@tangle-network/agent-eval 0.184.0 → 0.185.0

package/dist/analyst/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","names":[],"sources":["../../src/analyst/adapters.ts","../../src/analyst/bind.ts","../../src/analyst/define.ts","../../src/analyst/kinds/skill-usage.ts"],"sourcesContent":["/**\n * Adapter factory — lifts the semantic-concept judge into the Analyst\n * contract without re-implementing it.\n *\n * It builds an Analyst with a stable id (caller chooses; a default is given),\n * a version derived from the judge's version plus an adapter revision, and an\n * `analyze()` that calls the judge and lifts its output to `AnalystFinding[]`\n * through `makeFinding()`. The judge's `Severity` ('critical' | 'major' |\n * 'minor' | 'info') projects onto `AnalystSeverity`: 'major' becomes 'high'\n * and 'minor' becomes 'medium'.\n *\n * The adapter owns no state. Calling the factory twice is safe.\n */\n\nimport { CostLedger } from '../cost-ledger'\nimport type { Severity as LayerSeverity } from '../multi-layer-verifier'\nimport {\n  runSemanticConceptJudge,\n  SEMANTIC_CONCEPT_JUDGE_VERSION,\n  type SemanticConceptJudgeInput,\n  type SemanticConceptJudgeOptions,\n  type SemanticConceptJudgeResult,\n} from '../semantic-concept-judge'\nimport type { Analyst, AnalystFinding, AnalystSeverity } from './types'\nimport { makeFinding } from './types'\nimport { settleUsageReceiptFromCostLedger, validateUsageSettlementTimeout } from './usage-receipt'\n\nconst ADAPTER_REV = '1'\n\n// ── Severity bridges ───────────────────────────────────────────────\n\nfunction liftSeverity(s: LayerSeverity): AnalystSeverity {\n  switch (s) {\n    case 'critical':\n      return 'critical'\n    case 'major':\n      return 'high'\n    case 'minor':\n      return 'medium'\n    case 'info':\n      return 'info'\n  }\n}\n\n// ── SemanticConceptJudge → Analyst ─────────────────────────────────\n\nexport interface SemanticConceptJudgeAdapterOpts {\n  id?: string\n  area?: string\n  /** Registry context owns cancellation and the per-analyst cost ledger. */\n  options: Omit<SemanticConceptJudgeOptions, 'costLedger' | 'signal'>\n  /** Maximum post-cancellation wait for a provider receipt. Default 5 seconds. */\n  settlementTimeoutMs?: number\n}\n\nexport function createSemanticConceptJudgeAdapter(\n  opts: SemanticConceptJudgeAdapterOpts,\n): Analyst<SemanticConceptJudgeInput> {\n  const id = opts.id ?? 'semantic-concept-judge'\n  const area = opts.area ?? 'concept-coverage'\n  const settlementTimeoutMs = validateUsageSettlementTimeout(opts.settlementTimeoutMs)\n  return {\n    id,\n    description:\n      'Runs the semantic-concept judge and surfaces missing / weak concepts as findings.',\n    inputKind: 'custom',\n    cost: {\n      kind: 'llm',\n      models: opts.options.model ? [opts.options.model] : undefined,\n      settlement_timeout_ms: settlementTimeoutMs,\n    },\n    version: `${SEMANTIC_CONCEPT_JUDGE_VERSION}-adapter-${ADAPTER_REV}`,\n    async analyze(input, ctx) {\n      const costLedger = new CostLedger(ctx.budgetUsd)\n      let result: SemanticConceptJudgeResult\n      try {\n        result = await runSemanticConceptJudge(input, {\n          ...opts.options,\n          costLedger,\n          signal: ctx.signal,\n        })\n      } finally {\n        const usage = await settleUsageReceiptFromCostLedger(costLedger, {\n          channel: 'judge',\n          timeoutMs: settlementTimeoutMs,\n        })\n        if (!usage.settled) {\n          ctx.log?.('semantic-concept judge provider settlement timed out', {\n            pending_calls: usage.pendingCalls,\n            timeout_ms: settlementTimeoutMs,\n          })\n        }\n        ctx.recordUsage?.(usage.receipt)\n      }\n      if (!result.available) {\n        return [\n          makeFinding({\n            analyst_id: id,\n            area,\n            claim: 'semantic-concept judge unavailable',\n            rationale: result.error,\n            severity: 'info',\n            confidence: 1,\n            evidence_refs: [],\n            metadata: { reason: result.error },\n          }),\n        ]\n      }\n      const out: AnalystFinding[] = []\n      for (const f of result.findings) {\n        // Only surface gaps: missing concepts or low scores. Concepts at\n        // 7+/10 with present=true are not findings — they're successes.\n        if (f.present && f.score >= 7) continue\n        out.push(\n          makeFinding({\n            analyst_id: id,\n            area,\n            subject: f.concept,\n            claim: f.present\n              ? `concept \"${f.concept}\" is weak (${f.score}/10)`\n              : `concept \"${f.concept}\" is missing`,\n            rationale: f.evidence,\n            severity: liftSeverity(f.severity),\n            confidence: 0.85,\n            evidence_refs: [{ kind: 'artifact', uri: 'inline:evidence', excerpt: f.evidence }],\n            metadata: {\n              concept: f.concept,\n              present: f.present,\n              score_10: f.score,\n            },\n          }),\n        )\n      }\n      return out\n    },\n  }\n}\n","/**\n * bindAnalyst — compile an `AnalystDefinition` into a runnable arm.\n *\n * The definition declares protocol (question, task text, reply grammar,\n * evidence projection, budget, repair turns); the transport binding supplies\n * execution (a bridge endpoint, or the caller-owned model path). Each\n * projection mode dispatches to the same strategy its arm's entry point runs,\n * so a compiled definition and the historical creator send byte-identical\n * requests — the definition parity suite asserts this per arm.\n *\n * A projection × transport pair with no strategy fails loud with\n * `AnalystExpressivenessError` naming the pair: an arm shape this layer cannot\n * express must be reported, never approximated.\n */\n\nimport type { CustomTokenPricing } from '../cost-ledger'\nimport type { AnalystBenchmarkRunner } from './benchmark'\nimport { runChunkedAnalystDefinition } from './benchmark-public-model'\nimport { runReplVariableAnalystDefinition } from './benchmark-public-rlm'\nimport type { PublicAnalystBenchmarkModelConfig } from './benchmark-public-types'\nimport { runInlineAnalystDefinition } from './benchmark-runner-prime'\nimport { type AnalystDefinition, AnalystExpressivenessError } from './definition'\nimport type { PrimeBridgeTransport } from './prime-bridge-transport'\nimport type { AnalystRunInputs } from './types'\n\n/** Execution half of a binding: who actually reaches the model. */\nexport type AnalystTransportBinding =\n  | {\n      /** An OpenAI-compatible cli-bridge endpoint (inline projections). */\n      readonly kind: 'prime-bridge'\n      readonly baseUrl: string\n      readonly model: string\n      readonly transport?: PrimeBridgeTransport\n      readonly pricing?: CustomTokenPricing\n    }\n  | {\n      /** The caller-owned model path (chunked and repl-variable projections). */\n      readonly kind: 'model-owner'\n      readonly config: PublicAnalystBenchmarkModelConfig\n    }\n\n/**\n * Compile a definition plus a transport binding into a runnable arm. Dispatch\n * is total over the expressible pairs; everything else is a loud refusal.\n */\nexport function bindAnalyst<TRow, TAssignment, TBlock>(\n  definition: AnalystDefinition<TRow, TAssignment, TBlock>,\n  transports: AnalystTransportBinding,\n): AnalystBenchmarkRunner<AnalystRunInputs> {\n  const mode = definition.projection.mode\n  if (mode === 'inline' && transports.kind === 'prime-bridge') {\n    return runInlineAnalystDefinition(definition, {\n      baseUrl: transports.baseUrl,\n      model: transports.model,\n      ...(transports.transport ? { transport: transports.transport } : {}),\n      ...(transports.pricing ? { pricing: transports.pricing } : {}),\n    })\n  }\n  if (mode === 'chunked' && transports.kind === 'model-owner') {\n    return runChunkedAnalystDefinition(definition, transports.config)\n  }\n  if (mode === 'repl-variable' && transports.kind === 'model-owner') {\n    return runReplVariableAnalystDefinition(\n      // The repl-variable strategy is written against the engine's raw-finding\n      // row type; the definition's own row declaration carries it.\n      definition as Parameters<typeof runReplVariableAnalystDefinition>[0],\n      transports.config,\n    )\n  }\n  throw new AnalystExpressivenessError(\n    `no strategy compiles a '${mode}' projection over a '${transports.kind}' transport; ` +\n      `definition '${definition.id}' cannot be expressed by this layer yet`,\n  )\n}\n","import type { TraceAnalysisStore } from '../trace-analyst/store'\nimport type { ExactCapableAnalyst } from './exact-types'\nimport type { TraceAnalystDefinition } from './kind-factory'\nimport type { Analyst, AnalystCost } from './types'\n\n/**\n * Define a reusable trace-research question.\n *\n * The returned value contains no model, credentials, or execution state. Bind\n * it to any TraceAnalysisEngine with `runTraceAnalyst` or\n * `createTraceAnalyst`.\n */\nexport function defineTraceAnalyst(definition: TraceAnalystDefinition): TraceAnalystDefinition {\n  for (const [name, value] of [\n    ['id', definition.id],\n    ['description', definition.description],\n    ['area', definition.area],\n    ['version', definition.version],\n    ['instructions', definition.instructions],\n  ] as const) {\n    if (typeof value !