@tangle-network/agent-eval 0.23.0 → 0.24.0

package/dist/{chunk-42I2QC2L.js → chunk-6QDKWHLS.js}

@@ -10,6 +10,17 @@ __export(benchmarks_exports, {
   routing: () => routing_exports
 });
 
+// src/benchmarks/routing/index.ts
+var routing_exports = {};
+__export(routing_exports, {
+  ROUTING_DATASET: () => ROUTING_DATASET,
+  RoutingAdapter: () => RoutingAdapter,
+  assignSplit: () => assignSplit,
+  evaluate: () => evaluate,
+  extractRouteTokens: () => extractRouteTokens,
+  loadDataset: () => loadDataset
+});
+
 // src/benchmarks/types.ts
 function fnv1a32(input) {
   let h = 2166136261;
@@ -28,17 +39,6 @@ function deterministicSplit(itemId, seed = BENCHMARK_SPLIT_SEED) {
   return "holdout";
 }
 
-// src/benchmarks/routing/index.ts
-var routing_exports = {};
-__export(routing_exports, {
-  ROUTING_DATASET: () => ROUTING_DATASET,
-  RoutingAdapter: () => RoutingAdapter,
-  assignSplit: () => assignSplit,
-  evaluate: () => evaluate,
-  extractRouteTokens: () => extractRouteTokens,
-  loadDataset: () => loadDataset
-});
-
 // src/benchmarks/routing/dataset.ts
 var ROUTING_DATASET = [
   {
@@ -174,11 +174,15 @@ var ROUTING_DATASET = [
 // src/benchmarks/routing/index.ts
 var RoutingAdapter = class {
   async loadDataset(split) {
-    return ROUTING_DATASET.map((item) => ({ id: item.id, payload: item })).filter((it) => assignSplitImpl(it.id) === split);
+    return ROUTING_DATASET.map((item) => ({ id: item.id, payload: item })).filter(
+      (it) => assignSplitImpl(it.id) === split
+    );
   }
   async evaluate(item, response) {
     const tokens = extractRouteTokens(response);
-    const correct = new Set([item.payload.route, ...item.payload.synonyms].map((s) => s.toLowerCase()));
+    const correct = new Set(
+      [item.payload.route, ...item.payload.synonyms].map((s) => s.toLowerCase())
+    );
     const hardNeg = new Set(item.payload.hardNegatives.map((s) => s.toLowerCase()));
     const firstMatch = tokens.find((t) => correct.has(t.toLowerCase())) ?? null;
     const firstHardNeg = tokens.find((t) => hardNeg.has(t.toLowerCase())) ?? null;
@@ -216,4 +220,4 @@ export {
   routing_exports,
   benchmarks_exports
 };
-//# sourceMappingURL=chunk-42I2QC2L.js.map
+//# sourceMappingURL=chunk-6QDKWHLS.js.map

package/dist/chunk-6QDKWHLS.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/benchmarks/index.ts","../src/benchmarks/routing/index.ts","../src/benchmarks/types.ts","../src/benchmarks/routing/dataset.ts"],"sourcesContent":["/**\n * Reference benchmark wrappers — entry point.\n *\n * Core surface (exported here):\n *   - The `BenchmarkAdapter` contract.\n *   - `deterministicSplit` + `BENCHMARK_SPLIT_SEED` for split assignment.\n *   - `routing` — synthetic 16-task router benchmark. The only novel\n *     benchmark we built; ships in the package.\n *\n * Example wrappers (under `examples/benchmarks/`, NOT in the bundle):\n *   - `gsm8k`         — exact-match math reasoning (HF mirror, dataset\n *                       not bundled).\n *   - `swebench-lite` — 30-instance SWE-Bench subset via an external\n *                       grader command.\n *\n * The example wrappers are reference implementations of `BenchmarkAdapter`.\n * Read them, copy them, adapt them. They're intentionally not in the main\n * entry — every team will configure them differently.\n */\n\nexport * as routing from './routing/index'\nexport type {\n  BenchmarkAdapter,\n  BenchmarkDatasetItem,\n  BenchmarkEvaluation,\n} from './types'\nexport { BENCHMARK_SPLIT_SEED, deterministicSplit } from './types'\n","/**\n * Routing benchmark — synthetic, dependency-free, ships in the\n * package. 16 cross-category items in `dataset.ts`. See\n * `routing/README.md` for the format.\n *\n * `evaluate` does case-insensitive exact match against the canonical\n * route plus declared synonyms. The first valid route token in the\n * response wins; everything else is ignored. Wrong answers also\n * report whether they hit a hard negative — useful when triaging\n * \"always picks the popular route\" failure modes.\n */\n\nimport type { RunSplitTag } from '../../run-record'\nimport type { BenchmarkAdapter, BenchmarkDatasetItem, BenchmarkEvaluation } from '../types'\nimport { deterministicSplit } from '../types'\nimport { ROUTING_DATASET, type RoutingItem } from './dataset'\n\nexport type { RoutingItem }\nexport type RoutingPayload = RoutingItem\nexport type RoutingDatasetItem = BenchmarkDatasetItem<RoutingPayload>\n\nclass RoutingAdapter implements BenchmarkAdapter<RoutingDatasetItem, RoutingPayload> {\n  async loadDataset(split: RunSplitTag): Promise<RoutingDatasetItem[]> {\n    return ROUTING_DATASET.map((item) => ({ id: item.id, payload: item })).filter(\n      (it) => assignSplitImpl(it.id) === split,\n    )\n  }\n\n  async evaluate(item: RoutingDatasetItem, response: string): Promise<BenchmarkEvaluation> {\n    const tokens = extractRouteTokens(response)\n    const correct = new Set<string>(\n      [item.payload.route, ...item.payload.synonyms].map((s) => s.toLowerCase()),\n    )\n    const hardNeg = new Set<string>(item.payload.hardNegatives.map((s) => s.toLowerCase()))\n    const firstMatch = tokens.find((t) => correct.has(t.toLowerCase())) ?? null\n    const firstHardNeg = tokens.find((t) => hardNeg.has(t.toLowerCase())) ?? null\n    const score = firstMatch ? 1 : 0\n    return {\n      score,\n      raw: {\n        firstToken: tokens[0] ?? null,\n        matchedRoute: firstMatch,\n        hitHardNegative: Boolean(firstHardNeg),\n        hardNegativeRoute: firstHardNeg,\n        category: item.payload.category,\n      },\n    }\n  }\n\n  assignSplit(itemId: string): RunSplitTag {\n    return assignSplitImpl(itemId)\n  }\n}\n\nfunction assignSplitImpl(itemId: string): RunSplitTag {\n  return deterministicSplit(`routing::${itemId}`)\n}\n\n/**\n * Pull route-shaped tokens out of a model response. Routes look like\n * `category.action` (`fs.write`, `chat.reply`). Bare alphanumerics\n * are not routes, but `category.action` patterns are robust to most\n * model wrappers (JSON output, prose explanations, code fences).