@tangle-network/agent-eval 0.190.0 → 0.190.1

package/dist/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "@tangle-network/agent-eval — wire protocol",
-    "version": "0.190.0",
+    "version": "0.190.1",
     "description": "HTTP and stdio RPC interface to agent-eval. The TypeScript runtime is the source of truth; this spec is the contract that cross-language clients (Python, Rust, Go) generate from.\n\nWire-protocol version: 1.1.0. Bumps on breaking changes to request/response schemas.",
     "contact": {
       "name": "Tangle Network",

package/dist/pipelines/index.d.ts

@@ -1,7 +1,7 @@
 import { f as Run } from "../schema-2g7trUwH.js";
 import { a as RunFilter, s as TraceStore } from "../store-C4avFjIi.js";
 import { n as FailureClusterReport, r as failureClusterView, t as FailureCluster } from "../failure-cluster-CbgEDcFJ.js";
-import { t as DiagnosisSpan } from "../spans-D1RftP_w.js";
+import { t as DiagnosisSpan } from "../spans-BWtjk_9c.js";
 import { n as TrajectoryStep } from "../trajectory-Y1fmdlw7.js";
 import { c as computeToolUseMetrics, d as judgeAgreementView, f as BudgetBreachFinding, g as BaselineReport, h as BaselineOptions, i as toolWasteView, l as JudgeAgreementReport, m as budgetBreachView, n as ToolWasteOptions, p as BudgetBreachReport, r as ToolWasteReport, t as ToolWasteFinding, u as JudgePair } from "../tool-waste-CP5DO383.js";
 //#region src/pipelines/first-divergence.d.ts

package/dist/pipelines/index.js

@@ -167,7 +167,7 @@ function diffStepsFromSpans(spans) {
 		if (list) list.push(span);
 		else children.set(parent, [span]);
 	}
-	const byStart = (x, y) => (x.startMs ?? 0) - (y.startMs ?? 0) || x.spanId.localeCompare(y.spanId);
+	const byStart = (x, y) => (x.startMs ?? 0) - (y.startMs ?? 0) || (x.endMs ?? 0) - (y.endMs ?? 0) || x.kind.localeCompare(y.kind) || x.name.localeCompare(y.name) || x.spanId.localeCompare(y.spanId);
 	for (const list of children.values()) list.sort(byStart);
 	const ordered = [];
 	const visited = /* @__PURE__ */ new Set();

package/dist/pipelines/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","names":[],"sources":["../../src/pipelines/first-divergence.ts","../../src/pipelines/regression.ts","../../src/pipelines/stuck-loop.ts"],"sourcesContent":["/**\n * Step diff between two runs, with the first divergence marked.\n *\n * Steps pair in three passes, each over the steps still unpaired:\n *   1. `id`: the same step id in both runs, e.g. a forked run that kept its\n *      parent's span ids.\n *   2. `position`: the same index with the same name and kind.\n *   3. `name`: the same name and kind anywhere, first unpaired step in order.\n * A step left unpaired exists in only one run. The first divergence is the\n * first index at which the two runs stop agreeing step for step, and it says\n * how: a paired step changed, one run replaced, added or removed a step, or the\n * runs ran the same steps in a different order.\n *\n * `diffSteps` reads plain `DiffStep` lists, so the same diff serves agent-eval\n * trace stores (`firstDivergenceView`) and flat OTLP spans\n * (`diffStepsFromSpans`). Use it to attribute \"why is variant B better?\" to a\n * step, and to compare sibling or parent and child runs of a search tree.\n */\n\nimport type { DiagnosisSpan } from '../diagnosis/spans'\nimport type { Span } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\nimport { buildTrajectory, type TrajectoryStep } from '../trajectory'\n\n/** One step of a run, as the diff reads it. */\nexport interface DiffStep {\n  /** The step's id within its run. */\n  id: string\n  name: string\n  kind: string\n  /** Fields compared once two steps pair, such as status, tool or model. */\n  fields: Readonly<Record<string, string | number | boolean | null>>\n}\n\nexport type StepPairing = 'id' | 'position' | 'name'\n\nexport interface FieldDifference {\n  field: string\n  a: string | number | boolean | null\n  b: string | number | boolean | null\n}\n\nexport interface StepPair {\n  /** Index in run A. */\n  a: number\n  /** Index in run B. */\n  b: number\n  pairedBy: StepPairing\n  /** Empty when the paired steps agree on name, kind and every field. */\n  differences: FieldDifference[]\n}\n\nexport type StepDivergenceKind = 'changed' | 'replaced' | 'only-in-a' | 'only-in-b' | 'reordered'\n\nexport interface StepDivergence {\n  /** The first index at which the runs stop agreeing; equals `commonPrefixLen`. */\n  index: number\n  kind: StepDivergenceKind\n  /** Index of the step at `index` in each run, or null when that run has ended. */\n  a: number | null\n  b: number | null\n  /** Set when `kind` is `changed`. */\n  differences: FieldDifference[]\n  reason: string\n}\n\nexport interface StepDiff {\n  /** Paired steps in run A order. */\n  pairs: StepPair[]\n  /** Indexes of run A steps with no partner in run B. */\n  onlyInA: number[]\n  /** Indexes of run B steps with no partner in run A. */\n  onlyInB: number[]\n  /** Leading steps paired index for index with no difference. */\n  commonPrefixLen: number\n  /** Null when the runs agree step for step. */\n  firstDivergence: StepDivergence | null\n}\n\nexport function diffSteps(a: readonly DiffStep[], b: readonly DiffStep[]): StepDiff {\n  const partnerOfA = new Map<number, { b: number; pairedBy: StepPairing }>()\n  const pairedB = new Set<number>()\n  const pair = (i: number, j: number, pairedBy: StepPairing) => {\n    partnerOfA.set(i, { b: j, pairedBy })\n    pairedB.add(j)\n  }\n\n  const bById = new Map<string, number[]>()\n  b.forEach((step, j) => {\n    const list = bById.get(step.id)\n    if (list) list.push(j)\n    else bById.set(step.id, [j])\n  })\n  a.forEach((step, i) => {\n    const j = bById.get(step.id)?.find((candidate) => !pairedB.has(candidate))\n    if (j !== undefined) pair(i, j, 'id')\n  })\n  a.forEach((step, i) => {\n    if (partnerOfA.has(i) || i >= b.length || pairedB.has(i)) return\n    if (sameShape(step, b[i]!)) pair(i, i, 'position')\n  })\n  a.forEach((step, i) => {\n    if (partnerOfA.has(i)) return\n    const j = b.findIndex((candidate, index) => !pairedB.has(index) && sameShape(step, candidate))\n    if (j >= 0) pair(i, j, 'name')\n  })\n\n  const pairs: StepPair[] = [...partnerOfA.entries()]\n    .sort(([x], [y]) => x - y)\n    .map(([i, partner]) => ({\n      a: i,\n      b: partner.b,\n      pairedBy: partner.pairedBy,\n      differences: differencesBetween(a[i]!, b[partner.b]!),\n    }))\n  const pairByA = new Map(pairs.map((entry) => [entry.a, entry]))\n  const partnerOfB = new Map(pairs.map((entry) => [entry.b, entry.a]))\n\n  let prefix = 0\n  while (prefix < a.length && prefix < b.length) {\n    const entry = pairByA.get(prefix)\n    if (!entry || entry.b !== prefix || entry.differences.length > 0) break\n    prefix += 1\n  }\n\n  return {\n    pairs,\n    onlyInA: a.map((_, i) => i).filter((i) => !partnerOfA.has(i)),\n    onlyInB: b.map((_, j) => j).filter((j) => !pairedB.has(j)),\n    commonPrefixLen: prefix,\n    firstDivergence:\n      prefix === a.length && prefix === b.length\n        ? null\n        : divergenceAt(prefix, a, b, pairByA.get(prefix), partnerOfB),\n  }\n}\n\nfunction divergenceAt(\n  index: number,\n  a: readonly DiffStep[],\n  b: readonly DiffStep[],\n  pairOfA: StepPair | undefined,\n  partnerOfB: ReadonlyMap<number, number>,\n): StepDivergence {\n  const stepA = a[index]\n  const stepB = b[index]\n  const base = {\n    index,\n    a: stepA ? index : null,\n    b: stepB ? index : null,\n    differences: [] as FieldDifference[],\n  }\n  if (stepA && stepB && pairOfA?.b === index) {\n    return {\n      ...base,\n      kind: 'changed',\n      differences: pairOfA.differences,\n      reason: `at index ${index}, ${describe(stepA)} changed: ${pairOfA.differences\n        .map((d) => `${d.field} ${String(d.a)} vs ${String(d.b)}`)\n        .join('; ')}`,\n    }\n  }\n  const aUnpaired = stepA !== undefined && pairOfA === undefined\n  const bUnpaired = stepB !== undefined && !partnerOfB.has(index)\n  if (stepA && stepB && aUnpaired && bUnpaired) {\n    return {\n      ...base,\n      kind: 'replaced',\n      reason: `at index ${index}, A ran ${describe(stepA)} and B ran ${describe(stepB)}`,\n    }\n  }\n  if (stepA && aUnpaired) {\n    return {\n      ...base,\n      kind: 'only-in-a',\n      reason: `at index ${index}, only A has ${describe(stepA)}`,\n    }\n  }\n  if (stepB && bUnpaired) {\n    return {\n      ...base,\n      kind: 'only-in-b',\n      reason: `at index ${index}, only B has ${describe(stepB)}`,\n    }\n  }\n  return {\n    ...base,\n    kind: 'reordered',\n    reason: `at index ${index}, A ran ${describe(stepA!)} and B ran ${describe(stepB!)}; both steps occur in the other run at a different position`,\n  }\n}\n\nfunction sameShape(x: DiffStep, y: DiffStep): boolean {\n  return x.name === y.name && x.kind === y.kind\n}\n\nfunction differencesBetween(x: DiffStep, y: DiffStep): FieldDifference[] {\n  const differences: FieldDifference[] = []\n  if (x.kind !== y.kind) differences.push({ field: 'kind', a: x.kind, b: y.kind })\n  if (x.name !== y.name) differences.push({ field: 'name', a: x.name, b: y.name })\n  const fields = [...new Set([...Object.keys(x.fields), ...Object.keys(y.fields)])].sort()\n  for (const field of fields) {\n    const left = x.fields[field] ?? null\n    const right = y.fields[field] ?? null\n    if (left !== right) differences.push({ field, a: left, b: right })\n  }\n  return differences\n}\n\nfunction describe(step: DiffStep): string {\n  return `${step.kind} \"${step.name}\"`\n}\n\n// ── agent-eval trace store ────────────────────────────────────────────\n\nexport interface DivergenceReport {\n  runA: string\n  runB: string\n  /** Null when the runs agree step for step. */\n  firstDivergenceIndex: number | null\n  aStep?: TrajectoryStep\n  bStep?: TrajectoryStep\n  reason?: string\n  commonPrefixLen: number\n  /** The step diff behind the first divergence, over trajectory step indexes. */\n  diff: StepDiff\n}\n\nexport async function firstDivergenceView(\n  store: TraceStore,\n  runA: string,\n  runB: string,\n): Promise<DivergenceReport> {\n  const [a, b] = await Promise.all([buildTrajectory(store, runA), buildTrajectory(store, runB)])\n  const diff = diffSteps(a.steps.map(trajectoryDiffStep), b.steps.map(trajectoryDiffStep))\n  const divergence = diff.firstDivergence\n  return {\n    runA,\n    runB,\n    firstDivergenceIndex: divergence?.index ?? null,\n    ...(divergence?.a != null ? { aStep: a.steps[divergence.a] } : {}),\n    ...(divergence?.b != null ? { bStep: b.steps[divergence.b] } : {}),\n    ...(divergence ? { reason: divergence.reason } : {}),\n    commonPrefixLen: diff.commonPrefixLen,\n    diff,\n  }\n}\n\nfunction trajectoryDiffStep(step: TrajectoryStep): DiffStep {\n  const span: Span = step.span\n  const fields: Record<string, string | number | boolean | null> = {\n    status: span.status ?? null,\n  }\n  if (span.kind === 'tool') fields.tool = span.toolName\n  if (span.kind === 'llm') fields.model = span.model\n  if (span.kind === 'judge') fields.dimension = span.dimension\n  return { id: span.spanId, name: span.name, kind: span.kind, fields }\n}\n\n// ── flat OTLP spans ───────────────────────────────────────────────────\n\n/**\n * One run's spans as diff steps, in containment order: depth first, siblings\n * by start time, the same order `buildTrajectory` gives a trace store run.\n * Spans whose parent is absent, or whose parents form a cycle, start their own\n * subtree, so every span appears exactly once.\n */\nexport function diffStepsFromSpans(spans: readonly DiagnosisSpan[]): DiffStep[] {\n  const ids = new Set(spans.map((span) => span.spanId))\n  const children = new Map<string | null, DiagnosisSpan[]>()\n  for (const span of spans) {\n    const parent =\n      span.parentSpanId !