n\n let resolvedTask: BrowserTask;\n try {\n resolvedTask = resolveSecrets(task);\n } catch (error) {\n return { runId: task.runId, provider: \"kernel\", status: \"BLOCKED\", error: error instanceof Error ? error.message : String(error) };\n }\n\n const started = Date.now();\n const kernel = new Kernel({ apiKey });\n let sessionId: string | undefined;\n try {\n const kernelBrowser = await withTimeout(\n kernel.browsers.create({ headless: true, stealth: true }),\n 10_000,\n \"kernel browser create\",\n );\n sessionId = kernelBrowser.session_id;\n\n const response = await withTimeout(\n kernel.browsers.playwright.execute(sessionId, {\n code: buildKernelPlaywrightCode(resolvedTask),\n }),\n POLICY.browserTimeoutMs,\n \"kernel playwright execute\",\n );\n const result = (response as any).result ?? response;\n const verified = Boolean(result?.verified);\n return {\n runId: task.runId,\n provider: \"kernel\",\n status: verified ? \"PASS\" : \"PARTIAL\",\n finalUrl: result?.finalUrl,\n outputs: result?.outputs ?? {},\n checks: result?.checks ?? [],\n durationMs: Date.now() - started,\n providerMeta: { sessionId, headless: true, stealth: true },\n };\n } catch (error) {\n return {\n runId: task.runId,\n provider: \"kernel\",\n status: \"FAIL\",\n error: error instanceof Error ? error.message : String(error),\n durationMs: Date.now() - started,\n providerMeta: sessionId ? { sessionId } : {},\n };\n } finally {\n if (sessionId) {\n try {\n await withTimeout(kernel.browsers.deleteByID(sessionId), 5_000, \"kernel browser cleanup\");\n } catch (cleanupError) {\n console.error(\"kernel cleanup failed\", cleanupError);\n }\n }\n }\n}\n",[],[2084,2086],"26c3a975-f21b-42bd-bb91-c1a4c17ea40b","browserless.ts","01a077bc-2b99-740d-9881-3f2f98b23d50",["D",1788715215771],"import type { BrowserReceipt, BrowserTask } from \"../lib/types.ts\";\nimport { resolveSecrets } from \"../lib/dsl.ts\";\nimport { POLICY } from \"../lib/policy.ts\";\n\nconst FUNCTION_CODE = String.raw`\nexport default async ({ page, context }) => {\n const plan = context.plan;\n const outputs = {};\n const checks = [];\n const targetHost = new URL(plan.targetUrl).hostname;\n const allowedHosts = new Set([targetHost, ...(plan.allowedHosts || [])]);\n const allowedMethods = new Set([\"GET\", \"HEAD\", \"OPTIONS\"]);\n const assertUrl = (value) => {\n const u = new URL(value);\n if (![\"http:\", \"https:\"].includes(u.protocol)) throw new Error(\"blocked protocol\");\n if (!allowedHosts.has(u.hostname)) throw new Error(\"blocked host: \" + u.hostname);\n };\n\n await page.setRequestInterception(true);\n page.on(\"request\", (req) => {\n try {\n if (req.isNavigationRequest() && req.frame() === page.mainFrame()) assertUrl(req.url());\n if (plan.risk === \"read_only\" && !allowedMethods.has(req.method().toUpperCase())) return req.abort(\"blockedbyclient\");\n return req.continue();\n } catch {\n return req.abort(\"blockedbyclient\");\n }\n });\n page.on(\"popup\", async (popup) => { try { await popup.close(); } catch {} });\n\n assertUrl(plan.targetUrl);\n await page.goto(plan.targetUrl, { waitUntil: \"domcontentloaded\" });\n assertUrl(page.url());\n\n for (const step of plan.steps) {\n switch (step.action) {\n case \"goto\": {\n const url = step.url || plan.targetUrl;\n assertUrl(url);\n await page.goto(url, { waitUntil: \"domcontentloaded\" });\n break;\n }\n case \"wait_for\":\n await page.waitForSelector(step.selector, { visible: true, timeout: step.timeoutMs || 10000 });\n break;\n case \"click\":\n if (step.waitForNavigation) {\n await Promise.all([page.waitForNavigation({ waitUntil: \"domcontentloaded\" }), page.click(step.selector)]);\n } else await page.click(step.selector);\n break;\n case \"fill\":\n await page.click(step.selector, { clickCount: 3 });\n await page.keyboard.press(\"Backspace\");\n await page.type(step.selector, step.text);\n break;\n case \"type\":\n await page.type(step.selector, step.text, { delay: step.delayMs || 0 });\n break;\n case \"select\":\n await page.select(step.selector, step.value);\n break;\n case \"extract_text\":\n outputs[step.key] = await page.$eval(step.selector, (el) => el.textContent || \"\");\n break;\n case \"extract_attribute\":\n outputs[step.key] = await page.$eval(step.selector, (el, attr) => el.getAttribute(attr), step.attribute);\n break;\n case \"expect_text\": {\n const body = await page.evaluate(() => document.body?.innerText || \"\");\n checks.push({ type: \"text\", expected: step.text, ok: body.includes(step.text) });\n break;\n }\n case \"expect_url\": {\n const actual = page.url();\n checks.push({ type: \"url\", expected: step.contains, actual, ok: actual.includes(step.contains) });\n break;\n }\n }\n assertUrl(page.url());\n }\n\n return {\n data: {\n finalUrl: page.url(),\n outputs,\n checks,\n verified: checks.length > 0 && checks.every((x) => x.ok),\n },\n type: \"application/json\",\n };\n};\n`;\n\nexport function browserlessConfigured(): boolean {\n return Boolean(Deno.env.get(\"BROWSERLESS_TOKEN\"));\n}\n\nexport async function executeBrowserless(task: BrowserTask): Promise {\n const token = Deno.env.get(\"BROWSERLESS_TOKEN\");\n if (!token) return { runId: task.runId, provider: \"browserless\", status: \"BLOCKED\", error: \"BROWSERLESS_TOKEN missing\" };\n\n let resolvedTask: BrowserTask;\n try {\n resolvedTask = resolveSecrets(task);\n } catch (error) {\n return { runId: task.runId, provider: \"browserless\", status: \"BLOCKED\", error: error instanceof Error ? error.message : String(error) };\n }\n\n const started = Date.now();\n const base = (Deno.env.get(\"BROWSERLESS_BASE_URL\") || \"https://production-sfo.browserless.io\").replace(/\\/+$/, \"\");\n try {\n const response = await fetch(`${base}/function?token=${encodeURIComponent(token)}`, {\n method: \"POST\",\n headers: { \"content-type\": \"application/json\" },\n body: JSON.stringify({ code: FUNCTION_CODE, context: { plan: resolvedTask } }),\n signal: AbortSignal.timeout(POLICY.browserTimeoutMs),\n });\n const text = await response.text();\n let result: any;\n try { result = JSON.parse(text); } catch { result = { raw: text }; }\n if (!response.ok) {\n return { runId: task.runId, provider: \"browserless\", status: \"FAIL\", error: `HTTP ${response.status}: ${text.slice(0, 500)}`, durationMs: Date.now() - started };\n }\n return {\n runId: task.runId,\n provider: \"browserless\",\n status: result?.verified ? \"PASS\" : \"PARTIAL\",\n finalUrl: result?.finalUrl,\n outputs: result?.outputs ?? {},\n checks: result?.checks ?? [],\n durationMs: Date.now() - started,\n providerMeta: { base },\n };\n } catch (error) {\n return { runId: task.runId, provider: \"browserless\", status: \"FAIL\", error: error instanceof Error ? error.message : String(error), durationMs: Date.now() - started };\n }\n}\n",[],[2084,2093],"9e625300-a023-4458-8c07-bea2447f6601",["D",1788715215771],"directory",[300,301,302],[2084],"55c421c5-2ce5-4f82-9115-c76d40c7214a","url_security.ts","01a077bc-2b99-740d-9881-355e348197e4","01a077bc-2b99-740d-9881-3bf068d89182",["D",1788715215771],"import type { BrowserTask } from \"./types.ts\";\n\ntype DnsAnswer = { name?: string; type?: number; data?: string };\ntype DnsJson = { Status?: number; Answer?: DnsAnswer[] };\n\nconst BLOCKED_HOST_NAMES = new Set([\n \"localhost\",\n \"localhost.localdomain\",\n \"metadata.google.internal\",\n \"metadata\",\n]);\n\nfunction stripBrackets(host: string): string {\n return host.startsWith(\"[\") && host.endsWith(\"]\") ? host.slice(1, -1) : host;\n}\n\nfunction parseIpv4(value: string): number[] | null {\n const parts = value.split(\".\");\n if (parts.length !== 4) return null;\n const nums = parts.map((p) => Number(p));\n if (nums.some((n) => !Number.isInteger(n) || n < 0 || n > 255)) return null;\n return nums;\n}\n\nexport function isNonPublicIp(value: string): boolean {\n const ip = stripBrackets(value.trim().toLowerCase());\n const v4 = parseIpv4(ip);\n if (v4) {\n const [a, b, c] = v4;\n if (a === 0 || a === 10 || a === 127 || a >= 224) return true;\n if (a === 100 && b >= 64 && b <= 127) return true;\n if (a === 169 && b === 254) return true;\n if (a === 172 && b >= 16 && b <= 31) return true;\n if (a === 192 && b === 0 && c === 0) return true;\n if (a === 192 && b === 0 && c === 2) return true;\n if (a === 192 && b === 168) return true;\n if (a === 198 && (b === 18 || b === 19)) return true;\n if (a === 198 && b === 51 && c === 100) return true;\n if (a === 203 && b === 0 && c === 113) return true;\n return false;\n }\n\n if (ip.includes(\":\")) {\n if (ip === \"::\" || ip === \"::1\") return true;\n if (ip.startsWith(\"fc\") || ip.startsWith(\"fd\")) return true;\n if (/^fe[89ab]/.test(ip)) return true;\n if (ip.startsWith(\"ff\")) return true;\n if (ip.startsWith(\"2001:db8\")) return true;\n const mapped = ip.match(/::ffff:(\\d+\\.\\d+\\.\\d+\\.\\d+)$/);\n if (mapped) return isNonPublicIp(mapped[1]);\n return false;\n }\n\n return false;\n}\n\nfunction blockedHostname(hostname: string): boolean {\n const host = stripBrackets(hostname.toLowerCase().replace(/\\.$/, \"\"));\n if (BLOCKED_HOST_NAMES.has(host)) return true;\n if (host.endsWith(\".localhost\") || host.endsWith(\".local\") || host.endsWith(\".internal\") || host.endsWith(\".home.arpa\")) return true;\n if (isNonPublicIp(host)) return true;\n return false;\n}\n\nasync function doh(hostname: string, type: \"A\" | \"AAAA\"): Promise {\n const url = new URL(\"https://cloudflare-dns.com/dns-query\");\n url.searchParams.set(\"name\", hostname);\n url.searchParams.set(\"type\", type);\n const response = await fetch(url, {\n headers: { accept: \"application/dns-json\" },\n signal: AbortSignal.timeout(5_000),\n });\n if (!response.ok) throw new Error(`DoH ${type} failed with HTTP ${response.status}`);\n return await response.json() as DnsJson;\n}\n\nexport async function resolvePublicIps(hostname: string): Promise {\n const host = stripBrackets(hostname.toLowerCase().replace(/\\.$/, \"\"));\n if (blockedHostname(host)) throw new Error(`blocked hostname: ${host}`);\n if (parseIpv4(host) || host.includes(\":\")) return [host];\n\n const [a, aaaa] = await Promise.all([doh(host, \"A\"), doh(host, \"AAAA\")]);\n const answers = [...(a.Answer ?? []), ...(aaaa.Answer ?? [])];\n const ips = answers\n .filter((x) => x.type === 1 || x.type === 28)\n .map((x) => String(x.data ?? \"\"))\n .filter(Boolean);\n\n if (ips.length === 0) throw new Error(`no A/AAAA records resolved for ${host}`);\n const blocked = ips.filter(isNonPublicIp);\n if (blocked.length) throw new Error(`hostname ${host} resolved to blocked IP(s): ${blocked.join(\", \")}`);\n return [...new Set(ips)];\n}\n\nexport async function assertSafeUrl(value: string): Promise<{ url: string; host: string; ips: string[] }> {\n const url = new URL(value);\n if (![\"http:\", \"https:\"].includes(url.protocol)) throw new Error(\"only http/https URLs are allowed\");\n if (url.username || url.password) throw new Error(\"URL userinfo is blocked\");\n if (url.port && ![\"80\", \"443\"].includes(url.port)) throw new Error(`nonstandard port blocked: ${url.port}`);\n if (blockedHostname(url.hostname)) throw new Error(`blocked hostname: ${url.hostname}`);\n const ips = await resolvePublicIps(url.hostname);\n return { url: url.toString(), host: url.hostname, ips };\n}\n\nexport async function preflightTaskNetwork(task: BrowserTask) {\n const checked: Array<{ host: string; ips: string[] }> = [];\n const seen = new Set();\n\n const checkUrl = async (value: string) => {\n const result = await assertSafeUrl(value);\n if (!seen.has(result.host)) {\n seen.add(result.host);\n checked.push({ host: result.host, ips: result.ips });\n }\n };\n\n await checkUrl(task.targetUrl);\n for (const step of task.steps) {\n if (step.action === \"goto\" && step.url) await checkUrl(step.url);\n }\n for (const host of task.allowedHosts ?? []) {\n const clean = host.trim().toLowerCase();\n if (!clean || clean.includes(\"/\") || clean.includes(\"://\") || clean.includes(\"@\")) throw new Error(`invalid allowedHost: ${host}`);\n await checkUrl(`https://${clean}/`);\n }\n\n return checked;\n}\n",[],[2112,2105],"lib","b13f2e19-c7f4-4bcb-935b-b449dfda8b0e","types.ts","01a077bc-2b99-740d-9881-31a5d93088f8",["D",1788715215771],"export type Risk = \"read_only\" | \"reversible_write\" | \"consequential\";\nexport type ProviderName = \"kernel\" | \"browserless\";\nexport type EvidenceRetention = \"minimal\" | \"full\";\n\nexport type Step =\n | { action: \"goto\"; url?: string }\n | { action: \"wait_for\"; selector: string; timeoutMs?: number }\n | { action: \"click\"; selector: string; waitForNavigation?: boolean }\n | { action: \"fill\"; selector: string; text: string }\n | { action: \"fill_secret\"; selector: string; envVar: string }\n | { action: \"type\"; selector: string; text: string; delayMs?: number }\n | { action: \"select\"; selector: string; value: string }\n | { action: \"extract_text\"; selector: string; key: string }\n | { action: \"extract_attribute\"; selector: string; attribute: string; key: string }\n | { action: \"expect_text\"; text: string }\n | { action: \"expect_url\"; contains: string };\n\nexport type BrowserTask = {\n runId: string;\n targetUrl: string;\n risk: Risk;\n approved?: boolean;\n purpose: string;\n allowedHosts?: string[];\n preferredProvider?: ProviderName;\n evidenceRetention?: EvidenceRetention;\n steps: Step[];\n};\n\nexport type BrowserReceipt = {\n runId: string;\n provider: ProviderName;\n status: \"PASS\" | \"PARTIAL\" | \"FAIL\" | \"BLOCKED\";\n finalUrl?: string;\n outputs?: Record;\n checks?: Array>;\n error?: string;\n durationMs?: number;\n providerMeta?: Record;\n};\n",[],[2112,2114],"5257a923-09fd-451f-b4c5-eaa39b641d34","provider_health.ts","01a077bc-2b99-740d-9881-2d5b14823b37",["D",1788715215771],"import { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport type { BrowserReceipt, ProviderName } from \"./types.ts\";\nimport { POLICY } from \"./policy.ts\";\n\nexport async function providerAvailable(provider: ProviderName): Promise<{ available: boolean; reason?: string }> {\n const row = (await sqlite.execute({ sql: \"SELECT * FROM provider_health WHERE provider=?\", args: [provider] })).rows[0] as any;\n if (!row) return { available: true };\n if (row.circuit_state !== \"OPEN\") return { available: true };\n if (!row.open_until) return { available: false, reason: \"provider circuit open\" };\n const until = Date.parse(String(row.open_until));\n if (Number.isFinite(until) && Date.now() >= until) {\n await sqlite.execute({\n sql: \"UPDATE provider_health SET circuit_state='HALF_OPEN', updated_at=? WHERE provider=?\",\n args: [new Date().toISOString(), provider],\n });\n return { available: true };\n }\n return { available: false, reason: `provider circuit open until ${row.open_until}` };\n}\n\nexport async function recordProviderOutcome(provider: ProviderName, receipt: BrowserReceipt): Promise {\n const now = new Date().toISOString();\n if (receipt.status === \"PASS\" || receipt.status === \"PARTIAL\") {\n await sqlite.execute({\n sql: `UPDATE provider_health SET circuit_state='CLOSED', consecutive_failures=0, open_until=NULL,\n last_success_at=?, last_error=NULL, updated_at=? WHERE provider=?`,\n args: [now, now, provider],\n });\n return;\n }\n\n if (receipt.status === \"BLOCKED\") return;\n\n const row = (await sqlite.execute({ sql: \"SELECT consecutive_failures FROM provider_health WHERE provider=?\", args: [provider] })).rows[0] as any;\n const failures = Number(row?.consecutive_failures ?? 0) + 1;\n const shouldOpen = failures >= POLICY.providerCircuitFailureThreshold;\n const openUntil = shouldOpen ? new Date(Date.now() + POLICY.providerCircuitCooldownMs).toISOString() : null;\n await sqlite.execute({\n sql: `UPDATE provider_health SET circuit_state=?, consecutive_failures=?, open_until=?,\n last_failure_at=?, last_error=?, updated_at=? WHERE provider=?`,\n args: [shouldOpen ? \"OPEN\" : \"CLOSED\", failures, openUntil, now, receipt.error ?? receipt.status, now, provider],\n });\n}\n\nexport async function getProviderHealth() {\n return (await sqlite.execute(\"SELECT * FROM provider_health ORDER BY provider\")).rows;\n}\n",[],[2112,2121],"7ad8bdee-273b-4d7c-abcf-bb2f5d966c86","policy.ts","01a077bc-2b99-740d-9881-280da64a368a",["D",1788715215771],"import type { Risk } from \"./types.ts\";\n\nexport const POLICY = {\n version: \"0.5.0\",\n maxAttemptsPerTask: 2,\n retryDelayMs: 10_000,\n runningLeaseMs: 90_000,\n readOnlyFallbackAllowed: true,\n browserTimeoutMs: 25_000,\n providerFailureWindow: 5,\n providerCircuitFailureThreshold: 3,\n providerCircuitCooldownMs: 15 * 60_000,\n} as const;\n\nexport function isTerminalStatus(status: string): boolean {\n return [\"PASS\", \"PARTIAL\", \"FAIL\", \"BLOCKED\", \"CANCELLED\"].includes(status);\n}\n\nexport function decideDisposition(risk: Risk, receiptStatus: string, attemptCount: number) {\n const retryable = risk === \"read_only\" && receiptStatus === \"FAIL\" && attemptCount < POLICY.maxAttemptsPerTask;\n return {\n retryable,\n taskStatus: retryable ? \"PENDING\" : receiptStatus,\n retryDelayMs: retryable ? POLICY.retryDelayMs : 0,\n } as const;\n}\n\nexport async function withTimeout(promise: Promise, ms: number, label: string): Promise {\n let timer: number | undefined;\n const timeout = new Promise((_, reject) => {\n timer = setTimeout(() => reject(new Error(`${label} timed out after ${ms}ms`)), ms);\n });\n try {\n return await Promise.race([promise, timeout]);\n } finally {\n if (timer !== undefined) clearTimeout(timer);\n }\n}\n",[],[2112,2128],"7d31d3a0-a36d-43fe-acc2-6664bd750149","neuron_evolution.ts","7063f21a-aa18-11f1-84d5-1607ee4eb77e",["D",1788715736660],"export type SelectionCandidate = {\n agentId: string;\n netRevenue: number;\n operationallyEligible: boolean;\n terminal: boolean;\n};\n\nexport type SelectionState = \"INCUBATION\" | \"DEAD\" | \"SURVIVOR\";\n\nexport type SelectionResult = SelectionCandidate & {\n state: SelectionState;\n benchmark: number;\n rank?: number;\n offspringCount: number;\n};\n\nexport const MUTATION_ARCHETYPES = [\n \"faithful_replica\",\n \"tool_or_provider_mutation\",\n \"channel_or_market_mutation\",\n \"opposite_hypothesis\",\n \"causal_ablation\",\n] as const;\n\nexport function classifyCandidate(candidate: SelectionCandidate): SelectionResult {\n if (!candidate.operationallyEligible || !candidate.terminal) {\n return { ...candidate, state: \"INCUBATION\", benchmark: 0, offspringCount: 0 };\n }\n if (candidate.netRevenue <= 0) {\n return { ...candidate, state: \"DEAD\", benchmark: 0, offspringCount: 0 };\n }\n return { ...candidate, state: \"SURVIVOR\", benchmark: candidate.netRevenue, offspringCount: 1 };\n}\n\nexport function rankGeneration(candidates: SelectionCandidate[]): SelectionResult[] {\n const classified = candidates.map(classifyCandidate);\n const survivors = classified\n .filter((x) => x.state === \"SURVIVOR\")\n .sort((a, b) => b.netRevenue - a.netRevenue || a.agentId.localeCompare(b.agentId));\n\n const count = survivors.length;\n survivors.forEach((item, index) => {\n item.rank = index + 1;\n if (count === 1) item.offspringCount = 5;\n else {\n const percentile = index / count;\n item.offspringCount = percentile < 0.2 ? 5 : percentile < 0.5 ? 3 : 1;\n }\n });\n\n const byId = new Map(survivors.map((x) => [x.agentId, x]));\n return classified.map((x) => byId.get(x.agentId) ?? x);\n}\n\nexport function offspringPlan(parent: SelectionResult) {\n if (parent.state !== \"SURVIVOR\" || parent.offspringCount <= 0) return [];\n const count = Math.min(parent.offspringCount, MUTATION_ARCHETYPES.length);\n return Array.from({ length: count }, (_, index) => ({\n parentId: parent.agentId,\n childIndex: index + 1,\n mutation: MUTATION_ARCHETYPES[index],\n targetNetRevenue: parent.benchmark,\n }));\n}\n",[],[2112,2135],"ab63c4ed-e481-4aab-bec5-fdae429eae8c","kaggriculture_opponent_market.ts","1f8c3fca-aa39-11f1-b627-1607ee4eb77e",37,["D",1788729774423],"import type { AgentFn } from \"./kaggriculture_agents_v1.ts\";\nimport { withSellPolicy, AFTER_TOWN_HOURS } from \"./kaggriculture_market_mutations.ts\";\n\nfunction publicCropTiles(farm: any, crop: string) {\n let n = 0;\n for (const row of farm?.tiles ?? []) for (const tile of row ?? []) {\n if (tile && typeof tile === \"object\" && tile.kind === \"PLANT\" && tile.crop === crop) n++;\n }\n return n;\n}\n\nexport function opponentAwareCarrotMarket(baseFactory: () => AgentFn, threshold = 3): AgentFn {\n const aggressive = baseFactory(); // original immediate SELL-all behavior\n const reserve = withSellPolicy(baseFactory(), {\n name: \"reserve60_after\",\n item: \"CARROT\",\n allowedHours: AFTER_TOWN_HOURS,\n lotSize: 8,\n reservePrice: 60,\n forceShedLoad: 85,\n liquidationDay: 29,\n });\n\n return (obs: any) => {\n const me = Number(obs?.player ?? 0);\n const farms = obs?.farms ?? [];\n const opponent = farms.length === 2 ? farms[1 - me] : null;\n const exposure = publicCropTiles(opponent, \"CARROT\");\n // If the opponent is visibly committed to carrots, withholding supply raises\n // THEIR sale price too. Do not subsidize a same-product rival.\n return exposure >= threshold ? aggressive(obs) : reserve(obs);\n };\n}\n",[],[2112,2142],"9eb933ab-1945-4b60-8472-d88a83dc7ac7","kaggriculture_market_mutations.ts","f05a1d3a-aa38-11f1-bcf5-1607ee4eb77e",33,["D",1788729695212],"import type { AgentFn } from \"./kaggriculture_agents_v1.ts\";\n\nexport type SellPolicy = {\n name: string;\n item?: string;\n allowedHours?: number[] | null;\n lotSize?: number | null;\n reservePrice?: number | null;\n forceShedLoad?: number;\n liquidationDay?: number;\n};\n\nfunction shedTotal(shed: Record | undefined) {\n return Object.values(shed ?? {}).reduce((a, b) => a + Number(b || 0), 0);\n}\n\nexport function withSellPolicy(base: AgentFn, policy: SellPolicy): AgentFn {\n const item = policy.item ?? \"CARROT\";\n const allowed = policy.allowedHours ? new Set(policy.allowedHours) : null;\n const forceLoad = policy.forceShedLoad ?? 80;\n const liquidationDay = policy.liquidationDay ?? 29;\n\n return (obs: any) => {\n const action = base(obs);\n const market = Array.isArray(action?.market) ? action.market.filter((o: any) => !(Array.isArray(o) && o[0] === \"SELL\" && o[1] === item)) : [];\n const shed = obs?.private?.shed ?? {};\n const qty = Math.max(0, Math.floor(Number(shed[item] ?? 0)));\n if (qty <= 0) return { ...action, market };\n\n const day = Number(obs?.day ?? 0);\n const hour = Number(obs?.hour ?? 0);\n const price = Number(obs?.market?.prices?.[item] ?? 0);\n const mustLiquidate = day >= liquidationDay || shedTotal(shed) >= forceLoad;\n const timeOk = !allowed || allowed.has(hour);\n const priceOk = policy.reservePrice == null || price >= policy.reservePrice;\n\n if ((mustLiquidate || (timeOk && priceOk)) && market.length < 10) {\n const lot = mustLiquidate || policy.lotSize == null ? qty : Math.min(qty, policy.lotSize);\n // SELL first so proceeds can finance any later BUY_SEED/BUY_PRODUCT order.\n market.unshift([\"SELL\", item, lot]);\n }\n return { ...action, market: market.slice(0, 10) };\n };\n}\n\nexport const AFTER_TOWN_HOURS = [1, 5, 9, 13, 17, 21];\nexport const BEFORE_TOWN_HOURS = [0, 4, 8, 12, 16, 20];\n",[],[2112,2150],"9d2d3dd5-7d6c-4975-802b-0bb6322ae998","kaggriculture_livestock_portfolio.ts","0136cb6e-aa41-11f1-9721-1607ee4eb77e",141,["D",1788745336610],"export type AnimalName='COW'|'SHEEP'|'GOOSE';\nexport type ProductName='MILK'|'WOOL'|'EGG'|'WHEAT';\n\nexport const ANIMAL_ECONOMICS:Record}>={\n GOOSE:{cost:300,first:4,interval:1,product:'EGG'},\n COW:{cost:400,first:8,interval:2,product:'MILK'},\n SHEEP:{cost:500,first:6,interval:3,product:'WOOL'},\n};\n\nconst MARKET:any={\n WHEAT:{base:25,I0:10000,T:400,below_func:'sqrt',below_target:.80,above_func:'log',above_target:.20},\n EGG:{base:50,I0:10000,T:332,below_func:'hinge',below_target:.40,above_func:'log',above_target:.20},\n MILK:{base:160,I0:10000,T:122,below_func:'sqrt',below_target:.60,above_func:'linear',above_target:1.60},\n WOOL:{base:200,I0:10000,T:105,below_func:'log',below_target:.20,above_func:'sq',above_target:3.20},\n};\nconst SHOPS:Record={\n BAKERY:['EGG','WHEAT'],PIZZA_SHOP:['MILK','TOMATO','WHEAT'],BRUNCH_SPOT:['EGG','WHEAT','STRAWBERRY'],YARN_STORE:['WOOL'],ICE_CREAM_SHOP:['STRAWBERRY','MILK','WHEAT'],PET_CAFE:['CARROT'],SMOOTHIE_SHOP:['STRAWBERRY','MILK'],FARMERS_MARKET:['WHEAT','CARROT','TOMATO','STRAWBERRY']\n};\nconst SHOP_NAMES=Object.keys(SHOPS).sort();\nconst hingeGain=8;\nfunction shape(fn:string,x:number,T:number){x=Math.max(0,x);if(fn==='linear')return x;if(fn==='sq')return x*x;if(fn==='sqrt')return Math.sqrt(x);if(fn==='log')return Math.log(1+x);if(fn==='hinge'){const u=x/T;return u+hingeGain*Math.max(0,u-1)**2;}return x;}\nexport function marketPrice(item:ProductName,inventory:number){const p=MARKET[item];const x=Math.abs(inventory-p.I0);const below=inventorys+shopMultiplier(sh,item),0)*shopTicks;\n // Town center consumes every product except fertilizer once per day.\n const centerTicks=Math.max(0,Math.floor((episodeSteps-1)/24)-Math.ceil(step/24)+1);\n demand += centerTicks;\n // Future shop slots unlock at end of days 3,6,...24. Expected contribution is uniform over 8 shop types.\n const avgMult=SHOP_NAMES.reduce((s,sh)=>s+shopMultiplier(sh,item),0)/SHOP_NAMES.length;\n for(const unlockDay of [3,6,9,12,15,18,21,24]){\n if(unlockDay<=day || known.length>=8)continue;\n const unlockStep=unlockDay*24;\n const ticks=Math.max(0,Math.floor((episodeSteps-1)/4)-Math.ceil(unlockStep/4)+1);\n demand += avgMult*ticks;\n }\n return demand;\n}\n\nexport function caredProductionUnits(animal:AnimalName,currentDay:number,horizonDay=30){\n const a=ANIMAL_ECONOMICS[animal];const first=currentDay+a.first;if(first>=horizonDay)return 0;\n const events=1+Math.floor((horizonDay-1-first)/a.interval);\n // With daily feed+care, pending care bonus makes productive refreshes approximately 2 units/event.\n return Math.max(0,2*events);\n}\n\nexport function visibleAnimalSupplyRemaining(farm:any,animal:AnimalName,currentDay:number,horizonDay=30){\n const a=ANIMAL_ECONOMICS[animal];let units=0;\n for(const row of farm?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object'||t.animal!==animal)continue;const placed=Number(t.placed_day??currentDay);const first=placed+a.first;let next=first;while(next=horizonDay)continue;units+=2*(1+Math.floor((horizonDay-1-next)/a.interval));}\n return units;\n}\n\nexport type AnimalScoreInput={\n obs:any; animal:AnimalName; cashBuffer?:number; horizonDay?:number; discountPerDay?:number; strategicWeight?:number;\n liquidityFeedFloatDays?:number; liquidityStockCredit?:boolean;\n};\n\n/** Marginal NPV-like score for purchasing one more animal now. Designed for ranking, not accounting truth. */\nexport function scoreAnimalInvestment(x:AnimalScoreInput){\n const {obs,animal}=x;const a=ANIMAL_ECONOMICS[animal];const day=Number(obs?.day??0),h=x.horizonDay??30;\n const product=a.product;const inv=Number(obs?.market?.inventory?.[product]??10000);const wheatInv=Number(obs?.market?.inventory?.WHEAT??10000);\n const currentPrice=marketPrice(product,inv);const wheatPrice=marketPrice('WHEAT',wheatInv);\n const units=caredProductionUnits(animal,day,h);const feedDays=Math.max(0,h-day);\n const townDemand=expectedTownDemandRemaining(obs,product);\n const me=Number(obs?.player??0),farms=obs?.farms??[],mine=farms?.[me]??null,opp=farms.length===2?farms[1-me]:null;\n const ownCommittedSupply=visibleAnimalSupplyRemaining(mine,animal,day,h);\n const opponentSupply=visibleAnimalSupplyRemaining(opp,animal,day,h);\n // Cournot-like marginal accounting: compare the projected market before vs after the extra animal.\n // Extra supply changes the price received on OUR already-committed output and on the rival's output too.\n const baseProjectedInv=inv - 0.5*townDemand + 0.5*ownCommittedSupply + 0.5*opponentSupply;\n const priceBeforeMarginal=marketPrice(product,baseProjectedInv);\n const projectedInv=baseProjectedInv + 0.5*units;\n const projectedPrice=marketPrice(product,projectedInv);\n const avgPrice=(currentPrice+projectedPrice)/2;\n const disc=Math.pow(x.discountPerDay??0.997,Math.max(0,a.first));\n const newUnitsRevenue=units*avgPrice*disc;\n const ownPriceImpact=0.5*ownCommittedSupply*(projectedPrice-priceBeforeMarginal)*disc;\n const rivalPriceImpact=0.5*opponentSupply*(projectedPrice-priceBeforeMarginal)*disc;\n const relativeMarketContribution=newUnitsRevenue+ownPriceImpact-(x.strategicWeight??1.0)*rivalPriceImpact;\n const feedCost=feedDays*wheatPrice;\n const cash=Number(farms?.[me]?.money??0),buffer=x.cashBuffer??400;\n const postPurchaseCash=cash-a.cost;\n // Financial-feasibility gate mirrored from the reference scheduler: a positive-NPV animal is useless if the farm cannot fund its feed float and capital sequence.\n let flock=0;for(const row of mine?.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal)flock++;\n for(const k of ['COW','SHEEP','GOOSE'])flock+=Number(obs?.private?.shed?.[k]??0);\n const boughtLand=Math.max(0,(mine?.unlocked_quadrants??['NW']).length-1);\n const landPrices=[1000,2000,4000];\n const landReserve=(flock>=10 && boughtLand<2)?landPrices[boughtLand]:0;\n const feedFloatDays=x.liquidityFeedFloatDays??16;\n const ownedWheat=Number(obs?.private?.shed?.WHEAT??0)+(obs?.private?.inventories??[]).reduce((s:number,inv:any)=>s+Number(inv?.WHEAT??0),0);\n const stockCredit=(x.liquidityStockCredit??false)?ownedWheat*wheatPrice:0;\n const requiredFeedLiquidity=Math.max(0,(flock+1)*wheatPrice*feedFloatDays-stockCredit);\n const requiredLiquidity=a.cost+buffer+landReserve+requiredFeedLiquidity;\n const financeable=cash>=requiredLiquidity;\n const liquidityPenalty=financeable?0:1e6+(requiredLiquidity-cash)*1000;\n const net=relativeMarketContribution-a.cost-feedCost-liquidityPenalty;\n const timeToCash=a.first;\n return {animal,product,day,units,events:units/2,timeToCash,currentPrice,priceBeforeMarginal,projectedPrice,townDemand,ownCommittedSupply,opponentSupply,feedDays,wheatPrice,newUnitsRevenue,ownPriceImpact,rivalPriceImpact,relativeMarketContribution,animalCost:a.cost,feedCost,flock,boughtLand,landReserve,feedFloatDays,ownedWheat,stockCredit,requiredFeedLiquidity,requiredLiquidity,financeable,liquidityPenalty,net,roi:net/Math.max(1,a.cost),postPurchaseCash};\n}\n\nexport function rankAnimalInvestments(obs:any,opts?:Omit){\n return (['COW','SHEEP','GOOSE'] as AnimalName[]).map(animal=>scoreAnimalInvestment({obs,animal,...opts})).sort((a,b)=>b.net-a.net);\n}\n\nexport function choosePastureAnimal(obs:any,opts?:Omit){\n const xs=(['COW','SHEEP'] as AnimalName[]).map(animal=>scoreAnimalInvestment({obs,animal,...opts})).sort((a,b)=>b.net-a.net);\n return {choice:xs[0].net>0?xs[0].animal:null,ranking:xs};\n}\n",[],[2112,2158],"841bfd60-3bc8-4914-a54e-a0f010480893","kaggriculture_league.ts","476f455e-aa3f-11f1-8e55-1607ee4eb77e",48,["D",1788732418278],"export type MatchResult = \"WIN\" | \"LOSS\" | \"TIE\";\nexport type LeagueRole = \"MAIN\" | \"HISTORICAL\" | \"EXPLOITER\" | \"SPECIALIST\" | \"REFERENCE\";\n\nexport type LeagueMember = {\n id: string;\n role: LeagueRole;\n frozen?: boolean;\n lineage?: string;\n};\n\nexport type PairStats = { wins: number; losses: number; ties: number };\n\nexport class EmpiricalLeague {\n members = new Map();\n private pairs = new Map();\n\n add(member: LeagueMember) {\n this.members.set(member.id, { ...member });\n }\n\n private key(a: string, b: string) { return `${a}→${b}`; }\n\n record(a: string, b: string, result: MatchResult) {\n if (!this.members.has(a) || !this.members.has(b)) throw new Error(\"UNKNOWN_MEMBER\");\n const ab = this.pairs.get(this.key(a,b)) ?? { wins: 0, losses: 0, ties: 0 };\n const ba = this.pairs.get(this.key(b,a)) ?? { wins: 0, losses: 0, ties: 0 };\n if (result === \"WIN\") { ab.wins++; ba.losses++; }\n else if (result === \"LOSS\") { ab.losses++; ba.wins++; }\n else { ab.ties++; ba.ties++; }\n this.pairs.set(this.key(a,b), ab);\n this.pairs.set(this.key(b,a), ba);\n }\n\n stats(a: string, b: string): PairStats {\n return { ...(this.pairs.get(this.key(a,b)) ?? { wins: 0, losses: 0, ties: 0 }) };\n }\n\n /** Smoothed score in [0,1]. Ties count half; Beta(1,1)-style pseudocounts prevent zero-information extremes. */\n score(a: string, b: string): number {\n const s = this.stats(a,b);\n return (s.wins + 0.5*s.ties + 1) / (s.wins + s.losses + s.ties + 2);\n }\n\n /**\n * PFSP-like opponent weighting: prioritize opponents the focal policy does not reliably beat.\n * uncertaintyBonus keeps under-tested opponents alive so the pool does not collapse to one nemesis.\n */\n pfspWeights(focal: string, opponents: string[], alpha = 2, uncertaintyBonus = 0.25) {\n const raw = opponents.map(id => {\n const s = this.stats(focal,id);\n const n = s.wins+s.losses+s.ties;\n const p = this.score(focal,id);\n const difficulty = Math.pow(Math.max(1e-6, 1-p), alpha);\n const uncertainty = 1/Math.sqrt(n+1);\n return { id, weight: difficulty + uncertaintyBonus*uncertainty, n, p };\n });\n const total = raw.reduce((a,x)=>a+x.weight,0) || 1;\n return raw.map(x => ({ ...x, probability: x.weight/total }));\n }\n\n /** Weaknesses are opponents against which focal has lowest smoothed score. */\n weaknesses(focal: string, limit=5) {\n return [...this.members.keys()]\n .filter(id => id !== focal)\n .map(id => ({ id, score: this.score(focal,id), stats: this.stats(focal,id) }))\n .sort((a,b)=>a.score-b.score)\n .slice(0,limit);\n }\n\n /** Preserve frozen historical snapshots instead of overwriting them: catastrophic forgetting detector/control. */\n freezeSnapshot(sourceId: string, snapshotId: string) {\n const src = this.members.get(sourceId);\n if (!src) throw new Error(\"UNKNOWN_SOURCE\");\n this.add({ id: snapshotId, role: \"HISTORICAL\", frozen: true, lineage: src.lineage ?? sourceId });\n }\n\n payoffMatrix(ids: string[]) {\n return ids.map(a => ids.map(b => a===b ? 0 : 2*this.score(a,b)-1));\n }\n}\n\n/**\n * Regret-matching approximation of a symmetric zero-sum meta-strategy.\n * Useful as a low-cost PSRO meta-solver over an empirical payoff matrix.\n */\nexport function regretMatchingMix(payoff: number[][], iterations=20000) {\n const n = payoff.length;\n if (!n || payoff.some(r=>r.length!==n)) throw new Error(\"SQUARE_MATRIX_REQUIRED\");\n let regrets = Array(n).fill(0);\n let strategySum = Array(n).fill(0);\n let strategy = Array(n).fill(1/n);\n\n for (let t=0;tMath.max(0,r));\n const z = pos.reduce((a,b)=>a+b,0);\n strategy = z>1e-12 ? pos.map(x=>x/z) : Array(n).fill(1/n);\n for (let i=0;i row.reduce((s,v,j)=>s+v*strategy[j],0));\n const currentValue = strategy.reduce((s,p,i)=>s+p*actionValues[i],0);\n for (let i=0;ia+b,0) || 1;\n return strategySum.map(x=>x/z);\n}\n\nexport function mixEntropy(mix:number[]) {\n return -mix.filter(p=>p>0).reduce((s,p)=>s+p*Math.log(p),0);\n}\n",[],[2112,2166],"336a164c-cacf-4eed-b42f-572da6afe34e","kaggriculture_farm_economics.ts","5a82cd1e-aa3f-11f1-8146-1607ee4eb77e",50,["D",1788732450327],"export function mean(xs:number[]) {\n return xs.length ? xs.reduce((a,b)=>a+b,0)/xs.length : NaN;\n}\n\n/** Lower-tail CVaR for profit distributions. alpha=0.1 means average of worst 10%. */\nexport function lowerTailCVaR(xs:number[], alpha=0.1) {\n if (!xs.length) return NaN;\n const a=Math.min(1,Math.max(1/xs.length,alpha));\n const sorted=[...xs].sort((x,y)=>x-y);\n const k=Math.max(1,Math.ceil(sorted.length*a));\n return mean(sorted.slice(0,k));\n}\n\nexport function riskAdjustedProfit(xs:number[], lambda=0.25, alpha=0.1) {\n const m=mean(xs), c=lowerTailCVaR(xs,alpha);\n return m - lambda*(m-c);\n}\n\n/** Finite-difference shadow value: extra net profit generated by one extra unit of a resource. */\nexport function shadowValue(baseProfit:number, expandedProfit:number, extraUnits:number, extraCost=0) {\n if (extraUnits<=0) throw new Error(\"POSITIVE_EXTRA_UNITS_REQUIRED\");\n return (expandedProfit-baseProfit-extraCost)/extraUnits;\n}\n\n/**\n * Generic sell-vs-hold decision. Holding preserves an option to sell later but has operating/opportunity cost.\n * strategicExternality is positive when withholding benefits the opponent or harms our win probability.\n */\nexport function sellHoldDecision(x:{\n spotPrice:number;\n expectedFuturePrice:number;\n holdingCost:number;\n spoilageProbability?:number;\n salvageIfSpoiled?:number;\n discountFactor?:number;\n strategicExternality?:number;\n optionFlexibilityValue?:number;\n}) {\n const spoil=Math.min(1,Math.max(0,x.spoilageProbability??0));\n const salvage=x.salvageIfSpoiled??0;\n const disc=x.discountFactor??1;\n const futureExpected=(1-spoil)*x.expectedFuturePrice + spoil*salvage;\n const holdValue=disc*futureExpected - x.holdingCost + (x.optionFlexibilityValue??0) - (x.strategicExternality??0);\n return {\n sellNowValue:x.spotPrice,\n holdValue,\n action:x.spotPrice>=holdValue ? \"SELL\" as const : \"HOLD\" as const,\n threshold:holdValue,\n advantage:x.spotPrice-holdValue,\n };\n}\n\n/** Reserve is economically valuable if it prevents an interruption whose expected damage exceeds carrying cost. */\nexport function reserveOptionValue(x:{\n interruptionProbability:number;\n lossIfInterrupted:number;\n reserveCarryingCost:number;\n reservePurchaseCost:number;\n recoveryValue?:number;\n}) {\n const p=Math.min(1,Math.max(0,x.interruptionProbability));\n const avoidedLoss=p*x.lossIfInterrupted;\n const benefit=avoidedLoss+(x.recoveryValue??0);\n const cost=x.reserveCarryingCost+x.reservePurchaseCost;\n return { benefit, cost, netOptionValue:benefit-cost, shouldHoldReserve:benefit>cost };\n}\n\nexport type ResourceBottleneck = {\n resource:string;\n shadowValue:number;\n utilization:number;\n};\n\n/** Prioritize binding resources with positive marginal value; negative shadow value means expansion destroys value. */\nexport function rankBottlenecks(xs:ResourceBottleneck[]) {\n return [...xs].sort((a,b)=>{\n const sa=a.shadowValue*Math.max(0,a.utilization);\n const sb=b.shadowValue*Math.max(0,b.utilization);\n return sb-sa;\n });\n}\n\n/** Finite-horizon liquidation pressure. 0 = no pressure; 1 = liquidate aggressively. */\nexport function liquidationPressure(turnsRemaining:number, leadTime:number, terminalSalvage:number, currentPrice:number) {\n if (turnsRemaining<=0) return 1;\n const horizonRisk=Math.max(0,Math.min(1,(leadTime+1)/turnsRemaining));\n const salvagePenalty=currentPrice<=0 ? 1 : Math.max(0,Math.min(1,1-terminalSalvage/currentPrice));\n return Math.max(0,Math.min(1,0.65*horizonRisk+0.35*salvagePenalty));\n}\n\n/**\n * Capital-efficiency metric: return generated per unit of scarce capital tied up.\n * Useful when two strategies have similar win rate but one requires much larger herd/land/feed commitments.\n */\nexport function returnOnCommittedCapital(netIncrementalProfit:number, committedCapital:number) {\n if (committedCapital<=0) return netIncrementalProfit>0 ? Infinity : 0;\n return netIncrementalProfit/committedCapital;\n}\n",[],[2112,2174],"e2b3db4d-0adf-4758-9976-66328c59ed9d","kaggriculture_engine_loader.ts","6074d468-aa37-11f1-8d3f-1607ee4eb77e",20,["D",1788729024315],"const RAW_ROOT = \"https://raw.githubusercontent.com/Kaggle/kaggle-environments/master/kaggle_environments/envs/kaggriculture/\";\nconst ENGINE_BASE = `${RAW_ROOT}visualizer/playable/src/engine/`;\nconst SPEC_URL = `${RAW_ROOT}kaggriculture.json`;\nconst ENGINE_NAMES = [\"types\", \"constants\", \"market\", \"rng\", \"state\", \"interpreter\"] as const;\ntype EngineName = typeof ENGINE_NAMES[number];\n\nfunction toDataUrl(source: string): string {\n const bytes = new TextEncoder().encode(source);\n let binary = \"\";\n for (let i = 0; i < bytes.length; i += 0x8000) {\n binary += String.fromCharCode(...bytes.subarray(i, i + 0x8000));\n }\n return `data:application/typescript;base64,${btoa(binary)}`;\n}\n\nfunction stripTypeImports(src: string): string {\n return src.replace(/import\\s+type\\s+\\{[\\s\\S]*?\\}\\s+from\\s+['\"][^'\"]+['\"];?\\s*/g, \"\");\n}\n\nexport async function loadOfficialKaggricultureEngine() {\n const raw = new Map();\n for (const name of ENGINE_NAMES) {\n const res = await fetch(`${ENGINE_BASE}${name}.ts`);\n if (!res.ok) throw new Error(`fetch official ${name}.ts failed: ${res.status}`);\n raw.set(name, await res.text());\n }\n const specRes = await fetch(SPEC_URL);\n if (!specRes.ok) throw new Error(`fetch official kaggriculture.json failed: ${specRes.status}`);\n const specText = await specRes.text();\n JSON.parse(specText);\n\n const urls = new Map();\n for (const name of ENGINE_NAMES) {\n let src = stripTypeImports(raw.get(name)!);\n if (name === \"constants\") {\n src = src.replace(\n /import\\s+spec\\s+from\\s+['\"]\\.\\.\\/\\.\\.\\/\\.\\.\\/\\.\\.\\/kaggriculture\\.json['\"];?\\s*/,\n `const spec = ${specText};\\n`,\n );\n }\n src = src.replace(\n /import\\s+\\{([\\s\\S]*?)\\}\\s+from\\s+['\"]\\.\\/([A-Za-z0-9_-]+)['\"];?\\s*/g,\n (_all, symbols: string, dep: string) => {\n const depUrl = urls.get(dep as EngineName);\n if (!depUrl) throw new Error(`${name}: dependency ${dep} not compiled`);\n return `import {${symbols}} from ${JSON.stringify(depUrl)};\\n`;\n },\n );\n const unresolved = [...src.matchAll(/from\\s+['\"](\\.\\.?\\/[^'\"]+)['\"]/g)].map((m) => m[1]);\n if (unresolved.length) throw new Error(`${name}: unresolved imports ${unresolved.join(\", \")}`);\n urls.set(name, toDataUrl(src));\n }\n\n const [types, constants, market, rng, state, interpreter] = await Promise.all([\n import(urls.get(\"types\")!),\n import(urls.get(\"constants\")!),\n import(urls.get(\"market\")!),\n import(urls.get(\"rng\")!),\n import(urls.get(\"state\")!),\n import(urls.get(\"interpreter\")!),\n ]);\n\n return {\n types,\n constants,\n market,\n rng,\n state,\n interpreter,\n sourceMeta: Object.fromEntries(ENGINE_NAMES.map((name) => [name, { bytes: raw.get(name)!.length }])),\n };\n}\n\nexport function observationFor(state: any, player: number) {\n return {\n player,\n step: state.step,\n day: state.day,\n hour: state.hour,\n numAgents: state.numAgents,\n farms: state.farms,\n private: state.privates[player],\n market: state.market,\n town: state.town,\n };\n}\n",[],[2112,2182],"6835ce70-1a2f-421b-bb9e-5a0f960202ef","kaggriculture_bayesian_router.ts","505ca7ba-aa3f-11f1-8cab-1607ee4eb77e",49,["D",1788732433289],"export type Hypothesis = { id: string; prior: number };\nexport type LikelihoodEvidence = { name: string; likelihood: Record };\n\nexport class BayesianRouter {\n private posterior = new Map();\n private evidenceCount = 0;\n\n constructor(hypotheses: Hypothesis[]) {\n if (!hypotheses.length) throw new Error(\"EMPTY_HYPOTHESES\");\n const sum = hypotheses.reduce((s,h)=>s+Math.max(0,h.prior),0);\n if (sum <= 0) throw new Error(\"NONPOSITIVE_PRIOR_SUM\");\n for (const h of hypotheses) this.posterior.set(h.id, Math.max(0,h.prior)/sum);\n }\n\n distribution() {\n return [...this.posterior.entries()].map(([id,p])=>({id,p})).sort((a,b)=>b.p-a.p);\n }\n\n update(ev: LikelihoodEvidence, floor=1e-6) {\n const before = new Map(this.posterior);\n let z = 0;\n for (const [id,p] of this.posterior) {\n const like = Math.max(floor, Math.min(1, ev.likelihood[id] ?? floor));\n const q = p * like;\n this.posterior.set(id,q);\n z += q;\n }\n if (z <= 0) throw new Error(\"ZERO_POSTERIOR_MASS\");\n for (const [id,q] of this.posterior) this.posterior.set(id,q/z);\n this.evidenceCount++;\n return {\n evidence: ev.name,\n before: [...before.entries()].map(([id,p])=>({id,p})),\n after: this.distribution(),\n kl: klDivergence(this.posterior,before),\n entropy: this.entropy(),\n };\n }\n\n entropy() {\n let h=0;\n for (const p of this.posterior.values()) if (p>0) h -= p*Math.log(p);\n return h;\n }\n\n confidence() { return this.distribution()[0]?.p ?? 0; }\n best() { return this.distribution()[0]; }\n count() { return this.evidenceCount; }\n}\n\nfunction klDivergence(p: Map, q: Map) {\n let out=0;\n for (const [id,pv] of p) {\n const qv=Math.max(1e-12,q.get(id)??1e-12);\n if (pv>0) out += pv*Math.log(pv/qv);\n }\n return out;\n}\n\nexport type MacroReplanInput = {\n previousBest?: string;\n currentBest: string;\n posteriorConfidence: number;\n informationGainKL: number;\n event: \"NEW_DAY\"|\"NEW_SHOP\"|\"PRICE_SHIFT\"|\"OPPONENT_SHIFT\"|\"HORIZON_PHASE\"|\"NO_EVENT\";\n};\n\n/** Event-driven control: avoid turn-by-turn thrashing. */\nexport function shouldReplanMacro(x: MacroReplanInput) {\n if (x.event === \"NO_EVENT\") return false;\n if (x.previousBest && x.previousBest !== x.currentBest && x.posteriorConfidence >= 0.60) return true;\n if (x.informationGainKL >= 0.08 && x.posteriorConfidence >= 0.55) return true;\n if (x.event === \"HORIZON_PHASE\") return true;\n return false;\n}\n\n/** Simple feature-to-likelihood adapter for opponent family classification. */\nexport function opponentFamilyLikelihood(features: {\n cropShare:number;\n livestockCount:number;\n recentSellIntensity:number;\n landUnlocked:number;\n}) {\n const clamp=(x:number)=>Math.max(0.02,Math.min(0.98,x));\n const crop = clamp(0.25 + 0.65*features.cropShare - 0.03*features.livestockCount);\n const livestock = clamp(0.15 + 0.055*features.livestockCount + 0.05*features.landUnlocked - 0.35*features.cropShare);\n const marketHeavy = clamp(0.15 + 0.75*features.recentSellIntensity);\n const mixed = clamp(1 - Math.max(crop,livestock,marketHeavy)*0.72);\n return { CROP: crop, LIVESTOCK: livestock, MARKET_HEAVY: marketHeavy, MIXED: mixed };\n}\n",[],[2112,2190],"9fc2470f-749c-4d80-b0e8-d181bc8e7dc5","kaggriculture_battle_dossier_v1.ts","3641ed70-aa55-11f1-a952-1607ee4eb77e",102,["D",1788741838435],"export type DailyPoint={day:number,subject:{money:number,animals?:Record,crops?:Record},opponent:{money:number,animals?:Record,crops?:Record}};\nexport type Telemetry={hire?:number,land?:number,buyAnimal?:Record,buySeed?:Record,buyProduct?:Record,sell?:Record,first?:Record};\nexport type DossierInput={matchId:string,subjectId:string,opponentId:string,seed:number,seat:number,finalBank:number,opponentBank:number,timeline:DailyPoint[],telemetry:Telemetry,opponentTelemetry?:Telemetry};\nconst total=(o:Record|undefined)=>Object.values(o??{}).reduce((a,b)=>a+(Number(b)||0),0);\nconst diversity=(o:Record|undefined)=>Object.values(o??{}).filter(v=>Number(v)>0).length;\nexport function buildBattleDossier(x:DossierInput){const role=x.finalBank>x.opponentBank?'WINNER':x.finalBank({day:p.day,cash:+p.subject.money,opp:+p.opponent.money,margin:+p.subject.money-(+p.opponent.money)}));const min=pts.reduce((a,b)=>b.cashb.margin>a.margin?b:a,pts[0]??{day:0,margin:0});const maxDef=pts.reduce((a,b)=>b.marginp.margin>0)?.day??null,firstTrail=pts.find(p=>p.margin<0)?.day??null;const ownSell=x.telemetry.sell??{},oppSell=x.opponentTelemetry?.sell??{};const ownAnimals=x.telemetry.buyAnimal??{},oppAnimals=x.opponentTelemetry?.buyAnimal??{};const ownDiversity=diversity(ownSell),oppDiversity=diversity(oppSell);const strengths:string[]=[];const failures:string[]=[];const hypotheses:string[]=[];const counterfactuals:string[]=[];\nif(role==='WINNER')strengths.push(`terminal advantage ${x.finalBank-x.opponentBank}`);else failures.push(`terminal deficit ${x.opponentBank-x.finalBank}`);\nif(ownDiversity>oppDiversity)strengths.push(`broader realized sales mix ${ownDiversity} vs ${oppDiversity} product families`);if(ownDiversitytotal(oppAnimals))strengths.push(`higher livestock acquisition ${total(ownAnimals)} vs ${total(oppAnimals)}`);\nif((x.telemetry.land??0)<(x.opponentTelemetry?.land??0))failures.push(`lower land expansion ${x.telemetry.land??0} vs ${x.opponentTelemetry?.land??0}`);\nif(maxDef.margin<0&&maxDef.day>=12)failures.push(`largest observed deficit ${Math.abs(maxDef.margin)} occurred late at day ${maxDef.day}, consistent with compounding disadvantage`);\nconst early=pts.find(p=>p.day>=9),late=pts.find(p=>p.day>=21);if(early&&late){const growth=(late.cash-early.cash),oppGrowth=(late.opp-early.opp);if(growth${late.day}: ${growth} vs opponent ${oppGrowth}`);else strengths.push(`cash growth day ${early.day}->${late.day}: ${growth} vs opponent ${oppGrowth}`)}\nif(ownDiversity=12)hypotheses.push('loss may be driven by positive feedback: lower early capacity -> lower cashflow -> fewer later feasible investments');\ncounterfactuals.push('hold world/seat/opponent fixed and intervene only on first material investment timing');counterfactuals.push('hold production plan fixed and alter only market execution order to isolate sales-layer value');counterfactuals.push('hold market layer fixed and alter only enterprise mix/capacity timing to isolate production-layer value');\nreturn{dossier_id:`DOS-${x.matchId}-${x.subjectId}`,match_id:x.matchId,subject_id:x.subjectId,opponent_id:x.opponentId,subject_role:role,seed:x.seed,seat:x.seat,final_bank:x.finalBank,opponent_bank:x.opponentBank,margin:x.finalBank-x.opponentBank,min_cash:min.cash,min_cash_day:min.day,first_lead_day:firstLead,first_trail_day:firstTrail,max_lead:maxLead.margin,max_lead_day:maxLead.day,max_deficit:maxDef.margin,max_deficit_day:maxDef.day,investment_fingerprint:{hire:x.telemetry.hire??0,land:x.telemetry.land??0,buyAnimal:ownAnimals,buySeed:x.telemetry.buySeed??{},buyProduct:x.telemetry.buyProduct??{}},revenue_mix:ownSell,market_exposure:{salesFamilies:ownDiversity,opponentSalesFamilies:oppDiversity},observed_strengths:strengths,observed_failures:failures,causal_hypotheses:hypotheses,counterfactuals,evidence_state:'OBSERVED_FACTS_PLUS_UNVERIFIED_CAUSAL_HYPOTHESES'};}\nexport function mirrorTimeline(t:any[],subjectIsOurs:boolean):DailyPoint[]{return t.map(p=>subjectIsOurs?{day:p.day,subject:p.ours,opponent:p.ref}:{day:p.day,subject:p.ref,opponent:p.ours});}",[],[2112,2198],"81eab083-914b-4c12-a6ee-eb5d9b8f9e0b","kaggriculture_agents_v1.ts","7605724c-aa37-11f1-aaa8-1607ee4eb77e",21,["D",1788729060447],"export type AgentFn = (obs: any) => any;\n\nconst CROPS: Record = {\n WHEAT: { seed: 10, first: 2, maxDay: 4, maxYield: 6, ongoing: false },\n CARROT: { seed: 20, first: 2, maxDay: 3, maxYield: 4, ongoing: false },\n TOMATO: { seed: 50, first: 8, maxDay: 8, maxYield: 4, ongoing: true },\n STRAWBERRY: { seed: 100, first: 10, maxDay: 10, maxYield: 4, ongoing: true },\n MELON: { seed: 80, first: 10, maxDay: 12, maxYield: 6, ongoing: false },\n};\n\nconst PASS = [\"PASS\"];\nconst MEM = new Map }>();\n\nfunction manhattan(a: [number, number], b: [number, number]) {\n return Math.abs(a[0] - b[0]) + Math.abs(a[1] - b[1]);\n}\n\nfunction stepToward(pos: [number, number], target: [number, number]) {\n const dx = target[0] - pos[0];\n const dy = target[1] - pos[1];\n if (Math.abs(dx) >= Math.abs(dy) && dx !== 0) return [dx > 0 ? \"EAST\" : \"WEST\"];\n if (dy !== 0) return [dy > 0 ? \"SOUTH\" : \"NORTH\"];\n return PASS;\n}\n\nfunction ownedCoords(farm: any): Array<[number, number]> {\n const out: Array<[number, number]> = [];\n for (let y = 0; y < farm.tiles.length; y++) {\n for (let x = 0; x < farm.tiles[y].length; x++) {\n if (farm.tiles[y][x] !== \"LOCKED\") out.push([x, y]);\n }\n }\n return out;\n}\n\nfunction tileTask(tile: any, day: number, hour: number, crop: string): [number, string] | null {\n if (tile === null) return hour < 23 ? [60, \"PLANT\"] : null;\n if (tile === \"LOCKED\" || typeof tile !== \"object\") return null;\n if (tile.kind === \"WEED\") return [50, \"DIG\"];\n if (tile.kind !== \"PLANT\" || tile.crop !== crop) return null;\n\n const age = day - Number(tile.planted_day ?? day);\n const cfg = CROPS[crop];\n const watered = Boolean(tile.watered_today);\n const streak = Number(tile.consecutive_unwatered ?? 0);\n const yieldUnits = Number(tile.yield_units ?? 0);\n\n if (!watered && (streak >= 1 || hour >= 18)) return [5, \"WATER\"];\n if (yieldUnits > 0 && age >= cfg.first) {\n if (!cfg.ongoing && age <= cfg.maxDay && !watered && hour < 22) return [8, \"WATER\"];\n if (age >= cfg.maxDay || hour >= 22) return [10, \"HARVEST\"];\n }\n if (!watered) return [20, \"WATER\"];\n if (yieldUnits > 0 && age >= cfg.first) return [25, \"HARVEST\"];\n return null;\n}\n\nfunction collectTasks(farm: any, day: number, hour: number, crop: string, maxTiles: number | null) {\n let coords = ownedCoords(farm);\n if (maxTiles !== null) {\n const h = Math.floor(farm.tiles.length / 2);\n const shedCorner: [number, number] = [h - 1, h - 1];\n coords = coords.sort((a, b) => manhattan(a, shedCorner) - manhattan(b, shedCorner) || a[1] - b[1] || a[0] - b[0]).slice(0, maxTiles);\n }\n const tasks: Array<{ pri: number; kind: string; target: [number, number] }> = [];\n for (const [x, y] of coords) {\n const t = tileTask(farm.tiles[y][x], day, hour, crop);\n if (t) tasks.push({ pri: t[0], kind: t[1], target: [x, y] });\n }\n tasks.sort((a, b) => a.pri - b.pri || a.target[1] - b.target[1] || a.target[0] - b.target[0]);\n return tasks;\n}\n\nfunction assignmentValid(farm: any, item: { kind: string; target: [number, number] }, day: number, hour: number, crop: string) {\n const [x, y] = item.target;\n if (y < 0 || y >= farm.tiles.length || x < 0 || x >= farm.tiles[y].length) return false;\n const t = tileTask(farm.tiles[y][x], day, hour, crop);\n return Boolean(t && t[1] === item.kind);\n}\n\nfunction chooseAssignments(player: number, farm: any, day: number, hour: number, crop: string, positions: Array<[number, number]>, tasks: any[], persistent: boolean) {\n const key = `${player}:${crop}`;\n if (day === 0 && hour === 0) MEM.delete(key);\n let state = MEM.get(key) ?? { day, assign: new Map() };\n if (state.day !== day) state = { day, assign: new Map() };\n MEM.set(key, state);\n\n const assigns = persistent ? state.assign : new Map();\n for (const [uid, item] of [...assigns.entries()]) {\n if (uid >= positions.length || !assignmentValid(farm, item, day, hour, crop)) assigns.delete(uid);\n }\n const claimed = new Set([...assigns.values()].map((v: any) => `${v.target[0]},${v.target[1]}`));\n const available = tasks.filter((t: any) => !claimed.has(`${t.target[0]},${t.target[1]}`));\n\n for (let uid = 0; uid < positions.length; uid++) {\n if (assigns.has(uid) || available.length === 0) continue;\n let best = 0;\n for (let i = 1; i < available.length; i++) {\n const a = available[i], b = available[best];\n const ka = [a.pri, manhattan(positions[uid], a.target), a.target[1], a.target[0]];\n const kb = [b.pri, manhattan(positions[uid], b.target), b.target[1], b.target[0]];\n if (ka[0] < kb[0] || (ka[0] === kb[0] && (ka[1] < kb[1] || (ka[1] === kb[1] && (ka[2] < kb[2] || (ka[2] === kb[2] && ka[3] < kb[3])))))) best = i;\n }\n const chosen = available.splice(best, 1)[0];\n assigns.set(uid, { kind: chosen.kind, target: chosen.target });\n }\n if (persistent) state.assign = assigns;\n return new Map(assigns);\n}\n\nfunction marketOrders(obs: any, farm: any, priv: any, crop: string, seedBuffer: number, targetHands: number) {\n const out: any[] = [];\n const hour = Number(obs.hour ?? 0);\n const currentHands = (farm.hands ?? []).length;\n const hires = Math.max(0, targetHands - currentHands);\n if (hour === 0 && hires > 0) {\n for (let i = 0; i < Math.min(hires, 10); i++) out.push([\"HIRE\"]);\n if (out.length >= 10) return out;\n }\n const qty = Number(priv.shed?.[crop] ?? 0);\n if (qty > 0 && out.length < 10) out.push([\"SELL\", crop, qty]);\n const seeds = Number(priv.seeds?.[crop] ?? 0);\n const missing = Math.max(0, seedBuffer - seeds);\n const affordable = Math.floor(Number(farm.money ?? 0) / CROPS[crop].seed);\n if (missing > 0 && affordable > 0 && out.length < 10) out.push([\"BUY_SEED\", crop, Math.min(missing, affordable)]);\n return out.slice(0, 10);\n}\n\nexport function buildAgent(opts: { crop: string; targetHands: number; maxTiles: number | null; persistent: boolean; seedBuffer: number }): AgentFn {\n return (obs: any) => {\n try {\n const player = Number(obs.player ?? 0);\n const farms = obs.farms ?? [];\n if (!farms[player]) return { farmer: PASS, hands: [], market: [] };\n const farm = farms[player];\n const priv = obs.private ?? {};\n const day = Number(obs.day ?? 0), hour = Number(obs.hour ?? 0);\n const positions: Array<[number, number]> = [farm.farmer, ...(farm.hands ?? [])].map((x: any) => [Number(x[0]), Number(x[1])]);\n const tasks = collectTasks(farm, day, hour, opts.crop, opts.maxTiles);\n const assignments = chooseAssignments(player, farm, day, hour, opts.crop, positions, tasks, opts.persistent);\n let seedBudget = Number(priv.seeds?.[opts.crop] ?? 0);\n const actions: any[] = [];\n for (let uid = 0; uid < positions.length; uid++) {\n const item = assignments.get(uid);\n if (!item) { actions.push(PASS); continue; }\n const target = item.target as [number, number];\n if (positions[uid][0] !== target[0] || positions[uid][1] !== target[1]) { actions.push(stepToward(positions[uid], target)); continue; }\n if (item.kind === \"PLANT\") {\n if (hour < 23 && seedBudget > 0) { actions.push([\"PLANT\", opts.crop]); seedBudget--; }\n else actions.push(PASS);\n } else if ([\"WATER\", \"HARVEST\", \"DIG\"].includes(item.kind)) actions.push([item.kind]);\n else actions.push(PASS);\n }\n return {\n farmer: actions[0] ?? PASS,\n hands: actions.slice(1, 1 + (farm.hands ?? []).length),\n market: marketOrders(obs, farm, priv, opts.crop, opts.seedBuffer, opts.targetHands),\n };\n } catch {\n return { farmer: PASS, hands: [], market: [] };\n }\n };\n}\n\nexport const K0 = buildAgent({ crop: \"CARROT\", targetHands: 0, maxTiles: 4, persistent: false, seedBuffer: 6 });\nexport const K1 = buildAgent({ crop: \"CARROT\", targetHands: 10, maxTiles: 25, persistent: true, seedBuffer: 30 });\n\nexport const STARTER: AgentFn = (obs: any) => {\n const farm = obs.farms?.[obs.player];\n if (!farm) return { farmer: PASS, hands: [], market: [] };\n const [x, y] = farm.farmer;\n const tile = farm.tiles[y][x];\n const seeds = obs.private?.seeds ?? {};\n const shed = obs.private?.shed ?? {};\n const market: any[] = [];\n const q = Number(shed.CARROT ?? 0);\n if (q > 0) market.push([\"SELL\", \"CARROT\", q]);\n if (Number(seeds.CARROT ?? 0) === 0 && Number(farm.money) >= 20) market.push([\"BUY_SEED\", \"CARROT\", 1]);\n let farmer: any = PASS;\n if (tile === null && Number(seeds.CARROT ?? 0) > 0) farmer = [\"PLANT\", \"CARROT\"];\n else if (tile && typeof tile === \"object\" && tile.kind === \"PLANT\" && tile.crop === \"CARROT\") {\n const age = Number(obs.day) - Number(tile.planted_day);\n if (age >= 3) farmer = [\"HARVEST\"];\n else if (!tile.watered_today) farmer = [\"WATER\"];\n }\n return { farmer, hands: [], market };\n};\n",[],[2112,2206],"5a970014-78b8-486c-ad63-a0f09bc95d6f","kaggriculture_adaptive_racing.ts","6449343c-aa3f-11f1-a010-1607ee4eb77e",51,["D",1788732466721],"export type BernoulliRace = {\n wins:number;\n losses:number;\n p0?:number;\n p1?:number;\n alpha?:number;\n beta?:number;\n};\n\n/** Sequential Probability Ratio Test for H0:p=p0 vs H1:p=p1. Ties should be handled outside as paired margin diagnostics. */\nexport function sprt(x:BernoulliRace) {\n const p0=x.p0??0.50, p1=x.p1??0.60, alpha=x.alpha??0.05, beta=x.beta??0.10;\n if (!(0=upper ? \"PROMOTE_H1\" as const : llr<=lower ? \"REJECT_H1\" as const : \"CONTINUE\" as const,\n n:x.wins+x.losses,\n observed:x.wins/Math.max(1,x.wins+x.losses),\n };\n}\n\n/** Wilson score interval; descriptive CI for a fixed checkpoint, not claimed as an always-valid sequential confidence sequence. */\nexport function wilson(wins:number, losses:number, z=1.96) {\n const n=wins+losses;\n if (!n) return {low:0,high:1,center:0.5,n:0};\n const p=wins/n, z2=z*z;\n const den=1+z2/n;\n const center=(p+z2/(2*n))/den;\n const half=z*Math.sqrt((p*(1-p)+z2/(4*n))/n)/den;\n return {low:Math.max(0,center-half),high:Math.min(1,center+half),center,n};\n}\n\nexport type RaceCandidate = {\n id:string;\n wins:number;\n losses:number;\n avgMargin:number;\n runtimeErrorRate?:number;\n};\n\n/** Allocate future matches where information is most valuable: uncertain, close to decision boundary, and operationally stable. */\nexport function racePriority(c:RaceCandidate, target=0.55) {\n const ci=wilson(c.wins,c.losses);\n const width=ci.high-ci.low;\n const closeness=1-Math.min(1,Math.abs(ci.center-target)/0.5);\n const stability=1-Math.min(1,c.runtimeErrorRate??0);\n return width*(0.35+0.65*closeness)*stability;\n}\n\nexport function allocateBatch(candidates:RaceCandidate[], totalMatches:number) {\n const scores=candidates.map(c=>({id:c.id,score:racePriority(c)}));\n const z=scores.reduce((s,x)=>s+x.score,0);\n if (z<=0) return scores.map((x,i)=>({id:x.id,matches:i({id:x.id,q:totalMatches*x.score/z}));\n const base=raw.map(x=>({id:x.id,matches:Math.floor(x.q),frac:x.q-Math.floor(x.q)}));\n let left=totalMatches-base.reduce((s,x)=>s+x.matches,0);\n base.sort((a,b)=>b.frac-a.frac);\n for(let i=0;i0;i++,left--) base[i].matches++;\n return base.map(({id,matches})=>({id,matches}));\n}\n\nexport function dominanceDecision(a:RaceCandidate,b:RaceCandidate) {\n const A=wilson(a.wins,a.losses), B=wilson(b.wins,b.losses);\n if ((a.runtimeErrorRate??0)>0.02) return {decision:\"A_RUNTIME_BLOCKED\" as const,A,B};\n if ((b.runtimeErrorRate??0)>0.02) return {decision:\"B_RUNTIME_BLOCKED\" as const,A,B};\n if (A.low>B.high && a.avgMargin>=b.avgMargin) return {decision:\"A_DOMINATES\" as const,A,B};\n if (B.low>A.high && b.avgMargin>=a.avgMargin) return {decision:\"B_DOMINATES\" as const,A,B};\n return {decision:\"NEEDS_MORE_EVIDENCE\" as const,A,B};\n}\n",[],[2112,2214],"240f7455-1078-47c1-b200-2980a521996b","kag_teacher_stats_v2.ts","a58991a4-aa8e-11f1-892f-1607ee4eb77e",476,["D",1788766515526],"export type ActionStatV2={action:string,n:number,marginMean:number,marginSd:number,winRate:number,tieRate:number,lossRate:number,scoreMean:number,deltaMarginMean:number,deltaMarginLcb95:number,deltaMarginUcb95:number,deltaMarginPositiveRate:number,deltaMarginSignP:number,deltaScoreMean:number,deltaScorePositiveRate:number,deltaScoreSignP:number};\nexport const mean=(x:number[])=>x.reduce((a,b)=>a+b,0)/(x.length||1);\nexport function sampleSd(x:number[]){if(x.length<2)return 0;const m=mean(x);return Math.sqrt(x.reduce((s,v)=>s+(v-m)**2,0)/(x.length-1));}\nconst T95:Record={4:3.182,8:2.365,12:2.201,16:2.131,24:2.069,32:2.040,40:2.023,48:2.012,64:2.000};\nexport function t95(n:number){if(T95[n])return T95[n];return n>=120?1.98:n>=64?2.00:n>=48?2.012:n>=40?2.023:n>=32?2.040:n>=24?2.069:n>=16?2.131:n>=12?2.201:n>=8?2.365:3.182;}\nexport function pairedT95(delta:number[]){const n=delta.length,m=mean(delta),se=sampleSd(delta)/Math.sqrt(n||1),t=t95(n);return{n,mean:m,se,lcb:m-t*se,ucb:m+t*se,sd:sampleSd(delta)};}\nfunction comb(n:number,k:number){k=Math.min(k,n-k);let r=1;for(let i=1;i<=k;i++)r=r*(n-k+i)/i;return r;}\nexport function signTestOneSidedPositive(delta:number[]){const nz=delta.filter(x=>x!==0),n=nz.length,pos=nz.filter(x=>x>0).length;if(n===0||pos<=n/2)return{nonzero:n,positive:pos,p:1};let p=0;for(let k=pos;k<=n;k++)p+=comb(n,k)*Math.pow(.5,n);return{nonzero:n,positive:pos,p:Math.min(1,p)};}\nexport const score=(margin:number)=>margin>0?1:(margin===0?0.5:0);\nexport function summarizeAction(action:string,margins:number[],baseline:number[]):ActionStatV2{const n=margins.length,sc=margins.map(score),base=baseline.slice(0,n),baseSc=base.map(score),dm=margins.map((v,i)=>v-base[i]),ds=sc.map((v,i)=>v-baseSc[i]),mt=pairedT95(dm),ms=signTestOneSidedPositive(dm),ss=signTestOneSidedPositive(ds);return{action,n,marginMean:mean(margins),marginSd:sampleSd(margins),winRate:sc.filter(x=>x===1).length/n,tieRate:sc.filter(x=>x===.5).length/n,lossRate:sc.filter(x=>x===0).length/n,scoreMean:mean(sc),deltaMarginMean:mt.mean,deltaMarginLcb95:action==='N'?0:mt.lcb,deltaMarginUcb95:action==='N'?0:mt.ucb,deltaMarginPositiveRate:action==='N'?0:dm.filter(x=>x>0).length/n,deltaMarginSignP:action==='N'?1:ms.p,deltaScoreMean:action==='N'?0:mean(ds),deltaScorePositiveRate:action==='N'?0:ds.filter(x=>x>0).length/n,deltaScoreSignP:action==='N'?1:ss.p};}\nexport function chooseDualObjective(rows:ActionStatV2[]){const c=rows.filter(r=>r.action!=='N'&&r.deltaScoreMean>0&&r.deltaScoreSignP<.05).sort((a,b)=>b.deltaScoreMean-a.deltaScoreMean||b.deltaMarginLcb95-a.deltaMarginLcb95);if(c.length)return{pick:c[0].action,pickType:'COMPETITIVE',objectiveState:'COMPETITIVE_DUAL_OBJECTIVE',reason:'paired outcome-score improvement survives one-sided exact sign gate'};const s=rows.filter(r=>r.action!=='N'&&r.deltaMarginLcb95>0&&r.deltaMarginSignP<.05).sort((a,b)=>b.deltaMarginLcb95-a.deltaMarginLcb95);if(s.length)return{pick:s[0].action,pickType:'SHAPING_ONLY',objectiveState:'SHAPING_ONLY_MARGIN',reason:'no competitive outcome improvement; paired margin improvement only'};return{pick:'N',pickType:'NONE',objectiveState:'NO_ACTION_EVIDENCE',reason:'no action survives competitive or shaping gate'};}\n",[],[2112,2222],"e670ec9a-b7b2-40a2-9382-208ec1a6a9f4","kag_teacher_features_v31.ts","01a07a96-6a4d-7069-aaab-cd029a49449b",411,["D",1788763141556],"export const KAG_TEACHER_FEATURE_SCHEMA_V31='KAG_TEACHER_FEATURES_V3_1_STATEFUL';\nconst CROPS=['WHEAT','CARROT','TOMATO','STRAWBERRY','MELON'];const ANIMALS=['GOOSE','COW','SHEEP'];const PRODUCTS=['WHEAT','CARROT','TOMATO','STRAWBERRY','MELON','EGG','MILK','WOOL','FERTILIZER'];\nconst n=(x:any)=>Number.isFinite(+x)?+x:0;function blankCrop(){return{count:0,yield:0,watered:0,unwateredSum:0,unwateredMax:0,fertilized:0,ageSum:0,finiteLifespanCount:0,minLifeRemaining:1e9}}function blankAnimal(){return{count:0,yield:0,fed:0,cared:0,unfedSum:0,unfedMax:0,pendingCareBonus:0,fertilizerAvailable:0,ageSum:0}}\nfunction farmRich(f:any,step:number){const crops:any=Object.fromEntries(CROPS.map(k=>[k,blankCrop()])),animals:any=Object.fromEntries(ANIMALS.map(k=>[k,blankAnimal()])),structures:any={PASTURE:0,COOP:0,emptyPasture:0,emptyCoop:0};let cropTotal=0,animalTotal=0;for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT'&&t.crop){const c=crops[t.crop]??(crops[t.crop]=blankCrop());c.count++;cropTotal++;c.yield+=n(t.yield_units);c.watered+=t.watered_today?1:0;c.unwateredSum+=n(t.consecutive_unwatered);c.unwateredMax=Math.max(c.unwateredMax,n(t.consecutive_unwatered));c.fertilized+=n(t.fertilized_until_day)>=Math.floor(step/24)?1:0;c.ageSum+=Math.max(0,Math.floor(step/24)-n(t.planted_day));if(Number.isFinite(+t.max_lifespan_step)&&+t.max_lifespan_step>0){c.finiteLifespanCount++;c.minLifeRemaining=Math.min(c.minLifeRemaining,+t.max_lifespan_step-step)}}if(t.kind==='PASTURE'){structures.PASTURE++;if(!t.animal)structures.emptyPasture++}if(t.kind==='COOP'){structures.COOP++;if(!t.animal)structures.emptyCoop++}if(t.animal){const a=animals[t.animal]??(animals[t.animal]=blankAnimal());a.count++;animalTotal++;a.yield+=n(t.yield_units);a.fed+=t.fed_today?1:0;a.cared+=t.cared_today?1:0;a.unfedSum+=n(t.consecutive_unfed);a.unfedMax=Math.max(a.unfedMax,n(t.consecutive_unfed));a.pendingCareBonus+=n(t.pending_care_bonus);a.fertilizerAvailable+=t.fertilizer_available?1:0;a.ageSum+=Math.max(0,Math.floor(step/24)-n(t.placed_day))}}for(const c of Object.values(crops) as any[]){c.meanAge=c.count?c.ageSum/c.count:0;if(c.minLifeRemaining===1e9)c.minLifeRemaining=-1;delete c.ageSum}for(const a of Object.values(animals) as any[]){a.meanAge=a.count?a.ageSum/a.count:0;delete a.ageSum}return{cash:n(f?.money),land:(f?.unlocked_quadrants??[]).length,hands:(f?.hands??[]).length,hiresToday:n(f?.hires_today),farmer:[...(f?.farmer??[])],cropTotal,crops,animalTotal,animals,structures,pastureHeadroom:structures.PASTURE-animalTotal}}\nfunction bagTotals(inv:any[]){const z:any=Object.fromEntries(PRODUCTS.map(k=>[k,0]));for(const bag of inv??[])for(const [k,v] of Object.entries(bag??{}))z[k]=(z[k]??0)+n(v);return z}function workerSpatial(f:any){const pts:any[]=[...(f?.hands??[])];const farmer=[...(f?.farmer??[])];let dist=0,row=0,col=0;for(const p of pts){if(!Array.isArray(p))continue;row+=n(p[0]);col+=n(p[1]);dist+=Math.abs(n(p[0])-n(farmer[0]))+Math.abs(n(p[1])-n(farmer[1]))}return{count:pts.length,farmer,meanRow:pts.length?row/pts.length:0,meanCol:pts.length?col/pts.length:0,meanDistToFarmer:pts.length?dist/pts.length:0,uniquePositions:new Set(pts.map(p=>JSON.stringify(p))).size}}\nfunction dailyExecution(f:any){let watered=0,fed=0,cared=0,plants=0,animals=0;for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT'){plants++;watered+=t.watered_today?1:0}if(t.animal){animals++;fed+=t.fed_today?1:0;cared+=t.cared_today?1:0}}return{plants,watered,animals,fed,cared}}\nexport function extractTeacherFeaturesV31(o:any,prev:any=null){const me=+o.player,step=n(o.step),a=farmRich(o.farms?.[me],step),b=farmRich(o.farms?.[1-me],step),prices:any={},inventory:any={},marketDelta:any={};for(const p of PRODUCTS){prices[p]=n(o.market?.prices?.[p]);inventory[p]=n(o.market?.inventory?.[p]);marketDelta[p]=inventory[p]-10000}const shed:any={},seeds:any={};for(const p of PRODUCTS)shed[p]=n(o.private?.shed?.[p]);for(const c of CROPS)seeds[c]=n(o.private?.seeds?.[c]);const carried=bagTotals(o.private?.inventories??[]),shopCounts:any={};for(const s of o.town?.unlocked_shops??[])shopCounts[s]=(shopCounts[s]??0)+1;let temporal:any=null;if(prev){const pm=+prev.player,pa=prev.farms?.[pm],pb=prev.farms?.[1-pm];temporal={prevStep:n(prev.step),prevDay:n(prev.day),prevHour:n(prev.hour),me:{hands:(pa?.hands??[]).length,hiresToday:n(pa?.hires_today),workerSpatial:workerSpatial(pa),dailyExecution:dailyExecution(pa),carried:bagTotals(prev.private?.inventories??[])},opp:{hands:(pb?.hands??[]).length,hiresToday:n(pb?.hires_today),workerSpatial:workerSpatial(pb),dailyExecution:dailyExecution(pb)}}}return{schema:KAG_TEACHER_FEATURE_SCHEMA_V31,day:n(o.day),hour:n(o.hour),step,shops:[...(o.town?.unlocked_shops??[])],shopCounts,me:a,opp:b,gaps:{cash:a.cash-b.cash,land:a.land-b.land,hands:a.hands-b.hands,crops:a.cropTotal-b.cropTotal,animals:a.animalTotal-b.animalTotal,pastureHeadroom:a.pastureHeadroom-b.pastureHeadroom},workerSpatial:workerSpatial(o.farms?.[me]),market:{prices,inventory,deltaFromInitial:marketDelta},private:{shed,seeds,carried,carriedTotal:Object.values(carried).reduce((s:any,v:any)=>s+n(v),0),inventoryBags:(o.private?.inventories??[]).length},temporal}}\n",[],[2112,2230],"ee1de2c1-10c5-4fdc-bc0d-65f779efdd8e","kag_teacher_features_v3.ts","c9eca6c0-aa85-11f1-a35c-1607ee4eb77e",392,["D",1788762718911],"export const KAG_TEACHER_FEATURE_SCHEMA_V3='KAG_TEACHER_FEATURES_V3_STATEFUL';\nconst CROPS=['WHEAT','CARROT','TOMATO','STRAWBERRY','MELON'];const ANIMALS=['GOOSE','COW','SHEEP'];const PRODUCTS=['WHEAT','CARROT','TOMATO','STRAWBERRY','MELON','EGG','MILK','WOOL','FERTILIZER'];\nconst n=(x:any)=>Number.isFinite(+x)?+x:0;function blankCrop(){return{count:0,yield:0,watered:0,unwateredSum:0,unwateredMax:0,fertilized:0,ageSum:0,minLifeRemaining:1e9}}function blankAnimal(){return{count:0,yield:0,fed:0,cared:0,unfedSum:0,unfedMax:0,pendingCareBonus:0,fertilizerAvailable:0,ageSum:0}}\nfunction farmRich(f:any,step:number){const crops:any=Object.fromEntries(CROPS.map(k=>[k,blankCrop()])),animals:any=Object.fromEntries(ANIMALS.map(k=>[k,blankAnimal()])),structures:any={PASTURE:0,COOP:0,emptyPasture:0,emptyCoop:0};let cropTotal=0,animalTotal=0;for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT'&&t.crop){const c=crops[t.crop]??(crops[t.crop]=blankCrop());c.count++;cropTotal++;c.yield+=n(t.yield_units);c.watered+=t.watered_today?1:0;c.unwateredSum+=n(t.consecutive_unwatered);c.unwateredMax=Math.max(c.unwateredMax,n(t.consecutive_unwatered));c.fertilized+=n(t.fertilized_until_day)>=Math.floor(step/24)?1:0;c.ageSum+=Math.max(0,Math.floor(step/24)-n(t.planted_day));if(Number.isFinite(+t.max_lifespan_step))c.minLifeRemaining=Math.min(c.minLifeRemaining,+t.max_lifespan_step-step)}if(t.kind==='PASTURE'){structures.PASTURE++;if(!t.animal)structures.emptyPasture++}if(t.kind==='COOP'){structures.COOP++;if(!t.animal)structures.emptyCoop++}if(t.animal){const a=animals[t.animal]??(animals[t.animal]=blankAnimal());a.count++;animalTotal++;a.yield+=n(t.yield_units);a.fed+=t.fed_today?1:0;a.cared+=t.cared_today?1:0;a.unfedSum+=n(t.consecutive_unfed);a.unfedMax=Math.max(a.unfedMax,n(t.consecutive_unfed));a.pendingCareBonus+=n(t.pending_care_bonus);a.fertilizerAvailable+=t.fertilizer_available?1:0;a.ageSum+=Math.max(0,Math.floor(step/24)-n(t.placed_day))}}for(const c of Object.values(crops) as any[]){c.meanAge=c.count?c.ageSum/c.count:0;if(c.minLifeRemaining===1e9)c.minLifeRemaining=0;delete c.ageSum}for(const a of Object.values(animals) as any[]){a.meanAge=a.count?a.ageSum/a.count:0;delete a.ageSum}return{cash:n(f?.money),land:(f?.unlocked_quadrants??[]).length,hands:(f?.hands??[]).length,hiresToday:n(f?.hires_today),farmer:[...(f?.farmer??[])],cropTotal,crops,animalTotal,animals,structures,pastureHeadroom:structures.PASTURE-animalTotal}}\nfunction bagTotals(inv:any[]){const z:any=Object.fromEntries(PRODUCTS.map(k=>[k,0]));for(const bag of inv??[])for(const [k,v] of Object.entries(bag??{}))z[k]=(z[k]??0)+n(v);return z}function workerSpatial(f:any){const pts:any[]=[...(f?.hands??[])];const farmer=[...(f?.farmer??[])];let dist=0,row=0,col=0;for(const p of pts){if(!Array.isArray(p))continue;row+=n(p[0]);col+=n(p[1]);dist+=Math.abs(n(p[0])-n(farmer[0]))+Math.abs(n(p[1])-n(farmer[1]))}return{count:pts.length,farmer,meanRow:pts.length?row/pts.length:0,meanCol:pts.length?col/pts.length:0,meanDistToFarmer:pts.length?dist/pts.length:0,uniquePositions:new Set(pts.map(p=>JSON.stringify(p))).size}}\nfunction dailyExecution(f:any){let watered=0,fed=0,cared=0,plants=0,animals=0;for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT'){plants++;watered+=t.watered_today?1:0}if(t.animal){animals++;fed+=t.fed_today?1:0;cared+=t.cared_today?1:0}}return{plants,watered,animals,fed,cared}}\nexport function extractTeacherFeaturesV3(o:any,prev:any=null){const me=+o.player,step=n(o.step),a=farmRich(o.farms?.[me],step),b=farmRich(o.farms?.[1-me],step),prices:any={},inventory:any={},marketDelta:any={};for(const p of PRODUCTS){prices[p]=n(o.market?.prices?.[p]);inventory[p]=n(o.market?.inventory?.[p]);marketDelta[p]=inventory[p]-10000}const shed:any={},seeds:any={};for(const p of PRODUCTS)shed[p]=n(o.private?.shed?.[p]);for(const c of CROPS)seeds[c]=n(o.private?.seeds?.[c]);const carried=bagTotals(o.private?.inventories??[]),shopCounts:any={};for(const s of o.town?.unlocked_shops??[])shopCounts[s]=(shopCounts[s]??0)+1;let temporal:any=null;if(prev){const pm=+prev.player,pa=prev.farms?.[pm],pb=prev.farms?.[1-pm];temporal={prevStep:n(prev.step),prevDay:n(prev.day),prevHour:n(prev.hour),me:{hands:(pa?.hands??[]).length,hiresToday:n(pa?.hires_today),workerSpatial:workerSpatial(pa),dailyExecution:dailyExecution(pa),carried:bagTotals(prev.private?.inventories??[])},opp:{hands:(pb?.hands??[]).length,hiresToday:n(pb?.hires_today),workerSpatial:workerSpatial(pb),dailyExecution:dailyExecution(pb)}}}return{schema:KAG_TEACHER_FEATURE_SCHEMA_V3,day:n(o.day),hour:n(o.hour),step,shops:[...(o.town?.unlocked_shops??[])],shopCounts,me:a,opp:b,gaps:{cash:a.cash-b.cash,land:a.land-b.land,hands:a.hands-b.hands,crops:a.cropTotal-b.cropTotal,animals:a.animalTotal-b.animalTotal,pastureHeadroom:a.pastureHeadroom-b.pastureHeadroom},workerSpatial:workerSpatial(o.farms?.[me]),market:{prices,inventory,deltaFromInitial:marketDelta},private:{shed,seeds,carried,carriedTotal:Object.values(carried).reduce((s:any,v:any)=>s+n(v),0),inventoryBags:(o.private?.inventories??[]).length},temporal}}\n",[],[2112,2238],"0285a1ff-d2c1-433b-9d2c-bf77912088cc","kag_teacher_features_v2.ts","c8610acc-aa84-11f1-bb92-1607ee4eb77e",373,["D",1788762269930],"export const KAG_TEACHER_FEATURE_SCHEMA='KAG_TEACHER_FEATURES_V2';\nconst CROPS=['WHEAT','CARROT','TOMATO','STRAWBERRY','MELON'];\nconst ANIMALS=['GOOSE','COW','SHEEP'];\nconst PRODUCTS=['WHEAT','CARROT','TOMATO','STRAWBERRY','MELON','EGG','MILK','WOOL','FERTILIZER'];\nfunction farmFeatures(f:any){const crops:any=Object.fromEntries(CROPS.map(k=>[k,0])),animals:any=Object.fromEntries(ANIMALS.map(k=>[k,0])),structures:any={PASTURE:0,COOP:0};let cropTotal=0,animalTotal=0;for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PLANT'&&t.crop){crops[t.crop]=(crops[t.crop]??0)+1;cropTotal++}if(t.animal){animals[t.animal]=(animals[t.animal]??0)+1;animalTotal++}if(t.kind==='PASTURE')structures.PASTURE++;if(t.kind==='COOP')structures.COOP++}return{cash:+(f?.money??0),land:(f?.unlocked_quadrants??[]).length,hands:(f?.hands??[]).length,hiresToday:+(f?.hires_today??f?.hiresToday??0),cropTotal,crops,animalTotal,animals,structures,pastureHeadroom:structures.PASTURE-animalTotal}}\nexport function extractTeacherFeaturesV2(o:any){const me=+o.player,a=farmFeatures(o.farms[me]),b=farmFeatures(o.farms[1-me]),prices:any={},inventory:any={},marketDelta:any={};for(const p of PRODUCTS){prices[p]=+(o.market?.prices?.[p]??0);inventory[p]=+(o.market?.inventory?.[p]??0);marketDelta[p]=inventory[p]-10000}const shed:any={};for(const p of PRODUCTS)shed[p]=+(o.private?.shed?.[p]??0);const shopCounts:any={};for(const s of o.town?.unlocked_shops??[])shopCounts[s]=(shopCounts[s]??0)+1;return{schema:KAG_TEACHER_FEATURE_SCHEMA,day:+o.day,hour:+o.hour,shops:[...(o.town?.unlocked_shops??[])],shopCounts,me:a,opp:b,gaps:{cash:a.cash-b.cash,land:a.land-b.land,hands:a.hands-b.hands,crops:a.cropTotal-b.cropTotal,animals:a.animalTotal-b.animalTotal,pastureHeadroom:a.pastureHeadroom-b.pastureHeadroom},market:{prices,inventory,deltaFromInitial:marketDelta},shed}}\n",[],[2112,2246],"aaf519f1-f7ca-470e-8787-470a0c222e7d","kag_autolab_core.ts","8428507e-aa96-11f1-a3d5-1607ee4eb77e",509,["D",1788770374448],"import { unzipSync, strFromU8 } from 'npm:fflate';\nimport { loadPyodide } from 'npm:pyodide';\nimport { loadOfficialKaggricultureEngine, observationFor } from './kaggriculture_engine_loader.ts';\nimport { choosePastureAnimal } from './kaggriculture_livestock_portfolio.ts';\n\nexport const AUTOLAB_CORE_VERSION = 'KAG_AUTOLAB_CORE_V1';\nconst REF_DATASET='https://www.kaggle.com/api/v1/datasets/download/raykkretzschmar/kaggriculture-reference-agents';\nconst CHECK_DAYS=new Set([0,3,6,9,12,15,18,21,24,27,29]);\nlet runtimePromise:Promise|null=null;\n\nfunction avg(xs:number[]){return xs.length?xs.reduce((a,b)=>a+b,0)/xs.length:0}\nfunction profilePublic(f:any){let crops=0,pastures=0,animals=0;const cropMix:any={},animalMix:any={};for(const row of f?.tiles??[])for(const t of row??[]){if(!t||typeof t!=='object')continue;if(t.kind==='PASTURE')pastures++;if(t.animal){animals++;animalMix[t.animal]=(animalMix[t.animal]??0)+1}const c=t.crop??t.plant??(t.kind==='PLANT'?t.type:null);if(c){crops++;cropMix[c]=(cropMix[c]??0)+1}}return{cash:+(f?.money??0),land:(f?.unlocked_quadrants??[]).length,crops,pastures,animals,cropMix,animalMix}}\nfunction gapDiagnostics(games:any[],perspective:'A'|'OURS'){\n const days=[0,3,6,9,12,15,18,21,24,27,29];\n const trajectory=days.map(day=>{const rows=games.map(g=>(g.timeline??[]).find((x:any)=>x.day===day)).filter(Boolean);const gaps=rows.map((x:any)=>perspective==='A'?x.a.cash-x.b.cash:x.ours.cash-x.ref.cash);return{day,gapMean:avg(gaps),gapMin:gaps.length?Math.min(...gaps):0,gapMax:gaps.length?Math.max(...gaps):0}});\n const windows=trajectory.slice(1).map((x,i)=>({fromDay:trajectory[i].day,toDay:x.day,gapChange:x.gapMean-trajectory[i].gapMean}));\n const largestGapGrowth=[...windows].sort((a,b)=>Math.abs(b.gapChange)-Math.abs(a.gapChange))[0]??null;\n return{trajectory,windows,largestGapGrowth};\n}\nasync function ensureRuntime(){\n if(runtimePromise)return runtimePromise;\n runtimePromise=(async()=>{\n const E=await loadOfficialKaggricultureEngine();\n const rr=await fetch(REF_DATASET);if(!rr.ok)throw new Error(`reference dataset HTTP ${rr.status}`);\n const z=unzipSync(new Uint8Array(await rr.arrayBuffer()));\n const py=await loadPyodide();\n py.runPython(`import json,copy\\n_REG={}\\n_SRC={}\\n_D=None\\n_BASE=None`);\n const loaded=new Set();\n async function loadRef(name:string){if(loaded.has(name))return;if(!z[name])throw new Error(`missing reference agent ${name}`);py.globals.set('nm',name);py.globals.set('src',strFromU8(z[name]));py.runPython(`_SRC[nm]=src\\n_n={}\\nexec(src,_n)\\n_REG[nm]=_n['agent']`);loaded.add(name)}\n await loadRef('rancher_rita.py');\n py.globals.set('src',strFromU8(z['rancher_rita.py']));\n py.runPython(`_d={}\\nexec(src,_d)\\n_D=_d\\n_BASE=copy.deepcopy(_d['POLICY'])\\ndef reset_ref(n):\\n _n={}\\n exec(_SRC[n],_n)\\n _REG[n]=_n['agent']\\ndef call_ref(n,p): return json.dumps(_REG[n](json.loads(p)))\\ndef call_dyn(p): return json.dumps(_D['agent'](json.loads(p)))\\ndef reset_dyn(variant):\\n _D['POLICY'].clear();_D['POLICY'].update(copy.deepcopy(_BASE));_D['POLICY']['animal_target']={'COW':0,'SHEEP':0}\\n if variant in ('MIX11_FERT','M60_FERT'):_D['POLICY']['collect_fertilizer']=True\\n if variant=='M60_FERT':\\n _D['POLICY']['crops']=['WHEAT','MELON'];_D['POLICY']['crop_share']={'WHEAT':0.4,'MELON':0.6};_D['POLICY']['plant_until']={'WHEAT':25,'MELON':16};_D['POLICY']['sell_order']=['MELON','MILK','WOOL','WHEAT'];_D['POLICY']['price_floor']={'MELON':80};_D['POLICY']['sell_chunk']=12\\ndef set_herd(c,s,o):\\n _D['POLICY']['animal_target']={'COW':int(c),'SHEEP':int(s)};_D['POLICY']['animals']=json.loads(o)`);\n return{E,py,z,loadRef,resetRef:py.globals.get('reset_ref'),callRef:py.globals.get('call_ref'),callDyn:py.globals.get('call_dyn'),resetDyn:py.globals.get('reset_dyn'),setHerd:py.globals.get('set_herd')};\n })();\n return runtimePromise;\n}\nfunction own(o:any,a:string){const me=+o.player,f=o.farms[me];let n=+(o.private?.shed?.