== 'string' || !value.trim()) {\n      throw new TypeError(`defineTraceAnalyst: ${name} must be a non-empty string`)\n    }\n  }\n  return {\n    ...definition,\n    limits: definition.limits ? { ...definition.limits } : undefined,\n  }\n}\n\nexport interface DefineCustomAnalystOptions {\n  id: string\n  description: string\n  version?: string\n  cost: AnalystCost\n  analyze: Analyst<TraceAnalysisStore>['analyze']\n}\n\nexport interface DefineExactCustomAnalystOptions extends DefineCustomAnalystOptions {\n  /** Canonical JSON for behavior knobs not already bound by `version`. */\n  executionConfig: Readonly<Record<string, unknown>>\n}\n\n/** Construct a registrable analyst from a hand-written analyze function. */\nexport function defineCustomAnalyst(\n  options: DefineExactCustomAnalystOptions,\n): ExactCapableAnalyst<TraceAnalysisStore>\nexport function defineCustomAnalyst(\n  options: DefineCustomAnalystOptions,\n): Analyst<TraceAnalysisStore>\nexport function defineCustomAnalyst(\n  options: DefineCustomAnalystOptions | DefineExactCustomAnalystOptions,\n): Analyst<TraceAnalysisStore> | ExactCapableAnalyst<TraceAnalysisStore> {\n  if (!options.id.trim()) throw new TypeError('defineCustomAnalyst: id must not be empty')\n  if (!options.description.trim()) {\n    throw new TypeError('defineCustomAnalyst: description must not be empty')\n  }\n  if (options.cost === undefined) {\n    throw new TypeError('defineCustomAnalyst: cost must be declared')\n  }\n  if (\n    'executionConfig' in options &&\n    (!options.executionConfig ||\n      typeof options.executionConfig !== 'object' ||\n      Array.isArray(options.executionConfig))\n  ) {\n    throw new TypeError('defineCustomAnalyst: executionConfig must be an object')\n  }\n  return {\n    id: options.id,\n    description: options.description,\n    version: options.version ?? '1.0.0',\n    inputKind: 'trace-store',\n    cost: options.cost,\n    analyze: options.analyze,\n    ...('executionConfig' in options ? { executionConfig: options.executionConfig } : {}),\n  }\n}\n","/**\n * Skill-usage analyst — a DETERMINISTIC `Analyst` over a Claude/Codex skill\n * library + its trace corpus. Unlike the trace-store kinds (failure-mode,\n * improvement, ...) this kind calls no LLM: it mines real usage and skill\n * structure and emits findings by rule.\n *\n * It exists because the naive \"Skill-tool invocation count\" lies low — it\n * misses orchestrated sub-dispatch (a leaf skill run BY /pursue or /governor\n * logs under the parent), slash-command entry, local-script bypass, and\n * on-disk artifacts. The 2026-05-30 skill audit found 39/53 skills at zero\n * direct invocations, yet only one was a genuine cut: the rest were\n * measurement-invisible or discovery-limited. This analyst encodes that\n * lesson as a multi-signal usage model so a cheap repeatable pass can keep\n * the library honest, and so the expensive audit workflow's verdicts can\n * GEPA-distill it toward agreement (see `gold/skill-verdicts.gold.jsonl`).\n *\n * Report-building (`buildSkillUsageReport`, an fs scan) is separated from\n * finding emission (`SkillUsageAnalyst.analyze`, pure) so the slow scan runs\n * once at the registry boundary and the rule logic stays unit-testable.\n */\n\nimport { type Dirent, existsSync, readdirSync, readFileSync, statSync } from 'node:fs'\nimport { join } from 'node:path'\nimport type { ExactCapableAnalyst } from '../exact-types'\nimport type { AnalystContext, AnalystFinding, AnalystSeverity } from '../types'\nimport { computeFindingId } from '../types'\n\n// ── Input model ──────────────────────────────────────────────────────\n\nexport type SkillKind = 'public' | 'private'\n\n/** One skill's multi-signal usage + structure. All counts are deterministic. */\nexport interface SkillUsageRecord {\n  name: string\n  kind: SkillKind\n  /** Absolute path to the skill's SKILL.md. */\n  path: string\n  lines: number\n  /** `\"skill\":\"<name>\"` Skill-tool invocations across the trace corpus. */\n  directInvocations: number\n  /** `<command-name>/<name>` slash invocations across the trace corpus. */\n  slashInvocations: number\n  /** Sibling skills whose SKILL.md dispatches to this one (`/<name>`). Proxy\n   *  for orchestrated sub-dispatch the per-skill counter cannot see. */\n  inboundRefs: number\n  /** On-disk artifacts attributable to the skill (e.g. `.evolve/<name>/**`). */\n  artifactCount: number\n  /** Tangle-private reference count in the body (leak signal for public skills). */\n  tanglePrivateRefs: number\n  hasReferencesDir: boolean\n  hasEvalsDir: boolean\n  /** Body mentions `skill-runs.jsonl` (visible to /reflect + /governor). */\n  logsRuns: boolean\n  /** Description carries an explicit `Triggers:` clause / trigger phrases. */\n  hasTriggerPhrases: boolean\n}\n\nexport interface SkillUsageReport {\n  generatedFromTraces: number\n  records: SkillUsageRecord[]\n}\n\nexport interface SkillUsageScanConfig {\n  /** Dirs holding `*.jsonl` transcripts (Claude `~/.claude/projects`, Codex sessions). */\n  transcriptDirs: string[]\n  /** Skill roots to scan; each dir directly under `root` with a `SKILL.md` is a skill. */\n  skillRoots: { root: string; kind: SkillKind }[]\n  /** Roots scanned for `<root>/.evolve/<skill>` artifact dirs. */\n  artifactRoots?: string[]\n  /** Token-prefixed mappings: skill name → extra artifact subpaths under an artifactRoot\n   *  (e.g. reflect → `.evolve/reflections`). Catches non-eponymous artifact dirs. */\n  artifactAliases?: Record<string, string[]>\n  /** Cap files read per transcript dir (bounds a huge corpus); 0 = unbounded. */\n  maxTranscriptsPerDir?: number\n}\n\n// ── Deterministic thresholds ─────────────────────────────────────────\n\n/** Anthropic's authoring guidance keeps SKILL.md short; past this with no\n *  `references/` split the body burns context budget every session. */\nconst BLOAT_LINE_THRESHOLD = 300\n\nconst TANGLE_PRIVATE_RE =\n  /\\b(cli-bridge|tangletools|ops-board|drew-gtr-pro|@tangle-network\\/|~\\/company|tangle\\.tools|gtm-agent)\\b|\\bkimi\\b|\\btcloud\\b/gi\nconst TRIGGER_RE = /triggers?\\s*[:-]/i\n\n// ── Report builder (fs scan — slow, runs once at the registry boundary) ──\n\nfunction listSkillDirs(root: string): { name: string; path: string }[] {\n  if (!existsSync(root)) return []\n  const out: { name: string; path: string }[] = []\n  for (const entry of readdirSync(root, { withFileTypes: true })) {\n    if (!entry.isDirectory() && !entry.isSymbolicLink()) continue\n    const skillMd = join(root, entry.name, 'SKILL.md')\n    if (existsSync(skillMd)) out.push({ name: entry.name, path: skillMd })\n  }\n  return out\n}\n\nfunction walkJsonl(dir: string, cap: number): string[] {\n  if (!existsSync(dir)) return []\n  const files: string[] = []\n  const stack = [dir]\n  while (stack.length) {\n    const cur = stack.pop()!\n    let entries: Dirent[]\n    try {\n      entries = readdirSync(cur, { withFileTypes: true })\n    } catch {\n      continue\n    }\n    for (const e of entries) {\n      const full = join(cur, e.name)\n      if (e.isDirectory()) stack.push(full)\n      else if (e.name.endsWith('.jsonl')) {\n        files.push(full)\n        if (cap > 0 && files.length >= cap) return files\n      }\n    }\n  }\n  return files\n}\n\nfunction frontmatterDescription(body: string): string {\n  const fm = /^---\\n([\\s\\S]*?)\\n---/.exec(body)\n  const block = fm?.[1] ?? ''\n  const m = /description:\\s*(.+)/i.exec(block)\n  return m?.[1] ?? ''\n}\n\nfunction countArtifacts(roots: string[], name: string, aliases: string[]): number {\n  let n = 0\n  for (const root of roots) {\n    const candidates = [join(root, '.evolve', name), ...aliases.map((a) => join(root, a))]\n    for (const dir of candidates) {\n      if (!existsSync(dir)) continue\n      try {\n        if (statSync(dir).isDirectory()) n += readdirSync(dir).length\n        else n += 1\n      } catch {\n        /* unreadable — skip */\n      }\n    }\n  }\n  return n\n}\n\n/** Scan the corpus + skill roots into a {@link SkillUsageReport}. Deterministic. */\nexport function buildSkillUsageReport(config: SkillUsageScanConfig): SkillUsageReport {\n  const skills = config.skillRoots.flatMap(({ root, kind }) =>\n    listSkillDirs(root).map((s) => ({ ...s, kind })),\n  )\n  const names = skills.map((s) => s.name)\n\n  // One pass over the corpus accumulating direct + slash counts per skill.\n  const direct = new Map<string, number>(names.map((n) => [n, 0]))\n  const slash = new Map<string, number>(names.map((n) => [n, 0]))\n  const skillRe = /\"skill\"\\s*:\\s*\"([a-z0-9_:-]+)\"/g\n  const cmdRe = /<command-name>\\/?([a-z0-9_:-]+)<\\/command-name>/g\n  let transcripts = 0\n  for (const dir of config.transcriptDirs) {\n    for (const file of walkJsonl(dir, config.maxTranscriptsPerDir ?? 0)) {\n      transcripts += 1\n      let data: string\n      try {\n        data = readFileSync(file, 'utf8')\n      } catch {\n        continue\n      }\n      for (const m of data.matchAll(skillRe)) {\n        const g = m[1]\n        if (!g) continue\n        const n = g.split(':').pop() ?? g\n        const prev = direct.get(n)\n        if (prev !