== null && ids.has(span.parentSpanId) ? span.parentSpanId : null\n    const list = children.get(parent)\n    if (list) list.push(span)\n    else children.set(parent, [span])\n  }\n  const byStart = (x: DiagnosisSpan, y: DiagnosisSpan) =>\n    (x.startMs ?? 0) - (y.startMs ?? 0) || x.spanId.localeCompare(y.spanId)\n  for (const list of children.values()) list.sort(byStart)\n  const ordered: DiffStep[] = []\n  const visited = new Set<string>()\n  const visit = (span: DiagnosisSpan) => {\n    if (visited.has(span.spanId)) return\n    visited.add(span.spanId)\n    ordered.push({\n      id: span.spanId,\n      name: span.name,\n      kind: span.kind,\n      fields: { status: span.status, tool: span.toolName, model: span.model },\n    })\n    for (const child of children.get(span.spanId) ?? []) visit(child)\n  }\n  for (const root of children.get(null) ?? []) visit(root)\n  for (const span of [...spans].sort(byStart)) visit(span)\n  return ordered\n}\n","/**\n * RegressionView — compares a candidate slice to a baseline slice on a\n * named metric. Delegates the statistics (Welch's t-test, Cohen's d,\n * IQR stability) to `baseline.ts`.\n *\n * This is the entry point for CI regression gates: \"given runs tagged\n * release=A and release=B, did any metric regress?\"\n */\n\nimport { type BaselineOptions, type BaselineReport, compareToBaseline } from '../baseline'\nimport { runMetricExtractor } from '../trace/query'\nimport type { Run } from '../trace/schema'\nimport type { RunFilter, TraceStore } from '../trace/store'\n\nexport interface RegressionSpec {\n  metric: string\n  higherIsBetter: boolean\n  /** Extract a scalar from a run. Omit it and `metric` must name one of\n   *  `RUN_METRICS`; any other name is refused. */\n  extract?: (run: Run, store: TraceStore) => Promise<number | null>\n}\n\nexport interface RegressionOptions extends BaselineOptions {\n  baseline: RunFilter\n  candidate: RunFilter\n}\n\nexport async function regressionView(\n  store: TraceStore,\n  metrics: RegressionSpec[],\n  options: RegressionOptions,\n): Promise<BaselineReport> {\n  const baselineRuns = await store.listRuns(options.baseline)\n  const candidateRuns = await store.listRuns(options.candidate)\n  const samples = await Promise.all(\n    metrics.map(async (m) => {\n      const extract = m.extract ?? runMetricExtractor(m.metric)\n      const baseline = await extractAll(baselineRuns, extract, store)\n      const candidate = await extractAll(candidateRuns, extract, store)\n      return { metric: m.metric, higherIsBetter: m.higherIsBetter, baseline, candidate }\n    }),\n  )\n  return compareToBaseline(samples, options)\n}\n\nasync function extractAll(\n  runs: Run[],\n  extract: (r: Run, s: TraceStore) => Promise<number | null>,\n  store: TraceStore,\n): Promise<number[]> {\n  const out: number[] = []\n  for (const r of runs) {\n    const v = await extract(r, store)\n    if (v !== null && Number.isFinite(v)) out.push(v)\n  }\n  return out\n}\n","/**\n * StuckLoopView — detects when an agent calls the same tool with the\n * same (or structurally similar) arguments ≥ N times in a short window.\n *\n * Rationale: agents that loop are the number-one production failure\n * mode on long-horizon flows. The view returns (runId, toolName,\n * argHash, occurrences, windowMs) for each detected loop plus a\n * fraction of runs affected.\n */\n\nimport { executionTrackByLane } from '../trace/execution-tracks'\nimport { argHash, hasCapturedToolArgs } from '../trace/query'\nimport { isToolSpan, type Span, type ToolSpan } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nconst DEFAULT_MAX_WINDOW_MS = 60_000\nconst DEFAULT_MAX_INTERVENING_TOOL_CALLS = 0\n\ninterface ScopedToolCall {\n  span: ToolSpan\n  scopeSpanId: string | null\n  laneSpanId: string | null\n}\n\ninterface IndexedToolCall extends ScopedToolCall {\n  toolCallIndex: number\n  trackId: string\n}\n\nexport interface StuckLoopFinding {\n  runId: string\n  toolName: string\n  argHash: string\n  /** Calls in this episode's densest qualifying interval, not the whole-run total. */\n  occurrences: number\n  spanIds: string[]\n  /** Nearest agent ancestor, or the direct parent when ancestry is incomplete. */\n  scopeSpanId?: string\n  /** Milliseconds between first and last call in the loop. */\n  windowMs: number\n}\n\nexport interface StuckLoopReport {\n  findings: StuckLoopFinding[]\n  affectedRunRatio: number\n  totalRuns: number\n}\n\nexport interface StuckLoopOptions {\n  /** Minimum call count to flag a loop (default 3). */\n  minOccurrences?: number\n  /** Maximum time between the first and last repeated call (default 60 seconds). */\n  maxWindowMs?: number\n  /**\n   * Maximum other tool calls allowed between adjacent repeats (default 0).\n   * Set to 1 to detect alternating patterns such as A,B,A,B,A.\n   */\n  maxInterveningToolCalls?: number\n  /** Filter to a specific runId; omit to scan the entire corpus. */\n  runId?: string\n}\n\nexport async function stuckLoopView(\n  store: TraceStore,\n  options: StuckLoopOptions = {},\n): Promise<StuckLoopReport> {\n  const minOccurrences = options.minOccurrences ?? 3\n  const maxWindowMs = options.maxWindowMs ?? DEFAULT_MAX_WINDOW_MS\n  const maxInterveningToolCalls =\n    options.maxInterveningToolCalls ?? DEFAULT_MAX_INTERVENING_TOOL_CALLS\n  if (!Number.isInteger(minOccurrences) || minOccurrences < 1) {\n    throw new RangeError('minOccurrences must be a positive integer')\n  }\n  if (!Number.isFinite(maxWindowMs) || maxWindowMs < 0) {\n    throw new RangeError('maxWindowMs must be a finite non-negative number')\n  }\n  if (!Number.isInteger(maxInterveningToolCalls) || maxInterveningToolCalls < 0) {\n    throw new RangeError('maxInterveningToolCalls must be a non-negative integer')\n  }\n  const runs = options.runId\n    ? [{ runId: options.runId }]\n    : (await store.listRuns()).map((r) => ({ runId: r.runId }))\n\n  const findings: StuckLoopFinding[] = []\n  for (const { runId } of runs) {\n    const spans = await store.spans({ runId })\n    const spansById = new Map(spans.map((span) => [span.spanId, span]))\n    const scopedTools: ScopedToolCall[] = spans\n      .filter(isToolSpan)\n      .map((span, sourceIndex) => ({ span, sourceIndex }))\n      .sort((a, b) => a.span.startedAt - b.span.startedAt || a.sourceIndex - b.sourceIndex)\n      .map(({ span }) => ({ span, ...executionScope(span, spansById) }))\n    const trackByLane = executionTrackByLane(\n      scopedTools.map((call) => {\n        const direct = call.laneSpanId === null || call.laneSpanId === call.scopeSpanId\n        const timed = direct\n          ? call.span\n          : call.laneSpanId\n            ? spansById.get(call.laneSpanId)\n            : undefined\n        return {\n          key: executionKey(call),\n          scopeKey: JSON.stringify(call.scopeSpanId),\n          start: timed?.startedAt ?? null,\n          end: timed?.endedAt ?? null,\n        }\n      }),\n    )\n    const nextToolIndexByTrack = new Map<string, number>()\n    const orderedTools: IndexedToolCall[] = scopedTools.map((call) => {\n      const trackId = trackByLane.get(executionKey(call))!\n      const toolCallIndex = nextToolIndexByTrack.get(trackId) ?? 0\n      nextToolIndexByTrack.set(trackId, toolCallIndex + 1)\n      return { ...call, toolCallIndex, trackId }\n    })\n    const byKey = new Map<\n      string,\n      {\n        calls: IndexedToolCall[]\n        argHash: string\n        toolName: string\n        scopeSpanId: string | null\n      }\n    >()\n    for (const call of orderedTools) {\n      if (!hasCapturedToolArgs(call.span)) continue\n      const h = argHash(call.span.args)\n      const key = JSON.stringify([call.trackId, call.span.toolName, h])\n      const bucket = byKey.get(key) ?? {\n        calls: [],\n        argHash: h,\n        toolName: call.span.toolName,\n        scopeSpanId: call.scopeSpanId,\n      }\n      bucket.calls.push(call)\n      byKey.set(key, bucket)\n    }\n    for (const { calls, argHash: h, toolName, scopeSpanId } of byKey.values()) {\n      if (calls.length < minOccurrences) continue\n      let episodeStart = 0\n      for (let episodeEnd = 1; episodeEnd <= calls.length; episodeEnd += 1) {\n        const previous = calls[episodeEnd - 1]!\n        const next = calls[episodeEnd]\n        const episodeEnded =\n          next === undefined ||\n          next.span.startedAt - previous.span.startedAt > maxWindowMs ||\n          next.toolCallIndex - previous.toolCallIndex - 1 > maxInterveningToolCalls ||\n          !callsAreSerial(previous, next)\n        if (!episodeEnded) continue\n\n        const episode = calls.slice(episodeStart, episodeEnd)\n        let left = 0\n        let bestStart = 0\n        let bestEnd = -1\n        for (let right = 0; right < episode.length; right += 1) {\n          while (episode[right]!.span.startedAt - episode[left]!.span.startedAt > maxWindowMs) {\n            left += 1\n          }\n          if (right - left > bestEnd - bestStart) {\n            bestStart = left\n            bestEnd = right\n          }\n        }\n        if (bestEnd - bestStart + 1 >= minOccurrences) {\n          const loop = episode.slice(bestStart, bestEnd + 1)\n          const first = loop[0]!.span.startedAt\n          const last = loop[loop.length - 1]!.span.startedAt\n          findings.push({\n            runId,\n            toolName,\n            argHash: h,\n            occurrences: loop.length,\n            spanIds: loop.map((call) => call.span.spanId),\n            ...(scopeSpanId ? { scopeSpanId } : {}),\n            windowMs: last - first,\n          })\n        }\n        episodeStart = episodeEnd\n      }\n    }\n  }\n\n  const affectedRuns = new Set(findings.map((f) => f.runId))\n  return {\n    findings,\n    affectedRunRatio: runs.length > 0 ? affectedRuns.size / runs.length : 0,\n    totalRuns: runs.length,\n  }\n}\n\nfunction laneKey(call: Pick<ScopedToolCall, 'scopeSpanId' | 'laneSpanId'>): string {\n  return JSON.stringify([call.scopeSpanId, call.laneSpanId])\n}\n\nfunction executionKey(call: ScopedToolCall): string {\n  return call.laneSpanId === null || call.laneSpanId === call.scopeSpanId\n    ? JSON.stringify([call.scopeSpanId, call.span.spanId])\n    : laneKey(call)\n}\n\nfunction executionScope(\n  span: ToolSpan,\n  spansById: ReadonlyMap<string, Span>,\n): { scopeSpanId: string | null; laneSpanId: string | null } {\n  const directParent = span.parentSpanId\n  if (!directParent) return { scopeSpanId: null, laneSpanId: null }\n\n  let currentId: string | undefined = directParent\n  let laneSpanId: string | null = null\n  const seen = new Set<string>()\n  while (currentId && !seen.has(currentId)) {\n    seen.add(currentId)\n    const current = spansById.get(currentId)\n    if (!current) return { scopeSpanId: directParent, laneSpanId: directParent }\n    if (current.kind === 'agent') {\n      return { scopeSpanId: current.spanId, laneSpanId: laneSpanId ?? current.spanId }\n    }\n    laneSpanId = current.spanId\n    currentId = current.parentSpanId\n  }\n  return { scopeSpanId: directParent, laneSpanId: directParent }\n}\n\nfunction callsAreSerial(previous: IndexedToolCall, next: IndexedToolCall): boolean {\n  return previous.span.endedAt !