[a]??0);for(const i of o.private?.inventories??[])n+=+(i?.[a]??0);for(const row of f.tiles??[])for(const t of row??[])if(t&&typeof t==='object'&&t.animal===a)n++;return n}\nfunction enforceAnimals(o:any,act:any,t:any){const x:any={COW:own(o,'COW'),SHEEP:own(o,'SHEEP')},market:any[]=[];for(const m of act?.market??[]){if(Array.isArray(m)&&m[0]==='BUY_ANIMAL'&&(m[1]==='COW'||m[1]==='SHEEP')){const a=m[1],q=Math.max(0,Math.min(+m[2]||1,t[a]-x[a]));if(q){market.push(['BUY_ANIMAL',a,q]);x[a]+=q}}else market.push(m)}return{...act,market}}\nfunction addFertSell(o:any,act:any){const q=+(o.private?.shed?.FERTILIZER??0),p=+(o.market?.prices?.FERTILIZER??0);if(q>0&&p>=60&&(act.market?.length??0)<10)return{...act,market:[...(act.market??[]),['SELL','FERTILIZER',Math.min(20,q)]]};return act}\nasync function playRefVsRef(a:string,b:string,seed:number,aSeat:0|1){const R=await ensureRuntime();await R.loadRef(a);await R.loadRef(b);R.resetRef(a);R.resetRef(b);const cfg=R.E.state.resolveConfig({seed});let s=R.E.state.initGameState(2,cfg,seed),errors=0;const timeline:any[]=[];while(!s.done){if(+s.hour===0&&CHECK_DAYS.has(+s.day)){const fa=profilePublic(s.farms[aSeat]),fb=profilePublic(s.farms[1-aSeat]);timeline.push({day:+s.day,a:fa,b:fb,shops:[...(s.town?.unlocked_shops??[])],market:{WHEAT:+s.market.prices.WHEAT,MELON:+s.market.prices.MELON,MILK:+s.market.prices.MILK,WOOL:+s.market.prices.WOOL,STRAWBERRY:+s.market.prices.STRAWBERRY}})}const acts:any[]=[];for(const pid of [0,1] as const){const o=observationFor(s,pid),name=pid===aSeat?a:b;try{acts[pid]=JSON.parse(String(R.callRef(name,JSON.stringify(o))))}catch(e){errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]}}}s=R.E.interpreter.step(s,acts,cfg)}const aBank=+s.farms[aSeat].money,bBank=+s.farms[1-aSeat].money;return{seed,aSeat,aBank,bBank,margin:aBank-bBank,winner:aBank>bBank?'A':aBank=1&&ow.SHEEP>=1;if(!comm)tg={COW:Math.max(1,ow.COW),SHEEP:Math.max(1,ow.SHEEP)}}if(comm){const total=ow.COW+ow.SHEEP,c=choosePastureAnimal(o,{cashBuffer:400,strategicWeight:0});if(total<16&&c.choice)tg[c.choice]+=Math.min(2,16-total);R.setHerd(tg.COW,tg.SHEEP,JSON.stringify(c.ranking.map((x:any)=>x.animal).filter((x:string)=>x==='COW'||x==='SHEEP')))}else R.setHerd(tg.COW,tg.SHEEP,JSON.stringify(['SHEEP','COW']))}let a=enforceAnimals(o,JSON.parse(String(R.callDyn(JSON.stringify(o)))),tg);if(variant==='MIX11_FERT'||variant==='M60_FERT')a=addFertSell(o,a);acts[pid]=a}else acts[pid]=JSON.parse(String(R.callRef(ref,JSON.stringify(o))))}catch(e){errors++;acts[pid]={farmer:['PASS'],hands:[],market:[]}}}s=R.E.interpreter.step(s,acts,cfg)}const ourBank=+s.farms[ourSeat].money,refBank=+s.farms[1-ourSeat].money;return{variant,ref,seed,ourSeat,ourBank,refBank,margin:ourBank-refBank,winner:ourBank>refBank?'OURS':ourBankg.winner==='A').length,winsB=games.filter(g=>g.winner==='B').length,ties=games.filter(g=>g.winner==='TIE').length;\n return{kind:jobType,a,b,seeds,episodes:games.length,winsA,winsB,ties,scoreA:(winsA+0.5*ties)/games.length,avgMarginA:avg(games.map(g=>g.margin)),avgBankA:avg(games.map(g=>g.aBank)),avgBankB:avg(games.map(g=>g.bBank)),errors:games.reduce((s,g)=>s+g.errors,0),gap:gapDiagnostics(games,'A'),games};\n }\n if(jobType==='OUR_VS_REFERENCE'){\n const variant=String(payload.variant??'MIX11'),ref=String(payload.ref),seeds=(payload.seeds??[]).map(Number);if(!ref||!seeds.length)throw new Error('OUR_VS_REFERENCE requires ref,seeds');if(!['MIX11','MIX11_FERT','M60_FERT'].includes(variant))throw new Error(`unsupported variant ${variant}`);\n const games:any[]=[];for(const seed of seeds){games.push(await playOurVsRef(variant,ref,seed,0));games.push(await playOurVsRef(variant,ref,seed,1))}\n const wins=games.filter(g=>g.winner==='OURS').length,losses=games.filter(g=>g.winner==='REF').length,ties=games.filter(g=>g.winner==='TIE').length;\n return{kind:jobType,variant,ref,seeds,episodes:games.length,wins,losses,ties,score:(wins+0.5*ties)/games.length,avgMargin:avg(games.map(g=>g.margin)),avgOurBank:avg(games.map(g=>g.ourBank)),avgRefBank:avg(games.map(g=>g.refBank)),errors:games.reduce((s,g)=>s+g.errors,0),gap:gapDiagnostics(games,'OURS'),games};\n }\n throw new Error(`unsupported job_type ${jobType}`);\n}\n\nexport function renderArtifactMarkdown(job:any,result:any){const lines=[`# Kaggriculture Autolab Run`,``,`- Core: ${AUTOLAB_CORE_VERSION}` ,`- Job: ${job.job_id}` ,`- Type: ${job.job_type}` ,`- Evidence: AUTOLAB_COMPUTE_RESULT_NOT_CHAMPION_PROOF`,`- Generated: ${new Date().toISOString()}`,``];if(result.kind==='PUBLIC_CROSSPLAY'){lines.push(`## Public crossplay`,``,`**${result.a} vs ${result.b}** — ${result.winsA}-${result.winsB}-${result.ties}, score A ${(100*result.scoreA).toFixed(1)}%, avg margin A ${result.avgMarginA.toFixed(1)}.`,``,`Largest observed gap-change window: ${JSON.stringify(result.gap.largestGapGrowth)}.`)}else{lines.push(`## Our agent vs reference`,``,`**${result.variant} vs ${result.ref}** — W/L/T ${result.wins}/${result.losses}/${result.ties}, score ${(100*result.score).toFixed(1)}%, avg margin ${result.avgMargin.toFixed(1)}.`,``,`Largest observed gap-change window: ${JSON.stringify(result.gap.largestGapGrowth)}.`)}lines.push(``,`## Guardrails`,``,`This artifact is laboratory evidence only. No Kaggle submission, spend, KYC, token use, or promotion to champion occurred. Public high-tier agents were used as black-box opponents; no non-licensed TRACE content was incorporated.`);return lines.join('\\n')}\n",[],[2112,2254],"bc8cbefe-eadd-4ab1-a695-c92931a29c9c","economic_policy.ts","4fb3747e-aa17-11f1-8e41-1607ee4eb77e",["D",1788715252312],"export type EconomicActionClass =\n | \"R0_OBSERVE\"\n | \"R1_PREPARE\"\n | \"R2_REVERSIBLE_WRITE\"\n | \"R3_COMMIT_SUBMIT\"\n | \"R4_ECONOMIC_SPEND\"\n | \"R5_IDENTITY_LEGAL\";\n\nexport type EconomicActionRequest = {\n actionClass: EconomicActionClass;\n approved?: boolean;\n amount?: number;\n maxBudget?: number;\n};\n\nexport type EconomicActionDecision = {\n allowed: boolean;\n humanGate: boolean;\n reason: string;\n};\n\nexport const ECONOMIC_POLICY = {\n version: \"0.6.0-candidate\",\n defaultBudget: 0,\n actionClasses: {\n R0_OBSERVE: \"Research/read/analyze only\",\n R1_PREPARE: \"Create local/draft artifacts without external commitment\",\n R2_REVERSIBLE_WRITE: \"Reversible external writes such as saving a draft\",\n R3_COMMIT_SUBMIT: \"External commitment such as submit/publish/PR\",\n R4_ECONOMIC_SPEND: \"Any action that spends money or credits\",\n R5_IDENTITY_LEGAL: \"KYC, banking, payout setup, legal or identity declarations\",\n },\n} as const;\n\nexport function evaluateEconomicAction(req: EconomicActionRequest): EconomicActionDecision {\n switch (req.actionClass) {\n case \"R0_OBSERVE\":\n case \"R1_PREPARE\":\n case \"R2_REVERSIBLE_WRITE\":\n return { allowed: true, humanGate: false, reason: \"AUTO_ALLOWED_BY_CLASS\" };\n case \"R3_COMMIT_SUBMIT\":\n return req.approved\n ? { allowed: true, humanGate: false, reason: \"CONTROLLED_WRITE_APPROVED\" }\n : { allowed: false, humanGate: true, reason: \"NEEDS_COMMIT_APPROVAL\" };\n case \"R4_ECONOMIC_SPEND\": {\n const amount = req.amount ?? 0;\n const budget = req.maxBudget ?? ECONOMIC_POLICY.defaultBudget;\n if (!req.approved) return { allowed: false, humanGate: true, reason: \"NEEDS_SPEND_APPROVAL\" };\n if (amount < 0) return { allowed: false, humanGate: true, reason: \"INVALID_NEGATIVE_AMOUNT\" };\n if (amount > budget) return { allowed: false, humanGate: true, reason: \"BUDGET_EXCEEDED\" };\n return { allowed: true, humanGate: false, reason: \"SPEND_WITHIN_APPROVED_BUDGET\" };\n }\n case \"R5_IDENTITY_LEGAL\":\n return { allowed: false, humanGate: true, reason: \"HUMAN_IDENTITY_OR_LEGAL_GATE\" };\n }\n}\n\nexport function netRevenue(input: {\n grossRevenue?: number;\n browserCost?: number;\n llmCost?: number;\n apiCost?: number;\n hostingCost?: number;\n platformFee?: number;\n transactionCost?: number;\n}) {\n const gross = input.grossRevenue ?? 0;\n const costs = (input.browserCost ?? 0) + (input.llmCost ?? 0) + (input.apiCost ?? 0) +\n (input.hostingCost ?? 0) + (input.platformFee ?? 0) + (input.transactionCost ?? 0);\n return gross - costs;\n}\n",[],[2112,2262],"ebc09386-8d31-425e-a75d-ce3486658bf8","dsl.ts","01a077bc-2b99-740d-9881-2409b3ecffa5",["D",1788715215771],"import type { BrowserTask } from \"./types.ts\";\n\nconst ALLOWED_ACTIONS = new Set([\n \"goto\", \"wait_for\", \"click\", \"fill\", \"fill_secret\", \"type\", \"select\",\n \"extract_text\", \"extract_attribute\", \"expect_text\", \"expect_url\",\n]);\n\nexport function validateTask(task: BrowserTask): string[] {\n const errors: string[] = [];\n if (!task || typeof task !== \"object\") return [\"task must be an object\"];\n if (!task.runId || task.runId.length > 128) errors.push(\"invalid runId\");\n if (!task.purpose || task.purpose.length > 1000) errors.push(\"invalid purpose\");\n if (!Array.isArray(task.steps) || task.steps.length < 1 || task.steps.length > 30) errors.push(\"steps must contain 1..30 actions\");\n if (![\"read_only\", \"reversible_write\", \"consequential\"].includes(task.risk)) errors.push(\"invalid risk\");\n if (task.risk !== \"read_only\" && task.approved !== true) errors.push(\"NEEDS_APPROVAL\");\n if (task.preferredProvider && ![\"kernel\", \"browserless\"].includes(task.preferredProvider)) errors.push(\"invalid preferredProvider\");\n if (task.evidenceRetention && ![\"minimal\", \"full\"].includes(task.evidenceRetention)) errors.push(\"invalid evidenceRetention\");\n\n let target: URL | undefined;\n try {\n target = new URL(task.targetUrl);\n if (![\"http:\", \"https:\"].includes(target.protocol)) errors.push(\"target protocol blocked\");\n } catch {\n errors.push(\"invalid targetUrl\");\n }\n\n const targetHost = target?.hostname;\n const allowed = new Set([targetHost, ...(task.allowedHosts ?? [])].filter(Boolean) as string[]);\n for (const [i, step] of (task.steps ?? []).entries()) {\n if (!step || !ALLOWED_ACTIONS.has((step as any).action)) {\n errors.push(`step ${i}: unsupported action`);\n continue;\n }\n const maybe = step as any;\n if (typeof maybe.selector === \"string\" && maybe.selector.length > 1000) errors.push(`step ${i}: selector too long`);\n if (typeof maybe.text === \"string\" && maybe.text.length > 10000) errors.push(`step ${i}: text too long`);\n if (maybe.action === \"fill_secret\") {\n if (typeof maybe.envVar !== \"string\" || !/^[A-Za-z_][A-Za-z0-9_]*$/.test(maybe.envVar)) errors.push(`step ${i}: invalid secret env var`);\n if (task.approved !== true) errors.push(\"NEEDS_SECRET_APPROVAL\");\n }\n if (maybe.action === \"extract_attribute\" && String(maybe.attribute).toLowerCase() === \"value\") {\n errors.push(`step ${i}: extracting value attribute is blocked to reduce secret leakage risk`);\n }\n if (maybe.action === \"goto\" && maybe.url) {\n try {\n const u = new URL(maybe.url);\n if (![\"http:\", \"https:\"].includes(u.protocol) || !allowed.has(u.hostname)) errors.push(`step ${i}: navigation host blocked`);\n } catch {\n errors.push(`step ${i}: invalid goto url`);\n }\n }\n }\n return [...new Set(errors)];\n}\n\nexport function buildKernelPlaywrightCode(task: BrowserTask): string {\n const plan = JSON.stringify(task);\n return `\nconst plan = ${plan};\nconst outputs = {};\nconst checks = [];\nconst targetHost = new URL(plan.targetUrl).hostname;\nconst allowedHosts = new Set([targetHost, ...(plan.allowedHosts || [])]);\nconst allowedMethods = new Set([\"GET\", \"HEAD\", \"OPTIONS\"]);\n\nconst assertUrl = (value) => {\n const u = new URL(value);\n if (![\"http:\", \"https:\"].includes(u.protocol)) throw new Error(\"blocked protocol\");\n if (!allowedHosts.has(u.hostname)) throw new Error(\"blocked host: \" + u.hostname);\n};\n\nawait page.route(\"**/*\", async (route) => {\n const req = route.request();\n try {\n if (req.isNavigationRequest() && req.frame() === page.mainFrame()) assertUrl(req.url());\n if (plan.risk === \"read_only\" && !allowedMethods.has(req.method().toUpperCase())) return route.abort(\"blockedbyclient\");\n return route.continue();\n } catch {\n return route.abort(\"blockedbyclient\");\n }\n});\n\npage.on(\"popup\", async (popup) => { try { await popup.close(); } catch {} });\nassertUrl(plan.targetUrl);\nawait page.goto(plan.targetUrl, { waitUntil: \"domcontentloaded\" });\nassertUrl(page.url());\n\nfor (const step of plan.steps) {\n switch (step.action) {\n case \"goto\": {\n const url = step.url || plan.targetUrl;\n assertUrl(url);\n await page.goto(url, { waitUntil: \"domcontentloaded\" });\n break;\n }\n case \"wait_for\":\n await page.locator(step.selector).waitFor({ state: \"visible\", timeout: step.timeoutMs || 10000 });\n break;\n case \"click\":\n if (step.waitForNavigation) {\n await Promise.all([\n page.waitForLoadState(\"domcontentloaded\"),\n page.locator(step.selector).click(),\n ]);\n } else await page.locator(step.selector).click();\n break;\n case \"fill\":\n await page.locator(step.selector).fill(step.text);\n break;\n case \"type\":\n await page.locator(step.selector).pressSequentially(step.text, { delay: step.delayMs || 0 });\n break;\n case \"select\":\n await page.locator(step.selector).selectOption(step.value);\n break;\n case \"extract_text\":\n outputs[step.key] = await page.locator(step.selector).innerText();\n break;\n case \"extract_attribute\":\n outputs[step.key] = await page.locator(step.selector).getAttribute(step.attribute);\n break;\n case \"expect_text\": {\n const body = await page.locator(\"body\").innerText();\n checks.push({ type: \"text\", expected: step.text, ok: body.includes(step.text) });\n break;\n }\n case \"expect_url\": {\n const actual = page.url();\n checks.push({ type: \"url\", expected: step.contains, actual, ok: actual.includes(step.contains) });\n break;\n }\n }\n assertUrl(page.url());\n}\n\nreturn {\n finalUrl: page.url(),\n outputs,\n checks,\n verified: checks.length > 0 && checks.every((x) => x.ok),\n};\n`;\n}\n\nexport function resolveSecrets(task: BrowserTask): BrowserTask {\n const steps = task.steps.map((step) => {\n if (step.action !== \"fill_secret\") return step;\n const value = Deno.env.get(step.envVar);\n if (value === undefined) throw new Error(`secret env var missing: ${step.envVar}`);\n return { action: \"fill\" as const, selector: step.selector, text: value };\n });\n return { ...task, steps };\n}\n",[],[2112,2269],"ffa9eab5-1611-4a35-91c1-b027af8f5f45","cost.ts","01a077bc-2b99-740d-9881-21df979511ad",["D",1788715215771],"import type { BrowserReceipt } from \"./types.ts\";\n\nexport function estimateProviderUnits(receipt: BrowserReceipt): number | null {\n if (receipt.status === \"BLOCKED\") return null;\n if (receipt.provider !== \"browserless\") return null;\n if (receipt.durationMs == null) return null;\n return Math.max(1, Math.ceil(receipt.durationMs / 30_000));\n}\n\nexport const COST_NOTES = {\n browserless: \"Estimated units cover browser-time only. Actual cost can be higher with CAPTCHA solving or built-in proxies.\",\n kernel: \"No local unit estimate: Kernel bills usage by GB-second; actual provider usage must be observed from the account.\",\n} as const;\n",[],[2112,2276],"94b02e1c-a9b9-457c-8c32-1b19e33e4ad5",["D",1788715215771],[276,277,278,279,280,281,282,283,284,285,286,287,288,289,290,291,292,293,294,295,296,297,298],[2112],"9b11c5f5-b155-47a5-b4df-46bb3b72111d","TASK_SCHEMA.md","01a077bc-2b99-740d-9881-199895a0596b","01a077bc-2b99-740d-9881-1e90a3c35ec2",["D",1788715215771],"# Browser Task Schema — V0.5\n\nTasks are inserted into the val-scoped SQLite database by the authenticated ChatGPT Val Town plugin. There is no public enqueue endpoint.\n\n## Required fields\n\n- `runId`: unique stable id, <=128 chars.\n- `targetUrl`: http/https URL.\n- `risk`: `read_only`, `reversible_write`, or `consequential`.\n- `purpose`: human-readable reason for the run.\n- `steps`: 1..30 DSL actions.\n\n## Optional fields\n\n- `approved`: must be `true` for `reversible_write`, `consequential`, and any `fill_secret` action.\n- `allowedHosts`: extra main-navigation hosts, explicit only.\n- `preferredProvider`: `kernel` or `browserless`.\n- `evidenceRetention`: `minimal` (default) or `full`.\n\n## DSL actions\n\n- `goto`\n- `wait_for`\n- `click`\n- `fill`\n- `fill_secret`\n- `type`\n- `select`\n- `extract_text`\n- `extract_attribute` (attribute `value` is blocked)\n- `expect_text`\n- `expect_url`\n\n## Secret rule\n\nNever put a password, cookie, API token, or OTP secret in `task_json`.\nUse `fill_secret` with an environment-variable name. The value is resolved in execution memory only after approval.\n\n## Evidence retention\n\n`minimal` persists status, URL, checks, errors, duration and fallback metadata but omits extracted outputs and provider session ids.\n`full` persists the full receipt and should be used only when the extracted data is intentionally safe to retain.\n\n## Retry policy\n\n- Only `read_only` + provider `FAIL` is retryable.\n- Maximum 2 task attempts.\n- Retry delay: 10 seconds.\n- `BLOCKED`, `PARTIAL`, writes and consequential tasks are never automatically retried.\n- No cron/automatic runner is enabled, so a retry becomes eligible but does not consume browser credits until ChatGPT explicitly runs `runner.ts` again.\n\n## Provider fallback\n\n- Kernel is primary.\n- Browserless is fallback only for `read_only` tasks.\n- No cross-provider retry for writes.\n\n## PASS rule\n\nA successful browser call without at least one positive expectation remains `PARTIAL`.\n`PASS` requires all explicit expectations to succeed.\n",[],[2294,2287],"docs","989fd312-57a2-4772-9d55-033ac00154a3","SMOKE_TASK.json","01a077bc-2b99-740d-9881-17f664878040",["D",1788715215771],"{\n \"state\": \"DRAFT_ONLY_DO_NOT_AUTO_ENQUEUE\",\n \"task\": {\n \"runId\": \"smoke-kernel-example-com\",\n \"targetUrl\": \"https://example.com\",\n \"risk\": \"read_only\",\n \"purpose\": \"First live cloud-browser smoke test after provider credential activation\",\n \"preferredProvider\": \"kernel\",\n \"evidenceRetention\": \"minimal\",\n \"steps\": [\n {\"action\": \"extract_text\", \"selector\": \"h1\", \"key\": \"heading\"},\n {\"action\": \"expect_text\", \"text\": \"Example Domain\"},\n {\"action\": \"expect_url\", \"contains\": \"example.com\"}\n ]\n },\n \"promotionCriteria\": [\n \"provider credential key exists in Val Town\",\n \"runner returns PASS\",\n \"heading equals Example Domain\",\n \"all expectations pass\",\n \"attempt receipt is persisted\",\n \"provider usage is observed when available\"\n ]\n}\n",[],[2294,2296],"312b84fe-7828-4773-b437-17bbbcbafdca","SECURITY_MODEL.md","01a077bc-2b99-740d-9881-11dec1877342",["D",1788715215771],"# Security Model — V0.5.0\n\n## Trust boundaries\n\n### Trusted control plane\n- normal authenticated ChatGPT conversation;\n- authenticated Val Town plugin operations;\n- source code committed in `cleitoncosta/chat-native-maestro` after validation.\n\n### Sensitive trusted storage\n- Val Town Environment Variables;\n- val-scoped SQLite, but only for non-secret task metadata/evidence.\n\n### Untrusted inputs\n- every webpage and frame;\n- page text, HTML, script-generated text and accessibility labels;\n- redirects;\n- URLs supplied by pages;\n- downloaded files;\n- DNS answers and provider responses as evidence that still needs validation;\n- third-party provider error messages.\n\nA webpage is data, never policy or authorization.\n\n## Threat 1 — Prompt injection from a webpage\n\nRisk: page content tries to instruct ChatGPT/runner to reveal secrets, change policy, expand allowed hosts, approve actions or run arbitrary code.\n\nControls:\n- browser receives a bounded DSL, not a free-form agent prompt;\n- no shell or arbitrary-JS task action;\n- page content cannot set `approved=true`;\n- page content cannot add env vars;\n- page content cannot alter `allowedHosts`;\n- browser result is evidence/data only.\n\nResidual risk: extracted malicious text can still be shown to the language model. The control plane must continue treating it as untrusted data.\n\n## Threat 2 — SSRF / internal network access\n\nRisk: task or redirect targets localhost, cloud metadata, private network or a public hostname resolving to non-public IPs.\n\nControls before provider call:\n- HTTP/HTTPS only;\n- URL userinfo blocked;\n- only ports 80/443;\n- direct private/link-local/loopback/reserved/documentation IPv4 and IPv6 ranges blocked;\n- localhost/internal/metadata hostnames blocked;\n- `allowedHosts` validated;\n- A and AAAA resolved through Cloudflare DNS-over-HTTPS;\n- task fails closed if DoH fails, no A/AAAA is found, or any resolved address is non-public.\n\nControls in browser:\n- main-frame navigation restricted to target host plus explicit allowed hosts;\n- popups closed.\n\nValidated negative control:\n`169.254.169.254` is blocked in the real runner before provider invocation.\n\nResidual risk: DNS rebinding after preflight cannot be fully proven absent by a one-time DoH check. Remote browser isolation and navigation checks remain required.\n\n## Threat 3 — Secret leakage\n\nRisk: credentials enter task JSON, source, logs or persistent receipts.\n\nControls:\n- provider credentials only in Val Town env vars;\n- site credentials referenced by env-var name with `fill_secret`;\n- secret value resolved in memory immediately before execution;\n- `fill_secret` requires explicit approval;\n- `extract_attribute: value` blocked;\n- default `evidenceRetention=minimal` omits extracted outputs and provider session IDs;\n- snapshot/export must exclude env-var values and raw sensitive task payloads.\n\nResidual risk: a target site can reflect a submitted secret into ordinary DOM text. The orchestrator must not deliberately extract selectors expected to contain secrets, and `full` retention must not be used for sensitive workflows without a specific reason.\n\n## Threat 4 — Duplicate/replayed writes\n\nRisk: timeout or provider error occurs after a write happened, then another attempt repeats the effect.\n\nControls:\n- automatic retry only for `read_only + FAIL`;\n- no automatic retry for `reversible_write` or `consequential`;\n- no cross-provider fallback for writes;\n- unique `run_id` in SQLite;\n- write tasks require explicit approval.\n\nResidual risk: even a read-classified website can mutate state via GET. The read-only method firewall is strong but cannot prove zero side effects on a badly designed server.\n\n## Threat 5 — Read-only task causes state mutation\n\nControls:\n- browser request firewall allows only GET/HEAD/OPTIONS for `read_only`;\n- common POST/PUT/PATCH/DELETE side effects are blocked;\n- explicit expectations and post-action verification required.\n\nResidual risk:\n- GET endpoints may have side effects;\n- service-worker/browser behavior can be complex;\n- some legitimate read APIs use POST/GraphQL and will fail under strict mode. Do not weaken strict mode silently; use a connector/fetch route or explicitly reclassify the task.\n\n## Threat 6 — Provider degradation / repeated spend\n\nRisk: provider outage causes repeated browser sessions or runaway retries.\n\nControls:\n- max 2 task attempts;\n- 10-second retry eligibility delay;\n- no automatic scheduler;\n- provider circuit opens after 3 consecutive FAILs;\n- 15-minute cooldown;\n- BLOCKED config errors do not increment failure count;\n- PASS/PARTIAL resets provider failures;\n- browser timeouts;\n- Kernel cleanup in `finally`.\n\n## Threat 7 — Stuck RUNNING job\n\nRisk: Val Town/runtime interruption leaves queue permanently locked.\n\nControls:\n- 90-second lease;\n- stale lease recovery;\n- retry budget checked during recovery;\n- terminal failure when retry budget exhausted.\n\nValidated with a real stale-job fixture that was removed after testing.\n\n## Threat 8 — Cost surprise\n\n### Kernel\n- recurring free credits are finite;\n- no automatic runner;\n- short browser timeout;\n- cleanup in `finally`;\n- actual Kernel GB-second usage remains provider-authoritative.\n\n### Browserless\n- local estimate covers base browser-time units only;\n- CAPTCHA/proxy add-ons are not inferred as free;\n- before activation, provider overage spend limit must be set to `$0`;\n- no automatic browser cron;\n- compare local estimated units to provider accounting.\n\n## Threat 9 — False PASS\n\nControls:\n- `TOOL_SUCCESS_RESPONSE != BEHAVIOR_PROOF`;\n- no checks -> `PARTIAL`, not PASS;\n- PASS requires at least one positive expectation and all expectations true;\n- provider/module/pricing availability never promotes a tool to OPERATIONAL;\n- credential, browser creation, actual action, validation and usage evidence are separate states.\n\n## Threat 10 — Public Free-tier source exposure\n\nThe personal Val Town Free workspace requires public source visibility.\n\nControls:\n- no secrets in source;\n- public HTTP endpoint exposes liveness only;\n- detailed task/provider state stays behind authenticated plugin execution;\n- env-var values are never returned by normal Val Town listing tools;\n- source assumes it is publicly readable.\n\n## Evidence-state vocabulary\n\n- `AVAILABLE`: provider/service exists.\n- `ADAPTER_LOADED`: code imports in Val Town.\n- `CONFIGURED`: required env-var key exists.\n- `AUTHENTICATED`: credential proved usable.\n- `AUTHORIZED`: required action is permitted.\n- `EXECUTED`: real provider action occurred.\n- `VERIFIED`: postcondition proved.\n- `OPERATIONAL`: full required chain is repeatedly or sufficiently evidenced for intended scope.\n- `BLOCKED`: known prerequisite prevents execution.\n- `UNVERIFIED`: evidence is absent; not equivalent to failure.\n\nCurrent Kernel and Browserless live-browser state remains `UNVERIFIED` because no user-created provider env vars exist yet.\n",[],[2294,2303],"43c57926-d688-45f9-bdd4-705fbd5186c1","PROVIDER_CENSUS.md","01a077bc-2b99-740d-9881-0e23597c5715",["D",1788715215771],"# Browser Provider Census — V0.5.0\n\nResearch snapshot: 2026-09-06.\n\nRanking target: normal ChatGPT mobile/web, no user PC, recurring free allowance, deterministic browser control, strong isolation, minimal browser-open time and zero-surprise-cost preference.\n\n## 1. Kernel — CHAMPION\n\nRuntime state: `ACCOUNT_KEY_REQUIRED` / live browser `UNVERIFIED`.\n\nCurrent official Free plan:\n- $0/month plus usage costs\n- $5/month free credits\n- 5 concurrent browsers\n- live view\n- stealth mode including CAPTCHA solver/proxy\n- managed auth (3 connections)\n- one-day session replays\n- uploads/downloads\n- no card required to get started\n- usage billed by GB-second\n\nTechnical fit:\n- official Val Town browser-automation guide/starter\n- `@onkernel/sdk`\n- server-side Playwright execution in the browser VM\n- on-demand `browsers.create()` is explicitly recommended for occasional/bursty/one-off work\n- browser is destroyed in `finally`\n\nAdapter: `providers/kernel.ts`.\nRequired env var: `KERNEL_API_KEY`.\n\nOfficial sources:\n- https://www.kernel.sh/pricing\n- https://www.kernel.sh/docs\n- https://docs.val.town/guides/browser-automation/kernel\n\n## 2. Browserless — RECURRING-FREE FALLBACK\n\nRuntime state: `ACCOUNT_TOKEN_REQUIRED` / live browser `UNVERIFIED`.\n\nCurrent official Free plan:\n- 1,000 units/month\n- 2 concurrent browsers\n- 2-minute max session\n- no card required\n- Chrome / WebKit / Firefox\n- automatic CAPTCHA solving available\n- persisted sessions/replays\n\nUnit model:\n- browser time: 1 unit per 30-second connection block\n- partial blocks round up\n- successful CAPTCHA solve: 10 additional units\n- residential proxy: 6 units/MB\n- datacenter proxy: 2 units/MB\n\nSeptember 2026 provider control:\n- account overage spend limit can be set to `$0`\n- this should be mandatory before promotion to fallback OPERATIONAL under this project's zero-surprise-cost policy\n\nAdapter: `providers/browserless.ts`.\nRequired env var: `BROWSERLESS_TOKEN`.\nOptional: `BROWSERLESS_BASE_URL`.\n\nOfficial sources:\n- https://www.browserless.io/pricing\n- https://docs.browserless.io/overview/unit-consumption\n- https://www.browserless.io/changelog/2026-09\n\n## 3. Browserbase — RESEARCHED FALLBACK\n\nState: researched, not implemented.\n\nVal Town has an official Browserbase integration guide and starter. It remains behind Kernel/Browserless because the current combination above better matches recurring-free allowance + CAPTCHA/anti-bot + cost control for this project.\n\nSource:\n- https://docs.val.town/guides/browser-automation/browserbase\n\n## 4. Steel — RESERVE / ONE-TIME-CREDIT PATH\n\nState: researched, not implemented.\n\nUseful properties:\n- open-source browser API\n- Puppeteer/Playwright\n- official Val Town integration guide\n\nLower priority because one-time starter credit is less aligned with recurring zero-cost steady state than Kernel/Browserless.\n\nSource:\n- https://docs.val.town/guides/browser-automation/steel\n\n## Routing rules V0.5\n\n1. Explicit configured `preferredProvider` is honored unless its circuit is OPEN.\n2. Otherwise Kernel is primary.\n3. Browserless fallback is allowed only for `read_only`.\n4. Provider `BLOCKED` for missing configuration does not increment health failures.\n5. Provider `FAIL` increments consecutive failures.\n6. Three consecutive FAILs open the provider circuit for 15 minutes.\n7. PASS/PARTIAL resets provider failure state.\n8. Never automatically retry a write on another provider.\n9. Keep reasoning outside the browser; send only a short deterministic DSL plan.\n10. No automatic cron runs browsers.\n\n## Promotion rule\n\nA provider cannot become `OPERATIONAL` from docs/pricing/module-load alone. Required evidence:\n- env-var key exists\n- credential authenticates\n- real browser is created\n- network preflight passes\n- read-only smoke plan executes\n- explicit expectations PASS\n- attempt is persisted\n- provider health remains CLOSED\n- actual account consumption is observed if available\n\nUntil all of these are true, state remains `UNVERIFIED` even if the SDK call returns successfully.\n",[],[2294,2310],"875dc314-896d-458c-b402-0a0485726280","CURRENT_STATE.md","01a077bc-2b98-7186-8f2c-a206cdcc09fd",["D",1788715215771],"# CURRENT_STATE — V0.5.0\n\nUpdated: 2026-09-06\n\n## Canonical runtime\n\n`normal ChatGPT mobile/web -> authenticated Val Town plugin -> val-scoped SQLite -> runner.ts -> provider router -> cloud browser -> receipt`\n\nNo Codex and no user computer are required.\n\n## Val Town\n\n- Val: `cleitoncosta/chat-native-maestro`\n- Personal workspace tier: Free\n- Code visibility: public (Free-tier constraint)\n- Public HTTP: minimal liveness only\n- Detailed diagnostics: authenticated `status.ts`\n- Task execution: authenticated Val Town plugin only\n- Persistent state: val-scoped SQLite\n- Schema/app version: 0.5.0\n- Automatic runner/cron: disabled\n\n## Implemented files\n\n### Entrypoints\n- `main.ts` — minimal public GET-only liveness endpoint.\n- `status.ts` — authenticated provider/task/cost/security diagnostics.\n- `runner.ts` — queue lease, validation, network preflight, routed execution, retry decision, receipt persistence.\n- `selftest.ts` — stateless DSL/config regression tests.\n- `stateful_selftest.ts` — circuit breaker, timeout and retry-state regression tests with state restoration.\n- `url_selftest.ts` — URL/DNS/SSRF regression tests.\n\n### Libraries\n- `lib/types.ts` — typed task/receipt/evidence schema.\n- `lib/dsl.ts` — DSL validation, browser program generation, in-memory secret resolution.\n- `lib/policy.ts` — retry/lease/timeout/circuit constants and pure disposition logic.\n- `lib/provider_health.ts` — circuit breaker state machine.\n- `lib/cost.ts` — provider cost-estimation rules.\n- `lib/url_security.ts` — fail-closed URL/DNS-over-HTTPS preflight.\n\n### Providers\n- `providers/router.ts` — provider routing/fallback with health tracking.\n- `providers/kernel.ts` — primary browser adapter.\n- `providers/browserless.ts` — secondary browser adapter.\n\n### Documentation\n- `README.md`\n- `docs/CURRENT_STATE.md`\n- `docs/PROVIDER_CENSUS.md`\n- `docs/ACTIVATION_RUNBOOK.md`\n- `docs/TASK_SCHEMA.md`\n- `docs/SMOKE_TASK.json`\n\n## Database schema V0.5\n\n### `browser_tasks`\nDurable queue and terminal receipt state. New V0.5 fields:\n- `attempt_count`\n- `claimed_at`\n- `next_attempt_at`\n\n### `browser_attempts`\nOne row per provider attempt/fallback with:\n- attempt number\n- provider\n- timestamps\n- status/error\n- duration\n- sanitized receipt\n- estimated Browserless units\n\n### `provider_health`\nCircuit state per provider:\n- `CLOSED` / `HALF_OPEN` / `OPEN`\n- consecutive failures\n- cooldown timestamp\n- last success/failure/error\n\n### `provider_census`\nProvider ranking and current free-tier facts.\n\n### `schema_meta`\n`schema_version=0.5.0`, `app_version=0.5.0`.\n\n## Operational policy V0.5\n\n- max attempts per task: 2\n- read-only FAIL retry delay: 10 s\n- RUNNING lease: 90 s\n- browser execution timeout: 25 s\n- provider circuit opens at 3 consecutive FAILs\n- circuit cooldown: 15 min\n- automatic cross-provider fallback: read-only only\n- automatic scheduled execution: disabled\n\nA retry can become `PENDING`, but it cannot consume browser credits until ChatGPT explicitly invokes the runner again.\n\n## Network security\n\nBefore any provider execution:\n- only HTTP/HTTPS\n- URL credentials/userinfo blocked\n- nonstandard ports blocked\n- private/link-local/loopback/reserved/documentation IPv4/IPv6 blocked\n- localhost/internal/metadata hostnames blocked\n- allowed hosts validated\n- A/AAAA resolved through Cloudflare DNS-over-HTTPS\n- any resolved non-public IP blocks the task\n- DNS/preflight failure fails closed\n\nRunner-level negative-control proof:\n`ssrf-runner-v05 -> BLOCKED / NETWORK_PREFLIGHT_BLOCKED: blocked hostname: 169.254.169.254`\n\nThe fixture was deleted after verification.\n\n## Secret policy\n\nTask JSON never stores secret values.\n`fill_secret` stores only an env-var name and requires explicit approval.\nThe value is resolved in memory immediately before provider execution.\n\n`extract_attribute` of `value` is blocked to reduce accidental credential leakage.\n\nCurrent user-created Val Town env-var count: **0**.\n\n## Evidence retention\n\nDefault: `minimal`.\nPersisted by default:\n- run/provider/status\n- final URL\n- checks\n- errors\n- duration\n- fallback evidence\n- network preflight evidence\n\nNot persisted by default:\n- extracted outputs\n- provider session ID\n\n`full` must be intentionally requested.\n\n## Cost policy\n\n### Kernel\nOfficial current Free plan verified 2026-09-06:\n- $0/month plus usage\n- $5 monthly free credits\n- 5 concurrent browsers\n- no card required to start\n- billed by GB-second\n\nLocal Kernel cost estimate remains unsupported; actual account usage must be observed.\n\n### Browserless\nOfficial current Free plan verified 2026-09-06:\n- 1,000 units/month\n- 2 concurrent browsers\n- max session 2 min\n- no card\n- browser time 1 unit/30 s\n- CAPTCHA/proxy additional\n- overage cap should be set to $0 if Browserless is activated\n\nThe runtime estimates base browser-time units only; actual provider accounting remains authoritative.\n\n## Evidence states\n\n### PASS\n\n- ChatGPT Val Town plugin authentication.\n- Free workspace enumerated.\n- create/edit/run/deploy.\n- public liveness HTTP 200.\n- authenticated diagnostics.\n- SQLite schema/migrations/versioning.\n- atomic queue claim.\n- empty queue behavior.\n- stale RUNNING lease recovery.\n- task attempt persistence.\n- retry policy pure test.\n- timeout test.\n- provider circuit OPEN after three synthetic FAILs and restoration to CLOSED.\n- consequential action approval gate.\n- secret approval gate.\n- invalid provider rejection.\n- `value` attribute extraction block.\n- Kernel SDK adapter module load.\n- Browserless adapter module load.\n- missing provider credential -> BLOCKED.\n- DoH availability through normal Val Town HTTPS egress.\n- public-domain DNS preflight.\n- local/private/metadata/userinfo/nonstandard-port URL blocks.\n- runner-level SSRF block before provider invocation.\n- automatic schedule disabled.\n\n### UNVERIFIED\n\n- Kernel API authentication for this account.\n- live Kernel browser creation.\n- live Kernel server-side Playwright execution.\n- Kernel real credit consumption.\n- Browserless API authentication for this account.\n- live Browserless `/function` execution.\n- Browserless real unit consumption.\n\n## Persistent historical negative controls\n\nThese three V0.4 records intentionally remain in `browser_tasks` as historical evidence:\n1. `neg-control-no-kernel-key` -> BLOCKED / `KERNEL_API_KEY missing`.\n2. `neg-control-no-browserless-token` -> BLOCKED / `BROWSERLESS_TOKEN missing`.\n3. `neg-control-secret-not-approved` -> BLOCKED / `NEEDS_SECRET_APPROVAL`.\n\nV0.5 synthetic lease/circuit/SSRF fixtures were removed after tests so production metrics are not polluted.\n\n## Next incomplete milestone\n\nExactly one external action remains necessary: configure a recurring-free browser credential, preferably `KERNEL_API_KEY`, directly in Val Town env vars. Do not paste it into chat.\n\nAfter the key exists:\n1. confirm only the env-var **name** exists;\n2. enqueue `docs/SMOKE_TASK.json` explicitly;\n3. execute `runner.ts` once;\n4. inspect task + attempt receipt;\n5. verify `heading = Example Domain` and all checks PASS;\n6. inspect provider usage;\n7. only then promote Kernel from `ACCOUNT_KEY_REQUIRED/UNVERIFIED` to `OPERATIONAL`.\n",[],[2294,2317],"59787f7c-1e09-4175-b579-f752ee64d035","ACTIVATION_RUNBOOK.md","01a077bc-2b98-7186-8f2c-9ca4b3fc4a0e",["D",1788715215771],"# Activation Runbook — V0.5.0\n\n## Objetivo\n\nAtivar exatamente um browser cloud sem depender de Codex ou computador do usuário e sem criar execução automática que possa consumir franquia sozinha.\n\nO provider preferido para a primeira ativação é **Kernel**.\n\n## 1. Kernel — ativação preferencial\n\n### Conta e chave\n\n1. Criar ou entrar em uma conta Kernel Free.\n2. Gerar uma API key no dashboard Kernel.\n3. Abrir diretamente o editor de Environment Variables do Val Town.\n4. Criar `KERNEL_API_KEY` com o valor da chave.\n5. Não colar a chave no ChatGPT, README, código, SQLite, issue, log ou task JSON.\n\nVal Town env editor:\nhttps://www.val.town/x/cleitoncosta/chat-native-maestro/environment-variables?key=KERNEL_API_KEY\n\nKernel API key dashboard:\nhttps://dashboard.onkernel.com/api-keys\n\n### Após gravar a chave\n\nNão executar browser imediatamente.\n\n1. ChatGPT consulta `list_env_vars` e confirma apenas que o **nome** `KERNEL_API_KEY` existe. O valor não deve ser retornado.\n2. Executar `selftest.ts` e `status.ts`.\n3. Confirmar:\n - Kernel `configured=true`;\n - circuit `CLOSED`;\n - nenhum job `PENDING` inesperado;\n - `automaticSchedule=disabled`.\n4. Ler `docs/SMOKE_TASK.json`.\n5. Inserir explicitamente a task draft no SQLite como `PENDING` somente nesse momento.\n6. Executar `runner.ts` exatamente uma vez.\n7. Ler o registro de `browser_tasks` e `browser_attempts` do smoke.\n8. Exigir:\n - receipt `PASS`;\n - `finalUrl` contém `example.com`;\n - `expect_text` passou;\n - `expect_url` passou;\n - output corrente `heading` é `Example Domain`;\n - attempt persistido;\n - circuit continua `CLOSED`.\n9. Consultar uso/créditos no Kernel, quando disponível.\n10. Somente depois atualizar o estado do Kernel para `OPERATIONAL`.\n\n### Promoção rejeitada se\n\n- SDK retorna sucesso mas expectations falham;\n- receipt é `PARTIAL`;\n- browser cria mas não navega;\n- network preflight falha;\n- circuit abre;\n- consumo não pode ser observado quando isso é material à decisão;\n- qualquer etapa é apenas inferida, não executada.\n\n## 2. Browserless — fallback\n\nAtivar somente se Kernel onboarding estiver bloqueado ou para validar redundância.\n\n### Antes da chave\n\nNo dashboard Browserless, configurar o **overage spend limit para $0**. Isso é requisito deste projeto para evitar cobrança surpresa.\n\nDepois:\n1. criar/entrar na conta Browserless Free;\n2. obter o token;\n3. salvar como `BROWSERLESS_TOKEN` no Val Town;\n4. opcionalmente salvar `BROWSERLESS_BASE_URL` se o account endpoint regional diferir do default;\n5. nunca colar o token em chat/código/SQLite.\n\nVal Town env editor:\nhttps://www.val.town/x/cleitoncosta/chat-native-maestro/environment-variables?key=BROWSERLESS_TOKEN\n\n### Smoke Browserless\n\nUsar a mesma task de `docs/SMOKE_TASK.json`, alterando apenas:\n\n```json\n\"preferredProvider\": \"browserless\"\n```\n\nExigir os mesmos critérios de PASS e, adicionalmente:\n- registrar `estimated_units`;\n- comparar com units reais do provider;\n- lembrar que CAPTCHA/proxy podem gerar units adicionais e não estão incluídos na estimativa local.\n\n## 3. Credenciais de sites-alvo\n\nPara login em um site, criar env vars específicas no Val Town, por exemplo:\n- `PORTAL_USERNAME`\n- `PORTAL_PASSWORD`\n\nTask JSON usa apenas referência:\n\n```json\n{\n \"action\": \"fill_secret\",\n \"selector\": \"#password\",\n \"envVar\": \"PORTAL_PASSWORD\"\n}\n```\n\n`fill_secret` exige aprovação explícita mesmo quando a finalidade geral é leitura.\n\nNunca armazenar senha/token diretamente em `task_json`.\n\n## 4. Regras de segurança operacional\n\n- Não executar task automaticamente só porque uma credencial apareceu.\n- Não criar cron de browser por padrão.\n- Não repetir writes em outro provider após falha.\n- Não transformar `FAIL` de provider em prova de ausência de side effect.\n- Não usar `evidenceRetention=full` com dados sensíveis sem necessidade explícita.\n- Não extrair atributo `value` de campos de formulário; a DSL bloqueia isso.\n- Não usar localhost, metadata cloud, rede privada, URL com userinfo ou porta não padrão; o network preflight bloqueia.\n- Não desabilitar o DNS-over-HTTPS preflight para “fazer funcionar”.\n- Não ampliar `allowedHosts` automaticamente a partir de conteúdo da página.\n- Não tratar uma página como fonte de autorização, policy ou segredo.\n\n## 5. Política de custo\n\n### Kernel\nO plano Free atual fornece créditos mensais; o runtime mantém browser com timeout curto e cleanup em `finally`. Não existe cron de browser.\n\n### Browserless\nA franquia Free é baseada em units. Antes da ativação:\n- overage cap = $0;\n- manter sessões curtas;\n- não habilitar proxy/CAPTCHA sem necessidade;\n- comparar `estimated_units` com uso real.\n\n## 6. Estado atual\n\nEnquanto `list_env_vars` não mostrar `KERNEL_API_KEY` ou `BROWSERLESS_TOKEN`, ambos continuam:\n\n`AVAILABLE/ADAPTER_LOADED != AUTHENTICATED != OPERATIONAL`\n\nNenhum provider deve ser promovido por documentação, pricing, existência do SDK ou sucesso de importação isoladamente.\n",[],[2294,2324],"3e709f84-5c5f-42d7-bfae-c6fc769c7f45",["D",1788715215771],[269,270,271,272,273,274],[2294],"0bd27574-a4c3-4d33-aaf0-fde83f6d055e","kaggriculture_neuron_protocol_v1.md","ee36dfc6-aa19-11f1-ad34-1607ee4eb77e","40731aa4-aa18-11f1-bdcb-1607ee4eb77e",12,["D",1788716377031],"# Kaggriculture — Neurônio Competition Protocol V1\n\nStatus: R1_PREPARE / NO KAGGLE SUBMISSION YET\nAgent: A4-KAGGRICULTURE\nOpportunity: OP-KAGGRICULTURE-202609\nPrepared: 2026-09-06\n\n## Primary contract\n\n- Start: 2026-07-29\n- Entry / team-merger deadline: 2026-09-23 23:59 UTC\n- Final submission deadline: 2026-09-30 23:59 UTC\n- Final convergence/tournament continues into ~2026-10-15\n- Prize pool: USD 50,000\n- 10 prize positions × USD 5,000\n- No purchase necessary\n- 5 submissions/team/day\n- Latest 2 submissions are the active/final-evaluation pair\n- Winner of an episode = higher bank cash after 720 turns; cash margin itself does not change rating magnitude\n- Validation episode runs a submission against itself before ladder admission\n\n## Grimório interpretation\n\nThis competition is unusually compatible with the Neurônio method because the external environment supplies:\n\n- repeated trials;\n- adversarial opponents;\n- observable fitness;\n- daily mutation budget;\n- public replay evidence;\n- a final tournament;\n- an explicit economic objective.\n\nDo NOT use Kaggle rating as the only truth. Public discussion documents meaningful rating noise and large rating differences between identical submissions. Use paired local matches + ladder as independent evidence.\n\n## Champion / challenger slot policy\n\nBecause only the latest 2 submissions are used for final evaluation:\n\n1. Slot C = CURRENT_PROVEN_CHAMPION.\n2. Slot X = CHALLENGER.\n3. Never replace both slots with unproven mutations in the same generation.\n4. If X beats C in controlled evaluation and survives ladder falsification, X becomes C.\n5. Re-submit/freeze the champion when necessary to keep one latest slot occupied by a known-good artifact.\n6. Preserve exact code hash, parameters, tests, seed set and evidence for every champion epoch.\n\nThis is the competition-level equivalent of `CURRENT_PROVEN_CHAMPION`.\n\n## What current public meta evidence says\n\nCommunity evidence is discovery/benchmark evidence, not authority:\n\n- A fixed-asset strategy has been reported around 81% win rate and became widely copied.\n- A recently observed highly adaptive #1 agent was reported 40-0 over a 40-game public x-ray window with median margin around +13k before being retired by its team.\n- Another public report reached ~80k terminal cash with PPO but hit a wall and moved toward a hybrid policy + deterministic executor.\n- Public analysis repeatedly identifies path/walking inefficiency as a major bottleneck; early work reported 83% of unit-turns moving, improved toward ~55% by avoiding target thrashing.\n- The first ~10 daily hires are inexpensive relative to farm economics; labor-action capacity is generally cheaper than bad routing.\n- Static replay/meta agents remain strong, so novelty alone is not fitness.\n\n## Official-engine facts that matter to strategy\n\n- 30 days × 24 turns = 720 turns.\n- Starting money default: 3000.\n- Farm starts with one unlocked 5×5 quadrant.\n- Additional land: 1000, 2000, 4000 in fixed order.\n- Hiring resets daily and follows Fibonacci cost × multiplier.\n- Opponent farm is visible; opponent shed/private inventory is hidden.\n- Market inventory/prices are shared and persistent.\n- Selling pushes a resource into the glut branch for both players.\n- Town demand removes products from shared market inventory and can create scarcity.\n- Melon has no specialty-shop demand; town center is its principal exogenous sink.\n- Animals can create fertilizer daily; uncollected fertilizer does not accumulate.\n- Shed capacity is finite; excess end-of-day inventory can be discarded.\n- End-state winner uses cash in bank, so unliquidated inventory/capital is dangerous near horizon.\n\n## Architecture hypothesis\n\nThe strongest zero/low-compute direction is not pure LLM, pure RL, or a fixed tape.\n\nUse:\n\n`ADAPTIVE MACRO PLANNER → DETERMINISTIC TASK GRAPH → PATH/ACTION EXECUTOR → MARKET EXECUTOR`\n\n### Macro planner\n\nRecomputes only when meaningful state changes occur:\n\n- new day;\n- town shop unlock;\n- material price movement;\n- opponent commits to a new production cluster;\n- land/animal/crop lifecycle event;\n- endgame horizon threshold.\n\nOutputs desired asset mix, harvest/sell timing, land/hire budget and target inventory.\n\n### Deterministic executor\n\nConverts macro intent into explicit work orders and avoids per-turn retargeting:\n\n1. finish all useful same-tile operations before moving;\n2. assign persistent tasks to units;\n3. minimize Manhattan movement / use clustered layouts;\n4. prioritize survival actions (water/feed) over yield optimization;\n5. preserve item/shed capacity;\n6. use hands only when marginal task value exceeds hire + walking cost.\n\n### Market executor\n\nPredicts near-future shared inventory from:\n\n- current market inventory;\n- known town demand;\n- our visible production pipeline;\n- opponent visible crops/animals;\n- our hidden inventory;\n- horizon.\n\nThen chooses `sell now`, `hold`, or `produce another resource` based on expected marginal sale price, not base price.\n\n## Initial Neurônio branches\n\nThese are research variants, NOT economic offspring. Economic offspring are created only after positive real net revenue.\n\n### K0 — Deterministic control\n\nPurpose: stable control and regression oracle.\n\n- compact 1–3 quadrant layout;\n- deterministic task completion;\n- simple horizon-aware liquidation;\n- no opponent adaptation except basic price observation.\n\nMust be reproducible across seeds.\n\n### K1 — Labor/route optimizer\n\nMutation: K0 + workload-derived hands + persistent routing.\n\nHypothesis: reducing walking/thrashing provides larger win gain than adding a more sophisticated crop-value model.\n\nAblation: same macro policy with fixed hands / naive nearest-target routing.\n\n### K2 — Livestock/fertilizer flywheel\n\nMutation: K1 + animal-centric asset allocation.\n\n- value named product + fertilizer stream jointly;\n- feed/care/collect as co-located task bundle;\n- use fertilizer where marginal crop gain beats fertilizer sale value;\n- adapt animal type to observed shop demand.\n\nFalsifier: seasons with weak product demand; action/walking cost; wheat feed opportunity cost.\n\n### K3 — Opponent-aware anti-glut\n\nMutation: K1 + forecast opponent future supply from public farm state.\n\n- front-run likely opponent dumps;\n- avoid entering a resource whose predicted shared glut destroys both margins;\n- exploit complementary demand holes;\n- maintain uncertainty because opponent shed is hidden.\n\nFalsifier: opponent visibility adds noise without enough predictive power.\n\n### K4 — Demand/horizon portfolio\n\nMutation: K1 + explicit NPV by remaining turns and town draw.\n\n- only start an asset if expected payback occurs before wind-down;\n- demand-sensitive product allocation;\n- controlled inventory holding under shed limit;\n- late-game liquidation schedule;\n- land expansion only when marginal NPV beats capital + route overhead.\n\nFalsifier: static meta beats adaptivity because replanning causes execution instability.\n\n### K5 — Hybrid champion candidate\n\nOnly created after K1–K4 evidence.\n\n`K1 executor + best macro mechanisms from K2/K3/K4`\n\nDo not fuse every feature. Each mechanism must survive ablation first.\n\n## Evaluation harness when Kaggle runtime becomes available\n\nFor every candidate:\n\n1. validation episode/self mirror;\n2. fixed seed suite;\n3. both player seats;\n4. built-ins: pass/random/starter;\n5. public reference agents used only as attributed controls;\n6. CURRENT_CHAMPION head-to-head;\n7. paired seeds so both candidates face identical stochastic worlds;\n8. report win/loss/tie, terminal bank, movement %, pass %, task completion, discarded inventory, deaths/weeds/escapes, realized price per product and land/hire utilization.\n\nUse Bradley-Terry or paired win rate for ranking; terminal bank is diagnostic, not the competition's sole ranking signal.\n\n## Promotion gate\n\nA challenger can replace Champion only when all are true:\n\n- zero validation/runtime errors;\n- deterministic invariants pass;\n- >50% paired win rate vs champion with enough episodes to avoid obvious noise;\n- preferably >=60% before online promotion in early generations;\n- no material regression against reference-agent basket;\n- improvement mechanism is explainable or survives ablation;\n- exact submission artifact hash preserved.\n\n## Daily online use of 5 submission budget\n\nDo not spend 5 slots blindly.\n\nSuggested cadence after local harness exists:\n\n1. champion/canary if needed;\n2. strongest challenger;\n3. orthogonal challenger with high information gain;\n4–5 only when earlier evidence justifies them.\n\nBecause only the latest 2 are final-active, order matters. Preserve champion + one challenger at day end.\n\n## Negative-space questions\n\n- Are top agents intentionally retired to reduce replay harvesting?\n- Does a fixed-route meta dominate only because common opponents are copies of it?\n- Can opponent-aware adaptation exploit this monoculture?\n- Is 4th quadrant ever NPV-positive after walking + labor + delayed production?\n- Is fertilizer sale or use better under each current demand state?\n- Can shop draw be inferred early enough to materially reallocate assets?\n- Does expected-value demand planning lose to robust/worst-case planning?\n- Is market timing more valuable than production optimization?\n- Which features improve terminal bank but reduce head-to-head win rate?\n\n## Required next external gate\n\nKaggle account must accept the competition rules before 2026-09-23. This is an R3/terms/account gate and cannot be truthfully fabricated by the agent.\n\nAfter that, use the official `kaggle_environments` runtime as the oracle. Do not promote a hand-written substitute simulator as proof of game performance.\n",[],[175,2335],"00a862ef-d86a-4945-a833-760ba8728205","kaggriculture_joined_research_plan_v1.md","4178cfe2-aa32-11f1-9903-1607ee4eb77e",16,["D",1788726824853],"# Kaggriculture — Joined Research Plan V1\n\nStatus: JOINED_RESEARCHING\nAgent: A4-KAGGRICULTURE\nHuman gate: completed by user (competition joined; identity verification completed)\nDate: 2026-09-06\n\n## Official mechanics confirmed\n\n- 30 days × 24 turns = 720 turns.\n- One farmer action per turn; market orders can also be submitted, up to the configured per-turn limit.\n- Winning is determined by final bank balance; rating movement uses win/loss/tie rather than coin margin.\n- Up to 5 submissions/day; latest 2 submissions are tracked for active/final evaluation.\n- A validation episode runs the submission against a copy of itself before ladder entry.\n- Final evaluation continues after submission lock and culminates in a Bradley–Terry tournament.\n- Observation exposes both farms publicly, shared market state and town state; own shed/seeds/inventories are private.\n- Market inventory and prices are shared; player production can therefore alter the price environment faced by both agents.\n- Submission contract: root `main.py` with top-level `agent(obs)`.\n- No ingress/egress during episode evaluation.\n\n## Current strategic thesis\n\nDo not use an LLM per turn. Treat the problem as a deterministic control system with an adaptive macro layer.\n\nControl stack:\n\n1. macro allocation policy — crop/animal/land/labor/market regime;\n2. persistent task graph — avoid turn-by-turn retargeting;\n3. deterministic path/action executor — minimize walking/thrashing;\n4. market executor — exploit shared inventory/price dynamics;\n5. opponent model — infer likely supply pressure from public farm state;\n6. horizon controller — alter investment/liquidation as remaining turns shrink.\n\n## Neurônio variants\n\nK0 — Deterministic Control\n- stable regression oracle;\n- correct action schema;\n- no unnecessary complexity.\n\nK1 — Labor/Route Optimizer\n- hypothesis: reducing travel and retargeting improves win rate more than adding valuation complexity.\n\nK2 — Livestock/Fertilizer Flywheel\n- hypothesis: animal ROI is understated if product + fertilizer streams are not jointly valued.\n\nK3 — Opponent-Aware Anti-Glut\n- hypothesis: public opponent farm state predicts future shared-market supply enough to change profitable production/sale timing.\n\nK4 — Demand/Horizon Portfolio\n- hypothesis: explicit remaining-horizon value improves investment, allocation and liquidation timing.\n\nK5 — Hybrid Champion Candidate\n- blocked until K2/K3/K4 mechanisms individually survive ablation and regression testing.\n\n## Evidence discipline\n\nPublic replay data is useful for calibration, controls and matchup simulation, but copied action tapes are not a valid competitive strategy. Every external code/data source must satisfy competition licensing and accessibility rules. Public leaderboard/replay performance is historical evidence, not proof that a copied strategy will remain competitive.\n\n## Evaluation protocol\n\nFor every variant:\n\n1. validation/self-play smoke;\n2. deterministic seed set;\n3. mirrored seats where possible;\n4. fixed baseline opponents;\n5. replay-derived sparring opponents only as controls/reference;\n6. record wins/losses/ties, terminal bank, movement rate, pass rate, runtime errors;\n7. mutation/ablation against parent;\n8. promote only when the improvement survives repeated seed/opponent sets.\n\nPrimary selection metric: win rate / Bradley–Terry-relevant outcomes.\nSecondary diagnostics: terminal bank, movement efficiency, runtime stability and opponent-specific failure modes.\n\n## Immediate build order\n\n1. K0 exact-contract baseline.\n2. K1 persistent task graph + nearest/route-aware executor.\n3. Harness that runs paired/mirrored matches across seeds.\n4. Mechanic/reference calibration from official engine/reference data.\n5. K2/K3/K4 independently.\n6. K5 only after evidence-gated fusion.\n7. First Kaggle submission only after local/official-env smoke is clean.\n\n## Economic state\n\nGross revenue: 0\nNet revenue: 0\nTrial terminal: false\nSelection state: INCUBATION\nNo offspring allowed yet.\n",[],[175,2344],"0e19555c-7e69-4662-ba5f-cae3f13cdc4e","calle_feedback_candidate_v1.md","51f5ad64-aa18-11f1-a53a-1607ee4eb77e",["D",1788715685600],"# CALL-E Feedback Candidate V1\n\nStatus: R1_PREPARE / NOT SUBMITTED\nAgent: A2-CALLE\nOpportunity: OP-CALL-E-20260914\nPrepared: 2026-09-06\n\n## Why this candidate exists\n\nThe CALL-E rules allow one feedback submission per entrant and explicitly value actionable bug reports, UI improvements, documentation improvements, and suggested integrations. This artifact contains only findings supported by public documentation/repository evidence. It does **not** claim an authenticated CALL-E runtime test or a real phone call.\n\n## Finding 1 — Supported-region documentation can become stale relative to the maintained region backlog\n\n### Evidence\n\n- The current repository README lists Kenya (`KE`, +254, English, International) in the public “Supported Regions and Languages” table.\n- The current regional/language backlog in GitHub issue #102 records Kenya (`KE`, +254, Swahili) as `Requested`, added 2026-09-01, not `Enabled`.\n- The latest README commit returned for `README.md` was 2026-08-27, while the backlog contains later state changes through 2026-09-03.\n\n### User/agent impact\n\nAn agent that trusts the README as a capability contract can select a recipient/locale combination that the live service may reject. For autonomous callers this is more serious than a normal documentation typo: unsupported-route failures happen near a consequential action boundary and can break planning, retry logic, cost estimation, and user expectations.\n\n### Suggested improvement\n\nCreate one canonical machine-readable capability source and derive public documentation from it. For example:\n\n1. expose a read-only API/MCP capability such as `list_supported_regions` / `get_call_capabilities`;\n2. return region, calling code, supported locales/languages, line type/region, availability state, and effective date;\n3. generate the README table from that source in CI;\n4. fail CI when the generated table and committed documentation diverge;\n5. include a `last_updated`/capability-version value so agent clients can reason about staleness.\n\nThis avoids manually synchronized capability claims across README, issues, Discord and backend behavior.\n\n## Finding 2 — Capability discovery is missing from the 3-tool MCP workflow\n\nThe public MCP contract exposes only:\n\n- `plan_call`\n- `run_call`\n- `get_call_run`\n\nFor an agent, supported-region/language discovery therefore appears to happen only indirectly when `plan_call` rejects or clarifies a request. A read-only capability-discovery tool would let an agent validate a plan before generating a call workflow and would make the system friendlier for unattended automation.\n\n### Suggested tool\n\n`get_call_capabilities({ destination_country?, locale?, channel? })`\n\nPossible output:\n\n```json\n{\n \"capability_version\": \"...\",\n \"destination_country\": \"BR\",\n \"calling_code\": \"+55\",\n \"supported\": true,\n \"languages\": [\"Portuguese\", \"English\"],\n \"locales\": [\"pt-BR\", \"en-...\"],\n \"line_region\": \"Local\",\n \"effective_at\": \"...\"\n}\n```\n\nThis is especially useful for agents that need to choose among regions or languages before `plan_call`.\n\n## Finding 3 — Documentation currentness should be treated as a runtime contract\n\nThe repository is evolving quickly: recent public issues include Windows shell incompatibility (#111), SSE parsing loss (#110), CLI binary collision (#109), missing recovery guidance (#108), and an API-key placeholder mismatch (#107). These are already reported, so they should **not** be duplicated in the feedback survey as new bugs. However, together they show that integration documentation is a fast-moving operational surface.\n\n### Suggested improvement\n\nPublish a small generated compatibility manifest alongside the install guide, for example:\n\n```json\n{\n \"cli_version\": \"...\",\n \"mcp_protocol\": \"...\",\n \"supported_shells\": [\"bash\", \"pwsh\", \"cmd\"],\n \"mcp_tools\": [\"plan_call\", \"run_call\", \"get_call_run\"],\n \"capability_version\": \"...\",\n \"docs_commit\": \"...\"\n}\n```\n\nAgent integrations could verify this before using cached instructions. This would reduce the chance of an agent following stale command syntax or capability assumptions.\n\n## Recommended survey wording\n\n> I found a currentness problem in the supported-region documentation that is particularly important for autonomous agents. The current README lists Kenya (+254) as a supported English/International destination, while the maintained regional backlog in issue #102 records Kenya as Requested (added Sep 1, not Enabled). The README’s latest returned commit is Aug 27, so the two public capability surfaces can disagree as support changes weekly. I recommend making the region/language matrix machine-readable and generating the README from the same source, with a capability version/effective date. It would also be useful to expose a read-only MCP/API capability such as `get_call_capabilities` so agents can validate destination + locale before `plan_call` rather than learning support only through a failed/clarification path. A CI check could fail when generated docs diverge from the canonical capability source. I did not place a real call for this finding; this is a public docs/repository consistency audit.\n\n## Falsification / remaining work before submission\n\n- Re-check the live README immediately before submission.\n- Re-check issue #102 immediately before submission.\n- If a documented capability API/tool already exists elsewhere, revise Finding 2 from “missing” to “hard to discover in the public integration path”.\n- If authenticated CALL-E access is created, run a **plan-only** Brazil/Portuguese capability check first. Do not call `run_call` as a connectivity test.\n- Do not claim a phone call was attempted unless it actually was.\n\n## Evidence state\n\nFinding 1: STRONG / public contradiction observed\nFinding 2: PARTIAL / absence confirmed in public 3-tool MCP contract, backend-wide absence not proven\nFinding 3: INFERENCE / synthesis from multiple already-reported integration issues\n\n## Economic state\n\nGross revenue: 0\nCost recorded by this agent: 0\nNet revenue: 0\nStatus: PREPARED_NOT_SUBMITTED\n",[],[175,2352],"69d33785-05f9-4837-943b-cda43a5b3ea1","KAGGRICULTURE_REAL_WORLD_CURRICULUM_V1.md","385d7470-aa3d-11f1-8d8a-1607ee4eb77e",47,["D",1788731534043],"# Kaggriculture Real-World Curriculum V1\n\nStatus: ACTIVE RESEARCH DOCTRINE\nDate: 2026-09-06\nPurpose: turn Kaggriculture agents into domain experts rather than blind parameter searchers.\n\n## Principle\n\nResearch first, test second. External knowledge should generate mechanisms, priors, strategy families and falsifiable hypotheses. Simulation remains the validation laboratory because transferable theory can fail under Kaggriculture's exact rules.\n\n## Track A — Autonomous Competitive Agents\n\nCore disciplines:\n- Multi-agent reinforcement learning (MARL)\n- Stochastic / Markov games\n- Opponent modeling / Theory of Mind\n- Competitive coevolution\n- Population-based training (PBT)\n- Policy Space Response Oracles (PSRO)\n- League training / PFSP matchmaking\n- Self-play and historical snapshot pools\n- Evolutionary strategies / CMA-ES / genetic algorithms\n- Curriculum learning\n- Meta-learning / transfer learning\n- Imitation learning / behavioral cloning\n- Hierarchical reinforcement learning\n- Model-based RL and planning\n- Monte Carlo Tree Search\n- Counterfactual regret minimization / regret learning\n- Online learning / no-regret dynamics\n- Exploitability / Nash / best response / empirical game-theoretic analysis\n- Robust RL / distributionally robust Markov games\n- Operations research for task allocation and routing\n- Multi-agent path finding / scheduling\n- Hybrid learned-policy + deterministic executor architectures\n- Active learning / experiment design / Bayesian optimization\n- Ablation / causal attribution / mutation testing\n\nQuestions every agent-research branch must answer:\n1. What is the opponent set used for training and why?\n2. Does the method overfit one opponent or strategy family?\n3. Is fitness absolute or relative to an evolving population?\n4. Does the population preserve diversity?\n5. What triggers branch, mutation, reset and pruning?\n6. Can the agent infer opponent type online?\n7. What is the best response to each opponent family?\n8. How is robustness measured under unseen shops/seeds/seats/opponents?\n9. Which decisions should be learned/adaptive and which deterministic?\n10. What would falsify the claimed mechanism?\n\nHigh-value implementation families:\n- PSRO-like strategy population\n- AlphaStar-style league training\n- Historical champion pool\n- Exploiter agents specialized in finding champion weaknesses\n- Opponent classifier + conditional policy router\n- Hierarchical macro policy + deterministic operations executor\n- Population-based parameter evolution\n- Curriculum from weak reference agents to meta-line opponents\n- Robust selection on quantiles/CVaR/win probability, not average bank alone\n\n## Track B — Real-World Farm / Economics / Business Strategy\n\nCore disciplines:\n- Agricultural economics\n- Whole-farm planning\n- Farm management\n- Agribusiness\n- Production economics / theory of the firm\n- Operations management\n- Resource allocation / linear and mixed-integer programming\n- Dynamic programming / approximate DP\n- Stochastic optimization\n- Inventory theory / storage economics\n- Working capital / liquidity management\n- Cash conversion cycle\n- Marginal productivity of labor / land / capital\n- Diminishing returns\n- Opportunity cost / shadow price / bottleneck value\n- Commodity market microstructure\n- Supply and demand\n- Price elasticity\n- Scarcity / glut / endogenous price impact\n- Intertemporal selling / hold-vs-sell thresholds\n- Convenience yield / real options of inventory\n- Dynamic pricing under competition\n- Industrial organization / duopoly / oligopoly\n- Game theory / repeated games / differential games\n- Competitor response / strategic substitutes and complements\n- Capacity planning\n- Perishability / terminal salvage / liquidation\n- Portfolio allocation / diversification\n- Mean-variance / downside risk / CVaR\n- Risk management\n- Hedging concepts (transfer only where environment has analogous mechanism)\n- Interest / inflation / cost of capital (knowledge layer; not directly represented in current game)\n- Investment timing / NPV / real options\n- Supply chain / logistics / storage capacity / transport\n- Demand forecasting / Bayesian updating\n- Price discrimination / reservation price / market timing\n- Competitive strategy / positioning / resource-based view\n- OODA loop / wargaming / red teaming / adversarial planning\n\n## Kaggriculture ↔ Real World Mapping\n\n- farm money -> working capital / liquidity\n- workers -> labor capacity and marginal productivity\n- land -> capacity investment / capital expenditure\n- crop mix -> production portfolio\n- livestock -> capital-intensive productive assets with delayed payback\n- feed reserve -> operating liquidity / safety stock\n- shed -> inventory/storage capacity\n- crop/animal lifecycle -> lead time / biological production cycle\n- market inventory -> aggregate supply state\n- market price curve -> endogenous price elasticity\n- shops -> stochastic demand regimes / local demand shocks\n- market order sequence -> execution ordering + intraperiod financing\n- shared market -> duopoly / strategic price impact\n- opponent farm visibility -> competitor intelligence\n- unknown private inventory -> partial information / belief state\n- weeds -> operational shocks\n- 720-turn horizon -> finite planning horizon\n- final unsold stock -> zero terminal salvage / liquidation risk\n- buy land / animals -> irreversible investment / real option\n- choice to sell now or hold -> inventory real option / threshold policy\n\n## Things NOT directly represented — avoid fake realism\n\n- debt / loans / leverage\n- interest expense\n- inflation\n- taxation\n- insurance\n- futures/options contracts\n- regulation\n- weather and climate shocks beyond weeds\n- supplier defaults\n- reputation / relationships / contract enforcement\n- product quality grades\n- transportation distance/cost\n- multiple buyers with negotiated contracts\n- macroeconomic policy\n\nStudy these as transferable concepts, but never pretend the simulator already models them.\n\n## Research-to-Action Pipeline\n\nSOURCE -> CLAIM -> TRANSFERABLE MECHANISM -> KAGGRICULTURE ANALOGY -> HYPOTHESIS -> MINIMAL IMPLEMENTATION -> CONTROL -> SIMULATION -> FALSIFICATION -> PROMOTE/REJECT -> LESSON\n\nBlind parameter sweeps are prohibited when a researchable mechanism exists. Parameter search is allowed only after mechanism definition and should estimate thresholds/robust regions rather than replace reasoning.\n\n## Recommended Source Router\n\n- TinyFish research_paper: papers/theses/dissertations and targeted recent academic search\n- Exa: semantic discovery of theses, projects, labs, case studies, technical blogs, repositories and unusual adjacent fields\n- Firecrawl developer_search: GitHub README/issues/PRs/docs and implementation patterns\n- Firecrawl research_search_papers: broad paper discovery / alternate formulations\n- GitHub connector: full code/provenance/commit history once a relevant implementation is found\n- Reddit via TinyFish/Exa: practitioner failures, hidden assumptions, implementation pain, community consensus; never authority by itself\n- Consensus/Elicit/Scite: evidence synthesis, paper tables, supporting/contrasting citations\n- Wolfram when operational: symbolic/numeric optimization, sensitivity, probability, robust decision models, independent calculation oracle\n\n## Priority Research Waves\n\nWave A1: PSRO + league training + exploiters + opponent population management\nWave A2: opponent modeling + Bayesian strategy classification + best-response routing\nWave A3: hierarchical macro policy + deterministic task executor + routing/scheduling OR\nWave A4: competitive coevolution + diversity preservation + curriculum + transfer\nWave A5: robust selection / uncertainty / win-probability optimization\n\nWave B1: whole-farm optimization / MIP / dynamic programming\nWave B2: inventory/storage/hold-sell threshold policies and real options\nWave B3: working capital/liquidity/feed reserve/capital allocation\nWave B4: industrial organization / dynamic pricing / endogenous supply / competitor response\nWave B5: farm production economics / marginal returns / capacity bottlenecks\nWave B6: risk, CVaR, portfolio diversification and finite-horizon liquidation\n\n## Immediate research-derived hypotheses\n\nH-A1: opponent sequence and population composition materially affect learned policy quality; train against a curated evolving league, not one baseline.\nH-A2: preserve old champions as frozen opponents to prevent forgetting and cyclic collapse.\nH-A3: specialized exploiters should search for champion weaknesses and feed discovered failure modes back into the champion lineage.\nH-A4: online opponent classification should route among strategy families rather than force one universal policy.\nH-A5: macro adaptation should occur only at information-changing events; low-level logistics should remain deterministic/persistent.\n\nH-B1: optimal production is a joint production+inventory+liquidity decision, not yield maximization.\nH-B2: feed reserve has option value because it preserves the ability to operate through adverse price/supply states.\nH-B3: sell/hold decisions should use state-dependent thresholds rather than fixed reservation prices.\nH-B4: extra land/labor has a shadow value that becomes zero or negative after downstream demand/storage/logistics bottlenecks bind.\nH-B5: strategy should optimize probability of defeating the opponent, not expected own bank in isolation.\nH-B6: shop-world classification should alter the crop/livestock portfolio and liquidation timing.\n\n## Promotion rule\n\nNo concept becomes canonical merely because it appears in a paper, thesis, GitHub repo or successful public agent. Promotion requires a clear mechanism, compatible assumptions, current game mapping, an explicit counterfactual/ablation, and local evidence against a suitable control.\n",[],[175,2359],"05af4f80-7c0e-4dba-9cc2-f5000cae2a0d","HUMAN_GATE_KAGGRICULTURE_20260906.md","0e5a86d4-aa30-11f1-a6db-1607ee4eb77e",15,["D",1788725880082],"# HUMAN GATE — Kaggriculture\n\nStatus: READY_FOR_USER_ACTION\nChecked: 2026-09-06\nOpportunity: OP-KAGGRICULTURE-202609\nAgent: A4-KAGGRICULTURE\n\n## Proven current primary-source facts\n\n- Official competition: https://www.kaggle.com/competitions/kaggriculture\n- Official rules: https://www.kaggle.com/competitions/kaggriculture/rules\n- Sponsor: Google LLC\n- Total prizes: USD 50,000\n- 10 prizes of USD 5,000 each\n- Entry Deadline: 2026-09-23\n- Team Merger Deadline: 2026-09-23\n- Final Submission Deadline: 2026-09-30\n- Maximum submissions: 5/day\n- Up to 2 final submissions may be selected for judging\n- Rules explicitly permit AMLT/automated machine learning tools subject to licensing/rule compliance.\n- General eligibility is worldwide except listed sanctioned/restricted territories; Brazil is not among the listed exclusions in the current rules.\n- A Kaggle account and acceptance of the competition rules are required before the Entry Deadline.\n\n## Human action required\n\n1. Sign in to the user's real Kaggle account.\n2. Open the Kaggriculture competition page.\n3. Click `Join Competition`.\n4. Read and accept the official competition rules using the user's own identity/account.\n5. Do not create multiple Kaggle accounts; the rules prohibit multiple accounts.\n6. Stop after the competition shows the account as joined. Do not submit an agent yet.\n\n## Why this is a human gate\n\nAccepting binding competition rules is an identity/legal action (R5/HUMAN_GATE). The agent may prepare strategies, code, tests, ablations and submissions, but must not impersonate the user or accept binding terms on the user's behalf.\n\n## After the gate\n\nExpected next lifecycle state: `JOINED_RESEARCHING`.\nThe autonomous work resumes with K0→K5 strategy experiments, simulation/replay analysis, deterministic executor optimization, anti-glut/opponent-aware branches, falsification and champion selection.\n\n## Important correction from previous plan\n\nCALL-E/Devpost must NOT be used for this user's paid-entry experiment: the current official CALL-E rules explicitly list Brazil among excluded resident territories. That opportunity is now `INELIGIBLE` and the prior recommendation to register was withdrawn and recorded as negative knowledge.\n",[],[175,2367],"2bc213e2-cb21-4135-b42b-485c61c358d2","KAG_EVIDENCE_MASTER.md","ceb010de-acc9-11f1-b24f-1607ee4eb77e","5cf725d4-aa96-11f1-949e-1607ee4eb77e",843,["D",1789011818090],"# KAGGRICULTURE — EVIDENCE MASTER\n\n**Status:** CANONICAL_SINGLE_HUMAN_READABLE_EVIDENCE_LOG \n**Atualizado:** 2026-09-10 UTC \n**Autoridade estratégica:** `chat-native-maestro@economic-v0.6` \n**Compute:** `kaggriculture-autolab@main` \n**Baseline/champion preservado:** MIX11 \n**Fronteira principal:** Broker Bea \n**Upper tier:** Lena → Silas → Closer Cleo \n**Controle intermediário:** Rancher Rita \n**Regressão barata:** Rosa / Hana / Mateo\n\n> Este arquivo substitui os antigos logs cronológicos de hourly/audits/experiment Markdown como **registro humano canônico de evidências**. Evidência bruta/reprodutível continua no SQLite/Blob do sidecar e no código de testes; nada é considerado apagado cientificamente apenas porque os logs Markdown redundantes foram removidos.\n\n---\n\n## 0. REGRAS DE INTERPRETAÇÃO\n\n- `FOUND != READ != UNDERSTOOD != APPLIED != TESTED`.\n- `PASS != VALIDATOR_COMPETENCE`.\n- `W/L` é métrica primária; margem é diagnóstico.\n- `SAME_SEED != IDENTICAL_PRESTATE` depois que histórias divergem.\n- `TREATMENT_REQUESTED != TREATMENT_APPLIED`.\n- `PUBLIC_STATE_CHANGE != OPPONENT_REACTION_TO_US`.\n- `MARKET_MOVE_AFTER_OUR_ACTION != CAUSAL_RESPONSE`.\n- `PRIVATE_STATE_ESTIMATE != OBSERVED_PRIVATE_STATE`.\n- `RAW OPPONENT ACTION/TRACE = OFFLINE TEACHER ONLY / NONDEPLOYABLE`.\n- `CODE_PATH_EXISTS != ACTIVE_MECHANISM`.\n- `HISTORICAL_RESULT != CURRENT_RERUN`.\n- Um mecanismo causal local não vira estratégia sem materialidade econômica + heldout + regressão.\n- Uma variante só pode substituir MIX11 após superar gates de Broker e regressões sem reabrir conhecimento negativo por um único batch favorável.\n\n---\n\n# 1. ESTADO COMPETITIVO ACUMULADO\n\n## 1.1 Nossa família contra Broker — SQLite persistido\n\n| Variante | Episódios | W-L | Margem média por job | Estado |\n|---|---:|---:|---:|---|\n| MIX11 | 200 | 0-200 | -105603.15 | BASELINE, estruturalmente abaixo de Broker |\n| M60_FERT | 98 | 0-98 | -84017.69 | margem melhor, zero travessia |\n| M40_FERT | 22 | 0-22 | -83054.44 | zero travessia |\n| M50_FERT | 8 | 0-8 | -77376.25 | diagnóstico apenas |\n| DEPLOY10 | 30 | 0-30 | -82624.13 | falsificado como solução |\n| ADAPT_V1 | 8 | 0-8 | -79355.75 | rejeita router ingênuo, não adaptação em geral |\n| DP_B | 8 | 0-8 | -77288.75 | falsificado |\n| DP_H | 8 | 0-8 | -80739.00 | falsificado |\n| DP_W | 8 | 0-8 | -81825.25 | falsificado |\n| DP_COND | 8 | 0-8 | -86134.50 | falsificado |\n| DP_H_PULSE | 8 | 0-8 | -93275.75 | falsificado |\n| DP_B_PULSE | 8 | 0-8 | -92899.38 | falsificado |\n| DP_COND_PULSE | 8 | 0-8 | -90526.63 | falsificado |\n| M60_FERT_JIT | 16 | 0-16 | -93878.94 | falsificado |\n| MIX11_LIQ21 | 16 | 0-16 | -111885.25 | falsificado |\n| MIX11_LIQ24 | 16 | 0-16 | -111887.13 | falsificado |\n| HERD4 | 16 | 0-16 | -84422.69 | não transfere para Broker |\n| LAND6 | 8 | 0-8 | -85553.25 | não transfere |\n| PASTURE_RAMP | 32 | 0-32 | -108106.28 | HELDOUT_FALSIFIED |\n| LAND_AFTER6_CLEAN | 8 | 0-8 | -101961.63 | falsificado |\n| PORTFOLIO_D6_244 | 8 | 0-8 | -121734.13 | falsificado |\n| EARLY_MELON40 | 8 | 0-8 | -113501.00 | falsificado |\n| EARLY_HERD3 | 8 | 0-8 | -105514.13 | falsificado |\n| EARLY_HERD4 | 16 | 0-16 | -102676.88 | falsificado |\n| EARLY_HERD5 | 8 | 0-8 | -103231.38 | falsificado |\n| EARLY_BUILD4 | 16 | 0-16 | -104394.63 | falsificado |\n| EARLY_HERD_BUILD4 | 8 | 0-8 | -106726.50 | falsificado |\n| EARLY_HERD_BUILD8 | 8 | 0-8 | -110975.63 | falsificado |\n| STAGED_STRAW_D6_D12 | 12 | 0-12 | -112102.75 | harmful |\n| CAP2_D6 | 12 | 0-12 | -112072.67 | fresh-heldout falsified |\n| CAP2_D6_STAGED | 6 | 0-6 | -113869.83 | harmful |\n\n**Conclusão confirmada:** mais de uma centena de perturbações/episódios e muitas famílias diferentes continuam 0-win contra Broker. `MARGIN_PROXIMITY != IMMINENT_WIN`. Não é racional continuar uma busca local cega ao redor de MIX11.\n\n## 1.2 Nossa família nos outros adversários\n\n- MIX11 × Cleo: **0-84**, margem média por job ~**-105183.78**.\n- MIX11 × canonical Lena: **0-62**, ~**-103528.54**.\n- MIX11 × canonical Silas: **0-54**, ~**-99369.30**.\n- MIX11 × Rita: **121-37 / 158 episódios**, ~**+4884.09**.\n- MIX11 mantém vitórias perfeitas nos conjuntos persistidos recentes contra Rosa/Hana/Mateo (canonical: Rosa 14-0, Hana 14-0, Mateo 16-0 no conjunto acumulado consultado).\n\n**Inferência forte:** existe heterogeneidade de matchup. `GOOD_VS_RITA != TRANSFERS_TO_BROKER` e muito menos para o upper tier.\n\n## 1.3 Ladder público / teacher\n\n| Confronto | W-L acumulado |\n|---|---:|\n| Cleo vs Broker | **41-5 / 46** |\n| Cleo vs canonical Silas | **11-3 / 14** |\n| Cleo vs canonical Lena | **4-0** |\n| Cleo vs Rita | **4-0** |\n| canonical Silas vs canonical Lena | **26-2 / 28** |\n| canonical Lena vs Broker | **25-3 / 28** |\n| Broker vs Rita | **12-0** |\n\nAliases históricos permaneceram separados por proveniência (ex.: Lena antiga 98-6 Broker; Cleo 58-6 Silas antigo). A concordância entre épocas reforça, mas não autoriza colapsar identidades/versionamentos.