== undefined) direct.set(n, prev + 1)\n      }\n      for (const m of data.matchAll(cmdRe)) {\n        const g = m[1]\n        if (g === undefined) continue\n        const prev = slash.get(g)\n        if (prev !== undefined) slash.set(g, prev + 1)\n      }\n    }\n  }\n\n  // Read each skill body once; compute structure + inbound refs across siblings.\n  const bodies = new Map<string, string>()\n  for (const s of skills) {\n    try {\n      bodies.set(s.name, readFileSync(s.path, 'utf8'))\n    } catch {\n      bodies.set(s.name, '')\n    }\n  }\n  const inbound = new Map<string, number>(names.map((n) => [n, 0]))\n  for (const target of names) {\n    const ref = new RegExp(`/${target}\\\\b|\\\\[\\\\[${target}\\\\]\\\\]`)\n    for (const s of skills) {\n      if (s.name === target) continue\n      if (ref.test(bodies.get(s.name) ?? '')) inbound.set(target, inbound.get(target)! + 1)\n    }\n  }\n\n  const records: SkillUsageRecord[] = skills.map((s) => {\n    const body = bodies.get(s.name) ?? ''\n    const dir = s.path.replace(/\\/SKILL\\.md$/, '')\n    return {\n      name: s.name,\n      kind: s.kind,\n      path: s.path,\n      lines: body ? body.split('\\n').length : 0,\n      directInvocations: direct.get(s.name) ?? 0,\n      slashInvocations: slash.get(s.name) ?? 0,\n      inboundRefs: inbound.get(s.name) ?? 0,\n      artifactCount: countArtifacts(\n        config.artifactRoots ?? [],\n        s.name,\n        config.artifactAliases?.[s.name] ?? [],\n      ),\n      tanglePrivateRefs: (body.match(TANGLE_PRIVATE_RE) ?? []).length,\n      hasReferencesDir: existsSync(join(dir, 'references')),\n      hasEvalsDir: existsSync(join(dir, 'evals')),\n      logsRuns: body.includes('skill-runs.jsonl'),\n      hasTriggerPhrases: TRIGGER_RE.test(frontmatterDescription(body) || body.slice(0, 600)),\n    }\n  })\n  return { generatedFromTraces: transcripts, records }\n}\n\n// ── Finding emission (pure — unit-testable, no LLM, no fs) ────────────\n\nconst ANALYST_ID = 'skill-usage'\n\nfunction finding(\n  area: string,\n  subject: string,\n  claim: string,\n  severity: AnalystSeverity,\n  confidence: number,\n  producedAt: string,\n  recommended: string,\n  evidenceUri: string,\n  rationale?: string,\n): AnalystFinding {\n  return {\n    schema_version: '1.0.0',\n    finding_id: computeFindingId({ analyst_id: ANALYST_ID, area, subject, claim }),\n    analyst_id: ANALYST_ID,\n    produced_at: producedAt,\n    severity,\n    area,\n    claim,\n    rationale,\n    evidence_refs: [{ kind: 'artifact', uri: evidenceUri }],\n    recommended_action: recommended,\n    confidence,\n    subject,\n  }\n}\n\n/** Pure rule pass over a report → findings. Exported for direct/unit use. */\nexport function emitSkillUsageFindings(\n  report: SkillUsageReport,\n  producedAt: string,\n): AnalystFinding[] {\n  const out: AnalystFinding[] = []\n  for (const r of report.records) {\n    const directTotal = r.directInvocations + r.slashInvocations\n    const trueUsage = directTotal + r.inboundRefs + r.artifactCount\n\n    // 1. Dead: no usage signal of ANY kind. The only real deprecation candidate.\n    if (trueUsage === 0) {\n      out.push(\n        finding(\n          'skill-usage',\n          r.name,\n          `Skill '${r.name}' has zero usage across all signals (direct, slash, inbound-refs, artifacts)`,\n          'high',\n          0.6,\n          producedAt,\n          'Confirm the skill covers a real recurring job; if not, deprecate. Zero true usage is the only deterministic deprecation candidate.',\n          r.path,\n          'No Skill-tool call, no slash invocation, no sibling dispatches to it, and no on-disk artifacts.',\n        ),\n      )\n    } else if (directTotal === 0 && r.inboundRefs + r.artifactCount > 0) {\n      // 2. Measurement-invisible: real use via orchestration/artifacts, never invoked directly.\n      out.push(\n        finding(\n          'skill-usage',\n          r.name,\n          `Skill '${r.name}' shows 0 direct invocations but is used via orchestration/artifacts (inbound=${r.inboundRefs}, artifacts=${r.artifactCount})`,\n          'info',\n          0.8,\n          producedAt,\n          'Do NOT treat as unused — usage is real but logged under parent skills or on disk. Strengthen direct-invocation discovery only if direct use is desired.',\n          r.path,\n          'The Skill-tool counter undercounts orchestrated/chained leaf skills.',\n        ),\n      )\n    }\n\n    // 3. Discovery gap: low direct use AND weak trigger surface.\n    if (directTotal <= 2 && !r.hasTriggerPhrases) {\n      out.push(\n        finding(\n          'discoverability',\n          r.name,\n          `Skill '${r.name}' is rarely invoked directly and its description has no explicit trigger phrases`,\n          'medium',\n          0.7,\n          producedAt,\n          'Add a `Triggers:` clause with verbatim user phrases to the frontmatter description so the model auto-invokes it.',\n          r.path,\n        ),\n      )\n    }\n\n    // 4. Public-repo leak.\n    if (r.kind === 'public' && r.tanglePrivateRefs > 0) {\n      out.push(\n        finding(\n          'safety',\n          r.name,\n          `Public skill '${r.name}' carries ${r.tanglePrivateRefs} Tangle-private reference(s)`,\n          'high',\n          0.75,\n          producedAt,\n          'Sanitize incidental internal refs (cli-bridge/kimi/tcloud/~company/private repos) or relocate to a private repo. Verify @tangle-network/* refs are to PUBLISHED packages before treating as a leak.',\n          r.path,\n        ),\n      )\n    }\n\n    // 5. Bloat / no progressive disclosure.\n    if (r.lines > BLOAT_LINE_THRESHOLD && !r.hasReferencesDir) {\n      out.push(\n        finding(\n          'maintainability',\n          r.name,\n          `Skill '${r.name}' is ${r.lines} lines with no references/ split (progressive disclosure)`,\n          'medium',\n          0.8,\n          producedAt,\n          `Split detail into references/ loaded on demand; keep SKILL.md a short overview. ${r.lines} lines load into every session's context budget.`,\n          r.path,\n        ),\n      )\n    }\n\n    // 6. No evals (Anthropic's \">=3 evals before docs\" rule).\n    if (!r.hasEvalsDir) {\n      out.push(\n        finding(\n          'data-quality',\n          r.name,\n          `Skill '${r.name}' ships no evals/`,\n          'low',\n          0.6,\n          producedAt,\n          'Add evals/evals.json with >=3 scenarios proving the skill beats baseline; gives regression coverage.',\n          r.path,\n        ),\n      )\n    }\n\n    // 7. No run logging → invisible to /reflect and /governor.\n    if (!r.logsRuns) {\n      out.push(\n        finding(\n          'observability',\n          r.name,\n          `Skill '${r.name}' never appends to .evolve/skill-runs.jsonl`,\n          'low',\n          0.55,\n          producedAt,\n          'Append one run line to .evolve/skill-runs.jsonl on completion, or declare it a non-logging leaf, so the self-improvement loop can see it ran.',\n          r.path,\n        ),\n      )\n    }\n  }\n  return out\n}\n\n// ── The Analyst ──────────────────────────────────────────────────────\n\nclass SkillUsageAnalyst implements ExactCapableAnalyst<SkillUsageReport> {\n  readonly id = ANALYST_ID\n  readonly description =\n    'Deterministic multi-signal skill-usage analysis: flags dead skills, measurement-invisible (orchestrated) usage, discovery gaps, public-repo leaks, bloat, missing evals, and missing run-logging.'\n  readonly inputKind = 'custom' as const\n  readonly cost = { kind: 'deterministic' as const, est_usd_per_run: 0 }\n  readonly version = '1.0.0'\n  readonly executionConfig = {\n    kind: 'skill-usage',\n    bloat_line_threshold: BLOAT_LINE_THRESHOLD,\n    produced_at_source: 'tags.producedAt-or-system-clock',\n  } as const\n\n  async analyze(input: SkillUsageReport, ctx: AnalystContext): Promise<AnalystFinding[]> {\n    const producedAt = ctx.tags?.producedAt ?? new Date().toISOString()\n    ctx.log?.