== undefined && previous.span.endedAt <= next.span.startedAt\n}\n"],"mappings":";;;;;;;AA+EA,SAAgB,UAAU,GAAwB,GAAkC;CAClF,MAAM,6BAAa,IAAI,IAAkD;CACzE,MAAM,0BAAU,IAAI,IAAY;CAChC,MAAM,QAAQ,GAAW,GAAW,aAA0B;EAC5D,WAAW,IAAI,GAAG;GAAE,GAAG;GAAG;EAAS,CAAC;EACpC,QAAQ,IAAI,CAAC;CACf;CAEA,MAAM,wBAAQ,IAAI,IAAsB;CACxC,EAAE,SAAS,MAAM,MAAM;EACrB,MAAM,OAAO,MAAM,IAAI,KAAK,EAAE;EAC9B,IAAI,MAAM,KAAK,KAAK,CAAC;OAChB,MAAM,IAAI,KAAK,IAAI,CAAC,CAAC,CAAC;CAC7B,CAAC;CACD,EAAE,SAAS,MAAM,MAAM;EACrB,MAAM,IAAI,MAAM,IAAI,KAAK,EAAE,CAAC,EAAE,MAAM,cAAc,CAAC,QAAQ,IAAI,SAAS,CAAC;EACzE,IAAI,MAAM,KAAA,GAAW,KAAK,GAAG,GAAG,IAAI;CACtC,CAAC;CACD,EAAE,SAAS,MAAM,MAAM;EACrB,IAAI,WAAW,IAAI,CAAC,KAAK,KAAK,EAAE,UAAU,QAAQ,IAAI,CAAC,GAAG;EAC1D,IAAI,UAAU,MAAM,EAAE,EAAG,GAAG,KAAK,GAAG,GAAG,UAAU;CACnD,CAAC;CACD,EAAE,SAAS,MAAM,MAAM;EACrB,IAAI,WAAW,IAAI,CAAC,GAAG;EACvB,MAAM,IAAI,EAAE,WAAW,WAAW,UAAU,CAAC,QAAQ,IAAI,KAAK,KAAK,UAAU,MAAM,SAAS,CAAC;EAC7F,IAAI,KAAK,GAAG,KAAK,GAAG,GAAG,MAAM;CAC/B,CAAC;CAED,MAAM,QAAoB,CAAC,GAAG,WAAW,QAAQ,CAAC,CAAC,CAChD,MAAM,CAAC,IAAI,CAAC,OAAO,IAAI,CAAC,CAAC,CACzB,KAAK,CAAC,GAAG,cAAc;EACtB,GAAG;EACH,GAAG,QAAQ;EACX,UAAU,QAAQ;EAClB,aAAa,mBAAmB,EAAE,IAAK,EAAE,QAAQ,EAAG;CACtD,EAAE;CACJ,MAAM,UAAU,IAAI,IAAI,MAAM,KAAK,UAAU,CAAC,MAAM,GAAG,KAAK,CAAC,CAAC;CAC9D,MAAM,aAAa,IAAI,IAAI,MAAM,KAAK,UAAU,CAAC,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC;CAEnE,IAAI,SAAS;CACb,OAAO,SAAS,EAAE,UAAU,SAAS,EAAE,QAAQ;EAC7C,MAAM,QAAQ,QAAQ,IAAI,MAAM;EAChC,IAAI,CAAC,SAAS,MAAM,MAAM,UAAU,MAAM,YAAY,SAAS,GAAG;EAClE,UAAU;CACZ;CAEA,OAAO;EACL;EACA,SAAS,EAAE,KAAK,GAAG,MAAM,CAAC,CAAC,CAAC,QAAQ,MAAM,CAAC,WAAW,IAAI,CAAC,CAAC;EAC5D,SAAS,EAAE,KAAK,GAAG,MAAM,CAAC,CAAC,CAAC,QAAQ,MAAM,CAAC,QAAQ,IAAI,CAAC,CAAC;EACzD,iBAAiB;EACjB,iBACE,WAAW,EAAE,UAAU,WAAW,EAAE,SAChC,OACA,aAAa,QAAQ,GAAG,GAAG,QAAQ,IAAI,MAAM,GAAG,UAAU;CAClE;AACF;AAEA,SAAS,aACP,OACA,GACA,GACA,SACA,YACgB;CAChB,MAAM,QAAQ,EAAE;CAChB,MAAM,QAAQ,EAAE;CAChB,MAAM,OAAO;EACX;EACA,GAAG,QAAQ,QAAQ;EACnB,GAAG,QAAQ,QAAQ;EACnB,aAAa,CAAC;CAChB;CACA,IAAI,SAAS,SAAS,SAAS,MAAM,OACnC,OAAO;EACL,GAAG;EACH,MAAM;EACN,aAAa,QAAQ;EACrB,QAAQ,YAAY,MAAM,IAAI,SAAS,KAAK,EAAE,YAAY,QAAQ,YAC/D,KAAK,MAAM,GAAG,EAAE,MAAM,GAAG,OAAO,EAAE,CAAC,EAAE,MAAM,OAAO,EAAE,CAAC,GAAG,CAAC,CACzD,KAAK,IAAI;CACd;CAEF,MAAM,YAAY,UAAU,KAAA,KAAa,YAAY,KAAA;CACrD,MAAM,YAAY,UAAU,KAAA,KAAa,CAAC,WAAW,IAAI,KAAK;CAC9D,IAAI,SAAS,SAAS,aAAa,WACjC,OAAO;EACL,GAAG;EACH,MAAM;EACN,QAAQ,YAAY,MAAM,UAAU,SAAS,KAAK,EAAE,aAAa,SAAS,KAAK;CACjF;CAEF,IAAI,SAAS,WACX,OAAO;EACL,GAAG;EACH,MAAM;EACN,QAAQ,YAAY,MAAM,eAAe,SAAS,KAAK;CACzD;CAEF,IAAI,SAAS,WACX,OAAO;EACL,GAAG;EACH,MAAM;EACN,QAAQ,YAAY,MAAM,eAAe,SAAS,KAAK;CACzD;CAEF,OAAO;EACL,GAAG;EACH,MAAM;EACN,QAAQ,YAAY,MAAM,UAAU,SAAS,KAAM,EAAE,aAAa,SAAS,KAAM,EAAE;CACrF;AACF;AAEA,SAAS,UAAU,GAAa,GAAsB;CACpD,OAAO,EAAE,SAAS,EAAE,QAAQ,EAAE,SAAS,EAAE;AAC3C;AAEA,SAAS,mBAAmB,GAAa,GAAgC;CACvE,MAAM,cAAiC,CAAC;CACxC,IAAI,EAAE,SAAS,EAAE,MAAM,YAAY,KAAK;EAAE,OAAO;EAAQ,GAAG,EAAE;EAAM,GAAG,EAAE;CAAK,CAAC;CAC/E,IAAI,EAAE,SAAS,EAAE,MAAM,YAAY,KAAK;EAAE,OAAO;EAAQ,GAAG,EAAE;EAAM,GAAG,EAAE;CAAK,CAAC;CAC/E,MAAM,SAAS,CAAC,mBAAG,IAAI,IAAI,CAAC,GAAG,OAAO,KAAK,EAAE,MAAM,GAAG,GAAG,OAAO,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK;CACvF,KAAK,MAAM,SAAS,QAAQ;EAC1B,MAAM,OAAO,EAAE,OAAO,UAAU;EAChC,MAAM,QAAQ,EAAE,OAAO,UAAU;EACjC,IAAI,SAAS,OAAO,YAAY,KAAK;GAAE;GAAO,GAAG;GAAM,GAAG;EAAM,CAAC;CACnE;CACA,OAAO;AACT;AAEA,SAAS,SAAS,MAAwB;CACxC,OAAO,GAAG,KAAK,KAAK,IAAI,KAAK,KAAK;AACpC;AAiBA,eAAsB,oBACpB,OACA,MACA,MAC2B;CAC3B,MAAM,CAAC,GAAG,KAAK,MAAM,QAAQ,IAAI,CAAC,gBAAgB,OAAO,IAAI,GAAG,gBAAgB,OAAO,IAAI,CAAC,CAAC;CAC7F,MAAM,OAAO,UAAU,EAAE,MAAM,IAAI,kBAAkB,GAAG,EAAE,MAAM,IAAI,kBAAkB,CAAC;CACvF,MAAM,aAAa,KAAK;CACxB,OAAO;EACL;EACA;EACA,sBAAsB,YAAY,SAAS;EAC3C,GAAI,YAAY,KAAK,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,GAAG,IAAI,CAAC;EAChE,GAAI,YAAY,KAAK,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,GAAG,IAAI,CAAC;EAChE,GAAI,aAAa,EAAE,QAAQ,WAAW,OAAO,IAAI,CAAC;EAClD,iBAAiB,KAAK;EACtB;CACF;AACF;AAEA,SAAS,mBAAmB,MAAgC;CAC1D,MAAM,OAAa,KAAK;CACxB,MAAM,SAA2D,EAC/D,QAAQ,KAAK,UAAU,KACzB;CACA,IAAI,KAAK,SAAS,QAAQ,OAAO,OAAO,KAAK;CAC7C,IAAI,KAAK,SAAS,OAAO,OAAO,QAAQ,KAAK;CAC7C,IAAI,KAAK,SAAS,SAAS,OAAO,YAAY,KAAK;CACnD,OAAO;EAAE,IAAI,KAAK;EAAQ,MAAM,KAAK;EAAM,MAAM,KAAK;EAAM;CAAO;AACrE;;;;;;;AAUA,SAAgB,mBAAmB,OAA6C;CAC9E,MAAM,MAAM,IAAI,IAAI,MAAM,KAAK,SAAS,KAAK,MAAM,CAAC;CACpD,MAAM,2BAAW,IAAI,IAAoC;CACzD,KAAK,MAAM,QAAQ,OAAO;EACxB,MAAM,SACJ,KAAK,iBAAiB,QAAQ,IAAI,IAAI,KAAK,YAAY,IAAI,KAAK,eAAe;EACjF,MAAM,OAAO,SAAS,IAAI,MAAM;EAChC,IAAI,MAAM,KAAK,KAAK,IAAI;OACnB,SAAS,IAAI,QAAQ,CAAC,IAAI,CAAC;CAClC;CACA,MAAM,WAAW,GAAkB,OAChC,EAAE,WAAW,MAAM,EAAE,WAAW,MAAM,EAAE,OAAO,cAAc,EAAE,MAAM;CACxE,KAAK,MAAM,QAAQ,SAAS,OAAO,GAAG,KAAK,KAAK,OAAO;CACvD,MAAM,UAAsB,CAAC;CAC7B,MAAM,0BAAU,IAAI,IAAY;CAChC,MAAM,SAAS,SAAwB;EACrC,IAAI,QAAQ,IAAI,KAAK,MAAM,GAAG;EAC9B,QAAQ,IAAI,KAAK,MAAM;EACvB,QAAQ,KAAK;GACX,IAAI,KAAK;GACT,MAAM,KAAK;GACX,MAAM,KAAK;GACX,QAAQ;IAAE,QAAQ,KAAK;IAAQ,MAAM,KAAK;IAAU,OAAO,KAAK;GAAM;EACxE,CAAC;EACD,KAAK,MAAM,SAAS,SAAS,IAAI,KAAK,MAAM,KAAK,CAAC,GAAG,MAAM,KAAK;CAClE;CACA,KAAK,MAAM,QAAQ,SAAS,IAAI,IAAI,KAAK,CAAC,GAAG,MAAM,IAAI;CACvD,KAAK,MAAM,QAAQ,CAAC,GAAG,KAAK,CAAC,CAAC,KAAK,OAAO,GAAG,MAAM,IAAI;CACvD,OAAO;AACT;;;;;;;;;;;AC7QA,eAAsB,eACpB,OACA,SACA,SACyB;CACzB,MAAM,eAAe,MAAM,MAAM,SAAS,QAAQ,QAAQ;CAC1D,MAAM,gBAAgB,MAAM,MAAM,SAAS,QAAQ,SAAS;CAS5D,OAAO,kBAAkB,MARH,QAAQ,IAC5B,QAAQ,IAAI,OAAO,MAAM;EACvB,MAAM,UAAU,EAAE,WAAW,mBAAmB,EAAE,MAAM;EACxD,MAAM,WAAW,MAAM,WAAW,cAAc,SAAS,KAAK;EAC9D,MAAM,YAAY,MAAM,WAAW,eAAe,SAAS,KAAK;EAChE,OAAO;GAAE,QAAQ,EAAE;GAAQ,gBAAgB,EAAE;GAAgB;GAAU;EAAU;CACnF,CAAC,CACH,GACkC,OAAO;AAC3C;AAEA,eAAe,WACb,MACA,SACA,OACmB;CACnB,MAAM,MAAgB,CAAC;CACvB,KAAK,MAAM,KAAK,MAAM;EACpB,MAAM,IAAI,MAAM,QAAQ,GAAG,KAAK;EAChC,IAAI,MAAM,QAAQ,OAAO,SAAS,CAAC,GAAG,IAAI,KAAK,CAAC;CAClD;CACA,OAAO;AACT;;;;;;;;;;;;ACzCA,MAAM,wBAAwB;AAC9B,MAAM,qCAAqC;AA8C3C,eAAsB,cACpB,OACA,UAA4B,CAAC,GACH;CAC1B,MAAM,iBAAiB,QAAQ,kBAAkB;CACjD,MAAM,cAAc,QAAQ,eAAe;CAC3C,MAAM,0BACJ,QAAQ,2BAA2B;CACrC,IAAI,CAAC,OAAO,UAAU,cAAc,KAAK,iBAAiB,GACxD,MAAM,IAAI,WAAW,2CAA2C;CAElE,IAAI,CAAC,OAAO,SAAS,WAAW,KAAK,cAAc,GACjD,MAAM,IAAI,WAAW,kDAAkD;CAEzE,IAAI,CAAC,OAAO,UAAU,uBAAuB,KAAK,0BAA0B,GAC1E,MAAM,IAAI,WAAW,wDAAwD;CAE/E,MAAM,OAAO,QAAQ,QACjB,CAAC,EAAE,OAAO,QAAQ,MAAM,CAAC,KACxB,MAAM,MAAM,SAAS,EAAA,CAAG,KAAK,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE;CAE5D,MAAM,WAA+B,CAAC;CACtC,KAAK,MAAM,EAAE,WAAW,MAAM;EAC5B,MAAM,QAAQ,MAAM,MAAM,MAAM,EAAE,MAAM,CAAC;EACzC,MAAM,YAAY,IAAI,IAAI,MAAM,KAAK,SAAS,CAAC,KAAK,QAAQ,IAAI,CAAC,CAAC;EAClE,MAAM,cAAgC,MACnC,OAAO,UAAU,CAAC,CAClB,KAAK,MAAM,iBAAiB;GAAE;GAAM;EAAY,EAAE,CAAC,CACnD,MAAM,GAAG,MAAM,EAAE,KAAK,YAAY,EAAE,KAAK,aAAa,EAAE,cAAc,EAAE,WAAW,CAAC,CACpF,KAAK,EAAE,YAAY;GAAE;GAAM,GAAG,eAAe,MAAM,SAAS;EAAE,EAAE;EACnE,MAAM,cAAc,qBAClB,YAAY,KAAK,SAAS;GAExB,MAAM,QADS,KAAK,eAAe,QAAQ,KAAK,eAAe,KAAK,cAEhE,KAAK,OACL,KAAK,aACH,UAAU,IAAI,KAAK,UAAU,IAC7B,KAAA;GACN,OAAO;IACL,KAAK,aAAa,IAAI;IACtB,UAAU,KAAK,UAAU,KAAK,WAAW;IACzC,OAAO,OAAO,aAAa;IAC3B,KAAK,OAAO,WAAW;GACzB;EACF,CAAC,CACH;EACA,MAAM,uCAAuB,IAAI,IAAoB;EACrD,MAAM,eAAkC,YAAY,KAAK,SAAS;GAChE,MAAM,UAAU,YAAY,IAAI,aAAa,IAAI,CAAC;GAClD,MAAM,gBAAgB,qBAAqB,IAAI,OAAO,KAAK;GAC3D,qBAAqB,IAAI,SAAS,gBAAgB,CAAC;GACnD,OAAO;IAAE,GAAG;IAAM;IAAe;GAAQ;EAC3C,CAAC;EACD,MAAM,wBAAQ,IAAI,IAQhB;EACF,KAAK,MAAM,QAAQ,cAAc;GAC/B,IAAI,CAAC,oBAAoB,KAAK,IAAI,GAAG;GACrC,MAAM,IAAI,QAAQ,KAAK,KAAK,IAAI;GAChC,MAAM,MAAM,KAAK,UAAU;IAAC,KAAK;IAAS,KAAK,KAAK;IAAU;GAAC,CAAC;GAChE,MAAM,SAAS,MAAM,IAAI,GAAG,KAAK;IAC/B,OAAO,CAAC;IACR,SAAS;IACT,UAAU,KAAK,KAAK;IACpB,aAAa,KAAK;GACpB;GACA,OAAO,MAAM,KAAK,IAAI;GACtB,MAAM,IAAI,KAAK,MAAM;EACvB;EACA,KAAK,MAAM,EAAE,OAAO,SAAS,GAAG,UAAU,iBAAiB,MAAM,OAAO,GAAG;GACzE,IAAI,MAAM,SAAS,gBAAgB;GACnC,IAAI,eAAe;GACnB,KAAK,IAAI,aAAa,GAAG,cAAc,MAAM,QAAQ,cAAc,GAAG;IACpE,MAAM,WAAW,MAAM,aAAa;IACpC,MAAM,OAAO,MAAM;IAMnB,IAAI,EAJF,SAAS,KAAA,KACT,KAAK,KAAK,YAAY,SAAS,KAAK,YAAY,eAChD,KAAK,gBAAgB,SAAS,gBAAgB,IAAI,2BAClD,CAAC,eAAe,UAAU,IAAI,IACb;IAEnB,MAAM,UAAU,MAAM,MAAM,cAAc,UAAU;IACpD,IAAI,OAAO;IACX,IAAI,YAAY;IAChB,IAAI,UAAU;IACd,KAAK,IAAI,QAAQ,GAAG,QAAQ,QAAQ,QAAQ,SAAS,GAAG;KACtD,OAAO,QAAQ,MAAM,CAAE,KAAK,YAAY,QAAQ,KAAK,CAAE,KAAK,YAAY,aACtE,QAAQ;KAEV,IAAI,QAAQ,OAAO,UAAU,WAAW;MACtC,YAAY;MACZ,UAAU;KACZ;IACF;IACA,IAAI,UAAU,YAAY,KAAK,gBAAgB;KAC7C,MAAM,OAAO,QAAQ,MAAM,WAAW,UAAU,CAAC;KACjD,MAAM,QAAQ,KAAK,EAAE,CAAE,KAAK;KAC5B,MAAM,OAAO,KAAK,KAAK,SAAS,EAAE,CAAE,KAAK;KACzC,SAAS,KAAK;MACZ;MACA;MACA,SAAS;MACT,aAAa,KAAK;MAClB,SAAS,KAAK,KAAK,SAAS,KAAK,KAAK,MAAM;MAC5C,GAAI,cAAc,EAAE,YAAY,IAAI,CAAC;MACrC,UAAU,OAAO;KACnB,CAAC;IACH;IACA,eAAe;GACjB;EACF;CACF;CAEA,MAAM,eAAe,IAAI,IAAI,SAAS,KAAK,MAAM,EAAE,KAAK,CAAC;CACzD,OAAO;EACL;EACA,kBAAkB,KAAK,SAAS,IAAI,aAAa,OAAO,KAAK,SAAS;EACtE,WAAW,KAAK;CAClB;AACF;AAEA,SAAS,QAAQ,MAAkE;CACjF,OAAO,KAAK,UAAU,CAAC,KAAK,aAAa,KAAK,UAAU,CAAC;AAC3D;AAEA,SAAS,aAAa,MAA8B;CAClD,OAAO,KAAK,eAAe,QAAQ,KAAK,eAAe,KAAK,cACxD,KAAK,UAAU,CAAC,KAAK,aAAa,KAAK,KAAK,MAAM,CAAC,IACnD,QAAQ,IAAI;AAClB;AAEA,SAAS,eACP,MACA,WAC2D;CAC3D,MAAM,eAAe,KAAK;CAC1B,IAAI,CAAC,cAAc,OAAO;EAAE,aAAa;EAAM,YAAY;CAAK;CAEhE,IAAI,YAAgC;CACpC,IAAI,aAA4B;CAChC,MAAM,uBAAO,IAAI,IAAY;CAC7B,OAAO,aAAa,CAAC,KAAK,IAAI,SAAS,GAAG;EACxC,KAAK,IAAI,SAAS;EAClB,MAAM,UAAU,UAAU,IAAI,SAAS;EACvC,IAAI,CAAC,SAAS,OAAO;GAAE,aAAa;GAAc,YAAY;EAAa;EAC3E,IAAI,QAAQ,SAAS,SACnB,OAAO;GAAE,aAAa,QAAQ;GAAQ,YAAY,cAAc,QAAQ;EAAO;EAEjF,aAAa,QAAQ;EACrB,YAAY,QAAQ;CACtB;CACA,OAAO;EAAE,aAAa;EAAc,YAAY;CAAa;AAC/D;AAEA,SAAS,eAAe,UAA2B,MAAgC;CACjF,OAAO,SAAS,KAAK,YAAY,KAAA,KAAa,SAAS,KAAK,WAAW,KAAK,KAAK;AACnF"}
+{"version":3,"file":"index.js","names":[],"sources":["../../src/pipelines/first-divergence.ts","../../src/pipelines/regression.ts","../../src/pipelines/stuck-loop.ts"],"sourcesContent":["/**\n * Step diff between two runs, with the first divergence marked.\n *\n * Steps pair in three passes, each over the steps still unpaired:\n *   1. `id`: the same step id in both runs, e.g. a forked run that kept its\n *      parent's span ids.\n *   2. `position`: the same index with the same name and kind.\n *   3. `name`: the same name and kind anywhere, first unpaired step in order.\n * A step left unpaired exists in only one run. The first divergence is the\n * first index at which the two runs stop agreeing step for step, and it says\n * how: a paired step changed, one run replaced, added or removed a step, or the\n * runs ran the same steps in a different order.\n *\n * `diffSteps` reads plain `DiffStep` lists, so the same diff serves agent-eval\n * trace stores (`firstDivergenceView`) and flat OTLP spans\n * (`diffStepsFromSpans`). Use it to attribute \"why is variant B better?\" to a\n * step, and to compare sibling or parent and child runs of a search tree.\n */\n\nimport type { DiagnosisSpan } from '../diagnosis/spans'\nimport type { Span } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\nimport { buildTrajectory, type TrajectoryStep } from '../trajectory'\n\n/** One step of a run, as the diff reads it. */\nexport interface DiffStep {\n  /** The step's id within its run. */\n  id: string\n  name: string\n  kind: string\n  /** Fields compared once two steps pair, such as status, tool or model. */\n  fields: Readonly<Record<string, string | number | boolean | null>>\n}\n\nexport type StepPairing = 'id' | 'position' | 'name'\n\nexport interface FieldDifference {\n  field: string\n  a: string | number | boolean | null\n  b: string | number | boolean | null\n}\n\nexport interface StepPair {\n  /** Index in run A. */\n  a: number\n  /** Index in run B. */\n  b: number\n  pairedBy: StepPairing\n  /** Empty when the paired steps agree on name, kind and every field. */\n  differences: FieldDifference[]\n}\n\nexport type StepDivergenceKind = 'changed' | 'replaced' | 'only-in-a' | 'only-in-b' | 'reordered'\n\nexport interface StepDivergence {\n  /** The first index at which the runs stop agreeing; equals `commonPrefixLen`. */\n  index: number\n  kind: StepDivergenceKind\n  /** Index of the step at `index` in each run, or null when that run has ended. */\n  a: number | null\n  b: number | null\n  /** Set when `kind` is `changed`. */\n  differences: FieldDifference[]\n  reason: string\n}\n\nexport interface StepDiff {\n  /** Paired steps in run A order. */\n  pairs: StepPair[]\n  /** Indexes of run A steps with no partner in run B. */\n  onlyInA: number[]\n  /** Indexes of run B steps with no partner in run A. */\n  onlyInB: number[]\n  /** Leading steps paired index for index with no difference. */\n  commonPrefixLen: number\n  /** Null when the runs agree step for step. */\n  firstDivergence: StepDivergence | null\n}\n\nexport function diffSteps(a: readonly DiffStep[], b: readonly DiffStep[]): StepDiff {\n  const partnerOfA = new Map<number, { b: number; pairedBy: StepPairing }>()\n  const pairedB = new Set<number>()\n  const pair = (i: number, j: number, pairedBy: StepPairing) => {\n    partnerOfA.set(i, { b: j, pairedBy })\n    pairedB.add(j)\n  }\n\n  const bById = new Map<string, number[]>()\n  b.forEach((step, j) => {\n    const list = bById.get(step.id)\n    if (list) list.push(j)\n    else bById.set(step.id, [j])\n  })\n  a.forEach((step, i) => {\n    const j = bById.get(step.id)?.find((candidate) => !pairedB.has(candidate))\n    if (j !