\n */\nexport function extractRouteTokens(response: string): string[] {\n  const matches = response.match(/[a-z][a-z0-9_]*\\.[a-z][a-z0-9_]*/gi)\n  return matches ?? []\n}\n\nconst adapter = new RoutingAdapter()\n\nexport const loadDataset = adapter.loadDataset.bind(adapter)\nexport const evaluate = adapter.evaluate.bind(adapter)\nexport const assignSplit = adapter.assignSplit.bind(adapter)\nexport { ROUTING_DATASET, RoutingAdapter }\n","/**\n * Shared types for the reference benchmark wrappers under\n * `src/benchmarks/`. Each wrapper exports the three functions in\n * `BenchmarkAdapter` plus its own typed `DatasetItem` shape.\n */\n\nimport type { RunSplitTag } from '../run-record'\n\nexport interface BenchmarkDatasetItem<TPayload = unknown> {\n  /** Stable dataset-local item id (used for split assignment + paper\n   *  references). Unique within a benchmark. */\n  id: string\n  /** Free-form payload. Each benchmark defines its own shape. */\n  payload: TPayload\n}\n\nexport interface BenchmarkEvaluation {\n  /** [0, 1] score for the response on this item. Exact-match\n   *  benchmarks use 0/1; partial-credit benchmarks may return\n   *  fractional values. */\n  score: number\n  /** Optional bag of raw scoring signals — e.g. parsed numeric\n   *  answer, regex match, judge sub-scores. */\n  raw: Record<string, unknown>\n}\n\n/** Common signature implemented by every adapter under `src/benchmarks/*`. */\n// `TPayload` is the per-item payload type; `_TItem` is preserved for\n// downstream type-narrowing extensions (a richer `BenchmarkDatasetItem`\n// subclass that adds e.g. provenance metadata) but is intentionally\n// unused here. `noUnusedLocals` requires the leading underscore.\nexport interface BenchmarkAdapter<_TItem = unknown, TPayload = unknown> {\n  /** Load the dataset for the given split. May hit the network on\n   *  first call but should be cache-friendly. Adapters that don't\n   *  ship the dataset itself MUST throw a clearly-marked error\n   *  pointing the caller at the loader script. */\n  loadDataset(split: RunSplitTag): Promise<BenchmarkDatasetItem<TPayload>[]>\n  /** Score a single response. Pure with respect to the inputs. */\n  evaluate(item: BenchmarkDatasetItem<TPayload>, response: string): Promise<BenchmarkEvaluation>\n  /** Deterministic split assignment via item id hashing. The\n   *  fraction of items in each split is implementation-defined but\n   *  MUST be stable across processes and platforms. */\n  assignSplit(itemId: string): RunSplitTag\n}\n\n// ── Deterministic split assignment ───────────────────────────────────\n\n/**\n * 32-bit FNV-1a hash. Stable, allocation-free, deterministic across\n * runtimes. We use it to assign items to splits rather than depending\n * on a polyfilled crypto.subtle path.\n */\nfunction fnv1a32(input: string): number {\n  let h = 0x811c9dc5\n  for (let i = 0; i < input.length; i++) {\n    h ^= input.charCodeAt(i) & 0xff\n    h = (h + ((h << 1) + (h << 4) + (h << 7) + (h << 8) + (h << 24))) >>> 0\n  }\n  return h >>> 0\n}\n\n/** Split-assignment seed shared across all benchmarks. Bumping this\n *  value reshuffles every split — do NOT do that lightly. */\nexport const BENCHMARK_SPLIT_SEED = 'agent-eval-v1'\n\n/**\n * Assign an item id to one of `'search' | 'dev' | 'holdout'` using a\n * stable 32-bit hash of `${seed}::${id}`. Default proportions:\n *\n *   search: 60%   (optimization-readable)\n *   dev:    20%   (held-out for tuning, leak-on-purpose during dev)\n *   holdout:20%   (paper-grade held-out, gated reads)\n */\nexport function deterministicSplit(\n  itemId: string,\n  seed: string = BENCHMARK_SPLIT_SEED,\n): RunSplitTag {\n  const h = fnv1a32(`${seed}::${itemId}`)\n  const pos = h / 0x100000000\n  if (pos < 0.6) return 'search'\n  if (pos < 0.8) return 'dev'\n  return 'holdout'\n}\n","/**\n * Synthetic routing dataset. 16 tasks across 4 categories. Used as a\n * deterministic, dependency-free benchmark for any router that maps a\n * natural-language request to one of a fixed set of route labels.\n *\n * Format (see `routing/README.md` for prose):\n *\n *   {\n *     id:           stable per-task ID (matches across processes).\n *     category:     one of the four route labels.\n *     prompt:       the user-facing request the router must classify.\n *     route:        the ground-truth route the router should pick.\n *     synonyms:     other strings that count as a correct answer.\n *     hardNegatives:close-but-wrong route labels — used to detect the\n *                   \"always picks the popular route\" failure mode.\n *   }\n *\n * The four categories are intentionally cross-domain (file ops,\n * math, search, conversation) so a router that collapses to one\n * category is easy to spot.\n */\n\nexport interface RoutingItem {\n  id: string\n  category: 'file' | 'math' | 'search' | 'chat'\n  prompt: string\n  /** Canonical correct route label. */\n  route: string\n  /** Alternate route labels that also count as correct. */\n  synonyms: string[]\n  /** Wrong-but-tempting route labels (for analysis, not grading). */\n  hardNegatives: string[]\n}\n\nexport const ROUTING_DATASET: RoutingItem[] = [\n  {\n    id: 'file_001',\n    category: 'file',\n    prompt: 'Save the meeting notes to /tmp/notes-2025-04.md as markdown.',\n    route: 'fs.write',\n    synonyms: ['filesystem.write', 'write_file'],\n    hardNegatives: ['fs.read', 'chat.reply'],\n  },\n  {\n    id: 'file_002',\n    category: 'file',\n    prompt: 'Read the contents of /etc/hosts and summarize the entries.',\n    route: 'fs.read',\n    synonyms: ['filesystem.read', 'read_file'],\n    hardNegatives: ['fs.write', 'search.web'],\n  },\n  {\n    id: 'file_003',\n    category: 