(\n      `skill-usage: ${input.records.length} skills over ${input.generatedFromTraces} transcripts`,\n    )\n    return emitSkillUsageFindings(input, producedAt)\n  }\n}\n\nexport const SKILL_USAGE_ANALYST = new SkillUsageAnalyst()\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;AA2BA,MAAM,cAAc;AAIpB,SAAS,aAAa,GAAmC;CACvD,QAAQ,GAAR;EACE,KAAK,YACH,OAAO;EACT,KAAK,SACH,OAAO;EACT,KAAK,SACH,OAAO;EACT,KAAK,QACH,OAAO;CACX;AACF;AAaA,SAAgB,kCACd,MACoC;CACpC,MAAM,KAAK,KAAK,MAAM;CACtB,MAAM,OAAO,KAAK,QAAQ;CAC1B,MAAM,sBAAsB,+BAA+B,KAAK,mBAAmB;CACnF,OAAO;EACL;EACA,aACE;EACF,WAAW;EACX,MAAM;GACJ,MAAM;GACN,QAAQ,KAAK,QAAQ,QAAQ,CAAC,KAAK,QAAQ,KAAK,IAAI,KAAA;GACpD,uBAAuB;EACzB;EACA,SAAS,GAAG,+BAA+B,WAAW;EACtD,MAAM,QAAQ,OAAO,KAAK;GACxB,MAAM,aAAa,IAAI,WAAW,IAAI,SAAS;GAC/C,IAAI;GACJ,IAAI;IACF,SAAS,MAAM,wBAAwB,OAAO;KAC5C,GAAG,KAAK;KACR;KACA,QAAQ,IAAI;IACd,CAAC;GACH,UAAU;IACR,MAAM,QAAQ,MAAM,iCAAiC,YAAY;KAC/D,SAAS;KACT,WAAW;IACb,CAAC;IACD,IAAI,CAAC,MAAM,SACT,IAAI,MAAM,wDAAwD;KAChE,eAAe,MAAM;KACrB,YAAY;IACd,CAAC;IAEH,IAAI,cAAc,MAAM,OAAO;GACjC;GACA,IAAI,CAAC,OAAO,WACV,OAAO,CACL,YAAY;IACV,YAAY;IACZ;IACA,OAAO;IACP,WAAW,OAAO;IAClB,UAAU;IACV,YAAY;IACZ,eAAe,CAAC;IAChB,UAAU,EAAE,QAAQ,OAAO,MAAM;GACnC,CAAC,CACH;GAEF,MAAM,MAAwB,CAAC;GAC/B,KAAK,MAAM,KAAK,OAAO,UAAU;IAG/B,IAAI,EAAE,WAAW,EAAE,SAAS,GAAG;IAC/B,IAAI,KACF,YAAY;KACV,YAAY;KACZ;KACA,SAAS,EAAE;KACX,OAAO,EAAE,UACL,YAAY,EAAE,QAAQ,aAAa,EAAE,MAAM,QAC3C,YAAY,EAAE,QAAQ;KAC1B,WAAW,EAAE;KACb,UAAU,aAAa,EAAE,QAAQ;KACjC,YAAY;KACZ,eAAe,CAAC;MAAE,MAAM;MAAY,KAAK;MAAmB,SAAS,EAAE;KAAS,CAAC;KACjF,UAAU;MACR,SAAS,EAAE;MACX,SAAS,EAAE;MACX,UAAU,EAAE;KACd;IACF,CAAC,CACH;GACF;GACA,OAAO;EACT;CACF;AACF;;;;;;;AC3FA,SAAgB,YACd,YACA,YAC0C;CAC1C,MAAM,OAAO,WAAW,WAAW;CACnC,IAAI,SAAS,YAAY,WAAW,SAAS,gBAC3C,OAAO,2BAA2B,YAAY;EAC5C,SAAS,WAAW;EACpB,OAAO,WAAW;EAClB,GAAI,WAAW,YAAY,EAAE,WAAW,WAAW,UAAU,IAAI,CAAC;EAClE,GAAI,WAAW,UAAU,EAAE,SAAS,WAAW,QAAQ,IAAI,CAAC;CAC9D,CAAC;CAEH,IAAI,SAAS,aAAa,WAAW,SAAS,eAC5C,OAAO,4BAA4B,YAAY,WAAW,MAAM;CAElE,IAAI,SAAS,mBAAmB,WAAW,SAAS,eAClD,OAAO,iCAGL,YACA,WAAW,MACb;CAEF,MAAM,IAAI,2BACR,2BAA2B,KAAK,uBAAuB,WAAW,KAAK,2BACtD,WAAW,GAAG,wCACjC;AACF;;;;;;;;;;AC7DA,SAAgB,mBAAmB,YAA4D;CAC7F,KAAK,MAAM,CAAC,MAAM,UAAU;EAC1B,CAAC,MAAM,WAAW,EAAE;EACpB,CAAC,eAAe,WAAW,WAAW;EACtC,CAAC,QAAQ,WAAW,IAAI;EACxB,CAAC,WAAW,WAAW,OAAO;EAC9B,CAAC,gBAAgB,WAAW,YAAY;CAC1C,GACE,IAAI,OAAO,UAAU,YAAY,CAAC,MAAM,KAAK,GAC3C,MAAM,IAAI,UAAU,uBAAuB,KAAK,4BAA4B;CAGhF,OAAO;EACL,GAAG;EACH,QAAQ,WAAW,SAAS,EAAE,GAAG,WAAW,OAAO,IAAI,KAAA;CACzD;AACF;AAsBA,SAAgB,oBACd,SACuE;CACvE,IAAI,CAAC,QAAQ,GAAG,KAAK,GAAG,MAAM,IAAI,UAAU,2CAA2C;CACvF,IAAI,CAAC,QAAQ,YAAY,KAAK,GAC5B,MAAM,IAAI,UAAU,oDAAoD;CAE1E,IAAI,QAAQ,SAAS,KAAA,GACnB,MAAM,IAAI,UAAU,4CAA4C;CAElE,IACE,qBAAqB,YACpB,CAAC,QAAQ,mBACR,OAAO,QAAQ,oBAAoB,YACnC,MAAM,QAAQ,QAAQ,eAAe,IAEvC,MAAM,IAAI,UAAU,wDAAwD;CAE9E,OAAO;EACL,IAAI,QAAQ;EACZ,aAAa,QAAQ;EACrB,SAAS,QAAQ,WAAW;EAC5B,WAAW;EACX,MAAM,QAAQ;EACd,SAAS,QAAQ;EACjB,GAAI,qBAAqB,UAAU,EAAE,iBAAiB,QAAQ,gBAAgB,IAAI,CAAC;CACrF;AACF;;;;;;;;;;;;;;;;;;;;;;;;;ACGA,MAAM,uBAAuB;AAE7B,MAAM,oBACJ;AACF,MAAM,aAAa;AAInB,SAAS,cAAc,MAAgD;CACrE,IAAI,CAAC,WAAW,IAAI,GAAG,OAAO,CAAC;CAC/B,MAAM,MAAwC,CAAC;CAC/C,KAAK,MAAM,SAAS,YAAY,MAAM,EAAE,eAAe,KAAK,CAAC,GAAG;EAC9D,IAAI,CAAC,MAAM,YAAY,KAAK,CAAC,MAAM,eAAe,GAAG;EACrD,MAAM,UAAU,KAAK,MAAM,MAAM,MAAM,UAAU;EACjD,IAAI,WAAW,OAAO,GAAG,IAAI,KAAK;GAAE,MAAM,MAAM;GAAM,MAAM;EAAQ,CAAC;CACvE;CACA,OAAO;AACT;AAEA,SAAS,UAAU,KAAa,KAAuB;CACrD,IAAI,CAAC,WAAW,GAAG,GAAG,OAAO,CAAC;CAC9B,MAAM,QAAkB,CAAC;CACzB,MAAM,QAAQ,CAAC,GAAG;CAClB,OAAO,MAAM,QAAQ;EACnB,MAAM,MAAM,MAAM,IAAI;EACtB,IAAI;EACJ,IAAI;GACF,UAAU,YAAY,KAAK,EAAE,eAAe,KAAK,CAAC;EACpD,QAAQ;GACN;EACF;EACA,KAAK,MAAM,KAAK,SAAS;GACvB,MAAM,OAAO,KAAK,KAAK,EAAE,IAAI;GAC7B,IAAI,EAAE,YAAY,GAAG,MAAM,KAAK,IAAI;QAC/B,IAAI,EAAE,KAAK,SAAS,QAAQ,GAAG;IAClC,MAAM,KAAK,IAAI;IACf,IAAI,MAAM,KAAK,MAAM,UAAU,KAAK,OAAO;GAC7C;EACF;CACF;CACA,OAAO;AACT;AAEA,SAAS,uBAAuB,MAAsB;CAEpD,MAAM,QADK,wBAAwB,KAAK,IACzB,CAAC,GAAG,MAAM;CAEzB,OADU,uBAAuB,KAAK,KAC/B,CAAC,GAAG,MAAM;AACnB;AAEA,SAAS,eAAe,OAAiB,MAAc,SAA2B;CAChF,IAAI,IAAI;CACR,KAAK,MAAM,QAAQ,OAAO;EACxB,MAAM,aAAa,CAAC,KAAK,MAAM,WAAW,IAAI,GAAG,GAAG,QAAQ,KAAK,MAAM,KAAK,MAAM,CAAC,CAAC,CAAC;EACrF,KAAK,MAAM,OAAO,YAAY;GAC5B,IAAI,CAAC,WAAW,GAAG,GAAG;GACtB,IAAI;IACF,IAAI,SAAS,GAAG,CAAC,CAAC,YAAY,GAAG,KAAK,YAAY,GAAG,CAAC,CAAC;SAClD,KAAK;GACZ,QAAQ,CAER;EACF;CACF;CACA,OAAO;AACT;;AAGA,SAAgB,sBAAsB,QAAgD;CACpF,MAAM,SAAS,OAAO,WAAW,SAAS,EAAE,MAAM,WAChD,cAAc,IAAI,CAAC,CAAC,KAAK,OAAO;EAAE,GAAG;EAAG;CAAK,EAAE,CACjD;CACA,MAAM,QAAQ,OAAO,KAAK,MAAM,EAAE,IAAI;CAGtC,MAAM,SAAS,IAAI,IAAoB,MAAM,KAAK,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC;CAC/D,MAAM,QAAQ,IAAI,IAAoB,MAAM,KAAK,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC;CAC9D,MAAM,UAAU;CAChB,MAAM,QAAQ;CACd,IAAI,cAAc;CAClB,KAAK,MAAM,OAAO,OAAO,gBACvB,KAAK,MAAM,QAAQ,UAAU,KAAK,OAAO,wBAAwB,CAAC,GAAG;EACnE,eAAe;EACf,IAAI;EACJ,IAAI;GACF,OAAO,aAAa,MAAM,MAAM;EAClC,QAAQ;GACN;EACF;EACA,KAAK,MAAM,KAAK,KAAK,SAAS,OAAO,GAAG;GACtC,MAAM,IAAI,EAAE;GACZ,IAAI,CAAC,GAAG;GACR,MAAM,IAAI,EAAE,MAAM,GAAG,CAAC,CAAC,IAAI,KAAK;GAChC,MAAM,OAAO,OAAO,IAAI,CAAC;GACzB,IAAI,SAAS,KAAA,GAAW,OAAO,IAAI,GAAG,OAAO,CAAC;EAChD;EACA,KAAK,MAAM,KAAK,KAAK,SAAS,KAAK,GAAG;GACpC,MAAM,IAAI,EAAE;GACZ,IAAI,MAAM,KAAA,GAAW;GACrB,MAAM,OAAO,MAAM,IAAI,CAAC;GACxB,IAAI,SAAS,KAAA,GAAW,MAAM,IAAI,GAAG,OAAO,CAAC;EAC/C;CACF;CAIF,MAAM,yBAAS,IAAI,IAAoB;CACvC,KAAK,MAAM,KAAK,QACd,IAAI;EACF,OAAO,IAAI,EAAE,MAAM,aAAa,EAAE,MAAM,MAAM,CAAC;CACjD,QAAQ;EACN,OAAO,IAAI,EAAE,MAAM,EAAE;CACvB;CAEF,MAAM,UAAU,IAAI,IAAoB,MAAM,KAAK,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC;CAChE,KAAK,MAAM,UAAU,OAAO;EAC1B,MAAM,MAAM,IAAI,OAAO,IAAI,OAAO,YAAY,OAAO,OAAO;EAC5D,KAAK,MAAM,KAAK,QAAQ;GACtB,IAAI,EAAE,SAAS,QAAQ;GACvB,IAAI,IAAI,KAAK,OAAO,IAAI,EAAE,IAAI,KAAK,EAAE,GAAG,QAAQ,IAAI,QAAQ,QAAQ,IAAI,MAAM,IAAK,CAAC;EACtF;CACF;CAEA,MAAM,UAA8B,OAAO,KAAK,MAAM;EACpD,MAAM,OAAO,OAAO,IAAI,EAAE,IAAI,KAAK;EACnC,MAAM,MAAM,EAAE,KAAK,QAAQ,gBAAgB,EAAE;EAC7C,OAAO;GACL,MAAM,EAAE;GACR,MAAM,EAAE;GACR,MAAM,EAAE;GACR,OAAO,OAAO,KAAK,MAAM,IAAI,CAAC,CAAC,SAAS;GACxC,mBAAmB,OAAO,IAAI,EAAE,IAAI,KAAK;GACzC,kBAAkB,MAAM,IAAI,EAAE,IAAI,KAAK;GACvC,aAAa,QAAQ,IAAI,EAAE,IAAI,KAAK;GACpC,eAAe,eACb,OAAO,iBAAiB,CAAC,GACzB,EAAE,MACF,OAAO,kBAAkB,EAAE,SAAS,CAAC,CACvC;GACA,oBAAoB,KAAK,MAAM,iBAAiB,KAAK,CAAC,EAAA,CAAG;GACzD,kBAAkB,WAAW,KAAK,KAAK,YAAY,CAAC;GACpD,aAAa,WAAW,KAAK,KAAK,OAAO,CAAC;GAC1C,UAAU,KAAK,SAAS,kBAAkB;GAC1C,mBAAmB,WAAW,KAAK,uBAAuB,IAAI,KAAK,KAAK,MAAM,GAAG,GAAG,CAAC;EACvF;CACF,CAAC;CACD,OAAO;EAAE,qBAAqB;EAAa;CAAQ;AACrD;AAIA,MAAM,aAAa;AAEnB,SAAS,QACP,MACA,SACA,OACA,UACA,YACA,YACA,aACA,aACA,WACgB;CAChB,OAAO;EACL,gBAAgB;EAChB,YAAY,iBAAiB;GAAE,YAAY;GAAY;GAAM;GAAS;EAAM,CAAC;EAC7E,YAAY;EACZ,aAAa;EACb;EACA;EACA;EACA;EACA,eAAe,CAAC;GAAE,MAAM;GAAY,KAAK;EAAY,CAAC;EACtD,oBAAoB;EACpB;EACA;CACF;AACF;;AAGA,SAAgB,uBACd,QACA,YACkB;CAClB,MAAM,MAAwB,CAAC;CAC/B,KAAK,MAAM,KAAK,OAAO,SAAS;EAC9B,MAAM,cAAc,EAAE,oBAAoB,EAAE;EAI5C,IAHkB,cAAc,EAAE,cAAc,EAAE,kBAGhC,GAChB,IAAI,KACF,QACE,eACA,EAAE,MACF,UAAU,EAAE,KAAK,+EACjB,QACA,IACA,YACA,sIACA,EAAE,MACF,iGACF,CACF;OACK,IAAI,gBAAgB,KAAK,EAAE,cAAc,EAAE,gBAAgB,GAEhE,IAAI,KACF,QACE,eACA,EAAE,MACF,UAAU,EAAE,KAAK,gFAAgF,EAAE,YAAY,cAAc,EAAE,cAAc,IAC7I,QACA,IACA,YACA,2JACA,EAAE,MACF,sEACF,CACF;EAIF,IAAI,eAAe,KAAK,CAAC,EAAE,mBACzB,IAAI,KACF,QACE,mBACA,EAAE,MACF,UAAU,EAAE,KAAK,mFACjB,UACA,IACA,YACA,oHACA,EAAE,IACJ,CACF;EAIF,IAAI,EAAE,SAAS,YAAY,EAAE,oBAAoB,GAC/C,IAAI,KACF,QACE,UACA,EAAE,MACF,iBAAiB,EAAE,KAAK,YAAY,EAAE,kBAAkB,+BACxD,QACA,KACA,YACA,uMACA,EAAE,IACJ,CACF;EAIF,IAAI,EAAE,QAAQ,wBAAwB,CAAC,EAAE,kBACvC,IAAI,KACF,QACE,mBACA,EAAE,MACF,UAAU,EAAE,KAAK,OAAO,EAAE,MAAM,4DAChC,UACA,IACA,YACA,mFAAmF,EAAE,MAAM,mDAC3F,EAAE,IACJ,CACF;EAIF,IAAI,CAAC,EAAE,aACL,IAAI,KACF,QACE,gBACA,EAAE,MACF,UAAU,EAAE,KAAK,oBACjB,OACA,IACA,YACA,wGACA,EAAE,IACJ,CACF;EAIF,IAAI,CAAC,EAAE,UACL,IAAI,KACF,QACE,iBACA,EAAE,MACF,UAAU,EAAE,KAAK,8CACjB,OACA,KACA,YACA,iJACA,EAAE,IACJ,CACF;CAEJ;CACA,OAAO;AACT;AAIA,IAAM,oBAAN,MAAyE;CACvE,KAAc;CACd,cACE;CACF,YAAqB;CACrB,OAAgB;EAAE,MAAM;EAA0B,iBAAiB;CAAE;CACrE,UAAmB;CACnB,kBAA2B;EACzB,MAAM;EACN,sBAAsB;EACtB,oBAAoB;CACtB;CAEA,MAAM,QAAQ,OAAyB,KAAgD;EACrF,MAAM,aAAa,IAAI,MAAM,+BAAc,IAAI,KAAK,EAAA,CAAE,YAAY;EAClE,IAAI,MACF,gBAAgB,MAAM,QAAQ,OAAO,eAAe,MAAM,oBAAoB,aAChF;EACA,OAAO,uBAAuB,OAAO,UAAU;CACjD;AACF;AAEA,MAAa,sBAAsB,IAAI,kBAAkB"}
+{"version":3,"file":"index.js","names":[],"sources":["../../src/analyst/adapters.ts","../../src/analyst/bind.ts","../../src/analyst/define.ts","../../src/analyst/kinds/skill-usage.ts"],"sourcesContent":["/**\n * Adapter factory — lifts the semantic-concept judge into the Analyst\n * contract without re-implementing it.\n *\n * It builds an Analyst with a stable id (caller chooses; a default is given),\n * a version derived from the judge's version plus an adapter revision, and an\n * `analyze()` that calls the judge and lifts its output to `AnalystFinding[]`\n * through `makeFinding()`. The judge's `Severity` ('critical' | 'major' |\n * 'minor' | 'info') projects onto `AnalystSeverity`: 'major' becomes 'high'\n * and 'minor' becomes 'medium'.\n *\n * The adapter owns no state. Calling the factory twice is safe.\n */\n\nimport { CostLedger } from '../cost-ledger'\nimport type { Severity as LayerSeverity } from '../multi-layer-verifier'\nimport {\n  runSemanticConceptJudge,\n  SEMANTIC_CONCEPT_JUDGE_VERSION,\n  type SemanticConceptJudgeInput,\n  type SemanticConceptJudgeOptions,\n  type SemanticConceptJudgeResult,\n} from '../semantic-concept-judge'\nimport type { Analyst, AnalystFinding, AnalystSeverity } from './types'\nimport { makeFinding } from './types'\nimport { settleUsageReceiptFromCostLedger, validateUsageSettlementTimeout } from './usage-receipt'\n\nconst ADAPTER_REV = '1'\n\n// ── Severity bridges ───────────────────────────────────────────────\n\nfunction liftSeverity(s: LayerSeverity): AnalystSeverity {\n  switch (s) {\n    case 'critical':\n      return 'critical'\n    case 'major':\n      return 'high'\n    case 'minor':\n      return 'medium'\n    case 'info':\n      return 'info'\n  }\n}\n\n// ── SemanticConceptJudge → Analyst ─────────────────────────────────\n\nexport interface SemanticConceptJudgeAdapterOpts {\n  id?: string\n  area?: string\n  /** Registry context owns cancellation and the per-analyst cost ledger. */\n  options: Omit<SemanticConceptJudgeOptions, 'costLedger' | 'signal'>\n  /** Maximum post-cancellation wait for a provider receipt. Default 5 seconds. */\n  settlementTimeoutMs?: number\n}\n\nexport function createSemanticConceptJudgeAdapter(\n  opts: SemanticConceptJudgeAdapterOpts,\n): Analyst<SemanticConceptJudgeInput> {\n  const id = opts.id ?? 