== undefined) pair(i, j, 'id')\n  })\n  a.forEach((step, i) => {\n    if (partnerOfA.has(i) || i >= b.length || pairedB.has(i)) return\n    if (sameShape(step, b[i]!)) pair(i, i, 'position')\n  })\n  a.forEach((step, i) => {\n    if (partnerOfA.has(i)) return\n    const j = b.findIndex((candidate, index) => !pairedB.has(index) && sameShape(step, candidate))\n    if (j >= 0) pair(i, j, 'name')\n  })\n\n  const pairs: StepPair[] = [...partnerOfA.entries()]\n    .sort(([x], [y]) => x - y)\n    .map(([i, partner]) => ({\n      a: i,\n      b: partner.b,\n      pairedBy: partner.pairedBy,\n      differences: differencesBetween(a[i]!, b[partner.b]!),\n    }))\n  const pairByA = new Map(pairs.map((entry) => [entry.a, entry]))\n  const partnerOfB = new Map(pairs.map((entry) => [entry.b, entry.a]))\n\n  let prefix = 0\n  while (prefix < a.length && prefix < b.length) {\n    const entry = pairByA.get(prefix)\n    if (!entry || entry.b !== prefix || entry.differences.length > 0) break\n    prefix += 1\n  }\n\n  return {\n    pairs,\n    onlyInA: a.map((_, i) => i).filter((i) => !partnerOfA.has(i)),\n    onlyInB: b.map((_, j) => j).filter((j) => !pairedB.has(j)),\n    commonPrefixLen: prefix,\n    firstDivergence:\n      prefix === a.length && prefix === b.length\n        ? null\n        : divergenceAt(prefix, a, b, pairByA.get(prefix), partnerOfB),\n  }\n}\n\nfunction divergenceAt(\n  index: number,\n  a: readonly DiffStep[],\n  b: readonly DiffStep[],\n  pairOfA: StepPair | undefined,\n  partnerOfB: ReadonlyMap<number, number>,\n): StepDivergence {\n  const stepA = a[index]\n  const stepB = b[index]\n  const base = {\n    index,\n    a: stepA ? index : null,\n    b: stepB ? index : null,\n    differences: [] as FieldDifference[],\n  }\n  if (stepA && stepB && pairOfA?.b === index) {\n    return {\n      ...base,\n      kind: 'changed',\n      differences: pairOfA.differences,\n      reason: `at index ${index}, ${describe(stepA)} changed: ${pairOfA.differences\n        .map((d) => `${d.field} ${String(d.a)} vs ${String(d.b)}`)\n        .join('; ')}`,\n    }\n  }\n  const aUnpaired = stepA !== undefined && pairOfA === undefined\n  const bUnpaired = stepB !== undefined && !partnerOfB.has(index)\n  if (stepA && stepB && aUnpaired && bUnpaired) {\n    return {\n      ...base,\n      kind: 'replaced',\n      reason: `at index ${index}, A ran ${describe(stepA)} and B ran ${describe(stepB)}`,\n    }\n  }\n  if (stepA && aUnpaired) {\n    return {\n      ...base,\n      kind: 'only-in-a',\n      reason: `at index ${index}, only A has ${describe(stepA)}`,\n    }\n  }\n  if (stepB && bUnpaired) {\n    return {\n      ...base,\n      kind: 'only-in-b',\n      reason: `at index ${index}, only B has ${describe(stepB)}`,\n    }\n  }\n  return {\n    ...base,\n    kind: 'reordered',\n    reason: `at index ${index}, A ran ${describe(stepA!)} and B ran ${describe(stepB!)}; both steps occur in the other run at a different position`,\n  }\n}\n\nfunction sameShape(x: DiffStep, y: DiffStep): boolean {\n  return x.name === y.name && x.kind === y.kind\n}\n\nfunction differencesBetween(x: DiffStep, y: DiffStep): FieldDifference[] {\n  const differences: FieldDifference[] = []\n  if (x.kind !== y.kind) differences.push({ field: 'kind', a: x.kind, b: y.kind })\n  if (x.name !== y.name) differences.push({ field: 'name', a: x.name, b: y.name })\n  const fields = [...new Set([...Object.keys(x.fields), ...Object.keys(y.fields)])].sort()\n  for (const field of fields) {\n    const left = x.fields[field] ?? null\n    const right = y.fields[field] ?? null\n    if (left !== right) differences.push({ field, a: left, b: right })\n  }\n  return differences\n}\n\nfunction describe(step: DiffStep): string {\n  return `${step.kind} \"${step.name}\"`\n}\n\n// ── agent-eval trace store ────────────────────────────────────────────\n\nexport interface DivergenceReport {\n  runA: string\n  runB: string\n  /** Null when the runs agree step for step. */\n  firstDivergenceIndex: number | null\n  aStep?: TrajectoryStep\n  bStep?: TrajectoryStep\n  reason?: string\n  commonPrefixLen: number\n  /** The step diff behind the first divergence, over trajectory step indexes. */\n  diff: StepDiff\n}\n\nexport async function firstDivergenceView(\n  store: TraceStore,\n  runA: string,\n  runB: string,\n): Promise<DivergenceReport> {\n  const [a, b] = await Promise.all([buildTrajectory(store, runA), buildTrajectory(store, runB)])\n  const diff = diffSteps(a.steps.map(trajectoryDiffStep), b.steps.map(trajectoryDiffStep))\n  const divergence = diff.firstDivergence\n  return {\n    runA,\n    runB,\n    firstDivergenceIndex: divergence?.index ?? null,\n    ...(divergence?.a != null ? { aStep: a.steps[divergence.a] } : {}),\n    ...(divergence?.b != null ? { bStep: b.steps[divergence.b] } : {}),\n    ...(divergence ? { reason: divergence.reason } : {}),\n    commonPrefixLen: diff.commonPrefixLen,\n    diff,\n  }\n}\n\nfunction trajectoryDiffStep(step: TrajectoryStep): DiffStep {\n  const span: Span = step.span\n  const fields: Record<string, string | number | boolean | null> = {\n    status: span.status ?? null,\n  }\n  if (span.kind === 'tool') fields.tool = span.toolName\n  if (span.kind === 'llm') fields.model = span.model\n  if (span.kind === 'judge') fields.dimension = span.dimension\n  return { id: span.spanId, name: span.name, kind: span.kind, fields }\n}\n\n// ── flat OTLP spans ───────────────────────────────────────────────────\n\n/**\n * One run's spans as diff steps, in containment order: depth first, siblings\n * by start time, the same order `buildTrajectory` gives a trace store run.\n * Spans whose parent is absent, or whose parents form a cycle, start their own\n * subtree, so every span appears exactly once.\n */\nexport function diffStepsFromSpans(spans: readonly DiagnosisSpan[]): DiffStep[] {\n  const ids = new Set(spans.map((span) => span.spanId))\n  const children = new Map<string | null, DiagnosisSpan[]>()\n  for (const span of spans) {\n    const parent =\n      span.parentSpanId !== null && ids.has(span.parentSpanId) ? span.parentSpanId : null\n    const list = children.get(parent)\n    if (list) list.push(span)\n    else children.set(parent, [span])\n  }\n  // Span ids are random per run, so two runs of the same logical trace can carry\n  // different ids for \"the same\" span. Siblings that share a start time (a common\n  // case: parallel steps a driver starts together) must still sort the same way\n  // in both runs, or the diff below reports a false `reordered` divergence purely\n  // from spanId noise. End time, kind and name are content, not per-run identity,\n  // so they break the tie before spanId, which is the last resort.\n  const byStart = (x: DiagnosisSpan, y: DiagnosisSpan) =>\n    (x.startMs ?? 0) - (y.startMs ?? 0) ||\n    (x.endMs ?? 0) - (y.endMs ?? 0) ||\n    x.kind.localeCompare(y.kind) ||\n    x.name.localeCompare(y.name) ||\n    x.spanId.localeCompare(y.spanId)\n  for (const list of children.values()) list.sort(byStart)\n  const ordered: DiffStep[] = []\n  const visited = new Set<string>()\n  const visit = (span: DiagnosisSpan) => {\n    if (visited.has(span.spanId)) return\n    visited.add(span.spanId)\n    ordered.push({\n      id: span.spanId,\n      name: span.name,\n      kind: span.kind,\n      fields: { status: span.status, tool: span.toolName, model: span.model },\n    })\n    for (const child of children.get(span.spanId) ?? []) visit(child)\n  }\n  for (const root of children.get(null) ?? []) visit(root)\n  for (const span of [...spans].sort(byStart)) visit(span)\n  return ordered\n}\n","/**\n * RegressionView — compares a candidate slice to a baseline slice on a\n * named metric. Delegates the statistics (Welch's t-test, Cohen's d,\n * IQR stability) to `baseline.ts`.\n *\n * This is the entry point for CI regression gates: \"given runs tagged\n * release=A and release=B, did any metric regress?\"\n */\n\nimport { type BaselineOptions, type BaselineReport, compareToBaseline } from '../baseline'\nimport { runMetricExtractor } from '../trace/query'\nimport type { Run } from '../trace/schema'\nimport type { RunFilter, TraceStore } from '../trace/store'\n\nexport interface RegressionSpec {\n  metric: string\n  higherIsBetter: boolean\n  /** Extract a scalar from a run. Omit it and `metric` must name one of\n   *  `RUN_METRICS`; any other name is refused. */\n  extract?: (run: Run, store: TraceStore) => Promise<number | null>\n}\n\nexport interface RegressionOptions extends BaselineOptions {\n  baseline: RunFilter\n  candidate: RunFilter\n}\n\nexport async function regressionView(\n  store: TraceStore,\n  metrics: RegressionSpec[],\n  options: RegressionOptions,\n): Promise<BaselineReport> {\n  const baselineRuns = await store.listRuns(options.baseline)\n  const candidateRuns = await store.listRuns(options.candidate)\n  const samples = await Promise.all(\n    metrics.map(async (m) => {\n      const extract = m.extract ?? runMetricExtractor(m.metric)\n      const baseline = await extractAll(baselineRuns, extract, store)\n      const candidate = await extractAll(candidateRuns, extract, store)\n      return { metric: m.metric, higherIsBetter: m.higherIsBetter, baseline, candidate }\n    }),\n  )\n  return compareToBaseline(samples, options)\n}\n\nasync function extractAll(\n  runs: Run[],\n  extract: (r: Run, s: TraceStore) => Promise<number | null>,\n  store: TraceStore,\n): Promise<number[]> {\n  const out: number[] = []\n  for (const r of runs) {\n    const v = await extract(r, store)\n    if (v !== null && Number.isFinite(v)) out.push(v)\n  }\n  return out\n}\n","/**\n * StuckLoopView — detects when an agent calls the same tool with the\n * same (or structurally similar) arguments ≥ N times in a short window.\n *\n * Rationale: agents that loop are the number-one production failure\n * mode on long-horizon flows. The view returns (runId, toolName,\n * argHash, occurrences, windowMs) for each detected loop plus a\n * fraction of runs affected.\n */\n\nimport { executionTrackByLane } from '../trace/execution-tracks'\nimport { argHash, hasCapturedToolArgs } from '../trace/query'\nimport { isToolSpan, type Span, type ToolSpan } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\n\nconst DEFAULT_MAX_WINDOW_MS = 60_000\nconst DEFAULT_MAX_INTERVENING_TOOL_CALLS = 0\n\ninterface ScopedToolCall {\n  span: ToolSpan\n  scopeSpanId: string | null\n  laneSpanId: string | null\n}\n\ninterface IndexedToolCall extends ScopedToolCall {\n  toolCallIndex: number\n  trackId: string\n}\n\nexport interface StuckLoopFinding {\n  runId: string\n  toolName: string\n  argHash: string\n  /** Calls in this episode's densest qualifying interval, not the whole-run total. */\n  occurrences: number\n  spanIds: string[]\n  /** Nearest agent ancestor, or the direct parent when ancestry is incomplete. */\n  scopeSpanId?: string\n  /** Milliseconds between first and last call in the loop. */\n  windowMs: number\n}\n\nexport interface StuckLoopReport {\n  findings: StuckLoopFinding[]\n  affectedRunRatio: number\n  totalRuns: number\n}\n\nexport interface StuckLoopOptions {\n  /** Minimum call count to flag a loop (default 3). */\n  minOccurrences?: number\n  /** Maximum time between the first and last repeated call (default 60 seconds). */\n  maxWindowMs?: number\n  /**\n   * Maximum other tool calls allowed between adjacent repeats (default 0).\n   * Set to 1 to detect alternating patterns such as A,B,A,B,A.\n   */\n  maxInterveningToolCalls?: number\n  /** Filter to a specific runId; omit to scan the entire corpus. */\n  runId?: string\n}\n\nexport async function stuckLoopView(\n  store: TraceStore,\n  options: StuckLoopOptions = {},\n): Promise<StuckLoopReport> {\n  const minOccurrences = options.minOccurrences ?? 