'file',\n    prompt: 'List every Python file under src/ recursively.',\n    route: 'fs.list',\n    synonyms: ['filesystem.list', 'list_files'],\n    hardNegatives: ['fs.read', 'search.code'],\n  },\n  {\n    id: 'file_004',\n    category: 'file',\n    prompt: 'Delete the cached build at .turbo/cache.',\n    route: 'fs.delete',\n    synonyms: ['filesystem.delete', 'remove_file'],\n    hardNegatives: ['fs.write', 'fs.list'],\n  },\n  {\n    id: 'math_001',\n    category: 'math',\n    prompt: 'What is the integral of 3x^2 + 2x from 0 to 5?',\n    route: 'math.integral',\n    synonyms: ['calculator.integral', 'math.solve'],\n    hardNegatives: ['math.derivative', 'chat.reply'],\n  },\n  {\n    id: 'math_002',\n    category: 'math',\n    prompt: 'Compute the derivative of sin(x) * cos(x).',\n    route: 'math.derivative',\n    synonyms: ['calculator.derivative', 'math.solve'],\n    hardNegatives: ['math.integral', 'math.algebra'],\n  },\n  {\n    id: 'math_003',\n    category: 'math',\n    prompt: 'Solve 2x + 7 = 19 for x.',\n    route: 'math.algebra',\n    synonyms: ['calculator.algebra', 'math.solve'],\n    hardNegatives: ['math.derivative', 'math.integral'],\n  },\n  {\n    id: 'math_004',\n    category: 'math',\n    prompt: 'What is the prime factorization of 360?',\n    route: 'math.numbertheory',\n    synonyms: ['calculator.factor', 'math.solve'],\n    hardNegatives: ['math.algebra', 'search.web'],\n  },\n  {\n    id: 'search_001',\n    category: 'search',\n    prompt: 'Find recent papers on agent prompt optimization with held-out promotion gates.',\n    route: 'search.web',\n    synonyms: ['web.search', 'search.papers'],\n    hardNegatives: ['search.code', 'chat.reply'],\n  },\n  {\n    id: 'search_002',\n    category: 'search',\n    prompt: 'Search the codebase for every call site of `runProposeReview`.',\n    route: 'search.code',\n    synonyms: ['code.search', 'grep'],\n    hardNegatives: ['search.web', 'fs.read'],\n  },\n  {\n    id: 'search_003',\n    category: 'search',\n    prompt: 'What is the latest release of the Tangle network on GitHub?',\n    route: 'search.web',\n    synonyms: ['web.search', 'github.releases'],\n    hardNegatives: ['search.code', 'chat.reply'],\n  },\n  {\n    id: 'search_004',\n    category: 'search',\n    prompt: 'Find all TODO comments in the agent-eval src tree.',\n    route: 'search.code',\n    synonyms: ['code.search', 'grep'],\n    hardNegatives: ['search.web', 'fs.list'],\n  },\n  {\n    id: 'chat_001',\n    category: 'chat',\n    prompt: 'Hi there, how are you doing today?',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply'],\n    hardNegatives: ['search.web', 'fs.read'],\n  },\n  {\n    id: 'chat_002',\n    category: 'chat',\n    prompt: 'Please explain the difference between an LLM and a foundation model.',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply', 'qa.answer'],\n    hardNegatives: ['search.web', 'math.algebra'],\n  },\n  {\n    id: 'chat_003',\n    category: 'chat',\n    prompt: 'Tell me a short joke about distributed systems.',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply'],\n    hardNegatives: ['search.web', 'fs.read'],\n  },\n  {\n    id: 'chat_004',\n    category: 'chat',\n    prompt: 'Acknowledge my last message with a thumbs up.',\n    route: 'chat.reply',\n    synonyms: ['conversation.reply', 'react'],\n    hardNegatives: ['fs.write', 'search.web'],\n  },\n]\n"],"mappings":";;;;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;;;ACAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;;;ACoDA,SAAS,QAAQ,OAAuB;AACtC,MAAI,IAAI;AACR,WAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;AACrC,SAAK,MAAM,WAAW,CAAC,IAAI;AAC3B,QAAK,MAAM,KAAK,MAAM,KAAK,MAAM,KAAK,MAAM,KAAK,MAAM,KAAK,SAAU;AAAA,EACxE;AACA,SAAO,MAAM;AACf;AAIO,IAAM,uBAAuB;AAU7B,SAAS,mBACd,QACA,OAAe,sBACF;AACb,QAAM,IAAI,QAAQ,GAAG,IAAI,KAAK,MAAM,EAAE;AACtC,QAAM,MAAM,IAAI;AAChB,MAAI,MAAM,IAAK,QAAO;AACtB,MAAI,MAAM,IAAK,QAAO;AACtB,SAAO;AACT;;;AChDO,IAAM,kBAAiC;AAAA,EAC5C;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,oBAAoB,YAAY;AAAA,IAC3C,eAAe,CAAC,WAAW,YAAY;AAAA,EACzC;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,mBAAmB,WAAW;AAAA,IACzC,eAAe,CAAC,YAAY,YAAY;AAAA,EAC1C;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,mBAAmB,YAAY;AAAA,IAC1C,eAAe,CAAC,WAAW,aAAa;AAAA,EAC1C;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,qBAAqB,aAAa;AAAA,IAC7C,eAAe,CAAC,YAAY,SAAS;AAAA,EACvC;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,uBAAuB,YAAY;AAAA,IAC9C,eAAe,CAAC,mBAAmB,YAAY;AAAA,EACjD;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,yBAAyB,YAAY;AAAA,IAChD,eAAe,CAAC,iBAAiB,cAAc;AAAA,EACjD;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,sBAAsB,YAAY;AAAA,IAC7C,eAAe,CAAC,mBAAmB,eAAe;AAAA,EACpD;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,qBAAqB,YAAY;AAAA,IAC5C,eAAe,CAAC,gBAAgB,YAAY;AAAA,EAC9C;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,cAAc,eAAe;AAAA,IACxC,eAAe,CAAC,eAAe,YAAY;AAAA,EAC7C;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,eAAe,MAAM;AAAA,IAChC,eAAe,CAAC,cAAc,SAAS;AAAA,EACzC;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,cAAc,iBAAiB;AAAA,IAC1C,eAAe,CAAC,eAAe,YAAY;AAAA,EAC7C;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,eAAe,MAAM;AAAA,IAChC,eAAe,CAAC,cAAc,SAAS;AAAA,EACzC;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,oBAAoB;AAAA,IAC/B,eAAe,CAAC,cAAc,SAAS;AAAA,EACzC;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,sBAAsB,WAAW;AAAA,IAC5C,eAAe,CAAC,cAAc,cAAc;AAAA,EAC9C;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,oBAAoB;AAAA,IAC/B,eAAe,CAAC,cAAc,SAAS;AAAA,EACzC;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,UAAU;AAAA,IACV,QAAQ;AAAA,IACR,OAAO;AAAA,IACP,UAAU,CAAC,sBAAsB,OAAO;AAAA,IACxC,eAAe,CAAC,YAAY,YAAY;AAAA,EAC1C;AACF;;;AF9IA,IAAM,iBAAN,MAAqF;AAAA,EACnF,MAAM,YAAY,OAAmD;AACnE,WAAO,gBAAgB,IAAI,CAAC,UAAU,EAAE,IAAI,KAAK,IAAI,SAAS,KAAK,EAAE,EAAE;AAAA,MACrE,CAAC,OAAO,gBAAgB,GAAG,EAAE,MAAM;AAAA,IACrC;AAAA,EACF;AAAA,EAEA,MAAM,SAAS,MAA0B,UAAgD;AACvF,UAAM,SAAS,mBAAmB,QAAQ;AAC1C,UAAM,UAAU,IAAI;AAAA,MAClB,CAAC,KAAK,QAAQ,OAAO,GAAG,KAAK,QAAQ,QAAQ,EAAE,IAAI,CAAC,MAAM,EAAE,YAAY,CAAC;AAAA,IAC3E;AACA,UAAM,UAAU,IAAI,IAAY,KAAK,QAAQ,cAAc,IAAI,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC;AACtF,UAAM,aAAa,OAAO,KAAK,CAAC,MAAM,QAAQ,IAAI,EAAE,YAAY,CAAC,CAAC,KAAK;AACvE,UAAM,eAAe,OAAO,KAAK,CAAC,MAAM,QAAQ,IAAI,EAAE,YAAY,CAAC,CAAC,KAAK;AACzE,UAAM,QAAQ,aAAa,IAAI;AAC/B,WAAO;AAAA,MACL;AAAA,MACA,KAAK;AAAA,QACH,YAAY,OAAO,CAAC,KAAK;AAAA,QACzB,cAAc;AAAA,QACd,iBAAiB,QAAQ,YAAY;AAAA,QACrC,mBAAmB;AAAA,QACnB,UAAU,KAAK,QAAQ;AAAA,MACzB;AAAA,IACF;AAAA,EACF;AAAA,EAEA,YAAY,QAA6B;AACvC,WAAO,gBAAgB,MAAM;AAAA,EAC/B;AACF;AAEA,SAAS,gBAAgB,QAA6B;AACpD,SAAO,mBAAmB,YAAY,MAAM,EAAE;AAChD;AAQO,SAAS,mBAAmB,UAA4B;AAC7D,QAAM,UAAU,SAAS,MAAM,oCAAoC;AACnE,SAAO,WAAW,CAAC;AACrB;AAEA,IAAM,UAAU,IAAI,eAAe;AAE5B,IAAM,cAAc,QAAQ,YAAY,KAAK,OAAO;AACpD,IAAM,WAAW,QAAQ,SAAS,KAAK,OAAO;AAC9C,IAAM,cAAc,QAAQ,YAAY,KAAK,OAAO;","names":[]}

package/dist/chunk-I4MBDTY5.js

@@ -0,0 +1,272 @@
+import {
+  ValidationError
+} from "./chunk-NG236HPC.js";
+
+// src/statistics.ts
+var INVERTED_DIMENSIONS = /* @__PURE__ */ new Set(["hallucination", "false_confidence", "worst_failure"]);
+function normalizeScores(scores) {
+  return scores.map((s) => {
+    if (INVERTED_DIMENSIONS.has(s.dimension)) {
+      return s;
+    }
+    return s;
+  });
+}
+function weightedMean(scores) {
+  if (scores.length === 0) return 0;
+  let totalWeight = 0;
+  let weightedSum = 0;
+  for (const { score, weight } of scores) {
+    const w = weight ?? 1;
+    weightedSum += score * w;
+    totalWeight += w;
+  }
+  return totalWeight > 0 ? weightedSum / totalWeight : 0;
+}
+function confidenceInterval(scores, confidence = 0.95) {
+  if (scores.length === 0) return { mean: 0, lower: 0, upper: 0 };
+  if (scores.length === 1) return { mean: scores[0], lower: scores[0], upper: scores[0] };
+  const n = scores.length;
+  const mean = scores.reduce((a, b) => a + b, 0) / n;
+  const B = 1e3;
+  const bootstrapMeans = [];
+  for (let i = 0; i < B; i++) {
+    let sum = 0;
+    for (let j = 0; j < n; j++) {
+      sum += scores[Math.floor(Math.random() * n)];
+    }
+    bootstrapMeans.push(sum / n);
+  }
+  bootstrapMeans.sort((a, b) => a - b);
+  const alpha = 1 - confidence;
+  const lowerIdx = Math.floor(alpha / 2 * B);
+  const upperIdx = Math.floor((1 - alpha / 2) * B) - 1;
+  return {
+    mean,
+    lower: bootstrapMeans[lowerIdx],
+    upper: bootstrapMeans[Math.min(upperIdx, B - 1)]
+  };
+}
+function interRaterReliability(judgeScores) {
+  if (judgeScores.length < 2) return 1;
+  const dimensionMap = /* @__PURE__ */ new Map();
+  for (const judgeSet of judgeScores) {
+    for (const s of judgeSet) {
+      if (!dimensionMap.has(s.dimension)) dimensionMap.set(s.dimension, []);
+      const arr = dimensionMap.get(s.dimension);
+      if (arr.length === 0 || arr[arr.length - 1].length >= judgeScores.length) {
+        arr.push([s.score]);
+      } else {
+        arr[arr.length - 1].push(s.score);
+      }
+    }
+  }
+  const allValues = [];
+  const pairDiffs = [];
+  for (const items of dimensionMap.values()) {
+    for (const ratings of items) {
+      if (ratings.length < 2) continue;
+      for (const v of ratings) allValues.push(v);
+      for (let i = 0; i < ratings.length; i++) {
+        for (let j = i + 1; j < ratings.length; j++) {
+          pairDiffs.push((ratings[i] - ratings[j]) ** 2);
+        }
+      }
+    }
+  }
+  if (pairDiffs.length === 0 || allValues.length < 2) return 1;
+  const observedDisagreement = pairDiffs.reduce((a, b) => a + b, 0) / pairDiffs.length;
+  let expectedDisagreement = 0;
+  let expectedCount = 0;
+  for (let i = 0; i < allValues.length; i++) {
+    for (let j = i + 1; j < allValues.length; j++) {
+      expectedDisagreement += (allValues[i] - allValues[j]) ** 2;
+      expectedCount++;
+    }
+  }
+  expectedDisagreement = expectedCount > 0 ? expectedDisagreement / expectedCount : 0;
+  if (expectedDisagreement === 0) return 1;
+  return 1 - observedDisagreement / expectedDisagreement;
+}
+function mannWhitneyU(a, b) {
+  if (a.length === 0 || b.length === 0) return { u: 0, p: 1 };
+  const n1 = a.length;
+  const n2 = b.length;
+  const combined = [
+    ...a.map((v) => ({ v, group: "a" })),
+    ...b.map((v) => ({ v, group: "b" }))
+  ].sort((x, y) => x.v - y.v);
+  const ranks = new Array(combined.length);
+  let i = 0;
+  while (i < combined.length) {
+    let j = i;
+    while (j < combined.length && combined[j].v === combined[i].v) j++;
+    const avgRank = (i + 1 + j) / 2;
+    for (let k = i; k < j; k++) ranks[k] = avgRank;
+    i = j;
+  }
+  let r1 = 0;
+  for (let k = 0; k < combined.length; k++) {
+    if (combined[k].group === "a") r1 += ranks[k];
+  }
+  const u1 = r1 - n1 * (n1 + 1) / 2;
+  const u2 = n1 * n2 - u1;
+  const u = Math.min(u1, u2);
+  const mu = n1 * n2 / 2;
+  const sigma = Math.sqrt(n1 * n2 * (n1 + n2 + 1) / 12);
+  if (sigma === 0) return { u, p: 1 };
+  const z = Math.abs(u - mu) / sigma;
+  const p = 2 * (1 - normalCdf(z));
+  return { u, p };
+}
+function partialCredit(current, target) {
+  if (target <= 0) return 1;
+  return Math.min(1, Math.max(0, current / target));
+}
+function pairedTTest(before, after) {
+  if (before.length !== after.length) {
+    throw new ValidationError(
+      `pairedTTest: unequal sample sizes (${before.length} vs ${after.length})`
+    );
+  }
+  const n = before.length;
+  if (n < 2) return { t: 0, df: 0, p: 1 };
+  const diffs = before.map((b, i) => after[i] - b);
+  const mean = diffs.reduce((a, b) => a + b, 0) / n;
+  const variance = diffs.reduce((acc, d) => acc + (d - mean) ** 2, 0) / (n - 1);
+  const se = Math.sqrt(variance / n);
+  if (se === 0) return { t: mean === 0 ? 0 : Infinity, df: n - 1, p: mean === 0 ? 1 : 0 };
+  const t = mean / se;