'semantic-concept-judge'\n  const area = opts.area ?? 'concept-coverage'\n  const settlementTimeoutMs = validateUsageSettlementTimeout(opts.settlementTimeoutMs)\n  return {\n    id,\n    description:\n      'Runs the semantic-concept judge and surfaces missing / weak concepts as findings.',\n    inputKind: 'custom',\n    cost: {\n      kind: 'llm',\n      models: opts.options.model ? [opts.options.model] : undefined,\n      settlement_timeout_ms: settlementTimeoutMs,\n    },\n    version: `${SEMANTIC_CONCEPT_JUDGE_VERSION}-adapter-${ADAPTER_REV}`,\n    async analyze(input, ctx) {\n      const costLedger = new CostLedger(ctx.budgetUsd)\n      let result: SemanticConceptJudgeResult\n      try {\n        result = await runSemanticConceptJudge(input, {\n          ...opts.options,\n          costLedger,\n          signal: ctx.signal,\n        })\n      } finally {\n        const usage = await settleUsageReceiptFromCostLedger(costLedger, {\n          channel: 'judge',\n          timeoutMs: settlementTimeoutMs,\n        })\n        if (!usage.settled) {\n          ctx.log?.('semantic-concept judge provider settlement timed out', {\n            pending_calls: usage.pendingCalls,\n            timeout_ms: settlementTimeoutMs,\n          })\n        }\n        ctx.recordUsage?.(usage.receipt)\n      }\n      if (!result.available) {\n        return [\n          makeFinding({\n            analyst_id: id,\n            area,\n            claim: 'semantic-concept judge unavailable',\n            rationale: result.error,\n            severity: 'info',\n            confidence: 1,\n            evidence_refs: [],\n            metadata: { reason: result.error },\n          }),\n        ]\n      }\n      const out: AnalystFinding[] = []\n      for (const f of result.findings) {\n        // Only surface gaps: missing concepts or low scores. Concepts at\n        // 7+/10 with present=true are not findings — they're successes.\n        if (f.present && f.score >= 7) continue\n        out.push(\n          makeFinding({\n            analyst_id: id,\n            area,\n            subject: f.concept,\n            claim: f.present\n              ? `concept \"${f.concept}\" is weak (${f.score}/10)`\n              : `concept \"${f.concept}\" is missing`,\n            rationale: f.evidence,\n            severity: liftSeverity(f.severity),\n            confidence: 0.85,\n            evidence_refs: [{ kind: 'artifact', uri: 'inline:evidence', excerpt: f.evidence }],\n            metadata: {\n              concept: f.concept,\n              present: f.present,\n              score_10: f.score,\n            },\n          }),\n        )\n      }\n      return out\n    },\n  }\n}\n","/**\n * bindAnalyst — compile an `AnalystDefinition` into a runnable arm.\n *\n * The definition declares protocol (question, task text, reply grammar,\n * evidence projection, budget, repair turns); the transport binding supplies\n * execution (a bridge endpoint, or the caller-owned model path). Each\n * projection mode dispatches to the same strategy its arm's entry point runs,\n * so a compiled definition and the historical creator send byte-identical\n * requests — the definition parity suite asserts this per arm.\n *\n * A projection × transport pair with no strategy fails loud with\n * `AnalystExpressivenessError` naming the pair: an arm shape this layer cannot\n * express must be reported, never approximated.\n */\n\nimport type { CustomTokenPricing } from '../cost-ledger'\nimport type { AnalystBenchmarkRunner } from './benchmark'\nimport { runChunkedAnalystDefinition } from './benchmark-public-model'\nimport { runReplVariableAnalystDefinition } from './benchmark-public-rlm'\nimport type { PublicAnalystBenchmarkModelConfig } from './benchmark-public-types'\nimport { runInlineAnalystDefinition } from './benchmark-runner-prime'\nimport { type AnalystDefinition, AnalystExpressivenessError } from './definition'\nimport type { PrimeBridgeTransport } from './prime-bridge-transport'\nimport type { AnalystRunInputs } from './types'\n\n/** Execution half of a binding: who actually reaches the model. */\nexport type AnalystTransportBinding =\n  | {\n      /** An OpenAI-compatible cli-bridge endpoint (inline projections). */\n      readonly kind: 'prime-bridge'\n      readonly baseUrl: string\n      readonly model: string\n      readonly transport?: PrimeBridgeTransport\n      readonly pricing?: CustomTokenPricing\n    }\n  | {\n      /** The caller-owned model path (chunked and repl-variable projections). */\n      readonly kind: 'model-owner'\n      readonly config: PublicAnalystBenchmarkModelConfig\n    }\n\n/**\n * Compile a definition plus a transport binding into a runnable arm. Dispatch\n * is total over the expressible pairs; everything else is a loud refusal.\n */\nexport function bindAnalyst<TRow, TAssignment, TBlock>(\n  definition: AnalystDefinition<TRow, TAssignment, TBlock>,\n  transports: AnalystTransportBinding,\n): AnalystBenchmarkRunner<AnalystRunInputs> {\n  const mode = definition.projection.mode\n  if (mode === 'inline' && transports.kind === 'prime-bridge') {\n    return runInlineAnalystDefinition(definition, {\n      baseUrl: transports.baseUrl,\n      model: transports.model,\n      ...(transports.transport ? { transport: transports.transport } : {}),\n      ...(transports.pricing ? { pricing: transports.pricing } : {}),\n    })\n  }\n  if (mode === 'chunked' && transports.kind === 'model-owner') {\n    return runChunkedAnalystDefinition(definition, transports.config)\n  }\n  if (mode === 'repl-variable' && transports.kind === 'model-owner') {\n    return runReplVariableAnalystDefinition(\n      // The repl-variable strategy is written against the engine's raw-finding\n      // row type; the definition's own row declaration carries it.\n      definition as Parameters<typeof runReplVariableAnalystDefinition>[0],\n      transports.config,\n    )\n  }\n  throw new AnalystExpressivenessError(\n    `no strategy compiles a '${mode}' projection over a '${transports.kind}' transport; ` +\n      `definition '${definition.id}' cannot be expressed by this layer yet`,\n  )\n}\n","import type { TraceAnalysisStore } from '../trace-analyst/store'\nimport type { ExactCapableAnalyst } from './exact-types'\nimport type { TraceAnalystDefinition } from './kind-factory'\nimport type { Analyst, AnalystCost } from './types'\n\n/**\n * Define a reusable trace-research question.\n *\n * The returned value contains no model, credentials, or execution state. Bind\n * it to any TraceAnalysisEngine with `runTraceAnalyst` or\n * `createTraceAnalyst`.\n */\nexport function defineTraceAnalyst(definition: TraceAnalystDefinition): TraceAnalystDefinition {\n  for (const [name, value] of [\n    ['id', definition.id],\n    ['description', definition.description],\n    ['area', definition.area],\n    ['version', definition.version],\n    ['instructions', definition.instructions],\n  ] as const) {\n    if (typeof value !== 'string' || !value.trim()) {\n      throw new TypeError(`defineTraceAnalyst: ${name} must be a non-empty string`)\n    }\n  }\n  return {\n    ...definition,\n    limits: definition.limits ? { ...definition.limits } : undefined,\n  }\n}\n\nexport interface DefineCustomAnalystOptions {\n  id: string\n  description: string\n  version?: string\n  cost: AnalystCost\n  analyze: Analyst<TraceAnalysisStore>['analyze']\n}\n\nexport interface DefineExactCustomAnalystOptions extends DefineCustomAnalystOptions {\n  /** Canonical JSON for behavior knobs not already bound by `version`. */\n  executionConfig: Readonly<Record<string, unknown>>\n}\n\n/** Construct a registrable analyst from a hand-written analyze function. */\nexport function defineCustomAnalyst(\n  options: DefineExactCustomAnalystOptions,\n): ExactCapableAnalyst<TraceAnalysisStore>\nexport function defineCustomAnalyst(\n  options: DefineCustomAnalystOptions,\n): Analyst<TraceAnalysisStore>\nexport function defineCustomAnalyst(\n  options: DefineCustomAnalystOptions | DefineExactCustomAnalystOptions,\n): Analyst<TraceAnalysisStore> | ExactCapableAnalyst<TraceAnalysisStore> {\n  if (!options.id.trim()) throw new TypeError('defineCustomAnalyst: id must not be empty')\n  if (!options.description.trim()) {\n    throw new TypeError('defineCustomAnalyst: description must not be empty')\n  }\n  if (options.cost === undefined) {\n    throw new TypeError('defineCustomAnalyst: cost must be declared')\n  }\n  if (\n    'executionConfig' in options &&\n    (!options.executionConfig ||\n      typeof options.executionConfig !== 'object' ||\n      Array.isArray(options.executionConfig))\n  ) {\n    throw new TypeError('defineCustomAnalyst: executionConfig must be an object')\n  }\n  return {\n    id: options.id,\n    description: options.description,\n    version: options.version ?? '1.0.0',\n    inputKind: 'trace-store',\n    cost: options.cost,\n    analyze: options.analyze,\n    ...