[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8},"actionData","errors","root",{"_107":108,"_109":110},"routes/embed.x.$handle.$name.$",{"_9":10,"_11":12,"_13":14,"_15":16,"_17":18,"_19":20,"_21":-5},"handle","joaquinrubio","name","predmarket-monitor","path","llm-match.ts","project",{"_22":45,"_11":12,"_33":46,"_47":-5,"_48":42,"_49":50,"_51":50,"_52":53,"_54":-5,"_55":56,"_57":58,"_59":32,"_60":-5,"_61":-5,"_62":63,"_64":65,"_66":67,"_68":69,"_70":71,"_72":-5,"_73":74,"_75":74,"_76":77,"_78":77,"_79":77,"_80":74,"_81":74,"_82":83,"_84":85,"_86":87,"_88":42},"file",{"_22":23,"_11":14,"_24":25,"_26":-5,"_27":28,"_29":30,"_31":32,"_33":34,"_35":36,"_37":38,"_39":40,"_41":42,"_43":-5,"_13":44},"permalink","https://www.val.town/x/joaquinrubio/predmarket-monitor/code/llm-match.ts","extension","id","78c94703-df83-49a7-a1c8-0f3b8f7a1bea","fileId","ee9eef84-b258-11f1-9951-1607ee4eb77e","parentId","branchId","f1c2802b-7a05-402a-8e59-0a360e509b53","branchVersion",44,"latest",true,"createdAt",["D",1789623249708],"version",1,"fileType","script","content","// llm-match.ts — the LLM judge tier of the market matcher.\n//\n// The deterministic layer (matcher.ts) scans ~27k Kalshi markets for free and\n// hands over a short candidate list. This module asks a small, cheap model one\n// narrow question per Polymarket market: which candidate refers to the same\n// real-world event, and what KIND of relationship is it?\n//\n// Design notes:\n// - PROVIDER-AGNOSTIC. \"valtown\" uses Val Town's built-in OpenAI proxy: no API\n// key, no billing, nano-class model, ~10 requests/min. \"anthropic\" uses\n// Claude Haiku and needs ANTHROPIC_API_KEY. Default is valtown (free); if the\n// key exists, anthropic is preferred. Set explicitly to A/B the two.\n// - GRADED, not binary. A Kalshi market that isn't identical can still inform\n// us. \"same\" = tradeable 1:1. \"nested\" = a sub-option or threshold inside the\n// same question, still directly informative. \"proxy\" = correlated but\n// materially different, corroboration only, never a literal price\n// comparison. \"unrelated\" = drop.\n// - Reports SIDE ALIGNMENT (does Kalshi YES mean the same thing as Polymarket\n// YES), which is what lets us compute an honest price gap.\n// - It NEVER decides a trade. It classifies markets. The deterministic layer\n// still owns every trading decision.\n// - Degrades safely: any failure returns null and the caller falls back to\n// deterministic-only matching.\n// - Judgments are cached in SQLite (keyed by provider) so repeated 6h runs\n// don't re-pay, and so the two providers can be compared on the same pairs.\n\nimport { sqlite } from \"https://esm.town/v/std/sqlite/main.ts\";\nimport type { KMarket } from \"./matcher.ts\";\n\nconst HAIKU = \"claude-haiku-4-5-20251001\";\nconst NANO = \"gpt-4.5-nano\";\nconst ANTHROPIC_API = \"https://api.anthropic.com/v1/messages\";\nconst CACHE_TABLE = \"match_cache\";\n\nexport type Provider = \"valtown\" | \"anthropic\";\nexport type Relation = \"same\" | \"nested\" | \"proxy\" | \"unrelated\";\n\nexport interface Judgment {\n best: number | null; // 1-based index into the candidate list, or null\n relation: Relation;\n confidence: number; // 0..1\n aligned: boolean | null; // does Kalshi YES == Polymarket YES?\n reason: string;\n provider?: Provider;\n cached?: boolean;\n}\n\nconst SYSTEM = `You match prediction markets across two venues (Polymarket and Kalshi).\nGiven one Polymarket question and numbered Kalshi candidate markets, pick the candidate that refers to the same underlying real-world event and classify the relationship.