+  const df = n - 1;
+  const p = 2 * (1 - studentTCdf(Math.abs(t), df));
+  return { t, df, p };
+}
+function wilcoxonSignedRank(before, after) {
+  if (before.length !== after.length) {
+    throw new ValidationError(
+      `wilcoxonSignedRank: unequal sample sizes (${before.length} vs ${after.length})`
+    );
+  }
+  const diffs = before.map((b, i2) => after[i2] - b).filter((d) => d !== 0);
+  const n = diffs.length;
+  if (n < 6) return { w: 0, p: 1 };
+  const absRanks = diffs.map((d, i2) => ({ abs: Math.abs(d), sign: Math.sign(d), i: i2 })).sort((a, b) => a.abs - b.abs);
+  const ranks = new Array(n);
+  let i = 0;
+  while (i < n) {
+    let j = i;
+    while (j < n && absRanks[j].abs === absRanks[i].abs) j++;
+    const avg = (i + 1 + j) / 2;
+    for (let k = i; k < j; k++) ranks[absRanks[k].i] = avg;
+    i = j;
+  }
+  let wPlus = 0;
+  for (let k = 0; k < n; k++) if (diffs[k] > 0) wPlus += ranks[k];
+  const mean = n * (n + 1) / 4;
+  const variance = n * (n + 1) * (2 * n + 1) / 24;
+  const z = (wPlus - mean) / Math.sqrt(variance);
+  const p = 2 * (1 - normalCdf(Math.abs(z)));
+  return { w: wPlus, p };
+}
+function cohensD(a, b) {
+  if (a.length < 2 || b.length < 2) return 0;
+  const meanA = a.reduce((x, y) => x + y, 0) / a.length;
+  const meanB = b.reduce((x, y) => x + y, 0) / b.length;
+  const varA = a.reduce((acc, x) => acc + (x - meanA) ** 2, 0) / (a.length - 1);
+  const varB = b.reduce((acc, x) => acc + (x - meanB) ** 2, 0) / (b.length - 1);
+  const pooled = Math.sqrt(
+    ((a.length - 1) * varA + (b.length - 1) * varB) / (a.length + b.length - 2)
+  );
+  if (pooled === 0) return 0;
+  return (meanB - meanA) / pooled;
+}
+function studentTCdf(t, df) {
+  if (df <= 0) return 0.5;
+  if (df > 100) return normalCdf(t);
+  const x = df / (df + t * t);
+  const a = df / 2;
+  const b = 0.5;
+  const ib = incompleteBeta(x, a, b);
+  return t >= 0 ? 1 - 0.5 * ib : 0.5 * ib;
+}
+function incompleteBeta(x, a, b) {
+  if (x <= 0) return 0;
+  if (x >= 1) return 1;
+  const lnBeta = lnGamma(a) + lnGamma(b) - lnGamma(a + b);
+  const front = Math.exp(Math.log(x) * a + Math.log(1 - x) * b - lnBeta) / a;
+  const maxIter = 200;
+  const eps = 3e-7;
+  let c = 1;
+  let d = 1 - (a + b) * x / (a + 1);
+  if (Math.abs(d) < 1e-30) d = 1e-30;
+  d = 1 / d;
+  let f = d;
+  for (let m = 1; m <= maxIter; m++) {
+    const m2 = 2 * m;
+    let num = m * (b - m) * x / ((a + m2 - 1) * (a + m2));
+    d = 1 + num * d;
+    if (Math.abs(d) < 1e-30) d = 1e-30;
+    c = 1 + num / c;
+    if (Math.abs(c) < 1e-30) c = 1e-30;
+    d = 1 / d;
+    f *= d * c;
+    num = -((a + m) * (a + b + m) * x) / ((a + m2) * (a + m2 + 1));
+    d = 1 + num * d;
+    if (Math.abs(d) < 1e-30) d = 1e-30;
+    c = 1 + num / c;
+    if (Math.abs(c) < 1e-30) c = 1e-30;
+    d = 1 / d;
+    const delta = d * c;
+    f *= delta;
+    if (Math.abs(delta - 1) < eps) break;
+  }
+  return front * f;
+}
+function lnGamma(z) {
+  const g = 7;
+  const coefs = [
+    0.9999999999998099,
+    676.5203681218851,
+    -1259.1392167224028,
+    771.3234287776531,
+    -176.6150291621406,
+    12.507343278686905,
+    -0.13857109526572012,
+    9984369578019572e-21,
+    15056327351493116e-23
+  ];
+  if (z < 0.5) {
+    return Math.log(Math.PI / Math.sin(Math.PI * z)) - lnGamma(1 - z);
+  }
+  z -= 1;
+  let x = coefs[0];
+  for (let i = 1; i < g + 2; i++) x += coefs[i] / (z + i);
+  const t = z + g + 0.5;
+  return 0.5 * Math.log(2 * Math.PI) + (z + 0.5) * Math.log(t) - t + Math.log(x);
+}
+function normalCdf(x) {
+  const a1 = 0.254829592;
+  const a2 = -0.284496736;
+  const a3 = 1.421413741;
+  const a4 = -1.453152027;
+  const a5 = 1.061405429;
+  const p = 0.3275911;
+  const sign = x < 0 ? -1 : 1;
+  const absX = Math.abs(x);
+  const t = 1 / (1 + p * absX);
+  const y = 1 - ((((a5 * t + a4) * t + a3) * t + a2) * t + a1) * t * Math.exp(-absX * absX / 2);
+  return 0.5 * (1 + sign * y);
+}
+
+export {
+  normalizeScores,
+  weightedMean,
+  confidenceInterval,
+  interRaterReliability,
+  mannWhitneyU,
+  partialCredit,
+  pairedTTest,
+  wilcoxonSignedRank,
+  cohensD
+};
+//# sourceMappingURL=chunk-I4MBDTY5.js.map

package/dist/chunk-I4MBDTY5.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/statistics.ts"],"sourcesContent":["import { ValidationError } from './errors'\nimport type { JudgeScore } from './types'\n\n/** Dimensions where lower raw score = better outcome (inverted semantics) */\nconst INVERTED_DIMENSIONS = new Set(['hallucination', 'false_confidence', 'worst_failure'])\n\n/**\n * Normalize scores so all dimensions follow \"higher = better\".\n * Inverted dimensions (hallucination, false_confidence, worst_failure)\n * already use inverted scoring in the prompt (10 = no hallucination),\n * but this function ensures consistency if raw scores leak through.\n */\nexport function normalizeScores(scores: JudgeScore[]): JudgeScore[] {\n  return scores.map((s) => {\n    if (INVERTED_DIMENSIONS.has(s.dimension)) {\n      return s\n    }\n    return s\n  })\n}\n\n/** Weighted mean — falls back to uniform weights when omitted */\nexport function weightedMean(scores: { score: number; weight?: number }[]): number {\n  if (scores.length === 0) return 0\n  let totalWeight = 0\n  let weightedSum = 0\n  for (const { score, weight } of scores) {\n    const w = weight ?? 1\n    weightedSum += score * w\n    totalWeight += w\n  }\n  return totalWeight > 0 ? weightedSum / totalWeight : 0\n}\n\n/** Bootstrap confidence interval */\nexport function confidenceInterval(\n  scores: number[],\n  confidence = 0.95,\n): { mean: number; lower: number; upper: number } {\n  if (scores.length === 0) return { mean: 0, lower: 0, upper: 0 }\n  if (scores.length === 1) return { mean: scores[0]!, lower: scores[0]!, upper: scores[0]! }\n\n  const n = scores.length\n  const mean = scores.reduce((a, b) => a + b, 0) / n\n\n  const B = 1000\n  const bootstrapMeans: number[] = []\n\n  for (let i = 0; i < B; i++) {\n    let sum = 0\n    for (let j = 0; j < n; j++) {\n      sum += scores[Math.floor(Math.random() * n)]!