\n\n**Conclusão:** a família superior observável é `Rita << Broker < Lena < Silas < Cleo`. Cleo é o teacher prioritário para reconstrução operacional, não um recurso online.\n\n---\n\n# 2. O QUE APRENDEMOS SOBRE A FRONTEIRA BROKER\n\n## 2.1 Onde o MIX11 perde\n\nMúltiplos heldouts reproduziram a forma temporal: MIX11 pode liderar até aproximadamente D6/D9, depois a diferença degrada. Em uma decomposição representativa os maiores crescimentos negativos do gap ocorreram aproximadamente em D12-15, D18-21, D21-24, D24-27 e D27-30, com o fechamento entre os maiores colapsos. Isto aponta para throughput/conversão/execução composta, não para um único parâmetro inicial.\n\n## 2.2 Cash-awareness 199→200\n\nFixed-prestate positivo:\n- Broker/Lena/Silas: resposta de ação replicada em contextos validados 199→200.\n- Rita/Rosa/Hana/Mateo: controles negativos nos contextos testados.\n- Cleo: reação análoga aparece em contextos tardios/slot competition, não genericamente nos mesmos landmarks.\n\nMaterialidade econômica do mecanismo Broker ficou baixa/contextual: em uma rodada 18/133 estados comparáveis alteraram custo controlado; máximo observado ~4 moedas; média ~0.376 moeda. Em outra auditoria natural o nosso cash não chegava ao regime de 200 (mínimo observado ~1301). Portanto:\n`AWARENESS_REAL != PRIMARY_BOUNDARY_CAUSE`.\n\n## 2.3 Land / labor / hires_today\n\nFixed-prestate isolado em landmarks:\n- land visibility: 0 resposta nos contextos amostrados nos 8 adversários;\n- `hires_today`: 0 resposta nos contextos amostrados;\n- bundle de labor: também nulo, mas classificado como bundle/confundido.\n\nIsto não prova ausência universal. Prova apenas que perturbar estes campos isoladamente nos landmarks amostrados não gera resposta de ação.\n\n## 2.4 Shared-market snapshot\n\nIntervenções em preço/inventory de itens, shop multiplicity e cash amplo não explicaram as grandes assinaturas temporais de Broker/Cleo nos contextos testados. Rita mostrou sensibilidade causal replicada a WHEAT price e Mateo a WHEAT; Mateo FERTILIZER foi apenas candidato. Para Broker/Cleo, o lead migrou de snapshot instantâneo para trajetória/execução/throughput.\n\n## 2.5 Legal timing signatures\n\nCom self-action filtering, as assinaturas observacionais reproduziram padrões diferentes (Cleo forte lag0; Broker lag0 + eco lag2; Lena/Silas caudas distintas; Rita perto de zero) e sobreviveram qualitativamente a mudanças de política MIX11→M40/M60/DEPLOY em várias rodadas. Porém fixed-prestate mostrou que estes lifts não equivalem a direct cash-awareness. Portanto:\n`LEGAL_LAG_SIGNATURE = PROFILING_SIGNAL / NOT_CAUSAL_ELASTICITY`.\n\n---\n\n# 3. MECANISMOS CAUSAIS / SUCESSOS REAIS\n\n## 3.1 Sell-first funding\n\nFixed-prestate e heldout confirmaram causalmente que, sob cash baixo, executar SELL antes de BUY/HIRE pode permitir compras que falhariam se os mesmos comandos fossem reordenados. O efeito desaparece em cash natural suficiente. Entretanto auditoria natural mostrou que MIX11 já fazia SELL-first em **221/221** filas mistas contra Broker e **210/210** contra Cleo, incluindo oportunidades com cash <2000.\n\n**Estado:** mecanismo verdadeiro e útil como invariant/regression; `SELL_FIRST_AS_MISSING_CAPABILITY = FALSIFIED`.\n\n## 3.2 Clone/front-run history\n\nFixed-history controlado confirmou que Broker/Lena/Silas podem reagir a duas exposições históricas específicas mesmo com terminal public snapshot idêntico, produzindo diferenças premium-SELL e consequências imediatas. Weak controls ficaram nulos. Contudo auditoria natural de MIX11 encontrou distância mínima muito fora do regime de suporte em **720/720** checkpoint rows; confiança upper-bound ficou 0.\n\n**Estado:** `CAUSAL_CAPABILITY_CONFIRMED / NATURAL_SUPPORT_ABSENT`. Não é alvo de exploit prioritário.\n\n## 3.3 Supply-reserve path\n\nSource/runtime audit encontrou `_SUPPLY_DRIVER` populado, mas `_RESERVE={}` na configuração atual de Broker/Lena/Silas/Cleo. Logo o caminho `_opponent_scale → reserve_price` existe em código, porém está inativo na configuração atual.\n\n**Estado:** `CODE_PATH_PRESENT / CURRENT_CONFIG_INACTIVE`.\n\n---\n\n# 4. CAPACIDADE / PORTFÓLIO / SERVIÇO — FALSIFICAÇÕES IMPORTANTES\n\n## 4.1 Top-family trajectory\n\nTrajetórias públicas do upper tier reproduziram capacidade/composição muito maior que MIX11 nos primeiros dias (ex.: por D9 ~2 land/~30 crops/~11 animals e D12 ~3 land/~39 crops/~14 animals em amostras superiores). Mas copiar partes isoladas falhou.\n\n## 4.2 Labor D9-D15\n\n`LABOR10_D9_D15` em fresh heldout permaneceu 0-win Broker e efeito variou por seed; outro screen LABOR9/LABOR10 também piorou 8/8 em batch posterior. Fechado para sweep adjacente.\n\n## 4.3 Staged crop portfolio\n\n`STAGED_STRAW_D6_D12` foi harmful:\n- Broker: controle ~-94384 vs staged ~-113295, delta ~-18911;\n- Rita: controle ~+8989 vs staged ~-586, delta ~-9574.\n\n## 4.4 Capacity×portfolio\n\nFactorial B55: interação ~**-20673.67**; capacidade não resgatou staged portfolio. CAP2_D6 teve um falso sinal favorável de um batch e depois reverteu no heldout Broker; Rita depois caiu de 6-0 para 0-6. `CAP2_D6 = DO_NOT_REOPEN`.\n\n## 4.5 Routing/feed\n\nPersistent routing treatment fired, permaneceu 0-8 Broker e piorou stress de feed pretendido. Não promover.\n\n## 4.6 Task-service backlog\n\nBacklog/CARE/water/harvest existe tanto em loss-context Broker quanto em win-context Rita. Portanto não é causa específica da fronteira por si só.\n\n---\n\n# 5. WOOL — LINHA COMPLETA E POR QUE FOI DESPRIORIZADA\n\n1. Primeiro apareceu grande diferença contextual de WOOL entre Broker e Rita.\n2. Upper family formou cluster forte de footprint, mas o sinal de preço/inventory versus Rita virou por mundo; `STABLE_PRICE_SUPPRESSION` foi falsificado.\n3. Fixed-prestate WOOL price/inventory reaction foi nulo nos landmarks com cash-positive-control funcionando.\n4. Primeiros mediadores de `yieldSum` deram falso-zero devido a bug de consumer path; não foi tratado como ausência do fenômeno.\n5. Parity raw-public↔aggregate passou **4314/4314** por target e mutation control capturou corrupção **4314/4314**.\n6. Demanda WOOL foi corrigida: YARN_STORE consome ×2; ledger exato passou **3882/3882** e mutation control **3882/3882**.\n7. Modelo de sell próprio foi parity-validado, inclusive insuficiência de inventory e corte de 10 slots.\n8. Com conservação cada vez mais exata, associação residual upper-family caiu para ordem de ΔR² ~0.00030–0.00055, Rita ~0.000003, com forte heterogeneidade de mundo.\n9. Fixed-prestate forçado demonstrou causalmente que supply do oponente entra 1:1 no shared inventory e tem impacto de preço local não linear, mas isto prova mecânica do engine, não que upper agents vendem WOOL naturalmente de forma material.\n10. Census oracle esparso não encontrou supply WOOL nos checkpoints upper-family; insuficiente para declarar zero turn-level, mas contraevidência para uma história simples contínua.\n\n**Estado:** `WOOL = MECHANICALLY UNDERSTOOD / PRIMARY_BOUNDARY_MECHANISM DEPRIORITIZED / NO STUDENT MUTATION`.\n\n---\n\n# 6. MILK — DISCRIMINANTE ATUAL\n\n## 6.1 Fresh rank\n\nD18-D27, unchanged MIX11, fresh mirrored `[112001,112013,112027,112039]`:\n- Broker 0-8 / -94060.875; Rita 4-4 / +3017.\n- MILK executed qty: **632 vs 770**, delta **-138**.\n- realized value proxy: **103816 vs 120737**, delta **-16921**.\n- isto ocorreu apesar de preço MILK no Broker-context ser ~+7.46 maior.\n\n## 6.2 Fresh decomposition independente\n\n`[112101,112113,112127,112139]`:\n- Broker 0-8 / -114639.375; Rita 6-2 / +3983.875.\n- Broker-Rita por game: MILK sell -24.25; value -1025.25; private MILK shed -1.79; own cows -0.834; cow yieldSum -1.322; yield transitions -5.75; mean MILK price **+25.62**; unfed stress 0 nos dois.\n\n**Estado:** `MILK_OUTPUT_AVAILABILITY_PATH = REPLICATED_DIRECTIONAL_DIAGNOSTIC / WORLD_HETEROGENEOUS / NOT_CAUSAL`.\n\nA evidência enfraquece `LOW_PRICE_CAUSES_MILK_DEFICIT`; aponta para disponibilidade/produção/conversão, mas não autoriza aumentar herd cegamente.\n\n---\n\n# 7. NOVA LINHA V26 — TESTES INICIADOS 2026-09-10\n\n## 7.1 Objetivo\n\nTestar componentes derivados do teacher Cleo e do sinal MILK sem tocar no champion. Branch experimental inicial: `kaggriculture-autolab@meta-cleo-v1`; capacidade de laboratório V26 foi depois colocada em main com default MIX11 mecanicamente preservado no smoke de seed 113001.\n\nCore de laboratório: `KAG_AUTOLAB_CORE_V26_META_CLEO_DISTILL`.\n\n## 7.2 CLEO_SLOT_V1\n\nHeurística experimental: preservar non-SELL ops antes de SELL em cash normal; em cash <2500, até dois SELLs de maior valor primeiro para funding; máximo 10 market ops.\n\nSmoke seed113001:\n- Broker 0-2, -103868 vs MIX11 -103791.\n- Rita 2-0, +22236 vs MIX11 +20481.5.\n\nFresh gate `[113011,113023,113039,113051]`:\n- Broker 0-8, -120094.25 vs MIX11 -121709.25 (margin +1615 apenas).\n- Rita **2-6, -3051.375** vs MIX11 **6-2, +3735.875**.\n\n**Estado:** `CLEO_SLOT_V1 = SEVERE_RITA_REGRESSION / DO_NOT_PROMOTE`. O scheduler isolado não é a capacidade secreta do Cleo.\n\n## 7.3 MILK_BALANCE_V1\n\nCow-bias tardio experimental, sem intenção de aumentar herd total.\n\nFresh gate `[113011,113023,113039,113051]`:\n- Broker 0-8, -121709.25, igual ao MIX11 neste batch (tratamento efetivamente sem impacto suficiente).\n- Rita 6-2, +1679.875 vs +3735.875 baseline.\n\n**Estado:** `NO BROKER IMPROVEMENT / MARGIN REGRESSION RITA / DO_NOT PROMOTE`.\n\n## 7.4 META_CLEO_DISTILL_V1 inicial\n\nCombinação do slot scheduler + milk-balance V1.\nFresh gate:\n- Broker 0-8, -120094.25.\n- Rita **0-8, -4076.5**.\n\n**Estado:** `V1 FALSIFIED / SEVERE REGRESSION / NAME RETAINED AS PROGRAM, NOT AS THIS IMPLEMENTATION`.\n\nA conclusão estratégica de usar Cleo como teacher continua válida; a implementação V1 ingênua não.\n\n## 7.5 MILK_MIX_D6_V2\n\nNova tentativa mais restrita: mudar composição desejada para ~65% COW a partir de D6 **sem aumentar o target total de animais**, mantendo o restante da política.\n\nFresh `[113071,113083,113099,113111]`, 8 mirrored por adversário:\n- Broker: MIX11 0-8/-106648.625; V2 0-8/-106949.625 (**delta -301**, pior terminal). D24→D27 gap deterioration melhora diagnosticamente de -24371.75 para -22787.5, mas isto não compensa W/L/margem terminal.\n- Rita: ambos 6-2; margem cai +2044.375 → +529.75.\n- Rosa: ambos 8-0; margem 41326.375 → 40131.875.\n- Hana: ambos 8-0; 29848.25 → 25708.625.\n- Mateo: ambos 8-0; 30031.875 → 27036.75.\n\n**Estado:** `W/L REGRESSION PRESERVED IN THIS BATCH / BROKER 0-WIN / MARGINS WORSE / NOT PROMOTED`.\n\n---\n\n# 8. NOVO LOOP AUTOMÁTICO DE 15 MINUTOS\n\nArquivo ativo em `kaggriculture-autolab@main`: `meta_cleo_15min.ts`.\nConfiguração: delay **900000 ms = 15 minutos**.\n\nCada ciclo usa seed fresco determinístico e executa, no mesmo seed:\n- MIX11 × Broker;\n- candidate atual `MILK_MIX_D6_V2` × Broker;\n- MIX11 × Rita;\n- candidate × Rita;\n- Cleo × Broker como teacher/control upper-bound.\n\nO ciclo grava **uma única linha compacta** em SQLite `kag_meta_cleo_cycles`; não cria um Markdown por rodada. Isto substitui o antigo padrão de centenas de checkpoint/manifest/diff files.\n\nCircuit breaker: depois de 12 ciclos, se a candidata continuar com 0 Broker wins e média `delta_broker <= 0`, a candidata é desativada automaticamente; o loop não continua gastando compute cegamente nela.\n\n### Ciclo 1 — seed 114018\n- MIX11 Broker: 0-2 / -101820.\n- MILK_MIX_D6_V2 Broker: 0-2 / -104050. **delta -2230**.\n- MIX11 Rita: 2-0 / +1443.\n- Candidate Rita: 2-0 / +867.5. **delta -575.5**.\n- Cleo Broker: **2-0 / +2572**.\n\n### Ciclo 2 — seed 114035\n- MIX11 Broker: 0-2 / -78326.5.\n- Candidate Broker: 0-2 / -85554.5. **delta -7228**.\n- MIX11 Rita: 2-0 / +8136.5.\n- Candidate Rita: 2-0 / +1822.5. **delta -6314**.\n- Cleo Broker: **2-0 / +2418.5**.\n\n**Estado atual da candidate no loop:** 0-4 Broker em 2 ciclos; ambos os paired deltas Broker negativos. Isso é evidência inicial contra V2, ainda abaixo do circuit-breaker preregistrado de 12 ciclos.\n\n---\n\n# 9. CONHECIMENTO NEGATIVO CANÔNICO\n\nManter fechados, salvo **novo mecanismo + preregistered discriminant + fresh paired heldouts + regressão**:\n\n- fertilizer/static M40/M50/M60 como solução universal Broker;\n- static/persistent H/W/B/COND;\n- bounded D9-D12 deployment pulses;\n- generic D21/D24 liquidation;\n- unconditional pasture-JIT;\n- PASTURE_RAMP;\n- LAND_AFTER6_CLEAN;\n- PORTFOLIO_D6_244;\n- EARLY_MELON40;\n- EARLY_HERD3/4/5;\n- EARLY_BUILD4;\n- early herd+build;\n- LABOR9/LABOR10 D9-D15;\n- STAGED_STRAW_D6_D12;\n- CAP2_D6 e CAP2_D6_STAGED;\n- persistent routing como feed-continuity fix;\n- naive WOOL exploit;\n- direct cash200 exploit como boundary solution;\n- clone/front-run como natural MIX11 exploit;\n- CLEO_SLOT_V1 isolado;\n- META_CLEO_DISTILL_V1 primeira implementação.\n\n---\n\n# 10. PLANO DE AÇÃO PROFUNDO — PROGRAMA META-CLEO\n\n## Fase A — parar de copiar snapshots e reconstruir sequência operacional\n\n**Objetivo:** descobrir *quais decisões executadas* do Cleo geram valor e em que ordem, sem transformar raw teacher action em feature online.\n\nA1. Teacher decomposition por fase D0-6 / D6-12 / D12-18 / D18-24 / D24-30. Para cada fase, medir cadeia `cash → buy/hire → crop/animal lifecycle → collection → shed → sell → terminal cash`.\n\nA2. Separar capacidade em blocos independentes: movement/routing, task priority, market slot scheduler, funding, crop plan, animal composition, collection/conversion e terminal liquidation.\n\nA3. Para cada bloco, usar exact interpreter accounting e treatment detector. Nada de inferir valor de uma diferença de snapshot.\n\nA4. Criar `teacher parity signatures` legais: efeito observável da sequência, não raw opponent action como feature do student.\n\n## Fase B — mecanismo atual de maior EVSI: MILK conservation\n\nB1. Construir ledger MILK análogo ao WOOL: demanda determinística town/shop, own executed MILK sell, cow lifecycle legal e shared inventory delta.\n\nB2. Validar paridade do ledger contra engine com mutation control antes de qualquer causal conclusão.\n\nB3. Condicionar por mundo/regime; não agregar mundos heterogêneos em um único efeito.\n\nB4. Se sobrar residual material controlável, somente então fazer um fixed-prestate fork de **um estágio por vez**: cow acquisition, care/collection, conversion timing ou sell execution.\n\nB5. Se nenhum residual material sobreviver, fechar MILK sem novo herd sweep e avançar para próximo produto/throughput rank.\n\n## Fase C — distilação funcional do Cleo\n\nC1. Em vez de `COPY CLEO`, calcular delta funcional entre Cleo e Broker/Lena/Silas em same-world paired public crossplay.\n\nC2. Encontrar smallest sufficient mechanism: qual transformação do nosso planner aproxima o *efeito* do Cleo sem copiar seu estado interno.\n\nC3. Implementar candidatos em ordem causal e ablation:\n- scheduler somente onde slot pressure realmente existe;\n- funding somente quando interpreter prevê binding cash;\n- production conversion somente quando ledger detecta deficit;\n- closing controller separado D27-D30.\n\nC4. Todo candidato recebe treatment-firing rate; se quase nunca dispara, não pode explicar ganho.\n\n## Fase D — gates competitivos\n\nD0 Smoke: 1 seed × 2 seats contra Broker + Rita.\n\nD1 Fresh paired: >=4 seeds × 2 seats. Rejeitar se Broker delta terminal <=0 e/ou Rita sofre W/L regression material.\n\nD2 Boundary: fresh Broker com controle MIX11 same-seed. W/L primário. Margem só desempata evidência de direção.\n\nD3 Regression: Rita + Rosa/Hana/Mateo.\n\nD4 Upper: Lena/Silas/Cleo somente depois de atravessar Broker ou produzir mecanismo material preregistrado.\n\nD5 OOD/policy invariance e runtime <1s antes de qualquer promoção.\n\n## Fase E — inteligência de adversário depois de motor econômico\n\nOBG/profiler não entra no caminho crítico enquanto a política econômica não atravessar Broker. Depois:\n- posterior de família/identidade com UNKNOWN/abstain;\n- self-action normalization;\n- calibrated reaction latency/hysteresis;\n- active probes somente se EVSI > custo esperado;\n- route por opponent/regime apenas em suporte validado.\n\n---\n\n# 11. CRITÉRIO DE PROMOÇÃO\n\nNenhuma candidata é champion enquanto não cumprir simultaneamente:\n\n1. fresh paired improvement contra Broker;\n2. pelo menos uma vitória Broker em suporte não trivial, depois replicação independente;\n3. não destruir Rita/weak-trio regression;\n4. mecanismo/tratamento comprovadamente aplicado;\n5. heldout seed families diferentes das de desenvolvimento;\n6. runtime compatível com `actTimeout=1s`;\n7. zero dependência de private/oracle/raw opponent action;\n8. mutation/negative controls dos validadores relevantes;\n9. champion MIX11 preservado como rollback até gate final.\n\n---\n\n# 12. ESTADO OPERACIONAL ATUAL\n\n- Sidecar histórico: **241 DONE / 5 FAILED / 0 READY/RUNNING** na fila antiga.\n- Novo loop de 15 minutos independe desta fila e escreve `kag_meta_cleo_cycles`.\n- `worker.ts` histórico continua ativo, mas sem queue READY naquele snapshot.\n- `meta_cleo_15min.ts` está ativo em main a cada 15 min.\n- Core main de laboratório: V26; default MIX11 teve non-regression mecânico verificado no seed113001 versus V25.\n- Candidate automática atual: `MILK_MIX_D6_V2`; ainda **não promovida** e com sinais iniciais negativos.\n- Cleo continua teacher de maior prioridade porque nos mesmos ciclos 1 e 2 bateu Broker 2-0 +2572 e 2-0 +2418.5 enquanto MIX11 e candidate perderam 0-2.\n\n---\n\n# 13. ÚNICO REGISTRO HUMANO DAQUI EM DIANTE\n\nA partir desta versão:\n\n- **este arquivo é o log humano canônico**;\n- cada rodada de 15 minutos grava somente uma linha compacta estruturada no SQLite;\n- novas conclusões materiais devem ser consolidadas aqui em checkpoint periódico, em vez de criar `hourly checkpoint + manifest + diff` por rodada;\n- dados brutos continuam no sidecar para auditoria/reprodutibilidade;\n- scripts e contratos operacionais permanecem separados porque são executáveis/reprodutíveis, não logs redundantes.\n\n**Próximo problema científico prioritário:** fechar o ledger de MILK e, em paralelo, decompor o teacher Cleo em mecanismos executáveis causais. Não repetir sweep de herd/land/labor/portfolio já falsificado.\n",[],[175,176,2375],"f34c3bdb-608c-4749-8185-edc9e570e283","KAG_CONTEXT_MANIFEST.json","2f4aa21c-ab2e-11f1-bb9a-1607ee4eb77e",624,["D",1788838332504],"{\n \"schema\": \"KAG_CONTEXT_MANIFEST_V1\",\n \"strategy_source_of_truth\": \"cleitoncosta/chat-native-maestro@economic-v0.6\",\n \"compute_sidecar\": \"cleitoncosta/kaggriculture-autolab@main\",\n \"entrypoint\": \"artifacts/kaggriculture/00_READ_FIRST.md\",\n \"hard_gate\": \"KAG_CONTEXT_PREFLIGHT=PASS before hypothesis/mutation/interpretation/enqueue\",\n \"required\": [\n {\"path\":\"artifacts/kaggriculture/00_READ_FIRST.md\",\"role\":\"bootstrap hard gate\"},\n {\"path\":\"artifacts/kaggriculture/intelligence/KAG_STRATEGIC_INTELLIGENCE.md\",\"role\":\"strategic doctrine / OBG / causal probing\"},\n {\"path\":\"artifacts/kaggriculture/intelligence/KAG_OPPONENT_CAPABILITY_TAXONOMY.md\",\"role\":\"190-trait opponent profile schema\"},\n {\"path\":\"artifacts/KAGGRICULTURE_REAL_WORLD_CURRICULUM_V1.md\",\"role\":\"external disciplines and transfer mechanisms\"},\n {\"path\":\"artifacts/kaggriculture_neuron_protocol_v1.md\",\"role\":\"competition research architecture\"},\n {\"path\":\"artifacts/kaggriculture_joined_research_plan_v1.md\",\"role\":\"joined competition thesis and evaluation discipline\"},\n {\"path\":\"artifacts/kaggriculture/intelligence/KAG_OFFICIAL_RULES.md\",\"role\":\"game/observability/runtime contract\"},\n {\"path\":\"artifacts/kaggriculture/intelligence/KAG_RT_ARCHITECTURE.md\",\"role\":\"sub-1s teacher/student architecture\"},\n {\"path\":\"artifacts/kaggriculture/intelligence/KAG_CURRENT_STATE.md\",\"role\":\"current reconciled strategic state / contradictions\"},\n {\"path\":\"artifacts/kaggriculture/intelligence/KAG_NEGATIVE_KNOWLEDGE.md\",\"role\":\"hard rejected mechanisms / reopening gates\"},\n {\"path\":\"artifacts/kaggriculture/intelligence/KAG_15MIN_FULL_STRATEGIC_REANALYSIS_20260908.md\",\"role\":\"forensic reanalysis of every persisted compute-active 15-minute window under the full opponent-intelligence strategy\"},\n {\"path\":\"artifacts/kaggriculture/intelligence/KAG_REAL_SCENARIO_COVERAGE_V1_20260908.md\",\"role\":\"mandatory coverage contract for real Kaggriculture world/time/shop/market/capacity/order-budget/OOD conditions\"},\n {\"path\":\"artifacts/kaggriculture/intelligence/KAG_RECOVERY_RECONCILIATION_20260907.md\",\"role\":\"recovery provenance and known gaps\"},\n {\"path\":\"artifacts/kaggriculture/intelligence/KAG_OPPONENT_PROFILES/INDEX.md\",\"role\":\"dossier index / source-vs-observation discipline\"},\n {\"path\":\"artifacts/kaggriculture/intelligence/KAG_EXPERIMENT_MATRIX_OPPONENT_PROBES.md\",\"role\":\"probe program and promotion gates\"},\n {\"path\":\"artifacts/kaggriculture/intelligence/KAG_PROFILE_CAL_V1_20260907.md\",\"role\":\"current profiler calibration evidence\"},\n {\"path\":\"artifacts/kaggriculture/hourly/HOURLY_INDEX.md\",\"role\":\"latest checkpoint locator\"}\n ],\n \"dynamic_reads\": [\n \"target opponent dossier\",\n \"latest hourly checkpoint + manifest\",\n \"sidecar queue/run/state/blob reconciliation\"\n ],\n \"anti_drift\": [\n \"IDENTIFICATION != REACTION_MODEL\",\n \"FINGERPRINT != CAUSAL_MECHANISM\",\n \"DIRECT_AWARENESS != MATERIAL_ECONOMIC_EFFECT\",\n \"SOURCE_GROUND_TRUTH != LEGAL_ONLINE_OBSERVATION\",\n \"OPPONENT_RAW_ACTION_OR_TRACE != DEPLOYABLE_FEATURE\",\n \"GOOD_VS_RITA != TRANSFERS_TO_BROKER\",\n \"MARGIN_IMPROVEMENT != COMPETITIVE_BREAKTHROUGH\",\n \"ONE_FAVORABLE_BATCH != REOPEN_NEGATIVE_KNOWLEDGE\",\n \"CLASSIFIER_ACCURACY_SAME_POLICY != POLICY_INVARIANCE\",\n \"FOUND != READ\",\n \"CHAT_MEMORY != PROJECT_STATE\"\n ]\n}",[],[175,176,2384],"b6558037-f3a6-4a40-a6ed-80a30828f094","00_READ_FIRST.md","734330c0-ab2d-11f1-8323-1607ee4eb77e",846,["D",1789011862487],"# KAGGRICULTURE — 00_READ_FIRST\n\nStatus: CANONICAL_BOOTSTRAP_GATE_V1\nPurpose: impedir perda de contexto, regressão para W/L-only e redescoberta de hipóteses já estudadas.\n\n## HARD GATE\n\n**Nenhuma nova hipótese, mutação, experimento, interpretação de resultado ou mudança de estratégia pode ser formulada antes desta sequência de leitura/reconciliação.**\n\n`CHAT_MEMORY != PROJECT_STATE`.\n\nSe algum arquivo obrigatório estiver ausente, desatualizado, contraditório ou não puder ser lido, declarar `CONTEXT_PREFLIGHT = PARTIAL/BLOCKED` e resolver a lacuna antes de alterar estratégia. Não preencher por memória/plausibilidade.\n\n## Fonte de verdade\n\n1. Estratégia, doutrina, perfis, regras, negative knowledge e checkpoints: `cleitoncosta/chat-native-maestro@economic-v0.6`.\n2. Compute sidecar, fila, runs e blobs: `cleitoncosta/kaggriculture-autolab@main`.\n3. Branches experimentais são evidência/candidatos, não fonte canônica até reconciliação/promote explícito.\n\n## Ordem obrigatória de reidratação\n\n### L0 — Missão e invariantes\n1. `artifacts/kaggriculture/00_READ_FIRST.md` — este arquivo.\n2. `artifacts/kaggriculture/intelligence/KAG_STRATEGIC_INTELLIGENCE.md` — missão OBG/probing/decision theory.\n3. `artifacts/kaggriculture/intelligence/KAG_OPPONENT_CAPABILITY_TAXONOMY.md` — **schema completo do perfil vivo de cada oponente**.\n4. `artifacts/KAGGRICULTURE_REAL_WORLD_CURRICULUM_V1.md` — disciplinas/analogias externas e pesquisa→mecanismo.\n5. `artifacts/kaggriculture_neuron_protocol_v1.md` — arquitetura competitiva e método Neurônio.\n6. `artifacts/kaggriculture_joined_research_plan_v1.md` — tese inicial e evidência/promotions.\n\n### L1 — Contrato do jogo e runtime\n7. `artifacts/kaggriculture/intelligence/KAG_OFFICIAL_RULES.md`.\n8. `artifacts/kaggriculture/intelligence/KAG_RT_ARCHITECTURE.md`.\n\n### L2 — Estado estratégico vivo\n9. `artifacts/kaggriculture/KAG_EVIDENCE_MASTER.md` — **registro humano canônico único de testes, evidências, negative knowledge, estado competitivo e plano profundo**. Ler integralmente antes de qualquer nova hipótese.\n10. `artifacts/kaggriculture/intelligence/KAG_CURRENT_STATE.md` — estado operacional resumido; se divergir do Evidence Master, reconciliar e atualizar.\n11. `artifacts/kaggriculture/intelligence/KAG_NEGATIVE_KNOWLEDGE.md` — contrato operacional de bloqueio; o conteúdo histórico também está consolidado no Evidence Master.\n12. `artifacts/kaggriculture/intelligence/KAG_REAL_SCENARIO_COVERAGE_V1_20260908.md` — contrato de cobertura dos cenários/condições reais; obrigatório antes de generalizar.\n13. `artifacts/kaggriculture/intelligence/KAG_RECOVERY_RECONCILIATION_20260907.md` — histórico de recuperação, não log corrente.\n\n### L3 — Opponent Intelligence\n13. `artifacts/kaggriculture/intelligence/KAG_OPPONENT_PROFILES/INDEX.md`.\n14. Ler o dossier do oponente-alvo e dos controles/regressions relevantes.\n15. `artifacts/kaggriculture/intelligence/KAG_EXPERIMENT_MATRIX_OPPONENT_PROBES.md`.\n16. `artifacts/kaggriculture/intelligence/KAG_PROFILE_CAL_V1_20260907.md` e calibrações posteriores.\n\n### L4 — Evidência mais recente\n17. Reconciliar `artifacts/kaggriculture/KAG_EVIDENCE_MASTER.md` com o sidecar live.\n18. No sidecar: ler `kag_meta_cleo_cycles`, reconciliar filas/runs/artifacts/blobs e confirmar a configuração ativa de `meta_cleo_15min.ts`.\n19. Logs hourly antigos foram consolidados e não são mais fonte corrente. `SQLITE/Blob RAW EVIDENCE != HUMAN SUMMARY`; discrepâncias devem ser registradas no Evidence Master.\n\n## Receipt obrigatório antes de raciocinar\n\nAntes de criar o próximo teste, produzir internamente/registrar no checkpoint:\n\n```text\nKAG_CONTEXT_PREFLIGHT\nREAD_FIRST = READ\nSTRATEGIC_INTELLIGENCE = READ\nCAPABILITY_TAXONOMY = READ\nOFFICIAL_RULES = READ\nCURRENT_STATE = READ\nNEGATIVE_KNOWLEDGE = READ\nOPPONENT_INDEX = READ\nTARGET_DOSSIER = READ\nEXPERIMENT_MATRIX = READ\nEVIDENCE_MASTER = READ\nSIDECAR_STATE = RECONCILED\nCONTRADICTIONS = [...]\nSTALE_FILES = [...]\nCONTEXT_PREFLIGHT = PASS | PARTIAL | BLOCKED\n```\n\n`FOUND != READ`. Apenas listar os arquivos não satisfaz o gate.\n\n## Missão estratégica que não pode ser reduzida\n\nKaggriculture **não** é somente otimização de W/L/margem nem comparação de tapes de reference agents.\n\nA missão inclui simultaneamente:\n\n- identificar família/regime do adversário;\n- medir **como ele reage ao que nós fazemos**;\n- separar reação direta à nossa fazenda de reação mediada pelo mercado;\n- mapear advance/retreat, iniciativa, agressividade, investimento, liquidez, risco, commitment/hysteresis, adaptação, memória/histórico, substituição de portfolio, closing e OOD;\n- medir reaction functions, elasticidades e latência de adaptação;\n- executar probes controlados com positive/negative controls;\n- estudar bait/bluff/deception apenas como fenômeno operacional observável e causalmente demonstrável;\n- usar Bayes/HMM/RLS/CUSUM/SPRT/EVSI/EVPI/Value of Computation quando forem úteis;\n- aprender um OBG vivo com posterior, confiança, contraevidência, contexto e transfer scope;\n- usar professor offline pesado e student online confortável sob o limite de ~1 s;\n- preservar champion/baseline e negative knowledge;\n- tentar falsificar qualquer PASS antes de promover.\n\n## Perguntas obrigatórias sobre cada oponente\n\nAntes de dizer “entendemos este oponente”, o dossier precisa responder, com evidência ou `UNKNOWN`:\n\n1. Quem/família/regime ele parece ser?\n2. Ele avança, recua ou mantém posição em quais contextos?\n3. O que ele trata como sinal de oportunidade/ameaça?\n4. Ele reage ao nosso cash, land, labor, crops, animals, lead/gap ou apenas ao mercado compartilhado?\n5. Qual é a intensidade e latência da reação?\n6. A reação desaparece quando o estímulo desaparece ou existe hysteresis/commitment?\n7. Ele muda com histórico diferente mesmo em estado atual semelhante?\n8. Ele substitui recursos/portfolio ou insiste na linha original?\n9. Como responde a scarcity, glut, price shock e order-slot contention?\n10. É possível induzir uma reação útil por probe/bait/bluff e ela paga o próprio custo?\n11. Como muda no closing/horizon?\n12. Qual o nível de confiança, counterevidence e OOD?\n\nA taxonomia canônica expande essas perguntas para **190 dimensões explícitas** mensuráveis; campos ainda não medidos permanecem `UNKNOWN`.\n\n## Hard anti-drift rules\n\n- `IDENTIFICATION != REACTION_MODEL`.\n- `FINGERPRINT != CAUSAL_MECHANISM`.\n- `DIRECT_AWARENESS != MATERIAL_ECONOMIC_EFFECT`.\n- `PUBLIC_STATE_EQUALITY != POLICY_EQUALITY`.\n- `SAME_SEED != SAME_PRESTATE` após políticas divergirem.\n- `TREATMENT_REQUESTED != TREATMENT_FIRED`.\n- `GOOD_VS_RITA != TRANSFERS_TO_BROKER`.\n- `MARGIN_IMPROVEMENT != COMPETITIVE_BREAKTHROUGH`.\n- `SOURCE_GROUND_TRUTH != LEGAL_ONLINE_OBSERVATION`.\n- `OPPONENT_RAW_ACTION_OR_TRACE != DEPLOYABLE_FEATURE`.\n- `ONE_FAVORABLE_BATCH != REOPEN_NEGATIVE_KNOWLEDGE`.\n- `CLASSIFIER_ACCURACY_SAME_POLICY != POLICY_INVARIANCE`.\n- `HEURISTIC_SCORE != CALIBRATED_POSTERIOR`.\n\n## Drift detector\n\nSe as últimas rodadas estiverem predominantemente fazendo qualquer um destes padrões, interromper e reconciliar:\n\n- apenas W/L/margem sem opponent reaction telemetry;\n- mutações estáticas cegas;\n- decompor reference-agent tape por muitas rodadas sem perguntar como o oponente reage a nós;\n- criar “personalidade” sem teste causal;\n- esquecer negative knowledge;\n- usar raw opponent action/source como se fosse observável online;\n- aperfeiçoar classificador sem testar cross-policy/OOD;\n- implementar adaptação antes de calibrar positive + negative controls;\n- gastar compute em hipótese sem EVSI/VoC ou teste discriminante claro.\n\n## Regra de atualização\n\nQualquer descoberta material deve atualizar pelo menos um destes artefatos antes de continuar:\n\n- `KAG_CURRENT_STATE.md`;\n- dossier do oponente;\n- `KAG_NEGATIVE_KNOWLEDGE.md`;\n- `KAG_STRATEGIC_INTELLIGENCE.md`;\n- `KAG_OPPONENT_CAPABILITY_TAXONOMY.md` se surgir nova dimensão;\n- checkpoint horário/manifests.\n\nExperiência de um único run é candidato de aprendizado, não regra universal.\n\n## Estado deste bootstrap\n\nEste arquivo nasce porque uma auditoria em 08/09/2026 confirmou que já existiam vários documentos canônicos de opponent intelligence, mas **não existia ainda um START_HERE específico do Kaggriculture no branch canônico**. O Grimório Chat Agêntico já possuía o padrão `00_CORE/00_START_HERE.md`; este arquivo aplica a mesma disciplina ao Kaggriculture.\n",[],[175,176,2392],"b726245d-69fd-4cb9-9abd-adf88724bdfc","KAG_STRATEGIC_INTELLIGENCE.md","01a07e48-3be7-749c-b196-c370a8cea8d0","01a07e48-3be6-726a-80bb-d3fa4969c331",838,["D",1788916979348],25,"# KAG Strategic Intelligence — Opponent Modeling / Active Probing / Decision Theory\n\n> Bootstrap requirement: read `../00_READ_FIRST.md` first. The full opponent trait contract is `KAG_OPPONENT_CAPABILITY_TAXONOMY.md`; local experiments may measure a subset but may not silently reduce the project scope.\n\nStatus: RECONSTRUCTED_CANONICAL_DRAFT_V1\nScope: strategy intelligence only. `MODEL_HYPOTHESIS != IMPLEMENTED_STUDENT != CHAMPION_PROOF`.\n\n## 1. Objective\n\nThe agent must optimize the probability of defeating the current opponent under a finite 720-turn partially observed economic game, not merely maximize its own average bank. W/L is primary; bank/margin/trajectory are diagnostics and teacher signals.\n\nThe strategic intelligence layer answers five questions continuously:\n1. Which opponent family/regime is most probable now?\n2. What does this opponent appear to value/react to?\n3. Which observations discriminate competing hypotheses?\n4. Is a controlled probe worth its economic/action cost?\n5. Which prevalidated strategy family should be routed under the current posterior/OOD state?\n\n## 2. Opponent Belief Graph (OBG)\n\nOBG is a belief/evidence graph, not a personality label.\n\n### Node classes\n- `OPPONENT_ID`: known reference name when available; UNKNOWN online otherwise.\n- `STRATEGY_FAMILY`: expansion-rusher, livestock-heavy, market-reactive, portfolio-switcher, static-tape-like, closer, etc.\n- `LATENT_REGIME`: early build, accumulation, conversion, liquidation/closing, defensive response, exploration/OOD.\n- `TRAIT_HYPOTHESIS`: market reactivity, our-state awareness, adaptation speed, commitment/hysteresis, substitution flexibility, risk appetite, history sensitivity.\n- `MECHANISM`: reaction function/threshold/portfolio rule inferred from evidence.\n- `EVIDENCE_EVENT`: immutable raw event or paired experiment result.\n- `COUNTEREVIDENCE`: explicit contradictions and failed predictions.\n\n### Minimum state per hypothesis\n- posterior / score;\n- uncertainty/confidence interval or effective sample size;\n- supporting event IDs;\n- contradicting event IDs;\n- context/world regime;\n- last update time/turn;\n- decay/forgetting factor;\n- OOD score;\n- transfer scope: UNIVERSAL / MATCHUP / WORLD_REGIME / LOCAL;\n- evidence class: OBSERVATIONAL / FIXED_PRESTATE_CAUSAL / HELDOUT / STUDENT / CHAMPION.\n\n### Bayes update\nFor opponent type/regime `θ` and history `h_t`:\n\n`P(θ | h_t) ∝ P(o_t | θ, a_0:t-1, c_t) P(θ | h_t-1)`\n\nwhere `c_t` is observable context (shops, market, day/horizon, public farm state). Never omit our own prior actions from the likelihood when measuring opponent response.\n\nA practical log-domain incremental implementation:\n\n`L_t(θ)=ρ L_{t-1}(θ)+log p(o_t | θ,c_t)`\n\nthen normalize with log-sum-exp. `ρ<1` permits forgetting under nonstationarity.\n\n## 3. Hierarchical opponent classifier\n\nThe classifier should be staged, because static visible state has different discriminative power by family.\n\n### Confirmed current evidence\nA fresh early-signature probe showed:\n- Mateo is strongly distinguishable by D1: early land2, heavy wheat+melon, no animals.\n- Rita is distinguishable: land1/13 wheat early and livestock by D3.\n- Broker/Lena/Silas are visually identical in the tested public farm snapshots through D9 (same cash/land/crop/animal counts in that probe).\n\nTherefore a single static snapshot classifier is structurally insufficient for top-family discrimination.\n\n### Cascade\n**Stage 0 — cheap static signature**\n- day/hour;\n- land;\n- crop/animal counts and composition;\n- cash;\n- hands/hires;\n- town regime.\n\n**Stage 1 — temporal/action signature**\n- first differences and velocities;\n- buy/sell composition and timing where actions are available in laboratory telemetry;\n- crop replacement/transition timing;\n- expansion thresholds;\n- response lags;\n- market price/inventory co-movement.\n\n**Stage 2 — active discriminating probe**\nOnly if posterior entropy remains high and expected information value exceeds treatment cost.\n\n## 4. Hidden/regime dynamics — HMM\n\nLet latent opponent regime `z_t∈{BUILD, ACCUMULATE, CONVERT, LIQUIDATE, RESPOND, OOD,...}`.\n\n`P(z_t | z_{t-1}) = A[z_{t-1},z_t]`\n\n`P(y_t | z_t)=emission(y_t; φ_z)`\n\nObservable emission vector can include:\n`Δcash, Δland, Δcrop_mix, Δanimal_mix, Δhands, sell/buy vectors, market Δinventory/Δprice, dGap, horizon`.\n\nUse forward filtering online; Viterbi/smoothing only offline. Regime labels are hypotheses until predictive validity is established.\n\n## 5. Reaction functions / RLS\n\nTo estimate whether opponent behavior changes as a function of our observable interventions:\n\n`y_t = β_t^T x_t + ε_t`\n\nwhere `x_t` contains our intervention magnitude/state + shared-market/town confounders and `y_t` is an opponent response feature.\n\nRecursive least squares with forgetting:\n\n`K_t = P_{t-1} x_t / (λ + x_t^T P_{t-1} x_t)`\n\n`β_t = β_{t-1} + K_t (y_t - x_t^T β_{t-1})`\n\n`P_t = (P_{t-1} - K_t x_t^T P_{t-1}) / λ`\n\n`λ<1` allows adaptation. Persist `β`, covariance `P`, effective sample size and context. Do not call `β` causal without paired controls/interventions.\n\n## 6. Change/regime detection — CUSUM\n\nFor residual or feature `r_t` versus baseline mean:\n\n`S_t^+ = max(0, S_{t-1}^+ + r_t - k)`\n`S_t^- = min(0, S_{t-1}^- + r_t + k)`\n\nTrigger regime-change suspicion when `S_t^+>h` or `S_t^-<-h`. Thresholds must be calibrated from no-change controls to manage false alarms.\n\nUse this to detect:\n- market-policy switch;\n- sudden expansion;\n- liquidation transition;\n- changed reaction to our state;\n- OOD behavior.\n\n## 7. Sequential probe stopping — SPRT\n\nFor probe hypothesis `H1: response effect ≥ δ` vs `H0: effect ≤ 0`, accumulate:\n\n`LLR_n = Σ log[p(y_i|H1)/p(y_i|H0)]`\n\nStop early for H1 if `LLR_n ≥ log((1-β)/α)`; stop for H0 if `LLR_n ≤ log(β/(1-α))`.\n\nPurpose: avoid spending unnecessary episodes/probe cost once evidence is already discriminative. Exact likelihood family must match the metric (Bernoulli W/L, Gaussian-ish paired delta, categorical regime event, etc.).\n\n## 8. EVPI / EVSI / Value of Computation\n\n### EVPI\n`EVPI = E[max_a U(a,θ)] - max_a E[U(a,θ)]`\n\nUpper bound on value of knowing the latent state perfectly.\n\n### EVSI for a probe `p`\n`EVSI(p) = E_y[max_a E[U(a,θ)|y,p]] - max_a E[U(a,θ)]`\n\n### Net probe value\n`NPV_probe = EVSI - economic_cost - action_opportunity_cost - risk_penalty`\n\n### Value of Computation\nFor extra offline/deep compute `c`:\n`VOC(c) = E[V_after(c)-V_before] - latency_cost(c) - resource_cost(c)`.\n\nOnline, only execute deeper reasoning when VOC remains positive inside the time budget. In competition runtime, information-gathering actions must also beat their in-game opportunity cost.\n\n## 9. Market elasticity and reaction metrics\n\nLocal empirical own/market elasticity estimate:\n\n`ε_qp ≈ (Δq/q)/(Δp/p)`\n\nbut in the shared market this is endogenous and confounded by both players + town demand. Use paired perturbations or structural conservation accounting before interpreting elasticity.\n\nOpponent reaction elasticity to our treatment `u`:\n\n`R_yu = Δy_opp / Δu_ours`\n\nwith lag profile:\n`R_yu(τ) = Cov(Δu_t, Δy_{opp,t+τ}) / Var(Δu_t)`\n\nEstimate separately by world regime; pool only after a homogeneity test.\n\n## 10. Revealed preference\n\nInfer latent utility/priority weights from choices under observable constraints instead of assuming motives.\n\nCandidate utility:\n`U(a|s)=w_cash Δcash + w_capacity Δcapacity + w_inventory Δinventory_value + w_market Δmarket_position + w_horizon Δterminal_option + ...`\n\nObserved action `a*` implies approximately:\n`U(a*|s) ≥ U(a|s)` for feasible alternatives `a`.\n\nUse inverse optimization / inequality constraints offline. Since unobserved private inventory changes feasibility, inferred weights require uncertainty bounds.\n\n## 11. Quantal Response Equilibrium / bounded rationality\n\nQRE-style action probability:\n\n`P(a|s,λ) = exp(λ Q(a,s)) / Σ_b exp(λ Q(b,s))`\n\n`λ→∞` approaches deterministic best response; small `λ` allows noisy/suboptimal choices. This is a useful likelihood family for opponent modeling, not a claim that competitors literally implement QRE.\n\n## 12. Level-k / strategic depth\n\nMaintain competing hypotheses:\n- L0: environment-only/static policy;\n- L1: reacts to market/opponent visible state;\n- L2: anticipates that opponent reacts;\n- higher levels only if predictive evidence justifies complexity.\n\nDo not reward “cleverness” in the model. A simple static policy that predicts actions better should dominate a sophisticated Level-k story.\n\n## 13. Hubris / commitment / susceptibility to bait\n\nThese are operational properties, not anthropomorphic labels.\n\n- **commitment/hysteresis**: probability of maintaining a portfolio/capacity decision after its original trigger reverses;\n- **overcommitment/hubris proxy**: delayed correction after a measurable adverse regime shift;\n- **bait susceptibility**: causal response to an observable, intentionally temporary signal compared with no-bait control;\n- **history sensitivity**: different response to same current state after different observable histories.\n\nAny “bluff/deception” experiment is invalid unless:\n1. the manipulated signal is officially observable to the opponent;\n2. the treatment actually changes that signal;\n3. paired controls establish that the opponent response changed because of the signal;\n4. net strategic benefit exceeds the cost of creating the signal.\n\n## 14. OOD detection\n\nUse a conservative ensemble of signals:\n- posterior entropy high;\n- Mahalanobis/standardized distance from training signature;\n- low likelihood under all known opponent models;\n- CUSUM regime alarm;\n- disagreement between classifier heads;\n- reaction prediction residual spike.\n\nOOD policy: fall back to robust baseline/risk-limited router; collect observations; do not extrapolate an exploit with low support.\n\n## 15. Opponent profiling dimensions\n\nEach dossier must track at minimum:\n- static economic signature;\n- expansion timing;\n- crop/animal portfolio transitions;\n- market buy/sell timing and item mix;\n- response to our market perturbation;\n- response to our visible farm expansion/contraction;\n- adaptation latency distribution;\n- persistence/hysteresis;\n- substitution matrix between products/assets;\n- history sensitivity;\n- horizon/closing behavior;\n- OOD reactions;\n- confidence + counterexamples.\n\n## 16. Experiment causality tiers\n\nE0 observational correlation.\nE1 matched seeds/mirrored seats.\nE2 treatment-fired paired control.\nE3 fixed-prestate conditional future forks.\nE4 independent heldout worlds.\nE5 student implementation heldout + regression.\nE6 runtime/provenance/counterproof.\nE7 online/champion evidence.\n\nNo hypothesis skips tiers merely because it is mathematically elegant.\n\n## 17. Immediate priority\n\nBegin controlled profiling on Rosa, Hana and Mateo because they are cheap/weak references that MIX11 already defeats reliably on sampled B21 worlds. The objective is not more wins; it is to validate the measurement apparatus against opponents whose strategies are easier to distinguish, then move the same profiling stack to Rita → Broker → Lena/Silas/Cleo.\n\nPrimary first tests:\n1. baseline temporal signatures;\n2. market perturbation response;\n3. visible expansion signal response;\n4. adaptation latency/hysteresis;\n5. replay same-current-state/different-history where achievable;\n6. OOD shop/market regimes;\n7. negative control proving a non-observable/private manipulation cannot be attributed as opponent response.\n\n## Reaction-timing intelligence update — 2026-09-08 04:24 UTC\n\nA new legal-observation instrumentation layer is now available in sidecar main: `KAG_AUTOLAB_CORE_V23_LEGAL_SELF_MEMORY_MAIN`. It records only the agent's own market flow alongside public legal observations, allowing known self-caused market transitions to be filtered without using opponent raw actions. Same-seed V22→V23 mechanical nonregression matched terminal banks/margins/W/L and 719 decision calls exactly in both seats on the promotion gate.\n\nThe strongest new strategic intelligence is a replicated **temporal response signature**, not a causal mechanism:\n- Cleo: very large lag0 response after our large public cash changes, rapid decay;\n- Broker: smaller positive lag0 plus recurring lag2 echo;\n- Lena: positive lag0 plus recurring lags2–4 tail;\n- Silas: stronger lag0 plus smaller lag2 echo;\n- Rita: near-zero/nonpositive control.\n\nBroker/Rita/Cleo preregistered fresh heldout reproduced this ordering, and Lena/Silas fresh heldout reproduced their respective tails. A first material our-policy shift (MIX11 → M40_FERT) preserved qualitative Broker/Cleo/Rita ordering but changed numerical magnitude, so policy-invariant calibration is **not** established.\n\nSeparate causal work bounds Broker's known 199→200 cash-awareness effect: 13/101 fresh controlled WHEAT-contention states changed cost, maximum 4 coins. Treat this trait as calibration/identity evidence, not as the primary boundary mechanism.\n\nResearch routing consequence: stop spending high EVSI budget on exploiting the 199/200 Broker threshold. Prioritize (1) causal recovery of context-specific Cleo slot/funding reactions, (2) shared-market mediation attribution using V23 self-memory, (3) second policy-shift/OOD falsification, then (4) calibrated family/regime routing with abstention. No champion or online exploit promotion follows from these signatures.\n\n\n## Material update — 2026-09-08 05:35 UTC — snapshot reactivity narrowed\n\nContext authority gate passed before this epoch. B49 completed the preregistered third policy-shift OOD: DEPLOY10 remained 0-6 versus Broker and 0-6 versus Cleo while going 6-0 versus Rita. This strengthens `GOOD_VS_RITA != TRANSFERS_TO_BROKER`; no champion or policy promotion follows.\n\nA fixed-history attribution program then attacked the leading shared-market explanation directly. With exact prior legal-public-history replay for stateful Cleo and fixed prestates, current WHEAT price/inventory, every individual public market price, every individual public market inventory, a broad public-opponent-cash curve, and simple unlocked-shop removal/duplication were perturbed separately. The experimental apparatus recovered independent positive controls (Rita WHEAT-price response; Broker 199->200 cash response) and negative controls (Rosa), while Cleo history replay remained exact.\n\nMaterial falsification:\n- Cleo and Broker showed no action-level response to any individual current public price in 12 stratified market-active MID/LATE/CLOSING contexts each;\n- no response to any individual current public inventory under +/-1000 perturbations in the same design;\n- broad public-cash curves recovered only the already-known 199->200 slot reorderings and no additional threshold through 50,000; sampled action branches had zero immediate public consequence;\n- simple one-shop removal/duplication produced no action changes for Cleo/Broker/Rita/Rosa in 32 tests per direction/target.\n\nTherefore the large V23 legal lag signatures must not be explained as generic **instantaneous snapshot elasticity**. Current classification is `SNAPSHOT_REACTIVITY_AS_PRIMARY_LAG_CAUSE = FALSIFIED_IN_TESTED_CONTEXTS`, not universal absence.\n\nHighest-EVSI lane becomes `PATH_DEPENDENT_TIMING_ORDER_SLOT_FUNDING_ATTRIBUTION`: distinguish lagged/path-dependent market-history response, slot/funding interactions and endogenous timing correlation while holding the terminal observable snapshot fixed. Every claim still requires treatment-fired evidence, effective n/context, immediate/downstream economic consequence and transfer/OOD boundaries. No bait/bluff or adaptive routing is eligible yet.\n\nThe full 190-trait contract remains unchanged; no trait was removed. Relevant status updates: current snapshot price elasticity top-tier = tested-context null; current snapshot inventory sensitivity top-tier = tested-context null; simple shop-multiplicity response = tested-context null; broad our-cash magnitude elasticity = no new regime beyond known contextual 199->200; reaction latency/history/path-dependence remain active UNKNOWN/priority traits.\n\n\n## 2026-09-08 05:51 UTC — fixed-terminal history falsification\n\nTeacher-only fixed-terminal interventions further narrowed the lead mechanism without changing policy/champion.\n\n- `fixed_terminal_history_cash_path_v6.ts`: at step 144, public-cash 199→200 fired immediately in Broker 6/6 worlds and Cleo 2/6, recovering the known direct-awareness positive control. With the terminal legal observable snapshot held identical and subsequent replay history identical, the treatment produced **0 differences at lags 1/2/6/24/72** for both targets. Classification: `DIRECT_AWARENESS_REAL / SIMPLE_CASH_HISTORY_MEMORY_NULL_IN_TESTED_CONTEXT`.\n- `fixed_terminal_prev_market_path_v7.ts`: changing only the immediately previous legal public WHEAT/MILK price or inventory observation, while holding the terminal snapshot identical, produced **0/8 lag-1 action differences** for Cleo across D6/D18 sampled contexts. Classification: `ONE_STEP_MARKET_MEMORY_NULL_IN_TESTED_CONTEXT`.\n\nFalsified/weakening explanation: a short-memory reaction to a single previous public cash/price/inventory observation is not sufficient to explain the legal lag signatures observed in natural trajectories.\n\nLead hypothesis remains `PATH_DEPENDENT_TIMING_ORDER_SLOT_FUNDING_ATTRIBUTION`, but is narrowed toward endogenous trajectory/phase/order-slot/funding interactions or longer/sparse interaction history, not naive one-step memory. Opposite hypothesis retained: sparse context interaction was missed by the finite fixed-history support.\n\nNext discriminant: perturb a **multi-step legal public trajectory segment** while restoring the same terminal snapshot, and separately test order-slot/funding history, with positive/negative controls and treatment-fired accounting. This is teacher-only causal evidence; not deployable feature and not champion proof.\n\n\n## 2026-09-08 06:02 UTC — sell-first funding/order mechanism replicated\n\nA detector correction falsified the first V11 interpretation: comparing absolute next-money across cash arms was tautological because starting cash itself differs. Failure class: `ABSOLUTE_CASH_CARRYTHROUGH_FALSE_POSITIVE`. The corrected detector compares `moneyDelta` plus structural/public market state.\n\n### Corrected V11\n`funding_execution_fixed_prestate_v11.ts`, seed 107951, mirrored seats, D12/D24, same exact requested queue, cash arms 0/200/1000/5000:\n- Broker: structural execution changed 0/4.\n- Lena: 0/4.\n- Silas: 0/4.\n- Cleo: 2/4, both at D12; D24 0/2.\nInterpretation: Broker/Lena/Silas queues in these sampled states are robust to low starting cash; Cleo is funding-sensitive at D12. Raw action invariance alone was not sufficient evidence, so official interpreter execution was required.\n\n### V12 fixed-prestate order counterfactual\nSame prestate + same multiset of requested market operations; only order changed from natural to `SELLS_LAST`. Treatment counted only when queue contained both SELL and spend-like action. Low-cash arms 0/200; natural cash is negative control.\n- Broker: treatment 4/4; order effect low-cash 4/4; natural-cash effect 0.\n- Lena: 4/4; low-cash 4/4; natural 0.\n- Silas: 4/4; low-cash 4/4; natural 0.\n- Cleo: treatment 2/4; low-cash 2/2; natural 0.\n\n### V13 fresh-heldout falsification\nFresh seeds 108011/108023/108037, mirrored seats, D12/D24:\n- Broker: treatment fired 12/12; low-cash order effect 12/12; natural-cash control 0.\n- Cleo: treatment fired 6/12; low-cash order effect 6/6; natural-cash control 0.\n\nClassification: `SELL_FIRST_SELF_FUNDING_ORDER = CAUSAL_REPLICATED / FRESH_HELDOUT_SUPPORTED / REGIME_CONDITIONAL_LOW_CASH / OFFLINE_TEACHER_MECHANISM / NOT_CHAMPION_PROOF`.\n\nMechanism: when a queue contains both revenue-generating sells and cash-consuming operations, executing SELL before spend can finance later operations that would otherwise fail under binding low cash. High-cash control convergence supports funding/order attribution rather than a generic queue-order artifact.\n\nLimit: tested high-tier natural states were sufficiently funded, so this mechanism is **not yet shown to explain the natural ~100k competitive gap**. Next discriminant is opportunity/materiality: measure frequency/severity of binding mixed SELL+spend states in MIX11 trajectories and test a student-safe sell-before-spend scheduler only where the funding detector predicts a real execution difference.\n\n\n## 2026-09-08 06:04 UTC — materiality audit: sell-first is already present in MIX11\n\nV14 instrumented only **our own requested action order** on the reaction-attribution branch; no opponent raw action was used. Fresh seeds 108201/108213, mirrored seats:\n\n- vs Broker: 221 mixed SELL+spend turns; MIX11 was SELL-first **221/221**, spend-first 0. Cash <2000 in 16 cases and <5000 in 60. No mixed queue saturated the 10-op market limit.\n- vs Cleo: 210 mixed SELL+spend turns; MIX11 was SELL-first **210/210**, spend-first 0. Cash <2000 in 16 cases and <5000 in 56. No mixed queue saturated.\n\nThis falsifies `SELL_FIRST_AS_MISSING_MIX11_CAPABILITY`. The V12/V13 mechanism remains causal and useful as a regression invariant, but a new sell-first mutation would be redundant because MIX11 already exhibits the ordering in every observed mixed queue.\n\nUpdated classification:\n`SELL_FIRST_SELF_FUNDING_ORDER = CAUSAL_REPLICATED / ALREADY_PRESENT_IN_MIX11_OBSERVED_SAMPLE / REGRESSION_INVARIANT_CANDIDATE / NOT_GAP_EXPLANATION_BY_ABSENCE`.\n\nNegative-space implication: neither missing sell-before-spend ordering nor mixed-queue 10-slot saturation explains the observed top-tier gap in the sampled trajectories. Continue attribution toward what differs materially despite this shared funding discipline: deployment scale/timing, portfolio conversion, land/labor/capacity trajectories, order-slot composition and closing utilization, under existing negative-knowledge gates.\n\n\n## 2026-09-08 06:22 UTC — early-capacity attribution update\n\nFresh legal-public V15 capacity landmarks show Broker/Lena/Silas sharing the same sampled D3/D6/D9/D12 capacity trajectory and Cleo remaining close, while MIX11 carries substantially less land/crop/animal productive stock by D9/D12. This is a top-family descriptive signature, not a causal prescription.\n\nThe preregistered single-factor `LABOR10_D9_15` treatment failed transfer against Broker: 0-6 both baseline and treatment, paired delta positive in only 2/6 episodes and negative in 4/6, with D15 gap worsening in every seed family. Therefore the current evidence rejects `labor target 10 during D9→D15` as the missing standalone mechanism.\n\nStrategic implication: do not respond by sweeping adjacent hands targets or fusing hands+land/herd. The highest-EVSI next step is to decompose the **composition and timing** of the common top-family productive-capacity signature using legal public crop/animal/land lifecycle landmarks, then formulate a new single-factor causal treatment only if the decomposition identifies a discriminant not already blocked by negative knowledge. Preserve `A+B good != A+B good` and `descriptive signature != transferable mechanism`.\n\n\n## 2026-09-08 07:33 UTC — B53 replication + B54 staged-portfolio causal falsification\n\nCanonical preflight was repeated after B54 compute and returned PASS before interpretation. B53 fresh legal-public evidence strengthened the descriptive result: Broker/Lena/Silas were exactly equal episode-by-episode at D3/D6/D9/D12 on public cash/land/crop/pasture/animal composition in all 6 mirrored episodes; Cleo remained a close top-family variant, while Rita remained WHEAT-only/1-land intermediate control. Classification remains `LEGAL_PUBLIC_DESCRIPTIVE_REPLICATED / NOT_CAUSAL`.\n\nThe preregistered single-factor B54 then tested whether the *time path* of the upper-family crop portfolio transfers independently onto MIX11 while keeping land/labor/herd/funding/sell/closing logic unchanged. Treatment fired in all episodes (`stagedPortfolioCalls=3`) and actual crop composition changed materially by D9/D12/D15.\n\nResult:\n- Broker control MIX11 0-6/-94384; staged 0-6/-113295.17; paired delta **-18911.17**, negative 6/6.\n- Rita control 6-0/+8988.83; staged 4-2/-585.50; paired delta **-9574.33**, negative 6/6.\n- Against Broker the paired gap was worse in every episode at D9/D12/D15/D21/D24/D27/D29.\n\nClassification: `STAGED_STRAW_D6_D12 = HELDOUT_FALSIFIED / HARMFUL / DO_NOT_PROMOTE`.\n\nMechanistic update: `TOP_FAMILY_CROP_COMPOSITION != INDEPENDENTLY_TRANSFERABLE_CAUSE_ON_MIX11_SUBSTRATE`. The leading alternative is now an **interaction/substrate** explanation: observed diversification may be productive only when paired with sufficient capacity/headroom/throughput/funding state, or may simply be downstream of that state. This is a new causal question, not permission to fuse rejected land/herd/hands mutations.\n\nHighest-EVSI next lane: teacher-only fixed-prestate/factorial moderation test of `capacity_headroom × portfolio_transition`, with identical future forks, explicit treatment firing and no student/champion mutation. Its purpose is to decide whether the crop signature is causal only under a capacity substrate or merely a correlated marker. Adjacent portfolio-share sweeps are blocked.\n\n\n## 2026-09-08 07:52 UTC — B55 factorial moderation falsified\n\nAfter canonical preflight PASS, the B55 2x2 offline-teacher factorial tested whether limited D6 capacity headroom makes the already-falsified staged portfolio beneficial. V25 mechanical gate recorded equal D6 pretreatment hashes and actual treatment divergence.\n\nFresh Broker `[108801,108813,108827]`, mirrored seats: A MIX11 0-6/-113730.17; B staged 0-6/-110910.33; C CAP2_D6 0-6/-96016.00; D CAP2_D6+staged 0-6/-113869.83. The interaction is strongly negative: baseline portfolio effect +2819.83, portfolio-under-headroom effect -17853.83, `I=-20673.67`.\n\nClassification: `CAPACITY_HEADROOM_MODERATES_STAGED_PORTFOLIO = REJECTED_IN_B55_FORMULATION`. The top-family crop trajectory remains a replicated fingerprint, not a transferable recipe, and added capacity does not rescue it in this formulation.\n\nC-A is a separate one-batch headroom signal (+17714.17 diagnostic margin, still 0-6). Because it is adjacent to prior early-land negative knowledge, it remains `ONE_BATCH_REOPENING_CANDIDATE / NOT_REOPENED`. A preregistered fresh Broker+Rita B56 gate is the only authorized next test of CAP2_D6; no adjacent land or crop×land sweep is authorized.\n\n\n## 2026-09-08 08:20 UTC — B56 falsifies D6 headroom reopening\n\nThe preregistered fresh Broker reopening gate falsified `CAP2_D6`: MIX11 0-6/-116285.00 versus CAP2_D6 0-6/-128129.33 on `[108901,108913,108927]`, mirrored seats. The B55 favorable diagnostic did not replicate and W/L remained zero. `CAP2_D6` is now closed in this formulation; no adjacent early-land/capacity sweep follows.\n\nStrategic routing consequence: the top-family productive-capacity trajectory remains descriptive/fingerprint evidence, while standalone staged portfolio, capacity×portfolio moderation, bounded labor and D6 headroom prescriptions have each failed their causal transfer gates. Compute priority returns to the undermeasured opponent-reaction dimensions: legal history/path dependence, order-slot/funding context, reaction magnitude/latency/hysteresis, and direct-vs-market-mediated attribution. No opponent reaction trait or champion is promoted by B56.\n\n\n## 2026-09-08 09:15 UTC — B56 regression closure and reaction-budget return\n\nFresh B56 Rita control/treatment completed after canonical preflight: MIX11 6-0/+9630.50; CAP2_D6 0-6/-9005.00. Together with the fresh Broker reversal, this closes CAP2_D6 as severe-regression harmful and removes any remaining reopening rationale. No adjacent capacity mutation is authorized.\n\nThe queue is redirected to the undermeasured OBG lane. B57 preregisters a fresh legal-observation-only path/timing replication on Broker/Cleo/Rita with new seeds `[109001,109013,109027]`. Its purpose is to try to break the previously replicated qualitative timing signature under fresh worlds before investing in a more expensive fixed-prestate path/funding causal probe. Classification remains `OBSERVATIONAL_DISCRIMINANT / NOT_CAUSAL`; no raw opponent action is deployable and no adaptive routing follows from a B57 PASS alone.\n\n\n## 2026-09-08 10:43 UTC — B58 completes upper-family observational timing map\nFresh B58 `[109101,109113,109127]`, legal telemetry only, reproduced Lena's immediate response plus delayed tail and Silas's stronger immediate response plus lag2 echo. Lena lifts: lag0 +0.2926, lag2 +0.1220, lag3 +0.0276, lag4 +0.0626. Silas: lag0 +0.5857, lag2 +0.1421, lags3/4 negative. Together with B45/B48 this strengthens qualitative temporal discrimination across fresh worlds/policy epochs, but it remains observational association. `IDENTIFICATION/TIMING_SIGNATURE != REACTION_MODEL`; direct-vs-market-mediated attribution, history/path dependence and causal economic value remain unresolved. Next highest-EVSI step is a preregistered fixed-history/fixed-prestate path × order-slot/funding discriminant with treatment-fired accounting and controls, not another macro mutation.\n\n\n## 2026-09-08 — continuous public-supply reaction candidate rejected structurally\nPost-clone natural-support closure, source/root-cause audit examined `_opponent_scale → _reserve_price → _plan_sells` as a candidate continuously-active direct response to our public crop/animal supply. Although `_SUPPLY_DRIVER` maps MILK/WOOL/EGG/FERTILIZER and crops to public production drivers, runtime constants show `_RESERVE={}` in Broker/Lena/Silas/Cleo. Therefore the current configured sell planner does not traverse this path. `CODE_PATH_EXISTS != ACTIVE_MECHANISM` is directly applicable. Do not attribute natural timing signatures to public supply-scale elasticity from this path. The active reaction evidence remains sparse/contextual cash-threshold behavior plus controlled-history clone/front-run with absent natural MIX11 support. Highest-EVSI current-loss lane is now fixed-plan conversion/throughput and operational scheduling, while direct-reaction traits not otherwise measured remain UNKNOWN.\n\n\n## 2026-09-08 12:35 UTC — operational capability narrowing\nFresh own-private/own-action census plus a preregistered isolated routing intervention narrows the Broker gap away from simple WHEAT-stock procurement and route-memory persistence. MIX11 held WHEAT well above zero/near-zero in D9-D15 and D18-D27 while actively buying WHEAT. Enabling `persistent_routing` fired 8/8 but increased the preregistered missed-feed stress metric and remained 0-8 Broker. Research priority therefore moves downstream/upstream from raw feed quantity and generic persistence toward **work-conversion throughput and spatial/task-service bottlenecks**: what fraction of urgent FEED/WATER/HARVEST/PICKUP jobs is actually served in time; carrier travel/zone burden; product conversion and cash realization. Any next test must census the missing service capacity first and isolate one factor; do not reopen blind hands/build/land routing sweeps. `RESOURCE_PRESENT != RESOURCE_DELIVERED`; however `PERSISTENT_ROUTING != PROVEN_DELIVERY_FIX` in the tested formulation.\n\n## 2026-09-08 13:15 UTC — WOOL world-regime mediation and direct-reaction nulls\nThe late WOOL divergence is not a stable scalar exploit. Fresh upper-family worlds show a tight Broker/Lena/Silas/Cleo WOOL footprint, but relative to Rita the price/inventory sign flips across seeds while respecting the shared-market inventory→price direction. Fixed-prestate tests with calibrated cash positive controls found no top-family action response to current WOOL price/inventory (0/12 each) and no action response to isolated public crop-mix or animal-mix changes (0/15 each) in sampled D3–D27 landmarks. Thus `WOOL_CONTEXT_ASSOCIATION != DIRECT_WOOL_REACTION` and `PUBLIC_COMPOSITION_DIFFERENCE != DIRECT_COMPOSITION_RESPONSE`. The high-EVSI lane is now world-conditioned shared-market supply/conversion attribution: explain when top-family sheep/output pushes WOOL inventory enough to change our realizable conversion, while keeping direct-vs-mediated attribution explicit. No WOOL-price or crop/animal bait is eligible from these nulls.\n\n## 2026-09-08 15:20 UTC — WOOL mediation narrowed beyond opponent-sheep story\nFresh world-conditioned mirrored census `[110501,110513,110527,110539]` preserves a very tight Broker/Lena/Silas/Cleo WOOL-market cluster, but paired comparison to Rita falsifies a simple monotone `more opponent sheep -> more shared WOOL inventory` attribution: opponent sheep delta is positive in all paired upper-family worlds while inventory follows that sign in only 4/8 pairs per target. The downstream shared-market relation is much stronger: WOOL inventory and price have opposite signs in 8/8 paired worlds for each upper target. Our own sheep/output varies materially with world/matchup and remains a live endogenous mediator/confounder. Therefore treat `inventory -> price` as a replicated structural association under fresh world conditioning, but keep causal supply decomposition unresolved. Do not infer direct reaction, resource denial or a stable WOOL exploit; next discriminant must separate our-output contribution and exogenous town/shop demand from opponent-output contribution at finer temporal resolution.\n\n## 2026-09-08 17:17 UTC — WOOL temporal conservation narrowing\nFresh teacher-only turn-level census `[110601,110613,110627,110639]` tested whether opponent public sheep count adds explanatory power for `Δ shared WOOL inventory` after our requested WOOL sells and our public sheep count. Broker: ΔR²=0 because opponent sheep was constant across the sampled D18-D27 rows; Rita: ΔR²≈0.000689. Every simultaneous nonzero inventory/price move retained inverse sign (Broker 202/202; Rita 96/96). This strengthens `WOOL_INVENTORY_TO_PRICE_LINK` but weakens `PUBLIC_OPPONENT_SHEEP_COUNT_AS_INCREMENTAL_TURN_LEVEL_MEDIATOR`. Do not overread the regression: sheep count is stock/capacity, not actual WOOL output/sale flow, and the preliminary town/shop-demand marker was zero in this implementation. Next discriminant must instrument actual public pasture yield/collection transitions plus deterministic demand timing before causal/economic intervention.\n\n## 2026-09-08 18:25 UTC — WOOL output-flow detector falsified, not opponent-output hypothesis\nFresh V3 `[110701,110713,110727,110739]` corrected the prior zero demand marker: visible YARN_STORE periodic-exposure and town-center timing now have nonzero support. However the candidate public sheep `yieldSum` transition feature is identically zero for both farms across all 1,728 D18-D27 rows per matchup (Broker and Rita). Therefore it cannot serve as the requested production/collection-flow proxy in the current telemetry. `ZERO_VARIANCE_DETECTOR != ZERO_CAUSAL_EFFECT`: the V3 ΔR²=0 for opponent yield is invalid as a mediation null. Preserve the exact inventory→price inverse relation (208/208 Broker, 148/148 Rita) as structural evidence only. Next work must repair measurement by finding a varying legal-public lifecycle transition or by constructing an engine-validated conservation residual; no WOOL policy, direct-reaction or resource-denial claim follows.\n\n## 2026-09-08 19:35 UTC — WOOL measurement root cause: legal signal exists, V27 aggregate false-zero\nAfter full canonical preflight, fresh V4/V4b reproduced the prior zero-variation result when reading V27 `scenarioTrace.publicTileLifecycle`. A separate direct official-engine observation probe then falsified the implied observability limitation: public SHEEP `yield_units` and `pending_care_bonus` change repeatedly in legal farm observations (Broker self-play seed 110901: 44 yield-sum and 24 pending-care transitions per farm across D3-D27; e.g. yield 0→6 at D6H0 then 6→0 at D6H7). Hence `ZERO_VARIANCE_SCENARIO_AGGREGATE != ZERO_VARIANCE_PUBLIC_SIGNAL`. Current root cause is telemetry/aggregation-path defect, not absent public lifecycle information. Repair V27 lifecycle aggregation and add a direct-observation parity test before any WOOL conservation regression. No causal opponent-output, direct-reaction, resource-denial or exploit claim follows yet.\n\n## 2026-09-08 20:21 UTC — WOOL measurement repair: consumer bug, not V27 aggregate\nFresh scenario telemetry `[111001,111013,111027]` disproved the 19:35 attribution of the false-zero to the V27 core aggregate. Reading the actual path `scenarioTrace[i].publicTileLifecycle.{me,opp}.animals.SHEEP` recovers strong variation: Broker context 84 own/264 opponent yield transitions and 142/144 pending-care transitions; Rita 99/118 yield and 136/136 pending-care transitions over 3,456 rows each. The earlier V4/V4b helper read `x?.publicTileLifecycle` after already passing `x.publicTileLifecycle`, silently yielding `{}`. Failure class: `NESTED_TELEMETRY_CONSUMER_PATH_MISMATCH`. New hard lesson: `CORE_SIGNAL_PRESENT != CONSUMER_PATH_CORRECT`; measurement validators must include nonzero-support assertions plus mutation controls. This repairs observability but does not establish opponent-output mediation, direct reaction, resource denial or exploitability.\n\n\n## 2026-09-08 21:19 UTC — measurement parity closed; WOOL mediation signal reopens correctly\nA true same-turn raw-public-tile ↔ aggregate parity validator passed on fresh Broker/Rita `[111101,111113,111127]`: 0 mismatches in 4314 rows/target for SHEEP count/yield/pending-care/unfed, while a deliberate +1 yield corruption was caught 4314/4314. Thus the measurement layer is now competent against the exact false-null class that caused V4/V4b.\n\nRepaired fresh WOOL V4 `[111201,111213,111227]` shows opponent lifecycle variation is measurable and adds a small but consistent incremental association with turn-level shared WOOL inventory movement against Broker: ΔR²=0.00809 aggregate, positive in every mirrored world (~0.0051..0.0137). Rita control is ΔR²=0.000431. This is useful discrimination but remains observational/mediational: public pasture yield/pending state is not executed opponent WOOL sale/collection, and current demand variables are exposure clocks, not an exact executed-demand ledger. Therefore `MEASUREMENT_PARITY_PASS != CAUSAL_MEDIATION_PASS`. Highest-EVSI next step is to reconstruct/validate exact engine demand removals and use conservation residuals; only then test whether opponent lifecycle explains residual supply and whether that materially changes our realizable WOOL conversion.\n\n\n## 2026-09-08 22:2x UTC — exact town-demand ledger changes WOOL mediation scale\nThe executed WOOL demand ledger is now engine-parity validated. Initial assumption `YARN_STORE copy removes 1` failed 180/3882 no-action transitions; official source showed single-product shops use multiplier 2. Corrected exact rule (`step%4`, YARN_STORE×2; `step%24`, town center×1) matched 3882/3882 and mutation +1 was detected 3882/3882. This is a reusable measurement lesson: `SHOP_EXPOSURE != EXECUTED_DEMAND_QUANTITY` and source semantics require parity validation.\n\nApplying exact town demand to fresh residual mediation `[111401,111413,111427]` reduces the earlier Broker lifecycle signal from the preliminary V4 scale: upper-family ΔR² remains above Rita but is small (Broker .001679, Lena .002708, Silas .002481, Cleo .002071; Rita ~.000001) and heterogeneous by world. Therefore exact demand strengthens the family discrimination qualitatively while weakening any claim of a large stable mediator. Do not promote resource denial, direct reaction, elasticity or WOOL exploit. Next gate is exact **our executed WOOL sell** parity, because requested SELL may differ from executed quantity under inventory/funding/order-budget semantics; only then is a conservation residual fit for causal/economic mediation testing.\n\n\n## 2026-09-08 23:2x UTC — full WOOL residual closes large-mediator interpretation\nThe pending own-sell gate is now closed. Fresh natural MIX11 D18-D27 census found no requested-vs-executable WOOL mismatch in sampled Broker/Rita worlds, and an independent official-interpreter probe validated the execution semantics including inventory insufficiency and 10-order truncation (SELL7→7 with stock, →3 with stock3, slot11 dropped, slot10 executes; mutation control PASS). Applying exact town demand plus validated own executed sells on fresh `[111701,111713,111727]` leaves upper-family lifecycle ΔR² only ~0.00030–0.00055 versus Rita ~0.000003, with strong world heterogeneity. Thus the family signal survives qualitatively but is too small/unstable to support a large stable WOOL mediator or exploit. `MEASUREMENT_REPAIR != MATERIAL_MECHANISM`. Next EVSI should test whether the unexplained residual can be identified as latent opponent executed supply through fixed-prestate conservation forks or whether WOOL should be deprioritized in favor of other conversion channels; no WOOL student mutation is justified.\n\n\n## 2026-09-09 00:2x UTC — WOOL accounting identifiability PASS; branch deprioritized pending exact natural supply\nFixed-prestate/no-town-demand teacher forks now prove the structural conservation channel directly: forced opponent WOOL SELL quantities 0/1/3/7/10/20 changed shared inventory one-for-one in 36/36 fresh seed/seat forks, with mutation control 36/36. Immediate price response at the tested market state was nonlinear (0,0,-1,-3,-6,-23 respectively). Thus opponent executed supply is in principle identifiable from a properly controlled shared-market residual, and can materially move price at larger quantities. This is engine mechanism evidence, not opponent behavior evidence.\n\nA fresh sparse raw-action checkpoint census found upper-family WOOL sells absent at D18/D21/D24/D27 checkpoints while Rita sold 30 units; checkpoint sparsity prevents a universal/natural-supply null. Together with the prior full-residual ΔR² only ~0.00030–0.00055 and strong world heterogeneity, the EVSI of further WOOL-specific work is now below broader conversion/throughput channels unless exact turn-level teacher accounting shows a materially larger upper-family executed-supply process. `STRUCTURAL_MARKET_IMPACT != NATURAL_OPPONENT_POLICY`. Preserve WOOL as a calibrated market-mechanism benchmark; do not promote a WOOL exploit or resource-denial story.\n\n\n## 2026-09-09 01:2x UTC — non-WOOL conversion AVIÃO selects MILK as conditional next lane\nAfter canonical preflight PASS, a fresh exact-self-executed D18-D27 conversion ranking showed MILK as the largest non-WOOL Broker↔Rita realized-value differential: -138 executed units / -16,921 value proxy in Broker context despite slightly higher MILK price. A second fresh batch replicated lower Broker-context MILK sale, shed availability, own cow count and public cow yieldSum, while Broker-context MILK price again averaged higher. This shifts the mechanistic question from `sell-price timing` toward `output/availability/conversion volume`.\n\nThe effect is strongly world-conditioned: one Broker world collapsed to ~2 cows and near-floor MILK price while other worlds carried 9-11 cows and high price. Therefore MILK is a **high-EVSI discriminant**, not a stable exploit or herd prescription. Existing early-herd negative knowledge remains closed. Next test must first close MILK conservation accounting (exact own executed sale + exact town/shop MILK demand + legal own lifecycle/output) and separate world demand from opponent-mediated market effects before any fixed-prestate intervention. `REALIZED_CONVERSION_DIFFERENTIAL != CAUSAL_CAPABILITY_GAP`; `HIGHER_PRICE_WITH_LOWER_VOLUME` is evidence against a simple low-price selling explanation.",[],[175,176,177,2400],"55022f0f-4f67-40c6-9033-0a4f933b9390","KAG_RT_ARCHITECTURE.md","01a07e48-3be7-749c-b196-bd3888f81ccc",598,["D",1788835231808],"# KAG_RT_CASCADE — Sub-1s Strategic Runtime Architecture\n\nStatus: ARCHITECTURE_DRAFT_V1\nOfficial hard limit revalidated: `actTimeout = 1 s`.\nDesign rule: `HARD_LIMIT != TARGET_LATENCY`.\n\n## 1. Architecture\n\n`OFFLINE TEACHER / AUTOLAB`\n→ learn/calibrate signatures, conditional Q, posterior likelihoods, HMM transitions, reaction models, thresholds, strategy routing and negative knowledge\n→ distill compact artifacts\n→ `ONLINE STUDENT / KAG_RT_CASCADE`\n→ deterministic executor + guarded fallback.\n\nNo LLM call, network call, large simulation tree or heavy retraining is allowed in the per-turn Kaggle action path.\n\n## 2. Runtime paths\n\n### FAST path — every turn\nGoal: deterministic bounded latency.\n\n1. Parse observation into compact feature vector.\n2. Update incremental deltas/ring buffer.\n3. Update cheap opponent posterior/log-scores.\n4. Update CUSUM/OOD signals.\n5. Evaluate event trigger.\n6. If no meaningful event: continue persistent task/market plan.\n7. If event: route among a small set of prevalidated macro policies.\n8. Generate farmer action + prioritized market orders ≤10.\n9. Validate action schema/budget and return.\n\nComplexity target: O(F + K + A) with small fixed `F` features, `K` opponent hypotheses and `A` candidate routes; no search scaling with episode history.\n\n### DEEP path — event-gated and still bounded\nOnly when posterior uncertainty/regime change materially changes expected decision value.\n\nAllowed online deep operations:\n- top-k strategy scoring;\n- one-step/small-horizon lookup or compact model evaluation;\n- incremental RLS update;\n- HMM forward update over small state set;\n- EVSI/VOC check based on cached tables.\n\nDisallowed online:\n- full Monte Carlo teacher;\n- Pyodide/reference-agent simulation;\n- large matrix optimization;\n- broad parameter sweeps;\n- unbounded graph traversal;\n- anything requiring external I/O.\n\n### FALLBACK path\nIf time guard, exception, invalid posterior, OOD extreme or action validation failure:\n- preserve deterministic safe baseline action/task;\n- avoid speculative market orders;\n- obey market-order limit;\n- emit minimal valid action.\n\n## 3. Event triggers\n\nRecompute macro routing only on information-changing events such as:\n- new day;\n- shop unlock / town regime change;\n- material market price/inventory change;\n- opponent expansion/crop/animal regime transition;\n- CUSUM alarm;\n- posterior class switch with hysteresis;\n- key lifecycle/capacity event;\n- horizon threshold / closing phase;\n- treatment/probe completion.\n\nThis preserves persistent low-level execution and prevents turn-by-turn retarget thrashing.\n\n## 4. Runtime state\n\nUse fixed-size state only:\n\n```text\nRTState {\n turn/day/hour\n previous compact feature vector\n ring buffer N of selected deltas\n opponent log-posterior[K]\n HMM belief[Z]\n RLS beta[d], covariance[dxd] for tiny d only\n CUSUM +/- per monitored residual\n OOD score\n active macro route\n route confidence/hysteresis timer\n current persistent task graph IDs\n probe state / treatment detector\n time-budget telemetry\n}\n```\n\nNever retain unbounded full history online. Full event history belongs in offline artifacts/replays.\n\n## 5. Teacher → student distillation\n\nTeacher can be expensive because it runs outside the action deadline:\n- fixed-prestate conditional forks;\n- nested/paired Monte Carlo;\n- racing/futility stopping;\n- robust Q/LCB/CVaR;\n- reaction-function identification;\n- regime model fitting;\n- inverse/revealed-preference analysis;\n- exploitability/best-response search;\n- ablation/mutation testing.\n\nDistill to one or more:\n- lookup table indexed by coarse state/regime;\n- tiny decision tree;\n- linear/logistic model;\n- small HMM transition/emission table;\n- reaction coefficients + confidence;\n- strategy-router thresholds;\n- explicit negative-knowledge vetoes.\n\nA teacher result becomes student logic only after heldout + regression evidence.\n\n## 6. Value of Computation gate\n\nAt event `e`, estimate:\n`VOC = E[decision_value_after_deep - current_decision_value] - latency_penalty - failure_risk`.\n\nRun DEEP only when:\n- `VOC > 0`;\n- enough wall-clock headroom remains;\n- uncertainty is decision-relevant;\n- fallback remains available.\n\nIf all candidate actions are unchanged across posterior support, extra computation has near-zero decision value and should be skipped.\n\n## 7. Suggested engineering latency budget — recommendation, not official rule\n\nThe only official number is 1 s. Until measured in actual submission runtime, use conservative internal guards:\n- FAST normal path: aim p99 well below 100 ms;\n- event/deep path: aim p99 below ~250–400 ms;\n- stop optional deep work around ~600 ms elapsed;\n- hard local fallback guard around ~700 ms;\n- reserve the remaining headroom for interpreter/runtime variance.\n\nThese thresholds are provisional engineering controls and must be replaced by measured p50/p95/p99/max data under packaged submission conditions.\n\n## 8. Profiling cascade under time constraint\n\n### Stage 0: signature match\nCheap feature likelihoods classify obvious families such as Mateo/Rita.\n\n### Stage 1: temporal posterior\nUse deltas/regime transitions to distinguish families that are static-snapshot aliases (current evidence: Broker/Lena/Silas through D9 in the tested probe).\n\n### Stage 2: active probe\nDo not improvise probe planning per turn. Offline teacher precomputes which probe is discriminative in which belief/context bin. Online router performs only:\n`probe_id = table[belief_bin, context_bin]` if cached `net_EVSI > threshold`.\n\n### Stage 3: route\n`policy_id = router[posterior, regime, horizon, OOD]` with hysteresis and minimum-confidence thresholds.\n\n## 9. OBG online representation\n\nThe full graph is an offline artifact. Online student keeps only sufficient statistics:\n- log posterior by strategy family;\n- latent regime belief;\n- calibrated confidence;\n- last switch/context;\n- contradiction counter;\n- OOD score;\n- selected reaction coefficients.\n\nThis prevents graph reasoning from consuming the 1 s budget.