3\n  const maxWindowMs = options.maxWindowMs ?? DEFAULT_MAX_WINDOW_MS\n  const maxInterveningToolCalls =\n    options.maxInterveningToolCalls ?? DEFAULT_MAX_INTERVENING_TOOL_CALLS\n  if (!Number.isInteger(minOccurrences) || minOccurrences < 1) {\n    throw new RangeError('minOccurrences must be a positive integer')\n  }\n  if (!Number.isFinite(maxWindowMs) || maxWindowMs < 0) {\n    throw new RangeError('maxWindowMs must be a finite non-negative number')\n  }\n  if (!Number.isInteger(maxInterveningToolCalls) || maxInterveningToolCalls < 0) {\n    throw new RangeError('maxInterveningToolCalls must be a non-negative integer')\n  }\n  const runs = options.runId\n    ? [{ runId: options.runId }]\n    : (await store.listRuns()).map((r) => ({ runId: r.runId }))\n\n  const findings: StuckLoopFinding[] = []\n  for (const { runId } of runs) {\n    const spans = await store.spans({ runId })\n    const spansById = new Map(spans.map((span) => [span.spanId, span]))\n    const scopedTools: ScopedToolCall[] = spans\n      .filter(isToolSpan)\n      .map((span, sourceIndex) => ({ span, sourceIndex }))\n      .sort((a, b) => a.span.startedAt - b.span.startedAt || a.sourceIndex - b.sourceIndex)\n      .map(({ span }) => ({ span, ...executionScope(span, spansById) }))\n    const trackByLane = executionTrackByLane(\n      scopedTools.map((call) => {\n        const direct = call.laneSpanId === null || call.laneSpanId === call.scopeSpanId\n        const timed = direct\n          ? call.span\n          : call.laneSpanId\n            ? spansById.get(call.laneSpanId)\n            : undefined\n        return {\n          key: executionKey(call),\n          scopeKey: JSON.stringify(call.scopeSpanId),\n          start: timed?.startedAt ?? null,\n          end: timed?.endedAt ?? null,\n        }\n      }),\n    )\n    const nextToolIndexByTrack = new Map<string, number>()\n    const orderedTools: IndexedToolCall[] = scopedTools.map((call) => {\n      const trackId = trackByLane.get(executionKey(call))!\n      const toolCallIndex = nextToolIndexByTrack.get(trackId) ?? 0\n      nextToolIndexByTrack.set(trackId, toolCallIndex + 1)\n      return { ...call, toolCallIndex, trackId }\n    })\n    const byKey = new Map<\n      string,\n      {\n        calls: IndexedToolCall[]\n        argHash: string\n        toolName: string\n        scopeSpanId: string | null\n      }\n    >()\n    for (const call of orderedTools) {\n      if (!hasCapturedToolArgs(call.span)) continue\n      const h = argHash(call.span.args)\n      const key = JSON.stringify([call.trackId, call.span.toolName, h])\n      const bucket = byKey.get(key) ?? {\n        calls: [],\n        argHash: h,\n        toolName: call.span.toolName,\n        scopeSpanId: call.scopeSpanId,\n      }\n      bucket.calls.push(call)\n      byKey.set(key, bucket)\n    }\n    for (const { calls, argHash: h, toolName, scopeSpanId } of byKey.values()) {\n      if (calls.length < minOccurrences) continue\n      let episodeStart = 0\n      for (let episodeEnd = 1; episodeEnd <= calls.length; episodeEnd += 1) {\n        const previous = calls[episodeEnd - 1]!\n        const next = calls[episodeEnd]\n        const episodeEnded =\n          next === undefined ||\n          next.span.startedAt - previous.span.startedAt > maxWindowMs ||\n          next.toolCallIndex - previous.toolCallIndex - 1 > maxInterveningToolCalls ||\n          !callsAreSerial(previous, next)\n        if (!episodeEnded) continue\n\n        const episode = calls.slice(episodeStart, episodeEnd)\n        let left = 0\n        let bestStart = 0\n        let bestEnd = -1\n        for (let right = 0; right < episode.length; right += 1) {\n          while (episode[right]!.span.startedAt - episode[left]!.span.startedAt > maxWindowMs) {\n            left += 1\n          }\n          if (right - left > bestEnd - bestStart) {\n            bestStart = left\n            bestEnd = right\n          }\n        }\n        if (bestEnd - bestStart + 1 >= minOccurrences) {\n          const loop = episode.slice(bestStart, bestEnd + 1)\n          const first = loop[0]!.span.startedAt\n          const last = loop[loop.length - 1]!.span.startedAt\n          findings.push({\n            runId,\n            toolName,\n            argHash: h,\n            occurrences: loop.length,\n            spanIds: loop.map((call) => call.span.spanId),\n            ...(scopeSpanId ? { scopeSpanId } : {}),\n            windowMs: last - first,\n          })\n        }\n        episodeStart = episodeEnd\n      }\n    }\n  }\n\n  const affectedRuns = new Set(findings.map((f) => f.runId))\n  return {\n    findings,\n    affectedRunRatio: runs.length > 0 ? affectedRuns.size / runs.length : 0,\n    totalRuns: runs.length,\n  }\n}\n\nfunction laneKey(call: Pick<ScopedToolCall, 'scopeSpanId' | 'laneSpanId'>): string {\n  return JSON.stringify([call.scopeSpanId, call.laneSpanId])\n}\n\nfunction executionKey(call: ScopedToolCall): string {\n  return call.laneSpanId === null || call.laneSpanId === call.scopeSpanId\n    ? JSON.stringify([call.scopeSpanId, call.span.spanId])\n    : laneKey(call)\n}\n\nfunction executionScope(\n  span: ToolSpan,\n  spansById: ReadonlyMap<string, Span>,\n): { scopeSpanId: string | null; laneSpanId: string | null } {\n  const directParent = span.parentSpanId\n  if (!directParent) return { scopeSpanId: null, laneSpanId: null }\n\n  let currentId: string | undefined = directParent\n  let laneSpanId: string | null = null\n  const seen = new Set<string>()\n  while (currentId && !seen.has(currentId)) {\n    seen.add(currentId)\n    const current = spansById.get(currentId)\n    if (!current) return { scopeSpanId: directParent, laneSpanId: directParent }\n    if (current.kind === 'agent') {\n      return { scopeSpanId: current.spanId, laneSpanId: laneSpanId ?? current.spanId }\n    }\n    laneSpanId = current.spanId\n    currentId = current.parentSpanId\n  }\n  return { scopeSpanId: directParent, laneSpanId: directParent }\n}\n\nfunction callsAreSerial(previous: IndexedToolCall, next: IndexedToolCall): boolean {\n  return previous.span.endedAt !== undefined && previous.span.endedAt <= next.span.startedAt\n}\n"],"mappings":";;;;;;;AA+EA,SAAgB,UAAU,GAAwB,GAAkC;CAClF,MAAM,6BAAa,IAAI,IAAkD;CACzE,MAAM,0BAAU,IAAI,IAAY;CAChC,MAAM,QAAQ,GAAW,GAAW,aAA0B;EAC5D,WAAW,IAAI,GAAG;GAAE,GAAG;GAAG;EAAS,CAAC;EACpC,QAAQ,IAAI,CAAC;CACf;CAEA,MAAM,wBAAQ,IAAI,IAAsB;CACxC,EAAE,SAAS,MAAM,MAAM;EACrB,MAAM,OAAO,MAAM,IAAI,KAAK,EAAE;EAC9B,IAAI,MAAM,KAAK,KAAK,CAAC;OAChB,MAAM,IAAI,KAAK,IAAI,CAAC,CAAC,CAAC;CAC7B,CAAC;CACD,EAAE,SAAS,MAAM,MAAM;EACrB,MAAM,IAAI,MAAM,IAAI,KAAK,EAAE,CAAC,EAAE,MAAM,cAAc,CAAC,QAAQ,IAAI,SAAS,CAAC;EACzE,IAAI,MAAM,KAAA,GAAW,KAAK,GAAG,GAAG,IAAI;CACtC,CAAC;CACD,EAAE,SAAS,MAAM,MAAM;EACrB,IAAI,WAAW,IAAI,CAAC,KAAK,KAAK,EAAE,UAAU,QAAQ,IAAI,CAAC,GAAG;EAC1D,IAAI,UAAU,MAAM,EAAE,EAAG,GAAG,KAAK,GAAG,GAAG,UAAU;CACnD,CAAC;CACD,EAAE,SAAS,MAAM,MAAM;EACrB,IAAI,WAAW,IAAI,CAAC,GAAG;EACvB,MAAM,IAAI,EAAE,WAAW,WAAW,UAAU,CAAC,QAAQ,IAAI,KAAK,KAAK,UAAU,MAAM,SAAS,CAAC;EAC7F,IAAI,KAAK,GAAG,KAAK,GAAG,GAAG,MAAM;CAC/B,CAAC;CAED,MAAM,QAAoB,CAAC,GAAG,WAAW,QAAQ,CAAC,CAAC,CAChD,MAAM,CAAC,IAAI,CAAC,OAAO,IAAI,CAAC,CAAC,CACzB,KAAK,CAAC,GAAG,cAAc;EACtB,GAAG;EACH,GAAG,QAAQ;EACX,UAAU,QAAQ;EAClB,aAAa,mBAAmB,EAAE,IAAK,EAAE,QAAQ,EAAG;CACtD,EAAE;CACJ,MAAM,UAAU,IAAI,IAAI,MAAM,KAAK,UAAU,CAAC,MAAM,GAAG,KAAK,CAAC,CAAC;CAC9D,MAAM,aAAa,IAAI,IAAI,MAAM,KAAK,UAAU,CAAC,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC;CAEnE,IAAI,SAAS;CACb,OAAO,SAAS,EAAE,UAAU,SAAS,EAAE,QAAQ;EAC7C,MAAM,QAAQ,QAAQ,IAAI,MAAM;EAChC,IAAI,CAAC,SAAS,MAAM,MAAM,UAAU,MAAM,YAAY,SAAS,GAAG;EAClE,UAAU;CACZ;CAEA,OAAO;EACL;EACA,SAAS,EAAE,KAAK,GAAG,MAAM,CAAC,CAAC,CAAC,QAAQ,MAAM,CAAC,WAAW,IAAI,CAAC,CAAC;EAC5D,SAAS,EAAE,KAAK,GAAG,MAAM,CAAC,CAAC,CAAC,QAAQ,MAAM,CAAC,QAAQ,IAAI,CAAC,CAAC;EACzD,iBAAiB;EACjB,iBACE,WAAW,EAAE,UAAU,WAAW,EAAE,SAChC,OACA,aAAa,QAAQ,GAAG,GAAG,QAAQ,IAAI,MAAM,GAAG,UAAU;CAClE;AACF;AAEA,SAAS,aACP,OACA,GACA,GACA,SACA,YACgB;CAChB,MAAM,QAAQ,EAAE;CAChB,MAAM,QAAQ,EAAE;CAChB,MAAM,OAAO;EACX;EACA,GAAG,QAAQ,QAAQ;EACnB,GAAG,QAAQ,QAAQ;EACnB,aAAa,CAAC;CAChB;CACA,IAAI,SAAS,SAAS,SAAS,MAAM,OACnC,OAAO;EACL,GAAG;EACH,MAAM;EACN,aAAa,QAAQ;EACrB,QAAQ,YAAY,MAAM,IAAI,SAAS,KAAK,EAAE,YAAY,QAAQ,YAC/D,KAAK,MAAM,GAAG,EAAE,MAAM,GAAG,OAAO,EAAE,CAAC,EAAE,MAAM,OAAO,EAAE,CAAC,GAAG,CAAC,CACzD,KAAK,IAAI;CACd;CAEF,MAAM,YAAY,UAAU,KAAA,KAAa,YAAY,KAAA;CACrD,MAAM,YAAY,UAAU,KAAA,KAAa,CAAC,WAAW,IAAI,KAAK;CAC9D,IAAI,SAAS,SAAS,aAAa,WACjC,OAAO;EACL,GAAG;EACH,MAAM;EACN,QAAQ,YAAY,MAAM,UAAU,SAAS,KAAK,EAAE,aAAa,SAAS,KAAK;CACjF;CAEF,IAAI,SAAS,WACX,OAAO;EACL,GAAG;EACH,MAAM;EACN,QAAQ,YAAY,MAAM,eAAe,SAAS,KAAK;CACzD;CAEF,IAAI,SAAS,WACX,OAAO;EACL,GAAG;EACH,MAAM;EACN,QAAQ,YAAY,MAAM,eAAe,SAAS,KAAK;CACzD;CAEF,OAAO;EACL,GAAG;EACH,MAAM;EACN,QAAQ,YAAY,MAAM,UAAU,SAAS,KAAM,EAAE,aAAa,SAAS,KAAM,EAAE;CACrF;AACF;AAEA,SAAS,UAAU,GAAa,GAAsB;CACpD,OAAO,EAAE,SAAS,EAAE,QAAQ,EAAE,SAAS,EAAE;AAC3C;AAEA,SAAS,mBAAmB,GAAa,GAAgC;CACvE,MAAM,cAAiC,CAAC;CACxC,IAAI,EAAE,SAAS,EAAE,MAAM,YAAY,KAAK;EAAE,OAAO;EAAQ,GAAG,EAAE;EAAM,GAAG,EAAE;CAAK,CAAC;CAC/E,IAAI,EAAE,SAAS,EAAE,MAAM,YAAY,KAAK;EAAE,OAAO;EAAQ,GAAG,EAAE;EAAM,GAAG,EAAE;CAAK,CAAC;CAC/E,MAAM,SAAS,CAAC,mBAAG,IAAI,IAAI,CAAC,GAAG,OAAO,KAAK,EAAE,MAAM,GAAG,GAAG,OAAO,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK;CACvF,KAAK,MAAM,SAAS,QAAQ;EAC1B,MAAM,OAAO,EAAE,OAAO,UAAU;EAChC,MAAM,QAAQ,EAAE,OAAO,UAAU;EACjC,IAAI,SAAS,OAAO,YAAY,KAAK;GAAE;GAAO,GAAG;GAAM,GAAG;EAAM,CAAC;CACnE;CACA,OAAO;AACT;AAEA,SAAS,SAAS,MAAwB;CACxC,OAAO,GAAG,KAAK,KAAK,IAAI,KAAK,KAAK;AACpC;AAiBA,eAAsB,oBACpB,OACA,MACA,MAC2B;CAC3B,MAAM,CAAC,GAAG,KAAK,MAAM,QAAQ,IAAI,CAAC,gBAAgB,OAAO,IAAI,GAAG,gBAAgB,OAAO,IAAI,CAAC,CAAC;CAC7F,MAAM,OAAO,UAAU,EAAE,MAAM,IAAI,kBAAkB,GAAG,EAAE,MAAM,IAAI,kBAAkB,CAAC;CACvF,MAAM,aAAa,KAAK;CACxB,OAAO;EACL;EACA;EACA,sBAAsB,YAAY,SAAS;EAC3C,GAAI,YAAY,KAAK,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,GAAG,IAAI,CAAC;EAChE,GAAI,YAAY,KAAK,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,GAAG,IAAI,CAAC;EAChE,GAAI,aAAa,EAAE,QAAQ,WAAW,OAAO,IAAI,CAAC;EAClD,iBAAiB,KAAK;EACtB;CACF;AACF;AAEA,SAAS,mBAAmB,MAAgC;CAC1D,MAAM,OAAa,KAAK;CACxB,MAAM,SAA2D,EAC/D,QAAQ,KAAK,UAAU,KACzB;CACA,IAAI,KAAK,SAAS,QAAQ,OAAO,OAAO,KAAK;CAC7C,IAAI,KAAK,SAAS,OAAO,OAAO,QAAQ,KAAK;CAC7C,IAAI,KAAK,SAAS,SAAS,OAAO,YAAY,KAAK;CACnD,OAAO;EAAE,IAAI,KAAK;EAAQ,MAAM,KAAK;EAAM,MAAM,KAAK;EAAM;CAAO;AACrE;;;;;;;AAUA,SAAgB,mBAAmB,OAA6C;CAC9E,MAAM,MAAM,IAAI,IAAI,MAAM,KAAK,SAAS,KAAK,MAAM,CAAC;CACpD,MAAM,2BAAW,IAAI,IAAoC;CACzD,KAAK,MAAM,QAAQ,OAAO;EACxB,MAAM,SACJ,KAAK,iBAAiB,QAAQ,IAAI,IAAI,KAAK,YAAY,IAAI,KAAK,eAAe;EACjF,MAAM,OAAO,SAAS,IAAI,MAAM;EAChC,IAAI,MAAM,KAAK,KAAK,IAAI;OACnB,SAAS,IAAI,QAAQ,CAAC,IAAI,CAAC;CAClC;CAOA,MAAM,WAAW,GAAkB,OAChC,EAAE,WAAW,MAAM,EAAE,WAAW,OAChC,EAAE,SAAS,MAAM,EAAE,SAAS,MAC7B,EAAE,KAAK,cAAc,EAAE,IAAI,KAC3B,EAAE,KAAK,cAAc,EAAE,IAAI,KAC3B,EAAE,OAAO,cAAc,EAAE,MAAM;CACjC,KAAK,MAAM,QAAQ,SAAS,OAAO,GAAG,KAAK,KAAK,OAAO;CACvD,MAAM,UAAsB,CAAC;CAC7B,MAAM,0BAAU,IAAI,IAAY;CAChC,MAAM,SAAS,SAAwB;EACrC,IAAI,QAAQ,IAAI,KAAK,MAAM,GAAG;EAC9B,QAAQ,IAAI,KAAK,MAAM;EACvB,QAAQ,KAAK;GACX,IAAI,KAAK;GACT,MAAM,KAAK;GACX,MAAM,KAAK;GACX,QAAQ;IAAE,QAAQ,KAAK;IAAQ,MAAM,KAAK;IAAU,OAAO,KAAK;GAAM;EACxE,CAAC;EACD,KAAK,MAAM,SAAS,SAAS,IAAI,KAAK,MAAM,KAAK,CAAC,GAAG,MAAM,KAAK;CAClE;CACA,KAAK,MAAM,QAAQ,SAAS,IAAI,IAAI,KAAK,CAAC,GAAG,MAAM,IAAI;CACvD,KAAK,MAAM,QAAQ,CAAC,GAAG,KAAK,CAAC,CAAC,KAAK,OAAO,GAAG,MAAM,IAAI;CACvD,OAAO;AACT;;;;;;;;;;;ACvRA,eAAsB,eACpB,OACA,SACA,SACyB;CACzB,MAAM,eAAe,MAAM,MAAM,SAAS,QAAQ,QAAQ;CAC1D,MAAM,gBAAgB,MAAM,MAAM,SAAS,QAAQ,SAAS;CAS5D,OAAO,kBAAkB,MARH,QAAQ,IAC5B,QAAQ,IAAI,OAAO,MAAM;EACvB,MAAM,UAAU,EAAE,WAAW,mBAAmB,EAAE,MAAM;EACxD,MAAM,WAAW,MAAM,WAAW,cAAc,SAAS,KAAK;EAC9D,MAAM,YAAY,MAAM,WAAW,eAAe,SAAS,KAAK;EAChE,OAAO;GAAE,QAAQ,EAAE;GAAQ,gBAAgB,EAAE;GAAgB;GAAU;EAAU;CACnF,CAAC,CACH,GACkC,OAAO;AAC3C;AAEA,eAAe,WACb,MACA,SACA,OACmB;CACnB,MAAM,MAAgB,CAAC;CACvB,KAAK,MAAM,KAAK,MAAM;EACpB,MAAM,IAAI,MAAM,QAAQ,GAAG,KAAK;EAChC,IAAI,MAAM,QAAQ,OAAO,SAAS,CAAC,GAAG,IAAI,KAAK,CAAC;CAClD;CACA,OAAO;AACT;;;;;;;;;;;;ACzCA,MAAM,wBAAwB;AAC9B,MAAM,qCAAqC;AA8C3C,eAAsB,cACpB,OACA,UAA4B,CAAC,GACH;CAC1B,MAAM,iBAAiB,QAAQ,kBAAkB;CACjD,MAAM,cAAc,QAAQ,eAAe;CAC3C,MAAM,0BACJ,QAAQ,2BAA2B;CACrC,IAAI,CAAC,OAAO,UAAU,cAAc,KAAK,iBAAiB,GACxD,MAAM,IAAI,WAAW,2CAA2C;CAElE,IAAI,CAAC,OAAO,SAAS,WAAW,KAAK,cAAc,GACjD,MAAM,IAAI,WAAW,kDAAkD;CAEzE,IAAI,CAAC,OAAO,UAAU,uBAAuB,KAAK,0BAA0B,GAC1E,MAAM,IAAI,WAAW,wDAAwD;CAE/E,MAAM,OAAO,QAAQ,QACjB,CAAC,EAAE,OAAO,QAAQ,MAAM,CAAC,KACxB,MAAM,MAAM,SAAS,EAAA,CAAG,KAAK,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE;CAE5D,MAAM,WAA+B,CAAC;CACtC,KAAK,MAAM,EAAE,WAAW,MAAM;EAC5B,MAAM,QAAQ,MAAM,MAAM,MAAM,EAAE,MAAM,CAAC;EACzC,MAAM,YAAY,IAAI,IAAI,MAAM,KAAK,SAAS,CAAC,KAAK,QAAQ,IAAI,CAAC,CAAC;EAClE,MAAM,cAAgC,MACnC,OAAO,UAAU,CAAC,CAClB,KAAK,MAAM,iBAAiB;GAAE;GAAM;EAAY,EAAE,CAAC,CACnD,MAAM,GAAG,MAAM,EAAE,KAAK,YAAY,EAAE,KAAK,aAAa,EAAE,cAAc,EAAE,WAAW,CAAC,CACpF,KAAK,EAAE,YAAY;GAAE;GAAM,GAAG,eAAe,MAAM,SAAS;EAAE,EAAE;EACnE,MAAM,cAAc,qBAClB,YAAY,KAAK,SAAS;GAExB,MAAM,QADS,KAAK,eAAe,QAAQ,KAAK,eAAe,KAAK,cAEhE,KAAK,OACL,KAAK,aACH,UAAU,IAAI,KAAK,UAAU,IAC7B,KAAA;GACN,OAAO;IACL,KAAK,aAAa,IAAI;IACtB,UAAU,KAAK,UAAU,KAAK,WAAW;IACzC,OAAO,OAAO,aAAa;IAC3B,KAAK,OAAO,WAAW;GACzB;EACF,CAAC,CACH;EACA,MAAM,uCAAuB,IAAI,IAAoB;EACrD,MAAM,eAAkC,YAAY,KAAK,SAAS;GAChE,MAAM,UAAU,YAAY,IAAI,aAAa,IAAI,CAAC;GAClD,MAAM,gBAAgB,qBAAqB,IAAI,OAAO,KAAK;GAC3D,qBAAqB,IAAI,SAAS,gBAAgB,CAAC;GACnD,OAAO;IAAE,GAAG;IAAM;IAAe;GAAQ;EAC3C,CAAC;EACD,MAAM,wBAAQ,IAAI,IAQhB;EACF,KAAK,MAAM,QAAQ,cAAc;GAC/B,IAAI,CAAC,oBAAoB,KAAK,IAAI,GAAG;GACrC,MAAM,IAAI,QAAQ,KAAK,KAAK,IAAI;GAChC,MAAM,MAAM,KAAK,UAAU;IAAC,KAAK;IAAS,KAAK,KAAK;IAAU;GAAC,CAAC;GAChE,MAAM,SAAS,MAAM,IAAI,GAAG,KAAK;IAC/B,OAAO,CAAC;IACR,SAAS;IACT,UAAU,KAAK,KAAK;IACpB,aAAa,KAAK;GACpB;GACA,OAAO,MAAM,KAAK,IAAI;GACtB,MAAM,IAAI,KAAK,MAAM;EACvB;EACA,KAAK,MAAM,EAAE,OAAO,SAAS,GAAG,UAAU,iBAAiB,MAAM,OAAO,GAAG;GACzE,IAAI,MAAM,SAAS,gBAAgB;GACnC,IAAI,eAAe;GACnB,KAAK,IAAI,aAAa,GAAG,cAAc,MAAM,QAAQ,cAAc,GAAG;IACpE,MAAM,WAAW,MAAM,aAAa;IACpC,MAAM,OAAO,MAAM;IAMnB,IAAI,EAJF,SAAS,KAAA,KACT,KAAK,KAAK,YAAY,SAAS,KAAK,YAAY,eAChD,KAAK,gBAAgB,SAAS,gBAAgB,IAAI,2BAClD,CAAC,eAAe,UAAU,IAAI,IACb;IAEnB,MAAM,UAAU,MAAM,MAAM,cAAc,UAAU;IACpD,IAAI,OAAO;IACX,IAAI,YAAY;IAChB,IAAI,UAAU;IACd,KAAK,IAAI,QAAQ,GAAG,QAAQ,QAAQ,QAAQ,SAAS,GAAG;KACtD,OAAO,QAAQ,MAAM,CAAE,KAAK,YAAY,QAAQ,KAAK,CAAE,KAAK,YAAY,aACtE,QAAQ;KAEV,IAAI,QAAQ,OAAO,UAAU,WAAW;MACtC,YAAY;MACZ,UAAU;KACZ;IACF;IACA,IAAI,UAAU,YAAY,KAAK,gBAAgB;KAC7C,MAAM,OAAO,QAAQ,MAAM,WAAW,UAAU,CAAC;KACjD,MAAM,QAAQ,KAAK,EAAE,CAAE,KAAK;KAC5B,MAAM,OAAO,KAAK,KAAK,SAAS,EAAE,CAAE,KAAK;KACzC,SAAS,KAAK;MACZ;MACA;MACA,SAAS;MACT,aAAa,KAAK;MAClB,SAAS,KAAK,KAAK,SAAS,KAAK,KAAK,MAAM;MAC5C,GAAI,cAAc,EAAE,YAAY,IAAI,CAAC;MACrC,UAAU,OAAO;KACnB,CAAC;IACH;IACA,eAAe;GACjB;EACF;CACF;CAEA,MAAM,eAAe,IAAI,IAAI,SAAS,KAAK,MAAM,EAAE,KAAK,CAAC;CACzD,OAAO;EACL;EACA,kBAAkB,KAAK,SAAS,IAAI,aAAa,OAAO,KAAK,SAAS;EACtE,WAAW,KAAK;CAClB;AACF;AAEA,SAAS,QAAQ,MAAkE;CACjF,OAAO,KAAK,UAAU,CAAC,KAAK,aAAa,KAAK,UAAU,CAAC;AAC3D;AAEA,SAAS,aAAa,MAA8B;CAClD,OAAO,KAAK,eAAe,QAAQ,KAAK,eAAe,KAAK,cACxD,KAAK,UAAU,CAAC,KAAK,aAAa,KAAK,KAAK,MAAM,CAAC,IACnD,QAAQ,IAAI;AAClB;AAEA,SAAS,eACP,MACA,WAC2D;CAC3D,MAAM,eAAe,KAAK;CAC1B,IAAI,CAAC,cAAc,OAAO;EAAE,aAAa;EAAM,YAAY;CAAK;CAEhE,IAAI,YAAgC;CACpC,IAAI,aAA4B;CAChC,MAAM,uBAAO,IAAI,IAAY;CAC7B,OAAO,aAAa,CAAC,KAAK,IAAI,SAAS,GAAG;EACxC,KAAK,IAAI,SAAS;EAClB,MAAM,UAAU,UAAU,IAAI,SAAS;EACvC,IAAI,CAAC,SAAS,OAAO;GAAE,aAAa;GAAc,YAAY;EAAa;EAC3E,IAAI,QAAQ,SAAS,SACnB,OAAO;GAAE,aAAa,QAAQ;GAAQ,YAAY,cAAc,QAAQ;EAAO;EAEjF,aAAa,QAAQ;EACrB,YAAY,QAAQ;CACtB;CACA,OAAO;EAAE,aAAa;EAAc,YAAY;CAAa;AAC/D;AAEA,SAAS,eAAe,UAA2B,MAAgC;CACjF,OAAO,SAAS,KAAK,YAAY,KAAA,KAAa,SAAS,KAAK,WAAW,KAAK,KAAK;AACnF"}

package/dist/{produced-state-zSaGDC_Q.js → produced-state-BXjXJHJT.js}

@@ -1,10 +1,10 @@
 import { s as ValidationError } from "./errors-Dngq5h35.js";
 import { r as canonicalString } from "./canonical-BsZVgBhW.js";
-import { n as JudgeParseError } from "./llm-judge-bwbgDMP0.js";
+import { n as JudgeParseError } from "./llm-judge-BnxAv7ed.js";
 import { i as CostLedger } from "./cost-ledger-_IbkejCF.js";
-import { O as recoverTruncatedJson } from "./campaign-evidence-CWF8pEIk.js";
+import { O as recoverTruncatedJson } from "./campaign-evidence-D40-LHIL.js";
 import { t as certificationEvidenceDigest } from "./verdict-DhPoXApk.js";
-import { f as ModelSubstitutionError, h as assertServedModel, o as costReceiptFromLlm, s as costReceiptFromLlmError, u as maximumChargeForLlmRequest } from "./llm-client-BqmRnibA.js";
+import { f as ModelSubstitutionError, h as assertServedModel, o as costReceiptFromLlm, s as costReceiptFromLlmError, u as maximumChargeForLlmRequest } from "./llm-client-KYQBh9C8.js";
 import { createHash, randomUUID } from "node:crypto";
 import { harnessSupportsModel } from "@tangle-network/agent-interface";
 //#region src/agent-profile.ts
@@ -588,4 +588,4 @@ function extractProducedState(events) {
 //#endregion
 export { CODING_HARNESSES as a, agentProfileId as c, harnessAxisOf as d, verifyCompletion as i, agentProfileModelId as l, completionVerdict as n, HARNESS_NATIVE_MODEL as o, createLlmCorrectnessChecker as r, agentProfileHash as s, extractProducedState as t, expandProfileAxes as u };
 
-//# sourceMappingURL=produced-state-zSaGDC_Q.js.map
+//# sourceMappingURL=produced-state-BXjXJHJT.js.map

package/dist/{produced-state-zSaGDC_Q.js.map → produced-state-BXjXJHJT.js.map}

@@ -1 +1 @@
-{"version":3,"file":"produced-state-zSaGDC_Q.js","names":[],"sources":["../src/agent-profile.ts","../src/completion-verifier.ts","../src/produced-state.ts"],"sourcesContent":["import { createHash } from 'node:crypto'\nimport type { AgentProfile } from '@tangle-network/agent-interface'\nimport { type HarnessType, harnessSupportsModel } from '@tangle-network/agent-interface'\nimport { ValidationError } from './errors'\nimport { canonicalString } from './ledger-core/canonical'\n\nexport type { AgentProfile, HarnessType } from '@tangle-network/agent-interface'\n\n/**\n * The agentic coding harnesses an eval sweeps by default — the ones we care about\n * ranking. This is the SINGLE source of that list; consumers import it instead of\n * re-declaring their own (a re-declared list is how the fleet drifts). Pass an\n * explicit `harnesses` (e.g. `harnessTypeSchema.options` for literally every known\n * harness) to widen beyond these.\n */\nexport const CODING_HARNESSES: readonly HarnessType[] = [\n  'opencode',\n  'claude-code',\n  'codex',\n  'kimi-code',\n]\n\nexport interface ProfileAxisSpec {\n  /** The domain profile to sweep. Its prompt/tools/skills are held fixed; only the\n   *  harness and model vary. `model.default` is the fallback model. */\n  base: AgentProfile\n  /** Harnesses to cross. Default: {@link CODING_HARNESSES}. */\n  harnesses?: readonly HarnessType[]\n  /** Models to cross. Default: `[base.model.default]` — one model, i.e. today's\n   *  single-model behaviour, so omitting this never changes an existing run. */\n  models?: readonly string[]\n  /** Force every (harness, model) pair verbatim, even ones the harness can't run —\n   *  for deliberately testing failure modes. Default (false): SNAP instead — a\n   *  vendor-locked harness runs only the swept models in its family, or its native\n   *  default when it supports none, so no harness is dropped and none gets a\n   *  guaranteed-failing foreign-model cell. */\n  keepIncompatible?: boolean\n}\n\n/** Model sentinel for a vendor-locked harness that supports none of the swept models:\n *  it carries no provider prefix, so `harnessSupportsModel` accepts it and the harness\n *  resolves it to its own native default model at runtime (e.g. kimi-code → its Kimi\n *  model). Lets `expandProfileAxes` snap-instead-of-drop without a per-harness flagship\n *  table that would rot as router catalogs change. */\nexport const HARNESS_NATIVE_MODEL = 'default'\n\n/**\n * Expand a base profile across the harness × model matrix into the `AgentProfile[]`\n * that `runProfileMatrix` / `selfImprove` score — the ONE place \"which harnesses ×\n * which models do we evaluate\" lives, so no product hand-rolls its own harness list\n * or column→profile mapping (the pattern that let those copies drift and silently\n * break the harness pivot).\n *\n * Each cell clones `base`, sets the canonical top-level `harness` and `model.default`,\n * and stamps `metadata.harness` + `metadata.harnessModel` for matrix grouping. Both\n * metadata fields are hash-bearing, so every cell gets a distinct `agentProfileId` row\n * and results join back by harness/model via {@link harnessAxisOf} with no\n * hand-recomputed key. A vendor-locked harness snaps to its family's swept models — or\n * its native default ({@link HARNESS_NATIVE_MODEL}) when it supports none — so every\n * requested harness runs; `keepIncompatible` forces every pair verbatim.\n *\n * Omit `harnesses`/`models` to sweep the full default set — the \"turn it on for\n * everything we care about\" switch, identical in shape whether one harness or all.\n */\nexport function expandProfileAxes(spec: ProfileAxisSpec): AgentProfile[] {\n  const harnesses = spec.harnesses ?? CODING_HARNESSES\n  if (harnesses.length === 0) throw new ValidationError('expandProfileAxes: no harnesses to sweep')\n  const baseModel = spec.base.model?.default\n  const models = spec.models ?? (baseModel ? [baseModel] : [])\n  if (models.length === 0) {\n    throw new ValidationError(\n      'expandProfileAxes: no models to sweep — base profile has no model.default and none were supplied',\n    )\n  }\n  const out: AgentProfile[] = []\n  const seen = new Set<string>()\n  for (const harness of harnesses) {\n    // A universal (router-backed) harness — opencode/pi/claudish — runs every swept\n    // model. A vendor-locked harness — codex/claude-code/kimi-code — runs only the\n    // swept models in its own family; when it supports NONE of them it snaps to its\n    // native default (the `HARNESS_NATIVE_MODEL` sentinel it resolves at runtime)\n    // rather than being dropped, so every requested harness still appears in the\n    // sweep on a model it can actually run — e.g. sweeping `deepseek/x` puts opencode\n    // on deepseek and kimi-code on its own Kimi model, a real head-to-head.\n    // `keepIncompatible` forces every (harness, model) pair verbatim (failure-mode runs).\n    const supported = spec.keepIncompatible\n      ? models\n      : models.filter((model) => harnessSupportsModel(harness, model))\n    const effective = supported.length > 0 ? supported : [HARNESS_NATIVE_MODEL]\n    for (const model of effective) {\n      const profile: AgentProfile = {\n        ...spec.base,\n        harness,\n        name: `${spec.base.name ?? 'agent'}/${harness}/${model}`,\n        model: { ...spec.base.model, default: model },\n        metadata: { ...(spec.base.metadata ?? {}), harness, harnessModel: model },\n      }\n      const id = agentProfileId(profile)\n      if (seen.has(id)) continue\n      seen.add(id)\n      out.push(profile)\n    }\n  }\n  if (out.length === 0) {\n    // Unreachable in normal use — snapping guarantees ≥1 cell per harness — but keep a\n    // fail-closed guard so a future refactor can't silently produce an empty sweep.\n    throw new ValidationError(\n      `expandProfileAxes: produced no profiles (harnesses=[${harnesses.join(', ')}], models=[${models.join(', ')}]).`,\n    )\n  }\n  return out\n}\n\n/**\n * Read the (harness, model) a matrix cell ran under, off a profile or a result row's\n * profile — the join-back for a `byHarness` pivot. Returns undefined when the profile\n * wasn't produced by {@link expandProfileAxes}. Callers group `result.byProfile` by\n * this instead of recomputing an id (recomputing the wrong key is what broke the pivot\n * in the hand-rolled copies).