\n    }\n    bootstrapMeans.push(sum / n)\n  }\n\n  bootstrapMeans.sort((a, b) => a - b)\n\n  const alpha = 1 - confidence\n  const lowerIdx = Math.floor((alpha / 2) * B)\n  const upperIdx = Math.floor((1 - alpha / 2) * B) - 1\n\n  return {\n    mean,\n    lower: bootstrapMeans[lowerIdx]!,\n    upper: bootstrapMeans[Math.min(upperIdx, B - 1)]!,\n  }\n}\n\n/**\n * Inter-rater reliability — simplified Krippendorff's alpha.\n *\n * Each inner array is one judge's scores for all items.\n * All arrays must have the same length (same items scored).\n */\nexport function interRaterReliability(judgeScores: JudgeScore[][]): number {\n  if (judgeScores.length < 2) return 1\n\n  // Group scores by dimension across judges\n  const dimensionMap = new Map<string, number[][]>()\n  for (const judgeSet of judgeScores) {\n    for (const s of judgeSet) {\n      if (!dimensionMap.has(s.dimension)) dimensionMap.set(s.dimension, [])\n      const arr = dimensionMap.get(s.dimension)!\n      if (arr.length === 0 || arr[arr.length - 1]!.length >= judgeScores.length) {\n        arr.push([s.score])\n      } else {\n        arr[arr.length - 1]!.push(s.score)\n      }\n    }\n  }\n\n  // Collect all paired ratings\n  const allValues: number[] = []\n  const pairDiffs: number[] = []\n\n  for (const items of dimensionMap.values()) {\n    for (const ratings of items) {\n      if (ratings.length < 2) continue\n      for (const v of ratings) allValues.push(v)\n      for (let i = 0; i < ratings.length; i++) {\n        for (let j = i + 1; j < ratings.length; j++) {\n          pairDiffs.push((ratings[i]! - ratings[j]!) ** 2)\n        }\n      }\n    }\n  }\n\n  if (pairDiffs.length === 0 || allValues.length < 2) return 1\n\n  const observedDisagreement = pairDiffs.reduce((a, b) => a + b, 0) / pairDiffs.length\n\n  // Expected disagreement from all possible pairings of values\n  let expectedDisagreement = 0\n  let expectedCount = 0\n  for (let i = 0; i < allValues.length; i++) {\n    for (let j = i + 1; j < allValues.length; j++) {\n      expectedDisagreement += (allValues[i]! - allValues[j]!) ** 2\n      expectedCount++\n    }\n  }\n  expectedDisagreement = expectedCount > 0 ? expectedDisagreement / expectedCount : 0\n\n  if (expectedDisagreement === 0) return 1\n  return 1 - observedDisagreement / expectedDisagreement\n}\n\n/**\n * Mann-Whitney U test for comparing two independent groups.\n * Returns U statistic and approximate p-value (normal approximation).\n */\nexport function mannWhitneyU(a: number[], b: number[]): { u: number; p: number } {\n  if (a.length === 0 || b.length === 0) return { u: 0, p: 1 }\n\n  const n1 = a.length\n  const n2 = b.length\n\n  // Rank all values together\n  const combined = [\n    ...a.map((v) => ({ v, group: 'a' as const })),\n    ...b.map((v) => ({ v, group: 'b' as const })),\n  ].sort((x, y) => x.v - y.v)\n\n  // Assign ranks with tie handling\n  const ranks: number[] = new Array(combined.length)\n  let i = 0\n  while (i < combined.length) {\n    let j = i\n    while (j < combined.length && combined[j]!.v === combined[i]!.v) j++\n    const avgRank = (i + 1 + j) / 2\n    for (let k = i; k < j; k++) ranks[k] = avgRank\n    i = j\n  }\n\n  // Sum ranks for group a\n  let r1 = 0\n  for (let k = 0; k < combined.length; k++) {\n    if (combined[k]!.group === 'a') r1 += ranks[k]!\n  }\n\n  const u1 = r1 - (n1 * (n1 + 1)) / 2\n  const u2 = n1 * n2 - u1\n  const u = Math.min(u1, u2)\n\n  // Normal approximation for p-value\n  const mu = (n1 * n2) / 2\n  const sigma = Math.sqrt((n1 * n2 * (n1 + n2 + 1)) / 12)\n\n  if (sigma === 0) return { u, p: 1 }\n\n  const z = Math.abs(u - mu) / sigma\n  // Two-tailed p-value from z-score (approximation)\n  const p = 2 * (1 - normalCdf(z))\n\n  return { u, p }\n}\n\n/** Partial credit: returns 0-1 ratio of current toward target */\nexport function partialCredit(current: number, target: number): number {\n  if (target <= 0) return 1\n  return Math.min(1, Math.max(0, current / target))\n}\n\n/**\n * Paired t-test — before/after measurements on the SAME items.\n * Pairing removes inter-item variance, giving tighter significance than\n * an unpaired test when comparing prompt v1 vs prompt v2 on identical\n * scenarios.\n */\nexport function pairedTTest(\n  before: number[],\n  after: number[],\n): { t: number; df: number; p: number } {\n  if (before.length !== after.length) {\n    throw new ValidationError(\n      `pairedTTest: unequal sample sizes (${before.length} vs ${after.length})`,\n    )\n  }\n  const n = before.length\n  if (n < 2) return { t: 0, df: 0, p: 1 }\n\n  const diffs = before.map((b, i) => after[i]! - b)\n  const mean = diffs.reduce((a, b) => a + b, 0) / n\n  const variance = diffs.reduce((acc, d) => acc + (d - mean) ** 2, 0) / (n - 1)\n  const se = Math.sqrt(variance / n)\n  if (se === 0) return { t: mean === 0 ? 0 : Infinity, df: n - 1, p: mean === 0 ? 1 : 0 }\n\n  const t = mean / se\n  const df = n - 1\n  const p = 2 * (1 - studentTCdf(Math.abs(t), df))\n  return { t, df, p }\n}\n\n/**\n * Wilcoxon signed-rank test — paired non-parametric alternative.\n * Use when the differences aren't normally distributed.\n */\nexport function wilcoxonSignedRank(before: number[], after: number[]): { w: number; p: number } {\n  if (before.length !== after.length) {\n    throw new ValidationError(\n      `wilcoxonSignedRank: unequal sample sizes (${before.length} vs ${after.length})`,\n    )\n  }\n  const diffs = before.map((b, i) => after[i]! - b).filter((d) => d !