[{"_1":2,"_177":-5,"_178":-5},"loaderData",{"_3":4,"_93":94},"root",{"_5":6,"_36":37,"_66":67},"ctx",{"_7":8,"_23":-5,"_24":25,"_26":27,"_28":20,"_29":20,"_30":20,"_31":-5,"_32":33,"_34":20,"_35":20},"auth",{"_9":-5,"_10":-5,"_11":-5,"_12":-5,"_13":-5,"_14":-5,"_15":-5,"_16":-5,"_17":-5,"_18":-5,"_19":20,"_21":22},"sessionClaims","sessionId","sessionStatus","userId","actor","orgId","orgRole","orgSlug","orgPermissions","factorVerificationAge","isAuthenticated",false,"tokenType","session_token","user","preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesChatPanelState","notificationCount",0,"flagDenols","isTeamOrgMember","ENV",{"_38":39,"_40":41,"_42":43,"_44":45,"_46":47,"_48":49,"_50":51,"_52":53,"_54":55,"_56":57,"_58":59,"_60":61,"_62":63,"_64":65},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","VAL_RUNTIME_NAME","prod","RENDER_GIT_COMMIT","c9faa78320e89ae32855d559076900015c9debb4","RENDER_GIT_BRANCH","main","CLERK_PUBLISHABLE_KEY","pk_live_Y2xlcmsudmFsLnRvd24k","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","API_URL","https://api.val.town","clerkState",{"_68":69},"__internal_clerk_state",{"_70":71,"_72":47,"_73":74,"_75":74,"_76":20,"_77":74,"_78":74,"_79":74,"_80":74,"_81":82,"_91":20,"_92":20},"__clerk_ssr_state",{"_21":22,"_9":-5,"_10":-5,"_11":-5,"_12":-5,"_13":-5,"_14":-5,"_15":-5,"_16":-5,"_17":-5,"_18":-5,"_19":20},"__publishableKey","__proxyUrl","","__domain","__isSatellite","__signInUrl","__signUpUrl","__afterSignInUrl","__afterSignUpUrl","__clerk_debug",{"_83":20,"_19":20,"_84":74,"_85":86,"_87":74,"_88":47,"_89":20,"_90":74},"isSignedIn","proxyUrl","reason","session-token-and-uat-missing","message","publishableKey","isSatellite","domain","__telemetryDisabled","__telemetryDebug","routes/embed.x.$handle.$name.$",{"_95":96,"_97":98,"_99":100,"_101":102,"_155":156,"_173":174,"_175":176},"handle","dcm31","name","ev-goal-optimizer","path","optimizer.ts","project",{"_103":104,"_97":98,"_105":106,"_107":-5,"_108":20,"_109":110,"_111":112,"_113":-5,"_12":114,"_115":-5,"_116":20,"_117":-5,"_118":-5,"_119":33,"_14":120,"_121":-5,"_122":-5,"_123":-5,"_124":125,"_126":125,"_127":128,"_129":128,"_130":128,"_131":125,"_132":125,"_23":133,"_136":137,"_141":142,"_154":20},"id","0d4c8a1a-1b30-11f1-a769-42dde27851f2","createdAt",["D",1773002862068],"deletedAt","legacyEnvVars","privacy","public","description","Logistic regression goal optimizer via expected value","imageUrl","9319c175-2593-4f88-ba21-04f2ef7b1c28","pinnedFileId","pinPreview","pinMigrated","screenshotVersion","likeCount","d663a026-6449-42f6-9156-1d4a2d64d38c","tursoName","tursoHostname","pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm",{"_95":96,"_134":135},"profileImageUrl","https://imagedelivery.net/iHX6Ovru0O7AjmyT5yZRoA/260ec866-bb5d-4641-4bfe-526cca896a00/square512","org",{"_95":96,"_138":135,"_12":114,"_139":140},"avatar","tier","pro","branch",{"_143":144,"_97":45,"_103":145,"_146":104,"_147":-5,"_148":-5,"_149":-5,"_150":-5,"_151":-5,"_152":98,"_153":96},"version",3,"0d4cd83a-1b30-11f1-a769-42dde27851f2","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","isOwned","file",{"_103":157,"_97":100,"_158":159,"_160":-5,"_161":145,"_162":163,"_164":165,"_105":166,"_143":33,"_167":155,"_168":169,"_170":20,"_171":-5,"_99":172},"35d899ce-1b30-11f1-b6c3-42dde27851f2","fileId","35d89a78-1b30-11f1-b6c3-42dde27851f2","parentId","branchId","branchVersion",2,"latest",true,["D",1773002930080],"fileType","content","/**\n * Pure logistic regression + EV goal optimizer.