\n */\nexport function harnessAxisOf(\n  profile: Pick<AgentProfile, 'metadata'>,\n): { harness: HarnessType; model: string } | undefined {\n  const m = profile.metadata as Record<string, unknown> | undefined\n  const harness = m?.harness\n  const model = m?.harnessModel\n  if (typeof harness === 'string' && typeof model === 'string') {\n    return { harness: harness as HarnessType, model }\n  }\n  return undefined\n}\n\n/**\n * Collision-resistant, path-safe, human-readable profile id for eval artifacts.\n * Scorecard joins still use `agentProfileHash`; this id is for run ids, matrix\n * keys, and directory names where two profiles must not collapse onto one row.\n * The suffix is the first 64 bits of the behaviour hash, enough for ordinary\n * eval matrices while keeping filenames readable.\n */\nexport function agentProfileId(profile: AgentProfile): string {\n  const label = pathSafeProfileLabel(agentProfileDisplayLabel(profile)) ?? 'profile'\n  return `${label}-${agentProfileHash(profile).slice(0, 16)}`\n}\n\n/**\n * Model snapshot used for `RunRecord.model`. Eval surfaces require a concrete\n * model id because run records reject bare/missing model aliases.\n */\nexport function agentProfileModelId(profile: AgentProfile): string {\n  const model = profile.model?.default?.trim()\n  if (!model) {\n    const label = agentProfileDisplayLabel(profile) ?? 'unnamed profile'\n    throw new ValidationError(\n      `AgentProfile \"${label}\" has no model.default — cannot record eval run`,\n    )\n  }\n  return model\n}\n\nfunction agentProfileDisplayLabel(profile: AgentProfile): string | undefined {\n  return profile.name?.trim() || profile.version?.trim() || undefined\n}\n\nfunction pathSafeProfileLabel(label: string | undefined): string | undefined {\n  const safe = label\n    ?.trim()\n    .replace(/[^A-Za-z0-9._-]+/g, '-')\n    .replace(/-+/g, '-')\n    .replace(/^-|-$/g, '')\n  return safe || undefined\n}\n\nfunction compact<T extends Record<string, unknown>>(input: T): Partial<T> {\n  const out: Record<string, unknown> = {}\n  for (const [key, value] of Object.entries(input)) {\n    if (value !== undefined) out[key] = value\n  }\n  return out as Partial<T>\n}\n\n/**\n * Deterministic behaviour identity for the canonical\n * `@tangle-network/agent-interface` AgentProfile.\n *\n * `name` and `description` are labels and do not affect the hash. Profile\n * `version`, prompt, model hints, tools, resources, hooks, modes, permissions,\n * and extensions do affect the hash. Resource array order is hash-bearing\n * because mount order can change agent behaviour. Undefined fields are treated\n * as absent; explicit `null` fields remain hash-bearing.\n */\nexport function agentProfileHash(profile: AgentProfile): string {\n  const model = agentProfileModelId(profile)\n  const behaviour = compact({\n    ...profile,\n    name: undefined,\n    description: undefined,\n    tags: profile.tags ? [...profile.tags].sort() : undefined,\n    model: compact({ ...profile.model, default: model }),\n  })\n  return createHash('sha256').update(canonicalString(behaviour)).digest('hex')\n}\n","/**\n * Completion verifier — the task-completion oracle.\n *\n * Answers the only eval question that is not a proxy: did the agent actually\n * COMPLETE the task — produce every required deliverable, persisted and\n * correct — rather than describe what should be done. A fluent transcript\n * that never produces the artifact scores zero here.\n *\n * Per requirement, a two-stage check:\n *   1. Structural — a produced item (vault artifact / approved proposal /\n *      tool call) of the right kind is matched against the requirement and\n *      carries non-empty content. Deterministic; no LLM.\n *   2. Correctness — only if structurally present AND the matched item\n *      carries content, one targeted check decides whether that item\n *      actually fulfils the requirement. A hallucinated artifact fails here;\n *      an absent one already failed stage 1.\n *\n * `completionRate` is satisfied / MEASURABLE requirements (unmeasured rows —\n * checker failures — are excluded from the denominator, never scored as\n * zeros). Quality dimensions are meaningless on an incomplete task — callers\n * gate on `fullyComplete` / `completionRate` before scoring quality.\n */\n\nimport { randomUUID } from 'node:crypto'\nimport type { ChatClient, ChatRequest } from './analyst/chat-client'\nimport type { Artifact } from './artifact-validator'\nimport { CostLedger, type CostLedgerHandle } from './cost-ledger'\nimport { assertServedModel, ModelSubstitutionError } from './integrity/served-model'\nimport { recoverTruncatedJson } from './json-recovery'\nimport { JudgeParseError } from './judges'\nimport {\n  costReceiptFromLlm,\n  costReceiptFromLlmError,\n  maximumChargeForLlmRequest,\n} from './llm-client'\nimport { packageVersion } from './package-version'\nimport type { RawProviderEvent, RawProviderSink } from './trace/raw-provider-sink'\nimport {\n  certificationEvidenceDigest,\n  type DefaultVerdict,\n  type VerdictCertification,\n} from './verdict'\nimport type { CheckerIdentity, VerificationStrategySource } from './verification-strategy'\n\n/** What kind of produced state can satisfy a requirement structurally. */\nexport type SatisfiedBy = 'artifact' | 'proposal' | 'tool-call' | 'any'\n\nexport interface CompletionRequirement {\n  /** Stable id from the task gold (e.g. a persona's `expected_requirements[].req_id`). */\n  reqId: string\n  /** Human-readable description of the required deliverable. */\n  title: string\n  /** Optional kind/category hint, matched against a produced item's kind. */\n  category?: string\n  /** What produced state satisfies this requirement. Defaults to 'any'. */\n  satisfiedBy?: SatisfiedBy\n}\n\nexport interface TaskGold {\n  taskId: string\n  requirements: CompletionRequirement[]\n}\n\nexport interface ProducedProposal {\n  id: string\n  title: string\n  status: 'pending' | 'approved' | 'rejected'\n  /** Optional persisted body — when present, enables a correctness check. */\n  content?: string\n}\n\n/** Everything observable about what a run actually produced. */\nexport interface ProducedState {\n  /** Persisted vault artifacts. Reuses the shared `Artifact` shape. */\n  artifacts: Artifact[]\n  /** Proposals / filings the agent created. */\n  proposals: ProducedProposal[]\n  /** Names of tools the agent invoked. */\n  toolCalls: string[]\n}\n\nexport interface RequirementCheck {\n  reqId: string\n  title: string\n  /** A produced item of the right kind matched the requirement, non-empty. */\n  structurallyPresent: boolean\n  /**\n   * Whether the matched item actually fulfils the requirement. `null` when\n   * not structurally present, when the matched item carries no content\n   * to assess, or when the correctness check itself failed (`unmeasured`).\n   */\n  correct: boolean | null\n  /** structurallyPresent && !unmeasured && correct !== false. */\n  satisfied: boolean\n  /**\n   * Set when the correctness check itself errored (LLM call failure or an\n   * unparseable response after retry). The requirement's fulfilment is\n   * UNKNOWN — `correct` stays null, `satisfied` is false, and\n   * `completionVerdict` excludes the row from `completionRate`'s\n   * denominator. Never folded into a zero: a synthetic zero is\n   * indistinguishable from a real failure (see `JudgeParseError`).\n   */\n  unmeasured?: true\n  /** Why the correctness check could not be measured (present iff `unmeasured`). */\n  unmeasuredReason?: string\n  /** Human-readable evidence for the verdict. */\n  evidence: string[]\n}\n\n/** Extends the substrate verdict spine: `valid` = `fullyComplete` and\n *  `score` = `completionRate` — derived in `completionVerdict()`, the one\n *  place those equalities hold by construction. */\nexport interface CompletionVerdict extends DefaultVerdict {\n  taskId: string\n  requirements: RequirementCheck[]\n  /** satisfied / MEASURABLE requirements (unmeasured rows leave the denominator). */\n  completionRate: number\n  /** Every measurable requirement satisfied (false when anything is unmeasured). */\n  fullyComplete: boolean\n  /** Requirements whose correctness check errored — reported, never scored as zero. */\n  unmeasuredCount: number\n}\n\n/**\n * Construct a `CompletionVerdict` from the per-requirement checks, deriving\n * `completionRate` / `fullyComplete` and the spine fields (`valid` =\n * `fullyComplete`, `score` = `completionRate`) in one place. Throws on zero\n * requirements — a verdict over nothing is a misconfiguration, mirroring\n * `verifyCompletion`'s gold-spec guard.\n */\nexport function completionVerdict(input: {\n  taskId: string\n  requirements: RequirementCheck[]\n  /** What certified the correctness stage, when anything did. Omitted =\n   *  an uncertified verdict — the honest default for a bare checker. */\n  certification?: VerdictCertification\n}): CompletionVerdict {\n  if (input.requirements.length === 0) {\n    throw new Error(\n      `completionVerdict: task '${input.taskId}' has no requirement checks — nothing to derive a verdict from`,\n    )\n  }\n  const measurable = input.requirements.filter((r) => !r.unmeasured)\n  const unmeasuredCount = input.requirements.length - measurable.length\n  if (measurable.length === 0) {\n    // Every check errored: this is an infrastructure failure, not a scored\n    // run. A 0-rate verdict here would be a fabricated measurement.\n    throw new Error(\n      `completionVerdict: task '${input.taskId}' has no measurable requirements — all ${input.requirements.length} correctness checks failed (${input.requirements[0]?.unmeasuredReason ?? 'unknown reason'})`,\n    )\n  }\n  const satisfiedCount = measurable.filter((r) => r.satisfied).length\n  const completionRate = satisfiedCount / measurable.length\n  // A run with unmeasured rows can still report a rate over what WAS\n  // measured, but must not claim full completion.\n  const fullyComplete = unmeasuredCount === 0 && satisfiedCount === measurable.length\n  return {\n    taskId: input.taskId,\n    requirements: input.requirements,\n    completionRate,\n    fullyComplete,\n    unmeasuredCount,\n    valid: fullyComplete,\n    score: completionRate,\n    ...(input.certification === undefined ? {} : { certification: input.certification }),\n  }\n}\n\n/**\n * What a correctness checker declares about itself so the completion\n * verdict can carry a certification: which strategy member it discharges,\n * its exact identity, and the steps its answers rest on unverified.\n */\nexport interface CorrectnessCheckerAttestation {\n  strategy: VerificationStrategySource\n  checker: CheckerIdentity\n  assumptions: string[]\n}\n\n/**\n * Decides whether a produced item's content actually fulfils a requirement.\n * Injected so the structural verifier stays pure and unit-testable; the\n * production implementation is `createLlmCorrectnessChecker`.\n *\n * `attestation` is optional metadata on the function value: a checker that\n * carries one yields certified completion verdicts; a bare function yields\n * the same verdict uncertified. A plain arrow function remains a valid\n * checker.\n */\nexport interface CorrectnessChecker {\n  (\n    requirement: CompletionRequirement,\n    content: string,\n  ): Promise<{ correct: boolean; reason: string }>\n  attestation?: CorrectnessCheckerAttestation\n}\n\nconst STOPWORDS = new Set([\n  'the',\n  'a',\n  'an',\n  'of',\n  'for',\n  'and',\n  'or',\n  'to',\n  'in',\n  'on',\n  'with',\n  'by',\n])\n\n// Deliverable-FORM vocabulary — words that name the SHAPE of an output, not its\n// domain content. A correct \"swap-comparison view persisted as a ui/** artifact\"\n// is an OpenUI JSON whose body says nothing about \"artifact\" / \"persisted\" /\n// \"view\"; the discriminative tokens are the domain nouns (swap, comparison).\n// Stripped from the REQUIREMENT side of structural recall so a deliverable is\n// matched on what it IS about, not on the boilerplate describing its form. The\n// correctness checker strips the same class via TITLE_STOPWORDS. Anti-game holds:\n// the distinctive domain tokens remain, so an off-topic item still fails.