\n\n## 10. Safety / correctness guards\n\nBefore return:\n1. farmer action schema valid;\n2. market array length ≤10 after explicit priority sort;\n3. no action requests private opponent state;\n4. no stale plan crossing hard lifecycle/horizon constraints;\n5. numerical outputs finite/bounded;\n6. watchdog time check;\n7. fallback on exception.\n\n## 11. Telemetry required before promotion\n\nPer-turn local timing benchmark should record:\n- feature extraction µs/ms;\n- posterior update;\n- HMM/RLS/CUSUM;\n- route decision;\n- executor;\n- validation;\n- total latency;\n- fast/deep/fallback path counts;\n- p50/p95/p99/max;\n- exceptions/time-guard fallbacks;\n- output equivalence vs baseline when telemetry-only mode is enabled.\n\nA latency PASS must include negative controls that intentionally slow or corrupt a component and prove the watchdog/fallback catches it.\n\n## 12. Current implementation reality — reconciled 2026-09-08\n\nCanonical sidecar `kaggriculture-autolab@main` currently runs core `KAG_AUTOLAB_CORE_V22_LEGAL_TELEMETRY_SEPARATION` with worker `KAG_AUTOLAB_WORKER_V2_COMPACT_PERSISTENCE`.\n\nProgress since V19:\n- legal observation-only per-turn trace exists;\n- offline/raw opponent action telemetry is explicitly separated and marked non-deployable;\n- OBG trace instrumentation exists, but its current heuristic scores remain `HEURISTIC_CORRELATIONAL_UNCALIBRATED`;\n- profiler calibration recovered known weak-agent market thresholds and selected direct-awareness mechanisms;\n- reduced legal top-family classifier showed strong same-policy heldout performance but cross-policy Broker/Lena confusion exposed policy-confounding;\n- experimental V23 `LEGAL_SELF_MEMORY` stores our own action contribution so shared-market deltas can be self-normalized; it is not yet canonical/promoted;\n- full opponent-adaptive routing via OBG is still **not** promoted into the student.\n\nCurrent gate remains: measurement/calibration → cross-policy/OOD/history/reaction validation → compact distillation → latency validation → only then adaptive student routing. See `KAG_CURRENT_STATE.md` for the latest contradiction register and `KAG_OPPONENT_CAPABILITY_TAXONOMY.md` for the full profiling contract.",[],[175,176,177,2410],"f449a7b4-76ac-456f-9873-a46328853a4c","KAG_RECOVERY_RECONCILIATION_20260907.md","01a07e48-3be6-726a-80bc-0b2533197862",["D",1788825058278],"# Kaggriculture Recovery / Reconciliation — 2026-09-07\n\nStatus: RECOVERED_WITH_EXPLICIT_GAPS\nSource-of-truth parent: `cleitoncosta/chat-native-maestro@economic-v0.6` forked at version 567\nCompute sidecar parent: `cleitoncosta/kaggriculture-autolab@main` forked at version 107\nRecovery branch: `kag-intel-recovery-v1`\n\n## Recovered surfaces\n\n1. Original research/code history in `chat-native-maestro@economic-v0.6` (567-version branch history).\n2. Hourly checkpoints/manifests/diffs under `artifacts/kaggriculture/hourly/`.\n3. Autonomous compute evidence in `kaggriculture-autolab` SQLite + Blob artifacts.\n4. Current official Kaggriculture configuration and observation schema revalidated from the official Kaggle engine source.\n5. User-requested strategic architecture reconstructed from project/history context: opponent profiling, OBG, controlled probes, Bayes/HMM/RLS/CUSUM/SPRT/EVSI/EVPI/VoC, fast/deep runtime cascade, and weak-opponent-first profiling.\n\n## Negative result: GUH Knowledge Vault\n\nThe GUH Vault was queried for `Kaggriculture opponent profiling OBG CRAF`, `KAG_RT_CASCADE`, `Rosa/Hana/Mateo`, and adaptation/probe terms. Current Turso corpus has 6 conversations / 19 messages / 14 checkpoints / 10 knowledge items, but returned zero matching Kaggriculture items. Therefore the Vault is NOT evidence that the lost long conversation was ingested. Recovery relies on the Val Town project state/history + project conversation context, not a fabricated Vault transcript.\n\n## File Library search\n\nTargeted File Library search for `KAG_STRATEGIC_INTELLIGENCE`, `KAG_RT_CASCADE`, `KAGGRICULTURE_CRAF_TRANSFER_RECONSTRUCTION...`, and Rosa/Hana/Mateo profiling did not return those artifacts. CRAF-RW historical files were found, but not the Kaggriculture strategic documents requested here. Therefore these canonical documents are being reconstructed and persisted now rather than claimed as previously-existing files.\n\n## Reconciliation warning: evidence epochs\n\nHistorical hourly checkpoints and the current sidecar SQLite sometimes report materially different margins for similarly named batches (notably B2). Do not collapse them into one dataset. Preserve every result with:\n- evidence epoch;\n- source branch/version;\n- core version;\n- job id;\n- seeds;\n- artifact prefix/hash when available;\n- timestamp;\n- treatment detector state.\n\n`HISTORICAL_CHECKPOINT_RESULT != CURRENT_SIDECAR_RERUN_RESULT`.\n\n## Latest live correction after the 19:24 BRT hourly checkpoint\n\nAt 22:50:12Z the previously failed persistence retry `OUR-MIX11-BROKER-CTRL-B31` completed successfully.\n\nMatched B31 seeds: `[101401,101407,101421,101427]`.\n- `PASTURE_RAMP vs Broker`: 0-8, avg margin -96,945.25.\n- paired `MIX11 vs Broker`: 0-8, avg margin -76,344.625.\n- delta(PASTURE_RAMP - MIX11) = **-20,600.625**.\n- `PASTURE_RAMP vs Rita`: 4-4, avg margin +125.\n- paired `MIX11 vs Rita`: 4-4, avg margin +3,569.625.\n- delta Rita margin = **-3,444.625**.\n\nThis falsifies the prior provisional interpretation that PASTURE_RAMP had a stable directional Broker benefit. The earlier apparent B29/B31 directional improvement was caused by comparing before the paired B31 control had successfully persisted. Correct classification: `PASTURE_RAMP = HELDOUT_FALSIFIED / MATCHUP_UNSTABLE`, not promotable.\n\n## Current competitive state\n\n- Champion/proven baseline: unchanged.\n- Working lineage/baseline: MIX11.\n- Boundary: Broker Bea.\n- Stress: Closer Cleo.\n- Regression: Rancher Rita, plus cheap Rosa/Hana/Mateo suite.\n- Latest cheap suite: MIX11 8-0 each vs Mateo, Hana, Rosa (24-0 aggregate on B21 sampled worlds). Regression evidence only.\n- MIX11 remains 0-win against Broker in tested heldouts and 0-8 against Cleo in B31 (-104,137.875 avg margin).\n- Public ladder remains approximately `Broker < Lena < Silas < Cleo`, with non-deterministic small-N outcomes explicitly preserved.\n\n## Major causal negatives already established\n\nFalsified or quarantined as universal/boundary mechanisms:\n- fertilizer-only / static M40/M50/M60 melon shares;\n- static/persistent deployment H/W/B/COND;\n- bounded D9-D12 deployment pulses;\n- generic D21/D24 liquidation pulse (treatment verified, near-null);\n- unconditional pasture-JIT (independent heldout failures);\n- simple land-after-day6 threshold;\n- fixed D6 portfolio 2/4/4;\n- early melon40 / herd3/4/5 / build4 / herd+build variants;\n- PASTURE_RAMP after paired B31 control completed.\n\nDo not rediscover these by blind mutation.\n\n## Strong methodological lessons recovered\n\n1. W/L is the primary competition signal; terminal margin is diagnostic, not promotion proof.\n2. Same seed across diverging policies is comparable heldout evidence but not automatically an identical counterfactual pre-state.\n3. Fixed-prestate conditional forks are required for causal teacher estimates.\n4. A local teacher Q delta does not imply natural heldout W/L transfer.\n5. An intervention is not falsified until a treatment detector proves it actually fired.\n6. Infrastructure failures/alias/schema failures are not competitive losses.\n7. Public-state equality can hide economic-policy divergence; temporal actions/reactions are needed.\n8. Preserve negative knowledge and champion baseline before fusion.\n\n## First incomplete architectural step\n\nThe sidecar V19 contains OBG scaffolding but `OUR_VS_REFERENCE` does not actually call `obgUpdate`/`obgFastDecision`, and it does not persist opponent action-response telemetry. Therefore opponent adaptation has NOT yet been demonstrated. The next code branch must be telemetry-first, behavior-preserving, then controlled probes on Rosa/Hana/Mateo.",[],[175,176,177,2418],"906a4b6f-c124-4134-8301-a57922c255b9","KAG_REAL_SCENARIO_COVERAGE_V1_20260908.md","d7bdb3ec-ab35-11f1-92e6-1607ee4eb77e",659,["D",1788842127407],"# Kaggriculture — Real Scenario Coverage Contract V1 — 2026-09-08\n\nStatus: CANONICAL_SCENARIO_COVERAGE_V1\nPurpose: ensure every future opponent-intelligence claim is conditioned on the real Kaggriculture environment rather than a single convenient batch/world.\n\n## 1. Ground truth boundary\n\nCurrent official engine facts already revalidated:\n- 2 players;\n- 720 configured steps / 719 decision calls in the probed engine loop;\n- 30 days × 24 hours;\n- 1 s act timeout;\n- starting cash 3000;\n- 10×10 farm;\n- max 10 processed market orders/player/turn;\n- shed capacity 100 non-seed items with end-of-day overflow loss;\n- weed spawn probability 0.005 on eligible empty unlocked tiles;\n- shop unlock every 3 days, with replacement, repeated shops possible, max 8 instances;\n- each unlocked shop consumes every 4 turns;\n- town center consumes every 24 turns;\n- shared market prices/inventory;\n- opponent public farm observable, opponent private inventory not observable;\n- hidden environment seed to the agent.\n\nA finite sample cannot literally enumerate every stochastic state. Therefore `CONSIDER_ALL_REAL_SCENARIOS` means: define the full mechanic-derived scenario axes, measure coverage explicitly, preserve rare/OOD strata, and never generalize a reaction claim beyond the strata actually tested.\n\n## 2. Mandatory scenario axes for every material experiment\n\n### World/time\n- seed/evidence epoch;\n- seat 0/1;\n- step/day/hour;\n- phase EARLY / MID / LATE / CLOSING;\n- remaining decision calls;\n- world trajectory / prior path.\n\n### Town/shop regime\n- unlocked shop multiset;\n- number of unlocked instances;\n- repeated-shop multiplicity;\n- timing relative to 3-day unlock boundaries;\n- timing relative to 4-turn shop-consumption boundaries;\n- timing relative to 24-turn town-center boundaries.\n\n### Shared market regime\nFor WHEAT, CARROT, TOMATO, STRAWBERRY, MELON, MILK, WOOL, FERTILIZER:\n- price;\n- shared inventory;\n- scarcity/glut rank within sampled world;\n- recent delta/volatility;\n- whether our preceding action materially moved the same market;\n- market-order slot pressure.\n\n### Our legal private/capacity state\n- shed quantity and utilization against 100-item cap;\n- own carried/inventory quantity;\n- seed quantity where relevant;\n- market orders requested;\n- whether request count is <10, =10 or >10;\n- potential silently dropped orders;\n- own cash/land/crops/pastures/animals/hands.\n\n### Opponent legal public state\n- cash;\n- land/unlocked quadrants;\n- crop count/mix;\n- pastures/animals/mix;\n- hands/farmer public state;\n- observed tile kinds including weeds;\n- previous legal state and inferred public deltas.\n\n### Strategic context\n- identity/family posterior;\n- regime posterior;\n- our current policy;\n- treatment requested/fired;\n- direct-vs-market-mediated treatment classification;\n- lag since treatment;\n- whether current state is in-distribution for the fitted reaction model.\n\n## 3. New offline telemetry branch\n\nBranch: `cleitoncosta/kaggriculture-autolab@reaction-attribution-v1`.\nCore: `KAG_AUTOLAB_CORE_V24_REAL_SCENARIO_TRACE`.\n\nNew optional telemetry `scenarioTelemetry='FULL'` records, without changing actions:\n- legal public observer state;\n- our own private shed/inventory/seeds summary;\n- both farms' public tile-kind histograms;\n- shop multiplicity;\n- our own requested action summary;\n- 10-order budget saturation/drop-risk marker;\n- horizon phase/remaining decisions.\n\nEvidence state: `REAL_OFFICIAL_ENGINE_OWN_PRIVATE_PLUS_LEGAL_PUBLIC_TRACE`.\nThis is laboratory instrumentation only and is not merged into champion behavior.\n\n## 4. Instrumentation non-regression\n\n`real_scenario_trace_selftest_v1.ts` ran MIX11 against Rosa/Hana/Mateo on seed 100577, both seats.\n\nResult:\n- exact terminal/timeline/oracle-action signature equivalence between telemetry OFF and FULL: PASS for all 6 episodes;\n- exactly 719 scenario rows/game;\n- deliberate mutation detector caught the altered fixture;\n- state: `SCENARIO_TELEMETRY_NONREGRESSION_SELFTEST = PASS`.\n\nTherefore telemetry itself has not been shown to alter game behavior in this calibration set.\n\n## 5. First all-family real-world atlas\n\n`real_scenario_atlas_v1.ts`: MIX11, seed 100577, all 8 references, both seats = 16 official-engine games / 11,504 decision rows.\n\nObserved:\n- all four horizon phases represented;\n- repeated shops present in 7,184 rows;\n- maximum observed same-shop multiplicity = 4;\n- weeds observed on both farms;\n- own shed max = 70;\n- own requested market-order max = 9 in this sample;\n- own land range = 1..3 unlocked quadrants;\n- opponent land range = 1..3;\n- own cash range 1,290..59,705;\n- opponent public cash range 8..115,763.\n\nObserved market ranges in this sample:\n- WHEAT price 21..47, inventory 9,514..10,083;\n- CARROT 28..125, inventory 9,376..10,032;\n- TOMATO 60..302, inventory 9,599..10,000;\n- STRAWBERRY 1..339, inventory 9,322..10,062;\n- MELON 156..280, inventory 9,970..10,097;\n- MILK 1..301, inventory 9,735..10,076;\n- WOOL 198..249, inventory 9,686..10,006;\n- FERTILIZER 60..105, inventory 9,974..10,200.\n\nThese are sampled ranges, not engine limits.\n\n## 6. Targeted rare-condition scan\n\n`real_scenario_rare_scan_v1.ts`: refs Rita/Broker × variants MIX11, M60_FERT, MIX11_LIQ21, MIX11_LIQ24 × seeds 100591/100609 × both seats = 32 official-engine games.\n\nNew observed conditions:\n- own shed pressure reached 84 items;\n- 2 decision rows at shed >=80;\n- market-order budget saturation (=10 requested orders) occurred 20 times;\n- max same-shop multiplicity again reached 4;\n- weeds repeatedly observed.\n\nNot yet observed in these samples:\n- shed >=95 or actual 100-cap boundary;\n- >10 requested orders / real silent-drop risk in our emitted policy;\n- unlocked land 4+ in the measured atlas;\n- extreme duplicate-shop multiplicity above 4;\n- exhaustive scarcity/glut tails across seeds.\n\nThis does NOT mean those conditions are absent. They remain targeted rare/OOD strata.\n\n## 7. New hard interpretation rules\n\n- `SAMPLED_RANGE != ENGINE_LIMIT`.\n- `NOT_OBSERVED_IN_ATLAS != IMPOSSIBLE`.\n- `REACTION_EFFECT != UNIVERSAL_REACTION_EFFECT` unless replicated across materially different scenario strata.\n- `SAME_OPPONENT != SAME_MATCHUP_CONTEXT`; include our policy and world regime.\n- `DIRECT_RESPONSE_ESTIMATE` must control shared-market mediation.\n- `MARKET_RESPONSE_ESTIMATE` must record whether our own preceding orders changed the market.\n- `CLOSING_RESPONSE` must be tested separately from early/mid-game behavior.\n- `ORDERING_MECHANISM` must include 10-slot saturation context.\n- `CAPACITY_MECHANISM` must include shed-pressure/overflow-risk context.\n\n## 8. Scenario-stratified experiment minimum\n\nFor a material opponent-reaction claim, sample when applicable:\n1. both seats;\n2. multiple fresh seeds;\n3. at least EARLY/MID/LATE/CLOSING contexts or explicitly narrow scope;\n4. duplicate-shop and non-duplicate-shop worlds when relevant;\n5. low/median/high market-price/inventory strata for the affected item;\n6. opponent/our cash strata;\n7. order-budget unsaturated/saturated contexts for market-order claims;\n8. capacity-normal and capacity-pressure contexts for inventory/production claims;\n9. matched treatment/control with treatment-fired detector;\n10. lag/reversal observation window;\n11. OOD/rare-condition check before deployment.\n\n## 9. Immediate research implication\n\nThe next reaction-attribution program should use the real-scenario trace as the context layer. First target remains Broker because a true direct opponent-cash mechanism is already known and can serve as a positive control.\n\nThe improved question is no longer merely `does Broker notice our cash around 200?`.\nIt is:\n\n`under which real Kaggriculture regimes does the direct cash threshold fire, how often does it change an executed/economically relevant order, what is the lagged legal consequence, how does the effect interact with market price/inventory, shop regime, day/closing phase and 10-order slot pressure, and when is the effect economically negligible?`\n\nWeak trio provide negative/market-threshold controls; Rita remains intermediate regression; Lena/Silas/Cleo extend the same methodology to slot/funding/temporal mechanisms.\n\n## 10. Scenario-stratified reaction calibration — material update\n\nThe V24 branch now has three context-stratified teacher/control probes using natural official-engine prestates sampled every 6 in-game hours, fresh source calls only for the controlled counterfactual, and both seats.\n\n### Broker direct-cash positive control\n\n`broker_cash_context_probe_v1.ts`: Broker trajectories against Rosa/Rita/Cleo, seeds 100577/100609, both seats. At each sampled natural prestate, only the opponent-public-cash value presented to a fresh Broker action function was changed 199→200; all other observation fields were held fixed.\n\nResults:\n- sampled prestates: **1,440**;\n- Broker market-action difference: **86/1,440 = 5.97%**;\n- WHEAT sell-slot difference: **86/86** affected action branches;\n- selected immediate next-state digest differed after applying the two alternative Broker actions to the same official state: **40/86 = 46.51%**;\n- EARLY: 50/288 action differences, 36 with immediate digest difference;\n- MID: 36/576 action differences, 4 with immediate digest difference;\n- LATE: 0/432;\n- CLOSING: 0/144;\n- raw duplicate-shop stratification: 80/648 differences in non-duplicate contexts and 6/792 in duplicate contexts. This is **not** a causal shop effect because shop duplication is confounded with time/world trajectory.\n\nInterpretation: the known Broker cash threshold is a real direct-awareness mechanism, but it is **context-gated**. In this sample it was operational only in a minority of natural prestates and not at sampled late/closing prestates. Even when the action branch changed, immediate legal/economic consequences were not guaranteed.\n\n### Fresh-vs-stateful falsifier\n\n`broker_fresh_state_equivalence_v1.ts` compared the naturally stateful Broker action with a fresh reloaded source action on the exact same unmodified observation at all 1,440 sampled prestates above.\n\nResult: **0 mismatches / 1,440** across all four horizon phases and all three opponent contexts. This materially supports using the fresh action function for the 199/200 one-step teacher counterfactual **within the sampled regime**. It is not universal proof of statelessness outside sampled states.\n\n### Weak-trio direct-cash negative control\n\n`weak_cash_negative_context_v1.ts`: Rosa/Hana/Mateo against Rita, seeds 100577/100609, both seats, same 6-hour sampling and 199→200 opponent-cash perturbation.\n\nResult: **0 direct-cash action differences / 1,440 sampled prestates**, with 0 fresh-vs-stateful mismatches. Each target contributed 480 sampled states; all EARLY/MID/LATE/CLOSING strata stayed negative.\n\nThis is the calibrated negative control corresponding to Broker's positive control.\n\n### Known market-threshold positive controls under natural contexts\n\n`weak_price_context_positive_v1.ts` applied known source thresholds to natural prestates:\n- Hana CARROT 14→15: **61/480 = 12.71%** action/sell-quantity differences;\n- Hana TOMATO 24→25: **47/480 = 9.79%**;\n- Mateo MELON 119→120: **24/480 = 5.00%**;\n- fresh-vs-stateful mismatch: 0 for all three probes.\n\nThe phase distributions differ materially. For example, Hana CARROT threshold activity appeared in all four phases, while Mateo MELON was active only in sampled MID/LATE states. These are context-activation rates in this finite sample, not universal probabilities.\n\nNew methodological invariants:\n- `MECHANISM_PRESENT != MECHANISM_ACTIVE_IN_EVERY_PRESTATE`.\n- `KNOWN_THRESHOLD != ALWAYS_OBSERVABLE_FROM_ONE_STATE`.\n- `ACTION_BRANCH_DIFFERENCE != IMMEDIATE_ECONOMIC_STATE_DIFFERENCE`.\n- `RAW_CONTEXT_STRATIFICATION != CAUSAL_CONTEXT_MODERATOR` until the moderator itself is controlled.\n\nThese controls validate the direction of the new program: first learn **where a mechanism is active**, then its effect magnitude/lag and causal moderators, rather than assigning one scalar personality trait to the opponent.\n\n## 11. Promotion boundary\n\nThis V24 branch is measurement infrastructure only.\nNo champion/baseline change.\nNo adaptive routing enabled.\nNo opponent raw/private information may enter online student features.\nAny deployment candidate still requires current-engine provenance, fresh heldout, mirrored seats, regression, OOD, treatment detector, mutation control and measured runtime headroom under the 1 s action limit.\n\n## 12. Broker legal activation student — frozen fresh heldout\n\nA first all-in-one heldout script was terminated by the runtime before producing scientific output. This is classified as `COMPUTE_TERMINATION != MODEL_FAILURE`. The workload was then split into a frozen-training export and a heldout-only evaluator so the experimental design remained clean.\n\nTraining-only model (`mix11_broker_activation_train_export_v1.ts`):\n- train seeds `[106001,106013,106027,106039]`, both seats;\n- 960 legal-context rows sampled every 6 in-game hours;\n- 62 teacher-positive activation states;\n- teacher is the offline fixed-prestate Broker cash 199→200 counterfactual and is explicitly nondeployable;\n- features are deployable/legal observation context from our MIX11 view;\n- no fresh-vs-stateful baseline mismatches;\n- threshold frozen from training only = **0.78**;\n- train F1 **0.7222**, balanced accuracy **0.9027**.\n\nFrozen fresh-heldout evaluator (`mix11_broker_activation_heldout_frozen_v2.ts`) used **three previously unused seeds** `[106051,106063,106079]`, both seats, without refitting weights/scalers/threshold.\n\nHeldout result:\n- 720 rows;\n- 42 teacher-positive states;\n- baseline mismatch 0;\n- non-finite feature count 0;\n- AUC **0.9813176**;\n- TP 36 / FP 24 / TN 654 / FN 6;\n- precision **0.6000**;\n- recall **0.8571**;\n- specificity **0.9646**;\n- balanced accuracy **0.9109**;\n- F1 **0.7059**.\n\nEach of the three fresh seeds independently returned 12 TP / 8 FP / 218 TN / 2 FN (14 positives of 240), with per-seed AUC between **0.9810 and 0.9823**. This repeated structure is a replication signal, not proof of universality.\n\nPhase activation labels on heldout remained strongly context-gated:\n- EARLY 24/144 active;\n- MID 18/288 active;\n- LATE 0/216;\n- CLOSING 0/72.\n\nAn OOD diagnostic based on maximum absolute training z-score >6 flagged 6/720 heldout rows, max |z| 6.359. Those rows remain an explicit uncertainty stratum; OOD status is not itself a prediction error label.\n\nEvidence state: `MIX11_BROKER_LEGAL_ACTIVATION_FROZEN_FRESH_HELDOUT_V2 = PASS` for discrimination/generalization of the **activation label**, not for strategic exploitation or economic value.\n\nNew distinction: `CAN_PREDICT_HIDDEN_MECHANISM_ACTIVATION != KNOWS_HOW_TO_PROFIT_FROM_IT`. Next layer is effect magnitude/lag + selective uncertainty + runtime cost before any routing or probe exploitation.\n\n## 13. V25 legal public-state enrichment and Broker exploit-priority correction\n\nExperimental core advanced to `KAG_AUTOLAB_CORE_V25_PUBLIC_TILE_LIFECYCLE` on `kaggriculture-autolab@reaction-attribution-v1` without champion merge. `profilePublic` now includes legally visible `hands` and `hires_today`; scenario telemetry also preserves legal public tile lifecycle summaries derived from fields confirmed by `public_tile_schema_probe_v1.ts` (`crop`, `planted_day`, water stress, yield, lifespan/fertilization, pasture animal/age/feed/care/yield state, weeds). The telemetry non-regression suite remained PASS after these changes.\n\nAblation result: adding naive lifecycle aggregates to the Broker cash-activation student **hurt** leave-one-seed-out generalization relative to the leaner legal base. On the same 960 rows/62 positives, the base model reached AUC 0.9810 / F1 0.7838 / balanced accuracy 0.9522, whereas lifecycle-enhanced features reached AUC 0.9710 / F1 0.7160 / balanced accuracy 0.9444. Classification: `LIFECYCLE_FEATURES_AVAILABLE_BUT_NAIVE_FUSION_NOT_PROMOTED`. Richer observation is not automatically better prediction.\n\nThe improved base V2 student, trained only on seeds `[106001,106013,106027,106039]`, froze threshold 0.70 and was then evaluated without refit on fresh seeds `[106091,106103,106117]`, both seats. Fresh heldout: 720 rows / 44 positives, AUC **0.98252**, TP/FP/TN/FN **42/24/652/2**, precision **0.6364**, recall **0.9545**, specificity **0.9645**, balanced accuracy **0.9595**, F1 **0.7636**. OOD screen max |train-z|>6 flagged 6/720 rows. This strengthens discrimination of the hidden activation regime but active-call precision is still moderate; it is not an exploit gate.\n\nA train-only selective-gate audit also falsified an overconfident interpretation: there was **no threshold satisfying active precision >=0.90**. Best precision with at least four true positives was 0.7647 at threshold 0.86, recall 0.419. The inactive side could satisfy NPV>=0.99. Therefore `HIGH_AUC != HIGH_PRECISION_ACTIVE_GATE`.\n\n### Natural support of the 200-cash mechanism under MIX11\n\n`mix11_broker_natural_cash_support_v1.ts` ran six fresh seeds `[106131,106147,106159,106171,106183,106199]`, both seats = 12 MIX11-vs-Broker games / 8,628 legal decision rows. Our natural MIX11 cash was **never <=1000**, never <=500, never <=200 and never in 150..250. Minimum observed cash was **1301**; maximum 48,883. All 12 games lost to Broker.\n\nThis materially changes priority. Broker's 200-cash mechanism remains a real causal positive control and a useful hidden-mechanism benchmark, but it is outside MIX11's observed natural cash support in this fresh sample. Deliberately forcing our cash near 200 would require a large self-impoverishment intervention whose EV is currently unsupported.\n\nClassification: `BROKER_CASH200 = CALIBRATION_POSITIVE_CONTROL / LOW_DIRECT_EXPLOIT_PRIORITY`.\n\nHard distinction: `CAN_PREDICT_MECHANISM_ACTIVATION != NATURAL_TRIGGER_SUPPORT != PROFITABLE_INTERVENTION`.\n\n### Fresh Broker gap/action audit — where the actual loss develops\n\n`mix11_broker_gap_action_audit_v1.ts` used fresh seeds `[106211,106223,106237,106249]`, both seats. MIX11 went 0-8, average terminal margin **-105,169.5**. The parser was first normalized so BUY_LAND, BUY_ANIMAL, BUY_SEED, BUY_PRODUCT and HIRE are semantically separated; raw Broker actions remain offline-oracle evidence only.\n\nMean cash-gap changes by major 3-day windows show that we initially lead, then Broker's advantage compounds:\n- D0-D3: **+1,432** for us;\n- D3-D6: **+352.5**;\n- D6-D9: **-273.5**;\n- D9-D12: **-4,892**;\n- D12-D15: **-11,273.5**;\n- D15-D18: **-9,679.75**;\n- D18-D21: **-18,198.625**;\n- D21-D24: **-17,792.375**;\n- D24-D27: **-21,441.875**;\n- D27-D30: **-23,402.375**.\n\nOffline descriptive action evidence shows an upstream throughput transition rather than an early cash-threshold story. D9-D12 Broker averages 35 hires, 1 land buy, 109 WHEAT product purchases and 153 units sold, including 66 MELON. D12-D15 it averages 36 hires, 180 WHEAT product purchases and **501 units sold** (272 WHEAT, 81 MILK, 31 WOOL, 85 FERTILIZER plus smaller items), while our average sale volume is 57.5. D18-D21 Broker sells **499 units** versus our 40.875, with large STRAWBERRY/MILK/WHEAT flows. The worst gap deterioration is late-game, not at the direct cash-200 control.\n\nThis is descriptive and hypothesis-generating, not causal attribution. Current upstream candidate mechanisms are labor scale, WHEAT procurement/feed continuity, land timing, multi-product production conversion and market-order scheduling. Next causal program must isolate these components with treatment-fired paired controls rather than resurrecting previously falsified broad variants.",[],[175,176,177,2425],"9ed75c74-0fbf-4862-8548-9ad5235a8bfb","KAG_PROFILE_CAL_V1_20260907.md","01a07e48-3be6-726a-80bc-057e009974b2",["D",1788825058278],"# Kaggriculture Opponent Profiling Calibration Checkpoint — 2026-09-07 V1\n\nStatus: PROFILE_MEASUREMENT_STACK_PARTIALLY_CALIBRATED\nBranch: `cleitoncosta/chat-native-maestro@kag-intel-recovery-v1`\nAutolab telemetry branch: `cleitoncosta/kaggriculture-autolab@opponent-intel-v1`\n\n## What was validated\n\n### 1. Telemetry non-regression\n`KAG_AUTOLAB_CORE_V20_PROFILE_TELEMETRY_ONLY` reproduced all 24 terminal outcomes/banks from persisted V19 B21 weak-trio runs on seeds `[100577,100591,100609,100621]`, with zero mismatches. Each game produced 30 daily action aggregates + 11 action/OBG checkpoints, zero execution errors. Deliberate +1 expected-bank mutation was detected.\n\n### 2. Weak-trio fixed-prestate market calibration\nScript: `kag_weak_trio_fixed_prestate_floor_probe_v1.ts`.\n\n- Rosa CARROT 14/15/16 -> SELL 6/6/6; no floor detected; +50k opponent cash changed no action.\n- Hana CARROT 14/15/16 -> 0/3/3.\n- Hana TOMATO 24/25/26 -> 0/1/1.\n- Mateo MELON 119/120/121 -> 0/12/12.\n- Hana/Mateo floor suppression is overridden by shed-pressure and liquidation contexts, exactly as expected from source-ground-truth.\n- +50k opponent public cash changed no weak-trio action in tested prestates.\n\nThis validates both positive and negative controls and proves the detector must model interaction/context, not one global threshold.\n\n### 3. Direct opponent-cash awareness positive control\nScript: `kag_broker_opp_cash_fixed_prestate_probe_v1.ts`.\n\nBroker step-0 fixed prestate:\n- opponent cash 199 -> first WHEAT SELL at market slot 5;\n- 200 -> WHEAT SELL promoted to slot 0;\n- 201 -> same as 200.\n\nChanging Broker's own cash 199/200/201 while opponent cash remained 199 did not produce the transition. This is fixed-prestate causal evidence for direct opponent-cash awareness.\n\n### 4. Cash-sensitivity family screen\nScripts:\n- `kag_cash_awareness_blackbox_matrix_v1.ts`\n- `kag_cash_awareness_full_episode_firsthit_v1.ts`\n\nFindings:\n- Rosa: no step-0 sensitivity.\n- Hana: no step-0 sensitivity.\n- Mateo: no step-0 sensitivity.\n- Rita: no 199/200 action difference across 719 scanned steps in one natural Rita-vs-Rosa episode; classify as negative evidence, not universal absence.\n- Broker: sensitive immediately at step 0.\n- Lena: sensitive immediately at step 0.\n- Silas: sensitive immediately at step 0.\n- Cleo: no step-0 transition, but first fixed-prestate cash sensitivity appeared at D6 H0 / step 144, where WHEAT and FERTILIZER SELL ordering swapped.\n\nThis yields a family-level behavioral discriminator but does not separate Broker/Lena/Silas from each other.\n\n### 5. Authored market-controller source truth\n`kag_top_market_controller_truth_matrix_v1.ts` confirmed the authored market constants used for calibration:\n- Broker/Lena/Silas/Cleo share WHEAT opponent-cash promotion threshold 200.\n- Broker/Lena/Silas use `SELLS_FIRST=true` in the examined controller layer.\n- Cleo uses `SELLS_FIRST=false`, preserving non-sell/buy slots and only reordering sell positions.\n\nCaution: source contains common helper code paths that reference opponent state even when a current configuration may not activate them. `CODE_PATH_EXISTS != ACTIVE_MECHANISM`. Black-box treatment response remains required.\n\n### 6. Economic effect of Broker cash-awareness mechanism\n`kag_broker_cash_threshold_market_impact_v1.ts` scanned one natural Broker trajectory.\n\n33 fixed-prestate steps had a 199/200 WHEAT-slot transition. A controlled opponent WHEAT buy produced comparable acquisition in all 33. Only 6 had non-zero purchase-cost impact; largest observed saving at/above threshold was 4 coins for a five-unit buy (D6 H18, 139 -> 135).\n\nTherefore:\n`DIRECT_AWARENESS_CONFIRMED != PRIMARY_BOUNDARY_CAUSE`.\nThe measured immediate effect is real but tiny relative to MIX11's ~76k B31 average loss to Broker. This mechanism is useful for profiling/calibration but should not consume the main exploit-search budget unless a larger cumulative interaction is demonstrated.\n\n## New negative knowledge / corrections\n\n1. The weak agents provide excellent detector calibration precisely because they are *not* direct opponent-cash responders in tested/source-grounded policy paths.\n2. Broker/Lena/Silas share the same cash-threshold action response at step 0; cash awareness cannot distinguish those three.\n3. Cleo is cash-sensitive too, but the effect is masked until competing sell slots exist because its controller preserves buy slots.\n4. Raw opponent actions are laboratory-only evidence. The legal runtime observation does not expose opponent action lists. Any online classifier must infer these mechanisms from public farm + shared market/town state transitions.\n5. The Broker 200-cash threshold effect is low magnitude in the first causal purchase-cost screen. Do not promote a threshold-maintenance exploit on this evidence alone.\n\n## Next incomplete step\n\nBuild a **legal-observation-only temporal profiler** that never consumes opponent raw actions. It should record per-turn compact public/market deltas and attempt to recover known calibration distinctions indirectly:\n- weak trio market-floor signatures from shared-market/public-state consequences;\n- Broker/Lena/Silas/Cleo opponent-cash response from next-state market effects;\n- Cleo's sell-slot-preserving behavior versus the `SELLS_FIRST=true` family.\n\nOnly after the legal observer reproduces teacher labels should active in-game probes or adaptive strategy routing be enabled.",[],[175,176,177,2433],"ee7c8798-d798-467e-977b-8ea7a79a1124","KAG_OPPONENT_CAPABILITY_TAXONOMY.md","f48d83ce-ab2d-11f1-9c35-1607ee4eb77e",704,["D",1788852868258],"# KAGGRICULTURE — Opponent Capability Taxonomy\n\nStatus: CANONICAL_PROFILE_SCHEMA_V1\nPurpose: preservar a relação completa de capacidades/comportamentos que devem ser estimados para **cada** oponente. Um campo não medido é `UNKNOWN`, nunca ausência implícita.\n\n## 0. Regra epistemológica\n\nEste schema não é uma lista de adjetivos. Termos como “agressivo”, “investidor”, “recua”, “bluffa”, “teimoso” ou “adaptativo” só podem receber valor quando houver definição operacional + observação/probe + contexto + incerteza.\n\nCada trait deve armazenar:\n\n```text\ntrait_id\nestimate_or_posterior\nunits_or_scale\nconfidence / CI / effective_n\nstatus = UNKNOWN | OBSERVED | INFERRED | CAUSAL_LOCAL | HELDOUT_SUPPORTED | FALSIFIED\ncontext / world regime / day-horizon\nsupporting_event_ids\ncounterevidence_event_ids\nevidence_tier E0..E7\ntransfer_scope = LOCAL | WORLD_REGIME | MATCHUP | FAMILY | UNIVERSAL_CANDIDATE\nlast_update\nforgetting/decay if online\nfalsifier / next discriminating probe\n```\n\nScores 0..1 são permitidos apenas após calibração; antes disso use `UNKNOWN` ou estatística bruta.\n\n## 1. Identity / belief state\n\n1. `family_posterior` — distribuição sobre famílias conhecidas.\n2. `identity_posterior` — distribuição sobre oponentes específicos quando aplicável.\n3. `latent_regime_posterior` — BUILD/ACCUMULATE/CONVERT/LIQUIDATE/RESPOND/OOD/etc.\n4. `posterior_entropy` — ambiguidade atual.\n5. `classification_margin` — distância entre top-1 e top-2.\n6. `abstention_state` — UNKNOWN quando evidência insuficiente.\n7. `ood_score` — quão fora da distribuição conhecida está a trajetória.\n8. `classifier_policy_invariance` — estabilidade da classificação quando **nossa** política muda.\n9. `seat_invariance` — estabilidade entre assentos.\n10. `world_regime_invariance` — estabilidade entre shops/seeds/mercado.\n\n## 2. Initiative / tempo / estilo estratégico\n\n11. `initiative_score` — tendência a alterar o estado antes de ser forçado.\n12. `advance_score` — intensidade de expansão/commit de recursos em direção a maior capacidade/produção.\n13. `retreat_score` — intensidade de liquidação, redução de exposição, caixa defensivo ou interrupção de investimento.\n14. `hold_position_score` — persistência sem expansão/recuo material.\n15. `pressure_persistence` — duração da ofensiva após iniciar.\n16. `tempo_sensitivity` — quanto uma oportunidade de 1–N turns muda a decisão.\n17. `first_mover_propensity` — probabilidade de agir antes do rival/mercado confirmar tendência.\n18. `counterpunch_propensity` — preferência por responder em vez de iniciar.\n19. `simplification_preference` — tendência a reduzir variedade/risco quando à frente.\n20. `complexity_seeking` — tendência a aumentar diversidade/volatilidade quando atrás.\n21. `prophylaxis_preemption` — investimento para impedir/antecipar uma ameaça observável.\n22. `forcing_action_bias` — preferência por ações que alteram imediatamente restrições/mercado do rival versus construção lenta.\n\n### Definições-base\n\n`ADVANCE_t` deve combinar mudanças normalizadas de land, hands, productive assets, production pipeline e capital committed, descontando simples reposição.\n\n`RETREAT_t` deve combinar liquidation, queda de capex, cash reserve increase, portfolio contraction e abandono de capacidade, condicionado ao horizonte.\n\nNunca inferir “avanço” de uma variável isolada.\n\n## 3. Capital / investimento / liquidez\n\n23. `capex_appetite` — propensão a converter caixa em ativos irreversíveis.\n24. `land_expansion_propensity`.\n25. `labor_expansion_propensity`.\n26. `livestock_investment_propensity`.\n27. `crop_capacity_investment_propensity`.\n28. `working_capital_reserve` — caixa que tende a preservar antes de novo investimento.\n29. `liquidity_preference` — valor relativo de cash vs ativos/inventory.\n30. `cash_hoarding_score`.\n31. `capital_conversion_speed` — tempo entre acumular caixa e convertê-lo em capacidade.\n32. `reinvestment_rate` — fração aproximada de caixa incremental reinvestida.\n33. `reserve_buffer_sensitivity` — reação ao aproximar-se de caixa mínimo.\n34. `sunk_cost_commitment` — persistência após investimento adverso.\n35. `overcommitment_hubris_proxy` — demora em corrigir exposição após mudança adversa mensurável.\n36. `investment_reversal_threshold` — magnitude do choque necessária para interromper/reverter investimento.\n\n## 4. Market behavior / microstructure\n\n37. `market_reactivity` — mudança de política após mudança compartilhada de preço/inventory.\n38. `price_elasticity_by_item`.\n39. `inventory_elasticity_by_item`.\n40. `scarcity_response_by_item`.\n41. `glut_response_by_item`.\n42. `reservation_price_or_thresholds`.\n43. `threshold_context_overrides` — shed pressure, horizon etc.\n44. `sell_urgency`.\n45. `hold_propensity`.\n46. `front_run_propensity` — antecipação de venda/compra antes de deterioração esperada.\n47. `market_timing_score`.\n48. `buy_timing_score`.\n49. `sell_timing_score`.\n50. `order_slot_priority` — quais ações recebem slots escassos.\n51. `buy_slot_preservation` — proteção de compras quando SELL compete por slots.\n52. `funding_continuity` — manutenção de liquidez para compras/hires críticos.\n53. `market_order_budget_sensitivity` — degradação quando encosta no limite de 10 ordens.\n54. `market_impact_awareness` — comportamento compatível com antecipar o próprio impacto no shared market.\n55. `town_demand_sensitivity`.\n56. `shop_regime_sensitivity`.\n57. `product_substitution_matrix` — probabilidade/intensidade de migrar produto A→B sob choque.\n58. `strategic_complementarity` — tende a produzir o mesmo que nós quando produzimos mais?\n59. `strategic_substitutability` — tende a migrar para produto complementar quando entramos num mercado?\n\n## 5. Direct awareness of us vs market-mediated response\n\n60. `our_cash_sensitivity`.\n61. `our_cash_gap_sensitivity`.\n62. `our_land_sensitivity`.\n63. `our_labor_sensitivity`.\n64. `our_crop_mix_sensitivity`.\n65. `our_animal_mix_sensitivity`.\n66. `our_capacity_sensitivity`.\n67. `our_expansion_event_sensitivity`.\n68. `our_contraction_event_sensitivity`.\n69. `our_lead_sensitivity` — resposta à nossa vantagem observável.\n70. `our_deficit_sensitivity` — resposta quando estamos atrás.\n71. `our_market_flow_sensitivity` — resposta ao que nossa ação faz aparecer no mercado.\n72. `direct_awareness_probability` — posterior de reação direta ao nosso estado público.\n73. `market_mediated_reaction_probability` — posterior de reação explicada pelo shared market.\n74. `direct_vs_mediated_attribution_confidence`.\n\nRegra obrigatória:\n`REACTION_AFTER_OUR_ACTION != DIRECT_REACTION_TO_US`.\nPrecisamos controlar preços, inventory, town e demais mediadores observáveis.\n\n## 6. Reaction function / adaptation dynamics\n\n75. `reaction_magnitude` — efeito estimado do tratamento sobre a resposta.\n76. `reaction_elasticity` — Δresposta/Δtratamento.\n77. `reaction_latency_distribution` — τ de resposta, não apenas média.\n78. `adaptation_speed`.\n79. `adaptation_probability` — chance de mudar a política após sinal suficiente.\n80. `persistence_after_response`.\n81. `hysteresis_duration` — quanto a nova política persiste após remoção do estímulo.\n82. `reversal_speed`.\n83. `regime_switch_frequency`.\n84. `regime_switch_threshold`.\n85. `counter_adaptation` — adapta-se novamente quando percebe nossa resposta à primeira adaptação?\n86. `repeated_probe_learning` — resposta muda após repetirmos probes?\n87. `forgetting_decay` — peso da história recente versus antiga.\n88. `prediction_residual` — erro do reaction model; usado também como sinal OOD.\n\nEstimadores candidatos: RLS com forgetting, HMM, CUSUM, Bayes, SPRT e lagged response curves. `MODEL_FIT != CAUSALITY`.\n\n## 7. Risk / lead / comeback behavior\n\n89. `risk_appetite`.\n90. `variance_seeking`.\n91. `downside_aversion`.\n92. `tail_risk_tolerance`.\n93. `lead_defense` — de-risk quando à frente.\n94. `lead_pressing` — aumenta pressão quando à frente.\n95. `loss_chasing` — aumenta risco quando atrás.\n96. `comeback_aggression`.\n97. `draw_or_safe_outcome_preference` quando aplicável ao estado competitivo.\n98. `robustness_vs_exploration_tradeoff`.\n\n## 8. Portfolio / production / operations\n\n99. `portfolio_concentration`.\n100. `portfolio_diversification`.\n101. `portfolio_flexibility`.\n102. `portfolio_switch_cost` observado/inferido.\n103. `vertical_integration_score` — grau de cadeia interna feed→animal→produto/fertilizer etc.\n104. `feed_dependency`.\n105. `input_dependency_by_item`.\n106. `capacity_headroom`.\n107. `bottleneck_sensitivity` — reage a labor/storage/feed/land/market-slot bottlenecks?\n108. `labor_utilization`.\n109. `land_utilization`.\n110. `asset_utilization`.\n111. `operational_burstiness` — ações agrupadas versus fluxo suave.\n112. `execution_determinism` — variação de execução em estados semelhantes.\n113. `route_efficiency_proxy` quando publicamente inferível/lab-calibrável.\n114. `inventory_pressure_response`.\n\n## 9. History / memory / path dependence\n\n115. `history_sensitivity` — mesmo estado atual observável, histórias diferentes → resposta diferente?\n116. `path_dependence`.\n117. `recency_weight`.\n118. `long_memory_weight`.\n119. `state_alias_disambiguation_need` — estado atual é insuficiente para prever ação?\n120. `opponent_specific_memory` — existe comportamento compatível com condicionamento ao nosso padrão anterior?\n121. `previous_probe_effect`.\n122. `previous_market_shock_effect`.\n123. `previous_lead_state_effect`.\n\nTeste ideal: duas histórias que convergem ao mesmo estado observável/prestate tão próximo quanto possível e depois recebem future fork comum.\n\n## 10. Probe / bait / bluff / deception\n\n124. `probe_detectability` — o sinal criado por nós é realmente observável pelo adversário?\n125. `probe_response_rate`.\n126. `probe_response_magnitude`.\n127. `probe_evsi`.\n128. `probe_economic_cost`.\n129. `probe_net_value` = EVSI/strategic gain − opportunity/economic/risk cost.\n130. `bait_susceptibility` — muda de política diante de sinal temporário e reversível?\n131. `bait_overreaction` — resposta excede proporcionalidade esperada.\n132. `bluff_susceptibility` — reage a um sinal de intenção que não mantemos.\n133. `bluff_recovery_speed` — quão rápido corrige após o sinal desaparecer.\n134. `repeated_bluff_resistance` — aprende a não reagir à repetição?\n135. `trap_susceptibility` — resposta previsível que podemos explorar depois de provocar.\n136. `signal_discrimination` — distingue sinal persistente de ruído/transitório?\n137. `deception_propensity` — somente se houver fenômeno operacional observável em que o próprio oponente cria sinal temporário enganoso; nunca antropomorfizar por padrão.\n138. `probe_value_of_computation` — vale gastar compute/tempo para decidir o probe?\n\nBluff/bait só é válido se: sinal observável → treatment detector PASS → paired/fixed-prestate response difference → benefício líquido positivo. Caso contrário = narrativa.\n\n## 11. Strategic depth / game-theoretic behavior\n\n139. `level0_probability` — policy essencialmente environment/static.\n140. `level1_probability` — reage ao oponente/mercado.\n141. `level2_probability` — evidência de antecipar reação do rival.\n142. `best_response_strength`.\n143. `exploitability_estimate`.\n144. `exploit_detection_speed` — muda quando nossa exploração se repete?\n145. `anticipation_score` — age antes de consequência compartilhada tornar-se visível?\n146. `revealed_preference_weights` — pesos inferidos para cash/capacity/inventory/market/horizon.\n147. `qre_rationality_lambda` — parâmetro de decisão ruidosa se esse modelo passar validação.\n148. `strategic_consistency` — previsibilidade das escolhas dadas constraints semelhantes.\n149. `policy_entropy` — diversidade/noise das respostas.\n150. `response_oracle_family` — melhor modelo atual para prever a ação/resposta.\n\n## 12. Closing / finite horizon\n\n151. `horizon_awareness`.\n152. `investment_cutoff_timing`.\n153. `production_cutoff_timing`.\n154. `liquidation_start_timing`.\n155. `liquidation_aggression`.\n156. `terminal_conversion_efficiency`.\n157. `unsold_inventory_risk`.\n158. `late_cash_preservation`.\n159. `late_market_slot_priority`.\n160. `closing_pressure_response` — reage diferente à nossa posição perto do final?\n\n## 13. Robustness / OOD dimensions\n\n161. `seed_robustness`.\n162. `seat_robustness`.\n163. `shop_mix_robustness`.\n164. `duplicate_shop_robustness`.\n165. `price_shock_robustness`.\n166. `scarcity_robustness`.\n167. `glut_robustness`.\n168. `our_policy_shift_robustness`.\n169. `unseen_opponent_behavior_robustness`.\n170. `classifier_confidence_calibration`.\n171. `reaction_model_confidence_calibration`.\n172. `fallback_trigger_rate`.\n173. `abstention_quality` — quando abstém, evita erros de alta confiança?\n\n## 14. Recovered original OBG dimensions — explicit anti-loss additions\n\nA reconstruction check against the earlier conversation recovered several traits that were conceptually present in the original OBG but not explicit enough in sections 1–13. They are preserved here by name so they cannot be silently collapsed into neighboring metrics:\n\n174. `defensiveness_score` — tendência a proteger posição/capital/capacidade quando ameaçado, distinta de simplesmente `retreat`.\n175. `patience_score` — tolerância a esperar por payoff/condição melhor sem ação precipitada.\n176. `impulsivity_score` — propensão a reagir cedo a sinal fraco/transitório.\n177. `exploration_rate` — frequência/intensidade de testar alternativas novas sob incerteza.\n178. `exploitation_rate` — preferência por repetir a linha atualmente melhor conhecida.\n179. `behavioral_inertia` — resistência geral a mudar de política, separada de hysteresis após um tratamento específico.\n180. `opportunism_score` — intensidade de aproveitar janelas locais curtas criadas por preço, rival ou shops.\n181. `resource_denial_propensity` — comportamento que compra/ocupa/retira recurso/slot/capacidade para limitar opções do rival, quando causalmente demonstrável.\n182. `predictive_anticipation` — comportamento consistente com agir antes do evento compartilhado ocorrer.\n183. `reactive_following` — comportamento predominantemente posterior ao evento/sinal já materializado.\n184. `composure_under_shock` — estabilidade da política após choque adverso/volatilidade, sem overreaction.\n185. `persistence_under_adversity` — manutenção de plano/pressão mesmo após sequência de outcomes adversos.\n186. `counter_bluff_propensity` — resposta estratégica a sinal suspeito/temporário depois de evidência de bluff; requer probes controlados.\n187. `probe_awareness` — evidência de que o oponente detecta/neutraliza nossas ações de sondagem repetidas.\n188. `models_us_probability` — posterior de que a política depende de um modelo do nosso comportamento/histórico, além de simples leitura do estado atual.\n189. `deception_detection` — capacidade inferida de distinguir bait/bluff de mudança persistente; medir por resposta diferencial a sinais equivalentes com histories distintas.\n190. `saturation_sensitivity` — resposta a mercado/capacidade saturada mesmo quando preço isolado ainda não cruzou threshold.\n\nThese additions correspond to the recovered original concepts: `advance/retreat`, `bluff/counter-bluff`, `aggression/defensiveness`, `risk`, `commitment/reversibility`, `patience/impulsivity`, `exploration/exploitation`, `adaptability/inertia`, `memory/recency`, `awareness of our state/market`, `predictiveness/reactiveness`, `opportunism/resource-denial`, `price/supply/demand/cash sensitivity`, `scarcity/saturation`, `diversification/concentration`, `timing`, `closing`, `composure/persistence`, `reasoning depth`, `models-us` and `deception/probe awareness`.\n\n## 15. Composite profiles — apenas derivados\n\nOs termos abaixo são **views derivadas**, nunca observações primárias:\n\n- `AGGRESSOR` = initiative + advance + capex + pressure persistence − retreat, condicionado ao contexto.\n- `INVESTOR` = capex appetite + reinvestment + long-horizon assets + capital conversion pattern.\n- `DEFENDER` = lead defense + liquidity preference + de-risk/simplification behavior.\n- `CHASER` = comeback aggression + variance seeking + investment/risk increase quando atrás.\n- `MARKET_TRADER` = timing + elasticity + slot priority + market-impact awareness.\n- `ADAPTER` = treatment response + low latency + regime switching + counter-adaptation.\n- `COMMITTED` = persistence + hysteresis + sunk-cost commitment.\n- `FLEXIBLE_PORTFOLIO` = high substitution/diversification change, low switch cost.\n- `BAIT_SUSCEPTIBLE` = causal temporary-signal response + overreaction + exploitable net value.\n- `HISTORY_SENSITIVE` = current-state matched future actions depend materially on prior path.\n\nNenhuma label pode esconder os componentes e evidências que a geraram.\n\n## 16. Mandatory dossier completeness view\n\nPara cada Rosa/Hana/Mateo/Rita/Broker/Lena/Silas/Cleo, e posteriormente qualquer agente desconhecido, o dossier deve mostrar:\n\n```text\nMEASURED_TRAITS = n\nUNKNOWN_TRAITS = n\nCAUSAL_TRAITS = n\nHELDOUT_SUPPORTED_TRAITS = n\nFALSIFIED_TRAITS = n\nCURRENT_REGIME = ...\nTOP_POSTERIOR_FAMILY = ...\nOOD = ...\nTOP_UNCERTAINTIES = ...\nTOP_NEXT_DISCRIMINATING_PROBES = ...\n```\n\nNão é necessário copiar 190 campos em cada arquivo. Este schema é herdado automaticamente: qualquer trait não citado no dossier permanece `UNKNOWN`.\n\n## 17. Priority experimental order\n\n1. Weak trio para calibration/negative controls.\n2. Rita para threshold, livestock commitment e regression.\n3. Broker para direct awareness + reaction functions.\n4. Lena/Silas/Cleo para slot/funding/temporal mechanisms.\n5. Cross-policy invariance.\n6. History/hysteresis.\n7. Controlled bait/bluff only after detectors pass.\n8. OOD/fresh worlds.\n9. Adaptive student routing only after profiler quality gates.\n\n## 18. Anti-loss rule\n\nEsta taxonomia é **parte do contrato do projeto**, não backlog opcional. Uma rodada local pode focar 1–3 traits, mas não pode redefinir a missão como se os demais deixassem de existir. Nova capacidade descoberta deve ser adicionada aqui antes de ser tratada como conhecimento durável.\n\n\n## Measurement update — 2026-09-08 06:23 UTC\n\nNo taxonomy dimension was removed; universe remains 190 traits and unmeasured traits remain UNKNOWN.\n\nFresh legal-public evidence updates only the evidence state of existing dimensions:\n- `diversification/concentration / crop portfolio timing`: Broker/Lena/Silas = strong same-world descriptive family signature through D12; Cleo = near-match; Rita = negative/intermediate control. Pattern observed: early MELON diversification, gradual STRAWBERRY entry, then STRAWBERRY-heavy D9→D12 while retaining WHEAT/MELON.\n- `capital deployment / land timing`: top family public land expands from 1→2 by D9 and 3 by D12 in sampled worlds; Rita/MIX11 remain at 1 in the control. DESCRIPTIVE, NOT_CAUSAL.\n- `animal portfolio / productive capacity timing`: top family reaches ~11 animals by D9 and 14 by D12; Rita remains 4 through D9 and ~9.7 by D12. DESCRIPTIVE, NOT_CAUSAL.\n- `labor/capacity causal attribution`: bounded `LABOR10_D9_D15` formulation is HELDOUT_FALSIFIED/MATCHUP_UNSTABLE; broader labor timing remains UNKNOWN rather than removed.\n\nHard separation preserved: `LEGAL_PUBLIC_CAPACITY_SIGNATURE != TRANSFERABLE_CAUSAL_POLICY`.\n\n\n## Measurement update — 2026-09-08 07:34 UTC — portfolio timing transfer falsified\n\nTrait universe remains **190**; no dimension added or removed.\n\nEvidence-state changes only:\n- `portfolio_diversification` / `portfolio_concentration` / `portfolio_flexibility`: the upper-family MELON→STRAWBERRY trajectory is independently replicated as legal-public descriptive structure, but a staged single-factor transfer onto MIX11 is HELDOUT_FALSIFIED/HARMFUL.\n- `crop_capacity_investment_propensity`, `capacity_headroom`, `land_utilization`, `asset_utilization`: remain active explanatory candidates/UNKNOWN causal moderators. B54 does not establish them; it strengthens the need to test moderation rather than infer from fingerprint.\n- `product_substitution_matrix`: still UNKNOWN causally. The observed time sequence is not itself a validated substitution response function.\n\nNew hard separation: `REPLICATED_PORTFOLIO_TRAJECTORY != STANDALONE_TRANSFERABLE_PORTFOLIO_CAUSE`.\n",[],[175,176,177,2440],"71dac94c-be59-4d5b-922b-ec08b6579b85","KAG_OFFICIAL_RULES.md","01a07e48-3be6-726a-80bc-016a5de1ff2b",["D",1788825058278],"# Kaggriculture Official Rules / Observability Contract\n\nStatus: CURRENT_OFFICIAL_SOURCE_REVALIDATED_2026-09-07\nPrimary source: Kaggle `kaggle-environments` Kaggriculture spec/engine loaded directly by the project probe. Do not substitute remembered rules when a current official engine read is available.\n\n## Runtime contract — confirmed\n\n- Environment version observed: `0.1.0`.\n- Players: 2.\n- Configured episode horizon: **720 steps** / **30 days × 24 hours**.\n- **Decision-call nuance (official TypeScript engine probe):** `while (!state.done)` produced **719 action decisions**, from step 0 / D0 H0 through step 718 / D29 H22. The next state is step 719 / D29 H23 with `done=true`, so no additional action is requested by that engine loop. Treat `EPISODE_STEPS=720` and `AGENT_DECISION_CALLS=719` as distinct facts until the hosted Kaggle runner is independently traced.\n- `actTimeout`: **1 second**.\n- Starting money: **3000**.\n- Default board: **10 × 10** farm.\n- Maximum market orders processed per player per turn: **10**; extra orders are silently dropped.\n- Shed capacity: **100 non-seed items**; end-of-day overflow is discarded.\n- Default weed spawn chance: `0.005` per empty unlocked tile during end-of-day refresh.\n- Town shop unlock interval: every 3 days; draws are with replacement; repeated shops are possible; unlock stops after 8 instances.\n- Town shop consumption interval: every 4 turns; each unlocked shop instance consumes its products, duplicated shops consume once per copy.\n- Town center interval: every 24 turns; buys one of every non-fertilizer product per tick at the default configuration.\n- Daily farm-hand hire cost follows Fibonacci sequence × `farmHandCostMult`.\n- Input seed is hidden from the agent observation; the resolved seed is retained by the environment/replay.\n\n## Observation contract — confirmed by runtime probe\n\nTop-level observation fields:\n`player, step, day, hour, numAgents, farms, private, market, town`.\n\nPublic farm fields include for both farms:\n`money, tiles, farmer, hands, unlocked_quadrants, hires_today`.\n\nPrivate own-player fields:\n`shed, seeds, inventories`.\n\nShared market state includes shared inventory/prices. Town state is observable.\n\n### Consequence\n\nOpponent public farm state is available, but opponent private shed/seeds/inventories are not. Any opponent model that claims direct knowledge of private inventory is invalid unless inferred probabilistically from public actions/state transitions.\n\n## Core economic/game facts already established from official engine + project calibration\n\n- Winner of an episode is determined by final bank cash.\n- Terminal cash margin is useful diagnostic evidence but does not itself replace W/L as the competitive selection objective.\n- Market inventory/prices are shared; one player's orders can alter the price/supply environment encountered by the other.\n- Town demand removes products from shared market inventory and therefore creates endogenous scarcity pressure.\n- Land expansion is ordered/costed and should be treated as irreversible capacity investment during the finite horizon.\n- End-of-horizon unsold/private inventory has liquidation risk because the win condition is bank cash.\n\n## Action-order caution\n\n`maxMarketOrdersPerTurn=10` means an agent can generate more requested orders than the interpreter will process. The executor must explicitly prioritize order budget rather than assuming all emitted orders execute. Earlier project work already found `MARKET_ORDER_BUDGET_CONTENTION`; this remains a regression hazard.\n\n## Observability matrix for strategic intelligence\n\n### Directly observable\n- own + opponent cash;\n- own + opponent public tiles/crop/animal layouts;\n- own + opponent hands/farmer positions;\n- own + opponent unlocked land;\n- shared market prices/inventory;\n- town/shop state;\n- own private inventories/shed/seeds;\n- temporal public-state changes derived from successive observations.\n\n### Not directly observable\n- opponent shed;\n- opponent seed inventory;\n- opponent unit inventories/private carried items;\n- opponent internal policy, goals, beliefs, posterior, planned future actions;\n- environment seed in agent observation.\n\n### Inferable only with uncertainty\n- opponent production pipeline;\n- likely future sales;\n- latent strategy family;\n- response elasticity to our interventions;\n- commitment/hysteresis;\n- regime changes;\n- opponent awareness of our state;\n- private inventory bounds from conservation/state transitions.\n\n## Anti-false-positive rules\n\n1. `PUBLIC_STATE_CHANGE != OPPONENT_REACTION_TO_US`.\n2. `MARKET_MOVE_AFTER_OUR_ACTION != CAUSAL_RESPONSE` without paired control/fixed-prestate evidence.\n3. `SAME_SEED != IDENTICAL_PRESTATE` after policies have diverged.\n4. `PRIVATE_STATE_ESTIMATE != OBSERVED_PRIVATE_STATE`.\n5. `TREATMENT_REQUESTED != TREATMENT_APPLIED`; intervention detector required.\n6. `HTTP/FETCH_PASS != GAME_PARITY`; official engine/version provenance must be recorded.\n7. Any new game-mechanic claim must be linked to current official spec/engine or marked `UNVERIFIED`.\n\n## Runtime implication\n\nThe official hard boundary is 1 s per action. The strategy architecture must reserve substantial safety headroom for Python/Kaggle/runtime variance; targeting 1.000 s is not acceptable engineering. Exact internal p99/fallback budgets are implementation recommendations and must be measured before promotion.",[],[175,176,177,2448],"fb311e06-7704-4eeb-a790-809a8c164fed","KAG_NEGATIVE_KNOWLEDGE.md","dea3253e-ab1c-11f1-bbd8-1607ee4eb77e",769,["D",1788871484649],"# Kaggriculture Negative Knowledge\n\nEvidence policy: rejected/heldout-falsified mechanisms are hard gates. A single favorable directional batch does not reopen them.\n\n## Hard reopening gate\nA hypothesis marked `HELDOUT_FALSIFIED`, `REJECTED`, `HARMFUL`, `MATCHUP_UNSTABLE` or equivalent may be reopened only if all are present:\n1. explicit new mechanism explaining the prior failure;\n2. materially different measurable context/regime;\n3. preregistered discriminating test;\n4. at least two favorable fresh paired heldouts or statistically equivalent evidence;\n5. no severe regression damage.\n\n## PASTURE_RAMP\nState: `HELDOUT_FALSIFIED / MATCHUP_UNSTABLE / DO_NOT_PROMOTE`.\nPaired Broker evidence versus MIX11:\n- B31: delta -20600.625, both 0-8.\n- B32: delta +2102.5, both 0-8.\n- B33: delta -6321.0, both 0-8.\nConclusion: the B32 directional gain is unstable and does not reopen the branch. No adjacent pasture/build-ramp mutation without satisfying the full reopening gate.\n\n## Other preserved rejected formulations\n- LAND_AFTER6_CLEAN — rejected/heldout-falsified.\n- PORTFOLIO_D6_244 — rejected/heldout-falsified.\n- EARLY_MELON40 — rejected in tested formulation.\n- EARLY_HERD3 / EARLY_HERD4 / EARLY_HERD5 — rejected in tested formulations.\n- EARLY_BUILD4 — rejected in tested formulation.\n- EARLY herd+build combinations — rejected in tested formulations.\n- static 10/12 hands universal — rejected as universal rule.\n- early-land universal — rejected as universal rule.\n- excessive prefeed — rejected.\n- M70 universal — rejected.\n- direct rival-damage as primary universal mechanism — rejected in tested formulation.\n- strawberry 65/35 signal — falsified.\n\n## ADAPT_V1 — scope-correct rejection\nState: `REJECTED_IN_TESTED_FORMULATION_ONLY`.\n\n`ADAPT_V1` mapped a small visible-opponent snapshot (especially crop concentration/cash-gap heuristics) directly into our crop-share choice. It remained 0-win in the tested Broker batch. This rejects **that uncalibrated snapshot crop-router**, not opponent modeling, OBG, active probing or conditional adaptation as a class.\n\nHard interpretation rule:\n`ADAPT_V1_FAILURE != ADAPTIVE_OPPONENT_MODELING_FAILURE`.\n\n## Methodological negative knowledge from full 15-minute audit\n\nThe 2026-09-08 full reanalysis of 208 attempts / 1,272 DONE-game episodes establishes the following hard epistemic guards:\n\n- `RAW_CROSS_BATCH_MARGIN_DIFFERENCE != TREATMENT_EFFECT`. World/seed/regime variation is too large for unpaired margin attribution.\n- `PAIRED_CONTROL_INCOMPLETE => DO_NOT_INTERPRET_TREATMENT_DIRECTION`. PASTURE B31 is the concrete failure case: provisional direction reversed after control persistence recovery.\n- `LADDER_STRENGTH != MATCHUP_EFFECT_AGAINST_US`. Direct pairwise superiority is useful calibration, not a scalar predictor of the gap against our policy.\n- `OUR_POLICY_SHIFT_ROBUSTNESS_OF_OUTCOME != OPPONENT_ADAPTATION`. Broker defeating a broad basket of our variants does not prove it reacted online to those variants.\n- `PAIRWISE_STRENGTH_CALIBRATION != REACTION_MODEL`.\n- `LARGE_EPISODE_COUNT != LARGE_TRAIT_COVERAGE`; most reaction/adaptation traits can remain UNKNOWN despite many games.\n- infrastructure/schema/persistence failures are never competitive losses.\n\nThese are not hypotheses to reopen by ordinary parameter tuning; they are analysis-method constraints. To override one requires an explicit experimental design that directly tests the disputed equivalence.\n\n## Detector rule\nBefore queuing a mutation, search this document, `KAG_15MIN_FULL_STRATEGIC_REANALYSIS_20260908.md`, and the latest hourly checkpoint. If the parent mechanism is rejected, classify the proposed change as either `NEW_MECHANISM_WITH_REOPENING_EVIDENCE` or `BLIND_ADJACENT_MUTATION`. The latter must not enter the compute queue.\n\n\n## 2026-09-08 06:07 UTC — redundant mutation guard: SELL_FIRST\n\n### `SELL_FIRST_AS_MISSING_MIX11_CAPABILITY`\nStatus: `FALSIFIED_AS_ABSENCE / REDUNDANT_MUTATION / DO_NOT_BLINDLY_IMPLEMENT`.\n\nEvidence:\n- V12/V13 establish that SELL-before-spend is causally useful when starting cash is binding.\n- V14 then measured MIX11's **own** requested order directly using allowed self-action memory: 221/221 mixed SELL+spend turns vs Broker and 210/210 vs Cleo were already SELL-first; spend-first = 0. Mixed queue saturation = 0.\n\nCause of possible false lead: discovering a real opponent mechanism does not imply our agent lacks it (`MECHANISM_REAL != CAPABILITY_GAP`).\n\nControl: before porting any observed high-tier mechanism, run an own-agent capability census in matched contexts.\nDetector: if proposed mutation is behaviorally identical to baseline on the treatment detector, classify `REDUNDANT_ALREADY_PRESENT` before competitive enqueue.\nRegression: preserve SELL-before-spend ordering when mixed queues occur, especially under low cash; do not claim it explains the current top-tier gap by absence.\n\n\n## TEMPORALLY_BOUNDED_LABOR10_D9_D15 — HELDOUT_FALSIFIED / MATCHUP_UNSTABLE / DO_NOT_PROMOTE\n\nPreregistered experiment: `artifacts/kaggriculture/experiments/KAG_EXP_LABOR_D9_D15_CAUSAL_V1.md`.\n\nFresh paired Broker test on seeds `[108301,108313,108327]`, mirrored seats, explicit treatment detector:\n- MIX11 baseline: 0-6, avg margin -88401.67.\n- LABOR10_D9_15: 0-6, avg margin -87730.50.\n- paired avg delta: +671.17, but only 2/6 positive and 4/6 negative.\n- the two large positive deltas came from the same seed/world family (108301 mirrored seats: +20576/+19451); the other two seed families were negative.\n- D15 cash-gap versus Broker worsened under treatment in every fresh seed family; D18 improved only for 108301 and worsened for 108313/108327.\n- treatment detector: baseline laborApplied=0 in all 6 episodes; treatment laborApplied=6 in all 6; errors=0.\n\nInterpretation: the positive scalar average is dominated by one favorable world and does not establish a transferable labor bottleneck. W/L did not move. Per preregistration, the bounded D9→D15 formulation fails its first discriminant and is closed rather than sent automatically to a second heldout.\n\nThis does **not** erase the broader UNKNOWN question of labor/capacity timing. It specifically blocks blind adjacent `LABOR10_D9_15` mutations and prevents the old rejected static-hands family from being reopened by a margin-only artifact.\n\nReopening gate: require a genuinely new mechanism/context distinction, preregistered treatment detector, and the canonical >=2 favorable fresh paired heldouts/equivalent without severe regression.\n\n\n## STAGED_STRAW_D6_D12 — HELDOUT_FALSIFIED / HARMFUL / DO_NOT_PROMOTE — 2026-09-08\n\nPreregistered experiment: `artifacts/kaggriculture/experiments/KAG_EXP_STAGED_PORTFOLIO_D6_D12_V1.md`.\n\nFresh paired B54 `[108701,108713,108727]`, mirrored seats, sidecar core `KAG_AUTOLAB_CORE_V24_STAGED_PORTFOLIO_EXPERIMENT`:\n- Broker MIX11 control: 0-6, avg margin -94384.00.\n- Broker treatment: 0-6, -113295.17.\n- paired treatment-control delta: **-18911.17**, positive 0/6, negative 6/6, range -32975..-1965.\n- Rita MIX11 control: 6-0, +8988.83.\n- Rita treatment: 4-2, -585.50.\n- paired delta Rita: **-9574.33**, positive 0/6, negative 6/6.\n- errors: 0.\n\nTreatment detector passed rather than producing a requested-only false negative:\n- `stagedPortfolioCalls=3` in every treatment episode;\n- D6 public snapshot remained WHEAT13 because the D6 action had not yet propagated into the checkpoint;\n- by D9 treatment was materially present (typically W7/M3, some worlds W7/M5);\n- by D12 most worlds were W7/M4/S1, with seed-specific lag in one family;\n- by D15 the portfolio had rotated strongly to roughly W1/M5/S6 or M6/S6.\n\nBroker trajectory counterevidence is coherent: versus paired MIX11 the treatment gap was worse in 6/6 at D9, D12, D15, D21, D24, D27 and D29; mean terminal harm compounded rather than recovering late. Rita showed transient local D12/D15 improvement in some episodes but all six terminal outcomes were economically worse versus paired control and W/L regressed.\n\nInterpretation: the replicated public upper-family MELON→STRAWBERRY composition is **not independently transferable on the MIX11 capacity substrate** in this staged formulation. This strengthens the alternative that crop composition is coupled to an upstream capacity/throughput/funding substrate rather than being a standalone causal recipe.\n\nThis result does not reopen `PORTFOLIO_D6_244`, `EARLY_MELON40`, `PHASE3`, early-land/herd or static-hands families. It adds a new closed formulation. Do not sweep adjacent stage shares blindly.\n\nReopening requires the canonical hard gate plus a genuinely new measurable mechanism explaining why the same temporal crop transition would become beneficial in a different substrate/regime.\n\n\n## CAPACITY_HEADROOM_MODERATES_STAGED_PORTFOLIO — REJECTED_IN_B55_FORMULATION — 2026-09-08\n\nPreregistered 2x2 factorial: `KAG_EXP_CAPACITY_PORTFOLIO_INTERACTION_V1.md`, fresh Broker seeds `[108801,108813,108827]`, mirrored seats, V25 mechanical gate PASS with equal D6 pretreatment hashes and treatment-fired land/portfolio divergence.\n\nAggregate margins: A MIX11 -113730.17; B STAGED -110910.33; C CAP2_D6 -96016.00; D CAP2_D6_STAGED -113869.83; all arms 0-6. Portfolio effect at baseline = +2819.83; headroom-only = +17714.17; portfolio effect under headroom = -17853.83; interaction `I=(D-C)-(B-A)=-20673.67`.\n\nTherefore the proposed positive moderation mechanism is rejected in this formulation. Additional headroom did **not** rescue the staged portfolio; it reversed the same-epoch portfolio contrast strongly negative under headroom. `STAGED_STRAW_D6_D12` remains closed and no adjacent crop×land factorial sweep is allowed.\n\n`CAP2_D6` itself is not promoted: C-A is one favorable same-epoch diagnostic batch with unchanged 0-6 W/L and is adjacent to old early-land negative knowledge. It is classified `ONE_BATCH_REOPENING_CANDIDATE / NOT_REOPENED`. A dedicated preregistered B56 fresh Broker+Rita gate is required before any reopening claim.\n\n\n## CAP2_D6 — FRESH_HELDOUT_FALSIFIED / DO_NOT_REOPEN — 2026-09-08\n\nB55 produced a one-batch diagnostic Broker margin improvement for `CAP2_D6` (+17714.17 vs same-epoch MIX11, both 0-6), which was explicitly classified `ONE_BATCH_REOPENING_CANDIDATE / NOT_REOPENED` because it is adjacent to rejected early-land formulations.\n\nThe preregistered B56 hard gate used fresh seeds `[108901,108913,108927]`, mirrored seats. Broker control MIX11 = 0-6 / -116285.00; CAP2_D6 = 0-6 / -128129.33. Fresh paired aggregate delta = **-11844.33** in the harmful direction, with zero W/L improvement and zero runtime errors. The required fresh Broker directional replication therefore failed before the Rita arm was needed to establish reopening.\n\nClassification: `CAP2_D6 = FRESH_HELDOUT_FALSIFIED / DO_NOT_REOPEN / DO_NOT_PROMOTE` in this formulation. This also closes `LIMITED_D6_CAPACITY_HEADROOM_REOPENING_GATE`; do not sweep adjacent land timing/targets. Broader capacity timing remains UNKNOWN and may only be revisited under the canonical hard reopening gate with a genuinely new mechanism/context.\n\n\n## B56 RITA REGRESSION CLOSURE — CAP2_D6 SEVERE REGRESSION — 2026-09-08 09:15 UTC\n\nAfter canonical preflight PASS, the already-preregistered B56 Rita arms completed on fresh seeds `[108901,108913,108927]`, mirrored seats. MIX11 = **6-0 / +9630.50**; CAP2_D6 = **0-6 / -9005.00**, errors 0. The treatment therefore flips the regression matchup from 6-0 to 0-6 and changes mean margin by **-18635.50**.\n\nThis is independent reinforcement of the Broker falsification and closes the formulation more strongly: `CAP2_D6 = FRESH_HELDOUT_FALSIFIED / SEVERE_REGRESSION_HARM / DO_NOT_REOPEN / DO_NOT_PROMOTE`. It cannot satisfy the canonical reopening gate because the no-severe-regression condition fails decisively. No adjacent land/capacity sweep is authorized.\n\n\n## LABOR9_D9_D15 — POST-PREFLIGHT RECONCILED FRESH NEGATIVE — 2026-09-08\n\nAn adjacent hands-only screen was executed before the continuation round had completed the fresh canonical preflight. This is recorded as an ordering-process breach, not hidden. After full rehydration/reconciliation, its evidence can only strengthen the already-closed bounded-labor family; it cannot reopen strategy search.\n\nFresh Broker seeds `[106293,106307,106321,106333]`, mirrored seats, treatment detector verified exactly 6 active days/game and zero execution errors:\n- paired MIX11: 0-8, avg margin -104353.875;\n- `LABOR9_D9_15`: 0-8, -111207.5; paired mean delta **-6853.625**; positive 0/8, negative 8/8;\n- repeat `LABOR10_D9_15`: 0-8, -113096; paired mean delta **-8742.125**; positive 0/8, negative 8/8.\n\nClassification: `LABOR9_D9_D15 = FRESH_PAIRED_HARMFUL_IN_TESTED_FORMULATION / DO_NOT_PROMOTE`; `LABOR10_D9_D15` receives additional independent negative reinforcement. The broader scientific question of labor timing remains UNKNOWN, but **adjacent target sweeps are blocked**. Reopening requires a genuinely new mechanism/context satisfying the canonical hard reopening gate, not another 8/9/10/12-hands scan.\n\n\n## PERSISTENT_ROUTING_AS_FEED_CONTINUITY_FIX — FRESH_PAIRED_MECHANISM_FALSIFIED / DO_NOT_PROMOTE\nPreregistered fresh Broker test `[109501,109513,109527,109539]`, mirrored seats, isolated only `persistent_routing=True`; runtime detector proved treatment false for MIX11 and true for all 8 treatment episodes. MIX11 0-8/-102722.125; route-persist 0-8/-99815.625. Although terminal margin moved +2906.5 diagnostically, the preregistered mechanism worsened rather than improved: D9-D15 H0 `consecutive_unfed` sum 22→28; D18-D27 0→2. No animal H0 drops occurred in either arm in the audited windows. Therefore do not sweep persistence/route-memory parameters adjacent to this formulation. `MARGIN_IMPROVEMENT != MECHANISM_PASS != COMPETITIVE_BREAKTHROUGH`.\n\nRelated narrowing: fresh natural MIX11 census showed min own WHEAT 5 (D9-D15) and 12 (D18-D27), zero WHEAT<=2 rows, plus active WHEAT buys. `SIMPLE_WHEAT_STOCK_SHORTAGE_AS_MISSING_CAPABILITY = NOT_SUPPORTED_IN_TESTED_WINDOWS`; this does not prove universal feed/logistics sufficiency.\n\n## TASK_SERVICE_BACKLOG_AS_BOUNDARY_SPECIFIC_CAUSE — NOT_SUPPORTED / MATCHUP_NONSPECIFIC\nFresh natural control `[109701,109713,109727,109739]`, mirrored, compared unchanged MIX11 against Broker (0-8/-92701.375) and Rita (7-1/+5053.125). Despite the outcome split, our harvest backlog, water stress, CARE backlog, feed stress, pickup cadence, shed level and requested sell volume were broadly comparable, with several service metrics actually worse under the winning Rita regime. Therefore general task-service backlog is real but is not supported as a Broker-specific boundary explanation in this formulation. Do not launch blind service/routing/labor fixes from this census; require a differential conversion mechanism first.",[],[175,176,177,2455],"c3771a65-f8ac-4c18-a278-677b1ffaedf1","KAG_EXPERIMENT_MATRIX_OPPONENT_PROBES.md","01a07e48-3be6-726a-80bb-fc946c4aa6b1",632,["D",1788839067019],"# Kaggriculture Opponent-Profiling Experiment Matrix\n\nStatus: V1_TELEMETRY_FIRST\nRule: no behavior-changing adaptive policy promotion until the measurement layer passes positive + negative controls.\n\n## Phase P0 — instrumentation equivalence\n\nPurpose: prove telemetry itself does not alter MIX11.\n\nFor each weak opponent Rosa/Hana/Mateo:\n- same seeds, both seats;\n- baseline core vs telemetry-enabled core;\n- compare exact emitted our actions, opponent actions, terminal result, checkpoint state and RNG-relevant state where accessible;\n- required: exact behavior equivalence except telemetry metadata.\n\nMutation control: intentionally change one emitted action in a test-only fixture; equivalence validator must fail.\n\n## Phase P1 — baseline temporal signatures\n\nCapture at turn resolution or compact event windows:\n- both public farms;\n- own private state only;\n- shared market/town;\n- both agents' laboratory action outputs;\n- executed market order effects where interpreter semantics permit;\n- checkpoint profiles D0/D1/D3/D6/D9/D12/D15/D18/D21/D24/D27/D29;\n- delta vectors and regime-change alarms.\n\nOutput: immutable event log + profile features, not a personality narrative.\n\n## Phase P2 — calibration probes on known weak reference source\n\n### Rosa — negative direct-awareness control\nSource has no other-farm dependency and no explicit price floor.\n- Probe R-A: modify our visible expansion while matching market/town/prestate as tightly as possible.\n Expected source-ground-truth: no direct opponent policy change.\n- Probe R-M: price/inventory perturbation on staple goods.\n Expected: no sharp explicit threshold comparable to Hana/Mateo.\n\n### Hana — threshold positive control\nKnown carrot floor 15, tomato floor 25, sell chunk 30.\n- H-CARROT: fixed-prestate price sweep `{floor-δ, floor, floor+δ}`.\n- H-TOMATO: same.\n- H-AWARE: our-farm perturbation with shared market matched.\nExpected: market response yes; direct other-farm response no.\n\n### Mateo — strong threshold positive control\nKnown melon floor 120, chunk 12.\n- M-MELON: fixed-prestate melon price sweep around 120.\n- M-QUANTITY: abundant private/output condition to detect ≤12 sale chunk.\n- M-AWARE: our-farm perturbation with market matched.\nExpected: sharp market threshold + chunk cap; direct other-farm response no.\n\n## Phase P3 — reaction model estimators\n\nOnly after P2 passes.\n\nFor response feature `y` and treatment/context `x`:\n- paired mean/median deltas;\n- bootstrap/Wilson interval as appropriate;\n- RLS coefficient + covariance;\n- lagged response profile;\n- CUSUM calibrated false-alarm rate;\n- SPRT early-stopping decision;\n- Bayes posterior over competing mechanisms.\n\nReport effect magnitude, uncertainty, world context and counterexamples. `p-value alone != mechanism`.\n\n## Phase P4 — history sensitivity / hysteresis\n\nConstruct two histories that converge to the same observable current state as closely as engine constraints permit, then compare future opponent actions under common future forks.\n\nQuestions:\n- same current state, different prior market path -> same action?\n- after a threshold shock reverses -> immediate return or hysteresis?\n- after our visible lead disappears -> response persists?\n\nIf private opponent state cannot be equalized/observed, mark residual confounding explicitly.\n\n## Phase P5 — OOD\n\nVary official world conditions using controlled seeds/prefixes:\n- unusual duplicated shops;\n- scarcity/glut extremes;\n- atypical early price paths;\n- delayed/accelerated expansion contexts.\n\nMeasure:\n- classifier likelihood under all known families;\n- posterior entropy;\n- reaction-model residuals;\n- policy fallback rate;\n- stability of weak-trio ground-truth classifications.\n\n## Phase P6 — move to Rita\n\nValidate:\n- wheat buy threshold around source-described 55;\n- livestock build commitment/hysteresis;\n- direct-awareness remains UNKNOWN until isolated;\n- regression sensitivity for candidate changes.\n\n## Phase P7 — Broker direct-awareness positive control\n\nSource-ground-truth says wheat timing depends on opponent cash.\n\nTarget structural model:\n`P(BUY_WHEAT_t=1) = σ(β0 + β1*our_cash + β2*wheat_price + β3*market_inventory + β4*day + β5*own_cash + interactions)`.\n\nDiscriminating forks:\n- vary our visible cash while holding wheat market state;\n- vary wheat price while holding our cash;\n- vary market inventory while holding both;\n- test interactions by regime/day.\n\nNeed independent heldout validation before using the inferred reaction as an exploit.\n\n### P7 status update — real-scenario activation layer\nThe exact direct-cash threshold is now a passed scenario-stratified positive control, but its **activation is sparse/contextual** in the first broader screen: 86/1,440 sampled natural prestates changed Broker action under a fixed 199→200 opponent-cash switch, versus 0/1,440 on Rosa/Hana/Mateo direct-cash negative controls. Only 40/86 changed the selected immediate next-state cash/WHEAT digest.\n\nTherefore P7 now splits into two estimands:\n1. `P(mechanism active | legal context)` — classifier/activation model;\n2. `E(economic consequence | mechanism active, context)` — effect/lag model.\n\nDo not collapse them into one awareness scalar. Candidate moderators such as phase, shop duplication, price/inventory and order-slot pressure require controlled follow-up; raw stratified counts are not causal moderator evidence.\n\n## Phase P8 — Lena/Silas/Cleo slot-order mechanism\n\nBecause static early farm state aliases these top agents, instrument:\n- requested market orders by slot;\n- executed/retained orders after 10-order cap;\n- buy orders displaced by sells;\n- premium/staple ordering;\n- liquidity before/after slot execution.\n\nFixed-prestate counterfactuals compare ordering policies rather than blindly mutating crop share.