\n * Zero dependencies. Import and use in your own val.\n *\n * Usage:\n * import { recommend, explain } from \"https://esm.town/v/dcm31/ev-goal-optimizer/optimizer.ts\";\n * const result = recommend(myHistory, \"2026-03-08\");\n * // result.best = { goal: 28, prob: 0.52, ev: 14.6 }\n */\n\n// ─── Types ──────────────────────────────────────────────────────────────────\n\nexport type HistoryRow = {\n date: string; // YYYY-MM-DD\n goal: number; // what the target was\n actual?: number; // what was actually done (null/undefined = no data)\n achieved: 0 | 1; // did they hit the goal?\n};\n\nexport type Recommendation = { goal: number; prob: number; ev: number };\n\ntype Features = Record;\ntype Norm = { mu: Record; sigma: Record };\ntype Model = { w: number[]; b0: number };\n\n// ─── Feature Engineering ────────────────────────────────────────────────────\n\nfunction tailMean(arr: number[], n: number): number {\n const t = arr.slice(-n);\n return t.length ? t.reduce((a, b) => a + b, 0) / t.length : 0;\n}\n\nfunction signedStreak(arr: number[]): number {\n let s = 0;\n for (let i = arr.length - 1; i >= 0; i--) {\n if (arr[i] === 1) { if (s >= 0) s++; else break; }\n else if (arr[i] === 0) { if (s <= 0) s--; else break; }\n else break;\n }\n return s;\n}\n\nfunction weekday(dateStr: string): number {\n return (new Date(dateStr + \"T12:00:00Z\").getUTCDay() + 6) % 7;\n}\n\nconst DOW = [\"mon\", \"tue\", \"wed\", \"thu\", \"fri\", \"sat\", \"sun\"];\n\nexport const FEATURE_KEYS = [\n \"goal\", \"prev_goal\", \"goal_delta\", \"goal_ma7\",\n \"prev_ach\", \"streak\", \"roll7\", \"roll14\",\n \"prev_effort\", \"effort_ma7\",\n \"dow_sin\", \"dow_cos\",\n ...DOW.map(d => `dow_${d}`),\n];\n\nfunction buildFeatures(history: HistoryRow[], goalCandidate: number, targetDate: string): Features {\n const past = history.filter(r => r.date < targetDate);\n const w = weekday(targetDate);\n const dowOneHot: Record = {};\n DOW.forEach((d, i) => dowOneHot[`dow_${d}`] = i === w ? 1 : 0);\n\n if (past.length === 0) {\n const f: Features = {};\n FEATURE_KEYS.forEach(k => f[k] = 0);\n f.goal = goalCandidate;\n f.dow_sin = Math.sin(2 * Math.PI * w / 7);\n f.dow_cos = Math.cos(2 * Math.PI * w / 7);\n Object.assign(f, dowOneHot);\n return f;\n }\n\n const prev = past[past.length - 1];\n const goals = past.map(r => r.goal);\n const achs = past.map(r => r.achieved);\n const efforts = past.map(r => {\n if (r.actual == null || r.goal <= 0) return 0;\n return Math.min(1.5, r.actual / r.goal);\n });\n\n return {\n goal: goalCandidate,\n prev_goal: prev.goal,\n goal_delta: goalCandidate - prev.goal,\n goal_ma7: tailMean(goals, 7),\n prev_ach: prev.achieved,\n streak: signedStreak(achs),\n roll7: tailMean(achs, 7),\n roll14: tailMean(achs, 14),\n prev_effort: efforts[efforts.length - 1] ?? 0,\n effort_ma7: tailMean(efforts, 7),\n dow_sin: Math.sin(2 * Math.PI * w / 7),\n dow_cos: Math.cos(2 * Math.PI * w / 7),\n ...dowOneHot,\n };\n}\n\n// ─── Logistic Regression ────────────────────────────────────────────────────\n// Gradient descent with L2 regularization, z-score normalization.\n\nfunction fitNorm(data: { feat: Features; y: number }[]): Norm {\n const mu: Record = {};\n const sigma: Record = {};\n for (const k of FEATURE_KEYS) {\n const arr = data.map(d => d.feat[k] ?? 0);\n const m = arr.reduce((a, b) => a + b, 0) / arr.length;\n const v = arr.reduce((a, b) => a + (b - m) ** 2, 0) / arr.length;\n mu[k] = m;\n sigma[k] = Math.sqrt(v) || 1;\n }\n return { mu, sigma };\n}\n\nfunction standardize(feat: Features, norm: Norm): number[] {\n return FEATURE_KEYS.map(k => ((feat[k] ?? 