== 0)\n  const n = diffs.length\n  if (n < 6) return { w: 0, p: 1 }\n\n  const absRanks = diffs\n    .map((d, i) => ({ abs: Math.abs(d), sign: Math.sign(d), i }))\n    .sort((a, b) => a.abs - b.abs)\n  const ranks: number[] = new Array(n)\n  let i = 0\n  while (i < n) {\n    let j = i\n    while (j < n && absRanks[j]!.abs === absRanks[i]!.abs) j++\n    const avg = (i + 1 + j) / 2\n    for (let k = i; k < j; k++) ranks[absRanks[k]!.i] = avg\n    i = j\n  }\n  let wPlus = 0\n  for (let k = 0; k < n; k++) if (diffs[k]! > 0) wPlus += ranks[k]!\n\n  const mean = (n * (n + 1)) / 4\n  const variance = (n * (n + 1) * (2 * n + 1)) / 24\n  const z = (wPlus - mean) / Math.sqrt(variance)\n  const p = 2 * (1 - normalCdf(Math.abs(z)))\n  return { w: wPlus, p }\n}\n\n/**\n * Cohen's d — standardized effect size for two independent groups.\n * Positive d means group b has higher mean than group a.\n * Rule of thumb: |d| < 0.2 negligible, 0.2–0.5 small, 0.5–0.8 medium, > 0.8 large.\n */\nexport function cohensD(a: number[], b: number[]): number {\n  if (a.length < 2 || b.length < 2) return 0\n  const meanA = a.reduce((x, y) => x + y, 0) / a.length\n  const meanB = b.reduce((x, y) => x + y, 0) / b.length\n  const varA = a.reduce((acc, x) => acc + (x - meanA) ** 2, 0) / (a.length - 1)\n  const varB = b.reduce((acc, x) => acc + (x - meanB) ** 2, 0) / (b.length - 1)\n  const pooled = Math.sqrt(\n    ((a.length - 1) * varA + (b.length - 1) * varB) / (a.length + b.length - 2),\n  )\n  if (pooled === 0) return 0\n  return (meanB - meanA) / pooled\n}\n\n/** Student-t CDF approximation via Abramowitz-Stegun series. */\nfunction studentTCdf(t: number, df: number): number {\n  if (df <= 0) return 0.5\n  if (df > 100) return normalCdf(t)\n  const x = df / (df + t * t)\n  const a = df / 2\n  const b = 0.5\n  const ib = incompleteBeta(x, a, b)\n  return t >= 0 ? 1 - 0.5 * ib : 0.5 * ib\n}\n\n/** Regularized incomplete beta function via continued fraction (Lentz). */\nfunction incompleteBeta(x: number, a: number, b: number): number {\n  if (x <= 0) return 0\n  if (x >= 1) return 1\n  const lnBeta = lnGamma(a) + lnGamma(b) - lnGamma(a + b)\n  const front = Math.exp(Math.log(x) * a + Math.log(1 - x) * b - lnBeta) / a\n  const maxIter = 200\n  const eps = 3e-7\n  let c = 1\n  let d = 1 - ((a + b) * x) / (a + 1)\n  if (Math.abs(d) < 1e-30) d = 1e-30\n  d = 1 / d\n  let f = d\n  for (let m = 1; m <= maxIter; m++) {\n    const m2 = 2 * m\n    let num = (m * (b - m) * x) / ((a + m2 - 1) * (a + m2))\n    d = 1 + num * d\n    if (Math.abs(d) < 1e-30) d = 1e-30\n    c = 1 + num / c\n    if (Math.abs(c) < 1e-30) c = 1e-30\n    d = 1 / d\n    f *= d * c\n    num = -((a + m) * (a + b + m) * x) / ((a + m2) * (a + m2 + 1))\n    d = 1 + num * d\n    if (Math.abs(d) < 1e-30) d = 1e-30\n    c = 1 + num / c\n    if (Math.abs(c) < 1e-30) c = 1e-30\n    d = 1 / d\n    const delta = d * c\n    f *= delta\n    if (Math.abs(delta - 1) < eps) break\n  }\n  return front * f\n}\n\n/** Lanczos approximation to ln Γ(z). */\nfunction lnGamma(z: number): number {\n  const g = 7\n  const coefs = [\n    0.99999999999980993, 676.5203681218851, -1259.1392167224028, 771.32342877765313,\n    -176.61502916214059, 12.507343278686905, -0.13857109526572012, 9.9843695780195716e-6,\n    1.5056327351493116e-7,\n  ]\n  if (z < 0.5) {\n    return Math.log(Math.PI / Math.sin(Math.PI * z)) - lnGamma(1 - z)\n  }\n  z -= 1\n  let x = coefs[0]!\n  for (let i = 1; i < g + 2; i++) x += coefs[i]! / (z + i)\n  const t = z + g + 0.5\n  return 0.5 * Math.log(2 * Math.PI) + (z + 0.5) * Math.log(t) - t + Math.log(x)\n}\n\n// Standard normal CDF approximation (Abramowitz and Stegun)\nfunction normalCdf(x: number): number {\n  const a1 = 0.254829592\n  const a2 = -0.284496736\n  const a3 = 1.421413741\n  const a4 = -1.453152027\n  const a5 = 1.061405429\n  const p = 0.3275911\n\n  const sign = x < 0 ? -1 : 1\n  const absX = Math.abs(x)\n  const t = 1 / (1 + p * absX)\n  const y = 1 - ((((a5 * t + a4) * t + a3) * t + a2) * t + a1) * t * Math.exp((-absX * absX) / 2)\n\n  return 0.5 * (1 + sign * y)\n}\n"],"mappings":";;;;;AAIA,IAAM,sBAAsB,oBAAI,IAAI,CAAC,iBAAiB,oBAAoB,eAAe,CAAC;AAQnF,SAAS,gBAAgB,QAAoC;AAClE,SAAO,OAAO,IAAI,CAAC,MAAM;AACvB,QAAI,oBAAoB,IAAI,EAAE,SAAS,GAAG;AACxC,aAAO;AAAA,IACT;AACA,WAAO;AAAA,EACT,CAAC;AACH;AAGO,SAAS,aAAa,QAAsD;AACjF,MAAI,OAAO,WAAW,EAAG,QAAO;AAChC,MAAI,cAAc;AAClB,MAAI,cAAc;AAClB,aAAW,EAAE,OAAO,OAAO,KAAK,QAAQ;AACtC,UAAM,IAAI,UAAU;AACpB,mBAAe,QAAQ;AACvB,mBAAe;AAAA,EACjB;AACA,SAAO,cAAc,IAAI,cAAc,cAAc;AACvD;AAGO,SAAS,mBACd,QACA,aAAa,MACmC;AAChD,MAAI,OAAO,WAAW,EAAG,QAAO,EAAE,MAAM,GAAG,OAAO,GAAG,OAAO,EAAE;AAC9D,MAAI,OAAO,WAAW,EAAG,QAAO,EAAE,MAAM,OAAO,CAAC,GAAI,OAAO,OAAO,CAAC,GAAI,OAAO,OAAO,CAAC,EAAG;AAEzF,QAAM,IAAI,OAAO;AACjB,QAAM,OAAO,OAAO,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI;AAEjD,QAAM,IAAI;AACV,QAAM,iBAA2B,CAAC;AAElC,WAAS,IAAI,GAAG,IAAI,GAAG,KAAK;AAC1B,QAAI,MAAM;AACV,aAAS,IAAI,GAAG,IAAI,GAAG,KAAK;AAC1B,aAAO,OAAO,KAAK,MAAM,KAAK,OAAO,IAAI,CAAC,CAAC;AAAA,IAC7C;AACA,mBAAe,KAAK,MAAM,CAAC;AAAA,EAC7B;AAEA,iBAAe,KAAK,CAAC,GAAG,MAAM,IAAI,CAAC;AAEnC,QAAM,QAAQ,IAAI;AAClB,QAAM,WAAW,KAAK,MAAO,QAAQ,IAAK,CAAC;AAC3C,QAAM,WAAW,KAAK,OAAO,IAAI,QAAQ,KAAK,CAAC,IAAI;AAEnD,SAAO;AAAA,IACL;AAAA,IACA,OAAO,eAAe,QAAQ;AAAA,IAC9B,OAAO,eAAe,KAAK,IAAI,UAAU,IAAI,CAAC,CAAC;AAAA,EACjD;AACF;AAQO,SAAS,sBAAsB,aAAqC;AACzE,MAAI,YAAY,SAAS,EAAG,QAAO;AAGnC,QAAM,eAAe,oBAAI,IAAwB;AACjD,aAAW,YAAY,aAAa;AAClC,eAAW,KAAK,UAAU;AACxB,UAAI,CAAC,aAAa,IAAI,EAAE,SAAS,EAAG,cAAa,IAAI,EAAE,WAAW,CAAC,CAAC;AACpE,YAAM,MAAM,aAAa,IAAI,EAAE,SAAS;AACxC,UAAI,IAAI,WAAW,KAAK,IAAI,IAAI,SAAS,CAAC,EAAG,UAAU,YAAY,QAAQ;AACzE,YAAI,KAAK,CAAC,EAAE,KAAK,CAAC;AAAA,MACpB,OAAO;AACL,YAAI,IAAI,SAAS,CAAC,EAAG,KAAK,EAAE,KAAK;AAAA,MACnC;AAAA,IACF;AAAA,EACF;AAGA,QAAM,YAAsB,CAAC;AAC7B,QAAM,YAAsB,CAAC;AAE7B,aAAW,SAAS,aAAa,OAAO,GAAG;AACzC,eAAW,WAAW,OAAO;AAC3B,UAAI,QAAQ,SAAS,EAAG;AACxB,iBAAW,KAAK,QAAS,WAAU,KAAK,CAAC;AACzC,eAAS,IAAI,GAAG,IAAI,QAAQ,QAAQ,KAAK;AACvC,iBAAS,IAAI,IAAI,GAAG,IAAI,QAAQ,QAAQ,KAAK;AAC3C,oBAAU,MAAM,QAAQ,CAAC,IAAK,QAAQ,CAAC,MAAO,CAAC;AAAA,QACjD;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,MAAI,UAAU,WAAW,KAAK,UAAU,SAAS