('executionConfig' in options ? { executionConfig: options.executionConfig } : {}),\n  }\n}\n","/**\n * Skill-usage analyst — a DETERMINISTIC `Analyst` over a Claude/Codex skill\n * library + its trace corpus. Unlike the trace-store kinds (failure-mode,\n * improvement, ...) this kind calls no LLM: it mines real usage and skill\n * structure and emits findings by rule.\n *\n * It exists because the naive \"Skill-tool invocation count\" lies low — it\n * misses orchestrated sub-dispatch (a leaf skill run BY /pursue or /governor\n * logs under the parent), slash-command entry, local-script bypass, and\n * on-disk artifacts. The 2026-05-30 skill audit found 39/53 skills at zero\n * direct invocations, yet only one was a genuine cut: the rest were\n * measurement-invisible or discovery-limited. This analyst encodes that\n * lesson as a multi-signal usage model so a cheap repeatable pass can keep\n * the library honest, and so the expensive audit workflow's verdicts can\n * GEPA-distill it toward agreement (see `gold/skill-verdicts.gold.jsonl`).\n *\n * Report-building (`buildSkillUsageReport`, an fs scan) is separated from\n * finding emission (`SkillUsageAnalyst.analyze`, pure) so the slow scan runs\n * once at the registry boundary and the rule logic stays unit-testable.\n */\n\nimport { type Dirent, existsSync, readdirSync, readFileSync, statSync } from 'node:fs'\nimport { join } from 'node:path'\nimport type { ExactCapableAnalyst } from '../exact-types'\nimport type { AnalystContext, AnalystFinding, AnalystSeverity } from '../types'\nimport { computeFindingId } from '../types'\n\n// ── Input model ──────────────────────────────────────────────────────\n\nexport type SkillKind = 'public' | 'private'\n\n/** One skill's multi-signal usage + structure. All counts are deterministic. */\nexport interface SkillUsageRecord {\n  name: string\n  kind: SkillKind\n  /** Absolute path to the skill's SKILL.md. */\n  path: string\n  lines: number\n  /** `\"skill\":\"<name>\"` Skill-tool invocations across the trace corpus. */\n  directInvocations: number\n  /** `<command-name>/<name>` slash invocations across the trace corpus. */\n  slashInvocations: number\n  /** Sibling skills whose SKILL.md dispatches to this one (`/<name>`). Proxy\n   *  for orchestrated sub-dispatch the per-skill counter cannot see. */\n  inboundRefs: number\n  /** On-disk artifacts attributable to the skill (e.g. `.evolve/<name>/**`). */\n  artifactCount: number\n  /** Tangle-private reference count in the body (leak signal for public skills). */\n  tanglePrivateRefs: number\n  hasReferencesDir: boolean\n  hasEvalsDir: boolean\n  /** Body mentions `skill-runs.jsonl` (visible to /reflect + /governor). */\n  logsRuns: boolean\n  /** Description carries an explicit `Triggers:` clause / trigger phrases. */\n  hasTriggerPhrases: boolean\n}\n\nexport interface SkillUsageReport {\n  generatedFromTraces: number\n  records: SkillUsageRecord[]\n}\n\nexport interface SkillUsageScanConfig {\n  /** Dirs holding `*.jsonl` transcripts (Claude `~/.claude/projects`, Codex sessions). */\n  transcriptDirs: string[]\n  /** Skill roots to scan; each dir directly under `root` with a `SKILL.md` is a skill. */\n  skillRoots: { root: string; kind: SkillKind }[]\n  /** Roots scanned for `<root>/.evolve/<skill>` artifact dirs. */\n  artifactRoots?: string[]\n  /** Token-prefixed mappings: skill name → extra artifact subpaths under an artifactRoot\n   *  (e.g. reflect → `.evolve/reflections`). Catches non-eponymous artifact dirs. */\n  artifactAliases?: Record<string, string[]>\n  /** Cap files read per transcript dir (bounds a huge corpus); 0 = unbounded. */\n  maxTranscriptsPerDir?: number\n}\n\n// ── Deterministic thresholds ─────────────────────────────────────────\n\n/** Anthropic's authoring guidance keeps SKILL.md short; past this with no\n *  `references/` split the body burns context budget every session. */\nconst BLOAT_LINE_THRESHOLD = 300\n\nconst TANGLE_PRIVATE_RE =\n  /\\b(cli-bridge|tangletools|ops-board|drew-gtr-pro|@tangle-network\\/|~\\/company|tangle\\.tools|gtm-agent)\\b|\\bkimi\\b|\\btcloud\\b/gi\nconst TRIGGER_RE = /triggers?\\s*[:-]/i\n\n// ── Report builder (fs scan — slow, runs once at the registry boundary) ──\n\nfunction listSkillDirs(root: string): { name: string; path: string }[] {\n  if (!existsSync(root)) return []\n  const out: { name: string; path: string }[] = []\n  for (const entry of readdirSync(root, { withFileTypes: true })) {\n    if (!entry.isDirectory() && !entry.isSymbolicLink()) continue\n    const skillMd = join(root, entry.name, 'SKILL.md')\n    if (existsSync(skillMd)) out.push({ name: entry.name, path: skillMd })\n  }\n  return out\n}\n\nfunction walkJsonl(dir: string, cap: number): string[] {\n  if (!existsSync(dir)) return []\n  const files: string[] = []\n  const stack = [dir]\n  while (stack.length) {\n    const cur = stack.pop()!\n    let entries: Dirent[]\n    try {\n      entries = readdirSync(cur, { withFileTypes: true })\n    } catch {\n      continue\n    }\n    for (const e of entries) {\n      const full = join(cur, e.name)\n      if (e.isDirectory()) stack.push(full)\n      else if (e.name.endsWith('.jsonl')) {\n        files.push(full)\n        if (cap > 0 && files.length >= cap) return files\n      }\n    }\n  }\n  return files\n}\n\nfunction frontmatterDescription(body: string): string {\n  const fm = /^---\\n([\\s\\S]*?)\\n---/.exec(body)\n  const block = fm?.[1] ?? ''\n  const m = /description:\\s*(.+)/i.exec(block)\n  return m?.[1] ?? ''\n}\n\nfunction countArtifacts(roots: string[], name: string, aliases: string[]): number {\n  let n = 0\n  for (const root of roots) {\n    const candidates = [join(root, '.evolve', name), ...aliases.map((a) => join(root, a))]\n    for (const dir of candidates) {\n      if (!existsSync(dir)) continue\n      try {\n        if (statSync(dir).isDirectory()) n += readdirSync(dir).length\n        else n += 1\n      } catch {\n        /* unreadable — skip */\n      }\n    }\n  }\n  return n\n}\n\n/** Scan the corpus + skill roots into a {@link SkillUsageReport}. Deterministic. */\nexport function buildSkillUsageReport(config: SkillUsageScanConfig): SkillUsageReport {\n  const skills = config.skillRoots.flatMap(({ root, kind }) =>\n    listSkillDirs(root).map((s) => ({ ...s, kind })),\n  )\n  const names = skills.map((s) => s.name)\n\n  // One pass over the corpus accumulating direct + slash counts per skill.\n  const direct = new Map<string, number>(names.map((n) => [n, 0]))\n  const slash = new Map<string, number>(names.map((n) => [n, 0]))\n  const skillRe = /\"skill\"\\s*:\\s*\"([a-z0-9_:-]+)\"/g\n  const cmdRe = /<command-name>\\/?([a-z0-9_:-]+)<\\/command-name>/g\n  let transcripts = 0\n  for (const dir of config.transcriptDirs) {\n    for (const file of walkJsonl(dir, config.maxTranscriptsPerDir ?? 0)) {\n      transcripts += 1\n      let data: string\n      try {\n        data = readFileSync(file, 'utf8')\n      } catch {\n        continue\n      }\n      for (const m of data.matchAll(skillRe)) {\n        const g = m[1]\n        if (!g) continue\n        const n = g.split(':').pop() ?? g\n        const prev = direct.get(n)\n        if (prev !== undefined) direct.set(n, prev + 1)\n      }\n      for (const m of data.matchAll(cmdRe)) {\n        const g = m[1]\n        if (g === undefined) continue\n        const prev = slash.get(g)\n        if (prev !== undefined) slash.set(g, prev + 1)\n      }\n    }\n  }\n\n  // Read each skill body once; compute structure + inbound refs across siblings.\n  const bodies = new Map<string, string>()\n  for (const s of skills) {\n    try {\n      bodies.set(s.name, readFileSync(s.path, 'utf8'))\n    } catch {\n      bodies.set(s.name, '')\n    }\n  }\n  const inbound = new Map<string, number>(names.map((n) => [n, 0]))\n  for (const target of names) {\n    const ref = new RegExp(`/${target}\\\\b|\\\\[\\\\[${target}\\\\]\\\\]`)\n    for (const s of skills) {\n      if (s.name === target) continue\n      if (ref.test(bodies.get(s.name) ?? '')) inbound.set(target, inbound.get(target)! + 1)\n    }\n  }\n\n  const records: SkillUsageRecord[] = skills.map((s) => {\n    const body = bodies.get(s.name) ?? ''\n    const dir = s.path.replace(/\\/SKILL\\.md$/, '')\n    return {\n      name: s.name,\n      kind: s.kind,\n      path: s.path,\n      lines: body ? body.split('\\n').length : 0,\n      directInvocations: direct.get(s.name) ?? 0,\n      slashInvocations: slash.get(s.name) ?? 0,\n      inboundRefs: inbound.get(s.name) ?? 0,\n      artifactCount: countArtifacts(\n        config.artifactRoots ?? [],\n        s.name,\n        config.artifactAliases?.