0) - norm.mu[k]) / norm.sigma[k]);\n}\n\nfunction train(data: { feat: Features; y: number }[], norm: Norm): Model {\n const n = data.length;\n const p = FEATURE_KEYS.length;\n const w = new Array(p).fill(0);\n let b0 = 0;\n const lr = 0.1, l2 = 0.02, iters = 500;\n\n const X = data.map(d => standardize(d.feat, norm));\n const Y = data.map(d => d.y);\n\n for (let iter = 0; iter < iters; iter++) {\n let gb0 = 0;\n const gw = new Array(p).fill(0);\n for (let i = 0; i < n; i++) {\n const z = X[i].reduce((s, x, j) => s + w[j] * x, 0) + b0;\n const pred = 1 / (1 + Math.exp(-z));\n const diff = pred - Y[i];\n gb0 += diff;\n for (let j = 0; j < p; j++) gw[j] += diff * X[i][j];\n }\n b0 -= (lr * gb0) / n;\n for (let j = 0; j < p; j++) w[j] -= lr * (gw[j] / n + l2 * w[j]);\n }\n return { w, b0 };\n}\n\nfunction predictProb(feat: Features, model: Model, norm: Norm): number {\n const x = standardize(feat, norm);\n const z = x.reduce((s, v, j) => s + model.w[j] * v, 0) + model.b0;\n return Math.max(0.01, Math.min(0.99, 1 / (1 + Math.exp(-z))));\n}\n\n// ─── Public API ─────────────────────────────────────────────────────────────\n\n/**\n * Recommend the optimal goal for a target date.\n * Returns null if there's not enough history (<10 days).\n */\nexport function recommend(\n history: HistoryRow[],\n targetDate: string,\n goalMin = 1,\n goalMax = 50,\n): { best: Recommendation; sweep: Recommendation[] } | null {\n const dataset: { feat: Features; y: number }[] = [];\n for (let i = 1; i < history.length; i++) {\n const r = history[i];\n if (r.date >= targetDate) break;\n dataset.push({\n feat: buildFeatures(history.slice(0, i), r.goal, r.date),\n y: r.achieved,\n });\n }\n\n if (dataset.length < 10) return null;\n\n const norm = fitNorm(dataset);\n const model = train(dataset, norm);\n\n const sweep: Recommendation[] = [];\n for (let g = goalMin; g <= goalMax; g++) {\n const feat = buildFeatures(history, g, targetDate);\n const prob = predictProb(feat, model, norm);\n sweep.push({ goal: g, prob, ev: prob * g });\n }\n sweep.sort((a, b) => b.ev - a.ev);\n return { best: sweep[0], sweep };\n}\n\n/**\n * Explain why a specific goal was chosen — returns per-feature contributions.\n * Positive weight = feature pushes toward success. Negative = pushes away.\n */\nexport function explain(\n history: HistoryRow[],\n goal: number,\n targetDate: string,\n): { prob: number; ev: number; contributions: { feature: string; value: number; weight: number }[] } | null {\n const dataset: { feat: Features; y: number }[] = [];\n for (let i = 1; i < history.length; i++) {\n const r = history[i];\n if (r.date >= targetDate) break;\n dataset.push({\n feat: buildFeatures(history.slice(0, i), r.goal, r.date),\n y: r.achieved,\n });\n }\n if (dataset.length < 10) return null;\n\n const norm = fitNorm(dataset);\n const model = train(dataset, norm);\n const feat = buildFeatures(history, goal, targetDate);\n const prob = predictProb(feat, model, norm);\n const x = standardize(feat, norm);\n\n return {\n prob,\n ev: prob * goal,\n contributions: FEATURE_KEYS.map((k, i) => ({\n feature: k,\n value: feat[k] ?? 0,\n weight: x[i] * model.w[i],\n })).sort((a, b) => Math.abs(b.weight) - Math.abs(a.weight)),\n };\n}\n","deleted","label",[100],"permalink","https://www.val.town/x/dcm31/ev-goal-optimizer/code/optimizer.ts","extension","ts","actionData","errors"]