,EAAG,QAAO;AAE3D,QAAM,uBAAuB,UAAU,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,UAAU;AAG9E,MAAI,uBAAuB;AAC3B,MAAI,gBAAgB;AACpB,WAAS,IAAI,GAAG,IAAI,UAAU,QAAQ,KAAK;AACzC,aAAS,IAAI,IAAI,GAAG,IAAI,UAAU,QAAQ,KAAK;AAC7C,+BAAyB,UAAU,CAAC,IAAK,UAAU,CAAC,MAAO;AAC3D;AAAA,IACF;AAAA,EACF;AACA,yBAAuB,gBAAgB,IAAI,uBAAuB,gBAAgB;AAElF,MAAI,yBAAyB,EAAG,QAAO;AACvC,SAAO,IAAI,uBAAuB;AACpC;AAMO,SAAS,aAAa,GAAa,GAAuC;AAC/E,MAAI,EAAE,WAAW,KAAK,EAAE,WAAW,EAAG,QAAO,EAAE,GAAG,GAAG,GAAG,EAAE;AAE1D,QAAM,KAAK,EAAE;AACb,QAAM,KAAK,EAAE;AAGb,QAAM,WAAW;AAAA,IACf,GAAG,EAAE,IAAI,CAAC,OAAO,EAAE,GAAG,OAAO,IAAa,EAAE;AAAA,IAC5C,GAAG,EAAE,IAAI,CAAC,OAAO,EAAE,GAAG,OAAO,IAAa,EAAE;AAAA,EAC9C,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,IAAI,EAAE,CAAC;AAG1B,QAAM,QAAkB,IAAI,MAAM,SAAS,MAAM;AACjD,MAAI,IAAI;AACR,SAAO,IAAI,SAAS,QAAQ;AAC1B,QAAI,IAAI;AACR,WAAO,IAAI,SAAS,UAAU,SAAS,CAAC,EAAG,MAAM,SAAS,CAAC,EAAG,EAAG;AACjE,UAAM,WAAW,IAAI,IAAI,KAAK;AAC9B,aAAS,IAAI,GAAG,IAAI,GAAG,IAAK,OAAM,CAAC,IAAI;AACvC,QAAI;AAAA,EACN;AAGA,MAAI,KAAK;AACT,WAAS,IAAI,GAAG,IAAI,SAAS,QAAQ,KAAK;AACxC,QAAI,SAAS,CAAC,EAAG,UAAU,IAAK,OAAM,MAAM,CAAC;AAAA,EAC/C;AAEA,QAAM,KAAK,KAAM,MAAM,KAAK,KAAM;AAClC,QAAM,KAAK,KAAK,KAAK;AACrB,QAAM,IAAI,KAAK,IAAI,IAAI,EAAE;AAGzB,QAAM,KAAM,KAAK,KAAM;AACvB,QAAM,QAAQ,KAAK,KAAM,KAAK,MAAM,KAAK,KAAK,KAAM,EAAE;AAEtD,MAAI,UAAU,EAAG,QAAO,EAAE,GAAG,GAAG,EAAE;AAElC,QAAM,IAAI,KAAK,IAAI,IAAI,EAAE,IAAI;AAE7B,QAAM,IAAI,KAAK,IAAI,UAAU,CAAC;AAE9B,SAAO,EAAE,GAAG,EAAE;AAChB;AAGO,SAAS,cAAc,SAAiB,QAAwB;AACrE,MAAI,UAAU,EAAG,QAAO;AACxB,SAAO,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,UAAU,MAAM,CAAC;AAClD;AAQO,SAAS,YACd,QACA,OACsC;AACtC,MAAI,OAAO,WAAW,MAAM,QAAQ;AAClC,UAAM,IAAI;AAAA,MACR,sCAAsC,OAAO,MAAM,OAAO,MAAM,MAAM;AAAA,IACxE;AAAA,EACF;AACA,QAAM,IAAI,OAAO;AACjB,MAAI,IAAI,EAAG,QAAO,EAAE,GAAG,GAAG,IAAI,GAAG,GAAG,EAAE;AAEtC,QAAM,QAAQ,OAAO,IAAI,CAAC,GAAG,MAAM,MAAM,CAAC,IAAK,CAAC;AAChD,QAAM,OAAO,MAAM,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI;AAChD,QAAM,WAAW,MAAM,OAAO,CAAC,KAAK,MAAM,OAAO,IAAI,SAAS,GAAG,CAAC,KAAK,IAAI;AAC3E,QAAM,KAAK,KAAK,KAAK,WAAW,CAAC;AACjC,MAAI,OAAO,EAAG,QAAO,EAAE,GAAG,SAAS,IAAI,IAAI,UAAU,IAAI,IAAI,GAAG,GAAG,SAAS,IAAI,IAAI,EAAE;AAEtF,QAAM,IAAI,OAAO;AACjB,QAAM,KAAK,IAAI;AACf,QAAM,IAAI,KAAK,IAAI,YAAY,KAAK,IAAI,CAAC,GAAG,EAAE;AAC9C,SAAO,EAAE,GAAG,IAAI,EAAE;AACpB;AAMO,SAAS,mBAAmB,QAAkB,OAA2C;AAC9F,MAAI,OAAO,WAAW,MAAM,QAAQ;AAClC,UAAM,IAAI;AAAA,MACR,6CAA6C,OAAO,MAAM,OAAO,MAAM,MAAM;AAAA,IAC/E;AAAA,EACF;AACA,QAAM,QAAQ,OAAO,IAAI,CAAC,GAAGA,OAAM,MAAMA,EAAC,IAAK,CAAC,EAAE,OAAO,CAAC,MAAM,MAAM,CAAC;AACvE,QAAM,IAAI,MAAM;AAChB,MAAI,IAAI,EAAG,QAAO,EAAE,GAAG,GAAG,GAAG,EAAE;AAE/B,QAAM,WAAW,MACd,IAAI,CAAC,GAAGA,QAAO,EAAE,KAAK,KAAK,IAAI,CAAC,GAAG,MAAM,KAAK,KAAK,CAAC,GAAG,GAAAA,GAAE,EAAE,EAC3D,KAAK,CAAC,GAAG,MAAM,EAAE,MAAM,EAAE,GAAG;AAC/B,QAAM,QAAkB,IAAI,MAAM,CAAC;AACnC,MAAI,IAAI;AACR,SAAO,IAAI,GAAG;AACZ,QAAI,IAAI;AACR,WAAO,IAAI,KAAK,SAAS,CAAC,EAAG,QAAQ,SAAS,CAAC,EAAG,IAAK;AACvD,UAAM,OAAO,IAAI,IAAI,KAAK;AAC1B,aAAS,IAAI,GAAG,IAAI,GAAG,IAAK,OAAM,SAAS,CAAC,EAAG,CAAC,IAAI;AACpD,QAAI;AAAA,EACN;AACA,MAAI,QAAQ;AACZ,WAAS,IAAI,GAAG,IAAI,GAAG,IAAK,KAAI,MAAM,CAAC,IAAK,EAAG,UAAS,MAAM,CAAC;AAE/D,QAAM,OAAQ,KAAK,IAAI,KAAM;AAC7B,QAAM,WAAY,KAAK,IAAI,MAAM,IAAI,IAAI,KAAM;AAC/C,QAAM,KAAK,QAAQ,QAAQ,KAAK,KAAK,QAAQ;AAC7C,QAAM,IAAI,KAAK,IAAI,UAAU,KAAK,IAAI,CAAC,CAAC;AACxC,SAAO,EAAE,GAAG,OAAO,EAAE;AACvB;AAOO,SAAS,QAAQ,GAAa,GAAqB;AACxD,MAAI,EAAE,SAAS,KAAK,EAAE,SAAS,EAAG,QAAO;AACzC,QAAM,QAAQ,EAAE,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,EAAE;AAC/C,QAAM,QAAQ,EAAE,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,EAAE;AAC/C,QAAM,OAAO,EAAE,OAAO,CAAC,KAAK,MAAM,OAAO,IAAI,UAAU,GAAG,CAAC,KAAK,EAAE,SAAS;AAC3E,QAAM,OAAO,EAAE,OAAO,CAAC,KAAK,MAAM,OAAO,IAAI,UAAU,GAAG,CAAC,KAAK,EAAE,SAAS;AAC3E,QAAM,SAAS,KAAK;AAAA,MAChB,EAAE,SAAS,KAAK,QAAQ,EAAE,SAAS,KAAK,SAAS,EAAE,SAAS,EAAE,SAAS;AAAA,EAC3E;AACA,MAAI,WAAW,EAAG,QAAO;AACzB,UAAQ,QAAQ,SAAS;AAC3B;AAGA,SAAS,YAAY,GAAW,IAAoB;AAClD,MAAI,MAAM,EAAG,QAAO;AACpB,MAAI,KAAK,IAAK,QAAO,UAAU,CAAC;AAChC,QAAM,IAAI,MAAM,KAAK,IAAI;AACzB,QAAM,IAAI,KAAK;AACf,QAAM,IAAI;AACV,QAAM,KAAK,eAAe,GAAG,GAAG,CAAC;AACjC,SAAO,KAAK,IAAI,IAAI,MAAM,KAAK,MAAM;AACvC;AAGA,SAAS,eAAe,GAAW,GAAW,GAAmB;AAC/D,MAAI,KAAK,EAAG,QAAO;AACnB,MAAI,KAAK,EAAG,QAAO;AACnB,QAAM,SAAS,QAAQ,CAAC,IAAI,QAAQ,CAAC,IAAI,QAAQ,IAAI,CAAC;AACtD,QAAM,QAAQ,KAAK,IAAI,KAAK,IAAI,CAAC,IAAI,IAAI,KAAK,IAAI,IAAI,CAAC,IAAI,IAAI,MAAM,IAAI;AACzE,QAAM,UAAU;AAChB,QAAM,MAAM;AACZ,MAAI,IAAI;AACR,MAAI,IAAI,KAAM,IAAI,KAAK,KAAM,IAAI;AACjC,MAAI,KAAK,IAAI,CAAC,IAAI,MAAO,KAAI;AAC7B,MAAI,IAAI;AACR,MAAI,IAAI;AACR,WAAS,IAAI,GAAG,KAAK,SAAS,KAAK;AACjC,UAAM,KAAK,IAAI;AACf,QAAI,MAAO,KAAK,IAAI,KAAK,MAAO,IAAI,KAAK,MAAM,IAAI;AACnD,QAAI,IAAI,MAAM;AACd,QAAI,KAAK,IAAI,CAAC,IAAI,MAAO,KAAI;AAC7B,QAAI,IAAI,MAAM;AACd,QAAI,KAAK,IAAI,CAAC,IAAI,MAAO,KAAI;AAC7B,QAAI,IAAI;AACR,SAAK,IAAI;AACT,UAAM,GAAG,IAAI,MAAM,IAAI,IAAI,KAAK,OAAO,IAAI,OAAO,IAAI,KAAK;AAC3D,QAAI,IAAI,MAAM;AACd,QAAI,KAAK,IAAI,CAAC,IAAI,MAAO,KAAI;AAC7B,QAAI,IAAI,MAAM;AACd,QAAI,KAAK,IAAI,CAAC,IAAI,MAAO,KAAI;AAC7B,QAAI,IAAI;AACR,UAAM,QAAQ,IAAI;AAClB,SAAK;AACL,QAAI,KAAK,IAAI,QAAQ,CAAC,IAAI,IAAK;AAAA,EACjC;AACA,SAAO,QAAQ;AACjB;AAGA,SAAS,QAAQ,GAAmB;AAClC,QAAM,IAAI;AACV,QAAM,QAAQ;AAAA,IACZ;AAAA,IAAqB;AAAA,IAAmB;AAAA,IAAqB;AAAA,IAC7D;AAAA,IAAqB;AAAA,IAAoB;AAAA,IAAsB;AAAA,IAC/D;AAAA,EACF;AACA,MAAI,IAAI,KAAK;AACX,WAAO,KAAK,IAAI,KAAK,KAAK,KAAK,IAAI,KAAK,KAAK,CAAC,CAAC,IAAI,QAAQ,IAAI,CAAC;AAAA,EAClE;AACA,OAAK;AACL,MAAI,IAAI,MAAM,CAAC;AACf,WAAS,IAAI,GAAG,IAAI,IAAI,GAAG,IAAK,MAAK,MAAM,CAAC,KAAM,IAAI;AACtD,QAAM,IAAI,IAAI,IAAI;AAClB,SAAO,MAAM,KAAK,IAAI,IAAI,KAAK,EAAE,KAAK,IAAI,OAAO,KAAK,IAAI,CAAC,IAAI,IAAI,KAAK,IAAI,CAAC;AAC/E;AAGA,SAAS,UAAU,GAAmB;AACpC,QAAM,KAAK;AACX,QAAM,KAAK;AACX,QAAM,KAAK;AACX,QAAM,KAAK;AACX,QAAM,KAAK;AACX,QAAM,IAAI;AAEV,QAAM,OAAO,IAAI,IAAI,KAAK;AAC1B,QAAM,OAAO,KAAK,IAAI,CAAC;AACvB,QAAM,IAAI,KAAK,IAAI,IAAI;AACvB,QAAM,IAAI,QAAQ,KAAK,IAAI,MAAM,IAAI,MAAM,IAAI,MAAM,IAAI,MAAM,IAAI,KAAK,IAAK,CAAC,OAAO,OAAQ,CAAC;AAE9F,SAAO,OAAO,IAAI,OAAO;AAC3B;","names":["i"]}