[s.name] ?? [],\n      ),\n      tanglePrivateRefs: (body.match(TANGLE_PRIVATE_RE) ?? []).length,\n      hasReferencesDir: existsSync(join(dir, 'references')),\n      hasEvalsDir: existsSync(join(dir, 'evals')),\n      logsRuns: body.includes('skill-runs.jsonl'),\n      hasTriggerPhrases: TRIGGER_RE.test(frontmatterDescription(body) || body.slice(0, 600)),\n    }\n  })\n  return { generatedFromTraces: transcripts, records }\n}\n\n// ── Finding emission (pure — unit-testable, no LLM, no fs) ────────────\n\nconst ANALYST_ID = 'skill-usage'\n\nfunction finding(\n  area: string,\n  subject: string,\n  claim: string,\n  severity: AnalystSeverity,\n  confidence: number,\n  producedAt: string,\n  recommended: string,\n  evidenceUri: string,\n  rationale?: string,\n): AnalystFinding {\n  return {\n    schema_version: '1.0.0',\n    finding_id: computeFindingId({ analyst_id: ANALYST_ID, area, subject, claim }),\n    analyst_id: ANALYST_ID,\n    produced_at: producedAt,\n    severity,\n    area,\n    claim,\n    rationale,\n    evidence_refs: [{ kind: 'artifact', uri: evidenceUri }],\n    recommended_action: recommended,\n    confidence,\n    subject,\n  }\n}\n\n/** Pure rule pass over a report → findings. Exported for direct/unit use. */\nexport function emitSkillUsageFindings(\n  report: SkillUsageReport,\n  producedAt: string,\n): AnalystFinding[] {\n  const out: AnalystFinding[] = []\n  for (const r of report.records) {\n    const directTotal = r.directInvocations + r.slashInvocations\n    const trueUsage = directTotal + r.inboundRefs + r.artifactCount\n\n    // 1. Dead: no usage signal of ANY kind. The only real deprecation candidate.\n    if (trueUsage === 0) {\n      out.push(\n        finding(\n          'skill-usage',\n          r.name,\n          `Skill '${r.name}' has zero usage across all signals (direct, slash, inbound-refs, artifacts)`,\n          'high',\n          0.6,\n          producedAt,\n          'Confirm the skill covers a real recurring job; if not, deprecate. Zero true usage is the only deterministic deprecation candidate.',\n          r.path,\n          'No Skill-tool call, no slash invocation, no sibling dispatches to it, and no on-disk artifacts.',\n        ),\n      )\n    } else if (directTotal === 0 && r.inboundRefs + r.artifactCount > 0) {\n      // 2. Measurement-invisible: real use via orchestration/artifacts, never invoked directly.\n      out.push(\n        finding(\n          'skill-usage',\n          r.name,\n          `Skill '${r.name}' shows 0 direct invocations but is used via orchestration/artifacts (inbound=${r.inboundRefs}, artifacts=${r.artifactCount})`,\n          'info',\n          0.8,\n          producedAt,\n          'Do NOT treat as unused — usage is real but logged under parent skills or on disk. Strengthen direct-invocation discovery only if direct use is desired.',\n          r.path,\n          'The Skill-tool counter undercounts orchestrated/chained leaf skills.',\n        ),\n      )\n    }\n\n    // 3. Discovery gap: low direct use AND weak trigger surface.\n    if (directTotal <= 2 && !r.hasTriggerPhrases) {\n      out.push(\n        finding(\n          'discoverability',\n          r.name,\n          `Skill '${r.name}' is rarely invoked directly and its description has no explicit trigger phrases`,\n          'medium',\n          0.7,\n          producedAt,\n          'Add a `Triggers:` clause with verbatim user phrases to the frontmatter description so the model auto-invokes it.',\n          r.path,\n        ),\n      )\n    }\n\n    // 4. Public-repo leak.\n    if (r.kind === 'public' && r.tanglePrivateRefs > 0) {\n      out.push(\n        finding(\n          'safety',\n          r.name,\n          `Public skill '${r.name}' carries ${r.tanglePrivateRefs} Tangle-private reference(s)`,\n          'high',\n          0.75,\n          producedAt,\n          'Sanitize incidental internal refs (cli-bridge/kimi/tcloud/~company/private repos) or relocate to a private repo. Verify @tangle-network/* refs are to PUBLISHED packages before treating as a leak.',\n          r.path,\n        ),\n      )\n    }\n\n    // 5. Bloat / no progressive disclosure.\n    if (r.lines > BLOAT_LINE_THRESHOLD && !r.hasReferencesDir) {\n      out.push(\n        finding(\n          'maintainability',\n          r.name,\n          `Skill '${r.name}' is ${r.lines} lines with no references/ split (progressive disclosure)`,\n          'medium',\n          0.8,\n          producedAt,\n          `Split detail into references/ loaded on demand; keep SKILL.md a short overview. ${r.lines} lines load into every session's context budget.`,\n          r.path,\n        ),\n      )\n    }\n\n    // 6. No evals (Anthropic's \">=3 evals before docs\" rule).\n    if (!r.hasEvalsDir) {\n      out.push(\n        finding(\n          'data-quality',\n          r.name,\n          `Skill '${r.name}' ships no evals/`,\n          'low',\n          0.6,\n          producedAt,\n          'Add evals/evals.json with >=3 scenarios proving the skill beats baseline; gives regression coverage.',\n          r.path,\n        ),\n      )\n    }\n\n    // 7. No run logging → invisible to /reflect and /governor.\n    if (!r.logsRuns) {\n      out.push(\n        finding(\n          'observability',\n          r.name,\n          `Skill '${r.name}' never appends to .evolve/skill-runs.jsonl`,\n          'low',\n          0.55,\n          producedAt,\n          'Append one run line to .evolve/skill-runs.jsonl on completion, or declare it a non-logging leaf, so the self-improvement loop can see it ran.',\n          r.path,\n        ),\n      )\n    }\n  }\n  return out\n}\n\n// ── The Analyst ──────────────────────────────────────────────────────\n\nclass SkillUsageAnalyst implements ExactCapableAnalyst<SkillUsageReport> {\n  readonly id = ANALYST_ID\n  readonly description =\n    'Deterministic multi-signal skill-usage analysis: flags dead skills, measurement-invisible (orchestrated) usage, discovery gaps, public-repo leaks, bloat, missing evals, and missing run-logging.'\n  readonly inputKind = 'custom' as const\n  readonly cost = { kind: 'deterministic' as const, est_usd_per_run: 0 }\n  readonly version = '1.0.0'\n  readonly executionConfig = {\n    kind: 'skill-usage',\n    bloat_line_threshold: BLOAT_LINE_THRESHOLD,\n    produced_at_source: 'tags.producedAt-or-system-clock',\n  } as const\n\n  async analyze(input: SkillUsageReport, ctx: AnalystContext): Promise<AnalystFinding[]> {\n    const producedAt = ctx.tags?.producedAt ?? new Date().toISOString()\n    ctx.log?.(\n      `skill-usage: ${input.records.length} skills over ${input.generatedFromTraces} transcripts`,\n    )\n    return emitSkillUsageFindings(input, producedAt)\n  }\n}\n\nexport const SKILL_USAGE_ANALYST = new SkillUsageAnalyst()\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA2BA,MAAM,cAAc;AAIpB,SAAS,aAAa,GAAmC;CACvD,QAAQ,GAAR;EACE,KAAK,YACH,OAAO;EACT,KAAK,SACH,OAAO;EACT,KAAK,SACH,OAAO;EACT,KAAK,QACH,OAAO;CACX;AACF;AAaA,SAAgB,kCACd,MACoC;CACpC,MAAM,KAAK,KAAK,MAAM;CACtB,MAAM,OAAO,KAAK,QAAQ;CAC1B,MAAM,sBAAsB,+BAA+B,KAAK,mBAAmB;CACnF,OAAO;EACL;EACA,aACE;EACF,WAAW;EACX,MAAM;GACJ,MAAM;GACN,QAAQ,KAAK,QAAQ,QAAQ,CAAC,KAAK,QAAQ,KAAK,IAAI,KAAA;GACpD,uBAAuB;EACzB;EACA,SAAS,GAAG,+BAA+B,WAAW;EACtD,MAAM,QAAQ,OAAO,KAAK;GACxB,MAAM,aAAa,IAAI,WAAW,IAAI,SAAS;GAC/C,IAAI;GACJ,IAAI;IACF,SAAS,MAAM,wBAAwB,OAAO;KAC5C,GAAG,KAAK;KACR;KACA,QAAQ,IAAI;IACd,CAAC;GACH,UAAU;IACR,MAAM,QAAQ,MAAM,iCAAiC,YAAY;KAC/D,SAAS;KACT,WAAW;IACb,CAAC;IACD,IAAI,CAAC,MAAM,SACT,IAAI,MAAM,wDAAwD;KAChE,eAAe,MAAM;KACrB,YAAY;IACd,CAAC;IAEH,IAAI,cAAc,MAAM,OAAO;GACjC;GACA,IAAI,CAAC,OAAO,WACV,OAAO,CACL,YAAY;IACV,YAAY;IACZ;IACA,OAAO;IACP,WAAW,OAAO;IAClB,UAAU;IACV,YAAY;IACZ,eAAe,CAAC;IAChB,UAAU,EAAE,QAAQ,OAAO,MAAM;GACnC,CAAC,CACH;GAEF,MAAM,MAAwB,CAAC;GAC/B,KAAK,MAAM,KAAK,OAAO,UAAU;IAG/B,IAAI,EAAE,WAAW,EAAE,SAAS,GAAG;IAC/B,IAAI,KACF,YAAY;KACV,YAAY;KACZ;KACA,SAAS,EAAE;KACX,OAAO,EAAE,UACL,YAAY,EAAE,QAAQ,aAAa,EAAE,MAAM,QAC3C,YAAY,EAAE,QAAQ;KAC1B,WAAW,EAAE;KACb,UAAU,aAAa,EAAE,QAAQ;KACjC,YAAY;KACZ,eAAe,CAAC;MAAE,MAAM;MAAY,KAAK;MAAmB,SAAS,EAAE;KAAS,CAAC;KACjF,UAAU;MACR,SAAS,EAAE;MACX,SAAS,EAAE;MACX,UAAU,EAAE;KACd;IACF,CAAC,CACH;GACF;GACA,OAAO;EACT;CACF;AACF;;;;;;;AC3FA,SAAgB,YACd,YACA,YAC0C;CAC1C,MAAM,OAAO,WAAW,WAAW;CACnC,IAAI,SAAS,YAAY,WAAW,SAAS,gBAC3C,OAAO,2BAA2B,YAAY;EAC5C,SAAS,WAAW;EACpB,OAAO,WAAW;EAClB,GAAI,WAAW,YAAY,EAAE,WAAW,WAAW,UAAU,IAAI,CAAC;EAClE,GAAI,WAAW,UAAU,EAAE,SAAS,WAAW,QAAQ,IAAI,CAAC;CAC9D,CAAC;CAEH,IAAI,SAAS,aAAa,WAAW,SAAS,eAC5C,OAAO,4BAA4B,YAAY,WAAW,MAAM;CAElE,IAAI,SAAS,mBAAmB,WAAW,SAAS,eAClD,OAAO,iCAGL,YACA,WAAW,MACb;CAEF,MAAM,IAAI,2BACR,2BAA2B,KAAK,uBAAuB,WAAW,KAAK,2BACtD,WAAW,GAAG,wCACjC;AACF;;;;;;;;;;AC7DA,SAAgB,mBAAmB,YAA4D;CAC7F,KAAK,MAAM,CAAC,MAAM,UAAU;EAC1B,CAAC,MAAM,WAAW,EAAE;EACpB,CAAC,eAAe,WAAW,WAAW;EACtC,CAAC,QAAQ,WAAW,IAAI;EACxB,CAAC,WAAW,WAAW,OAAO;EAC9B,CAAC,gBAAgB,WAAW,YAAY;CAC1C,GACE,IAAI,OAAO,UAAU,YAAY,CAAC,MAAM,KAAK,GAC3C,MAAM,IAAI,UAAU,uBAAuB,KAAK,4BAA4B;CAGhF,OAAO;EACL,GAAG;EACH,QAAQ,WAAW,SAAS,EAAE,GAAG,WAAW,OAAO,IAAI,KAAA;CACzD;AACF;AAsBA,SAAgB,oBACd,SACuE;CACvE,IAAI,CAAC,QAAQ,GAAG,KAAK,GAAG,MAAM,IAAI,UAAU,2CAA2C;CACvF,IAAI,CAAC,QAAQ,YAAY,KAAK,GAC5B,MAAM,IAAI,UAAU,oDAAoD;CAE1E,IAAI,QAAQ,SAAS,KAAA,GACnB,MAAM,IAAI,UAAU,4CAA4C;CAElE,IACE,qBAAqB,YACpB,CAAC,QAAQ,mBACR,OAAO,QAAQ,oBAAoB,YACnC,MAAM,QAAQ,QAAQ,eAAe,IAEvC,MAAM,IAAI,UAAU,wDAAwD;CAE9E,OAAO;EACL,IAAI,QAAQ;EACZ,aAAa,QAAQ;EACrB,SAAS,QAAQ,WAAW;EAC5B,WAAW;EACX,MAAM,QAAQ;EACd,SAAS,QAAQ;EACjB,GAAI,qBAAqB,UAAU,EAAE,iBAAiB,QAAQ,gBAAgB,IAAI,CAAC;CACrF;AACF;;;;;;;;;;;;;;;;;;;;;;;;;ACGA,MAAM,uBAAuB;AAE7B,MAAM,oBACJ;AACF,MAAM,aAAa;AAInB,SAAS,cAAc,MAAgD;CACrE,IAAI,CAAC,WAAW,IAAI,GAAG,OAAO,CAAC;CAC/B,MAAM,MAAwC,CAAC;CAC/C,KAAK,MAAM,SAAS,YAAY,MAAM,EAAE,eAAe,KAAK,CAAC,GAAG;EAC9D,IAAI,CAAC,MAAM,YAAY,KAAK,CAAC,MAAM,eAAe,GAAG;EACrD,MAAM,UAAU,KAAK,MAAM,MAAM,MAAM,UAAU;EACjD,IAAI,WAAW,OAAO,GAAG,IAAI