\n\n## Metrics dashboard\n\nPer opponent/hypothesis:\n- posterior and entropy;\n- calibration Brier/log score where truth labels are available;\n- direct-awareness score;\n- market-reactivity score;\n- threshold estimates + CI;\n- reaction lag;\n- RLS coefficients + covariance;\n- CUSUM alarms / false alarms;\n- SPRT samples-to-decision;\n- adaptation/hysteresis duration;\n- OOD score;\n- probe economic cost;\n- EVSI/net probe value;\n- W/L and terminal margin only as outcome/regression metrics.\n\n## Promotion gates\n\nProfiler component may advance only if:\n1. positive control detected;\n2. negative control stays negative;\n3. telemetry equivalence passes;\n4. fresh heldout worlds reproduce calibration;\n5. mirrored-seat artifact exists;\n6. treatment detector proves perturbation fired;\n7. current official engine/version provenance is recorded;\n8. latency impact is measured;\n9. failure detector survives a deliberate mutation.\n\nAdaptive strategy route may advance only after profiler gate + independent evidence that routing improves W/L/robustness beyond baseline without hidden regression.",[],[175,176,177,2463],"2d431d5f-2f4d-416a-bcee-156d564d6985","KAG_CURRENT_STATE.md","0e287a8c-ab2e-11f1-8008-1607ee4eb77e",837,["D",1788916971521],"# KAGGRICULTURE — Current Strategic State\n\nStatus: CANONICAL_CURRENT_STATE_V1\nGenerated from canonical artifacts + live sidecar reconciliation on 2026-09-08 UTC / 2026-09-07 BRT boundary.\n\n## 1. Mission state\n\nThe project is **not** a W/L-only optimizer. Primary research direction is opponent intelligence:\n\n`LEGAL OBSERVATION → IDENTITY/REGIME BELIEF → REACTION MODEL → CONTROLLED PROBE → CAUSAL EFFECT → NET VALUE → CONDITIONAL POLICY ROUTING`\n\nThe full mandatory trait universe lives in `KAG_OPPONENT_CAPABILITY_TAXONOMY.md` and must not be silently narrowed by the current experiment.\n\n## 2. Competitive baseline\n\n- Proven/champion baseline: **unchanged**; no new champion promoted in the recovery/profiling work.\n- Working lineage/baseline: **MIX11**.\n- Main boundary: **Broker Bea**.\n- Stress/upper ladder: **Lena / Silas / Cleo**, with Cleo current stress ceiling.\n- Intermediate regression: **Rancher Rita**.\n- Cheap calibration/regression: **Rosa, Hana, Mateo**.\n- Weak trio evidence: MIX11 8-0 each in B21 sampled worlds; regression evidence, not universal proof.\n\n## 3. Sidecar runtime — live verified\n\n`cleitoncosta/kaggriculture-autolab@main`:\n\n- core: `KAG_AUTOLAB_CORE_V22_LEGAL_TELEMETRY_SEPARATION`;\n- worker: `KAG_AUTOLAB_WORKER_V2_COMPACT_PERSISTENCE`;\n- full large results are stored in Blob; SQLite stores compact index/summary when large;\n- latest verified successful large legal job persisted with `BLOB_FULL_SQL_COMPACT`;\n- B34 Broker/Lena legal jobs previously blocked by libSQL/hrana were recovered from Blob and marked DONE;\n- final full-audit queue snapshot: 171 DONE rows, 5 FAILED terminal historical/self-test/error rows, no READY/RUNNING jobs;\n- worker/core compute evidence remains `NOT_CHAMPION_PROOF`.\n\nExperimental branches:\n- `opponent-intel-v1`: core V23 `LEGAL_SELF_MEMORY`; records our own market-flow contribution for self-normalization. **Not canonical/promoted.**\n- `b34-analysis-v1`: deeper public-state/oracle root-cause experiments and counterfactual patch tooling. **Research branch only.**\n- `persistence-v2`: persistence/retry engineering branch whose compact-persistence worker changes are now represented on main.\n\n## 4. Opponent profiling stack — what is already established\n\n### Calibrated controls\n\n- Rosa: no explicit carrot floor in tested prestate; no direct opponent-cash response in calibration.\n- Hana: carrot floor 15 and tomato floor 25 recovered; direct opponent-cash negative control in tested prestates.\n- Mateo: melon floor 120 and sell cap 12 recovered; direct opponent-cash negative control.\n- Broker: direct opponent-cash threshold around 200 causally recovered in fixed prestate.\n- Broker threshold immediate economic effect was small in first impact screen: 6/33 treated states affected controlled purchase cost; max observed difference 4 coins for 5 wheat. Therefore `AWARENESS_REAL != PRIMARY_BOUNDARY_CAUSE`.\n- Broker/Lena/Silas share a related early cash-response signature; Cleo's analogous effect appears later under sell-slot competition.\n\n### Legal classifier\n\nEarlier reduced legal observer achieved strong same-policy top-4 classification and fresh heldout success, but later cross-policy tests showed Broker↔Lena confusion when our own policy changed.\n\nTherefore:\n`SAME_POLICY_HELDOUT_ACCURACY != POLICY_INVARIANCE`.\n\nV23's self-action normalization is a promising correction because our own market contribution is legally known to us, but it remains experimental pending broader validation.\n\n## 5. Important correction: reduced observer hid earlier divergence\n\nBefore the 2026-09-08 continuity repair, sidecar SQLite `hourly:current_hypothesis` still stated a D12H2 Broker↔Lena landmark. That state has now been superseded by the canonical context-reconciliation gate. The old D12H2 statement remains **historical/stale as a claim about first full-public divergence**.\n\nFresh `b34-analysis-v1` experiments compared the full public state and found:\n\n- in 4 fresh seeds × 2 seats = 8 paired worlds, Broker/Lena full public state was equal immediately before D10H4;\n- their own/reference private diagnostic state was also equal in the offline oracle at that point;\n- our emitted action was equal;\n- reference farmer action and market action were equal;\n- offline oracle action difference: Broker produced 12 hand commands versus Lena 11; the additional Broker command was `EAST` in all 8 pairs;\n- the next legal full-public state at **D10H5** diverged in all 8 pairs.\n\nClassification:\n- `D10H4 HAND DIFFERENCE = STRONG OFFLINE MECHANISM/FINGERPRINT CANDIDATE`;\n- `D10H5 FULL PUBLIC DIVERGENCE = LEGAL OBSERVABLE SIGNATURE`;\n- `D10H4 EXTRA EAST CAUSES MATERIAL TERMINAL ADVANTAGE = UNVERIFIED`.\n\nDo not confuse first divergence with value mechanism. A single-step counterfactual swap is prepared on the experimental branch but has not yet been promoted as evidence.\n\nAlso note: public reference `_TRACE`/raw action is **offline oracle only**. It cannot become an online feature. Legal student use must rely on public-state consequences and our own remembered actions.\n\n## 6. Negative knowledge — hard preserved\n\n`PASTURE_RAMP = HELDOUT_FALSIFIED / MATCHUP_UNSTABLE / DO_NOT_PROMOTE`.\nPaired Broker deltas vs MIX11:\n- B31 -20600.625;\n- B32 +2102.5;\n- B33 -6321;\n- all 0-8.\n\nAlso preserve rejections recorded in `KAG_NEGATIVE_KNOWLEDGE.md`, including LAND_AFTER6_CLEAN, PORTFOLIO_D6_244, EARLY_MELON40, EARLY_HERD3/4/5, EARLY_BUILD4 and tested early herd+build combinations.\n\nA favorable isolated batch cannot reopen them.\n\n## 7. What remains incomplete\n\nOpponent fingerprinting is ahead of opponent adaptation.\n\nCurrent rough maturity by capability:\n\n- game rules/provenance: strong;\n- telemetry separation: strong;\n- weak-agent calibration: strong;\n- top-family fingerprinting: strong same-policy, partial cross-policy;\n- direct-awareness causal calibration: partial/strong for selected mechanisms;\n- general reaction functions to **our interventions**: early;\n- advance/retreat/investor/risk profiles: schema now canonical, measurement incomplete;\n- adaptation latency/hysteresis: incomplete;\n- history/path sensitivity: incomplete;\n- substitution matrix: incomplete;\n- controlled bait/bluff: not yet validated;\n- Level-k/QRE/revealed-preference models: architecture/hypotheses, not operationally validated;\n- OOD/confidence calibration: partial;\n- adaptive online student <1s using OBG: not promoted.\n\n## 8. Priority order from here\n\n1. Run `00_READ_FIRST.md` preflight before anything else.\n2. Preserve the 190-trait taxonomy as the global profiling contract.\n3. Continue weak-trio/Rita positive-negative controls for reaction detectors where gaps remain.\n4. Separate direct reaction to us from market-mediated reaction.\n5. Test reaction magnitude + latency + hysteresis with controlled interventions.\n6. Only then progress to bait/bluff/history-sensitive experiments.\n7. Use fingerprinting work (including D10H4/D10H5) as supporting identity/mechanism research, not as replacement for reaction profiling.\n8. Connect adaptive student routing only after profiler calibration, cross-policy/OOD tests and latency gates.\n\n## 9. Current contradiction register\n\n### C1 — stale first-divergence hypothesis\n- stale state: D12H2 is first Broker↔Lena legal divergence in reduced observer.\n- newer evidence: full public state diverges D10H5 after D10H4 action branch in 8/8 fresh pairs.\n- reconciliation: D12 remains a reduced-feature landmark; it is not first full-public divergence.\n\n### C2 — raw action/oracle vs legal online feature\n- oracle action traces are useful for teacher/root-cause analysis.\n- student cannot use raw opponent actions or private state.\n- legal online signal must be reconstructed from public state + shared market + our own action memory.\n\n### C3 — profile plan vs measured coverage\n- strategic docs already defined many dimensions, but dossiers currently measure only a small subset.\n- new canonical taxonomy makes all unmeasured dimensions explicitly `UNKNOWN` instead of allowing them to disappear from scope.\n\n## 10. Full 15-minute historical reanalysis — material update\n\nCanonical audit: `KAG_15MIN_FULL_STRATEGIC_REANALYSIS_20260908.md`.\n\nForensic census covered every persisted compute-active 15-minute window in the sidecar through the audit cutoff: **208 run attempts, 171 DONE, 37 FAILED, 176 distinct jobs, 55 active 15-minute buckets and 1,272 game episodes** (782 our-vs-reference + 490 public crossplay). Historical empty/no-op scheduled ticks are not fully reconstructible because they were not append-logged.\n\nMaterial strategic conclusions:\n\n- MIX11 vs Broker aggregates 0-116 over persisted non-selftest batches; M60_FERT also remains 0-92. This is strong outcome-level robustness to our tested policy basket, **not proof of Broker online adaptation**.\n- MIX11 vs Rita aggregates 67-31 over 98 episodes with substantial batch variability; Rita remains matchup/regression evidence, not a universal proxy.\n- public ladder ordering is strong, but B38 shows that direct Silas>Lena superiority can almost disappear in the third-party M60_FERT matchup. New rule: `LADDER_STRENGTH != MATCHUP_EFFECT_AGAINST_US`.\n- world/seed/regime effects make unpaired cross-batch margin interpretation unsafe: `RAW_CROSS_BATCH_MARGIN_DIFFERENCE != TREATMENT_EFFECT`.\n- PASTURE B31 permanently establishes `PAIRED_CONTROL_INCOMPLETE => DO_NOT_INTERPRET_TREATMENT_DIRECTION`.\n- `ADAPT_V1` rejects its crude uncalibrated snapshot crop-router only; it does not reject adaptive opponent modeling.\n- the majority of historical compute targeted our own macro mutations and pairwise strength/fingerprinting. Most of the original 190-trait reaction/adaptation OBG remains UNKNOWN and should receive the next research budget.\n\nCurrent research allocation therefore changes from broad adjacent mutation search toward **causal opponent-reaction profiling**: legal turn-level traces → controlled observable probes → direct-vs-market-mediated attribution → reaction magnitude/lag → hysteresis/history → risk/advance-retreat/substitution → bait/bluff only after calibrated controls → OOD/cross-policy → student routing.\n\nA new mandatory scenario contract is active: `KAG_REAL_SCENARIO_COVERAGE_V1_20260908.md`. Experimental branch `kaggriculture-autolab@reaction-attribution-v1` now has core `KAG_AUTOLAB_CORE_V24_REAL_SCENARIO_TRACE`, which passed non-regression on Rosa/Hana/Mateo and sampled 48 official-engine games across an all-family atlas plus targeted rare-condition scan. Observed conditions now include all horizon phases, repeated shops up to multiplicity 4, weeds, shed pressure up to 84 items and 10-order saturation; shed >=95/100, >10 requested orders, land 4+ and more extreme OOD tails remain explicitly unfilled strata. `SAMPLED_RANGE != ENGINE_LIMIT` and `NOT_OBSERVED_IN_ATLAS != IMPOSSIBLE`.\n\nScenario-stratified controls have now advanced materially. Broker's known 199→200 opponent-cash mechanism changed its action in **86/1,440 = 5.97%** sampled natural prestates; all 86 involved WHEAT sell-slot movement, but only **40/86** changed the selected immediate cash/WHEAT-market digest. No sampled LATE/CLOSING Broker prestate fired the threshold in this test. Fresh-vs-stateful Broker action equivalence was 1,440/1,440. In matching weak-trio cash controls, Rosa/Hana/Mateo produced **0/1,440** action differences. Known market thresholds remained positive but context-gated: Hana CARROT 61/480, Hana TOMATO 47/480, Mateo MELON 24/480. New rules: `MECHANISM_PRESENT != MECHANISM_ACTIVE_IN_EVERY_PRESTATE` and `ACTION_BRANCH_DIFFERENCE != IMMEDIATE_ECONOMIC_STATE_DIFFERENCE`.\n\n## 11. Next checkpoint rule\n\nEvery material discovery must update this file or explicitly state why it does not change current strategic state. If a new run contradicts a line here, the contradiction must be recorded before new strategy is generated.\n\n\n## 12. Reaction-attribution calibration update — 2026-09-08 03:45 UTC\n\n`KAG_CONTEXT_PREFLIGHT = PASS` before this research epoch. Canonical L0–L4 sources, all relevant dossiers, latest checkpoint/manifest and sidecar queue/runs/states/artifacts/Blob were read/reconciled before new jobs were created.\n\n### B39/B40 legal baseline evidence\nFresh legal-observation-only MIX11 traces were collected on two independent seed sets against Broker plus the weak trio.\n\n- B39 `[106001,106013,106027]`: MIX11 vs Broker 0-6, avg margin -93,436.67; vs Rosa/Hana/Mateo 6-0 each.\n- B40 `[106211,106223,106237]`: MIX11 vs Broker 0-6, avg margin -104,281; vs Rosa/Hana/Mateo 6-0 each.\n\nA legal-trace-only observational screen found Broker's post-change response score associated with our large cash changes on both batches, including low-market-movement subsets. The weak trio did not show the same group-level cash pattern. Classification: `REPLICATED_OBSERVATIONAL_DETECTOR_CANDIDATE`, **not causal elasticity** and not a deployable posterior yet.\n\n### Fixed-prestate cash reaction calibration\nA fresh official-engine, black-box fixed-prestate probe varied only our publicly visible cash across `[0,199,200,500,1000,3000,10000,50000]`, while preserving the game prestate and our comparison action.\n\nAn initial version omitted the known positive-control contexts (step 0 and D6H0) and consequently failed to recover Broker's known 199→200 transition. This falsified the **probe design**, not the canonical mechanism. After adding step 0 / step 144, the detector recovered:\n\n- Broker: 199→200 action/treatment difference in 6/18 fresh prestates;\n- Rosa: 0/18;\n- Hana: 0/18;\n- Mateo: 0/18.\n\nIn the tested fixed baseline actions, those Broker action differences produced **0/18 immediate public next-state material differences**. This is compatible with the earlier canonical impact screen where only 6/33 contention states changed controlled purchase cost and the maximum immediate effect was four coins.\n\nUpdated classification:\n`BROKER_OUR_CASH_SENSITIVITY = CAUSAL_RECONFIRMED_ACTION_LEVEL / ECONOMIC_MATERIALITY_LOW_OR_CONTEXT_DEPENDENT`.\n\nStill UNKNOWN/not calibrated: general reaction magnitude curve, legal-online causal elasticity, market-mediated fraction, latency distribution, hysteresis/reversal, history dependence and net probe value.\n\nNew failure class: `POSITIVE_CONTROL_CONTEXT_OMISSION`.\nControl: every new reaction detector must include a previously validated positive-control context plus negative controls before null results can update opponent knowledge.\n\n### Runtime/queue state\nWeak-trio reference aliases were repaired against the live reference catalog (`rotation_rosa.py`, `homestead_hana.py`, `melon_mateo.py`), converting the first Rosa B39 missing-agent failure from a deterministic asset/schema failure into a successful rerun without treating it as competitive evidence.\n\nLive sidecar after B40 and B41 enqueue: 217 run attempts, 179 DONE attempts, 38 FAILED attempts, 188 queue jobs; queue rows = 179 DONE / 5 FAILED terminal / 4 READY.\n\nB41 READY jobs use fresh seeds `[106301,106313,106327]` and legal telemetry for Broker/Rita/Lena/Silas to test whether the legally recoverable cash-response signature generalizes across the known direct-awareness family and remains distinct from intermediate control Rita.\n\nChampion/baseline discipline unchanged: no promotion. Working baseline remains MIX11; Broker remains boundary; Cleo remains stress ceiling; Rita remains regression/intermediate.\n\n\n## Material update — 2026-09-08 03:55 UTC — cash-awareness family falsification\n\n### Context gate\n`KAG_CONTEXT_PREFLIGHT=PASS` for this round after full `00_READ_FIRST.md` L0→L4 rehydration, live sidecar queue/run/state/artifact/blob reconciliation, recovery-read of `KAG_AUTOLAB_DAY_AUDIT_20260908.md`, and reconstruction/readback of the missing `KAG_HOURLY_MANIFEST_20260908_0345.json`. No hypothesis or enqueue occurred before PASS.\n\n### B41 legal family calibration completed\nFresh seeds `[106301,106313,106327]`, legal telemetry only online, raw opponent actions absent from deployable telemetry:\n- Broker: MIX11 0-6, avg margin -95,711.67.\n- Rita: 3-3, +365.83.\n- Lena: 0-6, -95,896.33.\n- Silas: 0-6, -94,704.00.\nThese W/L results are `AUTOLAB_COMPUTE_RESULT_NOT_CHAMPION_PROOF` and are not reaction proof.\n\nObservational legal-trace cash-response lifts from B41:\n- Broker +0.3664; low-market subset +0.2647.\n- Rita +0.0989; low-market subset +0.1212.\n- Lena +0.3676; low-market subset +0.2615.\n- Silas +0.6025; low-market subset +0.3458.\nClassification: `OBSERVATIONAL_LEGAL_TRACE_NOT_CAUSAL`. The signal is not Broker-unique and may reflect family/timing structure.\n\n### Fixed-prestate family causal probe — fresh seeds `[106501,106513,106527]`\nSame 18 prestates across all reference agents; our publicly visible cash perception was intervened over the same cash grid. Raw reference action was used OFFLINE only as treatment-fired teacher evidence.\n\n199→200 action changes:\n- Broker: 6/18; material immediate public next-state changes 0/18.\n- Lena: 6/18; material 0/18.\n- Silas: 6/18; material 0/18.\n- Rita: 0/18.\n- Cleo: 0/18.\n- Rosa/Hana/Mateo: 0/18 each.\n\nThis falsifies `BROKER_UNIQUE_CASH_THRESHOLD_SIGNATURE`. Current classification:\n`OUR_CASH_199_200_AWARENESS = CAUSAL_FAMILY_TRAIT_BROKER_LENA_SILAS / CONTEXT_DEPENDENT / LOW_IMMEDIATE_MATERIALITY`.\nIt is useful as opponent-family structure, not as an individual identity mechanism and not as an explanation for the ~100k boundary gap.\n\n### Controlled market-mediation falsification\nFresh worlds `[106427,106439,106451,106463]` reproduced the Broker WHEAT slot-order mechanism:\n- 133 treatment-fired/comparable candidate states;\n- 18/133 nonzero controlled purchase-cost effects = 13.53%;\n- maximum observed saving = 4 coins;\n- mean saving across all comparable states = 0.376 coins.\n\nTherefore `DIRECT_AWARENESS != MATERIAL_ECONOMIC_EFFECT` is strengthened. Cash-awareness is now demoted from boundary-explanation candidate to calibration/family trait unless a new context produces materially larger downstream consequence under preregistered causal testing.\n\n### Next discriminant\nHighest-EVSI active reaction dimensions move to `our_land_sensitivity` and capacity/labor-related reaction, with fixed-prestate interventions and weak-trio/Rita/upper-family controls. Land associations from legal traces remain observational/confounded until such probes fire.\n\nB42 fresh legal family heldout `[106701,106713,106727]` was queued for Broker/Rita/Lena/Silas/Cleo to falsify cross-world stability of the observational family signature. It is profiling evidence only.\n\nChampion unchanged. Working baseline MIX11. Boundary Broker Bea. Stress ceiling Closer Cleo. Regression/intermediate Rita. PASTURE_RAMP and all canonical negative-knowledge families remain closed.\n\n\n### B42 fresh legal-family falsification — 2026-09-08 04:02 UTC\nFresh seeds `[106701,106713,106727]` completed for Broker/Rita/Lena/Silas/Cleo.\nCompetitive context: Broker 0-6/-117406.50; Rita 6-0/+7904.83; Lena 0-6/-117256.67; Silas 0-6/-116592.50; Cleo 0-6/-106777.83. Rita regression recovered strongly in this heldout; no champion implication.\n\nLegal observational cash-response lifts:\n- Broker +0.3955; low-market +0.3675.\n- Rita +0.0139; low-market +0.0330.\n- Lena +0.4061; low-market +0.3693.\n- Silas +0.9207; low-market +0.8540.\n- Cleo +1.5465; low-market +1.4892.\n\nCritical counterproof: Cleo showed the strongest observational cash lift while the fresh fixed-prestate family intervention measured `changed199200=0/18` for Cleo. Conversely Broker/Lena/Silas each had causal 6/18 threshold response. Therefore:\n\n`LEGAL_CASH_RESPONSE_LIFT != DIRECT_CASH_AWARENESS`.\n\nThe observational statistic is now classified as `DYNAMIC/TIMING/MEDIATION_ASSOCIATION`, useful for describing trajectories but not a direct-reaction detector without causal calibration. B39/B40/B41/B42 cash-lift evidence must not be used as direct-awareness proof. This strengthens the requirement to separate reaction after our action from direct reaction to us and to control shared-market mediation.\n\nLand lifts in B42 are also observational and small-n (`n=8` land events per upper opponent), so they are not promoted. Next causal lane remains fixed-prestate land/capacity intervention with treatment detector and controls.\n\n\n### Fixed-prestate public-land probe — 2026-09-08 04:04 UTC\nFresh seeds `[106901,106913,106927]`, 18 identical prestates at steps `[0,57,144,289,433,577]`, all eight public opponents. Only the opponent observation of our public `unlocked_quadrants` was varied over 1/2/3/4 quadrants; all other observed state remained fixed.\n\nResult: `responsivePrestates=0/18` for Broker, Rita, Lena, Silas, Cleo, Rosa, Hana and Mateo; mean distinct actions = 1 for every opponent; 1→3 and 1→4 quadrant contrasts were 0/18 throughout.\n\nClassification: `OUR_LAND_SENSITIVITY = NO_ACTION_LEVEL_RESPONSE_IN_TESTED_LANDMARKS / NOT_UNIVERSAL_ABSENCE`. The B42 observational land lifts are not supported as direct land-awareness by this causal screen and remain timing/confounding candidates. No economic materiality test is warranted for this formulation because the treatment detector never fired.\n\nFailure avoided: `TREATMENT_REQUESTED != TREATMENT_FIRED`. We do not interpret a no-action-response perception intervention as a material economic null beyond the tested contexts.\n\nNext separated factor: public `hands`/labor visibility, using only semantically existing hand objects from each prestate and varying visible count without changing land. Capacity remains decomposed; no fused land+labor probe.\n\n\n### Public labor bundle probe — methodological qualification — 2026-09-08 04:06 UTC\nFresh seeds `[107001,107013,107027]`, 12 prestates with >=5 visible hands at steps `[57,289,433,577]`. The intervention subsampled existing public `hands` objects and capped `hires_today` consistently. All eight opponents showed `responsivePrestates=0/12` and mean distinct actions = 1.\n\nThis result is classified only as `PUBLIC_LABOR_BUNDLE_NO_ACTION_RESPONSE_IN_TESTED_CONTEXTS`. It does **not** isolate `hands` sensitivity because two coupled public fields were changed to preserve semantic coherence. Recording it as `hands absent` would violate the anti-confounding rule. Next probe isolates `hires_today` while preserving the full hands list exactly.\n\n\n### Isolated `hires_today` probe — 2026-09-08 04:08 UTC\nFresh seeds `[107101,107113,107127]`, 12 identical prestates at steps `[57,289,433,577]`. Full public `hands` list was preserved exactly; only public `hires_today` varied over 0/1/3/actual.\n\nResult: all eight opponents `responsivePrestates=0/12`, mean distinct actions = 1, and 0→actual contrast = 0/12. Classification: `OUR_HIRES_TODAY_SENSITIVITY = NO_ACTION_LEVEL_RESPONSE_IN_TESTED_CONTEXTS / NOT_UNIVERSAL_ABSENCE`.\n\nCombined with the land and labor-bundle nulls, current EVSI for further blind capacity perturbations is low. Research pivots rather than stalls: next causal lane is shared-market sensitivity/mediation, beginning with single-factor WHEAT price perception at identical prestates. This directly tests market reactivity without claiming direct reaction to us.\n\n\n### Shared-market single-item price scan — 2026-09-08 04:10 UTC\nFresh fixed prestates `[107301,107313,107327]`, 18 states at steps `[0,57,144,289,433,577]`. Each of the 8 public market prices was intervened **one item at a time** at approximately 0.5×/1×/2× actual, with all other observation fields fixed.\n\nAction-level responses:\n- Broker: 0/18 for WHEAT, CARROT, TOMATO, STRAWBERRY, MELON, MILK, WOOL, FERTILIZER.\n- Lena: 0/18 for all eight.\n- Silas: 0/18 for all eight.\n- Cleo: 0/18 for all eight.\n- Rosa: 0/18 for all eight.\n- Hana: 0/18 for all eight.\n- Rita: WHEAT 9/18; all other seven 0/18.\n- Mateo: WHEAT 3/18; FERTILIZER 4/18; other six 0/18.\n\nIndependent prior WHEAT-only fresh screen `[107201,107213,107227]` had already produced Rita 6/18 and Mateo 3/18, with Broker/Lena/Silas/Cleo/Rosa/Hana 0/18. Thus Rita WHEAT price sensitivity and Mateo WHEAT price sensitivity have replicated directionally across two fresh prestate sets.\n\nClassification:\n- Rita `market_price_sensitivity.WHEAT = CAUSAL_REPLICATED_ACTION_LEVEL / CONTEXT_DEPENDENT`.\n- Mateo `market_price_sensitivity.WHEAT = CAUSAL_REPLICATED_ACTION_LEVEL / CONTEXT_DEPENDENT`.\n- Mateo `market_price_sensitivity.FERTILIZER = CAUSAL_CANDIDATE`, requiring independent replication.\n- All zero cells remain `NO_RESPONSE_IN_TESTED_LANDMARKS`, never universal absence.\n- These are shared-market reaction traits, **not direct reaction to us** and not yet economic-value proof.\n\nNext falsification: fresh focused WHEAT/FERTILIZER intervention for Rita/Mateo with Broker/Cleo negative controls; measure treatment firing and context stability before promoting the fertilizer trait.\n\n\n### 2026-09-08 — Broker activation student fresh-heldout\n\nFrozen legal-context student trained only on seeds 106001/106013/106027/106039 and threshold 0.78 was evaluated without refit on fresh seeds 106051/106063/106079, both seats: 720 rows, 42 positives, AUC 0.9813, TP/FP/TN/FN 36/24/654/6, precision 0.60, recall 0.8571, specificity 0.9646, balanced accuracy 0.9109, F1 0.7059. All three heldout seeds replicated the same confusion counts (12/8/218/2). Six rows exceeded max |train-z| >6. Evidence is strong for legal discrimination of Broker's hidden 199→200 activation regime, but not yet for exploitability or value. `CAN_PREDICT_HIDDEN_MECHANISM_ACTIVATION != KNOWS_HOW_TO_PROFIT_FROM_IT`. Champion remains MIX11; adaptive routing remains disabled.\n\n## 13. Legal reaction timing / V23 self-memory update — 2026-09-08 04:22 UTC\n\n`KAG_CONTEXT_PREFLIGHT = PASS` for this epoch. No post-bootstrap sidecar result was used before canonical L0–L4 rehydration and queue/run/state/artifact/blob reconciliation.\n\n### Causal cash-awareness calibration is now bounded\nFresh fixed-prestate black-box tests independently reproduced Broker's 199→200 opponent-cash treatment, with weak-trio negative controls. A separate controlled WHEAT-contention program then produced 101 comparable treatment-fired states across fresh seeds `[106401,106413,106427]`; 13 changed our controlled WHEAT purchase cost (12.87%) and the largest observed effect remained 4 coins. Combined with the prior canonical 6/33, max-4 screen, this supports:\n\n`BROKER_DIRECT_OUR_CASH_AWARENESS = CAUSAL_REPLICATED` \n`BROKER_CASH_AWARENESS_ECONOMIC_MATERIALITY = LOW / CONTEXT_DEPENDENT`.\n\nIt remains a calibration mechanism, not a plausible primary explanation for the ~100k boundary gap.\n\n### Upper-family causal label at validated contexts\nAt fresh fixed prestates step0 and D6H0/step144, 199→200 changed action-level treatment in Broker 6/6, Lena 6/6 and Silas 6/6; Rita 0/6 and Cleo 0/6. Cleo's null is context-specific and does not negate source-ground-truth slot competition sensitivity; it shows the threshold is not globally active in generic prestates.\n\n### V23 legal self-action memory\nSidecar main core advanced from V22 legal telemetry separation to `KAG_AUTOLAB_CORE_V23_LEGAL_SELF_MEMORY_MAIN`. V23 adds only our own legal market-flow memory (`selfActionTrace`), never opponent raw action. A same-seed mechanical nonregression gate on seed 106601 matched V22 exactly on both seats: terminal banks, margins, W/L, errors and 719 decision calls were identical; V23 additionally produced 719 self-action records per game. Classification: telemetry instrumentation promoted after mechanical nonreg, not policy/champion proof.\n\n### Replicated legal lag signatures\nUsing V23 legal public traces and excluding transitions where our own current market flow was non-zero, large changes in our public cash showed opponent-specific observational response timing. These are **signatures, not causal reaction claims**.\n\nB43 MIX11 fresh seeds `[106801,106813,106827]`:\n- Broker lag0 +0.3565, lag2 +0.1989;\n- Rita lag0 -0.0355;\n- Lena lag0 +0.2250 with lags2/3/4 +0.1824/+0.2571/+0.2850;\n- Silas lag0 +0.6656, lag2 +0.2000;\n- Cleo lag0 +2.4873 then rapid negative/near-zero decay.\n\nIndependent preregistered B44 fresh seeds `[106901,106913,106927]` reproduced the discriminant:\n- Broker lag0 +0.3815, lag2 +0.2021;\n- Rita lag0 +0.0089;\n- Cleo lag0 +2.2374 with immediate decay.\n\nB45 fresh seeds `[107001,107013,107027]` independently reproduced:\n- Lena lag0 +0.4482; lag2/3/4 +0.1983/+0.2278/+0.2334;\n- Silas lag0 +0.8589; lag2 +0.1767.\n\nClassification: `REPLICATED_LEGAL_OBSERVATIONAL_LAG_SIGNATURE / NOT_CAUSAL / NOT_CALIBRATED_POSTERIOR`.\n\n### Policy-shift OOD\nTo test `CLASSIFIER_ACCURACY_SAME_POLICY != POLICY_INVARIANCE`, B46 replaced MIX11 with M40_FERT solely as an instrumentation vehicle, not as policy promotion. Fresh seeds `[107101,107113,107127]` preserved qualitative ordering:\n- Broker lag0 +0.2321, lag2 +0.2464;\n- Rita lag0 -0.0348;\n- Cleo lag0 +1.0135 with subsequent decay.\n\nNumerical magnitude changed materially, so **quantitative policy invariance is not established**. Qualitative timing/order survived one policy shift. B47 with M60_FERT is preregistered/READY as a second policy-shift falsifier.\n\n### Trait state update\n- opponent reaction latency signature: `REPLICATED_OBSERVATIONAL_CANDIDATE` for Broker/Rita/Lena/Silas/Cleo under tested policies;\n- our-cash direct-awareness: causal for Broker and tested Broker/Lena/Silas calibration contexts; context-dependent/UNKNOWN in general;\n- immediate-vs-delayed response profile: observationally replicated, causal mechanism unresolved;\n- reaction magnitude: observational score only, not causal elasticity;\n- shared-market-mediated fraction: still UNKNOWN; our own direct market flow is now explicitly filterable;\n- hysteresis/reversal/history dependence: UNKNOWN;\n- bait/bluff/deception value: UNKNOWN and not yet eligible for exploit search;\n- calibrated posterior: NONE; abstention remains mandatory outside validated support.\n\nChampion/baseline unchanged. MIX11 remains working baseline; Broker boundary; Cleo stress ceiling; Rita regression/intermediate. No negative-knowledge family was reopened.\n\n\n### 2026-09-08 — V25 / natural-support / Broker upstream gap\n\n`reaction-attribution-v1` is now at `KAG_AUTOLAB_CORE_V25_PUBLIC_TILE_LIFECYCLE`; legal public hands/hires and public tile lifecycle telemetry were added with non-regression PASS. Naive lifecycle fusion hurt Broker activation CV, so it is telemetry-only; lean base V2 improved and fresh heldout `[106091,106103,106117]` reached AUC 0.9825, recall 0.9545, specificity 0.9645, balanced accuracy 0.9595, F1 0.7636, precision 0.6364. A >=0.90 active-precision gate was not feasible on train and must not be claimed.\n\nCritical exploit correction: over 12 fresh MIX11-vs-Broker games / 8,628 decision rows, our natural cash never fell <=1000 (minimum 1301), so the true 200-cash Broker mechanism is now `CALIBRATION_POSITIVE_CONTROL / LOW_DIRECT_EXPLOIT_PRIORITY`, not a current attack target. Fresh 8-game gap audit shows we lead through D6 and roughly D9, then lose increasingly from D9 onward; worst mean gap deterioration is D27-D30 (-23,402), D24-D27 (-21,442), D18-D21 (-18,199), D21-D24 (-17,792), D12-D15 (-11,274). Offline action evidence points upstream to labor/land/WHEAT procurement/feed continuity/multi-product throughput/order scheduling. These remain causal hypotheses, not conclusions. Champion remains MIX11; no adaptive routing.\n\n## 14. Cleo causal-context closure + second OOD — 2026-09-08 04:44 UTC\n\n### Second policy-shift robustness\nB47 M60_FERT fresh `[107201,107213,107227]` preserved the preregistered Broker/Rita/Cleo qualitative timing ordering:\n- Broker lag0 +0.1194, lag2 +0.2544;\n- Rita lag0 -0.0541;\n- Cleo lag0 +0.9837 with rapid decay.\n\nB48 M60_FERT fresh `[107301,107313,107327]` preserved upper-family shapes:\n- Lena lag0 +0.2066; lag2/3/4 +0.3768/+0.2116/+0.1188;\n- Silas lag0 +0.3255; lag2 +0.3251.\n\nThus the legal timing signatures have now transferred qualitatively across MIX11 and M60_FERT for Lena/Silas and across MIX11, M40_FERT and M60_FERT for Broker/Rita/Cleo. Numerical magnitudes vary materially; this is **not quantitative policy invariance**. B49 uses non-fertilizer DEPLOY10 as the next OOD falsifier solely as an instrumentation policy.\n\n### Cleo fixed-history causal cash screen\nA new fixed-prestate method was required because Cleo is stateful. For each candidate prestate, a fresh black-box Cleo instance replayed the exact sequence of prior legal public observations before receiving only the final opponent-cash intervention. This preserves reconstructible internal context and avoids `_TRACE`, private replay plans or opponent raw action as online features.\n\nGeneric H0 landmarks across cash `[199,200,1000,10000]` showed 0/5 sensitivity in seed 107401, falsifying the idea that day/phase alone activates Cleo cash response.\n\nThe screen then selected natural black-box contexts where Cleo emitted WHEAT plus at least one other SELL and tested 199→200 with exact history replay on three fresh worlds `[107411,107423,107437]`:\n- 30 candidate prestates;\n- 7/30 treatment-fired sell-slot reorderings;\n- concentrated in closing: mostly D29, earliest D28 H8;\n- 1/30 had any immediate public consequence under the natural Rosa action (Cleo cash -8 / Rosa +7);\n- 0/7 produced non-zero consequence when Rosa was replaced by a controlled BUY_PRODUCT WHEAT contention action (up to 5 units).\n\nUpdated classification:\n`CLEO_OUR_CASH_SLOT_REACTION = CAUSAL_REPLICATED / CLOSING_CONTEXTUAL / LOW_OBSERVED_MATERIALITY`.\n\nThis falsifies the simple hypothesis that Cleo's large legal lag0 cash-event signature is explained by a material WHEAT front-run threshold analogous to Broker. The lag signature and the direct threshold mechanism must remain separate evidence objects.\n\nNext causal budget moves to **shared-market / price-inventory / funding response**, with direct versus market-mediated attribution. Cash threshold remains useful as calibration evidence but is no longer the lead mechanism.\n\n\n## 15. Fixed-history snapshot attribution falsification — 2026-09-08 05:34 UTC\n\n`KAG_CONTEXT_PREFLIGHT=PASS` before this evidence epoch. Canonical bootstrap followed `00_READ_FIRST.md`; strategic intelligence, 190-trait taxonomy, official rules/RT architecture, current/negative state, dossiers, probe matrix/calibration, latest hourly checkpoint/manifest/index and live sidecar queue/runs/states/artifacts/Blob/evidence epochs were READ/RECONCILED. A stale schema assumption was detected and corrected: `kag_autolab_artifacts` stores `artifact_key`, not `artifact_prefix`. Failure class: `STALE_SCHEMA_ASSUMPTION`.\n\n### B49 third structural OOD\nDEPLOY10 was used only as an instrumentation-policy shift, never as promotion evidence. Fresh `[107501,107513,107527]`:\n- DEPLOY10 vs Broker: 0-6, avg margin -90,216.\n- DEPLOY10 vs Rita: 6-0, avg margin +26,843.17.\n- DEPLOY10 vs Cleo: 0-6, avg margin -80,498.33.\nThis reinforces `GOOD_VS_RITA != TRANSFERS_TO_BROKER` and leaves champion/boundary unchanged.\n\n### Shared-market/current-snapshot causal attribution\nFresh fixed-prestate/fixed-history black-box probes were run on `reaction-attribution-v1`. For stateful Cleo, the exact sequence of prior legal public observations was replayed; replay mismatch was zero in every tested context. Raw opponent action remained offline teacher/treatment-detector evidence only.\n\n1. WHEAT-only `[107601,107613]`, 24 market-active MID/LATE/CLOSING contexts per target:\n - Cleo: WHEAT price 0.5x/2x 0/24; WHEAT inventory +/-1000 0/24.\n - Broker: 0/24 and 0/24.\n - Rita positive control: WHEAT price 6/24 action changes, 4/24 immediate public effects.\n - Rosa negative control: 0/24.\n\n2. All eight current public prices `[107631,107643]`, one item at a time 0.5x/2x, 12 market-active contexts per top target:\n - Cleo: 0/12 for each of WHEAT/CARROT/TOMATO/STRAWBERRY/MELON/MILK/WOOL/FERTILIZER.\n - Broker: 0/12 for each item.\n - Rita WHEAT positive control: 8/24; Rosa: 0/24.\n\n3. All eight current public inventories, one item at a time +/-1000, same context stratification:\n - Cleo: 0/12 for every item.\n - Broker: 0/12 for every item.\n - Control apparatus again passed via Rita WHEAT-price 8/24 and Rosa 0/24.\n\n4. Broad public-opponent-cash perception curve `[0,199,200,1000,3000,10000,50000]`, fresh `[107651,107663]`:\n - Cleo: 2/12 responsive contexts, exclusively 199->200; 0 immediate public-effect pairs.\n - Broker: 4/12 responsive contexts, exclusively 199->200; 0 immediate public-effect pairs.\n - Rita/Rosa: 0/12 across the broad cash curve.\n - dedicated Broker 199->200 positive control step0/step144: 8/8.\nNo new broad cash-elasticity regime was found.\n\n5. Public shop-context intervention `[107681,107693]`: prices/inventories/cash/farms held fixed; one unlocked-shop instance removed or duplicated at a time.\n - Cleo/Broker/Rita/Rosa: 32 removals + 32 duplications each, 0 action changes in every target.\n\n### Updated inference\nThe hypothesis that the large replicated legal lag signatures of Cleo/Broker are caused mainly by **instantaneous public snapshot elasticity** to current price, current inventory, simple shop multiplicity or broad our-cash magnitude is falsified in these tested market-active contexts. This is a `TESTED_CONTEXT_NULL`, not universal absence.\n\nLead hypothesis becomes:\n`PATH_DEPENDENT_TIMING_ORDER_SLOT_FUNDING_ATTRIBUTION` — lag signatures may arise from legal public-history/path dependence, timing/order-slot/funding interactions, or endogenous trajectory structure rather than a single current-state scalar/vector. Opposite remains live: a sparse interaction-dependent snapshot mechanism may exist outside sampled activation support.\n\nNext discriminant: alter the **legal public market-history path while holding the terminal observable prestate snapshot fixed**, then measure treatment firing, lag, immediate/public economic effect and context support with exact-history replay + positive/negative controls. No competitive policy mutation is justified by these profiling results.\n\nB50 fresh MIX11 Broker/Rita/Cleo legal-context jobs `[107701,107713,107727]` are READY and unconsumed; their results must not enter strategy until the next context preflight.\n\nChampion unchanged; working lineage MIX11; Broker boundary; Cleo stress ceiling; Rita regression/intermediate. All canonical negative-knowledge families remain closed.\n\n\n## 2026-09-08 05:51 UTC — B50 + history-path narrowing\n\n- Fresh B50 V23 legal telemetry: MIX11 vs Broker = **0-6**, avg margin **-108608.83**; gap is positive through D6, near-flat by D9, negative by D12 and largest sampled deterioration is D24→D27 (**-22017.83**). `NOT_CHAMPION_PROOF`.\n- Fresh B50 regression: MIX11 vs Rita = **6-0**, avg margin **+4116**, errors 0.\n- B50 Cleo remains READY at this checkpoint cut; do not infer its result before completion.\n- Fixed-terminal cash-history V6 recovered immediate Broker/Cleo cash-awareness but found **0 persistence at lags 1/2/6/24/72** after restoring identical subsequent legal history.\n- Fixed-terminal previous-market V7 found **0/8** Cleo lag-1 differences from one-observation WHEAT/MILK price/inventory perturbations.\n- Strategic state: `PATH_DEPENDENT_TIMING_ORDER_SLOT_FUNDING_ATTRIBUTION` remains lead, narrowed away from naive one-step memory. Champion unchanged; MIX11 working lineage; Broker boundary; Cleo stress; Rita regression.\n\n\n## 2026-09-08 06:02 UTC — replicated funding/order mechanism\n\n- B50 completed: MIX11 vs Broker 0-6 / -108608.83; vs Rita 6-0 / +4116; vs Cleo 0-6 / -102071.33.\n- B51 fresh top-tier `[107801,107813,107827]`: Broker 0-6 / -118094.33; Lena 0-6 / -117895; Silas 0-6 / -117079.83; Cleo 0-6 / -106125.67. Ladder rank must not be inferred from scalar damage against MIX11.\n- V6/V7 falsified naive one-step cash/market memory as explanation of natural lag signatures in tested contexts.\n- V8 multi-step WHEAT/MILK path probe was `PARTIAL/NO_TREATMENT_FIRED`, so no null mechanism claim is allowed.\n- V9/V10 raw requested queues remained invariant under cash perturbation; official-interpreter V11 showed that raw-request invariance is not sufficient for funding attribution.\n- Corrected V11: Broker/Lena/Silas structurally invariant to starting cash 0..5000 at sampled D12/D24 states; Cleo funding-sensitive at D12.\n- V12 + fresh-heldout V13 causally replicated `SELL_FIRST_SELF_FUNDING_ORDER`: same action multiset, sells-last counterfactual changes execution only under low cash and converges at natural cash. Broker fresh heldout 12/12 treatment/effect; Cleo 6/6 treatment/effect; natural-cash controls 0.\n- Status: mechanism is `CAUSAL_REPLICATED / REGIME_CONDITIONAL_LOW_CASH / NOT_NATURAL_GAP_PROOF`.\n- Champion unchanged. Working lineage MIX11; Broker boundary; Cleo stress; Rita regression. No negative-knowledge hypothesis reopened.\n- Next: quantify opportunity/materiality of binding mixed SELL+spend states in MIX11 before any policy mutation.\n\n\n## 2026-09-08 06:07 UTC — sell-first materiality audit\n\n- V14 fresh own-action-memory audit found MIX11 already SELL-first in every observed mixed SELL+spend queue: 221/221 turns vs Broker and 210/210 vs Cleo on seeds 108201/108213, mirrored seats. Spend-first = 0; mixed queue saturation = 0.\n- There were real lower-cash opportunities (<2000: 16 turns in each matchup), so the observed ordering is not a trivial high-cash-only artifact.\n- `SELL_FIRST_AS_MISSING_MIX11_CAPABILITY` is falsified. Keep the causal sell-first funding mechanism as a regression/success-pattern invariant, not a candidate mutation.\n- B51 Rita fresh regression = 6-0 / +12751.33. B52 Broker = 0-6 / -105264.67; B52 Cleo = 0-6 / -116979.5.\n- Champion/lineage/league unchanged. No closed negative-knowledge family reopened.\n\n\n## 2026-09-08 06:22 UTC — contextual labor D9→D15 falsification\n\n`KAG_CONTEXT_PREFLIGHT=PASS` before experiment selection. The canonical bootstrap/manifest/190-trait taxonomy/current+negative state/rules/RT architecture/all opponent dossiers/probe matrix/calibration/latest hourly checkpoint+manifest/index and live sidecar queue/runs/states/artifacts+Blob were READ/RECONCILED. Live periodic sidecar queue was empty of READY/RUNNING work at the preflight cut (219 DONE / 5 FAILED terminal rows reported by the worker state), so no inertial refill was used.\n\nFresh legal-public capacity census V15 `[108401,108413]` showed a strong descriptive top-family commonality: Broker/Lena/Silas were identical at D3/D6/D9/D12 in sampled public capacity landmarks, while Cleo was very close. At D9, MIX11 averaged ~1 land / 11 crops / 3.5 animals versus Broker/Lena/Silas ~2 land / 30 crops / 11 animals; by D12 MIX11 ~1/13/5.5 versus ~3/39/14. This is `LEGAL_PUBLIC_DESCRIPTIVE_TOP4_CAPACITY_SIGNATURE`, not causal proof and not evidence that copying any single target transfers.\n\nA preregistered isolated contextual treatment then tested `LABOR10_D9_15` against Broker on fresh `[108301,108313,108327]`, mirrored seats. Treatment fired exactly as intended (baseline laborApplied=0/episode; treatment=6/episode; zero errors). Baseline MIX11 was 0-6 / -88401.67; treatment remained 0-6 / -87730.50. Paired mean delta +671.17 was misleading: only 2/6 episodes improved, both from the same seed family 108301; four episodes worsened. D15 gap worsened in every seed family. Classification: `TEMPORALLY_BOUNDED_LABOR10_D9_D15 = HELDOUT_FALSIFIED / MATCHUP_UNSTABLE / DO_NOT_PROMOTE`.\n\nThis does not reopen or alter the older rejected static/universal hands family; it adds an independently bounded negative result. The next research branch must explain the top-family early capacity signature without blind hands/land/herd target copying. Champion unchanged; MIX11 working lineage; Broker boundary; Cleo stress; Rita regression/intermediate.\n\n\n## 2026-09-08 07:34 UTC — B54 closes standalone staged portfolio\n\n`KAG_CONTEXT_PREFLIGHT=PASS` was repeated after B54 completion before scientific interpretation.\n\nFresh B53 replicated the upper-family legal-public D3→D12 capacity/composition signature. B54 then isolated the crop-transition time path without changing MIX11 land/labor/herd/funding/sell/closing logic. The treatment demonstrably fired and was harmful:\n- Broker: MIX11 0-6/-94384 vs staged 0-6/-113295.17; paired delta -18911.17; 0/6 positive.\n- Rita: MIX11 6-0/+8988.83 vs staged 4-2/-585.50; paired delta -9574.33; 0/6 positive.\n\nCurrent status: `STAGED_STRAW_D6_D12 = HELDOUT_FALSIFIED / HARMFUL / DO_NOT_PROMOTE`.\n\nThis narrows the productive-capacity hypothesis: the top-family MELON→STRAWBERRY sequence is a replicated public fingerprint/trajectory but **not a standalone transferable policy on MIX11's current capacity substrate**. The next teacher question is whether capacity/headroom causally moderates portfolio value. Existing early-land/herd/static-hands negative knowledge remains closed; no fused student mutation is authorized.\n\nLeague unchanged: champion unchanged; MIX11 working lineage; Broker boundary; Cleo stress; Rita regression/intermediate.\n\n\n## 2026-09-08 07:52 UTC — B55 factorial result\n\nB55 completed after B53/B54. The preregistered capacity×portfolio moderation hypothesis failed: Broker A MIX11 0-6/-113730.17, B staged 0-6/-110910.33, C CAP2_D6 0-6/-96016.00, D CAP2_D6+staged 0-6/-113869.83. The factorial interaction is `I=-20673.67`; capacity headroom did not rescue the staged portfolio. Classification: `CAPACITY_HEADROOM_MODERATES_STAGED_PORTFOLIO = REJECTED_IN_B55_FORMULATION`.\n\nThe C-A diagnostic (+17714.17 margin, unchanged 0-6) is not promotion evidence and does not reopen early-land negative knowledge. `CAP2_D6 = ONE_BATCH_REOPENING_CANDIDATE / NOT_REOPENED`; B56 fresh paired Broker+Rita is preregistered as the hard reopening gate. Champion unchanged; MIX11 working lineage; Broker boundary; Cleo stress; Rita regression.\n\n\n## 2026-09-08 08:20 UTC — B56 Broker gate closes CAP2_D6\n\nCanonical context preflight PASS preceded interpretation. B56 fresh Broker `[108901,108913,108927]`, mirrored seats: MIX11 0-6/-116285.00; CAP2_D6 0-6/-128129.33; aggregate paired-direction delta **-11844.33**, errors 0. This reverses the favorable B55 one-batch diagnostic and fails the preregistered Broker directional replication gate.\n\nState update: `CAP2_D6 = FRESH_HELDOUT_FALSIFIED / DO_NOT_REOPEN`. No adjacent land/capacity sweep is authorized. The two already-READY B56 Rita jobs remain useful regression evidence and may complete, but CAP2 reopening no longer depends on them. Research allocation returns to opponent reaction/path-dependence/order-slot/funding attribution under the 190-trait contract. Champion unchanged; MIX11 working lineage; Broker boundary; Cleo stress; Rita regression.\n\n\n## 2026-09-08 09:15 UTC — B56 Rita closes CAP2_D6 with severe regression\n\nCanonical preflight PASS preceded interpretation. The two B56 Rita arms completed on fresh `[108901,108913,108927]`, mirrored seats: MIX11 **6-0 / +9630.50** versus CAP2_D6 **0-6 / -9005.00**, errors 0. Mean-margin treatment-control delta is **-18635.50** and W/L flips from perfect regression pass to complete loss.\n\n`CAP2_D6` is therefore not merely Broker-heldout-falsified; it is now `FRESH_HELDOUT_FALSIFIED / SEVERE_REGRESSION_HARM / DO_NOT_REOPEN`. Champion/working lineage/league remain unchanged. The queue was empty after B56 closure and B57 fresh legal-path replication was preregistered for Broker/Cleo/Rita using legal telemetry only; B57 is observational/falsification evidence, not a reaction-mechanism or champion proof, and may only be interpreted after the next canonical preflight.\n\n\n## 2026-09-08 10:43 UTC — B58 upper-family timing completion\nAfter canonical preflight and post-compute reconciliation, fresh B58 `[109101,109113,109127]` completed: MIX11 vs Lena 0-6/-87274.33; vs Silas 0-6/-86227.67. Legal self-action-filtered detector reproduced Lena lag0 +0.2926 with positive lag2 +0.1220 and smaller positive lags3/4; Silas lag0 +0.5857 with lag2 +0.1421 and negative lags3/4. This directionally replicates the earlier B45/B48 distinction while magnitudes vary. Classification remains `MULTI_HELDOUT_REPLICATED_LEGAL_OBSERVATIONAL_TIMING_SIGNATURE / NOT_CAUSAL / NOT_ROUTING_PROOF`. No champion/negative-knowledge change. Next research gate is fixed-history/fixed-prestate causal attribution of path/order-slot/funding; no macro mutation is authorized from B58.\n\n\n## 2026-09-08 10:47 UTC — continuation preflight reconciliation\n\nFresh canonical preflight completed and persisted at `artifacts/kaggriculture/context/PREFLIGHT_20260908_1047.md`: PASS. Live sidecar main was read directly and is `KAG_AUTOLAB_CORE_V25_CAPACITY_PORTFOLIO_FACTORIAL`, with 241 DONE / 5 FAILED / 0 READY/RUNNING and B57/B58 reconciled. Older V22/V23 implementation labels in `KAG_RT_ARCHITECTURE.md` and early sections of this file are stale; V25 retains the legal self-action-memory capability introduced in V23.\n\nA continuation-ordering breach was also recovered transparently: before this fresh preflight finished, adjacent `LABOR9_D9_15`/`LABOR10_D9_15` screens were run experimentally. Post-preflight reconciliation quarantines them as negative reinforcement only. Fresh `[106293,106307,106321,106333]`, mirrored seats: MIX11 0-8/-104353.875; LABOR9 0-8/-111207.5, paired delta -6853.625, negative 8/8; LABOR10 0-8/-113096, paired delta -8742.125, negative 8/8; treatment fired 6 days/game, errors 0. LABOR9 is closed in this formulation; no adjacent hands sweep authorized.\n\nExperimental `reaction-attribution-v1` currently carries unpromoted telemetry-only label `KAG_AUTOLAB_CORE_V27_PRIVATE_RESOURCE_LIFECYCLE`; it is not sidecar main/champion logic. Correct non-regression selftest is still required before using that telemetry layer. Strategic first incomplete step remains unchanged: inspect prior V8–V14 teacher path/funding work, then implement/preregister the fixed-history/fixed-prestate multi-step path × order-slot/funding causal discriminant. No macro policy mutation before that gate.\n\n\n## 2026-09-08 — controlled history causality confirmed; natural MIX11 support absent\n\nPreregistered `KAG_EXP_CLONE_HISTORY_D12_V2.md` achieved `CONFIRMATORY_PASS` on fresh `[109241,109253,109267]`, mirrored seats. At identical terminal observations in D12 H1..H23, Broker/Lena/Silas each produced 6/138 TWO-vs-ZERO action differences, all 6/6 two-hit threshold-specific (`ONE==ZERO`), all premium-SELL deltas, all with immediate official-interpreter economic consequences. Weak controls Rosa/Hana/Mateo fired 0; terminal invariant failures 0; runtime errors 0. Cleo fired 0 and remains explicit context counterevidence. Evidence class: `CONTROLLED_HISTORY_CAUSAL_MODELS_US_FRONT_RUN_MECHANISM_CONFIRMED` for Broker/Lena/Silas; teacher/raw action remains nondeployable.\n\nPreregistered natural-support audit `KAG_EXP_CLONE_NATURAL_SUPPORT_V1.md` then tested unchanged MIX11 on fresh `[109281,109293,109307]`, mirrored seats, Broker/Lena/Silas/Cleo. Across 24 games and 720 clone-update checkpoint rows, the conservative legal-public signature-distance lower bound was >4 on **720/720** rows; minimum bound Broker/Lena/Silas 8, Cleo 7; optimistic confidence upper-bound stayed 0 in every game. Classification: `NATURAL_SUPPORT_ABSENT_UPPER_BOUND`. Thus the causal clone/front-run capability is real but is not a plausible explanation of current MIX11 losses in these worlds and is not a direct exploit target without separate cost/EV justification for deliberately entering the similarity regime.\n\nCanonical result: `artifacts/kaggriculture/experiments/KAG_RESULT_CLONE_HISTORY_D12_20260908.md`. Champion MIX11 unchanged. Current-loss branch pivots from clone/front-run to continuously active opponent-public-state channels and fixed-plan conversion mechanisms; B57/B58 timing signatures must not be retrospectively attributed to clone/front-run under MIX11.\n\n\n## 2026-09-08 — post-clone source-path falsification\nAfter a fresh canonical preflight, the next candidate continuously-active public-state channel was audited before any new policy mutation. Top-family source exposes `_opponent_scale(obs,item)` and `_reserve_price(...)`, which could in principle scale reservation prices from our public crop/animal supply. Runtime constant inspection on Broker/Lena/Silas/Cleo found `_SUPPLY_DRIVER` populated but `_RESERVE = {}` for all four current reference files. Consequently `_plan_sells` iterates no reserve-controlled items and the `_opponent_scale → reserve_price` path is not active in the current configuration. Classification: `PUBLIC_SUPPLY_RESERVE_RESPONSE_PATH = CODE_PATH_PRESENT / CURRENT_CONFIG_INACTIVE`; it cannot explain B57/B58 timing signatures or current MIX11 losses. This strengthens the pivot toward fixed-plan conversion/throughput plus the already-confirmed sparse cash/clone mechanisms rather than inventing a continuous reaction channel. Champion MIX11 unchanged; 190-trait taxonomy unchanged and unmeasured traits remain UNKNOWN.\n\n\n## 2026-09-08 12:35 UTC — WHEAT/feed capability census + persistent-routing falsification\nCanonical `KAG_CONTEXT_PREFLIGHT=PASS` preceded this epoch. Fresh natural MIX11 census `[109401,109413,109427,109439]`, mirrored vs Broker, found no simple WHEAT-stock shortage in the targeted failure windows: minimum own WHEAT was 5 in D9-D15 and 12 in D18-D27, with zero rows at WHEAT<=2; MIX11 already requested substantial WHEAT procurement. Official-engine semantics confirmed that `consecutive_unfed` is genuine missed-feed-day state and two consecutive missed days remove the animal.\n\nA preregistered single-factor fresh paired test `[109501,109513,109527,109539]` enabled only `POLICY['persistent_routing']=True`, with an explicit runtime treatment detector. Treatment fired 8/8; baseline detector remained false 8/8. Baseline MIX11 vs Broker was 0-8 / -102722.125; `MIX11_ROUTE_PERSIST` remained 0-8 / -99815.625 (diagnostic margin delta +2906.5 only). Critically, the intended mechanism moved the wrong way: D9-D15 H0 unfed-stress sum 22 -> 28 and D18-D27 0 -> 2. Classification: `PERSISTENT_ROUTING_AS_FEED_CONTINUITY_FIX = FRESH_PAIRED_MECHANISM_FALSIFIED / WINLESS / DO_NOT_PROMOTE`; no adjacent routing sweep. Simple procurement quantity is not the demonstrated missing capability in these tested windows. Champion/league unchanged.\n\n## 2026-09-08 12:45 UTC — task-service census falsifies boundary-specific backlog explanation\nFresh natural MIX11 census `[109701,109713,109727,109739]`, mirrored, compared Broker and Rita with identical telemetry and no opponent raw actions. Outcome separated sharply (Broker 0-8 / -92701.375; Rita 7-1 / +5053.125), but our own service-backlog metrics did not: D9-D15 harvest-yield backlog area was 23443 vs Broker and 22950 vs Rita; water-stress area 481 vs 546; CARE backlog 13056 vs 16992; requested sell qty 958 vs 970. D18-D27 harvest backlog was 41682 vs 42912; water stress 665 vs 734; requested sell qty 1347 vs 1512. Feed stress was not uniquely Broker-linked either (Broker 384 vs Rita 288 early; Broker 0 vs Rita 144 late). Shed stayed low in both matchups, never >=80% in these windows.\n\nClassification: `TASK_SERVICE_BACKLOG_AS_BOUNDARY_SPECIFIC_CAUSE = NOT_SUPPORTED / MATCHUP_NONSPECIFIC_IN_FRESH_CONTROL`. General service backlog is real, but because it persists under a strong Rita win regime it cannot by itself explain the Broker boundary. No service/routing/labor mutation is authorized from this census. Research moves to differential economic conversion/output composition and opponent interaction, preserving direct-vs-mediated attribution.\n\n## 2026-09-08 — differential conversion census V1\nAfter fresh canonical preflight PASS and continuity repair, a preregistered no-mutation census compared identical MIX11 policy on fresh mirrored `[109811,109823,109837,109849]` against Broker and Rita. D9-D15 requested sale quantity/mix was essentially identical (976 vs 974 units; WHEAT 816/814, MILK 96/96, WOOL 64/64), so no early quantity-conversion deficit is supported. D18-D27 diverged: Broker-context requested 1368 units vs Rita-context 1547, requested gross-value proxy 129633 vs 162436. Strongest product discriminant is WOOL: 182 vs 370 requested, mean public WOOL price 124.81 vs 209.33, mean private WOOL shed 1.07 vs 1.95. Classification: `LATE_WOOL_PRICE_AVAILABILITY_VOLUME_DIVERGENCE = FRESH_OBSERVATIONAL_DISCRIMINANT / NOT_CAUSAL / NOT_REALIZED_REVENUE`. This is compatible with opponent-mediated shared-market pressure and/or endogenous output divergence; direct reaction is not established. Independent fresh replication is required before causal mediation work.\n\n### WOOL shared-market attribution follow-up\nOn the preregistered replication worlds `[109901,109913,109927,109939]`, Broker-context versus Rita-context D18-D27 also showed mean shared WOOL inventory **+176.42** (9971.80 vs 9795.39), with WOOL inventory start +155.25 and end +198.13. The price gap was -106.99. Direction held in all four seed families: Broker-context WOOL inventory was higher and WOOL price lower. Public productive context differed concurrently: opponent sheep +2.64 in Broker-context, our sheep -2.86; YARN_STORE exposure was -0.42 on average. Classification: `BROKER_CONTEXT_WOOL_GLUT_PATH = REPLICATED_PAIRED_CONTEXTUAL_MECHANISM_CANDIDATE / NOT_DIRECT_CAUSAL_ATTRIBUTION`. The evidence supports shared-market supply pressure as a plausible mediator, but later prestates are not fixed and feedback through our own policy remains. Next discriminant: fresh family-level upper-ladder replication versus Rita before any causal intervention.\n\n### Fresh upper-family WOOL falsifier — world-regime interaction\nPreregistered fresh `[110101,110113,110127,110139]` compared MIX11 versus Broker/Lena/Silas/Cleo/Rita. Broker/Lena/Silas/Cleo formed an extremely tight WOOL-market cluster (mean prices ~125–126; opponent sheep=6), versus Rita mean price 182.36 and opponent sheep ~3.47. However the **direction against Rita flipped by world**: on seeds 110101/110113 top-family WOOL price was far lower than Rita while shared WOOL inventory was higher; on 110127/110139 top-family price was higher while shared WOOL inventory was lower. Thus `STABLE_TOP_FAMILY_WOOL_PRICE_SUPPRESSION` is **FALSIFIED AS UNIVERSAL**. Updated classification: `TOP_FAMILY_WOOL_MARKET_FOOTPRINT = STRONG_FAMILY_CLUSTER / WORLD_REGIME_INTERACTION / INVENTORY-CONSISTENT PRICE DIRECTION / NOT_STABLE_EXPLOIT`. The earlier Broker-context suppression remains real in its batches but is scope-limited; no WOOL-price intervention is authorized from aggregate direction alone.\n\n## 2026-09-08 13:15 UTC — WOOL mediation + direct-composition reaction narrowing\nFresh canonical preflight detected that Current State had advanced beyond the 12:45 hourly chain; the uncheckpointed conversion/WOOL evidence was reconciled before new interpretation. Fresh family census `[110101,110113,110127,110139]` confirms Broker/Lena/Silas/Cleo form a tight WOOL-market footprint cluster versus Rita, but the sign of the WOOL price/inventory difference versus Rita flips by world while remaining inventory-consistent. Therefore `STABLE_TOP_FAMILY_WOOL_PRICE_SUPPRESSION` is falsified as universal; the effect is `WORLD_REGIME_INTERACTION`, not a stable exploit.\n\nTwo fixed-prestate black-box discriminants then narrowed direct reaction. `wool_snapshot_fixedprestate_v1.ts`, fresh `[110401,110413,110427]`, found 0/12 action responses for every tested target to current WOOL price and 0/12 to current WOOL inventory at D18/D21/D24/D27 landmarks, while Broker/Lena/Silas cash 199→200 positive controls fired 3/6 and Rita/Rosa controls stayed 0. `public_mix_fixedprestate_v1.ts`, fresh `[110201,110213,110227]`, found 0/15 action responses in Broker/Lena/Silas/Cleo/Rita/Rosa to isolated public crop-mix and animal-mix perception changes across D3/D6/D12/D18/D24; Broker/Lena/Silas cash positive controls fired 6/6. Classification: `CURRENT_WOOL_SNAPSHOT_REACTION = TESTED_CONTEXT_NULL`; `OUR_CROP_MIX_SENSITIVITY / OUR_ANIMAL_MIX_SENSITIVITY = NO_ACTION_RESPONSE_IN_TESTED_LANDMARKS / NOT_UNIVERSAL_ABSENCE`. The replicated WOOL footprint is therefore better explained by endogenous shared-market supply/world regime than direct snapshot reaction in these tested contexts. Champion/league unchanged; no policy mutation authorized.\n\n## 2026-09-08 15:20 UTC — fresh world-conditioned WOOL mediation discriminant\nAfter canonical preflight PASS, fresh teacher census `[110501,110513,110527,110539]`, mirrored seats, compared unchanged MIX11 against Broker/Lena/Silas/Cleo/Rita with legal public + own-private telemetry only. The upper family again forms a tight WOOL footprint, but the structural mediation chain is not supported as a simple opponent-sheep causal story. Versus Rita, top-family opponent sheep was higher in every paired world, yet shared WOOL inventory moved in the same direction as that sheep difference in only 4/8 paired worlds per upper opponent; by contrast shared inventory and WOOL price moved in opposite directions in 8/8. Our own sheep/output changed strongly by world and often covaried with the inventory regime. Classification: `WOOL_INVENTORY_TO_PRICE_LINK = FRESH_WORLD_CONDITIONED_STRUCTURAL_REPLICATION`; `OPPONENT_SHEEP_TO_WOOL_INVENTORY_AS_SOLE_MEDIATOR = NOT_SUPPORTED / CONFOUNDED_BY_OUR_OUTPUT_AND_WORLD_DEMAND`; `DIRECT_REACTION = NOT_ESTABLISHED`. No WOOL policy mutation authorized. Champion/league unchanged.\n\n## 2026-09-08 17:17 UTC — WOOL temporal mediation V2\nAfter canonical preflight PASS, fresh teacher-only `[110601,110613,110627,110639]`, mirrored seats, compared unchanged MIX11 against Broker and Rita using scenario telemetry and no opponent raw actions. A turn-level D18-D27 diagnostic modeled `Δ shared WOOL inventory` from our requested WOOL sells, our public sheep count and a preliminary town/shop WOOL-demand marker, then added opponent public sheep count. Against Broker, opponent sheep was constant in the sampled window and added **ΔR²=0**; against Rita it added only **ΔR²≈0.000689**. Thus current evidence does not support public opponent sheep count as a useful incremental scalar predictor of turn-level WOOL inventory movement after these controls. The inventory→price sign relation remained exact on every nonzero simultaneous move (Broker 202/202; Rita 96/96). Important limitation: the preliminary town-demand feature evaluated to zero in this implementation, so this is a mediation narrowing result, not a completed conservation identity. `OPPONENT_SHEEP_COUNT != OPPONENT_WOOL_OUTPUT_FLOW`; actual production/collection/sale flow remains unresolved. Champion/league unchanged; no WOOL policy mutation authorized.\n\n## 2026-09-08 18:25 UTC — WOOL V3 measurement-layer falsification\nCanonical preflight PASS preceded the run. Fresh teacher-only `[110701,110713,110727,110739]`, mirrored, unchanged MIX11, attempted to replace sheep stock count with public pasture `yieldSum` transitions and corrected visible demand-exposure timing. The demand detector now fires (`YARN_STORE` exposure rows: Broker 288, Rita 432; town-center tick rows 72 each), but the proposed production-flow detector does **not**: `ourYieldDelta!=0` = 0/1728 rows and `oppYieldDelta!=0` = 0/1728 for both Broker and Rita. Consequently the apparent opponent-flow ΔR²=0 is non-informative and must not update causal mediation. Classification: `PUBLIC_YIELDSUM_DELTA_AS_WOOL_OUTPUT_FLOW_DETECTOR = FALSIFIED/NO_TREATMENT_VARIATION_IN_CURRENT_TELEMETRY`; `DEMAND_EXPOSURE_TIMING = DETECTOR_ACTIVE_BUT_NOT_EXECUTED_DEMAND_LEDGER`. Inventory→price inverse sign again held on every simultaneous nonzero move (Broker 208/208; Rita 148/148). Champion/league unchanged; no WOOL mutation authorized. Next gate is instrumentation/root-cause: identify an actually varying legal public production/collection proxy or derive conservation from shared inventory plus our known executed contribution and deterministic engine demand; do not interpret zero-variance yield telemetry as absence of opponent output.\n\n## 2026-09-08 19:35 UTC — WOOL telemetry aggregation bug isolated\nCanonical preflight PASS preceded this measurement-repair round. Fresh V4/V4b using existing V27 `scenarioTrace.publicTileLifecycle` again reported zero SHEEP yield/pending-care transitions, but an independent direct legal-observation engine probe on seed 110901 demonstrated that the underlying public pasture fields do vary materially: Broker self-play D3-D27 produced 44 yield-sum changes and 24 pending-care changes per farm, including yield 0→6 at D6H0 and 6→0 at D6H7. Therefore the zero-variance result is not an observability absence; it is an instrumentation/aggregation defect in the V27 scenario lifecycle path or its consumer. Classification: `PUBLIC_SHEEP_LIFECYCLE_SIGNAL = LEGALLY_OBSERVABLE_AND_VARIABLE`; `V27_SCENARIO_LIFECYCLE_AGGREGATE = FALSE_ZERO / DO_NOT_USE_FOR_MEDIATION_UNTIL_REPAIRED`. This invalidates V3/V4 zero-variation as evidence about production flow while preserving their measurement-failure lesson. No opponent reaction trait or WOOL policy is promoted.\n\n## 2026-09-08 20:21 UTC — lifecycle aggregate false-zero falsified on fresh scenario telemetry\nCanonical preflight PASS preceded this repair epoch. Fresh V27 scenario-telemetry validation `[111001,111013,111027]`, mirrored, directly consumed `scenarioTrace.publicTileLifecycle.animals.SHEEP` and recovered abundant nonzero transitions: versus Broker, 84 own-yield and 264 opponent-yield transitions plus 142/144 pending-care transitions across 3,456 D3-D27 rows; versus Rita, 99/118 yield and 136/136 pending-care transitions. Example Broker D6H0 opponent `yieldSum +6`, D6H7 `-6`. Therefore the prior V4/V4b zero result is not a core V27 aggregation defect: it was a **consumer path bug** (`publicTileLifecycle` was addressed from the wrong object level in the V4 scripts). Classification updated to `V27_PUBLIC_TILE_LIFECYCLE = SIGNAL_PRESENT/FRESH_REPLICATED`; `V4_V4B_ZERO = CONSUMER_PATH_BUG / MEASUREMENT_FALSE_NULL`. No opponent causal trait or WOOL policy is promoted. Next gate is a true direct-observation↔scenario aggregate parity validator with mutation control, then conservation mediation using the repaired signal.\n\n\n## 2026-09-08 21:19 UTC — lifecycle parity PASS + repaired WOOL mediation V4\nCanonical preflight PASS preceded this epoch. A true teacher same-turn validator compared raw legal public SHEEP tile fields against the scenario-style aggregate for `count/yieldSum/pendingCare/unfedStress` on fresh `[111101,111113,111127]`, mirrored, Broker and Rita. Exact parity held on **4314/4314 rows per target with 0 mismatches and 0 runtime errors**. Deliberately corrupting opponent `yieldSum` by +1 was detected on **4314/4314 rows per target**, so the validator itself passed its mutation control. Classification: `PUBLIC_SHEEP_LIFECYCLE_AGGREGATION_PARITY = FRESH_EXACT_PASS / MUTATION_CONTROL_PASS`; the prior V4/V4b false-zero is definitively a consumer-path failure, not missing signal or aggregate defect.\n\nWith the repaired path, fresh WOOL V4 `[111201,111213,111227]`, mirrored, now has nonzero opponent lifecycle support in D18-D27: Broker 138 yield + 54 pending-care transitions; Rita 74 + 50. Adding opponent yield/pending-care deltas to a preliminary model containing our requested WOOL sells, our lifecycle deltas and demand-exposure timing raises R² by **0.00809 Broker** versus **0.000431 Rita**; every Broker world is positive (~0.0051..0.0137), while Rita remains near zero. Inventory→price inverse sign remains exact on simultaneous moves (Broker 90/90; Rita 116/116). This is `FRESH_REPAIRED_MEDIATION_SIGNAL / OPPONENT_LIFECYCLE_INCREMENTAL_ASSOCIATION_BROKER_GT_RITA / NOT_CAUSAL / NOT_CONSERVATION_IDENTITY`: lifecycle state change is not opponent private collection/sale, and demand terms are exposure timing rather than exact executed demand. No WOOL mutation or reaction/resource-denial trait is promoted. Next gate: engine-validated executed-demand ledger + residual conservation before causal/economic interpretation.\n\n\n## 2026-09-08 22:2x UTC — executed WOOL demand ledger parity + residual mediation narrowing\nCanonical preflight PASS preceded this epoch. A first engine-demand ledger assumption failed parity in 180/3882 rows because YARN_STORE is a single-product shop and official `townConsume` applies multiplier 2 to single-product shops; this was treated as detector/model falsification, not opponent evidence. After correcting to exact official step semantics (`step%4==0`, each YARN_STORE copy removes 2 WOOL; `step%24==0` town center removes 1), fresh PASS-only engine trajectories `[111301,111313,111327]` matched predicted versus actual WOOL removal **3882/3882**, errors 0, and deliberate +1 corruption was caught 3882/3882. Classification: `WOOL_EXECUTED_TOWN_DEMAND_LEDGER = ENGINE_PARITY_PASS / MUTATION_CONTROL_PASS`.\n\nFresh residual diagnostic `[111401,111413,111427]`, mirrored, then used exact town demand and our requested WOOL sells: `residual = ΔsharedInventory + exactTownDemand - ourRequestedWoolSell`. Opponent public SHEEP lifecycle adds small incremental association: Broker ΔR²=0.001679, Lena 0.002708, Silas 0.002481, Cleo 0.002071 versus Rita ~0.000001. This preserves an upper-family-vs-Rita discrimination after exact town-demand correction, but world support is heterogeneous and our requested SELL quantity is not yet separately parity-validated as executed quantity. Classification: `UPPER_FAMILY_LIFECYCLE_TO_WOOL_RESIDUAL = FRESH_ASSOCIATION_CANDIDATE / NOT_CAUSAL / NOT_FULL_CONSERVATION_IDENTITY`. No direct reaction/resource-denial/WOOL exploit trait promoted; no policy mutation authorized.\n\n\n## 2026-09-08 23:2x UTC — own WOOL sell execution gate + full residual V9\nCanonical preflight PASS preceded this epoch. Fresh natural MIX11 census `[111501,111513,111527]`, mirrored, found requested WOOL sells equal the interpreter-semantics executed estimate in all observed D18-D27 sell opportunities: Broker 240 requested/240 executable across 24 rows; Rita 139/139 across 24; no inventory truncation, no 10-slot saturation and no WOOL sell beyond slot 10. An independent official-interpreter synthetic parity probe then validated the execution rule itself: sufficient inventory SELL7 executes 7; inventory=3 truncates SELL7 to 3; slot 11 is dropped; slot 10 executes; 4/4 parity and 4/4 deliberate +1 mutation detection. Classification: `OUR_WOOL_SELL_EXECUTION_MODEL = ENGINE_PARITY_PASS`; natural sampled MIX11 `REQUESTED_WOOL_SELL = EXECUTED_WOOL_SELL` in these windows, scope-limited.\n\nWith exact town demand + validated own executed-sell semantics, fresh full residual V9 `[111701,111713,111727]` leaves only a **very small** upper-family lifecycle association: ΔR² Broker 0.000303, Lena 0.000463, Silas 0.000553, Cleo 0.000415 versus Rita 0.00000316. Broker remains positive in every mirrored world but declines from ~0.000917 to ~0.000106 across the three worlds; Lena/Silas are concentrated almost entirely in seed 111713. This materially downgrades WOOL lifecycle as a large/stable mediator: `UPPER_FAMILY_LIFECYCLE_TO_FULL_WOOL_RESIDUAL = SMALL_FRESH_ASSOCIATION / WORLD_HETEROGENEOUS / NOT_CAUSAL / NOT_EXPLOIT`. No reaction/resource-denial/WOOL policy trait promoted. Champion/league unchanged.\n\n\n## 2026-09-09 00:2x UTC — WOOL conservation fork closes accounting identifiability, natural behavior remains unresolved\nCanonical preflight PASS preceded this epoch. Offline teacher `wool_fixedprestate_conservation_fork_v10.ts` used no-town-demand step 433 and forced only opponent executed WOOL SELL quantity across fresh seeds `[111801,111813,111827]`, both seats. Shared WOOL inventory delta matched executed opponent supply exactly **36/36**, and deliberate +1 corruption was detected **36/36**. The same fixed prestate showed nonlinear immediate price impact: SELL 1 -> 0 price change; 3 -> -1; 7 -> -3; 10 -> -6; 20 -> -23. Classification: `OPPONENT_EXECUTED_WOOL_SUPPLY_TO_SHARED_INVENTORY = ENGINE_CAUSAL_CONSERVATION_PARITY`; `EXECUTED_SUPPLY_TO_PRICE = ENGINE_CAUSAL_LOCAL_NONLINEAR_EFFECT`. This validates that a no-demand residual can identify executed opponent supply when other flows are controlled, but it is a forced teacher mechanism, not evidence that an opponent naturally chooses those sells.\n\nA sparse offline raw-action checkpoint census `wool_opponent_oracle_supply_census_v11.ts` on fresh `[111901,111913,111927]` found no WOOL sells at D18/D21/D24/D27 checkpoints for Broker/Lena/Silas/Cleo, while Rita had 30 units across 8/24 checkpoint rows. Because `actionTimeline` is checkpoint-sparse, `ZERO_AT_CHECKPOINTS != ZERO_TURN_LEVEL_SUPPLY`; this is counterevidence against a naive continuously-obvious upper-family sell story, not a natural-supply null. Combined with V9's tiny/world-heterogeneous lifecycle residual, WOOL is now **deprioritized as primary boundary mechanism** unless exact turn-level teacher accounting reveals materially larger upper-family executed supply. No WOOL student mutation or opponent reaction/resource-denial trait is promoted. Champion/league unchanged.\n\n\n## 2026-09-09 01:2x UTC — AVIÃO non-WOOL conversion ranking: MILK becomes next discriminant, but not yet stable mechanism\nCanonical preflight PASS preceded this epoch. Sidecar main was reconciled at 241 DONE / 5 FAILED / 0 READY/RUNNING; live SQLite `hourly:*` state is stale at the 21:19 epoch while canonical hourly artifacts are current through 00:25, so sidecar strategic state was explicitly treated as non-authoritative/stale rather than overriding canonical state.\n\nFresh self-executed-flow ranking `[112001,112013,112027,112039]`, mirrored, D18-D27, unchanged MIX11, found the largest **non-WOOL** Broker-vs-Rita realized-sale-value differential in MILK: Broker-context executed MILK 632 units / value proxy 103,816 versus Rita-context 770 / 120,737, delta -138 units / -16,921 despite Broker-context average realized public MILK price being +7.46 higher. WHEAT quantity was also lower (-63) but value was +3,033 because price was +10.33 higher. Classification: `MILK_LATE_CONVERSION_VOLUME = HIGH_EVSI_FRESH_DIAGNOSTIC_CANDIDATE / NOT_CAUSAL`; the signal points to availability/output conversion rather than simple low sale price.\n\nIndependent fresh decomposition `[112101,112113,112127,112139]` narrowed the claim. Broker remained 0-8/-114639.375 while Rita was 6-2/+3983.875. Broker-context mean MILK sell was 75.75/game versus Rita 100 (-24.25); mean private MILK shed 3.55 vs 5.34; public own COW count 7.87 vs 8.70; mean COW yieldSum 6.58 vs 7.90; mean MILK price was actually higher for Broker (+25.62). However world heterogeneity is large: one mirrored Broker world collapsed to ~2 cows / price ~7 while other worlds had 9-11 cows / price ~228-234. Therefore `STABLE_BROKER_MILK_DEFICIT = NOT_YET_ESTABLISHED`; current state is `MILK_OUTPUT/AVAILABILITY_PATH = REPLICATED_DIRECTIONAL_DIAGNOSTIC / WORLD_HETEROGENEOUS / NOT_CAUSAL`.\n\nNo student mutation, cow/herd reopening, direct-reaction trait, resource-denial claim or champion promotion follows. Existing early-herd negative knowledge remains closed; this is a different late conversion measurement question, not permission to sweep herd targets. Next gate is world-conditioned MILK conservation/output attribution with exact own executed sells + deterministic MILK town/shop demand, then a single fixed-prestate causal test only if the residual identifies a material controllable conversion loss.",[],[175,176,177,2471],"933dceaf-6424-4ea3-9c9f-144145478ef0","KAG_CONTEXT_CONTINUITY_AUDIT_20260908.md","ce3c5e10-ab2e-11f1-bd27-1607ee4eb77e",601,["D",1788835294192],"# KAGGRICULTURE — Context Continuity Audit 2026-09-08\n\nStatus: VERIFIED_BEST_AVAILABLE_RECOVERABLE_CONTEXT / LITERAL_ALL_CHAT_HISTORY_NOT_CLAIMED\n\n## Finding\n\nThe user's concern was valid. The project already contained substantial opponent-intelligence knowledge, but it was fragmented across strategic intelligence, experiment matrices, curricula, protocol docs, dossiers, checkpoints and sidecar state. There was no Kaggriculture-specific canonical START_HERE on `chat-native-maestro@economic-v0.6` before this repair.\n\nA general Grimório pattern already existed externally (`00_CORE/00_START_HERE.md`, artifact-first recovery, checkpoints as external memory, source-before-memory). The Kaggriculture project had not fully inherited that control structure.\n\n## Existing Kaggriculture knowledge found\n\n- `KAG_STRATEGIC_INTELLIGENCE.md`: OBG, Bayes, HMM, RLS, CUSUM, SPRT, EVSI/EVPI/VoC, reaction functions, revealed preference, QRE, Level-k, history/hysteresis, bait/bluff validity, OOD, evidence tiers.\n- `KAG_EXPERIMENT_MATRIX_OPPONENT_PROBES.md`: telemetry equivalence, weak-agent controls, reaction estimators, history sensitivity, OOD, Broker/Rita/top-family probes and promotion gates.\n- `KAG_OPPONENT_PROFILES/*`: live dossiers with SOURCE_GROUND_TRUTH/OBSERVED/INFERENCE/UNKNOWN/COUNTEREVIDENCE separation.\n- `KAG_PROFILE_CAL_V1_20260907.md`: calibrated market thresholds and selected opponent-awareness probes.\n- `KAGGRICULTURE_REAL_WORLD_CURRICULUM_V1.md`: opponent modeling/Theory of Mind, game theory, robust RL, active learning, causal attribution, economics/industrial organization, market microstructure, OODA/wargaming/red teaming.\n- `kaggriculture_neuron_protocol_v1.md`: competitive architecture and opponent-aware strategy research.\n- `kaggriculture_joined_research_plan_v1.md`: adaptive macro thesis, opponent-aware anti-glut and promotion discipline.\n- `KAG_NEGATIVE_KNOWLEDGE.md`: hard rejected hypotheses/reopening gates.\n- hourly checkpoints + Autolab SQLite/Blob evidence.\n\n## Recovered original OBG concepts from prior conversation\n\nThe prior project context additionally recovered an explicit multidimensional opponent genome containing:\n\n`advance/retreat; bluff/counter-bluff; aggression/defensiveness; risk; commitment/reversibility; patience/impulsivity; exploration/exploitation; adaptability/inertia; memory/recency; awareness of our state/market; predictiveness/reactiveness; opportunism/resource-denial; price/supply/demand/cash sensitivity; scarcity/saturation; diversification/concentration; timing; closing; composure/persistence; reasoning depth; models-us; deception/probe awareness`.\n\nThese were not all explicit as first-class fields in the reconstructed dossiers, creating a real risk of scope loss.\n\n## Corrections implemented\n\n1. Created canonical bootstrap `artifacts/kaggriculture/00_READ_FIRST.md`.\n2. Created `KAG_CONTEXT_MANIFEST.json` with mandatory source paths.\n3. Created `KAG_OPPONENT_CAPABILITY_TAXONOMY.md` with **190 explicit traits** and UNKNOWN inheritance.\n4. Created `KAG_CURRENT_STATE.md` with live current state, experimental branches and contradiction register.\n5. Added READ_FIRST banner to canonical Maestro root README.\n6. Added READ_FIRST/strategy-authority banners to both Autolab READMEs.\n7. Added taxonomy inheritance rule to opponent profile INDEX.\n8. Updated stale runtime-architecture implementation section from V19 to V22/V23 reality.\n9. Updated sidecar SQLite with `context:read_first`, `context:capability_taxonomy` and replaced stale `hourly:current_hypothesis` with a context-reconciliation gate.\n10. Updated the active hourly automation so **STEP ZERO** must read the canonical bootstrap and produce `KAG_CONTEXT_PREFLIGHT=PASS` before strategy/enqueue.\n11. Created an additional durable pointer in `cleitoncosta/grimorio-evidence-memory@main/KAGGRICULTURE_READ_FIRST_POINTER.md`.\n\n## New failure class\n\n`STRATEGIC_CONTEXT_DRIFT_BY_LOCAL_OPTIMIZATION`\n\nObserved pattern:\n- large strategic plan exists;\n- current subproblem becomes salient;\n- repeated local iterations optimize that subproblem;\n- broader mission is no longer loaded into active context;\n- local progress is incorrectly treated as project direction.