\n\nRelations:\n- \"same\": identical question AND resolution criteria. A position in one is economically equivalent to a position in the other.\n- \"nested\": the same broader question, but one side is a specific sub-option, threshold, bracket, or component of the other (e.g. a specific rate band inside a rate decision, or one party's outcome inside a control-of-chamber market). Directly informative.\n- \"proxy\": correlated but materially different question — different entity, different bar, different mechanism, or a comparative/\"which happens first\" framing. Useful only as soft corroboration, never as a literal price comparison.\n- \"unrelated\": no useful relationship.\n\nAlso report side alignment: does the Kalshi market's YES outcome mean the same real-world outcome as the Polymarket question's YES? true if yes, false if it is inverted, null if unclear or unrelated.\n\nBe strict about \"same\". If resolution criteria or dates could differ, use \"nested\" or \"proxy\".\n\nRespond with ONLY a JSON object, no prose, no code fences:\n{\"best\": <1-based number or null>, \"relation\": \"same\"|\"nested\"|\"proxy\"|\"unrelated\", \"confidence\": <0..1>, \"aligned\": true|false|null, \"reason\": \"\"}`;\n\nfunction hash(s: string): string {\n let h = 5381;\n for (let i = 0; i < s.length; i++) h = ((h << 5) + h + s.charCodeAt(i)) | 0;\n return (h >>> 0).toString(36);\n}\n\nexport function hasAnthropicKey(): boolean {\n return Boolean(Deno.env.get(\"ANTHROPIC_API_KEY\"));\n}\n\n// Free built-in proxy by default; upgrade to Haiku automatically if a key exists.\nexport function defaultProvider(): Provider {\n return hasAnthropicKey() ? \"anthropic\" : \"valtown\";\n}\n\nasync function ensureCache(): Promise {\n await sqlite.execute(\n `CREATE TABLE IF NOT EXISTS ${CACHE_TABLE} (k TEXT PRIMARY KEY, json TEXT NOT NULL, ts TEXT NOT NULL)`,\n );\n}\n\nfunction parseJudgment(text: string): Judgment | null {\n const m = text.match(/\\{[\\s\\S]*\\}/); // tolerate stray prose or code fences\n if (!m) return null;\n try {\n const o = JSON.parse(m[0]);\n const rel = [\"same\", \"nested\", \"proxy\", \"unrelated\"].includes(o.relation) ? o.relation as Relation : \"unrelated\";\n return {\n best: typeof o.best === \"number\" ? o.best : null,\n relation: rel,\n confidence: typeof o.confidence === \"number\" ? Math.max(0, Math.min(1, o.confidence)) : 0,\n aligned: o.aligned === true ? true : o.aligned === false ? false : null,\n reason: typeof o.reason === \"string\" ? o.reason.slice(0, 200) : \"\",\n };\n } catch {\n return null;\n }\n}\n\nasync function callAnthropic(user: string): Promise {\n const key = Deno.env.get(\"ANTHROPIC_API_KEY\");\n if (!key) return null;\n try {\n const res = await fetch(ANTHROPIC_API, {\n method: \"POST\",\n headers: { \"x-api-key\": key, \"anthropic-version\": \"2023-06-01\", \"content-type\": \"application/json\" },\n body: JSON.stringify({\n model: HAIKU,\n max_tokens: 300,\n temperature: 0,\n system: SYSTEM,\n messages: [{ role: \"user\", content: user }],\n }),\n signal: AbortSignal.timeout(20000),\n });\n if (!res.ok) return null;\n const j = await res.json();\n return (j.content ?? []).map((b: any) => b?.text ?? \"\").join(\"\");\n } catch {\n return null;\n }\n}\n\n// Val Town's built-in OpenAI proxy: no API key required.\n// Rate limited (~10 req/min), so callers should throttle and lean on the cache.\nasync function callValtown(user: string): Promise {\n try {\n const { OpenAI } = await import(\"https://esm.town/v/std/openai/main.ts\");\n const openai = new OpenAI();\n const completion = await openai.chat.completions.create({\n model: NANO,\n max_completion_tokens: 300,\n messages: [\n { role: \"system\", content: SYSTEM },\n { role: \"user\", content: user },\n ],\n });\n return completion.choices?.