,KAAK;GAAE,MAAM,MAAM;GAAM,MAAM;EAAQ,CAAC;CACvE;CACA,OAAO;AACT;AAEA,SAAS,UAAU,KAAa,KAAuB;CACrD,IAAI,CAAC,WAAW,GAAG,GAAG,OAAO,CAAC;CAC9B,MAAM,QAAkB,CAAC;CACzB,MAAM,QAAQ,CAAC,GAAG;CAClB,OAAO,MAAM,QAAQ;EACnB,MAAM,MAAM,MAAM,IAAI;EACtB,IAAI;EACJ,IAAI;GACF,UAAU,YAAY,KAAK,EAAE,eAAe,KAAK,CAAC;EACpD,QAAQ;GACN;EACF;EACA,KAAK,MAAM,KAAK,SAAS;GACvB,MAAM,OAAO,KAAK,KAAK,EAAE,IAAI;GAC7B,IAAI,EAAE,YAAY,GAAG,MAAM,KAAK,IAAI;QAC/B,IAAI,EAAE,KAAK,SAAS,QAAQ,GAAG;IAClC,MAAM,KAAK,IAAI;IACf,IAAI,MAAM,KAAK,MAAM,UAAU,KAAK,OAAO;GAC7C;EACF;CACF;CACA,OAAO;AACT;AAEA,SAAS,uBAAuB,MAAsB;CAEpD,MAAM,QADK,wBAAwB,KAAK,IACzB,CAAC,GAAG,MAAM;CAEzB,OADU,uBAAuB,KAAK,KAC/B,CAAC,GAAG,MAAM;AACnB;AAEA,SAAS,eAAe,OAAiB,MAAc,SAA2B;CAChF,IAAI,IAAI;CACR,KAAK,MAAM,QAAQ,OAAO;EACxB,MAAM,aAAa,CAAC,KAAK,MAAM,WAAW,IAAI,GAAG,GAAG,QAAQ,KAAK,MAAM,KAAK,MAAM,CAAC,CAAC,CAAC;EACrF,KAAK,MAAM,OAAO,YAAY;GAC5B,IAAI,CAAC,WAAW,GAAG,GAAG;GACtB,IAAI;IACF,IAAI,SAAS,GAAG,CAAC,CAAC,YAAY,GAAG,KAAK,YAAY,GAAG,CAAC,CAAC;SAClD,KAAK;GACZ,QAAQ,CAER;EACF;CACF;CACA,OAAO;AACT;;AAGA,SAAgB,sBAAsB,QAAgD;CACpF,MAAM,SAAS,OAAO,WAAW,SAAS,EAAE,MAAM,WAChD,cAAc,IAAI,CAAC,CAAC,KAAK,OAAO;EAAE,GAAG;EAAG;CAAK,EAAE,CACjD;CACA,MAAM,QAAQ,OAAO,KAAK,MAAM,EAAE,IAAI;CAGtC,MAAM,SAAS,IAAI,IAAoB,MAAM,KAAK,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC;CAC/D,MAAM,QAAQ,IAAI,IAAoB,MAAM,KAAK,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC;CAC9D,MAAM,UAAU;CAChB,MAAM,QAAQ;CACd,IAAI,cAAc;CAClB,KAAK,MAAM,OAAO,OAAO,gBACvB,KAAK,MAAM,QAAQ,UAAU,KAAK,OAAO,wBAAwB,CAAC,GAAG;EACnE,eAAe;EACf,IAAI;EACJ,IAAI;GACF,OAAO,aAAa,MAAM,MAAM;EAClC,QAAQ;GACN;EACF;EACA,KAAK,MAAM,KAAK,KAAK,SAAS,OAAO,GAAG;GACtC,MAAM,IAAI,EAAE;GACZ,IAAI,CAAC,GAAG;GACR,MAAM,IAAI,EAAE,MAAM,GAAG,CAAC,CAAC,IAAI,KAAK;GAChC,MAAM,OAAO,OAAO,IAAI,CAAC;GACzB,IAAI,SAAS,KAAA,GAAW,OAAO,IAAI,GAAG,OAAO,CAAC;EAChD;EACA,KAAK,MAAM,KAAK,KAAK,SAAS,KAAK,GAAG;GACpC,MAAM,IAAI,EAAE;GACZ,IAAI,MAAM,KAAA,GAAW;GACrB,MAAM,OAAO,MAAM,IAAI,CAAC;GACxB,IAAI,SAAS,KAAA,GAAW,MAAM,IAAI,GAAG,OAAO,CAAC;EAC/C;CACF;CAIF,MAAM,yBAAS,IAAI,IAAoB;CACvC,KAAK,MAAM,KAAK,QACd,IAAI;EACF,OAAO,IAAI,EAAE,MAAM,aAAa,EAAE,MAAM,MAAM,CAAC;CACjD,QAAQ;EACN,OAAO,IAAI,EAAE,MAAM,EAAE;CACvB;CAEF,MAAM,UAAU,IAAI,IAAoB,MAAM,KAAK,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC;CAChE,KAAK,MAAM,UAAU,OAAO;EAC1B,MAAM,MAAM,IAAI,OAAO,IAAI,OAAO,YAAY,OAAO,OAAO;EAC5D,KAAK,MAAM,KAAK,QAAQ;GACtB,IAAI,EAAE,SAAS,QAAQ;GACvB,IAAI,IAAI,KAAK,OAAO,IAAI,EAAE,IAAI,KAAK,EAAE,GAAG,QAAQ,IAAI,QAAQ,QAAQ,IAAI,MAAM,IAAK,CAAC;EACtF;CACF;CAEA,MAAM,UAA8B,OAAO,KAAK,MAAM;EACpD,MAAM,OAAO,OAAO,IAAI,EAAE,IAAI,KAAK;EACnC,MAAM,MAAM,EAAE,KAAK,QAAQ,gBAAgB,EAAE;EAC7C,OAAO;GACL,MAAM,EAAE;GACR,MAAM,EAAE;GACR,MAAM,EAAE;GACR,OAAO,OAAO,KAAK,MAAM,IAAI,CAAC,CAAC,SAAS;GACxC,mBAAmB,OAAO,IAAI,EAAE,IAAI,KAAK;GACzC,kBAAkB,MAAM,IAAI,EAAE,IAAI,KAAK;GACvC,aAAa,QAAQ,IAAI,EAAE,IAAI,KAAK;GACpC,eAAe,eACb,OAAO,iBAAiB,CAAC,GACzB,EAAE,MACF,OAAO,kBAAkB,EAAE,SAAS,CAAC,CACvC;GACA,oBAAoB,KAAK,MAAM,iBAAiB,KAAK,CAAC,EAAA,CAAG;GACzD,kBAAkB,WAAW,KAAK,KAAK,YAAY,CAAC;GACpD,aAAa,WAAW,KAAK,KAAK,OAAO,CAAC;GAC1C,UAAU,KAAK,SAAS,kBAAkB;GAC1C,mBAAmB,WAAW,KAAK,uBAAuB,IAAI,KAAK,KAAK,MAAM,GAAG,GAAG,CAAC;EACvF;CACF,CAAC;CACD,OAAO;EAAE,qBAAqB;EAAa;CAAQ;AACrD;AAIA,MAAM,aAAa;AAEnB,SAAS,QACP,MACA,SACA,OACA,UACA,YACA,YACA,aACA,aACA,WACgB;CAChB,OAAO;EACL,gBAAgB;EAChB,YAAY,iBAAiB;GAAE,YAAY;GAAY;GAAM;GAAS;EAAM,CAAC;EAC7E,YAAY;EACZ,aAAa;EACb;EACA;EACA;EACA;EACA,eAAe,CAAC;GAAE,MAAM;GAAY,KAAK;EAAY,CAAC;EACtD,oBAAoB;EACpB;EACA;CACF;AACF;;AAGA,SAAgB,uBACd,QACA,YACkB;CAClB,MAAM,MAAwB,CAAC;CAC/B,KAAK,MAAM,KAAK,OAAO,SAAS;EAC9B,MAAM,cAAc,EAAE,oBAAoB,EAAE;EAI5C,IAHkB,cAAc,EAAE,cAAc,EAAE,kBAGhC,GAChB,IAAI,KACF,QACE,eACA,EAAE,MACF,UAAU,EAAE,KAAK,+EACjB,QACA,IACA,YACA,sIACA,EAAE,MACF,iGACF,CACF;OACK,IAAI,gBAAgB,KAAK,EAAE,cAAc,EAAE,gBAAgB,GAEhE,IAAI,KACF,QACE,eACA,EAAE,MACF,UAAU,EAAE,KAAK,gFAAgF,EAAE,YAAY,cAAc,EAAE,cAAc,IAC7I,QACA,IACA,YACA,2JACA,EAAE,MACF,sEACF,CACF;EAIF,IAAI,eAAe,KAAK,CAAC,EAAE,mBACzB,IAAI,KACF,QACE,mBACA,EAAE,MACF,UAAU,EAAE,KAAK,mFACjB,UACA,IACA,YACA,oHACA,EAAE,IACJ,CACF;EAIF,IAAI,EAAE,SAAS,YAAY,EAAE,oBAAoB,GAC/C,IAAI,KACF,QACE,UACA,EAAE,MACF,iBAAiB,EAAE,KAAK,YAAY,EAAE,kBAAkB,+BACxD,QACA,KACA,YACA,uMACA,EAAE,IACJ,CACF;EAIF,IAAI,EAAE,QAAQ,wBAAwB,CAAC,EAAE,kBACvC,IAAI,KACF,QACE,mBACA,EAAE,MACF,UAAU,EAAE,KAAK,OAAO,EAAE,MAAM,4DAChC,UACA,IACA,YACA,mFAAmF,EAAE,MAAM,mDAC3F,EAAE,IACJ,CACF;EAIF,IAAI,CAAC,EAAE,aACL,IAAI,KACF,QACE,gBACA,EAAE,MACF,UAAU,EAAE,KAAK,oBACjB,OACA,IACA,YACA,wGACA,EAAE,IACJ,CACF;EAIF,IAAI,CAAC,EAAE,UACL,IAAI,KACF,QACE,iBACA,EAAE,MACF,UAAU,EAAE,KAAK,8CACjB,OACA,KACA,YACA,iJACA,EAAE,IACJ,CACF;CAEJ;CACA,OAAO;AACT;AAIA,IAAM,oBAAN,MAAyE;CACvE,KAAc;CACd,cACE;CACF,YAAqB;CACrB,OAAgB;EAAE,MAAM;EAA0B,iBAAiB;CAAE;CACrE,UAAmB;CACnB,kBAA2B;EACzB,MAAM;EACN,sBAAsB;EACtB,oBAAoB;CACtB;CAEA,MAAM,QAAQ,OAAyB,KAAgD;EACrF,MAAM,aAAa,IAAI,MAAM,+BAAc,IAAI,KAAK,EAAA,CAAE,YAAY;EAClE,IAAI,MACF,gBAAgB,MAAM,QAAQ,OAAO,eAAe,MAAM,oBAAoB,aAChF;EACA,OAAO,uBAAuB,OAAO,UAAU;CACjD;AACF;AAEA,MAAa,sBAAsB,IAAI,kBAAkB"}

package/dist/analyze-runs-CrYhqXop.d.ts

@@ -0,0 +1,75 @@
+import { a as RunRecord } from "./run-record-D9wS35fK.js";
+import { r as DatasetScenario } from "./dataset-DQqhOCPt.js";
+import "./summary-report-BzFS5UhZ.js";
+import { o as InsightReport, r as ExecutionInsight, t as CostProvenanceSummary } from "./insight-report-C3Bfl762.js";
+import { n as AnalystRegistry } from "./registry-D5Ly3GRV.js";
+//#region src/contract/analyze-runs.d.ts
+interface AnalyzeRunsOptions {
+  /** The runs to analyze. */
+  runs: RunRecord[];
+  /** Which split to score against when reading composite from RunOutcome.
+   *  Default: holdout when ANY run has a `holdoutScore`, else search. */
+  split?: 'search' | 'holdout' | 'auto';
+  /** Pairwise analysis configuration. When both `baselineCandidateId` and
+   *  `candidateCandidateId` are present, lift is computed on paired
+   *  (experimentId, scenarioId, seed) identities shared between the two sides.
+   *  Unmatched rows remain visible in the lift result. */
+  baselineCandidateId?: string;
+  candidateCandidateId?: string;
+  /** Average matched run scores within these units before estimating lift.
+   *  Raw run distributions and unmatched-run counts remain unchanged. */
+  independentUnitByScenarioId?: ReadonlyMap<string, string>;
+  /** Canary scenarios — checked against every run's raw output for
+   *  holdout contamination. */
+  canaryScenarios?: DatasetScenario[];
+  /** Analyst registry for failure clustering. When omitted, the
+   *  `failureClusters` section is left undefined. */
+  analyst?: AnalystRegistry;
+  /** Downstream outcome metric per run (e.g. engagement rate, approval
+   *  rate, downstream pass rate). When present, the report includes
+   *  `outcomeCorrelation` + a simple linear reward model fit. */
+  outcomeSignal?: {
+    metric: string;
+    valueByRunId: Record<string, number>;
+  };
+  /** Multi-rater feedback for inter-rater agreement. Each entry is one
+   *  rater's score for one run. Two or more raters → kappa + disagreement
+   *  triage list. */
+  raterScores?: Array<{
+    runId: string;
+    rater: string;
+    score: number;
+  }>;
+  /** Number of histogram bins for distributional summaries. Default 12. */
+  histogramBins?: number;
+  /** Decision threshold — the smallest composite lift the caller cares
+   *  about. Used by the recommendations engine to call ship vs hold.
+   *  Default 0.02. */
+  decisionThreshold?: number;
+  /** Optional prior-period runs. When set, the report includes
+   *  `priorPeriodComparison` with per-metric Welch-CI deltas and
+   *  recommendations fire on statistically significant regressions.
+   *  The two windows do NOT have to share scenarios — the comparison
+   *  is two-sample unpaired (the substrate's `lift` field uses paired
+   *  bootstrap on shared (experimentId, scenarioId, seed) identities; this is the
+   *  shape for "this week vs last week" rather than "candidate vs
+   *  baseline within a campaign"). */
+  baselineRuns?: RunRecord[];
+  /** Human-readable label for the baseline window, e.g. "vs prior 7
+   *  days", "vs v3.1 release". Surfaces in recommendations + UI. */
+  baselineLabel?: string;
+}
+interface SummarizeExecutionOptions {
+  runs: RunRecord[];
+  histogramBins?: number;
+}
+interface ExecutionReport {
+  execution: ExecutionInsight;
+  costProvenance: CostProvenanceSummary;
+}
+/** Summarize runtime facts without interpreting task quality or promotion readiness. */
+declare function summarizeExecution(opts: SummarizeExecutionOptions): ExecutionReport;
+declare function analyzeRuns(opts: AnalyzeRunsOptions): Promise<InsightReport>;
+//#endregion
+export { summarizeExecution as a, analyzeRuns as i, ExecutionReport as n, SummarizeExecutionOptions as r, AnalyzeRunsOptions as t };
+//# sourceMappingURL=analyze-runs-CrYhqXop.d.ts.map

package/dist/analyze-runs-CrYhqXop.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"analyze-runs-CrYhqXop.d.ts","names":[],"sources":["../src/contract/analyze-runs.ts"],"mappings":";;;;;;UAqEiB;;EAEf,MAAM;;;EAGN;;;;;EAKA;EACA;;;EAGA,8BAA8B;;;EAG9B,kBAAkB;;;EAGlB,UAAU;;;;EAIV;IACE;IACA,cAAc;;;;;EAKhB,cAAc;IAAQ;IAAe;IAAe;;;EAEpD;;;;EAIA;;;;;;;;;EASA,eAAe;;;EAGf;;UAGe;EACf,MAAM;EACN;;UAGe;EACf,WAAW;EACX,gBAAgB;;;iBAIF,mBAAmB,MAAM,4BAA4B;iBAgB/C,YAAY,MAAM,qBAAqB,QAAQ"}