\n\n### Control\n`00_READ_FIRST + manifest + taxonomy + current-state + mandatory preflight receipt + automation hard gate`.\n\n### Detector\nIf recent rounds are dominated by W/L-only, static mutations, tape/fingerprint decomposition, or classifier refinement without reaction/probing/history/adaptation coverage, mark `DRIFT_DETECTED` and run context preflight before further strategy.\n\n### Regression\nFuture hourly checkpoints must contain `KAG_CONTEXT_PREFLIGHT`. Absence means the round's strategic output is not eligible to become canonical even if compute results are valid.\n\n## Epistemic limit\n\nThis audit does **not** claim byte-for-byte enumeration of every historical account conversation. The runtime can recover project context, personal-context summaries, File Library search results, Val Town history/artifacts and current connected state, but cannot prove literal full-account transcript census. Therefore:\n\n`CURRENT_AND_RECOVERABLE_KAG_PROJECT_HISTORY_SYNTHESIZED = TRUE`\n\n`ALL_ACCOUNT_CHAT_TRANSCRIPTS_PHYSICALLY_ENUMERATED = FALSE`\n\nCritical knowledge is now externalized so future correctness does not depend on the unavailable literal transcript corpus.\n\n## Canonical entrypoint\n\nAlways begin Kaggriculture work here:\n\n`cleitoncosta/chat-native-maestro@economic-v0.6/artifacts/kaggriculture/00_READ_FIRST.md`\n",[],[175,176,177,2479],"5ef4af24-abeb-4e7e-be1e-5f1337b7f591","silas.md","01a07e48-3be6-726a-80bb-f997514b42e5",741,["D",1788864095261],"# Slotter Silas — Live Opponent Dossier\n\nRole: upper-ladder target above Lena.\n\n## SOURCE_GROUND_TRUTH\nPublic tier-8 strategy:\n- same meta field plan;\n- market queue is reordered so premium goods occupy the earliest SELL slots and glut-resistant staples fill behind them.\n\nDependency probe confirms direct other-farm dependency plus market-price/inventory references.\n\n## OBSERVED\n- MIX11 vs Silas B1 recovered: 0-4, strongly negative margin.\n- Public Silas > Lena in most recovered batches (often 6-2/7-1/8-0; some 5-3), supporting a robust but not deterministic ladder relationship.\n- Cleo usually beats Silas in recovered public crossplay.\n\n## INFERENCE\nSilas is a scheduling/market-slot opponent. Outcome differences may arise from priority ordering under the same 10-order-per-turn interpreter budget rather than gross macro farm composition.\n\nThis makes Silas especially relevant to our earlier `MARKET_ORDER_BUDGET_CONTENTION` lesson: the causal variable may be which economic action receives a scarce market slot, not merely requested quantity.\n\n## REQUIRED PROBES\n1. recover executed vs requested market-slot sequence;\n2. perturb premium/staple inventory while holding macro farm state fixed;\n3. inspect funding continuity and purchases lost to sell-slot pressure;\n4. compare Lena/Silas/Cleo from identical prestates;\n5. model slot priority as a discrete choice / revealed-preference problem rather than continuous crop-share tuning.\n\n## FALSIFIER\nIf a proposed Silas exploit changes crop quantities but leaves the binding market-slot sequence unchanged, require evidence that it hits the actual mechanism before promotion.\n\n## DIRECT CASH-SENSITIVITY CALIBRATION\nAt the step-0 fixed prestate, opponent cash 199→200 changes Silas's market queue at the same WHEAT promotion threshold as Broker/Lena. Below 200 the WHEAT sells remain behind other operations; at/above 200 they are promoted to the front.\n\nThis is causal evidence of direct opponent-cash awareness, but not a unique identity signature because Broker and Lena share the same step-0 response. Silas still needs differentiation through later sell-slot timing/ordering behavior.\n\n## FULL 15-MIN AUDIT UPDATE\n\nAfter canonicalizing historical aliases, direct Silas>Lena evidence aggregates **80-12 across 12 batches / 92 episodes**. B38 itself is 8-0 Silas over Lena with avg direct margin +1,964.375.\n\nHowever, on the exact same third-party M60_FERT seeds `[105401,105409,105421,105431]`, M60 lost to Silas by -71,641.125 and to Lena by -71,571.5 — only **69.625** coins apart. Thus direct Silas superiority is not a scalar prediction of Silas's extra damage to our policy. The priority is to learn which slot/funding/timing mechanism interacts with **our** trajectory, not merely rank Silas above Lena.\n\nHard rule: `LADDER_STRENGTH != MATCHUP_EFFECT_AGAINST_US`.\n\n## 2026-09-08 03:55 UTC — direct cash-awareness family evidence\nFresh fixed-prestate intervention on identical prestates found a 199→200 perceived-our-cash action change in 6/18 states, matching Broker and Lena; immediate material public next-state effect was 0/18. B41 legal observational cash lift was +0.6025 (low-market +0.3458), stronger than Broker/Lena but still observational. Classification: `CAUSAL_FAMILY_TRAIT_BROKER_LENA_SILAS`; magnitude in legal traces may differ, but causal economic materiality is unproven.\n\n\n## 2026-09-08 06:23 UTC — early-capacity composition\nFresh legal-public census `[108501,108513,108527]` matched Broker essentially exactly at D3/D6/D9/D12 in land, crop composition and animal composition. Classification: `LEGAL_PUBLIC_DESCRIPTIVE_TOP_FAMILY_CAPACITY_SIGNATURE / NOT_CAUSAL`. Preserve separate Silas timing/slot traits; common macro capacity does not imply identical behavior.\n\n\n## B58 legal timing fresh replication — 2026-09-08 10:43 UTC\nFresh `[109101,109113,109127]`, mirrored seats, MIX11: 0-6 / -86227.67. Legal self-action-filtered timing detector: lag0 +0.5857, lag2 +0.1421, lags3/4 -0.1820/-0.0818. Directionally consistent with prior Silas strong-immediate+lag2 evidence. Classification: `MULTI_HELDOUT_REPLICATED_OBSERVATIONAL_TIMING_SIGNATURE / NOT_CAUSAL`; direct-vs-mediated attribution, history/path and calibrated posterior remain UNKNOWN.\n",[],[175,176,177,178,2487],"5f6c5162-39a3-44eb-8446-8a19dd67458d","rosa.md","01a07e48-3be6-726a-80bb-f46b4c774e65",630,["D",1788839031002],"# Rotation Rosa — Live Opponent Dossier\n\nRole: weak-regression + negative-control opponent-awareness calibration.\n\n## SOURCE_GROUND_TRUTH\nPublic reference source describes a fixed tier-2 policy:\n- hires 4 hands every morning;\n- NW quadrant only; never buys land or livestock;\n- crop mix wheat/carrot/tomato = 50/30/20;\n- waits for full watering bonus window before one-time harvest;\n- seed batch 6; target seed stock 8;\n- sells tomato → carrot → wheat in chunks of 40;\n- shed pressure 70;\n- plants through day 25; liquidates from day 28.\n\nDependency probe found no reference to opponent/other farm in source. Therefore direct our-farm awareness should be absent when shared market/world effects are controlled.\n\n## OBSERVED\n- MIX11 B21: 8-0 against Rosa; positive avg-margin regime in recovered sidecar.\n- Early static source family should be identifiable by no expansion/no animals + staple rotation.\n\n## CALIBRATION HYPOTHESES\nH1 `our_state_awareness ≈ 0` under fixed market/world/private prestate.\nH2 price-threshold reactivity should be materially weaker than Hana/Mateo because Rosa policy has no explicit product price floor.\nH3 expansion/livestock response should remain absent even if our visible farm changes, unless mediated indirectly through shared market conditions.\n\n## REQUIRED PROBES\n1. no-op paired baseline;\n2. our-visible-expansion perturbation with market state controlled;\n3. market price perturbation for wheat/carrot/tomato;\n4. history sensitivity: same visible current state after different opponent-visible histories;\n5. OOD shop regimes.\n\n## FALSIFIER\nIf profiler assigns high direct our-state-awareness to Rosa while fixed-prestate controls keep shared market/town state equal, the profiler is wrong or the dependency/source assumptions changed. Treat as detector failure before applying to Broker/Lena/Silas/Cleo.\n\n## CALIBRATION RESULT — fixed-prestate black-box probe\nNatural prestate seed 460101, D4 H0, CARROT shed=6, pressure inactive. Sweeping CARROT price 14→15→16 produced identical CARROT SELL quantity 6 in all three cases. Raising opponent public cash by +50,000 changed no action. This is a passed negative control for explicit price-floor detection and direct opponent-cash awareness on this prestate.\n\nEvidence class: `FIXED_PRESTATE_BLACKBOX_POLICY_CALIBRATION`. Do not generalize to every possible opponent feature solely from this one intervention; source dependency scan remains the stronger structural negative for direct other-farm reads.\n\n## REAL-SCENARIO STRATIFIED CASH NEGATIVE CONTROL — V24\n\nNatural official-engine prestates against Rita, seeds 100577/100609, both seats, sampled every 6 in-game hours: opponent-cash 199→200 produced **0/480** Rosa action differences and 0 fresh-vs-stateful mismatches across sampled EARLY/MID/LATE/CLOSING strata.\n\nCombined with source dependency evidence, this strengthens Rosa as the direct-awareness negative control. Scope remains the tested cash dimension; it does not prove invariance to every conceivable observable feature.",[],[175,176,177,178,2495],"8970ab26-2a4d-427c-90e9-0c087849cc57","rita.md","01a07e48-3be6-726a-80bb-f22171a5114a",728,["D",1788859213518],"# Rancher Rita — Live Opponent Dossier\n\nRole: regression boundary below Broker; transition target after weak-trio calibration.\n\n## SOURCE_GROUND_TRUTH\nPublic tier-5 strategy:\n- buys two extra quadrants;\n- hires 8 hands;\n- builds 16 pastures;\n- targets 10 cows + 6 sheep;\n- grows wheat for feed and tops up wheat from market below 55/unit;\n- feeds and CAREs livestock daily.\n\nThe generic dependency probe saw the word `opponent` but did not detect direct other-farm indexing; direct our-state awareness is therefore **not yet proven** by this source scan.\n\n## OBSERVED\nFresh early signature: land1 / ~13 wheat at D1; by D3 3 cows + 1 sheep. This makes Rita distinguishable earlier than the top meta family.\n\nMIX11 empirical results vary substantially across batches, e.g. 8-0, 6-2, 5-3, 4-4 and one 3-5 slice, generally with positive mean margin in many current runs. This is a useful robustness/regression opponent rather than a solved deterministic fixture.\n\nMultiple Broker-targeted mutations harmed Rita even when they appeared directionally interesting against Broker. PASTURE_RAMP B31 ended 4-4 +125 vs paired MIX11 4-4 +3569.625, a -3444.625 margin delta.\n\n## INFERENCE\n- Livestock/wheat conversion gives a strong structural signature.\n- Rita is useful for detecting whether a Broker exploit destroys broad economic robustness.\n- Market wheat-price sensitivity is source-grounded (buy below 55), but direct opponent-state sensitivity remains UNKNOWN.\n\n## REQUIRED PROBES\n1. wheat price sweep around 55 from fixed prestate;\n2. own-state perturbation with market controlled;\n3. livestock build timing / commitment hysteresis;\n4. feed-scarcity OOD regimes;\n5. paired regression for every Broker-targeted candidate.\n\n## COUNTEREVIDENCE / GUARD\nDo not promote a mutation because it improves Broker margin if it causes statistically/materially worse Rita W/L or paired economic trajectory without a stronger matchup-routing justification.\n\n## CASH-AWARENESS SCREEN — negative result in one full natural episode\nA fixed-prestate 199-vs-200 opponent-cash perturbation was applied at every scanned step of one 719-step natural Rita-vs-Rosa episode (seed 471000). No Rita action changed. Classification: `NO_DIRECT_CASH_SENSITIVITY_OBSERVED_IN_THIS_EPISODE`, not universal absence.\n\nThis materially weakens any hypothesis that Rita uses the same opponent-cash threshold mechanism as Broker/Lena/Silas/Cleo. It does not rule out sensitivity to other opponent-visible variables.\n\n## FULL 15-MIN AUDIT UPDATE\n\nAcross persisted non-selftest MIX11 batches through the audit cutoff, Rita totals **67-31 across 98 episodes**, weighted avg margin about **+2,770.08**, with materially different batch outcomes ranging from strong wins to losing slices. Multiple herd/M60-like variants can improve or remain strong against Rita while failing completely against Broker.\n\nThis promotes Rita's role from a generic “weaker opponent” to a **matchup-specific regression and conditional-routing laboratory**. Her large world/batch variation should be used to study regime sensitivity, livestock commitment, wheat/feed dependence and robustness. It does not justify a universal strategy transfer to Broker.\n\nHard rule: `GOOD_VS_RITA != TRANSFERS_TO_BROKER`.\n\n## 2026-09-08 06:23 UTC — capacity-composition negative/intermediate control\nFresh legal-public control `[108501,108513,108527]`: Rita remained 1 land and WHEAT-only (13 crops) through D12; animals were 4 through D9 and ~9.7 by D12. MIX11 went 4-2 / +2426.83 in these worlds. Rita therefore does **not** share the top-family D3→D12 MELON→STRAWBERRY diversification/land-expansion sequence. Classification: `NEGATIVE_INTERMEDIATE_CONTROL_FOR_TOP_FAMILY_CAPACITY_COMPOSITION`, not proof that the signature alone causes upper-tier strength.\n\n\n## B54 regression update — 2026-09-08 07:35 UTC\n\nFresh paired `[108701,108713,108727]` reinforced Rita's role as a hard regression control for portfolio mutations. MIX11 was 6-0 / +8988.83; the staged post-D6 MELON→STRAWBERRY treatment fell to 4-2 / -585.50. Every paired episode had worse terminal margin under treatment (mean delta -9574.33; range -18017..-4492), despite some transient D12/D15 local gap improvement.\n\nInterpretation: a mutation that visually resembles upper-family composition can damage broad economic robustness even when a few intermediate landmarks look locally better. `TRANSIENT_TRAJECTORY_GAIN != REGRESSION_PASS`.\n\n\n## B56 CAP2_D6 REGRESSION CLOSURE — 2026-09-08 09:15 UTC\nFresh mirrored `[108901,108913,108927]`: MIX11 beat Rita **6-0 / +9630.50**, while CAP2_D6 lost **0-6 / -9005.00**. This is severe regression damage, not merely a margin warning. Rita therefore successfully serves as a hard regression discriminator here: the D6 headroom treatment that failed fresh Broker also destroys the intermediate matchup. No Rita trait is inferred from our-policy treatment; this is regression evidence only.\n",[],[175,176,177,178,2503],["D",1788839007738],[],[175,176,177,178,66],"640e83f2-320d-4371-a663-ae94fb626c5c","lena.md","01a07e48-3be6-726a-80bb-eadda252268f",740,["D",1788864090627],"# Ledger Lena — Live Opponent Dossier\n\nRole: upper-ladder target above Broker.\n\n## SOURCE_GROUND_TRUTH\nPublic tier-7 strategy:\n- same meta field plan family as Silas/Cleo;\n- same sell-reordering idea as Slotter Silas;\n- tuned to a different trade-off between selling premium goods early and keeping staples flowing.\n\nDependency probe confirms direct other-farm source dependency and market-price/inventory references.\n\n## OBSERVED\n- MIX11 vs Lena B1 recovered: 0-4, strongly negative margin.\n- Public Lena > Broker repeatedly in multiple recovered batches (commonly 6-2/7-1/8-0 slices).\n- In one analyzed trajectory, Lena and Broker could have equal public feature counts late while cash diverged; Lena sold more inventory around D21/D24. This was evidence for conversion/sell-order timing rather than simple planted-share differences.\n- Generic extra liquidation D21/D24 on our agent was explicitly treatment-verified and near-null against Broker, so “sell more at D21/D24” is not a universal transferable explanation.\n\n## INFERENCE\nThe distinction from Broker is likely in execution/market-slot ordering and timing rather than obvious early static farm composition. Static snapshots alone are insufficient.\n\n## REQUIRED PROBES\n1. action-window telemetry around market order slots;\n2. premium-vs-staple sell ordering and funding continuity;\n3. direct our-state reaction conditioned on market state;\n4. late-horizon conversion regime HMM;\n5. compare Lena/Silas/Cleo on identical fixed prestates to isolate ordering semantics.\n\n## FALSIFIER\nIf the classifier separates Lena from Silas/Cleo using only D1-D9 static farm counts, demand independent validation: current early-signature evidence shows top-family aliasing, so such confidence may be spurious.\n\n## DIRECT CASH-SENSITIVITY CALIBRATION\nAt the step-0 fixed prestate, opponent cash 199→200 changed Lena's market queue exactly at the same threshold seen in Broker: below 200 the WHEAT sells stayed behind hires/buys; at/above 200 they moved to the front. This is a causal direct-opponent-cash response in the black-box action function.\n\nImportant discriminator: this cash response alone does **not** separate Lena from Broker or Silas; all three exhibit the same step-0 transition. Classification among the trio therefore still requires additional temporal/market signatures.\n\n## FULL 15-MIN AUDIT UPDATE\n\nAfter canonicalizing historical aliases, direct public Lena>Broker evidence aggregates **123-9 across 17 batches / 132 episodes**. This is a strong relational ladder fact, but the full-day audit shows it must not be converted into a scalar expectation for our matchup.\n\nB38 adds a critical interaction clue: Silas beat Lena 8-0 directly on fresh worlds, yet M60_FERT lost to Silas and Lena by almost the same average margin on their exact shared third-party seed set (difference only 69.625 coins). Therefore Lena's effect on us depends on `opponent × our_policy × world_regime`, not only ordinal ladder strength.\n\nHard rule: `LADDER_STRENGTH != MATCHUP_EFFECT_AGAINST_US`.\n\n## 2026-09-08 03:55 UTC — direct cash-awareness family evidence\nFresh fixed-prestate intervention on identical prestates found a 199→200 perceived-our-cash action change in 6/18 states, exactly matching Broker in this screen; immediate material public next-state effect was 0/18. B41 legal observational cash lift was +0.3676 (low-market +0.2615), also close to Broker. Classification: `CAUSAL_FAMILY_TRAIT_BROKER_LENA_SILAS`, not unique identity and not material economic-effect proof. Continue reaction modeling on higher-EVSI land/capacity dimensions.\n\n\n## 2026-09-08 06:23 UTC — early-capacity composition\nFresh legal-public census `[108501,108513,108527]` matched Broker essentially exactly at D3/D6/D9/D12 in land, crop composition and animal composition. Classification: `LEGAL_PUBLIC_DESCRIPTIVE_TOP_FAMILY_CAPACITY_SIGNATURE / NOT_CAUSAL`. This strengthens family-level macro similarity but does not imply equal reaction functions or ladder strength.\n\n\n## B58 legal timing fresh replication — 2026-09-08 10:43 UTC\nFresh `[109101,109113,109127]`, mirrored seats, MIX11: 0-6 / -87274.33. Legal self-action-filtered timing detector: lag0 +0.2926, lag2 +0.1220, lag3 +0.0276, lag4 +0.0626. Directionally consistent with prior Lena immediate+tail evidence, but magnitude/tail shape varies. Classification: `MULTI_HELDOUT_REPLICATED_OBSERVATIONAL_TIMING_SIGNATURE / NOT_CAUSAL`; history/path, direct-vs-mediated attribution and calibrated posterior remain UNKNOWN.\n",[],[175,176,177,178,2514],"cfaa8708-982d-4d77-8ae4-ed95896778d2","hana.md","01a07e48-3be6-726a-80bb-e4a94af4ead7",628,["D",1788838992959],"# Homestead Hana — Live Opponent Dossier\n\nRole: weak-regression + market-threshold calibration.\n\n## SOURCE_GROUND_TRUTH\nPublic tier-3 policy:\n- buys one NE quadrant;\n- hands: 4 initially, ramps to 8 by day 4;\n- crop mix carrot/tomato/wheat = 40/20/40;\n- seed batch 8, target stock 10;\n- sells tomato → carrot → wheat in chunks of 30; max 3 sell orders;\n- explicit price floors: carrot 15, tomato 25;\n- invest through day 22; tomato planting ends day 17, staples day 25; liquidation from day 28.\n\nDependency probe: no opponent/other-farm read in source. Direct response to our public farm should therefore be absent under matched market/world state.\n\n## OBSERVED\n- MIX11 B21: 8-0 against Hana.\n- Source policy creates a known positive-control structure: price-floor thresholding can be inferred from behavior without opponent-state awareness.\n\n## CALIBRATION HYPOTHESES\nH1 detector should recover price-conditioned sell suppression near carrot=15 / tomato=25.\nH2 direct our-state-awareness posterior should stay near baseline/zero.\nH3 apparent reaction to us can occur indirectly if our actions move shared market prices; paired controls must separate this mediation.\n\n## REQUIRED PROBES\n1. sweep carrot/tomato price around thresholds from fixed prestate;\n2. hold our visible farm different while pinning market/town/prestate to test direct awareness;\n3. pulse market glut then restore price to measure hysteresis (expected mostly threshold-like, not long-memory opponent response);\n4. mirror seats and fresh future forks.\n\n## FALSIFIER\nA profiler that cannot distinguish `direct opponent awareness` from `shared-market mediated response` on Hana is not ready for Broker/Lena/Silas/Cleo.\n\n## CALIBRATION RESULT — fixed-prestate black-box floors\nTwo independent natural prestates passed exact floor tests:\n- CARROT: price 14 -> SELL 0; 15 -> SELL 3; 16 -> SELL 3.\n- TOMATO: price 24 -> SELL 0; 25 -> SELL 1; 26 -> SELL 1.\nThe known chunk cap remains <=30. For both items, forcing shed load to the pressure threshold or moving to liquidation day caused selling below the normal floor, proving the floor is context-gated rather than universal. Raising opponent public cash by +50,000 changed no action in both prestates.\n\nThis is a strong positive control for market-threshold detection and a negative control for direct opponent-cash awareness.\n\n## REAL-SCENARIO STRATIFIED CALIBRATION — V24\n\nNatural official-engine prestates against Rita, seeds 100577/100609, both seats, sampled every 6 in-game hours:\n- direct opponent-cash 199→200: **0/480** action differences; fresh-vs-stateful mismatch 0;\n- CARROT 14→15: **61/480 = 12.71%** action and sell-quantity differences;\n- TOMATO 24→25: **47/480 = 9.79%**.\n\nCARROT threshold activity appeared in sampled EARLY/MID/LATE/CLOSING contexts; TOMATO showed 0/96 EARLY differences but activity later. This does not redefine the exact floor. It shows that a true threshold is operational only when the rest of the real prestate makes that sell decision relevant.\n\nClassification: `MARKET_THRESHOLD_TRUE_CONTEXT_GATED`; `DIRECT_CASH_AWARENESS_NEGATIVE_IN_STRATIFIED_SAMPLE`. New lesson: `KNOWN_THRESHOLD != ALWAYS_ACTIVE`.",[],[175,176,177,178,2522],"86adf070-eedd-4932-a76b-5e5fee4650b0","cleo.md","01a07e48-3be6-726a-80bb-e032bf3322cc",691,["D",1788848846122],"# Closer Cleo — Live Opponent Dossier\n\nRole: current stress ceiling.\n\n## SOURCE_GROUND_TRUTH\nPublic tier-9 strategy:\n- same meta field plan;\n- SELL layer reorders only within slots already allocated to selling;\n- purchases retain their original slots, preserving funding/inputs while optimizing sale ordering.\n\nDependency probe confirms direct other-farm source dependency and market-price/inventory references.\n\n## OBSERVED\n- MIX11 vs Cleo: 0-4 in early B1; B26 0-8 around -101,687 avg margin; B31 0-8 at -104,137.875.\n- Cleo beats Broker and generally beats Silas in recovered public crossplay, though individual small-N slices can vary.\n- Cleo/Silas telemetry previously showed early melon lead and by ~D15 a mixed-crop lead before the later cash advantage; static melon-share mutations on our side still failed, indicating timing/capacity/execution interactions rather than simple share imitation.\n\n## INFERENCE\nCleo's source difference is explicitly about preserving buy slots while reordering sells. The relevant mechanism class is market-slot opportunity cost and funding continuity. This is a stronger hypothesis than “grow more of Cleo's crops.”\n\n## REQUIRED PROBES\n1. instrument exact requested/executed market order positions;\n2. measure purchases preserved/lost under alternative sell reordering;\n3. fixed-prestate slot-order counterfactuals;\n4. cash liquidity trajectory around purchase/sell collisions;\n5. use Cleo as stress regression for any Broker exploit but do not require every specialized Broker route to beat Cleo before it can be matchup-routed.\n\n## FALSIFIER\nA candidate that copies crop composition without reproducing/fixing the binding slot/funding mechanics should not be interpreted as a test of the Cleo mechanism.\n\n## DIRECT CASH-SENSITIVITY CALIBRATION\nCleo did not change its step-0 action under opponent cash 199→200, unlike Broker/Lena/Silas. However a full natural episode first-hit scan found a causal transition at D6 H0 (step 144): below 200 the sell-slot order was FERTILIZER then WHEAT; at/above 200 it became WHEAT then FERTILIZER. This is consistent with the authored controller constant `SELLS_FIRST=false`: Cleo preserves non-sell/buy slots and only reorders within the existing sell positions.\n\nThis gives a useful mechanistic discriminator: `early global WHEAT promotion` characterizes Broker/Lena/Silas in the tested step-0 state, while Cleo's cash-conditioned response can be delayed until a sell-slot competition exists. Online use still requires inference from legal next-state/market effects because the opponent's raw action list is not observable.\n\n## FULL 15-MIN AUDIT UPDATE\n\nAfter canonicalizing historical aliases, direct Cleo>Silas evidence aggregates **69-9 across 11 batches / 78 episodes**, and Cleo>Broker aggregates **41-5 across 7 batches**. These are strong stress/ladder relationships, but the same audit establishes that pairwise strength is not enough to predict the incremental effect on our policy.\n\nCleo should therefore remain the stress ceiling for robustness and a high-value target for `buy_slot_preservation`, `funding_continuity`, late cash sensitivity, closing and strategic-depth traits — not simply a stronger scalar label.\n\n## 2026-09-08 04:02 UTC — observational/causal separation counterproof\nB42 fresh legal trace produced the strongest cash-response association in the family (`cash lift +1.5465`, low-market `+1.4892`), yet the fresh fixed-prestate 199→200 cash intervention produced `0/18` action changes for Cleo. This is a direct counterexample to treating legal cash lift as direct cash-awareness. Classification: B42 cash lift = `DYNAMIC/TIMING/MEDIATION_ASSOCIATION`, not direct-reaction evidence. Cleo remains a high-value stress comparator for separating observational association from causal awareness.\n\n\n## V23 LEGAL TIMING UPDATE — 2026-09-08 04:23 UTC\n\nCleo shows the strongest legal self-action-filtered immediate response signature among tested upper-ladder opponents, but the mechanism is not yet causally identified. B43 MIX11 produced lag0 lift +2.4873 followed by rapid negative/near-zero decay; preregistered B44 fresh heldout reproduced lag0 +2.2374 and the same rapid decay. Under M40_FERT policy-shift OOD B46, lag0 remained positive and much larger than Rita (+1.0135 vs Rita -0.0348), while the magnitude was materially smaller than under MIX11.\n\nA generic fixed-prestate 199→200 probe at step0/D6H0 produced 0/6 Cleo treatment changes, unlike Broker/Lena/Silas 6/6. This is not evidence that Cleo ignores our cash: canonical source-ground-truth indicates context-dependent SELL-slot competition. It means the strong legal cash-event signature cannot be equated with Broker's 199/200 threshold.\n\nClassification: `CLEO_IMMEDIATE_LEGAL_SIGNATURE = REPLICATED_OBSERVATIONAL_CANDIDATE`; `CLEO_DIRECT_CASH_REACTION_MECHANISM = CONTEXT_DEPENDENT / CAUSAL_CONTEXT_NOT_YET_RECOVERED`. Next causal test must recover naturally occurring fixed prestates where Cleo's market-slot ordering actually changes, then measure public/economic consequences without using raw opponent action online.\n\n\n## FIXED-HISTORY CAUSAL CASH CONTEXT — 2026-09-08 04:44 UTC\n\nBecause Cleo maintains internal state, fixed-prestate intervention now reconstructs that context by replaying only the exact prior legal public observation history into a fresh black-box instance before branching the final observation. No `_TRACE`, private replay plan or online raw-action feature is used.\n\nGeneric H0 landmarks were insensitive even across cash `[199,200,1000,10000]`. Conditioning on natural prestates where Cleo black-box produced WHEAT plus another SELL changed the picture: across three fresh worlds and 30 candidates, 199→200 causally changed sell-slot ordering in 7 states, concentrated in D28–D29. Only one natural fork produced an immediate public consequence (Cleo -8 cash / Rosa +7), and controlled WHEAT BUY contention produced 0/7 non-zero effects.\n\nClassification: `OUR_CASH_SLOT_REACTION = CAUSAL_REPLICATED / CLOSING_CONTEXTUAL / LOW_OBSERVED_MATERIALITY`. This mechanism does **not** explain Cleo's much larger replicated legal lag0 cash-event signature. The next causal target is shared-market/price-inventory/funding response rather than further exploitation of the 199→200 threshold.\n\n\n## 2026-09-08 05:35 UTC — fixed-history snapshot-attribution update\n\nUsing exact replay of Cleo's prior legal public-observation history (0 replay mismatches):\n- WHEAT price/inventory single-factor intervention: 0/24 action responses each across market-active MID/LATE/CLOSING contexts;\n- every individual current public market price 0.5x/2x: 0/12 per item;\n- every individual current public market inventory +/-1000: 0/12 per item;\n- broad perceived-opponent-cash curve 0→50k: only 199→200 fired (2/12), with 0 immediate public-effect pairs; no broader cash-elasticity regime;\n- one-shop removal/duplication: 0/32 per direction.\n\nClassification: `INSTANTANEOUS_SNAPSHOT_REACTIVITY_AS_LARGE_LAG0_CAUSE = FALSIFIED_IN_TESTED_CONTEXTS`. This strengthens separation between Cleo's replicated legal lag0 signature and its contextual 199→200 sell-slot mechanism. Highest-EVSI unresolved dimensions are path/history dependence, temporal/order-slot/funding interaction and endogenous timing/trajectory mediation. Tested-context nulls do not imply universal absence.\n\n\n## 2026-09-08 06:02 UTC — funding/order trait\n\nCorrected V11 separates requested queue from official execution: Cleo was structurally funding-sensitive at sampled D12 states (2/2 mirrored seats) but not D24. V12 found two states with both SELL and spend operations; moving SELL to the end changed execution in both low-cash arms and not at natural cash. Fresh V13 seeds 108011/108023/108037 replicated: treatment fired 6/12 sampled prestates, and every fired treatment produced a low-cash order effect (6/6), with natural-cash control 0.\n\nTrait update: `funding_order_self_financing = CAUSAL_REPLICATED_WHEN_MIXED_QUEUE / SPARSE_CONTEXT_RELATIVE_TO_BROKER / LOW_CASH_CONDITIONAL`. This is not evidence that Cleo's natural advantage is caused by funding order; sampled natural cash was non-binding.\n\n\n## 2026-09-08 06:23 UTC — early-capacity composition\nFresh legal-public census `[108501,108513,108527]` places Cleo on almost the same D3→D12 productive sequence as Broker/Lena/Silas: D3 19 crops (11 WHEAT/8 MELON), 4 animals; D6 18 crops (6 WHEAT/10 MELON/2 STRAWBERRY), 6 animals; D9 29 crops (6 WHEAT/11 MELON/12 STRAWBERRY), 11 animals/2 land; D12 ~37.7 crops (~5 WHEAT/6 MELON/26.7 STRAWBERRY), 14 animals/3 land. Classification: `LEGAL_PUBLIC_DESCRIPTIVE_TOP_FAMILY_CAPACITY_SIGNATURE / NOT_CAUSAL`. The common macro signature does not erase Cleo-specific timing/closing reaction evidence.\n",[],[175,176,177,178,2530],"dde0775f-e4e1-49cf-9b2c-78d8a0bc1762","broker.md","01a07e48-3be6-726a-80bb-df4914be7e98",839,["D",1788916985736],"# Broker Bea — Live Opponent Dossier\n\nRole: current main competitive boundary.\n\n## SOURCE_GROUND_TRUTH\nPublic tier-6 strategy:\n- shared meta field plan: approximately 8 cows / 5 sheep / 6 strawberries across three quadrants with a full crew;\n- opportunistic wheat market schedule;\n- wheat purchase timing depends on the **opponent's cash position** rather than fixed cadence.\n\nDependency probe confirms direct other-farm source dependency plus market-price/inventory references. Broker is therefore a valid positive-control family for direct opponent-state awareness once the weak-trio detector has passed.\n\n## OBSERVED\n- MIX11 remains 0-win versus Broker on all current tested batches.\n- B31 paired control after retry: MIX11 0-8, avg margin -76,344.625.\n- PASTURE_RAMP B31: 0-8, -96,945.25; paired delta -20,600.625 -> falsified.\n- Public crossplay consistently places Broker above Rita and below Lena/Silas/Cleo in the recovered ladder evidence, while exact small-N outcomes are not deterministic.\n- Current failure trajectory frequently begins around D12→D18 capital/conversion windows and compounds later; this is a mechanism-search region, not yet a proven single cause.\n\n## SOURCE/OBSERVATION BRIDGE\nBroker source says wheat buying is conditioned on opponent cash. The online profiler should therefore attempt to infer a reaction surface:\n`P(buy_wheat | our_cash, wheat_price, market_inventory, day, own_state)`.\n\nThe causal question is not “does Broker observe us?” (source says yes) but:\n- which cash thresholds/relative gaps matter;\n- how strong the effect is versus price/inventory;\n- whether timing/hysteresis changes by day/regime;\n- whether our deliberate cash-path changes can improve our win probability after treatment cost.\n\n## REQUIRED PROBES\n1. fixed-prestate fork varying only opponent-visible cash where legally constructible;\n2. matched wheat-price/inventory controls;\n3. estimate logistic/QRE-style buy probability and RLS reaction slope;\n4. CUSUM for regime changes in opponent-cash sensitivity;\n5. active-probe EVSI versus economic cost;\n6. mirrored seats/fresh heldout worlds after any exploit route is learned.\n\n## FALSIFIERS\n- If a supposed Broker response disappears when wheat market state is matched, it was market mediation/confounding.\n- If a local fixed-prestate effect does not transfer heldout, classify LOCAL/WORLD_REGIME rather than universal.\n- Do not reuse PASTURE_RAMP, JIT pasture, static M40/50/60, D9-D12 deployment pulses, generic D21/D24 liquidation, LAND_AFTER6_CLEAN or PORTFOLIO_D6_244 as unqualified mechanisms; they are negative knowledge.\n\n## DIRECT OPPONENT-CASH MECHANISM — causal positive control\nPublic authored market-layer source exposes `WHEAT` promotion threshold at opponent money 200. A fixed-prestate black-box probe recovered the threshold exactly at step 0: opponent cash 199 kept the first WHEAT sell at market slot 5; cash 200 moved WHEAT sells to slots starting at 0; 201 matched 200. Changing Broker's own cash 199→200→201 while opponent cash stayed 199 did not produce that threshold transition in the same prestate.\n\nA second fixed-prestate market-impact screen scanned one natural episode and found 33 states where the 199/200 treatment changed Broker's WHEAT slot position. With a controlled 5-unit opponent WHEAT buy, 6/33 states produced non-zero purchase-cost effects; the largest observed saving from being at/above the threshold was 4 coins (D6 H18, cost 139→135 for the same 5 units). Therefore the mechanism is real but the measured immediate economic magnitude is small relative to the ~76k average B31 loss. Classification: `CAUSAL_REAL_LOW_MAGNITUDE_IN_TESTED_PURCHASE_SCREEN`; do not mistake direct awareness for the primary boundary cause.\n\n## FULL 15-MIN AUDIT UPDATE\n\nAcross the persisted non-selftest sidecar history through the 2026-09-08 audit cutoff, MIX11 is **0-116** against Broker over 15 batches (weighted avg margin about -104,409.5) and M60_FERT is **0-92** (about -84,133.76). Every major tested static/macro Broker-targeted family remained winless in its sampled Broker worlds.\n\nNew trait interpretation: this is strong evidence for **outcome-level `our_policy_shift_robustness` across the tested policy basket**, not evidence that Broker adapts online to each policy. The aggregate therefore strengthens the case for deeper reaction/operations/market-slot/capital-conversion study while explicitly leaving `adaptability`, most direct-awareness dimensions, hysteresis, history sensitivity, bait susceptibility and `models_us` largely UNKNOWN.\n\nHard rule: `OUR_POLICY_SHIFT_ROBUSTNESS_OF_OUTCOME != OPPONENT_ADAPTATION`.\n\n## REAL-SCENARIO CASH-THRESHOLD CALIBRATION — V24\n\nOn experimental branch `kaggriculture-autolab@reaction-attribution-v1`, the known 199→200 opponent-cash threshold was rescreened across **1,440 sampled natural official-engine prestates**: Broker vs Rosa/Rita/Cleo, seeds 100577/100609, both seats, one sample every 6 in-game hours.\n\nResults:\n- direct Broker action branch changed in **86/1,440 = 5.97%** sampled prestates;\n- all 86 differences changed WHEAT sell position;\n- applying those alternative Broker actions to the same official prestate changed a selected immediate cash/WHEAT-market digest in **40/86 = 46.51%**;\n- EARLY: 50/288 action differences, 36 immediate digest differences;\n- MID: 36/576, 4 immediate differences;\n- LATE: 0/432 in this sample;\n- CLOSING: 0/144 in this sample.\n\nA fresh-vs-stateful falsifier produced **0 mismatches / 1,440** sampled Broker prestates, supporting the fresh-source counterfactual within this sampled regime. Weak-trio direct-cash controls stayed 0/1,440.\n\nClassification: `DIRECT_CASH_AWARENESS_CAUSAL_REAL_CONTEXT_GATED`. Do not interpret the 5.97% as a natural trigger probability: it is the fraction of sampled natural prestates in which an artificial fixed-prestate 199→200 public-cash switch changed Broker's action function. Likewise, raw phase/shop stratification is not yet proof that phase or shop duplication causally moderates the threshold.\n\nNew rules: `MECHANISM_PRESENT != MECHANISM_ACTIVE_IN_EVERY_PRESTATE`; `ACTION_BRANCH_DIFFERENCE != IMMEDIATE_ECONOMIC_STATE_DIFFERENCE`.\n\n## REACTION-ATTRIBUTION UPDATE — B39/B40 / 2026-09-08\n\nA fresh fixed-prestate black-box perception intervention independently re-recovered Broker's opponent-cash threshold after requiring the canonical positive-control contexts. Across 18 fresh prestates (3 seeds × 6 landmarks), changing our visible cash 199→200 changed Broker's emitted offline-teacher action in 6 states; Rosa, Hana and Mateo changed in 0/18 each. This strengthens `our_cash_sensitivity` as a real Broker trait under tested contexts.\n\nThe same forks produced **no immediate public next-state material difference in 18/18 tested baseline-action states**. This does not contradict the earlier 33-state market-impact screen (6 non-zero effects; max 4 coins), because the effect requires contention/market context. Classification remains: `DIRECT_AWARENESS_CAUSAL_REAL / MATERIAL_EFFECT_LOW_OR_CONTEXT_DEPENDENT`.\n\nTwo independent legal-observation-only baseline batches also produced a directional Broker cash-response association distinct from the weak-trio group:\n- B39 legal trace: cash-event lift about +0.362; low-market subset +0.380.\n- B40 fresh legal trace: cash-event lift about +0.303; low-market subset +0.250.\n\nThese lifts are **observational detector candidates only**. They are not causal reaction magnitude, calibrated posterior, or proof that the online student can identify the mechanism under policy/OOD shift.\n\nFalsification lesson: an initial fixed-prestate detector that omitted step 0 / D6H0 failed its Broker positive control. Failure class `POSITIVE_CONTROL_CONTEXT_OMISSION`; future null reaction claims are invalid unless the detector first passes known positive and negative controls.\n\nNext discriminant: preserve the same fixed prestate while adding controlled WHEAT contention so the 199/200 action reorder has an opportunity to create a measurable public market consequence. Only after that should the cash mechanism be considered calibrated for legal consequence recovery; it still should not be treated as the main ~100k boundary mechanism.\n\n\n## 2026-09-08 03:55 UTC — family cash-awareness update\nFresh fixed-prestate family probe `[106501,106513,106527]`: Broker changed action at our perceived cash 199→200 in 6/18 identical prestates, but immediate material public next-state effect was 0/18. Lena and Silas reproduced the same 6/18 pattern; Rita, Cleo and the weak trio were 0/18. Therefore the threshold is reclassified from potentially Broker-distinctive to `CAUSAL_FAMILY_TRAIT_BROKER_LENA_SILAS`, context-dependent and low-materiality. Fresh controlled WHEAT mediation across four worlds produced only 18/133 nonzero effects, max 4 coins. Do not use this trait as the primary boundary-gap explanation or unique identity feature.\n\n\n## LEGAL ACTIVATION STUDENT — FROZEN FRESH HELDOUT\n\nThe first deployable-feature student for Broker's 199→200 cash-awareness **activation regime** has survived a true frozen heldout. Training used only seeds 106001/106013/106027/106039 (960 sampled legal-context rows, threshold 0.78 fixed on train). Heldout seeds 106051/106063/106079 were not used for fitting.\n\nHeldout: 720 rows / 42 positives, AUC 0.9813; TP 36, FP 24, TN 654, FN 6; precision 0.60, recall 0.8571, specificity 0.9646, balanced accuracy 0.9109, F1 0.7059. Each fresh seed independently replicated 12/8/218/2 confusion counts. Six heldout rows crossed the current OOD screen max |train-z| >6.\n\nClassification: `LEGAL_STUDENT_CAN_DISCRIMINATE_DIRECT_CASH_MECHANISM_ACTIVATION_HELDOUT_PASS`.\n\nScope restriction: this predicts whether the **offline teacher counterfactual** would change Broker's action branch at the sampled context. It does not establish that the natural opponent cash is near 200, that inducing such a state is cheap, or that exploiting the branch increases our terminal W/L. Strategic value remains UNVERIFIED.\n\n## V23 LEGAL TIMING UPDATE — 2026-09-08 04:22 UTC\n\nBroker's direct 199→200 cash-awareness mechanism is now independently re-recovered and economically bounded. Across fresh controlled WHEAT-contention seeds `[106401,106413,106427]`, 101 treatment-fired/comparable states yielded 13 nonzero controlled purchase-cost effects (12.87%), maximum 4 coins. This strengthens the causal trait while simultaneously downgrading its plausibility as the principal competitive-gap mechanism.\n\nWith V23 legal self-action memory, transitions where our own market flow is nonzero can be excluded. On B43, Broker's self-silent cash-event response lift was +0.3565 at lag0 and +0.1989 at lag2. B44 preregistered fresh heldout reproduced +0.3815 at lag0 and +0.2021 at lag2. Under M40_FERT policy-shift OOD B46, the qualitative pattern persisted (+0.2321 lag0, +0.2464 lag2), although magnitude changed.\n\nClassification: `IMMEDIATE_PLUS_LAG2_LEGAL_SIGNATURE = REPLICATED_OBSERVATIONAL_CANDIDATE`. It is not a causal latency curve, not a calibrated posterior, and not sufficient to attribute market-mediated versus direct opponent response. Fixed-prestate direct-awareness remains separately causal and low-materiality.\n\n\n## EXPLOIT PRIORITY CORRECTION — NATURAL SUPPORT + UPSTREAM THROUGHPUT\n\nSix fresh seeds `[106131,106147,106159,106171,106183,106199]`, both seats, produced 12 MIX11-vs-Broker games / 8,628 decision rows. Our natural MIX11 cash never reached <=1000; minimum was 1301. Therefore Broker's causal cash=200 mechanism is real but outside observed MIX11 natural support and is reclassified `CALIBRATION_POSITIVE_CONTROL / LOW_DIRECT_EXPLOIT_PRIORITY`. Forcing the signal would require deliberate self-impoverishment; no positive EV evidence currently supports that.\n\nA separate fresh 8-game audit `[106211,106223,106237,106249]` shows the economically material boundary developing after D9. We lead in mean cash gap D0-D6 and are roughly flat D6-D9; then the gap deteriorates -4,892 D9-D12, -11,273.5 D12-D15, -9,679.75 D15-D18, -18,198.625 D18-D21, -17,792.375 D21-D24, -21,441.875 D24-D27 and -23,402.375 D27-D30. Offline action telemetry shows Broker scaling labor/land/WHEAT procurement before very large multi-product sale throughput. This is descriptive evidence only.\n\nPriority hypotheses now: (H1) labor-scale timing, (H2) WHEAT procurement/feed continuity, (H3) land timing, (H4) multi-product conversion, (H5) market-order scheduling/SELLS_FIRST. Each requires an upstream matched causal intervention with treatment detector and fresh heldout. `SELLS_MORE != CAUSE_OF_WIN`.\n\n## 2026-09-08 05:35 UTC — fixed-history snapshot-attribution update\n\nFresh controlled evidence with valid positive/negative controls narrows Broker's reaction model:\n- individual current public price perturbations across all 8 market items: 0/12 action responses per item in sampled market-active MID/LATE/CLOSING contexts;\n- individual current public inventory perturbations +/-1000 across all 8 items: 0/12 per item;\n- broad perceived-opponent-cash curve 0→50k: only the already-known 199→200 branch fired (4/12 sampled contexts); no additional threshold and 0 immediate public-effect pairs in these contexts;\n- simple one-shop removal/duplication: 0/32 per direction.\n\nClassification: `CURRENT_SNAPSHOT_PRICE/INVENTORY/SHOP_ELASTICITY = NO_ACTION_RESPONSE_IN_TESTED_CONTEXTS`, never universal absence. Known 199→200 direct-awareness remains causal/calibration-only and low materiality. The primary open reaction lane shifts toward legal history/path dependence, timing/order-slot/funding interactions and endogenous trajectory structure.\n\n\n## 2026-09-08 06:02 UTC — funding/order trait\n\nTeacher-only fixed-prestate evidence isolates a robust execution-order trait. At sampled D12/D24 states, Broker's requested queue remained structurally executable from starting cash 0/200/1000/5000 (`corrected V11`: 0/4 structural funding differences). In V12, moving SELL operations behind HIRE/BUY while preserving the exact action multiset changed execution in every low-cash treatment (4/4) and never at natural cash. Fresh V13 seeds 108011/108023/108037 replicated this across mirrored seats: treatment fired 12/12, low-cash order effect 12/12, natural-cash control 0.\n\nTrait update: `funding_order_self_financing = CAUSAL_REPLICATED / FRESH_HELDOUT_SUPPORTED / LOW_CASH_CONDITIONAL`. This establishes a real SELL-before-spend funding mechanism, but does not show that it is materially active in Broker's naturally high-cash sampled states or explains the competitive gap. Raw requested actions remain offline teacher evidence only.\n\n\n## 2026-09-08 06:23 UTC — legal public early-capacity composition\nFresh legal-public composition census `[108501,108513,108527]` shows Broker following a stable early productive sequence under MIX11 worlds: D3 ≈ 19 crops (11 WHEAT/8 MELON), 4 animals (3 COW/1 SHEEP); D6 ≈ 18 crops (6 WHEAT/10 MELON/2 STRAWBERRY), 6 animals; D9 ≈ 30 crops (7 WHEAT/11 MELON/12 STRAWBERRY), 11 animals and 2 land; D12 ≈ 38.7 crops (6 WHEAT/6 MELON/26.7 STRAWBERRY), 14 animals and 3 land. Lena/Silas matched this sequence essentially exactly in the same fresh worlds; Cleo was very close. Rita did not: it remained WHEAT-only with 1 land through D12. Classification: `LEGAL_PUBLIC_DESCRIPTIVE_TOP_FAMILY_CAPACITY_SIGNATURE / NOT_CAUSAL / NOT_SOURCE_GROUND_TRUTH`. Do not convert this directly into a policy target; prior naive early land/herd/melon/portfolio formulations remain negative knowledge.\n\n\n## B53/B54 update — 2026-09-08 07:34 UTC\n\nFresh B53 replicated Broker's public D3→D12 productive composition exactly with Lena/Silas episode-by-episode in the tested worlds; this remains a family fingerprint/trajectory, not a mechanism.\n\nB54 tested a standalone staged MELON→STRAWBERRY transfer while preserving MIX11 land/labor/herd/funding/sell/closing logic. Treatment fired in all 6 Broker episodes and harmed every paired episode: MIX11 0-6/-94384 versus staged 0-6/-113295.17, paired delta -18911.17 (0/6 positive). The Broker gap was already worse by D9 and remained worse in 6/6 at D12/D15/D21/D24/D27/D29.\n\nTrait implication: Broker's portfolio composition cannot be treated as an independently transferable recipe. Capacity/headroom/throughput/funding remain candidate moderators, not established causes. Do not infer that Broker is reacting to the staged treatment; B54 is an our-policy causal screen, not an opponent-reaction claim.\n\n\n## B55 capacity×portfolio moderation — 2026-09-08 07:52 UTC\nFresh paired factorial `[108801,108813,108827]` kept all four arms 0-6 against Broker. Limited D6 headroom alone improved diagnostic mean margin versus same-epoch MIX11 (+17714.17), but combining that headroom with the staged portfolio produced a strongly negative interaction (`I=-20673.67`). Thus `capacity headroom rescues staged portfolio` is rejected in this formulation. This is an our-policy causal screen, not evidence that Broker reacted to either treatment. CAP2_D6 remains only a one-batch reopening candidate pending fresh Broker+Rita B56; no opponent reaction trait is promoted from B55.\n\n\n## B56 CAP2_D6 fresh falsification — 2026-09-08 08:20 UTC\nFresh paired Broker gate `[108901,108913,108927]`: MIX11 0-6/-116285.00; CAP2_D6 0-6/-128129.33; aggregate directional delta -11844.33, errors 0. This falsifies the one-batch B55 headroom improvement as a transferable Broker treatment. It is an our-policy result, **not** evidence that Broker reacted to capacity expansion. `CAP2_D6` is closed in this formulation; Broker reaction traits remain unchanged.\n\n\n## 2026-09-08 controlled-history reaction update\n\n**Confirmed causal capability, controlled history:** fresh fixed-history V2 (`[109241,109253,109267]`, mirrored) produced 6/138 action effects in Broker, all 6 threshold-specific (`TWO!=ZERO`, `ONE==ZERO`), all premium-SELL deltas and all with immediate official-engine economic consequences. Example D12H4: two-hit similarity history adds `SELL MILK 9` to a queue that otherwise has `SELL STRAWBERRY 6`, changing immediate Broker cash delta from 0 to +1289 in the tested state. This is direct evidence that Broker can model a repeated public-build similarity history and alter market behavior accordingly. Raw action/source is offline teacher evidence only.\n\n**Natural activation under current MIX11: absent in tested fresh worlds.** Fresh `[109281,109293,109307]`, mirrored, 180 authored update checkpoints: conservative legal-public distance lower bound >4 on 180/180, min 8, clone-confidence optimistic upper-bound 0 in all 6 games. Therefore the confirmed front-run mechanism is not a current MIX11-loss explanation in these worlds. Keep `models_us/history_sensitive_front_run = CAPABILITY_CONFIRMED`, but `current_MIX11_activation = ABSENT_UPPER_BOUND`.\n\n\n## 2026-09-08 13:15 UTC — WOOL/composition reaction scope\nFresh fixed-prestate controls narrow two direct-awareness candidates. On `[110401,110413,110427]`, changing current public WOOL price or inventory at D18/D21/D24/D27 changed Broker action 0/12 each; the same apparatus recovered Broker cash 199→200 in 3/6 positive-control contexts. On `[110201,110213,110227]`, isolated public crop-mix and animal-mix perception changes changed Broker action 0/15 each across D3/D6/D12/D18/D24, while cash positive control fired 6/6. Therefore `our_crop_mix_sensitivity`, `our_animal_mix_sensitivity` and current-WOOL-snapshot response are `NO_ACTION_RESPONSE_IN_TESTED_LANDMARKS`, not universal absence. The family WOOL footprint remains shared-market/world-regime evidence, not direct Broker reaction proof.\n\n## 2026-09-08 21:19 UTC — repaired WOOL lifecycle mediation candidate\nAfter exact lifecycle measurement parity + mutation-control PASS, fresh `[111201,111213,111227]` D18-D27 telemetry contains 138 nonzero Broker public SHEEP yield transitions and 54 pending-care transitions. Adding these public lifecycle deltas to a preliminary turn-level WOOL-inventory model increased R² by 0.00809 overall and positively in every mirrored world (~0.0051..0.0137), versus only 0.000431 for Rita. Classification: `BROKER_PUBLIC_SHEEP_LIFECYCLE_TO_WOOL_INVENTORY = FRESH_INCREMENTAL_ASSOCIATION_CANDIDATE / NOT_CAUSAL / NOT_EXECUTED_OUTPUT`. This does not establish direct reaction, resource denial, or exploitability; exact demand removal and opponent latent sale/collection remain unresolved mediators.\n\n\n## 2026-09-08 22:2x UTC — exact-demand residual qualification\nAfter engine-parity validation of executed WOOL town demand, fresh residual mediation `[111401,111413,111427]` gives Broker opponent-lifecycle incremental ΔR² **0.001679** versus Rita ~**0.000001**. The sign of incremental information is positive overall but support is heterogeneous by world (near-zero in seed 111427; larger in 111413). This downgrades the prior preliminary ΔR² scale while preserving `upper-family > Rita` discrimination. Status remains `FRESH_ASSOCIATION_CANDIDATE / NOT_CAUSAL / NOT_EXECUTED_OPPONENT_OUTPUT`; our requested WOOL SELL still requires executed-quantity parity before conservation attribution.\n\n\n## 2026-09-08 23:2x UTC — full WOOL residual qualification\nAfter exact town-demand parity and independent own-WOOL-sell interpreter parity, fresh `[111701,111713,111727]` gives Broker lifecycle incremental ΔR² only **0.000303** on the full residual. It remains positive in every mirrored world (~0.000917, ~0.000545, ~0.000106 by seed) but decreases materially across worlds. Classification: `BROKER_SHEEP_LIFECYCLE_TO_FULL_WOOL_RESIDUAL = SMALL_FRESH_ASSOCIATION / WORLD_HETEROGENEOUS / NOT_CAUSAL`. This further weakens WOOL as a primary boundary-gap explanation; direct reaction/resource denial/exploitability remain unproven.\n\n\n## 2026-09-09 00:2x UTC — WOOL branch qualification\nTeacher fixed-prestate conservation forks establish that executed opponent WOOL supply enters shared inventory one-for-one and can depress price nonlinearly at larger quantities, but this is a forced engine mechanism rather than Broker policy evidence. Fresh sparse D18/D21/D24/D27 raw-action checkpoints showed 0 Broker WOOL sells in 24 sampled checkpoint rows; because the telemetry is checkpoint-sparse, this is only counterevidence against a continuously-obvious sell story, not proof of zero turn-level supply. Combined with the tiny/world-heterogeneous full-residual lifecycle association, Broker WOOL supply is deprioritized as a primary boundary explanation pending exact turn-level teacher accounting. Direct reaction/resource denial remain unproven.\n\n\n## 2026-09-09 01:2x UTC — late MILK conversion diagnostic\nFresh unchanged-MIX11 ranking `[112001,112013,112027,112039]` identified MILK as the largest non-WOOL D18-D27 Broker-vs-Rita realized-sale-value differential: Broker context executed 632 MILK / value 103,816 versus Rita 770 / 120,737, while Broker realized public price was slightly higher. Independent `[112101,112113,112127,112139]` again showed lower Broker-context MILK sales (75.75 vs 100/game), lower own private MILK shed (3.55 vs 5.34), fewer own cows (7.87 vs 8.70) and lower public own cow yieldSum (6.58 vs 7.90), despite higher mean MILK price (+25.62). World heterogeneity is substantial, so classify `MILK_OUTPUT_AVAILABILITY_PATH = REPLICATED_DIRECTIONAL_DIAGNOSTIC / WORLD_HETEROGENEOUS / NOT_CAUSAL`. This is evidence about **our trajectory under the Broker matchup**, not proof Broker reacts to or suppresses MILK, and not permission to reopen rejected early-herd mutations.",[],[175,176,177,178,2538],"4b11e1b3-8222-4963-9611-38595a925298","INDEX.md","01a07e48-3be6-726a-80bb-d96abfbf39f6",589,["D",1788835004515],"# KAG Opponent Profiles — Live Dossier Index\n\nStatus: V1_RECOVERED_AND_SOURCE_CALIBRATED\n\n**Mandatory schema:** every dossier inherits `../KAG_OPPONENT_CAPABILITY_TAXONOMY.md`. Traits not explicitly measured in a dossier remain `UNKNOWN`; they do not disappear from project scope. Before profiling, read `../../00_READ_FIRST.md`.\n\nEach profile separates:\n- `SOURCE_GROUND_TRUTH`: facts read from public reference-agent source, useful only for laboratory calibration;\n- `OBSERVED`: behavior measured from runs/telemetry;\n- `INFERENCE`: model-derived claim requiring uncertainty;\n- `UNKNOWN`: not yet measured;\n- `COUNTEREVIDENCE`: contradictions/falsifications.\n\nDo not leak source-ground-truth assumptions into the competition classifier as if they were observed online. Public source is a teacher/calibration oracle; the student must recover the same distinctions from legal observations.\n\n## Current ladder / targets\n\n| Opponent | Source strategy family | Direct other-farm source dependency | Current role |\n|---|---|---:|---|\n| Rotation Rosa | fixed 3-crop rotation | no | cheap calibration / weak regression |\n| Homestead Hana | land+labor staples + price floors | no | cheap calibration / market-threshold positive control |\n| Melon Mateo | premium melon + strict price floor | no | cheap calibration / market-threshold strong positive control |\n| Rancher Rita | livestock/wheat feed chain | not proven by dependency probe | regression / transition opponent |\n| Broker Bea | meta field + opponent-cash-timed wheat buying | yes | main boundary |\n| Ledger Lena | meta field + sell-order trade-off | yes | upper ladder |\n| Slotter Silas | meta field + premium SELL slot ordering | yes | upper ladder |\n| Closer Cleo | meta field + sell reordering constrained to existing SELL slots | yes | stress ceiling |\n\n## Calibration insight\n\nThe weak trio is ideal for validating the profiler because source truth gives discriminating controls:\n- Rosa/Hana/Mateo do **not** read the other farm in their source: an `our-state-awareness` detector should not hallucinate direct opponent-state reactivity when market/world state is controlled.\n- Hana/Mateo have explicit price floors; Mateo's melon floor is 120 and sell chunk 12, Hana's carrot/tomato floors are 15/25 and sell chunk 30. A market-threshold detector should recover these effects from behavior.\n- Static early signatures already distinguish Mateo and Rita, while Broker/Lena/Silas were identical in tested public snapshots through D9; top-tier classification therefore requires temporal/action-response features.\n\n## Profiles\n- `rosa.md`\n- `hana.md`\n- `mateo.md`\n- `rita.md`\n- `broker.md`\n- `lena.md`\n- `silas.md`\n- `cleo.md`\n\n## Update rule\nEvery material new run appends raw event/evidence references before updating posterior claims. Never overwrite contradictory evidence; reduce confidence or split by context/regime.",[],[175,176,177,178,2546],"f835b737-188f-40d3-9404-55c9f090a2da",["D",1788825058278],[240,241,242,243,244,245,246,247,248],[175,176,177,178],"6a21222d-452e-4d8b-a7c8-13544e8121fd",["D",1788825058278],[239,249,250,251,252,253,254,255,256,257,258,259],[175,176,177],"fae84ad2-fbde-48bd-a520-1d83a4fb556b","KAG_HOURLY_CHECKPOINT_20260908_2119.md","753b22ea-abcb-11f1-822a-1607ee4eb77e","fff89766-aa98-11f1-a2f0-1607ee4eb77e",816,["D",1788902575857],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 21:19 UTC\n\nStatus: PASS / MATERIAL_MEASUREMENT_PARITY_AND_MEDIATION_SIGNAL / NOT_CHAMPION_PROOF\n\n## KAG_CONTEXT_PREFLIGHT\nREAD_FIRST=READ; STRATEGIC_INTELLIGENCE=READ; CAPABILITY_TAXONOMY_190=READ; CURRICULUM/NEURON/JOINED_PLAN=READ; OFFICIAL_RULES/RT=READ; CURRENT_STATE/NEGATIVE_KNOWLEDGE=READ; 15MIN_REANALYSIS/SCENARIO_COVERAGE/RECOVERY/LATEST_AUDIT=READ; OPPONENT_INDEX + Broker/Rita/Cleo relevant dossiers=READ; EXPERIMENT_MATRIX/PROFILE_CAL=READ; HOURLY_INDEX + checkpoint/manifest 20:21=READ; SIDECAR queue/runs/states/artifacts/evidence epochs=RECONCILED (241 DONE / 5 FAILED / 0 READY/RUNNING; no post-20:21 main jobs/artifacts). Final **PASS**.\n\n## Material change\nTrue teacher same-turn parity validator `lifecycle_same_turn_parity_v8.ts` used fresh `[111101,111113,111127]`, mirrored, Broker/Rita. Raw legal public SHEEP tiles and scenario-style aggregation matched exactly for count/yieldSum/pendingCare/unfedStress on **4314/4314 rows per target**, mismatch=0, errors=0. Deliberate +1 corruption of opponent yield was caught **4314/4314** per target. Classification: `PUBLIC_SHEEP_LIFECYCLE_AGGREGATION_PARITY = FRESH_EXACT_PASS / MUTATION_CONTROL_PASS`.\n\nRepaired `wool_temporal_conservation_v4.ts`, fresh `[111201,111213,111227]`, mirrored: Broker has 138 opponent-yield and 54 opponent-pending transitions in D18-D27; Rita 74/50. Preliminary base model uses our requested WOOL sells, our lifecycle deltas and demand-exposure timing. Adding opponent lifecycle deltas gives Broker ΔR² **0.00809199**, positive in every world (~0.00511..0.01370); Rita ΔR² **0.00043070**. Inventory→price inverse sign remains 90/90 Broker and 116/116 Rita.\n\n## Traits / hypotheses / falsifications\nNo taxonomy dimension added/removed; universe remains 190. Measurement hypothesis promoted only: lifecycle aggregate parity is validated. `BROKER_PUBLIC_SHEEP_LIFECYCLE_TO_WOOL_INVENTORY = FRESH_INCREMENTAL_ASSOCIATION_CANDIDATE / NOT_CAUSAL / NOT_EXECUTED_OUTPUT`. Direct-vs-mediated attribution, reaction magnitude, resource denial, strategic substitution and WOOL exploit remain UNKNOWN/unproven. V4/V4b false-zero remains falsified as consumer-path bug. No negative knowledge reopened.\n\n## League / jobs\nFresh diagnostic competitive context: MIX11 0-6 Broker / -127512.67; 4-2 Rita / +2684.33. These are NOT_CHAMPION_PROOF. Champion unchanged; working lineage MIX11; boundary Broker Bea; stress Closer Cleo; regression Rancher Rita. Main sidecar remains 241 DONE / 5 FAILED / 0 READY/RUNNING; no filler jobs enqueued.\n\n## Artifacts updated\nTeacher: `lifecycle_same_turn_parity_v8.ts`, `wool_temporal_conservation_v4.ts`. Canonical: `KAG_CURRENT_STATE.md`, `KAG_STRATEGIC_INTELLIGENCE.md`, Broker dossier, this checkpoint + manifest + HOURLY_INDEX. Sidecar context/hypothesis states reconciled after persistence.\n\n## First incomplete step\nBuild an **engine-validated executed WOOL demand ledger** (YARN_STORE copies + town-center exact removal semantics/timing) with parity/mutation controls, then form a conservation residual from shared inventory + our known contribution. Only after that test whether opponent public lifecycle adds stable explanatory power to residual supply across fresh Broker/Rita/upper-family worlds and whether the mediated price effect is economically material. No WOOL policy mutation before this gate.\n",[],[175,176,2570,2562],"hourly","74337ec4-6749-4b30-a7bf-f2bd3f3909d7","KAG_HOURLY_CHECKPOINT_20260908_2021.md","3b99fe60-abc3-11f1-847b-1607ee4eb77e",810,["D",1788899043173],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 20:21 UTC\n\nStatus: PASS / MATERIAL_MEASUREMENT_REPAIR / NOT_CHAMPION_PROOF\n\n## KAG_CONTEXT_PREFLIGHT\nREAD_FIRST=READ; STRATEGIC_INTELLIGENCE=READ; CAPABILITY_TAXONOMY_190=READ; CURRICULUM/NEURON/JOINED_PLAN=READ; OFFICIAL_RULES/RT=READ; CURRENT_STATE/NEGATIVE_KNOWLEDGE=READ; 15MIN_REANALYSIS/SCENARIO_COVERAGE/RECOVERY/LATEST_AUDIT=READ; OPPONENT_INDEX + Broker/Rita dossiers=READ; EXPERIMENT_MATRIX/PROFILE_CAL=READ; HOURLY_INDEX + checkpoint/manifest 19:35=READ; SIDECAR queue/runs/states/artifacts/evidence epochs=RECONCILED (241 DONE / 5 FAILED / 0 READY/RUNNING; no post-19:35 main jobs). Final **PASS**.\n\n## Material change\nFresh V27 scenario telemetry `[111001,111013,111027]`, mirrored, directly read `scenarioTrace.publicTileLifecycle.{me,opp}.animals.SHEEP`. Signal is strongly nonzero: Broker context over 3,456 D3-D27 rows: own yield transitions 84, opponent 264, own pending-care 142, opponent 144; Rita: 99, 118, 136, 136. Example Broker opponent yield `+6` at D6H0 and `-6` at D6H7.\n\nRoot cause correction: V4/V4b called helper `sh(a?.publicTileLifecycle?.me)` while `sh()` itself dereferenced `.publicTileLifecycle`, producing an empty object. Thus the 19:35 classification `V27 aggregate bug` is superseded by `V4/V4B CONSUMER PATH BUG`. Direct V5 and corrected scenario V7 now agree qualitatively that legal lifecycle signal exists and varies.\n\n## Traits / hypotheses / falsifications\nNo opponent trait promoted. `PUBLIC_SHEEP_LIFECYCLE_SIGNAL = LEGALLY_OBSERVABLE / FRESH_SCENARIO_REPLICATED`. `V27_PUBLIC_TILE_LIFECYCLE = SIGNAL_PRESENT`; `V4_V4B_ZERO = MEASUREMENT_FALSE_NULL / NESTED_CONSUMER_PATH_MISMATCH`. Opponent-output mediation, direct-vs-mediated attribution, resource denial, reaction magnitude, hysteresis and WOOL exploit remain UNKNOWN/unproven. Taxonomy remains 190.\n\n## League / negative knowledge\nChampion unchanged; working lineage MIX11; boundary Broker Bea; stress Closer Cleo; regression Rancher Rita. No negative knowledge reopened; no WOOL mutation authorized.\n\n## Jobs / artifacts\nMain queue unchanged 241 DONE / 5 FAILED / 0 READY/RUNNING. No main jobs enqueued. Teacher-only scripts created/run: `lifecycle_parity_validator_v6.ts` (mutation-control scaffold; not a true direct parity test because compact legalTrace lacks raw tiles) and `lifecycle_sheep_signal_v7.ts` (fresh corrected scenario signal validation).\n\n## First incomplete step\nImplement a **true same-turn direct-observation ↔ scenario aggregate parity hook** inside teacher instrumentation, not via compact `legalTrace`, and require exact equality for SHEEP count/yield/pendingCare/unfed plus a deliberate corrupted-aggregate mutation that the validator catches. After parity PASS, rerun WOOL conservation mediation with repaired yield flow and executed deterministic demand ledger. No policy mutation before that gate.\n",[],[175,176,2570,2572],"6ded8df4-a751-4544-a94d-e6e7458cbb0c","KAG_HOURLY_CHECKPOINT_20260908_1935.md","2cca19c2-abbb-11f1-8485-1607ee4eb77e",805,["D",1788895582369],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 19:35 UTC\n\nStatus: PASS / MATERIAL_MEASUREMENT_ROOT_CAUSE / NOT_CHAMPION_PROOF\n\n## KAG_CONTEXT_PREFLIGHT\nREAD_FIRST=READ; STRATEGIC_INTELLIGENCE=READ; CAPABILITY_TAXONOMY_190=READ; CURRICULUM/NEURON/JOINED_PLAN=READ; OFFICIAL_RULES/RT=READ; CURRENT_STATE/NEGATIVE_KNOWLEDGE=READ; 15MIN_REANALYSIS/SCENARIO_COVERAGE/RECOVERY/LATEST_AUDIT=READ; OPPONENT_INDEX + Broker/Rita relevant dossiers=READ; EXPERIMENT_MATRIX/PROFILE_CAL=READ; HOURLY_INDEX + checkpoint/manifest 18:25=READ; SIDECAR queue/runs/states/artifacts/evidence epochs=RECONCILED (241 DONE / 5 FAILED / 0 READY/RUNNING; no post-18:25 main jobs). Final **PASS**.\n\n## Material change\nV4/V4b fresh scenario-trace checks `[110801,110813,110827,110839]` and `[110851,110863]` again returned zero SHEEP lifecycle deltas through the V27 aggregate. Independent direct official-engine legal observation probe `wool_public_tile_direct_v5.ts` then falsified the implied signal absence: Broker self-play seed 110901 showed 44 SHEEP yield-sum transitions and 24 pending-care transitions per farm across D3-D27, including yield 0→6 at D6H0 and 6→0 at D6H7. Therefore the underlying legal public signal is variable; V27 scenario lifecycle aggregation/consumer is false-zero.\n\n## Traits / falsification\nNo opponent trait promoted. `PUBLIC_SHEEP_LIFECYCLE_SIGNAL = LEGALLY_OBSERVABLE_AND_VARIABLE`. `V27_SCENARIO_LIFECYCLE_AGGREGATE = FALSE_ZERO / MEASUREMENT_BUG / DO_NOT_USE_FOR_MEDIATION`. V3/V4 zero-variation cannot update opponent-output mediation. Direct reaction, resource denial, reaction magnitude and WOOL exploit remain UNKNOWN/unproven. Taxonomy remains 190.\n\n## League / negative knowledge\nChampion unchanged; working lineage MIX11; boundary Broker Bea; stress Closer Cleo; regression Rancher Rita. No negative knowledge reopened; no WOOL mutation authorized.\n\n## Sidecar/jobs\nMain remains 241 DONE / 5 FAILED / 0 READY/RUNNING. No main jobs consumed/enqueued. Teacher-only measurement scripts created/run: `wool_public_yield_variation_v4.ts`, `wool_public_yield_variation_v4b.ts`, `wool_public_tile_direct_v5.ts`.\n\n## First incomplete step\nRepair `tileLifecycleFeature/scenarioFeature` parity so aggregated SHEEP `yieldSum` and `pendingCare` match direct legal observations turn-by-turn; add a mutation/negative control proving the parity validator fails on a deliberately corrupted aggregate. Only after parity PASS rerun fresh WOOL conservation/mediation with executed deterministic demand accounting.",[],[175,176,2570,2580],"76912176-0bfe-4519-bb13-c1fd66730257","KAG_HOURLY_CHECKPOINT_20260908_1825.md","6b9b0386-abb2-11f1-b8ec-1607ee4eb77e",800,["D",1788891822288],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 18:25 UTC\n\nStatus: PASS / MATERIAL_MEASUREMENT_FALSIFICATION / NOT_CHAMPION_PROOF\n\n## KAG_CONTEXT_PREFLIGHT\nREAD_FIRST=READ; STRATEGIC_INTELLIGENCE=READ; CAPABILITY_TAXONOMY_190=READ; CURRICULUM/NEURON/JOINED_PLAN=READ; OFFICIAL_RULES/RT=READ; CURRENT_STATE/NEGATIVE_KNOWLEDGE=READ; 15MIN_REANALYSIS/SCENARIO_COVERAGE/RECOVERY/LATEST_AUDIT=READ; OPPONENT_INDEX + Broker/Rita relevant dossiers=READ; EXPERIMENT_MATRIX/PROFILE_CAL=READ; HOURLY_INDEX + checkpoint/manifest 17:17=READ; SIDECAR queue/runs/states/artifacts/evidence epochs=RECONCILED (241 DONE / 5 FAILED / 0 READY/RUNNING; no post-17:17 main jobs). Final **PASS**.\n\n## Material evidence\nTeacher `wool_temporal_conservation_v3.ts`, fresh `[110701,110713,110727,110739]`, mirrored seats, unchanged MIX11, V27 scenario telemetry, opponent raw action NONE. Broker 0-8/-113330.5; Rita 8-0/+3839.375; errors 0.\n\nV3 corrected the previous demand-support bug: YARN_STORE periodic exposure is nonzero (Broker 288 rows, Rita 432) and town-center timing contributes 72 rows each. But the candidate public production-flow proxy failed its detector: `ourYieldDelta!=0=0/1728` and `oppYieldDelta!=0=0/1728` in both matchups. Thus opponent-yield ΔR²=0 is not interpretable as a causal/mediation null.\n\nInventory/price simultaneous nonzero moves remained inverse-sign Broker 208/208 and Rita 148/148.\n\n## Traits / falsification\nNo opponent trait promoted. `PUBLIC_YIELDSUM_DELTA_AS_WOOL_OUTPUT_FLOW_DETECTOR = FALSIFIED/NO_VARIATION_IN_CURRENT_TELEMETRY`. `DEMAND_EXPOSURE_TIMING = ACTIVE_PROXY / NOT_EXECUTED_DEMAND_LEDGER`. `ZERO_VARIANCE_DETECTOR != ZERO_CAUSAL_EFFECT`. Direct reaction, resource denial, opponent output mediation, reaction magnitude and WOOL exploit remain UNKNOWN/unproven. Taxonomy remains 190.\n\n## League / negative knowledge\nChampion unchanged; working lineage MIX11; boundary Broker Bea; stress Closer Cleo; regression Rancher Rita. No negative knowledge reopened; no WOOL mutation authorized.\n\n## Sidecar\nMain remains 241 DONE / 5 FAILED / 0 READY/RUNNING. No blind refill.\n\n## First incomplete step\nRepair the measurement layer before another mediation regression: inspect legal public pasture lifecycle fields across actual collection transitions and official interpreter demand timing, then build a treatment/detector that demonstrably varies. If no legal public output-flow proxy exists, switch to an engine-validated conservation residual using shared Δinventory, our known executed contribution, and deterministic town/shop demand, explicitly leaving opponent private collection latent.",[],[175,176,2570,2588],"73127a27-8c89-460f-a37e-bcbd142b6487","KAG_HOURLY_CHECKPOINT_20260908_1717.md","d65b69d0-aba9-11f1-902d-1607ee4eb77e",795,["D",1788888135913],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 17:17 UTC\n\nStatus: PASS / MATERIAL_MEDIATION_NARROWING / NOT_CHAMPION_PROOF\n\n## KAG_CONTEXT_PREFLIGHT\nREAD_FIRST=READ; STRATEGIC_INTELLIGENCE=READ; CAPABILITY_TAXONOMY_190=READ; CURRICULUM/NEURON/JOINED_PLAN=READ; OFFICIAL_RULES/RT=READ; CURRENT_STATE/NEGATIVE_KNOWLEDGE=READ; 15MIN_REANALYSIS/SCENARIO_COVERAGE/RECOVERY/LATEST_AUDIT=READ; OPPONENT_INDEX + Broker/Rita/Lena/Silas/Cleo relevant dossiers=READ; EXPERIMENT_MATRIX/PROFILE_CAL=READ; HOURLY_INDEX + checkpoint/manifest 15:20=READ; SIDECAR queue/runs/states/artifacts/evidence epochs=RECONCILED (241 DONE / 5 FAILED / 0 READY/RUNNING; no post-15:20 main jobs). Final **PASS**.\n\n## Material evidence\nNew teacher `wool_temporal_conservation_v2.ts`, fresh `[110601,110613,110627,110639]`, mirrored seats, unchanged MIX11, scenario telemetry FULL, opponent raw action NONE. Broker 0-8/-96493.75; Rita 8-0/+14490.5, errors 0.\n\nD18-D27 turn-level diagnostic: baseline regression for `Δ shared WOOL inventory` used our requested WOOL sells + our public sheep + preliminary town-WOOL marker. Adding opponent public sheep produced ΔR²=0 against Broker and ≈0.000689 against Rita. Broker opponent sheep was constant in the sampled window, so the zero incremental result is partly a support/variation limitation, not universal absence. Inventory/price nonzero simultaneous moves had inverse signs Broker 202/202 and Rita 96/96.\n\n## Traits / hypotheses / falsification\n`WOOL_INVENTORY_TO_PRICE_LINK` strengthened as structural turn-level relation in tested fresh worlds. `PUBLIC_OPPONENT_SHEEP_COUNT_AS_INCREMENTAL_TURN_LEVEL_MEDIATOR` weakened/not supported in this formulation. No direct reaction, resource-denial, substitution, reaction-magnitude or bait/bluff trait promoted. Taxonomy remains 190; unmeasured traits remain UNKNOWN.\n\nCritical falsifier/limitation: `OPPONENT_SHEEP_COUNT != OPPONENT_WOOL_OUTPUT_FLOW`. The preliminary deterministic town/shop demand marker was zero, so V2 is not a completed conservation identity and cannot establish causal mediation. Next test must use actual public pasture-yield/collection transitions and corrected demand timing.\n\n## League / negative knowledge\nChampion unchanged; working lineage MIX11; boundary Broker Bea; stress Closer Cleo; regression Rancher Rita. No negative knowledge reopened; no WOOL mutation authorized.\n\n## Sidecar\nMain remains 241 DONE / 5 FAILED / 0 READY/RUNNING. No blind refill; teacher work is branch-local until its measurement layer is complete.\n\n## First incomplete step\nImplement V3 conservation instrumentation: derive opponent/our public SHEEP yield-state transitions as production-flow proxies, our known WOOL sale flow, corrected YARN/town-center deterministic consumption timing, and shared `Δinventory/Δprice`; then test incremental explanatory value by world/seat. Only after this measurement passes controls consider a fixed-prestate economic treatment.",[],[175,176,2570,2596],"3b1e3ab9-57ad-43d3-beb4-1ff4f6be89d0","KAG_HOURLY_CHECKPOINT_20260908_1520.md","7e228efc-ab99-11f1-bd04-1607ee4eb77e",790,["D",1788881115957],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 15:20 UTC\n\nStatus: PASS / MATERIAL_MEDIATION_NARROWING / NOT_CHAMPION_PROOF\n\n## KAG_CONTEXT_PREFLIGHT\nREAD_FIRST=READ; STRATEGIC_INTELLIGENCE=READ; CAPABILITY_TAXONOMY_190=READ; CURRICULUM/NEURON/JOINED_PLAN=READ; OFFICIAL_RULES/RT=READ; CURRENT_STATE/NEGATIVE_KNOWLEDGE=READ; 15MIN_REANALYSIS/SCENARIO_COVERAGE/RECOVERY/LATEST_AUDIT=READ; OPPONENT_INDEX + Broker/Rita/Cleo relevant dossiers=READ; EXPERIMENT_MATRIX/PROFILE_CAL=READ; HOURLY_INDEX + checkpoint/manifest 13:15=READ; SIDECAR queue/runs/states/artifacts/evidence epochs=RECONCILED (241 DONE / 5 FAILED / 0 READY/RUNNING; no post-13:15 main jobs). Final **PASS**.\n\n## Material evidence\nFresh teacher-only world-conditioned census `wool_mediation_world_v1.ts`, `[110501,110513,110527,110539]`, mirrored seats, unchanged MIX11, legal public + own-private telemetry, opponent raw actions NONE.\n\nUpper family remained tightly clustered. Broker example: 0-8/-92497.625; mean D18-D27 WOOL inventory 10012.04, price 81.66, opponent sheep 6. Rita control: opponent sheep materially lower but world-conditioned inventory direction was not monotone.\n\nPaired upper-family vs Rita: opponent-sheep delta was positive throughout, but `sign(dOppSheep)==sign(dWoolInventory)` held only 4/8 pairs per upper opponent. Conversely `sign(dInventory)==-sign(dPrice)` held 8/8 pairs per upper opponent. Our sheep/output changed strongly by world/matchup and remains an endogenous mediator/confounder.\n\n## Traits / falsification\n`market inventory -> price` WOOL relation: `FRESH_WORLD_CONDITIONED_STRUCTURAL_REPLICATION`.\n`opponent sheep/output -> shared WOOL inventory as sole/simple mediator`: `NOT_SUPPORTED_IN_THIS_FORMULATION`; more opponent sheep is insufficient to determine inventory sign once our output and world demand differ.\nNo promotion of direct reaction, resource denial, substitution, bait/bluff, or reaction magnitude. Unmeasured taxonomy traits remain UNKNOWN.\n\n## League / negative knowledge\nChampion unchanged; working lineage MIX11; boundary Broker Bea; stress Closer Cleo; regression Rancher Rita. No negative knowledge reopened and no WOOL policy mutation authorized.\n\n## Sidecar\nMain queue unchanged: 241 DONE / 5 FAILED / 0 READY/RUNNING. No blind refill. Teacher script persisted on `reaction-attribution-v1` as `wool_mediation_world_v1.ts`.\n\n## First incomplete step\nBuild a finer temporal conservation/mediation decomposition over WOOL that separately accounts for our visible sheep/output, opponent visible sheep/output, our known requested WOOL sales, shared inventory delta, price delta, and deterministic town/shop consumption timing. The discriminant must test whether opponent-output adds explanatory power after controlling our contribution + exogenous demand. Only after that consider any economic treatment.",[],[175,176,2570,2604],"2ae9a76f-c648-4400-8097-7e65c7103862","KAG_HOURLY_CHECKPOINT_20260908_1315.md","1032f6b8-ab88-11f1-b0ad-1607ee4eb77e",785,["D",1788873630049],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 13:15 UTC\n\nStatus: PASS / CONTINUITY_REPAIR + MATERIAL_CAUSAL_NARROWING / NOT_CHAMPION_PROOF\n\n## KAG_CONTEXT_PREFLIGHT\n`00_READ_FIRST.md`=READ; Strategic Intelligence=READ; Capability Taxonomy=READ (190 traits); Curriculum/Neuron/Joined Plan=READ; Official Rules/RT=READ; Current State/Negative Knowledge=READ; 15-min reanalysis/scenario coverage/recovery/latest audit=READ; Opponent Index + Broker/Rita/Lena/Silas/Cleo target/control dossiers=READ; Experiment Matrix/Profile Calibration=READ; HOURLY_INDEX + checkpoint/manifest 12:45=READ; sidecar main queue/runs/states/artifacts/evidence epochs=RECONCILED (241 DONE / 5 FAILED / 0 READY/RUNNING). Contradiction: Current State contained material post-12:45 conversion/WOOL evidence not yet closed by hourly checkpoint. Reconciled by re-running fresh family WOOL census and the two fixed-prestate discriminants before strategic update. Final **PASS**.\n\n## Material evidence\nFresh upper-family WOOL census `[110101,110113,110127,110139]`: Broker/Lena/Silas/Cleo remain a tight WOOL footprint cluster versus Rita, but the direction of price/inventory difference versus Rita flips across world seeds. `STABLE_TOP_FAMILY_WOOL_PRICE_SUPPRESSION = FALSIFIED_AS_UNIVERSAL`; classification is `WORLD_REGIME_INTERACTION / INVENTORY-CONSISTENT_PRICE_DIRECTION / NOT_STABLE_EXPLOIT`.\n\n`wool_snapshot_fixedprestate_v1.ts`, fresh `[110401,110413,110427]`: current WOOL price response 0/12 and inventory response 0/12 for Broker/Lena/Silas/Cleo/Rita/Rosa at D18/D21/D24/D27. Broker/Lena/Silas cash 199→200 positive control fired 3/6; Rita/Rosa 0. Apparatus therefore did not simply fail to detect all reactions.\n\n`public_mix_fixedprestate_v1.ts`, fresh `[110201,110213,110227]`: isolated public crop-mix and animal-mix perception interventions produced 0/15 action responses for Broker/Lena/Silas/Cleo/Rita/Rosa across D3/D6/D12/D18/D24. Broker/Lena/Silas cash positive controls fired 6/6.\n\n## Trait/hypothesis update\n`our_crop_mix_sensitivity` and `our_animal_mix_sensitivity`: `NO_ACTION_RESPONSE_IN_TESTED_LANDMARKS / NOT_UNIVERSAL_ABSENCE` for tested targets. Current WOOL snapshot price/inventory reaction: tested-context null. Reaction magnitude/latency causality, hysteresis, broader history dependence, substitution, risk, bait/bluff and unmeasured taxonomy traits remain UNKNOWN. The WOOL footprint is better supported as endogenous shared-market supply/world-regime mediation than direct snapshot reaction.\n\n## League / negative knowledge\nChampion unchanged; working lineage MIX11; boundary Broker Bea; stress Closer Cleo; regression Rancher Rita. No negative-knowledge family reopened. No WOOL-price, crop-mix or animal-mix bait/mutation authorized.\n\n## Sidecar\nMain remains 241 DONE / 5 FAILED / 0 READY/RUNNING. No blind queue refill; these teacher-only fixed-prestate tests run on the research branch and are not champion evidence.\n\n## First incomplete step\nBuild a world-conditioned shared-market mediation discriminant for late WOOL: within matched fresh worlds, quantify whether top-family sheep/output → shared WOOL inventory → price → our WOOL sale quantity/value forms a consistent causal/structural chain, versus the alternative that our own sheep/output divergence drives most of the effect. Require mediation accounting and fresh world strata before any policy treatment.",[],[175,176,2570,2612],"7c85d9e5-34c4-4777-a2e4-a4cbe00e0b27","KAG_HOURLY_CHECKPOINT_20260908_1245.md","46b2e1d4-ab84-11f1-b8c4-1607ee4eb77e",770,["D",1788872003522],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 12:45 UTC\n\nStatus: PASS / CONTINUITY_REPAIR + MATERIAL_CENSUS_FALSIFICATION / NOT_CHAMPION_PROOF\n\n## KAG_CONTEXT_PREFLIGHT\nFinal: **PASS**.\n- `00_READ_FIRST.md`: READ integralmente.\n- `KAG_STRATEGIC_INTELLIGENCE.md`: READ.\n- `KAG_OPPONENT_CAPABILITY_TAXONOMY.md`: READ — exatamente 190 traits; UNKNOWN preservados.\n- Curriculum / Neuron Protocol / Joined Research Plan: READ.\n- Official Rules / RT Architecture: READ.\n- Current State / Negative Knowledge: READ.\n- 15-min reanalysis / scenario coverage / recovery / latest day audit: READ.\n- Opponent index + Rosa/Hana/Mateo/Rita/Broker/Lena/Silas/Cleo dossiers: READ.\n- Experiment Matrix / Profile Calibration: READ.\n- `HOURLY_INDEX.md` + checkpoint/manifest 12:36: READ.