[0]?.message?.content ?? null;\n } catch {\n return null;\n }\n}\n\n// Judge one Polymarket market against its candidate shortlist.\n// Returns null if the model is unavailable or the call fails — callers must\n// fall back to deterministic matching rather than treating null as \"unrelated\".\nexport async function judge(\n polyText: string,\n cands: KMarket[],\n opts: { useCache?: boolean; provider?: Provider } = {},\n): Promise {\n if (cands.length === 0) return null;\n const provider = opts.provider ?? defaultProvider();\n const useCache = opts.useCache !== false;\n\n const listing = cands.map((k, i) => `${i + 1}. ${k.text}`).join(\"\\n\");\n const user = `Polymarket question:\\n${polyText}\\n\\nKalshi candidates:\\n${listing}`;\n const cacheKey = provider + \":\" + hash(polyText + \"||\" + cands.map((c) => c.ticker).join(\",\"));\n\n if (useCache) {\n try {\n await ensureCache();\n const r = await sqlite.execute({ sql: `SELECT json FROM ${CACHE_TABLE} WHERE k = ?`, args: [cacheKey] });\n const row = r.rows?.[0] as any;\n if (row) {\n const raw = typeof row === \"object\" && \"json\" in row ? row.json : row[0];\n return { ...JSON.parse(raw), provider, cached: true };\n }\n } catch {\n // cache is best-effort\n }\n }\n\n const text = provider === \"anthropic\" ? await callAnthropic(user) : await callValtown(user);\n if (!text) return null;\n\n const verdict = parseJudgment(text);\n if (!verdict) return null;\n if (verdict.best != null && (verdict.best < 1 || verdict.best > cands.length)) verdict.best = null;\n verdict.provider = provider;\n\n if (useCache) {\n try {\n await sqlite.execute({\n sql: `INSERT OR REPLACE INTO ${CACHE_TABLE} (k, json, ts) VALUES (?, ?, ?)`,\n args: [cacheKey, JSON.stringify(verdict), new Date().toISOString()],\n });\n } catch {\n // best-effort\n }\n }\n return verdict;\n}\n","deleted",false,"label",[14],"7b98160e-759a-451e-ac63-bac16ce91a02",["D",1789595019450],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","Live Kalshi/Polymarket arbitrage monitor (detect-only)","imageUrl","userId","5a6bd924-f7ef-46ed-95a2-11ef633bd193","pinnedFileId","01a0ac2c-e909-715a-acf6-adfd5fd204b2","pinPreview","pinMigrated","screenshotVersion","likeCount",0,"orgId","01a0ac28-941c-777f-a124-9765c1807b9c","tursoName","7b98160e759a451eac63bac16ce91a02","tursoHostname","7b98160e759a451eac63bac16ce91a02-val-town.aws-us-east-2.turso.io","tursoMeasuredAt",["D",1789596070409],"pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_9":10,"_106":102},"org",{"_9":10,"_99":100,"_101":102,"_55":56,"_103":104,"_105":42},"branch",{"_35":89,"_11":90,"_22":28,"_91":45,"_92":-5,"_93":-5,"_94":-5,"_95":-5,"_96":-5,"_97":12,"_98":10},"isOwned",60,"main","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","displayName","Joaquin Rubio","avatar","https://lh3.googleusercontent.com/a/ACg8ocKu9e2FTAA0T_eao8qJSdToUv1ioGcVQUV0-zlGd4zBo0YpRA=s96-c","tier","free","canRestrictVals","profileImageUrl","ctx",{"_82":-5,"_140":141,"_142":143,"_144":145,"_146":42,"_147":42,"_148":42,"_149":32,"_150":-5,"_151":-7,"_152":-7,"_153":63,"_154":42,"_155":42,"_156":-5,"_157":42,"_158":42,"_159":42,"_160":-5},"ENV",{"_111":112,"_113":114,"_115":90,"_116":117,"_118":119,"_120":121,"_122":123,"_124":125,"_126":127,"_128":129,"_130":131,"_132":133,"_134":135,"_136":137,"_138":139},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","RENDER_GIT_COMMIT","c73f21da490985bff19c279eb0e09db24d688e4f","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","orgs",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]