\n- main sidecar queue/runs/states/artifact rows: RECONCILED — 241 DONE / 5 FAILED / 0 READY/RUNNING.\n- latest main-sidecar physical B58 blobs Lena/Silas: RECONCILED / readable.\n- evidence epochs: RECONCILED — main V25 evidence remains `AUTOLAB_COMPUTE_RESULT_NOT_CHAMPION_PROOF`; teacher census uses `reaction-attribution-v1` V27.\n- contradiction repaired: `KAG_CURRENT_STATE.md` and `KAG_NEGATIVE_KNOWLEDGE.md` had material 12:45 task-service results not represented in `HOURLY_INDEX` or matching checkpoint/manifest. `task_service_census_v2.ts` was READ and reexecuted; output reproduced the canonical values exactly. Continuity chain repaired here.\n\n## Reconciled task-service census\nFresh natural MIX11 `[109701,109713,109727,109739]`, mirrored, same telemetry against Broker and Rita, no raw opponent action online.\n- Broker: 0-8 / avg margin -92701.375.\n- Rita: 7-1 / +5053.125.\n- D9-D15 own requested sell qty: Broker-context 958 vs Rita-context 970; harvest-yield backlog area 23443 vs 22950; water stress 481 vs 546; care backlog 13056 vs 16992; feed stress 384 vs 288.\n- D18-D27 sell qty: 1347 vs 1512; harvest backlog 41682 vs 42912; water stress 665 vs 734; feed stress 0 vs 144; care backlog 34080 vs 31368.\n- shed-high rows >=80%: 0 in both contexts/windows.\n\nClassification: `TASK_SERVICE_BACKLOG_AS_BOUNDARY_SPECIFIC_CAUSE = NOT_SUPPORTED / MATCHUP_NONSPECIFIC_IN_FRESH_CONTROL`. General service backlog exists, but it is not a discriminant that explains Broker losses versus Rita wins in this formulation. No blind service/routing/labor mutation is authorized.\n\n## Traits / opponent model\nNo opponent reaction trait promoted by this census. It narrows `bottleneck_sensitivity / route_efficiency_proxy / service_capacity` as a boundary explanation. Direct-vs-market-mediated opponent effects, product substitution, output-value conversion, price/supply sensitivity and interaction with our policy remain unresolved/UNKNOWN where not already measured.\n\n## Negative knowledge / league\nNo closed family reopened. PASTURE_RAMP, CAP2_D6, staged portfolio, bounded labor, persistent routing and existing negative knowledge remain closed.\nChampion unchanged; working lineage MIX11; boundary Broker Bea; stress Closer Cleo; regression Rancher Rita.\n\n## First incomplete step\nPerform a fresh, no-policy-mutation **differential economic conversion census** under identical MIX11 policy and fresh mirrored worlds: decompose our requested/realized sales, product mix, realized price/revenue, private inventory/shed and shared-market movement versus Broker and Rita. Goal: discriminate `OUR_OUTPUT_COMPOSITION/CAPACITY` from `OPPONENT-MEDIATED_MARKET_CONVERSION` before any treatment.",[],[175,176,2570,2620],"81377de9-dd90-41bf-9116-806ed84d14db","KAG_HOURLY_CHECKPOINT_20260908_1236.md","d042d0ba-ab81-11f1-a816-1607ee4eb77e",763,["D",1788870945811],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 12:36 UTC\n\nStatus: PASS / MATERIAL_MECHANISM_FALSIFICATION / NOT_CHAMPION_PROOF\n\n## KAG_CONTEXT_PREFLIGHT\nFinal: **PASS**.\n- `00_READ_FIRST.md`: READ integralmente.\n- Strategic Intelligence: READ.\n- Capability Taxonomy: READ — exatamente 190 traits; UNKNOWN preservados.\n- Curriculum / Neuron Protocol / Joined Research Plan: READ.\n- Official Rules / RT Architecture: READ.\n- Current State / Negative Knowledge: READ.\n- 15-min reanalysis / scenario coverage / recovery / latest day audit: READ.\n- Opponent index + Broker/Rita/Lena/Silas/Cleo/Rosa/Hana/Mateo dossiers: READ.\n- Experiment Matrix / Profile Calibration: READ.\n- HOURLY_INDEX + latest checkpoint/manifest 11:22: READ.\n- sidecar queue/runs/states/artifact rows: RECONCILED.\n- latest physical B58 blobs Lena/Silas: RECONCILED / readable.\n- evidence state: RECONCILED; main sidecar remains compute evidence `NOT_CHAMPION_PROOF`; teacher work used experimental V27 branch only.\n- contradiction: sidecar `context:last_preflight`, `hourly:last_analysis`, `hourly:current_hypothesis` were stale versus canonical 11:22. Canonical precedence applied; no blocker.\n\n## AVIÃO / capability census\nFresh natural MIX11 `[109401,109413,109427,109439]`, mirrored vs Broker: D9-D15 minimum own WHEAT=5; D18-D27 minimum=12; WHEAT<=2 rows=0 in both windows. MIX11 already requested substantial BUY_PRODUCT WHEAT. Thus `simple missing WHEAT procurement` is not supported in the tested windows.\n\nOfficial engine verification: FEED consumes one WHEAT from the acting unit's carried inventory; `consecutive_unfed` increments after an unfed day and at >=2 the animal is removed. Therefore the observed stress is real service failure evidence rather than an arbitrary proxy.\n\n## Preregistered causal discriminant\nPreregistration: `artifacts/kaggriculture/experiments/KAG_EXP_PERSISTENT_FEED_ROUTING_V1.md`.\nFresh `[109501,109513,109527,109539]`, mirrored Broker, only treatment `persistent_routing=True`; macro policy otherwise preserved.\n- detector: MIX11 false 8/8; treatment true 8/8; errors 0.\n- MIX11: 0-8 / avg margin -102722.125.\n- MIX11_ROUTE_PERSIST: 0-8 / -99815.625; margin-only delta +2906.5.\n- D9-D15 H0 unfed-stress sum: 22 -> 28.\n- D18-D27 H0 unfed-stress sum: 0 -> 2.\n- no H0 animal-count drops in audited windows.\n- WHEAT remained non-scarce in both arms.\n\nClassification: `PERSISTENT_ROUTING_AS_FEED_CONTINUITY_FIX = FRESH_PAIRED_MECHANISM_FALSIFIED / WINLESS / DO_NOT_PROMOTE`. No adjacent routing sweep. Margin improvement does not override mechanism failure.\n\n## Traits / opponent model\nNo opponent reaction trait promoted: this is an our-capability intervention. Existing 190-trait scope remains. Evidence narrows operational bottleneck hypotheses toward task-service/spatial conversion throughput, not raw WHEAT quantity. Direct-vs-market-mediated reaction, hysteresis, risk, bait/bluff and other unmeasured dimensions remain UNKNOWN.\n\n## Negative knowledge / league\nNegative knowledge updated; no prior rejected family reopened. PASTURE_RAMP, CAP2_D6, staged portfolio, bounded labor and previous negatives remain closed.\nChampion unchanged; working lineage MIX11; boundary Broker Bea; stress Closer Cleo; regression Rancher Rita.\n\n## Sidecar\nReconciled main queue before teacher experiment: 241 DONE / 5 FAILED / 0 READY/RUNNING. No sidecar compute jobs consumed/enqueued in this epoch; causal experiment executed teacher-only on `reaction-attribution-v1` after PASS.\n\n## Updated artifacts\n- `KAG_CURRENT_STATE.md`\n- `KAG_NEGATIVE_KNOWLEDGE.md`\n- `KAG_STRATEGIC_INTELLIGENCE.md`\n- `KAG_EXP_PERSISTENT_FEED_ROUTING_V1.md`\n- experimental tools: `mix11_wheat_feed_capability_census_v1.ts`, `official_feed_semantics_probe_v1.ts`, routing/source slices, `mix11_route_persist_feed_causal_v1.ts`, V27 treatment detector.\n\n## First incomplete step\nAfter a fresh canonical preflight, census **task-service throughput** without policy mutation: quantify urgent FEED/PICKUP/WATER/HARVEST demand versus actual completion, travel/zone burden and product-to-cash conversion around D9-D15 and D18-D27. Only a demonstrated service-capacity gap may authorize the next isolated treatment.",[],[175,176,2570,2628],"ae9d946d-44df-42f0-a134-d49a6be54376","KAG_HOURLY_CHECKPOINT_20260908_1122.md","722ad7fc-ab77-11f1-9d93-1607ee4eb77e",756,["D",1788866492991],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 11:22 UTC\n\nStatus: PASS / MATERIAL_MECHANISM_FALSIFICATION / NOT_CHAMPION_PROOF\n\n## KAG_CONTEXT_PREFLIGHT\nREAD_FIRST=READ; STRATEGIC_INTELLIGENCE=READ; CAPABILITY_TAXONOMY_190=READ; CURRICULUM=READ; NEURON_PROTOCOL=READ; JOINED_PLAN=READ; OFFICIAL_RULES=READ; RT_ARCHITECTURE=READ; CURRENT_STATE=READ; NEGATIVE_KNOWLEDGE=READ; FULL_15MIN_REANALYSIS=READ; REAL_SCENARIO_COVERAGE=READ; RECOVERY=READ; latest DAY_AUDIT=READ; OPPONENT_INDEX=READ; dossiers Broker/Lena/Silas/Cleo/Rita + weak controls=READ; EXPERIMENT_MATRIX=READ; PROFILE_CAL=READ; HOURLY_INDEX=READ; latest hourly checkpoint/manifest 10:43=READ; continuation preflight 10:47=READ; clone-history result + natural-support result=READ; SIDECAR queue/runs/states/artifacts=RECONCILED (241 DONE / 5 FAILED / 0 READY/RUNNING). Canonical state is ahead of sidecar strategy states; canonical precedence applied. Final **PASS**.\n\n## Material change\nThe controlled clone/history mechanism is already canonically CONFIRMED for Broker/Lena/Silas but natural support is absent under MIX11. The next candidate continuously-active public-state reaction path was audited: `_opponent_scale(obs,item)` feeds `_reserve_price`, but runtime source constants for Broker/Lena/Silas/Cleo have `_RESERVE={}`. Therefore `_plan_sells` has no reserve-controlled items and the public supply-scale reaction path is inactive in the current configuration.\n\nClassification: `PUBLIC_SUPPLY_RESERVE_RESPONSE_PATH = CODE_PATH_PRESENT / CURRENT_CONFIG_INACTIVE / NOT_CURRENT_LOSS_EXPLANATION`.\n\nThis is a falsification, not a universal claim that top agents never use our public state: sparse cash-threshold reactions and controlled clone/history front-run remain real in their validated contexts. It specifically rejects this apparent continuous supply-elasticity channel.\n\n## Traits/hypotheses\nNo new trait promoted. `our_crop_mix_sensitivity`, `our_animal_mix_sensitivity`, `direct_awareness_probability`, `reaction_magnitude`, history/path and models-us retain only their already-established scoped evidence; all unmeasured taxonomy fields remain UNKNOWN. Clone/history capability remains causal but natural activation absent in tested MIX11 worlds.\n\n## League / negative knowledge\nChampion unchanged; working lineage MIX11; boundary Broker Bea; stress Closer Cleo; regression Rancher Rita. No negative-knowledge family reopened. CAP2_D6, staged portfolio, bounded labor and PASTURE_RAMP remain closed.\n\n## Sidecar / tools\nMain queue remains 241 DONE / 5 FAILED / 0 READY/RUNNING. No blind refill. Experimental teacher tools added/extended: `top_stateful_logic_slice_v1.ts`, `top_supply_driver_constants_v1.ts`, `top_runtime_market_constants_v1.ts`. They are offline source/root-cause tools, nondeployable.\n\n## First incomplete step\nPivot current-loss branch to fixed-plan conversion/throughput: isolate a genuinely active mechanism before mutation, prioritizing WHEAT procurement/feed continuity and conversion/sale throughput around D9→D15 and D18→D27. First action is source+own-capability census and treatment-opportunity detector; only if a missing capability is demonstrated may a preregistered single-factor causal treatment be created. No adjacent land/labor/crop-share sweep.\n",[],[175,176,2570,2636],"e390eae5-d108-4141-abd5-e9f8ea496ff3","KAG_HOURLY_CHECKPOINT_20260908_1043.md","e3f6fe70-ab71-11f1-89d1-1607ee4eb77e",742,["D",1788864106925],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 10:43 UTC\n\nStatus: PASS / MATERIAL_OBSERVATIONAL_REPLICATION / NOT_CHAMPION_PROOF\n\n## KAG_CONTEXT_PREFLIGHT\nREAD_FIRST=READ; STRATEGIC_INTELLIGENCE=READ; CAPABILITY_TAXONOMY_190=READ; CURRICULUM/NEURON/JOINED_PLAN=READ; OFFICIAL_RULES/RT=READ; CURRENT_STATE/NEGATIVE_KNOWLEDGE=READ; STRATEGIC_REANALYSIS/SCENARIO_COVERAGE/RECOVERY/LATEST_AUDIT=READ; OPPONENT_INDEX=READ; TARGET_DOSSIERS Lena/Silas + controls/regression structure=READ; EXPERIMENT_MATRIX/PROFILE_CAL=READ; HOURLY_INDEX + checkpoint/manifest 10:05=READ; SIDECAR queue/runs/states/artifacts=RECONCILED; B58 blobs 2/2=RECONCILED. CONTRADICTIONS: sidecar hourly:last_analysis/current_hypothesis/context:last_preflight are stale versus canonical 10:05 checkpoint; canonical precedence applied. Final: **PASS**.\n\n## B58 consumed\nFresh `[109101,109113,109127]`, mirrored seats, MIX11, legal telemetry FULL, raw opponent action NONE:\n- Lena: 0-6, avg margin -87274.33; artifact `artifacts/kaggriculture/autolab/20260908T104028Z_B58-V25-PATH-LENA`.\n- Silas: 0-6, avg margin -86227.67; artifact `artifacts/kaggriculture/autolab/20260908T104033Z_B58-V25-PATH-SILAS`.\n\nReproducible detector `tools/analyze_b58_cash_event_lag.ts`:\n- Lena: lag0 +0.2926; lag2 +0.1220; lag3 +0.0276; lag4 +0.0626.\n- Silas: lag0 +0.5857; lag2 +0.1421; lag3 -0.1820; lag4 -0.0818.\nThis directionally replicates prior Lena immediate+tail and Silas stronger-immediate+lag2 signatures. Classification remains `MULTI_HELDOUT_REPLICATED_LEGAL_OBSERVATIONAL_TIMING_SIGNATURE / NOT_CAUSAL / NOT_ROUTING_PROOF`. Magnitudes are not policy/world invariant.\n\n## Traits/hypotheses\nEvidence strengthened only for reaction-timing observational signatures (`reaction_latency_distribution` descriptive proxy / temporal fingerprint). `reaction_magnitude` causal elasticity, direct-vs-mediated attribution, history/path dependence, hysteresis, models-us, bait/bluff and calibrated posterior remain UNKNOWN/unpromoted. No negative knowledge reopened.\n\n## League\nChampion unchanged. Working lineage MIX11. Boundary Broker Bea. Stress Closer Cleo. Regression Rancher Rita.\n\n## Queue\nAfter B58: 241 DONE / 5 FAILED / 0 READY/RUNNING. No blind refill. The next executable scientific step requires implementing/preregistering the fixed-history/fixed-prestate path × order-slot/funding causal discriminant; it is not representable as an existing safe queue job without new teacher tooling, so queue emptiness is intentional rather than inertial.\n\n## First incomplete step\nImplement and preregister teacher-only causal discriminant: two legal public history paths converging to the same terminal observable prestate, isolate order-slot/funding mediator, require treatment-fired detector, positive/negative controls, immediate/downstream economic effect, and direct-vs-market-mediated attribution. Only then enqueue/run fresh causal forks. No macro policy mutation before this gate.\n",[],[175,176,2570,2644],"90d45edb-adf3-4c61-9645-ede05849d708","KAG_HOURLY_CHECKPOINT_20260908_1005.md","618b7dc8-ab6f-11f1-8172-1607ee4eb77e",734,["D",1788863029131],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 10:05 UTC\n\nStatus: PASS / MATERIAL_OBSERVATIONAL_REPLICATION / NOT_CHAMPION_PROOF\n\n## KAG_CONTEXT_PREFLIGHT\nFinal state: **PASS**.\n- `00_READ_FIRST.md`: READ integralmente.\n- `KAG_STRATEGIC_INTELLIGENCE.md`: READ.\n- `KAG_OPPONENT_CAPABILITY_TAXONOMY.md`: READ — 190 traits preservados; UNKNOWN não removidos.\n- curriculum / neuron protocol / joined research plan: READ.\n- `KAG_OFFICIAL_RULES.md` / `KAG_RT_ARCHITECTURE.md`: READ.\n- `KAG_CURRENT_STATE.md` / `KAG_NEGATIVE_KNOWLEDGE.md`: READ.\n- strategic reanalysis / real-scenario coverage / recovery / latest day audit: READ.\n- opponent index + dossiers relevantes Broker/Cleo/Rita e canonical profile structure: READ; demais dossiers permanecem canônicos do preflight anterior sem nova claim nesta rodada.\n- experiment matrix + profile calibration: READ.\n- `HOURLY_INDEX.md` + latest checkpoint/manifest 09:15: READ.\n- sidecar queue/runs/states/artifact rows: RECONCILED.\n- B57 physical JSON blobs Broker/Cleo/Rita: RECONCILED — 3/3 present and readable.\n- evidence/core: RECONCILED — V25; B57 evidence `AUTOLAB_COMPUTE_RESULT_NOT_CHAMPION_PROOF`.\n\n## Consumed B57\nFresh `[109001,109013,109027]`, mirrored seats, MIX11, legal telemetry FULL, opponent raw action NONE:\n- Broker: 0-6, avg margin -105993.33.\n- Cleo: 0-6, avg margin -107848.33.\n- Rita: 4-2, avg margin +7724.33.\n\nArtifact prefixes:\n- `artifacts/kaggriculture/autolab/20260908T093122Z_B57-V25-PATH-BROKER`\n- `artifacts/kaggriculture/autolab/20260908T093127Z_B57-V25-PATH-CLEO`\n- `artifacts/kaggriculture/autolab/20260908T095215Z_B57-V25-PATH-RITA`\n\n## B57 legal self-action-filtered lag falsifier\nA fresh analyzer reused the preregistered V23 detector: event = |our public cash delta| >=500; opponent-response metric uses only legal public opponent cash/land/crops/animals; rows with our own market flow at measured lag are excluded.\n\nFresh lifts:\n- Broker: lag0 **+0.3461**, lag2 **+0.1719**; other nearby lags negative.\n- Cleo: lag0 **+2.4884**, then lags1-4 negative; lag5 +0.1837.\n- Rita: lag0 **+0.0520**, lags1-6 negative.\n\nThis independently preserves the preregistered qualitative ordering `Cleo immediate >> Broker immediate`, Broker's lag2 echo, and Rita near-null. Classification advances only to `LEGAL_OBSERVATIONAL_TIMING_SIGNATURE = MULTI_HELDOUT_REPLICATED / NOT_CAUSAL / NOT_ROUTING_PROOF`.\n\nHard guard: this does not prove direct reaction to us. Shared-market mediation, endogenous trajectory and common timing remain live alternatives. Cleo's fixed-history snapshot interventions already falsified simple instantaneous price/inventory/shop/broad-cash explanations in tested contexts, so the highest-EVSI causal branch remains path/history × order-slot/funding.\n\n## Negative knowledge / champion\nNo rejected policy reopened. PASTURE_RAMP and CAP2_D6 remain closed. Champion unchanged; working lineage MIX11; boundary Broker; stress Cleo; regression Rita.\n\n## Queue\nPre-enqueue: 239 DONE / 5 FAILED / 0 READY/RUNNING.\nB58 queued with fresh `[109101,109113,109127]` to complete upper-family observational map before the more expensive causal path experiment:\n- `B58-V25-PATH-LENA`\n- `B58-V25-PATH-SILAS`\nBoth legal telemetry FULL, opponent raw action NONE, unique dedupe keys.\n\n## Failure learned this round\nAn initial SQLite batch enqueue attempt used unsupported parameter binding semantics for this connector path and was atomically rejected; no partial jobs were created. Control: verify rows after write and use explicit SQL only for fixed internal values. This was infrastructure/tooling, not scientific evidence.\n\n## First incomplete step\nAfter the next canonical preflight, consume B58. Then preregister a fixed-history/fixed-prestate causal discriminant that separates direct/path-dependent response from shared-market mediation and tests order-slot/funding as mediator, with treatment-fired detector and positive/negative controls. Do not create another macro policy mutation before that attribution test.\n",[],[175,176,2570,2652],"a8fa4a75-b02d-4901-b359-cc57ecfef553","KAG_HOURLY_CHECKPOINT_20260908_0915.md","8c80df54-ab66-11f1-b00c-1607ee4eb77e",730,["D",1788859235733],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 09:15 UTC\n\nStatus: PASS / MATERIAL_REGRESSION_FALSIFICATION / NOT_CHAMPION_PROOF\n\n## KAG_CONTEXT_PREFLIGHT\n\nFinal state: **PASS**.\n\n- `00_READ_FIRST.md`: READ\n- `KAG_STRATEGIC_INTELLIGENCE.md`: READ\n- `KAG_OPPONENT_CAPABILITY_TAXONOMY.md`: READ — 190 traits preserved\n- curriculum / neuron protocol / joined research plan: READ\n- `KAG_OFFICIAL_RULES.md` / `KAG_RT_ARCHITECTURE.md`: READ\n- `KAG_CURRENT_STATE.md` / `KAG_NEGATIVE_KNOWLEDGE.md`: READ\n- strategic reanalysis / real-scenario coverage / recovery / latest day audit: READ\n- opponent index + Broker/Rita/Cleo/Lena/Silas/Rosa/Hana/Mateo dossiers: READ\n- experiment matrix + profile calibration: READ\n- `HOURLY_INDEX.md` + latest checkpoint/manifest 0820: READ\n- sidecar queue/runs/states/artifact rows: RECONCILED\n- B56 Rita physical JSON blobs: RECONCILED — 2/2 present\n- evidence/core: RECONCILED — V25\n\nContradiction reconciled: `HOURLY_INDEX.md` was stale and omitted the persisted 08:20 checkpoint; this round repairs the index. Sidecar `hourly:last_analysis/current_hypothesis` was also stale relative to canonical 08:20 state and is not treated as strategy authority.\n\n## Consumed sidecar evidence\n\n- `artifacts/kaggriculture/autolab/20260908T082358Z_B56-V25-R0-MIX11-RITA`\n- `artifacts/kaggriculture/autolab/20260908T082404Z_B56-V25-R1-CAP2-RITA`\n\nFresh `[108901,108913,108927]`, mirrored seats:\n- MIX11 vs Rita: **6-0 / +9630.50**, avg our bank 32326.17, errors 0.\n- CAP2_D6 vs Rita: **0-6 / -9005.00**, avg our bank 16960, errors 0.\n- treatment-control margin delta: **-18635.50**.\n\n## Material falsification\n\nCAP2_D6 now fails both required dimensions:\n1. fresh Broker gate: 0-6 and -11844.33 directional delta versus paired MIX11;\n2. fresh Rita regression: W/L flips 6-0 → 0-6 and margin delta -18635.50.\n\nClassification hardened to `FRESH_HELDOUT_FALSIFIED / SEVERE_REGRESSION_HARM / DO_NOT_REOPEN / DO_NOT_PROMOTE`. No adjacent land/capacity sweep is authorized.\n\n## Opponent model / traits\n\nNo new opponent trait is promoted by B56 because CAP2_D6 is our-policy intervention. All unmeasured taxonomy fields remain UNKNOWN. Priority returns to reaction magnitude/latency, history/path dependence, direct-vs-market mediation, order-slot/funding context and hysteresis.\n\nB57 is preregistered as a cheap fresh-world falsifier of the existing legal observational timing signature, not as causal proof: Broker/Cleo/Rita, MIX11, fresh seeds `[109001,109013,109027]`, legal telemetry FULL, opponent raw action NONE. Prediction: Cleo immediate signature > Broker, Broker lag2 echo, Rita near-zero. Collapse of this ordering downgrades transfer scope and blocks routing.\n\n## Queue after action\n\nAt pre-enqueue cut: 236 DONE / 5 FAILED / 0 READY/RUNNING. Enqueued 3 deduplicated B57 jobs:\n- `B57-V25-PATH-BROKER`\n- `B57-V25-PATH-CLEO`\n- `B57-V25-PATH-RITA`\n\n## Champion / league\n\nChampion unchanged. Working lineage MIX11. Boundary Broker Bea. Stress Closer Cleo. Regression Rancher Rita. Cheap controls Rosa/Hana/Mateo.\n\n## First incomplete step\n\nAfter the next canonical preflight, consume B57 if completed and recompute the self-action-filtered legal lag/path signature. If it survives fresh worlds, proceed to a fixed-prestate/fixed-terminal multi-step path or order-slot/funding causal discriminant with positive/negative controls; if it collapses, downgrade the observational reaction-timing model rather than tuning it to fit.\n",[],[175,176,2570,2660],"abb18346-d3bd-4f43-aa0f-033c76938fe7","KAG_HOURLY_CHECKPOINT_20260908_0820.md","1f4da1d6-ab5e-11f1-a480-1607ee4eb77e",725,["D",1788855641102],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 08:20 UTC\n\nStatus: PASS / MATERIAL_FALSIFICATION / NOT_CHAMPION_PROOF / ARTIFACT_PARTIAL\n\nArtifact note: mandatory `HOURLY_INDEX.md` append was attempted after checkpoint/manifest/diff persistence but was blocked by the external tool safety layer. Scientific artifacts are persisted; index update remains incomplete and must be reconciled next round.\n\n## KAG_CONTEXT_PREFLIGHT\n\nFinal state: **PASS**.\n\n| Item | Receipt |\n|---|---|\n| `00_READ_FIRST.md` | READ |\n| `KAG_STRATEGIC_INTELLIGENCE.md` | READ |\n| `KAG_OPPONENT_CAPABILITY_TAXONOMY.md` — 190 traits preserved | READ |\n| curriculum / neuron protocol / joined research plan | READ |\n| `KAG_OFFICIAL_RULES.md` / `KAG_RT_ARCHITECTURE.md` | READ |\n| `KAG_CURRENT_STATE.md` / `KAG_NEGATIVE_KNOWLEDGE.md` | READ |\n| strategic reanalysis / real-scenario coverage / recovery / latest day audit | READ |\n| opponent index + Rosa/Hana/Mateo/Rita/Broker/Lena/Silas/Cleo dossiers | READ |\n| experiment matrix + profile calibration | READ |\n| `HOURLY_INDEX.md` + checkpoint/manifest 0752 | READ |\n| sidecar queue / runs / states | RECONCILED |\n| sidecar artifact rows | RECONCILED |\n| B56 Broker physical JSON Blobs | RECONCILED — 2/2 present |\n| evidence epoch/core | RECONCILED — V25 capacity/portfolio factorial |\n\nContradictions: none unresolved. Canonical project state takes precedence over chat memory and sidecar strategy state.\n\n## Sidecar consumed since 0752\n\nB56 Broker control and treatment completed:\n- `artifacts/kaggriculture/autolab/20260908T080824Z_B56-V25-A-MIX11-BROKER`\n- `artifacts/kaggriculture/autolab/20260908T080831Z_B56-V25-C-CAP2-BROKER`\n\nLive queue at reconciliation: 234 DONE / 5 FAILED terminal / 2 READY / 0 RUNNING. The two READY jobs are the preregistered same-epoch Rita regression arms; they remain useful and are not cancelled.\n\n## Material falsification — CAP2_D6 reopening gate fails\n\nFresh seeds `[108901,108913,108927]`, mirrored seats:\n- MIX11 vs Broker: 0-6, avg margin -116285.00, avg our bank 27041.50, errors 0.\n- CAP2_D6 vs Broker: 0-6, avg margin -128129.33, avg our bank 16825.50, errors 0.\n- aggregate treatment-control directional delta = **-11844.33**.\n\nB55 had shown a one-batch +17714.17 diagnostic margin direction for CAP2_D6, still 0-6. B56 reverses that direction on fresh worlds and therefore fails the preregistered Broker replication requirement.\n\nClassification: `CAP2_D6 = FRESH_HELDOUT_FALSIFIED / DO_NOT_REOPEN / DO_NOT_PROMOTE`.\n\nNo adjacent land/capacity target sweep is authorized. Broader capacity timing remains UNKNOWN and can only return under the canonical hard reopening gate with a new mechanism and context.\n\n## Opponent-model / taxonomy\n\nNo opponent reaction trait changes from B56 because this is an our-policy intervention, not a reaction-function experiment. The 190-trait universe remains intact; unmeasured traits remain UNKNOWN. Highest-EVSI strategic lane returns to legal opponent reaction/path dependence/order-slot/funding attribution, with direct-vs-market-mediated controls and causal treatment detectors.\n\n## Champion / league\n\nChampion unchanged. Working lineage MIX11. Boundary Broker Bea. Stress Closer Cleo. Regression/intermediate Rancher Rita. Cheap controls Rosa/Hana/Mateo.\n\n## First incomplete step\n\nConsume the already-READY B56 Rita control/treatment after the next canonical preflight. They can strengthen regression negative knowledge but cannot rescue CAP2_D6 reopening after the fresh Broker directional gate failed. After Rita closure, use the next compute budget on a discriminating opponent-reaction/path/slot/funding experiment rather than adjacent capacity mutation.\n",[],[175,176,2570,2668],"53e13bfb-03ad-49cf-b3a8-f73f66e26d6a","KAG_HOURLY_CHECKPOINT_20260908_0752.md","717dbdaa-ab5a-11f1-90fd-1607ee4eb77e",714,["D",1788854036450],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 07:52 UTC\n\nStatus: PASS / MATERIAL_UPDATE / NOT_CHAMPION_PROOF\n\n## KAG_CONTEXT_PREFLIGHT\n\nFinal state: **PASS**.\n\n| Item | Receipt |\n|---|---|\n| `00_READ_FIRST.md` | READ |\n| `KAG_CONTEXT_MANIFEST.json` | READ |\n| `KAG_STRATEGIC_INTELLIGENCE.md` | READ |\n| `KAG_OPPONENT_CAPABILITY_TAXONOMY.md` — 190 traits preserved | READ |\n| curriculum / neuron protocol / joined research plan | READ |\n| `KAG_OFFICIAL_RULES.md` / `KAG_RT_ARCHITECTURE.md` | READ |\n| `KAG_CURRENT_STATE.md` / `KAG_NEGATIVE_KNOWLEDGE.md` | READ |\n| strategic reanalysis / real-scenario coverage / recovery / day audit | READ |\n| opponent dossier index + Rosa/Hana/Mateo/Rita/Broker/Lena/Silas/Cleo | READ |\n| experiment matrix + profile calibration | READ |\n| `HOURLY_INDEX.md` + latest checkpoint 0628 + manifest | READ |\n| sidecar queue / runs / states | RECONCILED |\n| sidecar artifact rows | RECONCILED |\n| B55 physical Blob JSON+MD set | RECONCILED — 8/8 present |\n| evidence epoch/core | RECONCILED — main V25 capacity/portfolio factorial |\n\nProcess defect recorded: a B56 planner draft was created before the final five dossier reads in this tool turn. It was not executed. After detecting the ordering breach, the remaining mandatory dossiers + manifest + B55 physical Blobs were read/reconciled; B56 was then explicitly revalidated and only then ratified as READY. Failure class: `PREFLIGHT_ORDERING_BREACH_WITHOUT_COMPUTE_EXECUTION`; control: no future enqueue evidence state becomes executable/ratified until the receipt is complete PASS.\n\n## Sidecar consumed since 0628\n\nB53 fresh capacity family completed: Broker/Lena/Silas remained 0-6 against MIX11; Cleo 0-6; Rita 6-0 for MIX11. This replicated the descriptive top-family public trajectory but did not make it causal.\n\nB54 single-factor staged portfolio completed and is already persisted as `HELDOUT_FALSIFIED / HARMFUL`: Broker control 0-6/-94384 vs staged 0-6/-113295.17; Rita control 6-0/+8988.83 vs staged 4-2/-585.50.\n\nB55 2x2 factorial artifact prefixes consumed/reconciled:\n- `artifacts/kaggriculture/autolab/20260908T074036Z_B55-V25-A-MIX11-BROKER`\n- `artifacts/kaggriculture/autolab/20260908T074042Z_B55-V25-B-STAGED-BROKER`\n- `artifacts/kaggriculture/autolab/20260908T074058Z_B55-V25-C-CAP2-BROKER`\n- `artifacts/kaggriculture/autolab/20260908T074107Z_B55-V25-D-CAP2-STAGED-BROKER`\n\nV25 mechanical state: pretreatment hash equality + capacity treatment fired + portfolio treatment fired + actual land divergence = PASS.\n\n## B55 falsification — capacity does not rescue staged portfolio\n\nFresh Broker seeds `[108801,108813,108827]`, mirrored seats:\n- A MIX11: 0-6, margin -113730.17.\n- B STAGED: 0-6, -110910.33.\n- C CAP2_D6: 0-6, -96016.00.\n- D CAP2_D6_STAGED: 0-6, -113869.83.\n\nFactorial estimands:\n- B-A portfolio effect at baseline substrate = +2819.83.\n- C-A headroom-only = +17714.17.\n- D-C portfolio effect under headroom = -17853.83.\n- interaction `I=(D-C)-(B-A)=-20673.67`.\n\nClassification: `CAPACITY_HEADROOM_MODERATES_STAGED_PORTFOLIO = REJECTED_IN_B55_FORMULATION`.\n\nThe interaction is not merely absent; it changes the portfolio contrast strongly in the harmful direction. Therefore no adjacent crop×land factorial sweep is authorized. `STAGED_STRAW_D6_D12` remains closed.\n\n## CAP2_D6 — hard reopening gate, not promotion\n\nC-A is diagnostically favorable but all C episodes still lost to Broker. CAP2_D6 is adjacent to prior early-land negative knowledge and cannot be reopened from one batch. New explicit mechanism/context is narrower: limited D6 capacity headroom before D9 throughput inversion, with actual land divergence, rather than universal early-land expansion.\n\nState: `CAP2_D6 = ONE_BATCH_REOPENING_CANDIDATE / NOT_REOPENED`.\n\nPreregistered `KAG_EXP_CAP2_REOPENING_GATE_V1.md` requires a second fresh paired Broker heldout plus same-epoch Rita regression before any reopening claim.\n\n## Opponent-model / taxonomy\n\nNo new opponent reaction trait was promoted from B54/B55. These are our-policy causal screens, not evidence that Broker reacted to our treatment. The replicated top-family composition remains `FINGERPRINT / DESCRIPTIVE_TRAJECTORY`, not reaction model or causal mechanism. All 190 taxonomy traits remain in scope; unmeasured traits remain UNKNOWN.\n\n## Queue B56\n\nAfter full preflight PASS, four preregistered jobs are READY on fresh seeds `[108901,108913,108927]`:\n- `B56-V25-A-MIX11-BROKER`\n- `B56-V25-C-CAP2-BROKER`\n- `B56-V25-R0-MIX11-RITA`\n- `B56-V25-R1-CAP2-RITA`\n\nQueue after ratification: 232 DONE / 5 FAILED / 4 READY / 0 RUNNING.\n\n## Champion / league\n\nChampion unchanged. Working lineage MIX11. Boundary Broker Bea. Stress Closer Cleo. Regression/intermediate Rancher Rita. Cheap controls Rosa/Hana/Mateo. No negative family has been reopened.\n\n## First incomplete step\n\nConsume B56 only after the next canonical context preflight. If CAP2_D6 fails fresh Broker direction or causes material Rita regression, close the headroom lane and pivot away from adjacent land/capacity mutation. If it satisfies the reopening gate, classify it only as reopened-for-further-causal-testing, not champion evidence, and return research budget to opponent reaction/path/slot/funding traits rather than broad mutation sweeps.\n",[],[175,176,2570,2676],"f47b6305-44bd-46e4-b504-4617d59fcc10","KAG_HOURLY_CHECKPOINT_20260908_0628.md","866b42b6-ab4e-11f1-9778-1607ee4eb77e",696,["D",1788848917590],"# KAGGRICULTURE — Hourly Checkpoint 2026-09-08 06:28 UTC\n\nStatus: PASS / MATERIAL_UPDATE / NOT_CHAMPION_PROOF\n\n## KAG_CONTEXT_PREFLIGHT\n\nFinal state: **PASS**.\n\n| Item | Receipt |\n|---|---|\n| `artifacts/kaggriculture/00_READ_FIRST.md` | READ |\n| `artifacts/kaggriculture/KAG_CONTEXT_MANIFEST.json` | READ |\n| `KAG_STRATEGIC_INTELLIGENCE.md` | READ |\n| `KAG_OPPONENT_CAPABILITY_TAXONOMY.md` — 190 traits preserved | READ |\n| real-world curriculum / neuron protocol / joined research plan | READ |\n| `KAG_OFFICIAL_RULES.md` / `KAG_RT_ARCHITECTURE.md` | READ |\n| `KAG_CURRENT_STATE.md` / `KAG_NEGATIVE_KNOWLEDGE.md` | READ |\n| strategic reanalysis / real-scenario coverage / recovery reconciliation / autolab day audit | READ |\n| all 8 opponent dossiers + dossier index | READ |\n| experiment matrix + profile calibration | READ |\n| `HOURLY_INDEX.md` + latest checkpoint `0535` + manifest | READ |\n| sidecar queue + runs + states | RECONCILED |\n| `kag_autolab_artifacts` schema/content indexes | RECONCILED |\n| physical sidecar Blobs for latest B50/B51/B52 evidence epoch | RECONCILED |\n| evidence epoch/core | RECONCILED — sidecar main V23 / worker V2; experimental reaction-attribution branch independently identified |\n\nPrecedence applied: canonical primary artifacts superseded stale chat/sidecar state. `FOUND != READ` and `CHAT_MEMORY != PROJECT_STATE` preserved.\n\n## Sidecar state at preflight\n\nLatest completed evidence reconciled through B52:\n- B50 Broker: `artifacts/kaggriculture/autolab/20260908T054112Z_B50-V23-FRESH-MIX11-BROKER` — 0-6 / -108608.83.\n- B50 Rita: `artifacts/kaggriculture/autolab/20260908T054117Z_B50-V23-FRESH-MIX11-RITA` — 6-0 / +4116.\n- B50 Cleo: `artifacts/kaggriculture/autolab/20260908T055211Z_B50-V23-FRESH-MIX11-CLEO` — 0-6 / -102071.33.\n- B51 top family prefixes reconciled for Broker/Lena/Silas/Cleo/Rita.\n- B52 Broker: `artifacts/kaggriculture/autolab/20260908T060329Z_B52-V23-FRESH-MATERIALITY-BROKER` — 0-6 / -105264.67.\n- B52 Cleo: `artifacts/kaggriculture/autolab/20260908T060353Z_B52-V23-FRESH-MATERIALITY-CLEO` — 0-6 / -116979.50.\n\nAt 06:13 tick the queue contained 219 DONE / 5 FAILED / 0 READY-RUNNING. No result newer than B52 was interpreted before PASS.\n\n## Falsification — contextual labor\n\nPreregistered: `artifacts/kaggriculture/experiments/KAG_EXP_LABOR_D9_D15_CAUSAL_V1.md`.\n\nFresh paired Broker test `[108301,108313,108327]`, mirrored seats:\n- MIX11: 0-6, avg margin -88401.67.\n- `LABOR10_D9_15`: 0-6, avg margin -87730.50.\n- paired mean delta +671.17 but positive only 2/6; negative 4/6.\n- both large positive episodes belong to seed 108301 mirrored seats (+20576/+19451).\n- D15 gap worsened under treatment in every seed family.\n- treatment detector passed: baseline laborApplied 0/episode; treatment 6/episode; errors 0.\n\nClassification: `TEMPORALLY_BOUNDED_LABOR10_D9_D15 = HELDOUT_FALSIFIED / MATCHUP_UNSTABLE / DO_NOT_PROMOTE`.\n\nThe scalar positive average is a local-world artifact, not competitive breakthrough. Per preregistration no automatic second heldout is justified. This does not reopen the rejected static/universal hands family and does not claim labor is universally irrelevant.\n\n## Opponent-model progress — top-family productive-capacity composition\n\nPreregistered descriptive decomposition: `KAG_EXP_TOP_CAPACITY_COMPOSITION_V1.md`.\n\nFresh legal-public worlds `[108501,108513,108527]` show Broker/Lena/Silas following essentially identical D3→D12 productive composition; Cleo is nearly identical. Rita is a discriminating intermediate control.\n\nTop-family common pattern:\n- D3: ~19 crops = 11 WHEAT + 8 MELON; 4 animals ≈ 3 COW + 1 SHEEP; 1 land.\n- D6: ~18 crops = 6 WHEAT + 10 MELON + 2 STRAWBERRY; 6 animals; 1 land.\n- D9: ~29–30 crops = ~6–7 WHEAT + 11 MELON + 12 STRAWBERRY; 11 animals; 2 land.\n- D12: ~37.7–38.7 crops = ~5–6 WHEAT + 6 MELON + 26.7 STRAWBERRY; 14 animals; 3 land.\n\nMIX11 in the same census remains WHEAT-only through D12 and substantially below the public productive stock. Rita control also remains WHEAT-only with 1 land through D12; it has 4 animals through D9 and ~9.7 at D12. MIX11 vs Rita in the control worlds was 4-2 / +2426.83.\n\nClassification: `LEGAL_PUBLIC_DESCRIPTIVE_TOP_FAMILY_CAPACITY_SIGNATURE / STRONG_CANDIDATE / NOT_CAUSAL`.\n\nThis updates existing taxonomy dimensions — diversification/concentration, crop-portfolio timing, land deployment and animal-capacity timing — but does not add/remove taxonomy traits. All other unmeasured traits remain UNKNOWN.\n\nImportant anti-drift: this signature does **not** license copying the public trajectory. Existing negative knowledge for EARLY_MELON40, PORTFOLIO_D6_244, PHASE3/related tested portfolio formulations, early land/herd and static hands remains intact. The next causal treatment must isolate a genuinely new timing/composition discriminant.\n\n## Queue refill — B53\n\nThe queue was refilled only after PASS and after the new descriptive signature was identified. Five fresh V23 legal-observation jobs use seeds `[108601,108613,108627]`:\n- B53 Broker\n- B53 Lena\n- B53 Silas\n- B53 Cleo\n- B53 Rita control\n\nPurpose: independent fresh-world replication/falsification of the top-family composition signature under canonical sidecar telemetry. Dedupe keys are unique; all five are READY, attempts=0. Queue snapshot after enqueue: 219 DONE / 5 FAILED / 5 READY.\n\nA manual worker trigger through the connector was blocked by the external execution safety layer; this is not a compute/competitive failure. The normal 15-minute worker remains the valid execution path.\n\n## Champion / league\n\n- Champion: unchanged.\n- Working lineage: MIX11.\n- Boundary: Broker Bea.\n- Stress: Closer Cleo.\n- Regression/intermediate: Rancher Rita.\n- Cheap controls: Rosa/Hana/Mateo.\n- No closed negative-knowledge family reopened.\n\n## First incomplete step\n\nConsume B53 after the normal worker executes it and re-run the full context preflight before interpreting. If the D3→D12 top-family composition survives fresh worlds and Rita remains a negative/intermediate control, decompose the new signal into a preregistered **single-factor temporal portfolio** discriminant; do not sweep adjacent shares or fuse crop+land+herd.\n",[],[175,176,2570,2684],"11240c62-2a3c-4f99-82eb-889ebe1e006e","KAG_HOURLY_CHECKPOINT_20260908_0535.md","31d9f65e-ab47-11f1-bbbd-1607ee4eb77e",672,["D",1788845769240],"# KAGGRICULTURE HOURLY CHECKPOINT — 2026-09-08 05:35 UTC\n\nStatus: MATERIAL_UPDATE / CONTEXT_PREFLIGHT_PASS\n\n## KAG_CONTEXT_PREFLIGHT\n\n- `00_READ_FIRST.md`: READ\n- `KAG_STRATEGIC_INTELLIGENCE.md`: READ\n- `KAG_OPPONENT_CAPABILITY_TAXONOMY.md` (190 traits): READ\n- canonical foundations/curriculum/neuron/joined plan: READ\n- `KAG_OFFICIAL_RULES.md`: READ\n- `KAG_RT_ARCHITECTURE.md`: READ\n- `KAG_CURRENT_STATE.md`: READ\n- `KAG_NEGATIVE_KNOWLEDGE.md`: READ\n- reanalysis/coverage/recovery/day-audits: READ\n- opponent profile index + Broker/Cleo/Rita/Lena/Silas/Rosa/Hana/Mateo dossiers: READ\n- `KAG_EXPERIMENT_MATRIX_OPPONENT_PROBES.md`: READ\n- `KAG_PROFILE_CAL_V1_20260907.md`: READ\n- latest prior checkpoint `KAG_HOURLY_CHECKPOINT_20260908_0444.md`: READ\n- latest prior manifest + `HOURLY_INDEX.md`: READ\n- sidecar queue/runs/states: RECONCILED\n- sidecar artifacts/Blob: RECONCILED\n- evidence epochs: RECONCILED\n- contradiction/failure detected: `STALE_SCHEMA_ASSUMPTION` — live `kag_autolab_artifacts` uses `artifact_key`, not `artifact_prefix`; schema was re-read with PRAGMA before evidence interpretation.\n- authority resolution: `chat-native-maestro@economic-v0.6` remains strategy/code/artifact source of truth; sidecar remains compute/evidence only.\n\n**KAG_CONTEXT_PREFLIGHT = PASS**\n\n## Sidecar evidence consumed\n\n### B49 third policy-shift OOD — DEPLOY10 instrumentation only\nFresh seeds `[107501,107513,107527]`:\n- DEPLOY10 × Broker: 0-6, avg margin -90,216.0. Prefix: `artifacts/kaggriculture/autolab/20260908T050915Z_B49-V23-OOD-DEPLOY10-BROKER`.\n- DEPLOY10 × Rita: 6-0, avg margin +26,843.17. Prefix: `artifacts/kaggriculture/autolab/20260908T050921Z_B49-V23-OOD-DEPLOY10-RITA`.\n- DEPLOY10 × Cleo: 0-6, avg margin -80,498.33. Prefix: `artifacts/kaggriculture/autolab/20260908T052413Z_B49-V23-OOD-DEPLOY10-CLEO`.\n\nInterpretation: reinforces `GOOD_VS_RITA != TRANSFERS_TO_BROKER`; no champion/candidate promotion.\n\n## Causal reaction-attribution experiments\n\nAll were run on `kaggriculture-autolab@reaction-attribution-v1`, offline teacher evidence only. Cleo used exact replay of prior **legal public observation history**; raw opponent action was used only as an offline treatment detector, never as a deployable feature.\n\n### 1. WHEAT fixed-history attribution\nScript: `shared_market_fixed_history_attribution_v1.ts`; seeds `[107601,107613]`.\n- 24 market-active MID/LATE/CLOSING contexts per target.\n- Cleo: WHEAT price 0.5x/2x = 0/24 action differences; WHEAT inventory +/-1000 = 0/24; replay mismatch 0.\n- Broker: 0/24 price; 0/24 inventory.\n- Rita positive control: WHEAT price 6/24 action differences, 4/24 immediate public effects.\n- Rosa negative control: 0/24.\n\n### 2. All-item current-price attribution\nScript: `shared_market_all_price_fixed_history_v2.ts`; fresh `[107631,107643]`.\n- Cleo: 0/12 action differences for each of WHEAT/CARROT/TOMATO/STRAWBERRY/MELON/MILK/WOOL/FERTILIZER.\n- Broker: 0/12 for each item.\n- Rita WHEAT positive control: 8/24.\n- Rosa: 0/24.\n\n### 3. All-item current-inventory attribution\nScript: `shared_market_all_inventory_fixed_history_v3.ts`; same fresh set/context stratification.\n- Cleo: 0/12 for every item under +/-1000 inventory intervention; replay mismatch 0.\n- Broker: 0/12 for every item.\n- apparatus control remained positive via Rita WHEAT-price 8/24; Rosa 0/24.\n\n### 4. Broad public-cash curve\nScript: `public_cash_fixed_history_curve_v4.ts`; fresh `[107651,107663]`; grid `[0,199,200,1000,3000,10000,50000]`.\n- Cleo: 2/12 responsive contexts, exclusively 199->200; 0 immediate public-effect pairs.\n- Broker: 4/12 responsive contexts, exclusively 199->200; 0 immediate public-effect pairs.\n- Rita: 0/12; Rosa: 0/12.\n- dedicated Broker 199->200 positive control at step0/step144: 8/8.\n\nNo broad cash-elasticity regime beyond the already-known contextual 199->200 slot behavior was found.\n\n### 5. Public shop/demand-context screen\nScript: `public_shop_fixed_history_probe_v5.ts`; fresh `[107681,107693]`.\nHolding prices/inventories/cash/farms fixed, remove or duplicate one unlocked-shop instance at a time:\n- Cleo/Broker/Rita/Rosa: 32 removal tests + 32 duplication tests per target; 0 action differences throughout.\n- Cleo replay mismatch: 0.\n\n## Material falsification / model update\n\nThe current leading explanation `shared-market instantaneous snapshot elasticity` is **falsified in tested market-active contexts** for Cleo/Broker:\n- not current single-item price;\n- not current single-item inventory;\n- not simple shop multiplicity;\n- not broad our-cash magnitude beyond the already-known contextual 199->200 slot branch.\n\nThis is `TESTED_CONTEXT_NULL`, **not universal absence** and not evidence that market mediation as a whole is absent.\n\nNew lead hypothesis:\n`PATH_DEPENDENT_TIMING_ORDER_SLOT_FUNDING_ATTRIBUTION`.\n\nMechanistic alternatives still live:\n1. legal public-history/path dependence;\n2. lagged market response rather than current snapshot response;\n3. interactions among order slots, funding/cash availability and timing;\n4. endogenous trajectory/time correlation generating the V23 lag signature;\n5. sparse interaction-dependent snapshot support missed by current fixed-context screens.\n\nNext discriminant is preregistered: alter the **legal public market-history path** while restoring/holding the terminal observable snapshot fixed, then measure treatment firing, reaction lag and public/economic consequence with exact-history replay and calibrated controls.\n\n## Trait status changes\n\n190-trait taxonomy preserved; no trait removed.\n- `market_price_sensitivity` top-tier current-snapshot formulation: `NO_ACTION_RESPONSE_IN_TESTED_CONTEXTS`.\n- `market_inventory_sensitivity` top-tier current-snapshot formulation: `NO_ACTION_RESPONSE_IN_TESTED_CONTEXTS`.\n- simple `shop/demand_multiplicity_response`: `NO_ACTION_RESPONSE_IN_TESTED_CONTEXTS`.\n- broad `our_cash_sensitivity`: no new regime through 50k; known 199->200 remains causal/contextual/low-materiality calibration trait.\n- `reaction_latency`, `memory/recency`, `history/path_dependence`, `order-slot/funding interaction`: remain UNKNOWN/priority active traits.\n- fingerprint/lag signature remains `REPLICATED_LEGAL_OBSERVATIONAL / NOT_CAUSAL / NOT_CALIBRATED_POSTERIOR`.\n\n## Negative knowledge / league / champion\n\n- PASTURE_RAMP remains `HELDOUT_FALSIFIED / MATCHUP_UNSTABLE / DO_NOT_PROMOTE`.\n- LAND_AFTER6_CLEAN, PORTFOLIO_D6_244 and recorded early herd/build/melon families remain closed.\n- Champion: unchanged.\n- Working lineage: MIX11.\n- Boundary: Broker Bea.\n- Stress ceiling: Closer Cleo.\n- Regression/intermediate: Rancher Rita.\n- Cheap controls: Rosa/Hana/Mateo.\n\n## Queue after analysis\n\nLive snapshot at 2026-09-08 05:34:34 UTC: 209 DONE / 5 FAILED / 3 READY.\nB50 READY with fresh `[107701,107713,107727]`:\n- `B50-V23-FRESH-MIX11-BROKER`\n- `B50-V23-FRESH-MIX11-RITA`\n- `B50-V23-FRESH-MIX11-CLEO`\n\nThey were queued **after** this round's context preflight. By hard gate, any resulting evidence must wait for the next full context preflight before entering strategy.\n\n## First incomplete step\n\nRun the next context preflight after B50 changes state, consume B50 if eligible, then execute the fixed-terminal-snapshot **history-path intervention** to distinguish path-dependent reaction from endogenous timing/order-slot correlation. No competitive policy mutation is currently justified.\n",[],[175,176,2570,2692],"e34d4617-200d-4a51-bc6f-68f3166dff98","KAG_HOURLY_CHECKPOINT_20260908_0444.md","022d94d0-ab40-11f1-a67e-1607ee4eb77e",664,["D",1788842682780],"# Kaggriculture Hourly Checkpoint — 2026-09-08 04:44 UTC\n\nStatus: MATERIAL_CAUSAL_FALSIFICATION / NO_CHAMPION_PROMOTION\n\n## KAG_CONTEXT_PREFLIGHT\nCanonical bootstrap receipt inherited from this uninterrupted evidence epoch and still valid:\n- L0 canonical mission/intelligence/taxonomy/curriculum/protocol/plans = READ\n- L1 official rules/runtime architecture = READ\n- L2 current state/negative knowledge/audits/reconciliation = READ\n- L3 all opponent dossiers + experiment matrix + calibration = READ\n- L4 hourly index + previous checkpoint/manifest = READ\n- sidecar queue/runs/states/artifacts/blob/evidence epochs = RECONCILED\n- `FOUND != READ` satisfied\n- `KAG_CONTEXT_PREFLIGHT` = **PASS**\n\nNo conflicting canonical artifact appeared during the epoch.\n\n## Material changes after 04:24 checkpoint\n\n### B47 second policy-shift OOD — M60_FERT\nFresh `[107201,107213,107227]`:\n- Broker lag0 +0.1194, lag2 +0.2544\n- Rita lag0 -0.0541\n- Cleo lag0 +0.9837 with rapid decay\n\nThe qualitative ordering/shape survived a second material policy shift. Magnitudes remain policy-dependent.\n\n### B48 Lena/Silas M60 OOD\nFresh `[107301,107313,107327]`:\n- Lena lag0 +0.2066; lag2/3/4 +0.3768/+0.2116/+0.1188\n- Silas lag0 +0.3255; lag2 +0.3251\n\nLena's delayed tail and Silas's immediate+lag2 shape therefore survive MIX11→M60_FERT qualitatively. This remains observational, not a causal latency model or calibrated posterior.\n\n### Cleo exact-history fixed-prestate causal program\nCleo is stateful, so the causal detector was changed from a naive fresh call to **exact legal-public-history replay** into a fresh black-box instance before branching the final observation. This reconstructs observable history-dependent context without `_TRACE`, private replay-plan or online opponent raw-action features.\n\nGeneric H0 landmarks in seed 107401 showed 0/5 sensitivity even across cash `[199,200,1000,10000]`, falsifying `day/phase alone activates Cleo cash response`.\n\nNatural candidate states were then selected only when Cleo black-box emitted WHEAT plus another SELL. Across fresh seeds `[107411,107423,107437]`:\n- 30 candidate prestates\n- 7/30 199→200 treatment-fired sell-slot reorderings\n- timing concentrated D28–D29, mostly D29\n- 1/30 natural forks produced any immediate public consequence: Cleo cash -8 / Rosa +7\n- controlled BUY_PRODUCT WHEAT contention up to 5 units produced 0/7 non-zero effects\n\nClassification:\n`CLEO_OUR_CASH_SLOT_REACTION = CAUSAL_REPLICATED / CLOSING_CONTEXTUAL / LOW_OBSERVED_MATERIALITY`.\n\nFalsified simple mechanism:\n`CLEO_LARGE_LEGAL_LAG0_SIGNATURE == MATERIAL_WHEAT_FRONT_RUN_THRESHOLD` = **FALSE in tested formulations**.\n\nThe large legal lag0 signature and the small direct cash-threshold mechanism are separate evidence objects. Next causal target is shared-market / price-inventory / funding response.\n\n## Champion / regression\n- Champion: unchanged\n- Working baseline: MIX11\n- Boundary: Broker Bea\n- Stress: Closer Cleo\n- Regression/intermediate: Rancher Rita\n- Negative knowledge: unchanged; no closed policy/hypothesis reopened\n- M40/M60/DEPLOY10 are instrumentation/OOD vehicles only, not promotion candidates\n\n## Sidecar\nConsumed since prior checkpoint:\n- B47 Broker: `artifacts/kaggriculture/autolab/20260908T042639Z_B47-V23-OOD-M60-BROKER`\n- B47 Rita: `.../20260908T042644Z_B47-V23-OOD-M60-RITA`\n- B47 Cleo: `.../20260908T042708Z_B47-V23-OOD-M60-CLEO`\n- B48 Lena: `.../20260908T044055Z_B48-V23-OOD-M60-LENA`\n- B48 Silas: `.../20260908T044100Z_B48-V23-OOD-M60-SILAS`\n\nFinal census:\n- 206 DONE queue jobs\n- 5 FAILED terminal queue jobs\n- 3 READY\n- 244 run attempts = 206 DONE + 38 FAILED attempts\n- 214 distinct queue jobs\n\nREADY B49 non-fertilizer OOD, fresh/deduped `[107501,107513,107527]`:\n- `B49-V23-OOD-DEPLOY10-BROKER`\n- `B49-V23-OOD-DEPLOY10-RITA`\n- `B49-V23-OOD-DEPLOY10-CLEO`\n\nPurpose: determine whether qualitative lag ordering transfers to a structurally different non-fertilizer policy shift. DEPLOY10 is not reopened/promoted.\n\n## Durable updates\n- `KAG_CURRENT_STATE.md`\n- `KAG_OPPONENT_PROFILES/cleo.md`\n- sidecar states `hourly:policy_shift_ood_v2`, `hourly:cleo_cash_causal_v2`\n- checkpoint/manifest/diff/index\n\nTaxonomy remains 190 traits; unmeasured traits remain UNKNOWN.\n\n## First incomplete step\n1. Consume B49 and classify policy-conditioning versus qualitative robustness.\n2. Start fixed-history shared-market intervention: perturb legal shared market price/inventory while holding opponent-cash treatment fixed, with positive/negative controls, to separate direct cash response from market-mediated/funding response.\n3. Measure treatment-fired rate, public consequence, lag, effective n and OOD before considering any online routing or bait/probe strategy.\n",[],[175,176,2570,2700],"5ac6f1a1-222f-41a1-98b2-dca34caee392","KAG_HOURLY_CHECKPOINT_20260908_0424.md","310b8918-ab3d-11f1-9345-1607ee4eb77e",655,["D",1788841472729],"# Kaggriculture Hourly Checkpoint — 2026-09-08 04:24 UTC\n\nStatus: MATERIAL_OPPONENT_MODEL_PROGRESS / NO_CHAMPION_PROMOTION\n\n## KAG_CONTEXT_PREFLIGHT\n\n- `00_READ_FIRST.md` = READ\n- `KAG_STRATEGIC_INTELLIGENCE.md` = READ\n- `KAG_OPPONENT_CAPABILITY_TAXONOMY.md` = READ\n- `KAGGRICULTURE_REAL_WORLD_CURRICULUM_V1.md` = READ\n- neuron protocol = READ\n- joined research plan = READ\n- `KAG_OFFICIAL_RULES.md` = READ\n- `KAG_RT_ARCHITECTURE.md` = READ\n- `KAG_CURRENT_STATE.md` = READ\n- `KAG_NEGATIVE_KNOWLEDGE.md` = READ\n- full strategic reanalysis / recovery / autolab audit / context continuity audit = READ\n- opponent profile index = READ\n- Broker/Rosa/Hana/Mateo/Rita/Lena/Silas/Cleo dossiers = READ\n- opponent probe matrix = READ\n- profile calibration = READ\n- `HOURLY_INDEX.md` = READ\n- previous checkpoint `KAG_HOURLY_CHECKPOINT_20260908_0246.md` = READ\n- previous manifest `KAG_HOURLY_MANIFEST_20260908_0246.json` = READ\n- sidecar queue = RECONCILED\n- sidecar runs = RECONCILED\n- sidecar states = RECONCILED\n- sidecar artifacts = RECONCILED\n- latest blob = RECONCILED\n- evidence epochs = RECONCILED against canonical full audit\n- contradictions = initial new cash probe omitted known positive-control contexts; classified and corrected before inference\n- stale canonical files affecting this round = NONE DETECTED\n- `KAG_CONTEXT_PREFLIGHT` = **PASS**\n\nNo result was interpreted and no new strategic job was enqueued before PASS.\n\n## Material causal/falsification updates\n\n### Broker cash awareness bounded\nFresh fixed-prestate tests recovered Broker 199→200 treatment firing while weak trio remained negative. Controlled WHEAT-contention follow-up on fresh seeds `[106401,106413,106427]` produced 101 comparable treatment-fired states; 13 had non-zero economic effect (12.87%), maximum 4 coins. This independently confirms the mechanism but bounds its materiality as low/contextual.\n\nClassification: `CAUSAL_REPLICATED / LOW_MATERIALITY / CALIBRATION_TRAIT`, not a primary boundary explanation.\n\nFailure learned: `POSITIVE_CONTROL_CONTEXT_OMISSION`. An initial detector omitted step0/D6H0 and produced a false null. Null reaction claims now require known positive + negative controls to pass first.\n\n### Upper-family fixed-prestate label\nAt fresh step0 and step144/D6H0 prestates:\n- Broker 199→200 changed treatment 6/6;\n- Lena 6/6;\n- Silas 6/6;\n- Cleo 0/6;\n- Rita 0/6.\n\nCleo's null is contextual, not global evidence of no awareness; causal Cleo context remains unresolved.\n\n## V23 telemetry promotion\n\nSidecar main advanced to `KAG_AUTOLAB_CORE_V23_LEGAL_SELF_MEMORY_MAIN`, adding only our own legal market-flow memory. Same fresh seed 106601 before/after V22→V23 produced identical terminal banks, margins, W/L, errors and 719 decision calls in both seats; V23 added 719 self-action records/game. Scope: telemetry-only mechanical promotion, not policy/champion proof.\n\n## Replicated legal opponent timing signatures\n\nAll following are `LEGAL_OBSERVATIONAL / SELF_ACTION_FILTERED / NOT_CAUSAL`.\n\nB43 MIX11 `[106801,106813,106827]`:\n- Broker: lag0 +0.3565, lag2 +0.1989\n- Rita: lag0 -0.0355\n- Lena: lag0 +0.2250; lag2/3/4 +0.1824/+0.2571/+0.2850\n- Silas: lag0 +0.6656; lag2 +0.2000\n- Cleo: lag0 +2.4873, rapid decay\n\nB44 preregistered fresh Broker/Rita/Cleo `[106901,106913,106927]`:\n- Broker: +0.3815 lag0, +0.2021 lag2\n- Rita: +0.0089 lag0\n- Cleo: +2.2374 lag0, rapid decay\nResult: preregistered ordering/shape PASS.\n\nB45 independent Lena/Silas `[107001,107013,107027]`:\n- Lena: +0.4482 lag0; +0.1983/+0.2278/+0.2334 lags2/3/4\n- Silas: +0.8589 lag0; +0.1767 lag2\nResult: both signatures independently replicated.\n\n### Policy-shift OOD B46\nM40_FERT was used only as an instrumentation policy shift, not reopened/promoted. Fresh `[107101,107113,107127]`:\n- Broker: +0.2321 lag0; +0.2464 lag2\n- Rita: -0.0348 lag0\n- Cleo: +1.0135 lag0 followed by decay\n\nQualitative ordering survived, numerical magnitude changed. Therefore `QUALITATIVE_POLICY_SHIFT_TRANSFER = ONE_OOD_PASS`; quantitative policy invariance remains unproven.\n\n## Competitive/regression observations\n\nNo champion change.\n\nFresh calibration batches continued to show:\n- MIX11 consistently 0-6 against Broker/Lena/Silas/Cleo in these upper-family traces;\n- Rita varied by seed set but remained regression/intermediate (B42 6-0, B43 6-0, B44 5-1 for MIX11);\n- M40_FERT B46 was 0-6 vs Broker, 6-0 vs Rita, 0-6 vs Cleo and is **not** a promotion claim.\n\nLeague remains:\n- working baseline/candidate lineage: MIX11\n- boundary: Broker Bea\n- stress ceiling: Closer Cleo\n- regression/intermediate: Rancher Rita\n- cheap controls: Rosa/Hana/Mateo\n\nNegative knowledge unchanged; PASTURE_RAMP and all previously closed families remain closed under canonical reopening gates.\n\n## Sidecar jobs consumed this epoch\n\nConsumed/validated artifact prefixes include:\n- B39 Broker/Rosa/Hana/Mateo: `20260908T033115Z`, `033259Z`, `033305Z`, `033327Z`\n- B40 Broker/Rosa/Hana/Mateo: `033956Z`, `034001Z`, `034037Z`, `034042Z`\n- B41 Broker/Rita/Lena/Silas: `034611Z`, `034617Z`, `035302Z`, `035307Z`\n- B42 Broker/Rita/Lena/Silas/Cleo: `035919Z`, `035925Z`, `035944Z`, `035950Z`, `040007Z`\n- B43 Broker/Rita/Lena/Silas/Cleo: `040645Z`, `040651Z`, `040711Z`, `040717Z`, `040736Z`\n- B44 Broker/Rita/Cleo: `041219Z`, `041225Z`, `041246Z`\n- B45 Lena/Silas: `041801Z`, `041806Z`\n- B46 Broker/Rita/Cleo: `042008Z`, `042014Z`, `042034Z`\n\nAll are autolab compute/lab evidence, not champion proof.\n\nFinal sidecar census at checkpoint:\n- queue: 201 DONE / 5 FAILED / 3 READY\n- run attempts: 239 = 201 DONE + 38 FAILED attempts\n- distinct queue jobs: 209\n- latest completed run: 2026-09-08 04:20:34 UTC\n\nREADY B47, fresh/deduped `[107201,107213,107227]`:\n- `B47-V23-OOD-M60-BROKER`\n- `B47-V23-OOD-M60-RITA`\n- `B47-V23-OOD-M60-CLEO`\n\nPurpose: second materially different our-policy-shift falsifier of the legal timing signature. M60_FERT is an instrumentation vehicle only; this does not reopen/promote it.\n\n## Durable changes\n\nUpdated:\n- `KAG_CURRENT_STATE.md`\n- `KAG_STRATEGIC_INTELLIGENCE.md`\n- Broker dossier\n- Cleo dossier\n- sidecar V23 core telemetry and state ledgers\n- this checkpoint + manifest + diff + hourly index\n\nTaxonomy remains canonical at 190 traits; no new dimension was required. Unmeasured traits remain UNKNOWN.\n\n## First incomplete step\n\n1. Consume B47 and determine whether qualitative lag ordering survives a second policy shift.\n2. Recover naturally occurring fixed-prestate Cleo contexts where visible-cash intervention actually fires a slot/funding treatment; then quantify public/economic consequence.\n3. Use V23 self-memory to separate our direct market contribution from remaining shared-market-mediated response; do not promote a causal latency model or calibrated posterior before those gates.\n",[],[175,176,2570,2708],"2b6f818f-e167-4eca-be58-9204f61a233a","KAG_HOURLY_CHECKPOINT_20260908_0345.md","d274bb72-ab37-11f1-ac71-1607ee4eb77e",635,["D",1788839166739],"# Kaggriculture Hourly Checkpoint — 2026-09-08 03:45 UTC\n\nStatus: MATERIAL_REACTION_ATTRIBUTION_CALIBRATION / NO_CHAMPION_PROMOTION\n\n## KAG_CONTEXT_PREFLIGHT\n\n- READ_FIRST = READ\n- STRATEGIC_INTELLIGENCE = READ\n- CAPABILITY_TAXONOMY = READ\n- REAL_WORLD_CURRICULUM = READ\n- NEURON_PROTOCOL = READ\n- JOINED_RESEARCH_PLAN = READ\n- OFFICIAL_RULES = READ\n- RT_ARCHITECTURE = READ\n- CURRENT_STATE = READ\n- NEGATIVE_KNOWLEDGE = READ\n- FULL_15MIN_STRATEGIC_REANALYSIS = READ\n- RECOVERY_RECONCILIATION = READ\n- AUTOLAB_DAY_AUDIT = READ\n- CONTEXT_CONTINUITY_AUDIT = READ\n- OPPONENT_INDEX = READ\n- TARGET_DOSSIER_BROKER = READ\n- CONTROL_AND_LADDER_DOSSIERS = READ_ALL_8\n- EXPERIMENT_MATRIX = READ\n- PROFILE_CALIBRATION = READ\n- HOURLY_INDEX = READ\n- LATEST_CHECKPOINT_0246 = READ\n- LATEST_MANIFEST_0246 = READ\n- SIDECAR_QUEUE = RECONCILED\n- SIDECAR_RUNS = RECONCILED\n- SIDECAR_STATE = RECONCILED\n- SIDECAR_ARTIFACTS = RECONCILED\n- SIDECAR_BLOB = RECONCILED\n- EVIDENCE_EPOCHS = RECONCILED_WITH_CANONICAL_FULL_AUDIT\n- CONTRADICTIONS = [`initial reaction_cash_fixedprestate_v1 omitted known positive-control contexts and produced a false null; probe design corrected before opponent inference`]\n- STALE_FILES = []\n- CONTEXT_PREFLIGHT = PASS\n\n`FOUND != READ` satisfied. No sidecar result was interpreted and no new queue work was created before this preflight reached PASS.\n\n## Material evidence consumed/generated\n\n### B39 legal calibration — fresh seeds `[106001,106013,106027]`\n- MIX11 × Broker: 0-6, avg margin -93,436.67; artifact prefix `artifacts/kaggriculture/autolab/20260908T033115Z_B39-LEGAL-BASELINE-BROKER`.\n- MIX11 × Rosa: 6-0, +40,190.17; prefix `artifacts/kaggriculture/autolab/20260908T033259Z_B39-LEGAL-BASELINE-ROSA`.\n- MIX11 × Hana: 6-0, +35,574.67; prefix `artifacts/kaggriculture/autolab/20260908T033305Z_B39-LEGAL-BASELINE-HANA`.\n- MIX11 × Mateo: 6-0, +26,763.00; prefix `artifacts/kaggriculture/autolab/20260908T033327Z_B39-LEGAL-BASELINE-MATEO`.\n\nThe first Rosa attempt failed deterministically because historical alias `rookie_rosa.py` was absent from the live reference catalog. The catalog established canonical weak names `rotation_rosa.py`, `homestead_hana.py`, `melon_mateo.py`. Sidecar alias normalization was repaired before rerun. Classification: `ASSET_ALIAS_SCHEMA`, not competitive failure.\n\n### B40 legal fresh holdout — seeds `[106211,106223,106237]`\n- MIX11 × Broker: 0-6, -104,281.00; prefix `artifacts/kaggriculture/autolab/20260908T033956Z_B40-LEGAL-HOLDOUT-BROKER`.\n- MIX11 × Rosa: 6-0, +42,941.00; prefix `artifacts/kaggriculture/autolab/20260908T034001Z_B40-LEGAL-HOLDOUT-ROSA`.\n- MIX11 × Hana: 6-0, +34,645.17; prefix `artifacts/kaggriculture/autolab/20260908T034037Z_B40-LEGAL-HOLDOUT-HANA`.\n- MIX11 × Mateo: 6-0, +32,267.17; prefix `artifacts/kaggriculture/autolab/20260908T034042Z_B40-LEGAL-HOLDOUT-MATEO`.\n\nThese W/L results are calibration/regression context; they are not reaction proof.\n\n## Opponent-model progress\n\nA legal-trace-only observational screen was run without opponent raw actions. Broker showed a replicated association between large changes in our cash and subsequent public opponent-response score:\n\n- B39: cash lift +0.362; low-shared-market-movement subset +0.380.\n- B40: cash lift +0.303; low-market subset +0.250.\n\nWeak-trio cash lifts did not reproduce a comparable group pattern. This updates `legal cash-response signature` to `REPLICATED_OBSERVATIONAL_DETECTOR_CANDIDATE`, not causal elasticity and not a calibrated posterior.\n\nLand/animals also produced observational associations, but land had substantial Rosa association in B40 and animals moved across controls; these remain confounded candidate traits and were not promoted.\n\n## Fixed-prestate causal calibration\n\nTool: `kaggriculture-autolab@main/tools/reaction_cash_fixedprestate_v1.ts`.\n\nFresh official-engine fixed prestates were generated independently, then only our publicly visible cash in the opponent observation was varied over `[0,199,200,500,1000,3000,10000,50000]`. Raw opponent action was used only as an OFFLINE treatment-fired teacher label; it is not a deployable feature.\n\nThe first version targeted steps `[57,145,289,433,577]`. It failed the known Broker positive control. Canonical calibration was reread and showed that step 0 and D6H0/step144 are necessary known contexts. After correcting the probe to `[0,57,144,289,433,577]`:\n\n- Broker 199→200 action difference: 6/18 fresh prestates.\n- Rosa: 0/18.\n- Hana: 0/18.\n- Mateo: 0/18.\n- Broker immediate material public next-state difference under the tested fixed baseline action: 0/18.\n\nTherefore:\n- `BROKER_DIRECT_OUR_CASH_AWARENESS = CAUSAL_RECONFIRMED_ACTION_LEVEL`.\n- `IMMEDIATE_ECONOMIC_MATERIALITY = ZERO_IN_THIS_SCREEN / KNOWN_CONTEXT_DEPENDENT_SMALL_EFFECT_FROM_PRIOR_SCREEN`.\n- `DIRECT_AWARENESS != PRIMARY_BOUNDARY_CAUSE` remains enforced.\n\nThe zero immediate effect does not contradict the prior canonical 33-state contention screen: only 6/33 states changed controlled WHEAT purchase cost, max 4 coins.\n\n## Falsification / learning\n\nOBSERVATION: initial new cash probe returned no Broker 199→200 response.\nCAUSE: target-state set omitted validated positive-control contexts step0 and D6H0.\nFAILURE CLASS: `POSITIVE_CONTROL_CONTEXT_OMISSION`.\nCONTROL: every new reaction detector must include at least one canonical positive-control context and relevant negative controls before a null can update opponent knowledge.\nDETECTOR: fail the measurement stack if known Broker threshold or weak-trio controls do not reproduce.\nREGRESSION: repaired probe now recovers Broker 6/18 and weak trio 0/18.\n\n## Trait / hypothesis updates\n\n- `our_cash_sensitivity` / Broker: stronger causal evidence; context-dependent.\n- weak trio `our_cash_sensitivity`: negative evidence strengthened in tested fixed prestates; not universal absence.\n- `direct_awareness_probability`: qualitatively strengthened for Broker, no numeric posterior claimed.\n- `reaction_magnitude`: still not calibrated; immediate public effect zero in this fixed-action screen.\n- `market_mediated_reaction_probability`: UNKNOWN.\n- `reaction_latency_distribution`: UNKNOWN.\n- `hysteresis_duration`, `reversal_speed`, `history/path dependence`: UNKNOWN.\n- land/capacity/animals reaction traits: observational candidates only; no causal promotion.\n- remaining taxonomy fields remain UNKNOWN unless prior canonical evidence exists.\n\nCurrent hypothesis: `REACTION_ATTRIBUTION_CASH_MEDIATION_CALIBRATION` — cash awareness is a valid calibration mechanism/legal signature, but likely low strategic materiality unless market contention turns slot ordering into consequence. Opposite hypothesis remains that legal cash association is mostly endogenous timing/world-state correlation.\n\n## Champion / league\n\n- Champion: unchanged / no promotion.\n- Working baseline: MIX11.\n- Boundary: Broker Bea.\n- Stress ceiling: Closer Cleo.\n- Regression/intermediate: Rancher Rita.\n- Cheap controls: Rosa / Hana / Mateo.\n- Negative knowledge unchanged; PASTURE_RAMP and prior rejected families remain closed under their reopening gates.\n\n## Sidecar state and jobs\n\nLive census after B40 and B41 enqueue:\n- run attempts: 217;\n- DONE attempts: 179;\n- FAILED attempts: 38;\n- distinct queue jobs: 188;\n- queue rows: 179 DONE / 5 FAILED terminal / 4 READY.\n\nREADY, fresh/deduped B41 family-calibration jobs, seeds `[106301,106313,106327]`:\n- `B41-LEGAL-FAMILY-BROKER`;\n- `B41-LEGAL-FAMILY-RITA`;\n- `B41-LEGAL-FAMILY-LENA`;\n- `B41-LEGAL-FAMILY-SILAS`.\n\nPurpose: calibrate whether legal cash-response signatures distinguish known direct-awareness upper family from Rita/intermediate under fresh worlds. These are detector-calibration jobs, not champion proof.\n\n## Durable updates\n\nUpdated:\n- `artifacts/kaggriculture/intelligence/KAG_CURRENT_STATE.md`;\n- `artifacts/kaggriculture/intelligence/KAG_OPPONENT_PROFILES/broker.md`;\n- sidecar alias normalization and reaction-analysis state;\n- this checkpoint + corresponding manifest + hourly index.\n\nNo taxonomy dimension was added; the canonical 190-trait schema remains sufficient. Negative knowledge did not change.\n\n## First incomplete step\n\nRun a controlled fixed-prestate **WHEAT-contention mediation probe**: preserve state and our action except for a controlled WHEAT purchase/market contention opportunity, force the known Broker 199/200 perception treatment, verify treatment fired, and measure whether the action reorder produces a legal public-market consequence. This is a detector/materiality calibration, not an exploit search.\n\nIn parallel, consume B41 when DONE and test cross-family/OOD stability before any online reaction router is considered.\n",[],[175,176,2570,2716],"6dd09e0a-51bc-42c4-b55e-530887046885","KAG_HOURLY_CHECKPOINT_20260908_0246.md","2be778c2-ab31-11f1-94a7-1607ee4eb77e",619,["D",1788836310320],"# Kaggriculture Hourly Checkpoint — 2026-09-08 02:46Z\n\nStatus: MATERIAL_STRATEGIC_REANALYSIS / NO_CHAMPION_PROMOTION\nLocal time: 2026-09-07 23:46 BRT\nCanonical audit: `../intelligence/KAG_15MIN_FULL_STRATEGIC_REANALYSIS_20260908.md`\n\n## KAG_CONTEXT_PREFLIGHT\n\nREAD_FIRST=READ; STRATEGIC_INTELLIGENCE=READ; CAPABILITY_TAXONOMY=READ; OFFICIAL_RULES=READ; CURRENT_STATE=READ; NEGATIVE_KNOWLEDGE=READ; OPPONENT_INDEX=READ; ALL_8_DOSSIERS=READ; EXPERIMENT_MATRIX=READ; PROFILE_CAL=READ; HOURLY_INDEX=READ; SIDECAR_STATE=RECONCILED; `CONTEXT_PREFLIGHT=PASS`.\n\n## Full 15-minute census\n\nAll persisted compute-active 15-minute windows in sidecar run history were located and reinterpreted:\n- 208 attempts;\n- 171 DONE;\n- 37 FAILED;\n- 176 distinct job IDs;\n- 55 compute-active 15-minute buckets;\n- 1,272 game episodes: 782 OUR_VS_REFERENCE + 490 PUBLIC_CROSSPLAY;\n- first final-snapshot run 2026-09-07 08:38:32Z;\n- latest 2026-09-08 02:40:35Z.\n\nBoundary: empty/no-op scheduled ticks are not fully reconstructible because historical no-work ticks were not append-logged.\n\n## Strategic relabeling\n\nMost early compute was own-policy mutation + ladder/fingerprint work, not causal opponent-reaction profiling. It remains valuable as failure atlas, negative knowledge, matchup/world-regime evidence and mechanism-screen evidence.\n\nMaterial aggregate conclusions:\n- MIX11 vs Broker 0-116; M60_FERT vs Broker 0-92. Broker outcome is robust to the tested our-policy basket, but `OUR_POLICY_SHIFT_ROBUSTNESS_OF_OUTCOME != OPPONENT_ADAPTATION`.\n- MIX11 vs Rita 67-31 across 98 episodes; Rita is matchup-specific regression evidence, not a universal proxy.\n- canonical direct ladder: Lena>Broker 123-9; Silas>Lena 80-12; Cleo>Silas 69-9; Cleo>Broker 41-5; Broker>Rita 12-0.\n- B38: direct Silas>Lena 8-0 while M60_FERT loses to both on same third-party seeds with only 69.625 average-margin difference. New rule: `LADDER_STRENGTH != MATCHUP_EFFECT_AGAINST_US`.\n- PASTURE B31 remains the concrete proof that `PAIRED_CONTROL_INCOMPLETE => DO_NOT_INTERPRET_TREATMENT_DIRECTION`.\n- `ADAPT_V1` rejects only the tested crude uncalibrated snapshot crop-router, not opponent modeling/adaptation as a research class.\n- all 37 failed attempts are infrastructure/schema/test failures, not competitive losses.\n\n## Trait-coverage correction\n\nThe 1,272 episodes strongly/partially populate identity/family, ladder, selected price/cash sensitivity, some slot-ordering, outcome robustness and coarse regime timing. Most original OBG reaction/adaptation traits remain UNKNOWN: advance/retreat in response to us, aggression/defensiveness, direct land/labor/crop/animal/capacity sensitivity, effect/lag curves, adaptation, hysteresis/reversal, history/path dependence, repeated-probe learning, risk/lead/comeback, substitution matrix, opportunism/resource denial, patience/impulsivity, exploration/exploitation, composure, bait/bluff/counter-bluff, `models_us`, reasoning depth, deception detection and probe awareness.\n\n`LARGE_EPISODE_COUNT != LARGE_TRAIT_COVERAGE`.\n\n## Strategy update\n\nCurrent live hypothesis advanced to `REACTION_ATTRIBUTION_PROGRAM_NEXT`:\nlegal turn-level traces → controlled observable probe → direct-vs-market-mediated attribution → reaction magnitude/lag → reversal/hysteresis → history/repeated probes → advance/retreat/risk/substitution → bait/bluff only after calibrated detectors → OOD/cross-policy → student routing.\n\nWeak trio remain positive/negative controls; Rita intermediate/regression; Broker boundary reaction-function target; Lena/Silas/Cleo upper-ladder temporal/slot/funding/strategic-depth targets.\n\n## Sidecar state\n\nFinal audit queue snapshot: 171 DONE rows, 5 FAILED terminal rows, no READY/RUNNING jobs. Core/worker evidence remains `NOT_CHAMPION_PROOF`. No champion/baseline change.\n\n## Durable changes\n\nUpdated/created:\n- `../intelligence/KAG_15MIN_FULL_STRATEGIC_REANALYSIS_20260908.md`;\n- `../intelligence/KAG_CURRENT_STATE.md`;\n- `../intelligence/KAG_NEGATIVE_KNOWLEDGE.md`;\n- Broker/Rita/Lena/Silas/Cleo dossiers;\n- `../00_READ_FIRST.md`;\n- `../KAG_CONTEXT_MANIFEST.json`;\n- sidecar context state + current hypothesis.\n\n## First incomplete step\n\nDesign the first post-audit reaction-attribution experiment with treatment-fired controls and legal next-state recovery. Do not resume adjacent static mutation search.",[],[175,176,2570,2724],"0b3e18c3-c666-4110-a353-8fbd219e2f46","KAG_HOURLY_CHECKPOINT_20260908_0227.md","b528c590-ab2d-11f1-baee-1607ee4eb77e",583,["D",1788834822626],"# Kaggriculture Hourly Checkpoint — 2026-09-08 02:27 UTC\n\nEvidence state: MATERIAL_UPDATE / NO_CHAMPION_PROMOTION\n\n## RECONCILIATION\nSource of truth reconciled before hypothesis formation: strategic intelligence, official rules, negative knowledge, Broker/Lena dossiers, latest hourly checkpoint, autolab day audit. PASTURE_RAMP remains HELDOUT_FALSIFIED / MATCHUP_UNSTABLE / DO_NOT_PROMOTE. No rejected branch was reopened.\n\n## SIDECAR RESULTS CONSUMED\n### B34 recovery\n- `OUR-MIX11-BROKER-LEGAL-B34`: DONE after blob recovery; evidence `AUTOLAB_COMPUTE_RECOVERED_FROM_BLOB_NOT_CHAMPION_PROOF`; artifact prefix `artifacts/kaggriculture/autolab/20260908T015219Z_OUR-MIX11-BROKER-LEGAL-B34`; MIX11 0-8 Broker, avg margin -97206.125.\n- `OUR-MIX11-LENA-LEGAL-B34`: DONE after blob recovery; artifact prefix `artifacts/kaggriculture/autolab/20260908T015226Z_OUR-MIX11-LENA-LEGAL-B34`; MIX11 0-8 Lena, avg margin -96945.25.\n- Full blobs were already written before the previous SQL-index Hrana failure; compact recovery reconciled compute vs persistence. Infrastructure failure was not counted as competitive evidence.\n\n### B35\n- `PUBLIC-LENA-BROKER-LEGAL-B35`: DONE; artifact prefix `artifacts/kaggriculture/autolab/20260908T022304Z_PUBLIC-LENA-BROKER-LEGAL-B35`; Lena 8-0 Broker, avg margin +3264.875.\n- `OUR-MIX11-BROKER-LEGAL-B35`: DONE; artifact prefix `artifacts/kaggriculture/autolab/20260908T022311Z_OUR-MIX11-BROKER-LEGAL-B35`; MIX11 0-8 Broker, avg margin -126451.375.\n\n## OBG / PROFILER MATERIAL UPDATE\nFresh B34 paired legal traces show Broker and Lena are identical through step 289 and first diverge at step 290 = D12 H2 in 8/8 episodes across four seeds and both seats.\nAt the first divergence, static public farm assets remain identical. The repeated signature is Broker retaining roughly 114-152 more cash while shared WHEAT inventory is 3-4 units higher than in the Lena counterfactual; WHEAT price sometimes differs by one tick.\nAn older cross-policy fixed experiment independently reproduces the same D12 H2 separation across MIX11, M60_FERT, ADAPT_V1, DP_H, M50_FERT and other policies: Lena consumes exactly four more WHEAT units than Broker in the tested paired worlds.\nHowever the frozen generic-feature classifier degrades under cross-policy OOD (75-87.5% on tested variants, with Broker→Lena errors). Therefore generic feature compression is not robust enough. The durable direction is a policy-normalized legal landmark: market delta minus known contribution of our own policy/action, with abstention when residual evidence is weak.\nClassification of D12H2 landmark: STRONG_CROSSPOLICY_LEGAL_IDENTITY_LANDMARK. It is not yet an economic exploit.\n\n## CAUSAL FALSIFICATION\nDirect fresh Lena×Broker B35 provides the counterproof against treating D12 H2 as the primary economic mechanism. Lena wins 8-0, but mean public cash gap is 0 through D12, -65 at D15, +200 D18, +625.5 D21, +1050.125 D24, +2503 D27 and +3399.25 D29. Sampled public land/crops/pastures/animals/crop mix stay equal on average at all these landmarks.\nLargest Lena advantage growth is D24→D27 (+1452.875), not D12. Thus D12H2 is promoted as profiler landmark but downgraded as causal economic explanation.\nThe late advantage is best described legally as PUBLIC-STATE-NEUTRAL CASH CONVERSION / MARKET-EXECUTION TIMING. Private inventory is not observed and no hidden action/replay telemetry is used.\nGeneric SELL reordering is not the answer: existing `kag_market_slot_ablation_v1.ts` shows INPLACE_SELL_SORT and HOIST_SELLS both slightly worse than unchanged overall and still 0-win vs Broker. Do not reopen generic late liquidation/sell-sort branches.\n\n## ACTIVE HYPOTHESIS\n`LATE_CONVERSION_RESIDUAL_V1` — the transferable mechanism, if any, is not static farm composition and not generic sell hoisting; it is a conditional market-execution/conversion rule that accumulates cash advantage after D21 while preserving the same sampled public farm footprint.\nOpposite hypothesis: the late cash separation depends on opponent-specific/private inventory state and is not transferable from public observations alone.\nDiscriminant: fresh direct Lena×Broker replication plus Silas×Lena and second paired MIX11 Broker/Lena legal heldout. Only if the late cash-with-equal-public-footprint pattern repeats will a single-factor teacher treatment be preregistered.\n\n## LEAGUE\nChampion: unchanged.\nCandidate lineage: MIX11.\nBOUNDARY: Broker Bea.\nSTRESS: Closer Cleo.\nREGRESSION: Rancher Rita; Rosa/Hana/Mateo remain cheap regression suite.\n\n## B36 QUEUED\nFresh, deduplicated, legal-only:\n- `OUR-MIX11-BROKER-LEGAL-B36` seeds 53621-53624.\n- `OUR-MIX11-LENA-LEGAL-B36` seeds 53621-53624.\n- `PUBLIC-LENA-BROKER-LEGAL-B36` seeds 53601-53604.\n- `PUBLIC-SILAS-LENA-LEGAL-B36` seeds 53611-53614.\n\n## FIRST INCOMPLETE STEP\nConsume B36 and determine whether (1) D12H2 residual landmark repeats on the second fresh heldout and (2) late public-state-neutral cash conversion generalizes from Lena>Broker to Silas>Lena. No online policy mutation before that discriminant.",[],[175,176,2570,2732],"7f7a3c22-0af0-48e4-932f-f198f4ab5e80","KAG_HOURLY_CHECKPOINT_20260907_2220.md","c3d18532-ab23-11f1-bc2b-1607ee4eb77e",578,["D",1788830552231],"# KAG Hourly Checkpoint — 2026-09-07 22:20 America/Fortaleza\n\nEvidence state: ARTIFACT_PARTIAL (legal B34 probes blocked by repeated sidecar persistence/runtime infrastructure error; no competitive inference from those failures).\n\n## Reconciled source of truth\nRead before hypothesis: KAG_STRATEGIC_INTELLIGENCE.md, KAG_NEGATIVE_KNOWLEDGE.md, and latest checkpoint 20260907_2131. Negative-knowledge hard gate preserved. PASTURE_RAMP remains HELDOUT_FALSIFIED / MATCHUP_UNSTABLE / DO NOT PROMOTE; no adjacent blind mutation queued.\n\n## Sidecar consumed / reconciled\n- PUBLIC-CLEO-BROKER-B34: DONE, artifact_prefix `artifacts/kaggriculture/autolab/20260908T003113Z_PUBLIC-CLEO-BROKER-B34`. Cleo 8-0 Broker, avg margin +4005.125. Largest observed public gap growth D9→D12 = +4871.625. This strengthens localization of the superior mechanism to the transition entering D12, but does not identify a causal factor by itself.\n- OUR-MIX11-BROKER-LEGAL-B34 and OUR-MIX11-LENA-LEGAL-B34: repeated FAILED attempts with `TypeError: resp.body?.cancel is not a function` in @libsql/hrana-client during the worker execution/persistence path. No result_json/artifact_prefix. Classified INFRA_TRANSIENT / NO_COMPETITIVE_EVIDENCE.\n- MIX11×Rita B33 remains 6-2, +483.75 and is already incorporated regression evidence.\n\n## LOOP decision\nNo strategy mutation is justified before legal Broker/Lena telemetry exists. The discriminant remains: identify the first material legal-observation difference around D12 H1-H2 and connect it to a measurable economic mechanism, rather than infer from hidden actions or reopen rejected static knobs.\n\nThe two B34 legal jobs were reset READY with attempts=0/max_attempts=6 and fresh retry authorization because the failure is infrastructure, not hypothesis falsification. An immediate worker retry reproduced the same hrana error for both, leaving both READY. Queue therefore remains informative and non-empty without adding redundant jobs.\n\n## League\nChampion: unchanged.\nCandidate lineage: MIX11.\nBOUNDARY: Broker Bea.\nSTRESS: Closer Cleo.\nREGRESSION: Rancher Rita + cheap Rosa/Hana/Mateo suite.\n\n## First incomplete step\nObtain persisted legal telemetry for MIX11×Broker and MIX11×Lena B34. If the hrana failure persists across future ticks, repair/isolate sidecar persistence so compute result cannot be lost after execution; only then pre-register a single-factor D9→D12/D12-entry causal treatment.",[],[175,176,2570,2740],"07e20bfd-da07-4421-bea9-2221df546cbb","KAG_HOURLY_CHECKPOINT_20260907_2131.md","c6e921c8-ab1c-11f1-ac7c-1607ee4eb77e",573,["D",1788827550945],"# KAG Hourly Checkpoint — 2026-09-07 21:31 BRT\n\nEvidence state: MATERIAL_UPDATE / NO_CHAMPION_PROMOTION\n\n## Reconciliation\n- Source of truth: `chat-native-maestro@economic-v0.6`.\n- Sidecar consumed only after strategy/rules/intelligence reconciliation.\n- Boundary remains Broker Bea; stress remains Closer Cleo; candidate lineage remains MIX11; regression anchor remains Rancher Rita.\n- Negative-knowledge hard gate enforced. No adjacent mutation of PASTURE_RAMP or previously rejected static targets was opened.\n\n## Consumed sidecar evidence since 20260907_2022\n1. `artifacts/kaggriculture/autolab/20260907T232649Z_OUR-PASTURE-RAMP-BROKER-B33`\n - PASTURE_RAMP vs Broker: 0-8, avg margin -132223.25.\n2. `artifacts/kaggriculture/autolab/20260907T232657Z_OUR-MIX11-BROKER-B33`\n - paired MIX11 control vs Broker: 0-8, avg margin -125902.25.\n - paired delta PASTURE_RAMP - MIX11 = -6321.00.\n3. `artifacts/kaggriculture/autolab/20260907T234416Z_OUR-MIX11-RITA-B33`\n - MIX11 vs Rita: 6-2, avg margin +483.75.\n4. `artifacts/kaggriculture/autolab/20260908T003113Z_PUBLIC-CLEO-BROKER-B34`\n - public black-box crossplay Cleo vs Broker: 8-0, avg margin Cleo +4005.125.\n\n## Falsification / learning\nPASTURE_RAMP is confirmed `HELDOUT_FALSIFIED / MATCHUP_UNSTABLE / DO_NOT_PROMOTE` under the dominant reopening gate. Known paired deltas versus MIX11 against Broker are B31 -20600.625, B32 +2102.5, B33 -6321; all three batches were 0-8. The single B32 directional margin gain is not sufficient to reopen the branch.\n\nFailure chain:\nOBSERVATION: B33 repeated 0-8 and underperformed paired MIX11 by 6321 average margin.\nCAUSE: tested pasture/build ramp is not the missing conversion mechanism and its occasional directional margin gain is unstable across fresh worlds.\nFAILURE CLASS: MATCHUP_UNSTABLE / SMALL_N_DIRECTIONAL_FALSE_LEAD.\nCONTROL: hard reopening gate; no adjacent blind mutations.\nDETECTOR: require explicit new mechanism + measurable regime difference + preregistered discriminant + >=2 favorable paired fresh heldouts/equivalent + no severe regression damage.\nREGRESSION: retain B31/B32/B33 paired evidence as counterproof set.\n\n## Current causal branch\nActive hypothesis remains upstream temporal capital conversion / portfolio transition around D12→D18, but no new treatment is promoted yet. The next discriminant is legal-observation-only telemetry comparing Broker/Lena behavior around their first reliably separable window near D12 H1-H2, not another static target knob.\n\n## OBG / profiler\nNo new profiler promotion this round. Existing V21 legal-landmark profiler remains `STRONG_HELDOUT_CANDIDATE`, not universal proof. A direct fresh heldout script attempt returned an upstream 504 and therefore produced no evidence. Two fresh legal-telemetry jobs were queued on identical seed sets for MIX11×Broker and MIX11×Lena; their first executions encountered transient libSQL/hrana persistence failures and were returned to READY for retry. This is `INFRA_TRANSIENT`, not competitive evidence.\n\n## Queue / next work\n- READY retry: `OUR-MIX11-BROKER-LEGAL-B34`, seeds 53401-53404, legalTelemetry FULL.\n- READY retry: `OUR-MIX11-LENA-LEGAL-B34`, seeds 53401-53404, legalTelemetry FULL.\n- Completed: `PUBLIC-CLEO-BROKER-B34`, seeds 53411-53414.\n\nThe queue is not empty. Retries are informative because they target the D12 legal divergence and do not reopen falsified treatment branches.\n\n## Champion discipline\nNo champion promotion. MIX11 remains preserved candidate/baseline lineage. Broker remains boundary because our lineage remains 0-win against it in current heldouts; Cleo remains stress/public ceiling evidence. TS execution remains mechanical evidence only; official CPython parity remains a future gate.\n\n## First incomplete step\nConsume the two B34 legal-telemetry retries, quantify the Broker↔Lena D12 transition using only licensed public observations, then preregister one single-factor temporal conversion treatment only if that evidence identifies a materially discriminating mechanism.",[],[175,176,2570,2748],"658bbf44-3796-4a91-be47-5ca484199642","KAG_HOURLY_CHECKPOINT_20260907_2022.md","5c156446-ab13-11f1-b420-1607ee4eb77e",568,["D",1788823506271],"# KAG Hourly Checkpoint — 2026-09-07 20:22 America/Fortaleza\n\nStatus: MATERIAL_UPDATE.\n\n## Sidecar consumed / executed\n- `artifacts/kaggriculture/autolab/20260907T232352Z_OUR-MIX11-BROKER-B32` — MIX11 vs Broker: 0-8, avg margin -109760.625.\n- `artifacts/kaggriculture/autolab/20260907T232359Z_OUR-PASTURE-RAMP-BROKER-B32` — PASTURE_RAMP vs Broker: 0-8, avg margin -107658.125.\n- `artifacts/kaggriculture/autolab/20260907T232417Z_OUR-PASTURE-RAMP-RITA-B32` — PASTURE_RAMP vs Rita: 3-5, avg margin +1600.375.\n- `artifacts/kaggriculture/autolab/20260907T232424Z_OUR-MIX11-CLEO-B32` — MIX11 vs Closer Cleo: 0-8, avg margin -101894.\n- `artifacts/kaggriculture/autolab/20260907T232441Z_X-CLEO-BROKER-B32` — Cleo vs Broker: 7-1, avg margin +4474.875.\n\nNo new DONE/FAILED existed after the prior persisted analysis timestamp before B32 was deliberately enqueued; worker was healthy and queue was empty, so B32 was created and executed immediately.\n\n## Material learning / falsification\nPASTURE_RAMP repeated only a small paired margin improvement against Broker on B32: +2102.5 relative to MIX11 on identical seeds, but both remained 0-8. This is diagnostic evidence, not crossing evidence. Against Rita, PASTURE_RAMP was 3-5 despite positive average margin, another direct warning that margin sign must not override W/L.\n\nThe hypothesis `PASTURE_RAMP is the likely crossing mechanism` is weakened. Classification remains MATCHUP/PROVISIONAL only. The broader upstream hypothesis `D12→D18 capital conversion regime` remains active because static capacity, land, herd, portfolio and liquidation knobs have not crossed Broker.\n\nCloser Cleo again crossed Broker strongly (7-1) on fresh B32 worlds while MIX11 remained 0-8 vs Cleo. The public mechanism remains observable and materially above the internal lineage.\n\n## League / champion\n- CHAMPION: unchanged.\n- Candidate lineage: MIX11 retained.\n- BOUNDARY: Broker Bea.\n- STRESS: Closer Cleo.\n- REGRESSION: Rancher Rita; Rosa/Hana/Mateo remain cheap regression suite.\n\n## Enqueued next\nB33 READY with fresh paired seeds 53311-53314:\n- PASTURE_RAMP vs Broker — independent replication.\n- MIX11 vs Broker — paired control.\n- MIX11 vs Rita — regression control.\n\nAll have unique dedupe_key. Queue intentionally remains non-empty.\n\n## First incomplete step\nConsume B33. If PASTURE_RAMP again fails to change W/L, demote it from lead mechanism and implement/test one single-factor D12→D18 capital-conversion timing treatment rather than another static target.",[],[175,176,2570,2756],"c26c1d48-eee5-436f-8092-d01b0b0a9b5b","KAG_HOURLY_CHECKPOINT_20260907_2000.md","e40b87b2-aaf6-11f1-8baa-1607ee4eb77e",553,["D",1788811278954],"# KAG Hourly Checkpoint — 2026-09-07 17:00 America/Fortaleza\n\nStatus: MATERIAL_UPDATE.\n\n## Sidecar consumed since previous checkpoint\n- `artifacts/kaggriculture/autolab/20260907T192856Z_OUR-MIX11-MATEO-B21` — MIX11 vs Mateo: 8-0, avg margin +24694.25.\n- `artifacts/kaggriculture/autolab/20260907T192903Z_OUR-MIX11-HANA-B21` — MIX11 vs Hana: 8-0, avg margin +38521.375.\n- `artifacts/kaggriculture/autolab/20260907T192925Z_OUR-MIX11-ROSA-B21` — MIX11 vs Rosa: 8-0, avg margin +38046.625.\n- `artifacts/kaggriculture/autolab/20260907T193211Z_OUR-ADAPT-V1-BROKER-B22` — ADAPT_V1 vs Broker: 0-8, avg margin -79355.75.\n- `artifacts/kaggriculture/autolab/20260907T193219Z_OUR-DEPLOY10-BROKER-B22` — DEPLOY10 vs Broker: 0-8, avg margin -72885.75.\n- `artifacts/kaggriculture/autolab/20260907T193337Z_OUR-M60F-BROKER-B22` — M60_FERT vs Broker: 0-8, avg margin -76154.875.\n- `artifacts/kaggriculture/autolab/20260907T193345Z_X-LENA-BROKER-B22` — Lena vs Broker: 8-0, avg margin +3583.375.\n- `artifacts/kaggriculture/autolab/20260907T193637Z_OUR-HERD4-BROKER-B22` — HERD4 vs Broker: 0-8, avg margin -77981.125.\n- `artifacts/kaggriculture/autolab/20260907T193645Z_X-CLEO-SILAS-B23` — Cleo vs Silas: 8-0, avg margin +3855.875.\n- `artifacts/kaggriculture/autolab/20260907T193934Z_OUR-LAND6-BROKER-B22` — LAND6 vs Broker: 0-8, avg margin -85553.25.\n- `artifacts/kaggriculture/autolab/20260907T193941Z_OUR-DEPLOY10-BROKER-B24` — DEPLOY10 vs Broker: 0-8, avg margin -84529.375.\n- `artifacts/kaggriculture/autolab/20260907T194013Z_OUR-M60F-BROKER-B24` — M60_FERT vs Broker: 0-8, avg margin -86518.625.\n- `artifacts/kaggriculture/autolab/20260907T194021Z_X-LENA-BROKER-B24` — Lena vs Broker: 8-0, avg margin +2797.75.\n- `artifacts/kaggriculture/autolab/20260907T194326Z_OUR-PHASE3-BROKER-B22` — PHASE3 vs Broker: 0-8, avg margin -90041.75.\n- `artifacts/kaggriculture/autolab/20260907T195728Z_X-CLEO-SILAS-B25` — Cleo vs Silas: 6-2, avg margin +1782.625.\n- `artifacts/kaggriculture/autolab/20260907T195744Z_X-SILAS-LENA-B25` — Silas vs Lena: 8-0, avg margin +1827.75.\n- `artifacts/kaggriculture/autolab/20260907T195852Z_X-LENA-BROKER-B25` — Lena vs Broker: 8-0, avg margin +4707.\n- `artifacts/kaggriculture/autolab/20260907T195859Z_OUR-DEPLOY10-BROKER-B25` — DEPLOY10 vs Broker: 0-8, avg margin -82865.375.\n- `artifacts/kaggriculture/autolab/20260907T195923Z_OUR-HERD4-BROKER-B25` — HERD4 vs Broker: 0-8, avg margin -90864.25.\n- `artifacts/kaggriculture/autolab/20260907T195930Z_OUR-M40F-BROKER-B25` — M40_FERT vs Broker: 0-8, avg margin -100720.125.\n\n## Material learning\nCheap regression is now strong on the sampled B21 worlds: MIX11 went 24-0 across Rosa/Hana/Mateo. This is regression evidence only; champion remains unchanged because Broker remains uncrossed.\n\nBroker is not a near-boundary for the current internal lineage: every tested internal variant remains winless against Broker. Earlier margin improvements for M40_FERT/HERD4/DEPLOY10 were not robust proxies for crossing probability; B25 independently worsened those margins while W/L stayed 0-8. This is a concrete SMALL_N/MARGIN_OVERCOMMITMENT falsification.\n\nThe public ladder is robust across additional heldout worlds: Lena repeatedly beats Broker, Silas beats Lena, and Cleo beats Silas. The useful mechanism therefore remains observable in public policy behavior rather than in current isolated internal knobs.\n\n## Mechanism / causal focus\nThe recurring internal failure transition is upstream: our variants are competitive/positive through roughly D12 and then suffer a large negative gap jump D12→D15 followed by compounding losses. Late liquidation, static melon fraction, pasture-JIT, generic deployment persistence, herd/land static targets and PHASE3 did not repair it.\n\nWinning research mechanism for next branch: `D12→D18 capital conversion regime` — identify how Lena changes inventory creation, crop portfolio, capacity utilization, purchases/sales and cash realization while Broker is crossed. Classification: MATCHUP hypothesis; not UNIVERSAL until perturbation and heldout proof.\n\n## Falsification / infrastructure learning\n- `REFERENCE_CROSSPLAY` is unsupported; corrected to `PUBLIC_CROSSPLAY`.\n- `cleo_closer.py` was an invalid alias; canonical `closer_cleo.py` subsequently executed successfully, so Cleo is no longer blocked.\n- Margin-only ranking of zero-win variants is diagnostic only, not promotion evidence.\n\n## League / champion\n- CHAMPION: unchanged.\n- Candidate lineage: MIX11 retained as regression-capable baseline; no Broker-crossing candidate exists.\n- BOUNDARY: Broker Bea.\n- STRESS: Closer Cleo, executable again with canonical alias.\n- REGRESSION: Rancher Rita; Rosa/Hana/Mateo cheap suite all passed 8-0 each in B21.\n\n## Enqueued next\nB26 READY with deduplicated fresh seeds:\n- OUR-MIX11-RITA-B26 — regression persistence.\n- OUR-MIX11-CLEO-B26 — direct stress ceiling.\n- OUR-DEPLOY10-LENA-B26 — locate internal variant on the public ladder.\n- OUR-M40F-LENA-B26 — test whether previous Broker-margin signal transfers upward.\n- X-CLEO-BROKER-B26 — fresh public ladder/trajectory control.\n\n## First incomplete step\nConsume B26 and use its gradient to decide whether the next code change should implement a D12→D18 state-conditioned capital-conversion/crop-transition treatment. Do not fuse multiple mechanisms before single-factor evidence.",[],[175,176,2570,2764],"7cb5717f-a638-4b74-8739-7231b64105f1","KAG_HOURLY_CHECKPOINT_20260907_1924.md","009a6a4c-ab0b-11f1-94c4-1607ee4eb77e",564,["D",1788819916808],"# KAG Hourly Checkpoint — 2026-09-07 19:24 BRT\n\nEvidence state: LAB_VERIFIED_NOT_CHAMPION_PROOF\n\n## Material results consumed\n- B29 PASTURE_RAMP vs Broker: 0/8, avg margin -95,598.5; paired MIX11 control 0/8, -101,299. Delta margin +5,700.5 but no W/L conversion.\n- B29 PASTURE_RAMP vs Rita: 1/7, -5,697.9. New B31 heldout: 4/4, +125. This is unstable/matchup-sensitive and not promotable.\n- B30 LAND_AFTER6_CLEAN vs Broker: 0/8, -101,961.6 vs paired MIX11 0/8, -96,700.5: harmful delta -5,261.1.\n- B30 LAND_AFTER6_CLEAN vs Rita: 2/6, -1,867.1 vs MIX11 7/1, +3,665.3: severe regression. FALSIFIED as useful single-factor.\n- Earlier isolated Broker probes all remained 0/8: EARLY_MELON40 -113,501; EARLY_HERD3 -105,514; EARLY_HERD4/5 -103,231; EARLY_BUILD4 -106,645; EARLY_HERD_BUILD4 -106,726; EARLY_HERD_BUILD8 -110,976.\n- B31 heldout PASTURE_RAMP vs Broker: 0/8, -96,945.3. Improvement in margin relative to prior controls is not a boundary crossing.\n- B32 PORTFOLIO_D6_244 vs Broker: 0/8, -121,734.1 vs MIX11 control 0/8, -107,712.8: strongly harmful. vs Rita 2/6, -2,506.3 vs MIX11 7/1, +4,412.4: severe regression. FALSIFIED.\n- MIX11 Rita controls remain variable but viable: B30 7/1 +3,665; B31 4/4 +3,570; B32 7/1 +4,412.\n\n## Causal interpretation\nPASTURE_RAMP is the only recent isolated factor with repeated directional Broker margin improvement, but its Rita response ranges from 1/7 to 4/4. Classify MATCHUP/PROVISIONAL, not universal. LAND_AFTER6_CLEAN and PORTFOLIO_D6_244 are falsified. Early melon/herd/build changes do not explain Broker's advantage. The causal search remains centered on conversion efficiency after the early game, but portfolio D6 and simple land threshold are ruled out.\n\n## Infrastructure finding\nA B31 MIX11 Broker control completed compute but blob persistence suffered transient local connection closure; worker correctly returned it to READY. This is INFRA_TRANSIENT, not competitive evidence.\n\n## League\nChampion: unchanged preserved baseline.\nCandidate lineage: MIX11.\nBOUNDARY: Broker Bea.\nSTRESS: Closer Cleo.\nREGRESSION: Rancher Rita; Rosa/Hana/Mateo cheap suite.\n\n## Next queued work\nB31 heldout paired controls/retries and MIX11 stress remain queued; do not promote PASTURE_RAMP unless W/L improves under further heldout while Rita regression remains acceptable. Next causal branch after these results should target a single conversion/timing factor rather than another broad portfolio fusion.\n",[],[175,176,2570,2772],"161f40ba-c60d-4f53-ade0-0c8579b60a91","KAG_HOURLY_CHECKPOINT_20260907_1820.md","cf06667e-aae8-11f1-9069-1607ee4eb77e",550,["D",1788805230754],"# KAG Hourly Checkpoint — 2026-09-07 15:20 America/Fortaleza\n\nStatus: MATERIAL_UPDATE.\n\n## Sidecar consumed\n- artifacts/kaggriculture/autolab/20260907T181852Z_OUR-M60F-BROKER-B18 — M60_FERT vs Broker: 0/8/0, avg margin -72081.\n- artifacts/kaggriculture/autolab/20260907T181859Z_X-LENA-BROKER-B18 — Lena vs Broker: 7/1/0, avg margin +2574.\n- artifacts/kaggriculture/autolab/20260907T181915Z_OUR-MIX11-RITA-REG-B18 — MIX11 vs Rita: 5/3/0, avg margin +1065.25.\n- artifacts/kaggriculture/autolab/20260907T181940Z_OUR-M60F-BROKER-B19 — M60_FERT vs Broker: 0/8/0, avg margin -84792.125.\n- artifacts/kaggriculture/autolab/20260907T181947Z_X-SILAS-LENA-B19 — Silas vs Lena: 7/1/0, avg margin +1720.875.\n- artifacts/kaggriculture/autolab/20260907T182004Z_X-LENA-BROKER-B19 — Lena vs Broker: 7/1/0, avg margin +3629.\n\n## Material learning\nBroker remains a hard boundary for our M60_FERT formulation: two new independent heldout batches remain 0-8. In contrast, Lena crossed Broker 7-1 in two independent new batches, while Silas crossed Lena 7-1 in B19. This strengthens the public ordering Broker < Lena < Silas for these sampled worlds and indicates the useful mechanism gap is not simply raw Broker strength: Lena contains an observable strategy that repeatedly converts against Broker while our lineage does not.\n\nRegression MIX11 vs Rita remains positive but not dominant (5-3); champion unchanged.\n\n## Falsification / failure learning\nTwo old B17 jobs failed because they were queued as unsupported job_type REFERENCE_CROSSPLAY while the core accepts PUBLIC_CROSSPLAY. A new B18 Cleo job then exposed a separate configuration failure: the current reference ZIP does not contain cleo_closer.py. The retry was stopped rather than burned repeatedly. FAILURE CLASS: JOB_SCHEMA_MISMATCH + REFERENCE_ALIAS/MISSING_ASSET. Control: future crossplay uses PUBLIC_CROSSPLAY and validates canonical reference availability before enqueue.\n\n## Mechanism state\nWinner for investigation: extract Lena-vs-Broker behavioral delta (capital deployment, land/hands timing, crop mix, market sell-by-item, demand exposure and D21→D24 conversion) and contrast with M60_FERT-vs-Broker on matched/new worlds. Classification remains MATCHUP/WORLD_REGIME until causal perturbation proves otherwise.\n\n## League/champion\nCHAMPION unchanged. BOUNDARY=Broker Bea. STRESS=Closer Cleo conceptually, but current sidecar asset for Cleo is BLOCKED/MISSING and must not be treated as executable until reference naming/source is reconciled. REGRESSION=Rancher Rita.\n\n## Enqueued\nB20 READY with new seeds 100487,100511,100537,100559: X-SILAS-LENA-B20; X-LENA-BROKER-B20; OUR-M60F-BROKER-B20. Dedupe keys unique. Queue non-empty at checkpoint.\n\n## First incomplete step\nConsume B20, then implement an instrumentation/perturbation core that records opponent-response features and supports safe official-parity probes. Current V5 core can observe action timelines for public crossplay but cannot yet causally perturb our policy while measuring Broker response; this is the next code-level gap before claiming opponent adaptation.",[],[175,176,2570,2780],"18bf3cdc-1262-4a3a-8a3a-51d5ff3dad9c","KAG_HOURLY_CHECKPOINT_20260907_1716.md","617cb962-aaf9-11f1-b273-1607ee4eb77e",557,["D",1788812348400],"# KAG Hourly Checkpoint — 2026-09-07 17:16 BRT\n\nEvidence state: VERIFIED_AUTOLAB_COMPUTE + PROVISIONAL_CAUSAL_HYPOTHESIS\n\n## Material results consumed\n- `X-CLEO-BROKER-B26`: Closer Cleo 5–3 Broker Bea, avgMarginA +1,506.5, 8 episodes. artifact_prefix `artifacts/kaggriculture/autolab/20260907T200239Z_X-CLEO-BROKER-B26`.\n- `OUR-DEPLOY10-LENA-B26`: DEPLOY10 0–8 Lena, avgMargin -95,349.375. artifact_prefix `artifacts/kaggriculture/autolab/20260907T200312Z_OUR-DEPLOY10-LENA-B26`.\n- `OUR-MIX11-CLEO-B26`: MIX11 0–8 Closer Cleo, avgMargin -101,686.625. artifact_prefix `artifacts/kaggriculture/autolab/20260907T200319Z_OUR-MIX11-CLEO-B26`.\n- `OUR-MIX11-RITA-B26`: MIX11 5–3 Rancher Rita, avgMargin +2,764. artifact_prefix `artifacts/kaggriculture/autolab/20260907T200346Z_OUR-MIX11-RITA-B26`.\n- `OUR-M40F-LENA-B26`: compute reached persistence but blob write failed transiently; queue returned READY for retry. Do not interpret as competitive FAIL.\n\n## Reconciliation / falsification\n- MIX11 remains regression-capable against Rita on this heldout batch, but is nowhere near STRESS: 0–8 Cleo.\n- DEPLOY10 does not transfer upward to Lena: 0–8. This falsifies any promotion of simple capital-deployment timing as sufficient mechanism.\n- Cleo > Broker remains observable on fresh B26 seeds (5–3), but weaker than earlier public-crossplay batches; therefore mechanism extraction must use trajectory/dGap windows rather than winner label alone.\n- Boundary remains Broker Bea. Stress remains Closer Cleo. Champion unchanged; MIX11 remains candidate/baseline lineage only.\n\n## Current mechanism hypothesis\n`D12→D18 capital conversion + crop transition` remains MATCHUP_HYPOTHESIS, not universal. Existing variants show that deployment-only, fertilizer-only and static policy changes are insufficient. Next causal treatment must be single-factor and activated around the first dGap acceleration window, with pre-state-identical/fixed-world comparisons before fusion.\n\n## Sidecar actions\nQueued fresh non-duplicated B27 jobs:\n1. `OUR-MIX11-BROKER-B27` — direct boundary measurement, seeds 101003/101009/101021/101027.\n2. `OUR-MIX11-RITA-B27` — regression guard on identical seeds.\n3. `X-CLEO-BROKER-B27` — public crossplay refresh, seeds 101033/101039/101051/101063.\nPending automatic retry: `OUR-M40F-LENA-B26`.\n\nQueue intentionally remains non-empty. No champion promotion.\n\n## First incomplete step\nConsume B27 + M40F retry, then use the Broker/Cleo differential to specify the first isolated D12–D18 treatment. If no current sidecar variant expresses that treatment, implement it in the source-of-truth branch only after preserving baseline and then enqueue fixed-world/heldout/regression tests.",[],[175,176,2570,2788],"6cd8c841-f3b0-48eb-92d6-dcb3b8fa5ffa","KAG_HOURLY_CHECKPOINT_20260907_1534.md","ef66f49a-aad1-11f1-b64f-1607ee4eb77e",541,["D",1788795767110],"# KAG Hourly Checkpoint — 2026-09-07 15:34Z\n\nStatus: COMPLETE_FOR_THIS_ROUND\nEvidence class: AUTOLAB_COMPUTE_RESULT_NOT_CHAMPION_PROOF\n\n## Material changes\n- Champion/baseline unchanged. Boundary remains Broker Bea; stress remains Closer Cleo.\n- B9 public crossplay: Lena→Broker 7-1 (+2,857.375), Silas→Lena 5-3 (+1,151.875), Cleo→Silas 5-3 (+1,698.875). B10 independently restored Cleo→Silas to 7-1 (+3,897), so the upper ladder remains ordered but is not deterministic at small N.\n- The previous `inventory_to_cash_conversion / sell_timing` candidate is downgraded from leading general mechanism to LOCAL/MATCHUP: Lena sells materially more than Broker at D21/D24, but Silas beats Lena with essentially identical D21/D24 liquidation and Cleo beats Silas while liquidating much less before closing.\n- Added valid treatment detectors for liquidation ablations. On B11, MIX11_LIQ21 and MIX11_LIQ24 each fired in 8/8 episodes with 290 and 284 eligible inventory units observed. They remained 0-8 vs Broker and changed paired mean margin by only -1.0 and -2.25 relative to MIX11 control. Therefore `force extra liquidation at D21/D24` is causally near-null in this student formulation; B10's identical result was correctly treated as a possible no-op until the detector existed.\n- Melon frontier B12 with fresh seeds: M40_FERT 0-8 (-77,081.875), M50_FERT 0-8 (-77,376.25), M60_FERT 0-8 (-73,253.25) vs Broker. M50_FERT retained regression strength vs Rita 8-0 (+17,731). Static melon share improves diagnostics/lower-tier strength but does not cross Broker.\n- Cleo→Silas B9+B10 black-box telemetry points upstream: Cleo averages about +0.94 MELON at D3/D6; around D15 it carries +5 crops, decomposed about +3.94 MELON and +1.06 STRAWBERRY, with a multi-thousand cash advantage. Since static melon share fails, the next mechanism is crop/inventory creation timing or capacity transition, not a fixed share.\n\n## Falsifications / controls\n1. `sell more / sell earlier at D21 or D24` is not a universal ladder mechanism and is near-null when applied as a generic value-based liquidation pulse to MIX11.\n2. The treatment validator was strengthened: exact outcome equality is not accepted as a causal FAIL unless the intervention detector proves the treatment fired.\n3. M40/M50/M60 static melon shares all fail the primary W/L boundary gate on B12; M60 wins on margin only, which is diagnostic and not promotion evidence.\n4. Lena→Broker's sell-volume pattern is real but does not transfer to Silas→Lena or Cleo→Silas; it is LOCAL/MATCHUP until further evidence.\n\n## Winning mechanism candidate\nClassification: MATCHUP / WORLD-REGIME / UNVERIFIED_FOR_STUDENT\nName: upstream_inventory_creation / crop_portfolio_transition / capacity_timing.\n\nSupporting evidence: the stress ladder's strongest agent creates an early melon lead, then a larger mixed-crop lead by D15, before/while maintaining a cash advantage. Counterproof: static melon share alone does not cross Broker. Existing source-of-truth lesson `KAG-LESSON-PASTURE-JIT-20260907` independently reports PARTIAL evidence that just-in-time pasture capacity frees early crop tiles. Therefore the discriminating next experiment is pasture-capacity timing isolated from hands/labor, followed by opponent-aware crop transition if it survives.\n\n## Regression / league\n- Champion: unchanged; no promotion.\n- Boundary: Broker Bea.\n- Stress: Closer Cleo.\n- Regression: Rancher Rita. MIX11 B10 = 6-2 (+2,640.125); M50_FERT B12 = 8-0 (+17,731).\n- Public ladder remains provisionally Broker < Lena < Silas < Cleo, with small-N variability now explicit.\n\n## Sidecar artifacts consumed\n- artifacts/kaggriculture/autolab/20260907T144119Z_X-LENA-BROKER-SELLITEM-B9\n- artifacts/kaggriculture/autolab/20260907T144127Z_X-SILAS-LENA-SELLITEM-B9\n- artifacts/kaggriculture/autolab/20260907T145915Z_X-CLEO-SILAS-SELLITEM-B9\n- artifacts/kaggriculture/autolab/20260907T152733Z_OUR-MIX11-LIQ21-BROKER-B10\n- artifacts/kaggriculture/autolab/20260907T152740Z_OUR-MIX11-LIQ24-BROKER-B10\n- artifacts/kaggriculture/autolab/20260907T152756Z_OUR-MIX11-BROKER-B10\n- artifacts/kaggriculture/autolab/20260907T152803Z_OUR-MIX11-RITA-B10\n- artifacts/kaggriculture/autolab/20260907T152924Z_OUR-MIX11-LIQ21-BROKER-B11\n- artifacts/kaggriculture/autolab/20260907T152931Z_OUR-MIX11-LIQ24-BROKER-B11\n- artifacts/kaggriculture/autolab/20260907T152956Z_OUR-MIX11-BROKER-B11\n- artifacts/kaggriculture/autolab/20260907T153003Z_X-CLEO-SILAS-B10\n- artifacts/kaggriculture/autolab/20260907T153321Z_OUR-M40-FERT-BROKER-B12\n- artifacts/kaggriculture/autolab/20260907T153328Z_OUR-M50-FERT-BROKER-B12\n- artifacts/kaggriculture/autolab/20260907T153345Z_OUR-M60-FERT-BROKER-B12\n- artifacts/kaggriculture/autolab/20260907T153353Z_OUR-M50-FERT-RITA-B12\n\n## Sidecar changes / queued work\n- Core advanced through `KAG_AUTOLAB_CORE_V3_LIQUIDATION_ABLATION` and `KAG_AUTOLAB_CORE_V4_MELON_FRONTIER`.\n- Added liquidation treatment detector (`liqChecks`, `liqApplied`, `liqOpportunityQty`) to prevent FALSE_PRUNE from no-op interventions.\n- Added fresh M40/M50/M60 melon-frontier support and ran B12.\n- Queue was found empty at round start (49 DONE, 0 READY/RUNNING), treated as continuity failure; it was refilled repeatedly during the LOOP.\n- At round close, four fresh B14 jobs remain READY with unique seeds 100057/100069/100103/100109: M60_FERT×Broker, Lena×Broker, Silas×Lena, Cleo×Silas.\n\n## Blocker / first incomplete step\nA targeted code edit to create `M60_JITPAST` (pasture target only, no hands change) was blocked once by the tool safety layer during this execution. This is a transient tooling blocker, not an experimental FAIL, and the automation remains active. First incomplete step: retry that isolated implementation, validate that only pasture target changed, then run paired Broker heldout vs M60 control; only if positive proceed to Rita regression and then stress. If JIT pasture is falsified, move to opponent-aware crop-transition timing rather than another static crop share.\n\n## Continuation before close\n- B14 completed after the initial checkpoint write: M60_FERT×Broker 0-8 (-92,829.75), Lena×Broker 7-1 (+3,662.125), Silas×Lena 7-1 (+1,993.125), Cleo×Silas 8-0 (+3,090.375). Prefixes: `artifacts/kaggriculture/autolab/20260907T153613Z_OUR-M60-FERT-BROKER-B14`, `.../20260907T153620Z_X-LENA-BROKER-B14`, `.../20260907T153800Z_X-SILAS-LENA-B14`, `.../20260907T153808Z_X-CLEO-SILAS-B14`.\n- The transient edit blocker was resolved. Core advanced to `KAG_AUTOLAB_CORE_V5_PASTURE_JIT_ISOLATE`; `M60_FERT_JIT` changes only pasture build target to `clamp(owned livestock + 2, 4, 16)` while preserving M60 crop/fertilizer policy and not changing hands.\n- B15 paired heldout validated the treatment: `M60_FERT_JIT` invoked the build setter 240 times across 8 episodes (`jitTargetSum=2469`) and finished 0-8 vs Broker at -86,754.25. Identical-seed M60_FERT control finished 0-8 at -71,357.625. JIT therefore worsened mean margin by -15,396.625 without generating a win.\n- `KAG-LESSON-PASTURE-JIT-HELDOUT-FAIL-20260907` was persisted in `kag_strategy_lessons` as `HELDOUT_FALSIFIED`; the earlier fixed-world PARTIAL success is not promoted or erased, but its scope is now explicitly limited.\n- Winning mechanism candidate narrows to `opponent/world-conditioned crop portfolio transition / upstream inventory creation`; static melon share, generic liquidation pulse, deployment pulse and unconditional pasture-JIT are now separately falsified at Broker.\n- Fresh B16 queue left READY (not drained) with seeds 100183/100189/100193/100207: M60_FERT×Broker plus Lena×Broker, Silas×Lena and Cleo×Silas. These preserve sidecar continuity while the next round designs the conditional crop-transition student.\n\n## Revised first incomplete step\nDerive the smallest observable opponent/world-conditioned crop-transition rule from existing teacher-Q / public black-box evidence, with hysteresis and an explicit treatment detector; test it paired against Broker before regression/stress. Do not try another static melon share or universal capacity rule.\n",[],[175,176,2570,2796],"41b3b5fe-5056-4fe6-8472-1f3c94aab7cf","KAG_HOURLY_CHECKPOINT_20260907_1436.md","8e65f37e-aac9-11f1-b0b1-1607ee4eb77e",533,["D",1788791897059],"# KAG Hourly Checkpoint — 2026-09-07 14:36Z\n\nStatus: COMPLETE_FOR_THIS_ROUND\nEvidence class: AUTOLAB_COMPUTE_RESULT_NOT_CHAMPION_PROOF\n\n## Material changes\n- Boundary remains Broker Bea; stress remains Closer Cleo; champion/baseline unchanged.\n- Persistent deployment family is falsified at Broker: DP_H 0-8 (-80,739), DP_W 0-8 (-81,825.25), DP_B 0-8 (-77,288.75), DP_COND 0-8 (-86,134.5).\n- Bounded D9→D12 pulse also failed: DP_H_PULSE 0-8 (-93,275.75), DP_B_PULSE 0-8 (-92,899.375), DP_COND_PULSE 0-8 (-90,526.625). Duration alone does not rescue deployment.\n- Latest public crossplay: Lena→Broker 8-0 (+5,762 B3; +4,453.5 action B6; +2,827.25 sell-item B7), Silas→Lena 8-0 B3 then 6-2 B6 (+2,563.5), Cleo→Silas 8-0 B3 (+3,081). This preserves the provisional ordering but reduces confidence that Silas→Lena is deterministic.\n- Largest failure window for our deployment family moved later than previously hypothesized: typically D18→D27, not D12→D15.\n\n## Winning mechanism candidate\nClassification: MATCHUP / WORLD-REGIME / SHAPING-ONLY / UNVERIFIED_FOR_STUDENT\nName: inventory_to_cash_conversion / sell_timing / closing routing.\n\nPublic black-box dossier shows Lena and Broker with essentially identical land, crop counts, pasture, animals and planted crop mix through D18-D27, while Lena creates a cash advantage. At hour 0 checkpoints, Lena sells more inventory with identical planted mix: D21 STRAWBERRY 21 vs 14 (MELON 2 vs 2); D24 STRAWBERRY 44 vs 29, MELON 8 vs 6, FERTILIZER 7 vs 7. This is evidence for conversion/harvest/inventory timing, not for the previously falsified static strawberry 65/35 planting rule.\n\n## Falsifications learned\n1. Persistent hands/build/both/conditional deployment after D9 does not cross Broker.\n2. Simply bounding those interventions to D9→D12 does not cross Broker and worsened mean margin on this heldout seed set.\n3. Previous D12→D15 collapse hypothesis was not stable across heldout: the strongest gap deterioration in B4/B5 often occurs D18→D27.\n4. SAME FEATURE COUNTS != SAME ECONOMIC TRAJECTORY: public agents can expose equal land/crops/animals while diverging materially in cash through inventory conversion/routing.\n\n## Regression / league\n- Champion: unchanged; no promotion.\n- Boundary: Broker Bea.\n- Stress: Closer Cleo.\n- Regression: Rancher Rita remains a cheap regression target, but wins against Rita are insufficient promotion evidence.\n- Provisional public ladder remains Broker < Lena < Silas < Cleo, with Silas>Lena now 6-2 on one independent B6 rather than universal 8-0.\n\n## Sidecar artifacts consumed\n- artifacts/kaggriculture/autolab/20260907T134521Z_OUR-DP-B-BROKER-B4\n- artifacts/kaggriculture/autolab/20260907T134528Z_OUR-DP-COND-BROKER-B4\n- artifacts/kaggriculture/autolab/20260907T140804Z_X-LENA-BROKER-B3\n- artifacts/kaggriculture/autolab/20260907T140811Z_X-SILAS-LENA-B3\n- artifacts/kaggriculture/autolab/20260907T142424Z_X-CLEO-SILAS-B3\n- artifacts/kaggriculture/autolab/20260907T143059Z_OUR-DP-H-PULSE-BROKER-B5\n- artifacts/kaggriculture/autolab/20260907T143106Z_OUR-DP-B-PULSE-BROKER-B5\n- artifacts/kaggriculture/autolab/20260907T143124Z_OUR-DP-COND-PULSE-BROKER-B5\n- artifacts/kaggriculture/autolab/20260907T143300Z_X-LENA-BROKER-ACTION-B6\n- artifacts/kaggriculture/autolab/20260907T143336Z_X-SILAS-LENA-ACTION-B6\n- artifacts/kaggriculture/autolab/20260907T143459Z_X-LENA-BROKER-SELLITEM-B7\n\n## Sidecar changes / queued work\nCompute core instrumented with bounded DP_H_PULSE / DP_B_PULSE / DP_COND_PULSE and public black-box action diagnostics including market volumes and sell-by-item. No internal TRACE/replay plan is consumed.\nQueued with fresh dedupe keys: X-LENA-BROKER-SELLITEM-B8 and X-BROKER-RITA-SELLITEM-B8. X-CLEO-SILAS-ACTION-B6 remains retryable after a transient blob-persistence failure; failed persistence is not counted as competitive evidence.\n\n## First incomplete step\nUse independent B8/counterproof to decide whether the sell-volume signal generalizes, then isolate harvest/inventory availability from liquidation timing before modifying the student. Do not fuse strawberry production/share with closing logic; test inventory creation and sell timing separately. Any student candidate must then face Broker heldout, seats, natural RNG, Rita regression, runtime <1s, provenance and counterproof gates.\n\n## Continuation before round close\n- Independent B8: Lena vs Broker 8-0, avg margin +5,197.5; artifact `artifacts/kaggriculture/autolab/20260907T143727Z_X-LENA-BROKER-SELLITEM-B8`.\n- Counterproof B8: Broker vs Rita 8-0, avg margin +111,983.125; artifact `artifacts/kaggriculture/autolab/20260907T143737Z_X-BROKER-RITA-SELLITEM-B8`.\n- Lena→Broker repeated the same D21/D24 sell pattern on fresh seeds. Broker→Rita did not express a universal 'sell more strawberry' rule; it used a different product/liquidation regime. Therefore the candidate remains MATCHUP/WORLD-REGIME, not UNIVERSAL.\n- New B9 diagnostics queued with fresh seeds 99809/99817/99833/99859: Lena×Broker, Silas×Lena, Cleo×Silas, all with sell-item instrumentation. The transient Cleo×Silas B6 persistence retry remains separate.\n",[],[175,176,2570,2804],"db9d2fb6-fb47-4362-99f6-396e7863d027","KAG_HOURLY_CHECKPOINT_20260907_1420.md","9bcf57c8-aae0-11f1-935d-1607ee4eb77e",547,["D",1788801708849],"# KAG Hourly Checkpoint — 2026-09-07 14:20 America/Fortaleza\n\nStatus: MATERIAL_UPDATE; ARTIFACT_PARTIAL(sidecar SQL/worker invocation blocked by safety layer; blob evidence consumed directly).\n\n## Consumed sidecar evidence\n- artifacts/kaggriculture/autolab/20260907T163052Z_OUR-M60-FERT-JIT-BROKER-B17.md — M60_FERT_JIT vs Broker: 0/8/0, avg margin -101003.6; largest gap deterioration D21→D24 -21759.125.\n- artifacts/kaggriculture/autolab/20260907T163059Z_OUR-M60-FERT-BROKER-B17.md — control M60_FERT vs Broker: 0/8/0, avg margin -93380.1; largest gap deterioration D21→D24 -17391.375.\n- artifacts/kaggriculture/autolab/20260907T164608Z_X-SILAS-LENA-B16.md — Silas vs Lena: 6/2/0, avg margin +1178.8; largest gap-change D21→D24 +1173.125.\n\n## Falsification / learning\nM60_FERT_JIT independently fails again and is worse than matched B17 control by 7623.5 avg margin with identical W/L. This independently confirms the V5 pasture-JIT formulation should be QUARANTINED, not promoted or fused. The dominant deterioration window aligns at D21→D24 for both JIT and control, so pasture headroom alone is not the root cause of the boundary failure.\n\nThe public Silas→Lena edge remains positive but non-deterministic at 6–2. This supports preserving the provisional league order while preventing overclaim of deterministic dominance.\n\n## Mechanism state\nLeader: capital deployment + land/crop compounding conditioned on observable state + portfolio/demand routing, with D21→D24 conversion/closing as a discriminating diagnostic window. Classification: MATCHUP + WORLD_REGIME; UNVERIFIED_FOR_STUDENT.\n\n## League/champion\nCHAMPION unchanged. BOUNDARY=Broker Bea. STRESS=Closer Cleo. REGRESSION=Rancher Rita / strongest already beaten. No promotion.\n\n## Blocker\nDirect sidecar SQLite reads and manual worker execution were blocked by the platform safety layer in this run. This is treated as tooling blocker, not experimental failure. Existing scheduled sidecar remains the compute path. Queue state could not be truthfully verified/replenished via SQL in this execution, so queue status is UNVERIFIED rather than claimed healthy.\n\n## First incomplete step\nConsume the next sidecar artifacts after B17 and test separated state-conditioned land/crop compounding versus portfolio/demand routing at Broker, using heldout seeds and mirrored seats. Preserve JIT as quarantined negative knowledge.",[],[175,176,2570,2812],"b9afd174-bd15-41d0-98fb-4d6b132ae560","KAG_HOURLY_CHECKPOINT_20260907_1332.md","1a1b3f9a-aac1-11f1-9f8f-1607ee4eb77e",527,["D",1788788176847],"# KAG Hourly Checkpoint — 2026-09-07 13:32 UTC\n\n## Estado\n- Evidence state: `HOURLY_CODE_AND_BOUNDARY_ABLATION_PARTIAL_NO_PROMOTION`\n- Champion: **inalterado**.\n- Boundary: **Broker Bea**.\n- Stress: **Closer Cleo**.\n- Candidato atual: `bounded_state_conditioned_deployment_pulse` — **shaping-only / UNVERIFIED**.\n- Nenhuma promoção autorizada.\n\n## Delta consumido do sidecar\n### DONE — evidência válida\n1. `OUR-DP-H-BROKER-B4`\n - artifact_prefix: `artifacts/kaggriculture/autolab/20260907T133217Z_OUR-DP-H-BROKER-B4`\n - 8 episódios, assentos espelhados, seeds 99317/99331/99347/99371.\n - W/L/T = **0/8/0**.\n - margem média = **-80,739**.\n - erros de execução = 0.\n - trajetória média do gap: D9 +1,654.75; D12 +3,291.625; D15 -8,817.75; D24 -45,243; D29 -67,834.625.\n - maior deterioração observada: D21→D24 = -16,816.75.\n\n2. `OUR-DP-W-BROKER-B4`\n - artifact_prefix: `artifacts/kaggriculture/autolab/20260907T133224Z_OUR-DP-W-BROKER-B4`\n - 8 episódios, mesmos seeds/assentos do lote B4.\n - W/L/T = **0/8/0**.\n - margem média = **-81,825.25**.\n - erros de execução = 0.\n - trajetória média do gap: D9 +1,654.75; D12 +3,357.25; D15 -9,716.125; D21 -30,937.625; D29 -68,184.75.\n - maior deterioração observada: D18→D21 = -15,461.875.\n\n### FAILED — evidência de infraestrutura, não de força competitiva\n- `X-LENA-BROKER-B3`: falhou por referência histórica `lena_optimist.py` inexistente no dataset atual.\n- `X-SILAS-LENA-B3`: falhou por referência histórica `silas_strategist.py` inexistente.\n- `X-CLEO-SILAS-B3`: primeira tentativa falhou pelo mesmo alias de Silas.\n- Esses FAILs **não contam como derrota competitiva**.\n\n## Correção de causa raiz no compute sidecar\nArquivo alterado: `cleitoncosta/kaggriculture-autolab@main/lib/kag_autolab_core.ts`.\n\nMudanças materiais:\n1. versão do core → `KAG_AUTOLAB_CORE_V2_DEPLOY_PULSE`;\n2. aliases canônicos: `lena_optimist.py→ledger_lena.py`, `silas_strategist.py→slotter_silas.py`;\n3. canonicalização aplicada em crossplay e OUR_VS_REFERENCE;\n4. `set_ops(h,b)` para manipular hands/build/land timing no professor-port;\n5. quatro ablações adicionadas: `DP_H`, `DP_W`, `DP_B`, `DP_COND`;\n6. DP parte da linhagem M60_FERT/Melon apenas como base experimental; isso **não** promove M60_FERT nem fertilizer a regra universal;\n7. worker importou/rodou o core V2 com PASS antes das novas medições.\n\n## Falsificações desta rodada\n1. **Static-H pós-D9 foi rejeitado como solução de boundary**: 0–8 vs Broker.\n2. **Static-W pós-D9 foi rejeitado como solução de boundary**: 0–8 vs Broker.\n3. `teacher fixed-world positive delta != heldout W/L improvement`: ganhos locais de action-value do professor não transferiram automaticamente para vitória natural contra Broker.\n4. `persistent intervention != bounded causal pulse`: a formulação atual mantém H/W depois do gatilho, enquanto os forks positivos do professor são locais. O sinal de reversão D12→D15 sugere que a duração da intervenção é agora uma variável causal crítica.\n5. Mesmo lote de seeds não é tratado como mundo idêntico entre políticas; B4 serve como heldout comparável, não como prova contrafactual de pre-state idêntico.\n\n## Mecanismo vencedor / hipótese líder\nO antigo `conditional_deployment_pulse` foi refinado para **`bounded_state_conditioned_deployment_pulse`**.\n\nMecanismo observável provisório:\n- existe subdeployment claro antes de D9/D12 em land/crops/animals contra Broker;\n- H e W permanentes não resolvem;\n- a política precisa alterar deployment **apenas na janela causal necessária**, depois devolver controle ao portfólio/mercado;\n- a ruptura recorrente D12→D15 desloca o próximo NEURÔNIO para `crop portfolio + demand exposure + market slots + sell timing`, não simplesmente para mais capacidade.\n\nClassificação: **MATCHUP + WORLD_REGIME + shaping-only**. Ainda não é student nem candidate de champion.\n\n## Fila após validação\nREADY e deduplicados:\n- `OUR-DP-B-BROKER-B4` — `our_dp_b_broker_b4_99317_99371`.\n- `OUR-DP-COND-BROKER-B4` — `our_dp_cond_broker_b4_99317_99371`.\n- `X-LENA-BROKER-B3` — reaberto após correção de alias.\n- `X-SILAS-LENA-B3` — reaberto após correção de alias.\n- `X-CLEO-SILAS-B3` — aguardando retry com alias corrigido.\n\nUma tentativa manual adicional de drenar imediatamente o worker foi bloqueada pela camada externa de execução; os jobs permaneceram READY e serão consumidos pelo worker normal. Isso é blocker transitório de ferramenta, não falha experimental.\n\n## Champion discipline\n- Champion/baseline preservado.\n- Broker continua boundary.\n- Cleo continua stress.\n- DP_H/DP_W não passam gate primário W/L.\n- Nenhum resultado de margem isolado foi usado para promoção.\n- TS-port continua evidência mecânica, não exact CPython parity.\n\n## Primeira etapa incompleta\n1. Consumir `DP_B` e `DP_COND` no mesmo B4.\n2. Se ambos continuarem 0–8, **não** fazer mais uma fusão estática: implementar ablação temporal explícita `PULSE_D9_12`/pulso único e medir D12→D15.\n3. Separar então `crop portfolio`, `market exposure`, `sell timing` e `closing` antes de fusão.\n4. Revalidar aliases pelos três crossplays B3 reparados.\n5. Só após sinal W/L heldout considerar versão student <1 s, seguida de regression/runtime/provenance/counterproof.\n",[],[175,176,2570,2820],"68145ee8-61ab-4241-bc2f-70d87f446286","KAG_HOURLY_CHECKPOINT_20260907_1317.md","61ef77ca-aad8-11f1-ba73-1607ee4eb77e",545,["D",1788798175764],"# KAG Hourly Checkpoint — 2026-09-07 13:17 America/Fortaleza\n\n## Estado\n- Champion: PRESERVADO / sem promoção.\n- BOUNDARY: Broker Bea.\n- STRESS: Closer Cleo.\n- Core sidecar observado: `KAG_AUTOLAB_CORE_V5_PASTURE_JIT_ISOLATE`.\n\n## Resultados materiais consumidos desde o checkpoint anterior\n- `OUR-M60-FERT-JIT-BROKER-B15`: 0-8, avgMargin -86754.25, errors=0. artifact_prefix `artifacts/kaggriculture/autolab/20260907T154142Z_OUR-M60-FERT-JIT-BROKER-B15`.\n- `OUR-M60-FERT-BROKER-B15`: 0-8, avgMargin -71357.625, errors=0. artifact_prefix `artifacts/kaggriculture/autolab/20260907T154150Z_OUR-M60-FERT-BROKER-B15`.\n\n## Falsificação\n`M60_FERT_JIT` piorou a margem média versus o controle M60_FERT no mesmo seed set em ~15396.6 e não alterou W/L (ambos 0-8). Logo, pasture-JIT na formulação V5 está FALSIFICADO para Broker; não promover nem fundir.\n\n## Mecanismo causal atualizado\nO JIT não corrige a causa dominante. Em M60_FERT_JIT, o agente chega a D12 ainda com apenas 1 land e ~18-19 crops enquanto Broker já está em 3 land e ~39 crops; a ruptura de caixa/produção aparece depois. A hipótese líder volta para `capital deployment + land/crop compounding + portfolio/demand routing`, com inventory-to-cash como downstream, não causa isolada.\n\n## Infra / failure learning\n`X-SILAS-LENA-B16` falhou após compute ao persistir blob (connection closed), foi corretamente devolvido a READY com `run_after=2026-09-07 16:28:01`. Classe: ARTIFACT_PERSIST_TRANSIENT, não resultado competitivo. Detector: last_error em queue + ausência de DONE correspondente. Não contar como W/L.\n\n## Fila\nFila observada antes do refill: DONE=70, READY=1 (retry transitório B16). Foram enfileirados com dedupe_key e seeds inéditos 100213/100237/100267/100271:\n- OUR-M60-FERT-JIT-BROKER-B17 (priority 29)\n- OUR-M60-FERT-BROKER-B17 (priority 30)\n- X-LENA-BROKER-B17 (priority 32)\n- X-CLEO-SILAS-B17 (priority 33)\n\n## Próxima etapa incompleta\nConsumir retry `X-SILAS-LENA-B16` e B17. Se B17 repetir a degradação JIT, quarentenar `M60_FERT_JIT` definitivamente e implementar a próxima ablação separando early land/crop compounding condicionado ao estado (não regra universal) de portfolio/demand routing, sempre contra Broker e com regressão Rita antes de qualquer promoção.",[],[175,176,2570,2828],"b09ca78f-68c7-45be-aeff-ce42ccaab3e7","KAG_HOURLY_CHECKPOINT_20260907_1222.md","29b39fb0-aab7-11f1-964a-1607ee4eb77e",524,["D",1788783908046],"# Kaggriculture Hourly Checkpoint — 2026-09-07 12:22 UTC\n\nEvidence: HOURLY_HELDOUT_UPDATE_NO_CHAMPION_PROMOTION\n\n## Material update\nConsumed 8 B2 sidecar jobs. All completed DONE with no execution error reported in the run summaries.\n\n### Our variants vs boundary/regression\n- MIX11 vs Broker Bea: 0-8, avg margin -75,798.75.\n- MIX11_FERT vs Broker Bea: 0-8, avg margin -74,754.375.\n- M60_FERT vs Broker Bea: 0-8, avg margin -75,595.5.\n- M60_FERT vs Rancher Rita: 8-0, avg margin +39,086.625.\n\nInterpretation: the upward-transfer failure is now heldout-repeated. Fertilizer/M60 improves or preserves performance against Rita but does not cross Broker. This is not champion evidence.\n\n### Public crossplay B2\n- Closer Cleo vs Broker Bea: 8-0, avg margin +6,589.0.\n- Silas Strategist vs Broker Bea: 8-0, avg margin +4,564.25.\n- Silas Strategist vs Lena Optimist: 8-0, avg margin +870.625.\n- Lena Optimist vs Broker Bea: 8-0, avg margin +3,919.75.\n\nThe observed ladder is therefore Broker < Lena < Silas < Cleo on this B2 set. Broker remains boundary because our variants still lose 0-8 to it; promotion of boundary is not warranted until our student crosses Broker.\n\n## Falsifications / mechanism\n- FALSIFIED: M60_FERT as a transferable boundary solution.\n- FALSIFIED: fertilizer sell as sufficient boundary mechanism.\n- CONFIRMED AS SHAPING-ONLY: conditional_deployment_pulse remains the best causal candidate, but current sidecar cannot execute that student variant; it only supports MIX11, MIX11_FERT, M60_FERT. No promotion without an implemented student and heldout.\n- Regression separation strengthened: M60_FERT 8-0 Rita while 0-8 Broker indicates Rita is a weak regression gate relative to the current boundary.\n\n## Champion discipline\nChampion/baseline unchanged. Candidate mechanism: conditional_deployment_pulse, MATCHUP + WORLD_REGIME, shaping-only. Boundary=Broker Bea. Stress=Closer Cleo. Regression=Rancher Rita.\n\n## Sidecar artifacts consumed\n- artifacts/kaggriculture/autolab/20260907T120146Z_OUR-MIX11-BROKER-B2\n- artifacts/kaggriculture/autolab/20260907T120158Z_OUR-MIX11F-BROKER-B2\n- artifacts/kaggriculture/autolab/20260907T120413Z_OUR-M60F-BROKER-B2\n- artifacts/kaggriculture/autolab/20260907T120426Z_OUR-M60F-RITA-REG-B2\n- artifacts/kaggriculture/autolab/20260907T120628Z_X-CLEO-BROKER-B2\n- artifacts/kaggriculture/autolab/20260907T120639Z_X-SILAS-BROKER-B2\n- artifacts/kaggriculture/autolab/20260907T121318Z_X-SILAS-LENA-B2\n- artifacts/kaggriculture/autolab/20260907T121329Z_X-LENA-BROKER-B2\n\n## Queue replenishment\nEnqueued four deduplicated B3 jobs using unseen seeds 99173/99181/99191/99223: MIX11 vs Rita regression, Lena vs Broker, Silas vs Lena, Cleo vs Silas. These are information-gathering jobs, not champion gates.\n\n## First incomplete step\nThe highest-value missing capability is sidecar execution of the conditional_deployment_pulse student/ablation family. Until that variant is implemented in the compute core, further crossplay mainly refines ladder confidence rather than improving our agent. Next material code change should add the smallest observable <1s deployment-pulse variant with explicit ablation switches, then test fixed-world/forks offline and heldout Broker + regression before any champion consideration.\n",[],[175,176,2570,2836],"7d9c28c8-e9ed-41cf-b890-f3384eb971ea","KAG_HOURLY_CHECKPOINT_20260907_1021.md","cfeacd16-aaa5-11f1-b40b-1607ee4eb77e",521,["D",1788776455965],"# Kaggriculture Hourly Checkpoint — 2026-09-07 10:21 UTC\n\n## Material changes\n- **Champion:** unchanged; no hard promotion gate passed.\n- **Boundary:** Broker Bea remains boundary.\n- **Stress:** Closer Cleo remains stress.\n- **Regression:** Rancher Rita remains strongest defeated regression in the current slice.\n\n## Sidecar results consumed this round\nAll are `AUTOLAB_COMPUTE_RESULT_NOT_CHAMPION_PROOF` and completed without execution errors.\n1. `artifacts/kaggriculture/autolab/20260907T102009Z_OUR-M60F-LENA-B1` — M60_FERT vs Lena: **0-4**, avg margin **-81726**.\n2. `artifacts/kaggriculture/autolab/20260907T102013Z_X-BROKER-RITA-B1` — Broker vs Rita: **4-0**, avg margin Broker **+92278.25**.\n3. `artifacts/kaggriculture/autolab/20260907T102030Z_OUR-MIX11-SILAS-B1` — MIX11 vs Silas: **0-4**, avg margin **-94283.5**.\n4. `artifacts/kaggriculture/autolab/20260907T102034Z_OUR-M60F-SILAS-B1` — M60_FERT vs Silas: **0-4**, avg margin **-88101.25**.\n\n## New falsifications / mechanism update\n- **M60_FERT does not generalize upward in the public ladder:** it is 0-4 vs Lena and 0-4 vs Silas despite the earlier 4-0 vs Rita. Fertilizer remains contextual/shaping-only, not a boundary mechanism.\n- **MIX11 does not cross Silas:** 0-4 with very large negative margin. This reinforces that the current student family has a structural deployment/compounding deficit rather than a narrow Broker-only exploit problem.\n- **Broker > Rita 4-0 by +92k average margin** confirms Rita is a cheap regression target substantially below the current boundary; beating Rita is therefore weak evidence for promotion.\n- The prior `conditional_deployment_pulse` remains the leading causal candidate, but it is still **MATCHUP + WORLD_REGIME / shaping-only** until student-side heldout proof exists.\n\n## League interpretation\nThe public ladder separation is becoming clearer: current student variants lose decisively not only to Broker but also to Lena/Silas. Therefore boundary stays Broker; no premature promotion. Existing public evidence still supports Cleo/Silas above Broker, but heldout B2 remains required for formal league movement.\n\n## Queue / execution state\nTwo worker invocations consumed four READY B1 jobs; queue moved from 10 READY to **8 READY** after the second invocation, with **21 DONE** total at that point. A third immediate worker invocation was blocked by the tool safety layer; this is a transient execution blocker, not evidence of job failure. Existing queue remains non-empty, so no duplicate replenishment was added in this checkpoint.\n\n## Guardrails\nPrimary metric W/L/tie preserved. Margins are diagnostic only. No champion promotion, Kaggle submission, spend, KYC/payout/identity/legal action, secret exposure, or unlicensed TRACE/replay incorporation occurred. TS-port remains mechanical evidence rather than exact CPython parity.\n\n## First incomplete step\nConsume the remaining READY jobs, prioritizing B2 heldouts already queued. Then reconcile public B2 ordering and test the smallest observable `<1s` student implementation of the conditional deployment pulse only if heldout evidence survives. Require Broker boundary heldout, Rita regression, runtime, provenance, and counterproof before promotion.",[],[175,176,2570,2844],"3aca245d-95b5-4efd-a43a-45729d72a886","KAG_HOURLY_CHECKPOINT_20260907_0945.md","e5461fa8-aaa0-11f1-b97f-1607ee4eb77e",518,["D",1788774344303],"# Kaggriculture Hourly Checkpoint — 2026-09-07 09:45 UTC\n\n## Material changes\n- **Champion:** unchanged. No promotion gate passed.\n- **Boundary:** Broker Bea remains boundary; every tested student-side variant still lost all 4 mirrored episodes in B1.\n- **Stress:** Closer Cleo remains stress. MIX11 lost 0-4, avg margin -90078.25; largest observed mean gap deterioration was day 24→27 (-20494.5).\n- **Regression:** Rancher Rita remains strongest defeated regression reference in this slice; M60_FERT beat Rita 4-0, avg margin +38597, but this is small-N and not sufficient for promotion.\n\n## Sidecar results consumed since prior checkpoint\nAll are `AUTOLAB_COMPUTE_RESULT_NOT_CHAMPION_PROOF` and had 0 execution errors.\n1. `artifacts/kaggriculture/autolab/20260907T090204Z_OUR-M60F-BROKER-B1` — M60_FERT vs Broker: 0-4, avg margin -99124.75.\n2. `artifacts/kaggriculture/autolab/20260907T090208Z_X-CLEO-BROKER-B1` — Cleo vs Broker: 3-1, avg margin +1327.5.\n3. `artifacts/kaggriculture/autolab/20260907T091910Z_OUR-M60F-RITA-REG-B1` — M60_FERT vs Rita: 4-0, avg margin +38597.\n4. `artifacts/kaggriculture/autolab/20260907T091914Z_X-SILAS-LENA-B1` — Silas vs Lena: 4-0, avg margin +923.\n5. `artifacts/kaggriculture/autolab/20260907T093422Z_OUR-MIX11-CLEO-STRESS-B1` — MIX11 vs Cleo: 0-4, avg margin -90078.25.\n6. `artifacts/kaggriculture/autolab/20260907T093426Z_X-SILAS-BROKER-B1` — Silas vs Broker: 4-0, avg margin +4678.75.\nNo FAILED sidecar jobs were present.\n\n## Falsifications / negative knowledge\n- **M60_FERT is not a boundary solution:** despite 4-0 vs Rita, it went 0-4 vs Broker and was worse in average margin than MIX11 on its B1 seeds. Classification: `REGRESSION_STRONG / BOUNDARY_REJECTED`.\n- **Fertilizer sell is not sufficient:** MIX11_FERT improved B1 average margin versus plain MIX11 (-86275.5 vs -97315.5) but remained 0-4; retain as contextual shaping only.\n- **Early pasture reduction alone is not the missing universal:** fixed-world pasture/labor frontier found J4B2_H8 only +3394 margin over baseline and still deeply losing. This is supportive/local evidence, not a universal rule.\n- **A single fixed operations pulse is not universal:** Broker fixed-fork teacher outputs change by shop regime and decision day; day-12 evidence did not confirm a positive action.\n\n## Winning mechanism candidate — conditional deployment pulse\nClassification: **MATCHUP + WORLD-REGIME, teacher/shaping-only**.\n\nIndependent fixed-prestate evidence now repeats a causal pattern: when our crop/animal/land deployment lags at day 6–9, temporarily increasing labor and/or pasture build target can improve future value, but the correct intervention depends on the observed shop regime.\n\n- Broker D6, `FARMERS_MARKET,FARMERS_MARKET`: combined `B` (hands 9 + build target 16) at N=16: Δ vs N +2817.2, 95% LCB +1066.6, positive-rate 81.25%.\n- Broker D9, `PIZZA_SHOP,PET_CAFE,ICE_CREAM_SHOP`: `H` (hands 9) at N=12: Δ +12684.4, LCB +2748.4, positive-rate 75%.\n- Broker D9, `ICE_CREAM_SHOP,YARN_STORE,FARMERS_MARKET`: `B` at N=12: Δ +6051.6, LCB +2638.6, positive-rate 83.3%.\n- Broker D12, `BAKERY,BAKERY,BRUNCH_SPOT,YARN_STORE`: no action cleared the teacher persistence gate; `N` retained.\n- A Silas D9 persisted context independently preferred build expansion (`W/B`) with positive LCB, showing the observable mechanism is not Broker-exclusive, while the action choice remains context-dependent.\n\nMechanistic interpretation: the current family frequently carries excess early pasture/headroom while under-deploying productive crop tiles and land relative to top opponents. The useful control is **not “always buy land/hands/build more”**; it is a state-conditioned deployment pulse before the opponent’s compounding phase. This directly preserves the prior rejection of universal early-land/static-hands rules.\n\n## Public league evidence\nSmall-N only, therefore no league promotion yet:\n- Silas > Broker 4-0 and Silas > Lena 4-0 in B1.\n- Cleo > Broker 3-1 in B1 and previously Cleo > Silas 4-0 in B1.\nCurrent ordering hypothesis remains `Cleo > Silas > Lena/Broker`, but heldout B2 is required before changing the dynamic league.\n\n## Queue replenishment\nEight deduplicated heldout jobs were added after the existing READY queue: MIX11/MIX11_FERT/M60_FERT vs Broker B2; M60_FERT vs Rita B2; Cleo×Broker, Silas×Broker, Silas×Lena, Lena×Broker B2. Each B2 uses four fresh seeds mirrored by the worker (8 episodes/job). Existing B1 queue was preserved.\n\n## Evidence/guardrails\n- Primary metric remains W/L/tie. Margins and gap trajectories are diagnostics only.\n- Fixed-prestate teacher evidence is offline shaping evidence, not student/champion proof.\n- TS/Pyodide sidecar is mechanical evidence; exact CPython parity remains a future gate.\n- No Kaggle submission, spend, KYC/payout/identity/legal action, secret exposure, or unlicensed TRACE incorporation occurred.\n\n## First incomplete step\nConsume the remaining B1 jobs, then B2 heldouts. If the conditional deployment pulse persists across independent Broker/Silas contexts, implement the smallest <1s student gate in the source-of-truth lineage and require boundary heldout + Rita regression + runtime/provenance/counterproof before any champion consideration.",[],[175,176,2570,2852],"2780bbcd-e7bd-4d7c-ba84-17faeac7d2db","KAG_HOURLY_CHECKPOINT_20260907_0849.md","095508d0-aa99-11f1-95fa-1607ee4eb77e",516,["D",1788771034062],"# Kaggriculture Hourly Checkpoint — 2026-09-07 08:49 UTC\n\n## State\n- Source of truth: `cleitoncosta/chat-native-maestro@economic-v0.6`.\n- Autonomous compute sidecar: `cleitoncosta/kaggriculture-autolab@main`.\n- Sidecar trigger: `worker.ts`, active every 15 minutes (Free-tier minimum).\n- Worker: `KAG_AUTOLAB_WORKER_V1`; core: `KAG_AUTOLAB_CORE_V1`.\n- Queue supports `PUBLIC_CROSSPLAY` and `OUR_VS_REFERENCE` (`MIX11`, `MIX11_FERT`, `M60_FERT`).\n- Worker persists queue/runs/state/artifacts in its own SQLite and writes Markdown/JSON blobs under `artifacts/kaggriculture/autolab/`.\n\n## Validation performed\n- Manual worker smoke: `MIX11 vs Rancher Rita`, mirrored seat, 2 episodes, 2 wins, avg margin +9511.\n- Manual public crossplay smoke: `Closer Cleo vs Slotter Silas`, 2 episodes, 1-1, avg margin Cleo +90.5.\n- Inter-episode reference contamination mitigation added: reference modules are re-executed/reset before every episode.\n- Reset-ref control smoke: `Closer Cleo vs Broker Bea`, 2 episodes, Cleo 2-0, avg margin +5874.\n- First production tick: `MIX11 vs Broker`, 4 episodes, 0-4, avg margin -97315.5; `Cleo vs Silas`, 4 episodes, Cleo 4-0, avg margin +2423.75.\n- First autonomous scheduled tick was independently observed before this checkpoint closed: `MIX11_FERT vs Broker`, 4 episodes, 0-4, avg margin -86275.5; `Cleo vs Lena`, 4 episodes, Cleo 4-0, avg margin +4772. This proves the 15-minute trigger is firing without a chat turn.\n\n## Queue\nAn initial production queue of 18 jobs was seeded, interleaving our variants against boundary/stress/regression opponents with high-tier public crossplay. The manual production tick reduced it to 16 READY; the first autonomous scheduled tick then processed two more, leaving 14 READY at checkpoint close.\n\n## Evidence state\nAll sidecar outcomes are `AUTOLAB_COMPUTE_RESULT_NOT_CHAMPION_PROOF`. W/L/tie remains primary; margin/gap trajectory is diagnostic only. No champion promotion occurred.\n\n## Guardrails\nNo Kaggle submission, spending, KYC/payout, identity/legal action, token use or account mutation was performed. Public high-tier agents are black-box opponents only; non-licensed TRACE/replay-plan content is excluded.\n\n## Next loop\nConsume the remaining sidecar jobs, compare MIX11/MIX11_FERT/M60_FERT on Broker/Rita/Cleo/Lena/Silas, mine public crossplay winner/loser gap windows, then use the hourly reasoning loop to create causal mutations rather than merely extending brute-force benchmarks.\n",[],[175,176,2570,2860],"0a0d871d-9265-4fea-bf65-45a51597b8a8","HOURLY_INDEX.md","171b015e-aa99-11f1-9994-1607ee4eb77e",842,["D",1788917010819],39,"# Kaggriculture Hourly Index\n\n| UTC | Checkpoint | Manifest | Evidence |\n|---|---|---|---|\n| 2026-09-07 08:49 | `KAG_HOURLY_CHECKPOINT_20260907_0849.md` | `KAG_HOURLY_MANIFEST_20260907_0849.json` | AUTOLAB_INFRA_OPERATIONAL_COMPUTE_RESULTS_NOT_CHAMPION_PROOF |\n| 2026-09-07 09:45 | `KAG_HOURLY_CHECKPOINT_20260907_0945.md` | `KAG_HOURLY_MANIFEST_20260907_0945.json` | HOURLY_CAUSAL_UPDATE_NO_CHAMPION_PROMOTION |\n\nThe hourly routine appends one deduplicated row per completed reasoning round. Sidecar machine artifacts are stored separately in `cleitoncosta/kaggriculture-autolab` under blob prefix `artifacts/kaggriculture/autolab/` and mirrored in `kag_autolab_artifacts`.\n\n- 2026-09-07 10:21 UTC — `KAG_HOURLY_CHECKPOINT_20260907_1021.md` / `KAG_HOURLY_MANIFEST_20260907_1021.json` — M60_FERT rejected upward vs Lena/Silas; Broker>Rita separation confirmed; conditional deployment pulse remains shaping-only; queue 8 READY, transient immediate-run blocker.\n- 2026-09-07 12:22 UTC — `KAG_HOURLY_CHECKPOINT_20260907_1222.md` / `KAG_HOURLY_MANIFEST_20260907_1222.json` — B2 heldout: all current variants 0-8 vs Broker; M60_FERT 8-0 Rita; public ladder BrokerBroker e Broker>Rita; B9 enfileirado.\n\n- 2026-09-07 15:34Z — `KAG_HOURLY_CHECKPOINT_20260907_1534.md` / `KAG_HOURLY_MANIFEST_20260907_1534.json` / `KAG_HOURLY_DIFF_20260907_1534.md` — sell-timing downgraded LOCAL; treatment detector added; D21/D24 liquidation near-null after 8/8 confirmed application; M40/M50/M60 static melon all 0-8 Broker; mechanism moved upstream to inventory creation/crop transition/capacity timing; B14 queue replenished.\n- 2026-09-07 15:42Z continuation — B14 public ladder stabilized; V5 isolated pasture-JIT and falsified it heldout (0-8, materially worse paired margin); negative lesson persisted; B16 left READY; next branch is conditioned crop-portfolio transition.\n\n- 2026-09-07 14:20 — `KAG_HOURLY_CHECKPOINT_20260907_1420.md` / `KAG_HOURLY_MANIFEST_20260907_1420.json` — MATERIAL_UPDATE; JIT independently falsified B17; Silas>Lena 6–2; queue UNVERIFIED due platform safety blocker.\n\n- 2026-09-07 15:20 America/Fortaleza — [checkpoint](KAG_HOURLY_CHECKPOINT_20260907_1820.md) · [manifest](KAG_HOURLY_MANIFEST_20260907_1820.json) — MATERIAL_UPDATE: Lena→Broker 7-1 repeated B18/B19; Silas→Lena 7-1; our M60_FERT remains 0-8 vs Broker; Cleo sidecar reference asset missing; B20 queued.\n\n- 2026-09-07 20:00Z — `KAG_HOURLY_CHECKPOINT_20260907_2000.md` / `KAG_HOURLY_MANIFEST_20260907_2000.json` / `KAG_HOURLY_DIFF_20260907_2000.md` — MATERIAL_UPDATE: MIX11 cheap regression 24-0 Rosa/Hana/Mateo; internal lineage still zero-win vs Broker; margin-proximity proxy falsified; public ladder Lena>Broker> Broker immediate, Broker lag2 echo, Rita near-zero under legal self-action-filtered detector; still observational/not causal; B58 Lena/Silas queued; no champion promotion.\n\n## 2026-09-08 10:43 UTC — MATERIAL_UPPER_FAMILY_TIMING_REPLICATION\n- Checkpoint: `KAG_HOURLY_CHECKPOINT_20260908_1043.md`\n- Manifest: `KAG_HOURLY_MANIFEST_20260908_1043.json`\n- Context preflight: PASS\n- Material: B58 fresh Lena/Silas legal self-action-filtered timing signatures directionally replicate prior immediate/tail patterns; remains observational/not causal; queue intentionally empty pending new teacher causal tooling; no champion promotion.\n\n\n## 2026-09-08 11:22 UTC — MATERIAL_MECHANISM_FALSIFICATION\n- Checkpoint: `KAG_HOURLY_CHECKPOINT_20260908_1122.md`\n- Manifest: `KAG_HOURLY_MANIFEST_20260908_1122.json`\n- Context preflight: PASS\n- Material: reconciled controlled clone/history capability with absent natural MIX11 support; audited apparent continuous `_opponent_scale→reserve_price` public-supply reaction path and found current Broker/Lena/Silas/Cleo `_RESERVE={}`, so path is configured inactive; no reaction/champion promotion.\n- Next: fixed-plan conversion/throughput source+own-capability census, prioritizing WHEAT procurement/feed continuity before any treatment.\n\n\n## 2026-09-08 12:36 UTC — MATERIAL_MECHANISM_FALSIFICATION\n- Checkpoint: `KAG_HOURLY_CHECKPOINT_20260908_1236.md`\n- Manifest: `KAG_HOURLY_MANIFEST_20260908_1236.json`\n- Diff: `KAG_HOURLY_DIFF_20260908_1236.md`\n- Context preflight: PASS\n- Material: simple WHEAT-stock shortage not supported in tested Broker windows; treatment-verified persistent routing worsened missed-feed stress and remained 0-8; formulation closed, task-service throughput becomes next census lane; no champion promotion.\n\n## 2026-09-08 12:45 UTC — CONTINUITY_REPAIR + MATCHUP_NONSPECIFIC_SERVICE_FALSIFICATION\n- Checkpoint: `KAG_HOURLY_CHECKPOINT_20260908_1245.md`\n- Manifest: `KAG_HOURLY_MANIFEST_20260908_1245.json`\n- KAG_CONTEXT_PREFLIGHT: PASS\n- Material: canonical 12:45 task-service census was reproduced after detecting it had been written to Current State/Negative Knowledge without hourly closure; service backlog persists in both Broker-loss and Rita-win contexts and is not supported as a Broker-specific cause. Next: differential economic-conversion census before mutation.\n\n\n## 2026-09-08 13:15 UTC — MATERIAL_WOOL_MEDIATION_NARROWING\n- Checkpoint: `KAG_HOURLY_CHECKPOINT_20260908_1315.md`\n- Manifest: `KAG_HOURLY_MANIFEST_20260908_1315.json`\n- KAG_CONTEXT_PREFLIGHT: PASS\n- Material: top-family WOOL footprint remains family/world-regime structure but stable suppression vs Rita is falsified as universal; fixed-prestate WOOL price/inventory and our crop/animal-mix direct reaction screens are null in tested landmarks with valid cash positive controls; no champion/policy promotion.\n\n\n## 2026-09-08 15:20 UTC — MATERIAL_WOOL_MEDIATION_NARROWING\n- Checkpoint: `KAG_HOURLY_CHECKPOINT_20260908_1520.md`\n- Manifest: `KAG_HOURLY_MANIFEST_20260908_1520.json`\n- KAG_CONTEXT_PREFLIGHT: PASS\n- Material: fresh world-conditioned WOOL census preserves upper-family cluster and inventory→price inverse relation in 8/8 paired worlds per target, but simple opponent-sheep→inventory sole-mediator story is not supported (4/8 sign alignment); our output/world demand remain confounders; no champion/policy promotion.\n\n## 2026-09-08 17:17 UTC — MATERIAL_WOOL_TEMPORAL_MEDIATION_NARROWING\n- Checkpoint: `KAG_HOURLY_CHECKPOINT_20260908_1717.md`\n- Manifest: `KAG_HOURLY_MANIFEST_20260908_1717.json`\n- KAG_CONTEXT_PREFLIGHT: PASS\n- Material: fresh turn-level WOOL diagnostic found opponent public sheep count adds ~zero incremental explanatory power for Δ shared inventory after preliminary controls; inventory→price inverse relation remains exact on nonzero simultaneous moves; V2 demand marker incomplete, so next gate is actual output-flow + corrected deterministic demand instrumentation; no champion/policy promotion.\n\n## 2026-09-08 18:25 UTC — MATERIAL_MEASUREMENT_FALSIFICATION\n- Checkpoint: `KAG_HOURLY_CHECKPOINT_20260908_1825.md`\n- Manifest: `KAG_HOURLY_MANIFEST_20260908_1825.json`\n- KAG_CONTEXT_PREFLIGHT: PASS\n- Material: V3 fixed the zero-support demand-exposure marker but public sheep `yieldSum` delta was zero-variance across all tested D18-D27 rows, falsifying it as an output-flow detector rather than falsifying opponent output; inventory→price inverse relation replicated; no champion/policy promotion.\n\n## 2026-09-08 19:35 UTC — MATERIAL_MEASUREMENT_ROOT_CAUSE\n- Checkpoint: `KAG_HOURLY_CHECKPOINT_20260908_1935.md`\n- Manifest: `KAG_HOURLY_MANIFEST_20260908_1935.json`\n- KAG_CONTEXT_PREFLIGHT: PASS\n- Material: direct official-engine legal public SHEEP yield/pending-care signals vary, falsifying the apparent observability absence; V27 scenario lifecycle aggregate/consumer is false-zero and blocked for mediation until parity repair; no champion/policy promotion.\n\n## 2026-09-08 20:21 UTC — MATERIAL_MEASUREMENT_REPAIR\n- Checkpoint: `KAG_HOURLY_CHECKPOINT_20260908_2021.md`\n- Manifest: `KAG_HOURLY_MANIFEST_20260908_2021.json`\n- KAG_CONTEXT_PREFLIGHT: PASS\n- Material: fresh corrected V27 scenario telemetry recovers abundant SHEEP yield/pending-care transitions; prior V4/V4b false-zero traced to nested consumer-path mismatch, superseding the 19:35 aggregate-bug attribution; no champion/policy promotion.\n\n\n## 2026-09-08 21:19 UTC — MATERIAL_MEASUREMENT_PARITY_AND_MEDIATION_SIGNAL\n- Checkpoint: `KAG_HOURLY_CHECKPOINT_20260908_2119.md`\n- Manifest: `KAG_HOURLY_MANIFEST_20260908_2119.json`\n- KAG_CONTEXT_PREFLIGHT: PASS\n- Material: true same-turn SHEEP lifecycle raw-public↔aggregate parity passed 4314/4314 rows per Broker/Rita target and deliberate corruption was caught 4314/4314; repaired fresh WOOL diagnostic now shows small consistent opponent-lifecycle incremental association for Broker (ΔR² 0.00809) versus near-zero Rita (0.000431), still observational/not causal; no champion/policy promotion.\n\n\n## 2026-09-08 22:25 UTC — MATERIAL_DEMAND_LEDGER_PARITY_AND_MEDIATION_NARROWING\n- Checkpoint: `KAG_HOURLY_CHECKPOINT_20260908_2225.md`\n- Manifest: `KAG_HOURLY_MANIFEST_20260908_2225.json`\n- KAG_CONTEXT_PREFLIGHT: PASS\n- Material: naive YARN_STORE×1 demand assumption failed parity and was corrected from official engine semantics to ×2; exact executed WOOL town-demand ledger then passed 3882/3882 with mutation control 3882/3882. Fresh exact-demand residual retains small upper-family lifecycle association above Rita but is heterogeneous/not causal; no champion/policy promotion.\n\n\n## 2026-09-08 23:23 UTC — MATERIAL_OWN_SELL_PARITY_AND_FULL_RESIDUAL_NARROWING\n- Checkpoint: `KAG_HOURLY_CHECKPOINT_20260908_2323.md`\n- Manifest: `KAG_HOURLY_MANIFEST_20260908_2323.json`\n- KAG_CONTEXT_PREFLIGHT: PASS\n- Material: own WOOL SELL interpreter semantics validated including inventory insufficiency and 10-order truncation; fresh full conservation residual shrinks upper-family lifecycle association to ~0.00030–0.00055 ΔR² vs Rita ~0.000003, with strong world heterogeneity; no champion/policy/reaction-trait promotion.\n\n\n## 2026-09-09 00:25 UTC — MATERIAL_WOOL_CONSERVATION_CAUSAL_PARITY_AND_DEPRIORITIZATION\n- Checkpoint: `KAG_HOURLY_CHECKPOINT_20260909_0025.md`\n- Manifest: `KAG_HOURLY_MANIFEST_20260909_0025.json`\n- KAG_CONTEXT_PREFLIGHT: PASS\n- Material: fixed-prestate/no-town-demand forced opponent WOOL supply maps exactly one-for-one into shared inventory with mutation-control PASS and local nonlinear price impact; sparse upper-family checkpoint actions provide no evidence of continuous natural WOOL selling; WOOL branch deprioritized as primary boundary mechanism; no champion/policy/reaction-trait promotion.\n\n\n## 2026-09-09 01:23 UTC — MATERIAL_NON_WOOL_CONVERSION_RANKING\n- Checkpoint: `KAG_HOURLY_CHECKPOINT_20260909_0123.md`\n- Manifest: `KAG_HOURLY_MANIFEST_20260909_0123.json`\n- KAG_CONTEXT_PREFLIGHT: PASS\n- Material: fresh exact-self-executed AVIÃO ranking selects MILK as largest non-WOOL Broker-vs-Rita late conversion-value differential; independent fresh decomposition replicates lower MILK sale/shed/cow/yield in Broker context despite higher price, but with large world heterogeneity; no champion/policy/reaction-trait promotion.\n",[],[175,176,2570,2868],"34d3faf3-74bf-4f42-8f12-74c09924862b",511,["D",1788770953126],[199,200,201,202,203,204,205,206,207,208,209,210,211,212,213,214,215,216,217,218,219,220,221,222,223,224,225,226,227,228,229,230,231,232,233,234,235,236,237],[175,176,2570],"4d19e772-5ecf-423c-b025-1974ae892fd9","experiments","e35d199c-ab06-11f1-8660-1607ee4eb77e",561,["D",1788818149777],[],[175,176,2882],"6b9ff7b2-96de-4b4e-a15d-b1171e92d020","PREFLIGHT_20260908_1047.md","a9f64414-ab72-11f1-9e39-1607ee4eb77e","55449dab-6696-482d-b71e-d82cdda0f9d5",745,["D",1788864439059],"# KAG_CONTEXT_PREFLIGHT — 2026-09-08 continuation epoch\n\nStatus: PASS / POST-REHYDRATION RECONCILED\n\nREAD_FIRST = READ\nSTRATEGIC_INTELLIGENCE = READ\nCAPABILITY_TAXONOMY = READ (190 traits)\nCURRICULUM = READ\nNEURON_PROTOCOL = READ\nJOINED_RESEARCH_PLAN = READ\nOFFICIAL_RULES = READ\nRT_ARCHITECTURE = READ\nCURRENT_STATE = READ\nNEGATIVE_KNOWLEDGE = READ\nFULL_15MIN_REANALYSIS = READ\nREAL_SCENARIO_COVERAGE = READ\nRECOVERY_RECONCILIATION = READ\nDAY_AUDIT_20260907 = READ\nDAY_AUDIT_20260908 = READ\nOPPONENT_INDEX = READ\nTARGET_DOSSIERS = READ (Broker, Cleo, Rita, Lena, Silas; weak-trio calibration inherited from INDEX/PROFILE_CAL)\nEXPERIMENT_MATRIX = READ\nPROFILE_CAL = READ\nHOURLY_INDEX = READ\nLATEST_CHECKPOINT = READ (`KAG_HOURLY_CHECKPOINT_20260908_1043.md`)\nLATEST_MANIFEST = READ (`KAG_HOURLY_MANIFEST_20260908_1043.json`)\nSIDECAR_QUEUE = RECONCILED (241 DONE / 5 FAILED / 0 READY / 0 RUNNING)\nSIDECAR_LATEST_RUNS = RECONCILED (B57/B58)\nSIDECAR_STATE = RECONCILED\nSIDECAR_MAIN_CORE = LIVE-READ as `KAG_AUTOLAB_CORE_V25_CAPACITY_PORTFOLIO_FACTORIAL`\n\n## Contradictions/stale state\n1. `KAG_RT_ARCHITECTURE.md` section 12 still describes main V22/V23-era implementation. This is stale implementation text; live sidecar main source + latest checkpoint take precedence.\n2. Early section of `KAG_CURRENT_STATE.md` still mentions V22, while later timestamped sections and live sidecar show subsequent promotions. Live main is V25; V25 retains legal self-action memory introduced by V23.\n3. Sidecar strategy state is current at the 10:43 checkpoint (`PATH_ORDER_SLOT_FUNDING_CAUSAL_ATTRIBUTION`).\n\n## Ordering breach recovered\nBefore this fresh rehydration finished, the continuation turn executed an adjacent hands-only screen (`LABOR9_D9_15` and repeat `LABOR10_D9_15`) on the experimental branch. This violated the canonical `00_READ_FIRST` ordering rule because current canonical state already closed bounded LABOR10 and prohibited adjacent blind hands sweeps.\n\nThe runs are quarantined from strategy generation until this receipt. Post-preflight reconciliation shows they only strengthen existing negative knowledge:\n- fresh `[106293,106307,106321,106333]`, mirrored seats;\n- MIX11 0-8, avg margin -104353.875;\n- LABOR9_D9_15 0-8, avg margin -111207.5, paired mean delta -6853.625, positive 0/8, negative 8/8;\n- LABOR10_D9_15 0-8, avg margin -113096, paired mean delta -8742.125, positive 0/8, negative 8/8;\n- treatment detector: 30 daily checks/game, 6 treatment days/game, errors 0.\n\nClassification: `PRECHECK_EXECUTED_BUT_POST_PREFLIGHT_RECONCILED / REINFORCING_NEGATIVE_ONLY`. No adjacent hands sweep is authorized.\n\n## Experimental branch status\n`kaggriculture-autolab@reaction-attribution-v1` has an unpromoted telemetry-only experimental core label `KAG_AUTOLAB_CORE_V27_PRIVATE_RESOURCE_LIFECYCLE`, adding own-private item totals to scenario telemetry. It is not sidecar main, not champion logic, and still requires the correct non-regression selftest before it may be used as measurement infrastructure.\n\n## Final gate\nCONTEXT_PREFLIGHT = PASS.\n\nNext authorized scientific step remains the canonical first incomplete step: inspect prior V8–V14 path/order/funding teacher work, then implement/preregister a fixed-history + fixed-terminal/prestate multi-step path × order-slot/funding discriminant with treatment-fired accounting, positive/negative controls, and no macro policy mutation.",[],[175,176,2897,2889],"context","b03d307e-e59b-497a-bcf1-4c494b1dfd48","KAG_CONTEXT_PREFLIGHT_20260908_0240.md","1db52bca-ab2f-11f1-88b0-1607ee4eb77e",603,["D",1788835427529],"# KAG_CONTEXT_PREFLIGHT — 2026-09-08 02:40Z\n\nStatus: PASS\nScope: best available recoverable Kaggriculture project context; **not** a claim of byte-for-byte enumeration of every account conversation.\n\n## Receipt\n\n```text\nKAG_CONTEXT_PREFLIGHT\nREAD_FIRST = READ\nSTRATEGIC_INTELLIGENCE = READ\nCAPABILITY_TAXONOMY = READ (190 explicit traits)\nREAL_WORLD_CURRICULUM = READ\nNEURON_PROTOCOL = READ\nJOINED_RESEARCH_PLAN = READ\nOFFICIAL_RULES = READ\nRT_ARCHITECTURE = READ\nCURRENT_STATE = READ\nNEGATIVE_KNOWLEDGE = READ\nRECOVERY_RECONCILIATION = READ\nDAY_AUDIT = READ\nOPPONENT_INDEX = READ\nTARGET_DOSSIER = READ (Broker + weak controls + upper-ladder dossiers reconciled)\nEXPERIMENT_MATRIX = READ\nPROFILE_CALIBRATION = READ\nHOURLY_INDEX = READ\nLATEST_CHECKPOINT = READ (20260907_2220; operationally stale where superseded below)\nSIDECAR_STATE = RECONCILED\nEXPERIMENTAL_BRANCHES = RECONCILED\nCONTRADICTIONS = RESOLVED/REGISTERED\nCONTEXT_PREFLIGHT = PASS\n```\n\n## Contradictions reconciled\n\n### C1 — B34 infrastructure status\nHistorical checkpoint `KAG_HOURLY_CHECKPOINT_20260907_2220.md` says Broker/Lena legal B34 were blocked by hrana persistence failure. Current sidecar evidence supersedes this operational state: B34 was recovered from Blob and marked DONE; worker V2 compact persistence now supports large legal artifacts.\n\nDisposition: historical checkpoint remains provenance; `KAG_CURRENT_STATE.md` is current authority.\n\n### C2 — Broker↔Lena first divergence\nReduced observer/old state localized the first discriminating landmark at D12H2. New full-public fresh paired evidence on `b34-analysis-v1` found first full-public divergence at D10H5 after a D10H4 offline action branch in 8/8 paired worlds.\n\nDisposition: D12H2 remains a reduced-feature landmark, not the first full-public divergence. The material value of the D10H4 extra action remains UNVERIFIED.\n\n### C3 — original OBG scope loss\nThe recovered strategic docs already contained opponent modeling/probing, but several original multidimensional OBG concepts from prior project context were not explicit first-class fields in the reconstructed dossiers.\n\nDisposition: `KAG_OPPONENT_CAPABILITY_TAXONOMY.md` now contains 190 explicit traits, including advance/retreat, bluff/counter-bluff, aggression/defensiveness, patience/impulsivity, exploration/exploitation, adaptability/inertia, models-us, deception/probe awareness and related dimensions. Unmeasured = UNKNOWN, never absent.\n\n### C4 — classifier success versus policy invariance\nStrong same-policy top-family heldout classification did not transfer perfectly when our policy changed; Broker/Lena confusion appeared.\n\nDisposition: same-policy accuracy cannot be promoted to policy-invariant classifier. Experimental V23 self-action normalization remains non-canonical pending validation.\n\n## Stale/superseded material detected\n\n- Latest hourly checkpoint/index ends before persistence-v2 repair, B34 Blob recovery, B35 and the context-continuity repair.\n- Old sidecar `hourly:current_hypothesis` D12-local framing was superseded by `CONTEXT_RECONCILIATION_REQUIRED_BEFORE_NEXT_HYPOTHESIS`.\n- Old `KAG_RT_ARCHITECTURE.md` V19 implementation-status section was updated to V22/V23 reality.\n\nThese do not block the preflight because newer canonical state and explicit contradiction dispositions now exist.\n\n## Canonical direction after preflight\n\nPrimary mission:\n\n`LEGAL OBSERVATION → IDENTITY/REGIME BELIEF → REACTION MODEL → CONTROLLED PROBE → CAUSAL EFFECT → NET VALUE → CONDITIONAL POLICY ROUTING`\n\nFingerprint/tape/root-cause studies are supporting subproblems. They cannot replace measurement of how opponents react to our actions, to shared-market changes caused by us, to history, probes, shocks, regime changes and strategic signals.\n\n## Hard gate for future rounds\n\nAny future strategic round lacking a fresh/valid `KAG_CONTEXT_PREFLIGHT` is **not eligible to update canonical strategy**, even if its compute result is valid.\n\n`COMPUTE_VALID != STRATEGY_CONTEXT_VALID`.\n",[],[175,176,2897,2899],"80ecda6f-3ef9-4b78-a635-c408c175d235",["D",1788835427529],[195,196],[175,176,2897],"1a82d0cc-c03a-4cd8-8cff-4de7b85f32f9","d83d55be-aa97-11f1-b62f-1607ee4eb77e","63b5fd96-aa96-11f1-8b2f-1607ee4eb77e",588,["D",1788834989505],"# Kaggriculture Autolab\n\n> **START GATE:** strategy/replanning must begin by reading `../00_READ_FIRST.md` from `artifacts/kaggriculture/` and completing `KAG_CONTEXT_PREFLIGHT`. This sidecar computes evidence; it does **not** define strategy independently.\n\nAutonomous compute sidecar for `chat-native-maestro/economic-v0.6`.\n\n## Runtime split\n\n- Reasoning/replanning: ChatGPT hourly automation `Kaggriculture Evolução Horária`.\n- Repetitive compute: `cleitoncosta/kaggriculture-autolab`, cron worker every 15 minutes on the Free-tier minimum.\n- Source-of-truth strategy/code: `cleitoncosta/chat-native-maestro`, branch `economic-v0.6`.\n- Worker queue/results: SQLite in `cleitoncosta/kaggriculture-autolab`.\n- Per-run machine artifacts: project-scoped Blob keys under `artifacts/kaggriculture/autolab/` plus mirrored rows in `kag_autolab_artifacts`.\n- Hourly human/audit checkpoints: versioned Markdown/JSON files generated by the ChatGPT hourly routine in this directory.\n\n## Queue contract\n\nSupported worker jobs in V1:\n\n1. `PUBLIC_CROSSPLAY` — reference agent A vs B over explicit seeds and mirrored seats.\n2. `OUR_VS_REFERENCE` — our independently authored/allowed variants (`MIX11`, `MIX11_FERT`, `M60_FERT`) vs a public reference opponent.\n\nThe worker processes up to 2 jobs per tick with a soft 35 s budget, retries failed jobs up to 3 times with a 15-minute backoff, and persists every result before continuing.\n\n## Evidence discipline\n\nAutolab results are `AUTOLAB_COMPUTE_RESULT_NOT_CHAMPION_PROOF`. W/L/tie is the primary competitive metric. Margin, bank, gap trajectory and bottleneck windows are secondary diagnostics. A worker result cannot promote a champion by itself; promotion requires heldout, regression, runtime, provenance and counter-proof gates.\n\n## Guardrails\n\nNo Kaggle submission, spend, KYC/payout, token handling or account changes are performed. Public high-tier agents are used as black-box opponents for behavioral/mechanism study. Non-licensed `_TRACE`/reconstructed replay plans are not incorporated.\n",[],[175,176,2918,2069],"autolab","18d5c1d9-7db9-42ba-b3be-d1571974c3b1",497,["D",1788769831958],[193],[175,176,2918],"15714ce7-0402-4b30-b72e-4bae7bf7d496",496,["D",1788769820625],[192,194,197,198,238,260,261,262],[175,176],"92a4504f-1375-4b96-be86-04027652b123",["D",1788715656185],[191,263,264,265,266,267],[175],["D",1788707065045],["D",1789575110896],{"_11":12,"_136":71},{"_11":12,"_131":132,"_70":71,"_30":69,"_133":134,"_135":24},{"_39":120,"_13":16,"_33":36,"_121":81,"_122":123,"_124":125,"_126":86,"_127":81,"_128":42,"_129":14,"_130":12},"ctx",{"_54":-5,"_156":2971,"_2972":2973,"_2974":2975,"_2976":24,"_2977":24,"_2978":24,"_2979":94,"_2980":-5,"_2981":-7,"_2982":-7,"_2983":26,"_2984":24,"_2985":24,"_2986":-5,"_2987":24,"_2988":24,"_2989":24,"_2990":-5},"ENV",{"_2942":2943,"_2944":2945,"_2946":162,"_2947":2948,"_2949":2950,"_2951":2952,"_2953":2954,"_2955":2956,"_2957":2958,"_2959":2960,"_2961":2962,"_2963":2964,"_2965":2966,"_2967":2968,"_2969":2970},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","RENDER_GIT_COMMIT","5015b8df46402d8bba4f485ecc37aec202b0e06b","RENDER_GIT_BRANCH","POSTHOG_API_TOKEN","phc_nmoyuAkStEZuUp69rYlg1DZgrIt19Gx0Y4Hwf8tGgJP","POSTHOG_API_HOST","https://wart.val.town","WEB_API_TEMPLATE","https://{val}.web.val.run","APP_URL","https://www.val.town","MODULE_URL","https://esm.town","VANITY_WEB_API_TEMPLATE","https://{label}.val.run/","DENOLS_URL","https://lsp.val.town","NODE_ENV","production","CLOUDFLARE_TURNSTILE_SITE_KEY","0x4AAAAAACtaIZJH7s9rBMZ6","GOOGLE_CLIENT_ID","875989696899-jtsqkhlbdv5jrlsk8631p7ipjg3c6o62.apps.googleusercontent.com","STRIPE_PUBLISHABLE_KEY","pk_live_51N2zvWEpZTTLipIkVEwXrUG9AT0yeOOEYxszarZT1jpsZV58Ly6wlg3SltZpHN26CUalOBmUSY7adQRQiSkCZz2500qSXNkf8Y","API_URL","https://api.val.town",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]

History · mateo.md