\nconst REQUIREMENT_FORM_STOPWORDS = new Set([\n  'generated',\n  'generate',\n  'view',\n  'render',\n  'rendered',\n  'persisted',\n  'persist',\n  'artifact',\n  'file',\n  'document',\n  'note',\n  'proposal',\n  'deliverable',\n  'output',\n  'created',\n  'create',\n  'produce',\n  'produced',\n  'flag',\n])\n\nconst MATCH_THRESHOLD = 0.5\nconst MIN_CONTENT_CHARS = 50\n\nfunction tokens(s: string, extraStop?: Set<string>): Set<string> {\n  return new Set(\n    s\n      .toLowerCase()\n      .split(/[^a-z0-9]+/)\n      .filter((t) => t.length > 1 && !STOPWORDS.has(t) && !extraStop?.has(t)),\n  )\n}\n\n/**\n * Recall of the requirement's tokens within a candidate's identifying text.\n * Recall, not Jaccard — a candidate's path/id legitimately carries extra\n * tokens the requirement does not name. The requirement side drops\n * deliverable-FORM vocabulary so recall keys on the distinctive domain tokens.\n */\nfunction tokenRecall(requirementText: string, candidateText: string): number {\n  const req = tokens(requirementText, REQUIREMENT_FORM_STOPWORDS)\n  if (req.size === 0) return 0\n  const cand = tokens(candidateText)\n  let hit = 0\n  for (const t of req) if (cand.has(t)) hit++\n  return hit / req.size\n}\n\ninterface Candidate {\n  reqIndex: number\n  /** Unique key for a produced item — each item satisfies at most one requirement. */\n  itemKey: string\n  score: number\n  evidence: string\n  /** Content to correctness-check, or null when the matched item has none. */\n  content: string | null\n}\n\nfunction artifactCandidates(\n  req: CompletionRequirement,\n  reqIndex: number,\n  artifacts: Artifact[],\n): Candidate[] {\n  const reqText = `${req.title} ${req.category ?? ''}`\n  const out: Candidate[] = []\n  artifacts.forEach((a, i) => {\n    if ((a.content ?? '').trim().length < MIN_CONTENT_CHARS) return\n    // Match against the artifact CONTENT too, not just its path + kind — a\n    // generated view / note whose path is generic still satisfies a requirement\n    // when its body covers it (e.g. an OpenUI comparison grounded in the on-file\n    // figures). Bounded slice keeps the recall text cheap; MATCH_THRESHOLD holds.\n    let score = tokenRecall(\n      reqText,\n      `${a.path ?? ''} ${a.kind} ${(a.content ?? '').slice(0, 4000)}`,\n    )\n    if (req.category && a.kind && req.category.toLowerCase() === a.kind.toLowerCase()) {\n      score = Math.max(score, 1)\n    }\n    if (score < MATCH_THRESHOLD) return\n    out.push({\n      reqIndex,\n      itemKey: `artifact:${i}`,\n      score,\n      evidence: `artifact '${a.path ?? a.kind}' matched (token recall ${score.toFixed(2)})`,\n      content: a.content ?? null,\n    })\n  })\n  return out\n}\n\nfunction proposalCandidates(\n  req: CompletionRequirement,\n  reqIndex: number,\n  proposals: ProducedProposal[],\n): Candidate[] {\n  const reqText = `${req.title} ${req.category ?? ''}`\n  const out: Candidate[] = []\n  for (const p of proposals) {\n    // Pending or rejected work is not a completed deliverable.\n    if (p.status !== 'approved') continue\n    // A proposal needs an assessable BODY to be a deliverable. A bare title is\n    // not completion: correctness cannot be judged on it, so a title-only match\n    // would auto-pass the oracle (structurallyPresent && correct===null →\n    // satisfied) with no verifiable content. Tool calls are the only\n    // legitimately content-less deliverable (`toolCallCandidates`).\n    const body = (p.content ?? '').trim()\n    if (body.length < MIN_CONTENT_CHARS) continue\n    // Match against the body as well as the (often short) title — a refusal /\n    // flag / analysis proposal whose title is a label still satisfies a\n    // descriptively-worded requirement when its content covers it. MATCH_THRESHOLD\n    // + the requirement's distinctive tokens keep an off-topic proposal out;\n    // correctness (a SEMANTIC checker, NOT this lexical pass) then judges\n    // polarity/fulfilment, so a negation that merely contains the tokens fails.\n    // Structural and correctness must use different evidence or the two-stage\n    // check collapses to one lexical gate.\n    const score = tokenRecall(reqText, `${p.title} ${body}`)\n    if (score < MATCH_THRESHOLD) continue\n    out.push({\n      reqIndex,\n      itemKey: `proposal:${p.id}`,\n      score,\n      evidence: `approved proposal '${p.title}' matched (token recall ${score.toFixed(2)})`,\n      content: body,\n    })\n  }\n  return out\n}\n\nfunction toolCallCandidates(\n  req: CompletionRequirement,\n  reqIndex: number,\n  toolCalls: string[],\n): Candidate[] {\n  const out: Candidate[] = []\n  toolCalls.forEach((name, i) => {\n    const score = tokenRecall(req.title, name)\n    if (score < MATCH_THRESHOLD) return\n    out.push({\n      reqIndex,\n      itemKey: `tool:${i}`,\n      score,\n      evidence: `tool call '${name}' matched (token recall ${score.toFixed(2)})`,\n      content: null,\n    })\n  })\n  return out\n}\n\n/**\n * Verify whether a run completed the task. `checkCorrectness` is injected —\n * `createLlmCorrectnessChecker` for production, a deterministic stub in tests.\n *\n * Throws on a gold spec with no requirements: an eval task that requires\n * nothing is a misconfiguration, not a vacuously-complete task.\n */\nexport async function verifyCompletion(\n  gold: TaskGold,\n  state: ProducedState,\n  checkCorrectness: CorrectnessChecker,\n): Promise<CompletionVerdict> {\n  if (gold.requirements.length === 0) {\n    throw new Error(\n      `verifyCompletion: task '${gold.taskId}' has no requirements — malformed gold spec`,\n    )\n  }\n\n  // Collect every above-threshold (requirement, produced-item) candidate, then\n  // assign greedily by descending score: each requirement and each produced\n  // item is used at most once. One deliverable fulfils one requirement.\n  const candidates: Candidate[] = []\n  gold.requirements.forEach((req, i) => {\n    const by = req.satisfiedBy ?? 'any'\n    if (by === 'artifact' || by === 'any') {\n      candidates.push(...artifactCandidates(req, i, state.artifacts))\n    }\n    if (by === 'proposal' || by === 'any') {\n      candidates.push(...proposalCandidates(req, i, state.proposals))\n    }\n    if (by === 'tool-call' || by === 'any') {\n      candidates.push(...toolCallCandidates(req, i, state.toolCalls))\n    }\n  })\n  candidates.sort((a, b) => b.score - a.score)\n\n  const assigned = new Map<number, Candidate>()\n  const itemTaken = new Set<string>()\n  for (const c of candidates) {\n    if (assigned.has(c.reqIndex) || itemTaken.has(c.itemKey)) continue\n    assigned.set(c.reqIndex, c)\n    itemTaken.add(c.itemKey)\n  }\n\n  const requirements: RequirementCheck[] = []\n  for (let i = 0; i < gold.requirements.length; i++) {\n    const req = gold.requirements[i]!\n    const match = assigned.get(i)\n    const evidence: string[] = []\n    let correct: boolean | null = null\n    let unmeasuredReason: string | undefined\n\n    if (match) {\n      evidence.push(match.evidence)\n      if (match.content !== null) {\n        try {\n          const r = await checkCorrectness(req, match.content)\n          correct = r.correct\n          evidence.push(`correctness: ${r.correct ? 'pass' : 'fail'} — ${r.reason}`)\n        } catch (err) {\n          // The CHECKER failed, not the requirement. Recording this as a\n          // zero would fabricate a model failure out of an infrastructure\n          // one; the requirement is unmeasured and leaves the denominator.\n          unmeasuredReason =\n            err instanceof JudgeParseError\n              ? `checker response unparseable after retry: ${err.raw.slice(0, 200)}`\n              : `checker call failed: ${err instanceof Error ? err.message : String(err)}`\n          evidence.push(`correctness: UNMEASURED — ${unmeasuredReason}`)\n        }\n      } else {\n        evidence.push('correctness: not assessed — matched item carries no content')\n      }\n    } else {\n      const by = req.satisfiedBy ?? 'any'\n      const kind = by === 'any' ? 'artifact/proposal/tool-call' : by\n      evidence.push(`no produced ${kind} matched this requirement`)\n    }\n\n    const structurallyPresent = match !== undefined\n    const unmeasured = unmeasuredReason !== undefined\n    const satisfied = structurallyPresent && !unmeasured && correct !== false\n    requirements.push({\n      reqId: req.reqId,\n      title: req.title,\n      structurallyPresent,\n      correct,\n      satisfied,\n      ...(unmeasured ? { unmeasured: true as const, unmeasuredReason } : {}),\n      evidence,\n    })\n  }\n\n  // The verdict blends the deterministic structural stage with the injected\n  // correctness stage, so the certification names the checker's own member\n  // and carries the structural stage's lexical nature as an assumption. A\n  // checker without an attestation yields an uncertified verdict.\n  const attestation = checkCorrectness.attestation\n  const certification: VerdictCertification | undefined = attestation\n    ? {\n        strategy: attestation.strategy,\n        checker: attestation.checker,\n        assumptions: [\n          'structural matching is lexical token recall over produced items — the correctness stage only sees items it matched',\n          ...attestation.assumptions,\n        ],\n        evidenceDigest: certificationEvidenceDigest({ taskId: gold.taskId, requirements }),\n      }\n    : undefined\n\n  return completionVerdict({\n    taskId: gold.taskId,\n    requirements,\n    ...(certification === undefined ? {} : { certification }),\n  })\n}\n\nexport interface LlmCorrectnessCheckerOpts {\n  model?: string\n  /** Optional ledger for direct use. */\n  costLedger?: CostLedgerHandle\n  costPhase?: string\n  costTags?: Record<string, string>\n  signal?: AbortSignal\n  /** Max chars of artifact content sent to the checker. */\n  maxContentChars?: number\n  /**\n   * Checker LLM calls per requirement before giving up (parse failures and\n   * call errors both consume attempts). The failure then surfaces as an\n   * `unmeasured` requirement, never a zero.\n   */\n  maxAttempts?: number\n  /**\n   * Forensic capture of every checker request/response/error — without it a\n   * checker failure is unauditable (the agent-turn raws never contain the\n   * checker's own calls). Same sink contract as `LlmClient`.\n   */\n  rawSink?: RawProviderSink\n}\n\n/**\n * Parse the correctness checker's model response. Tolerates a response\n * truncated mid-JSON (max_tokens cap) by auto-closing the prefix — the\n * verdict boolean usually lands in the first few tokens, so a recovered\n * prefix with a boolean `correct` is a real measurement, not a guess.\n * Fails loud (JudgeParseError) when no boolean verdict is recoverable.\n */\nexport function parseCorrectnessResponse(raw: string): { correct: boolean; reason: string } {\n  const readVerdict = (candidate: unknown): { correct: boolean; reason: string } | null => {\n    if (candidate === null || typeof candidate !== 'object') return null\n    const { correct, reason } = candidate as { correct?: unknown; reason?: unknown }\n    if (typeof correct !== 'boolean') return null\n    return { correct, reason: typeof reason === 'string' ? reason : '' }\n  }\n\n  const match = raw.match(/\\{[\\s\\S]*\\}/)\n  if (match) {\n    try {\n      const strict = readVerdict(JSON.parse(match[0]))\n      if (strict) return strict\n    } catch {\n      // fall through to truncation recovery\n    }\n  }\n  // The strict path needs a closing `}`; a cap-hit response has none. Take\n  // everything from the first `{` and auto-close it.\n  const start = raw.indexOf('{')\n  if (start !== -1) {\n    const recovered = readVerdict(recoverTruncatedJson(raw.slice(start)))\n    if (recovered) return recovered\n  }\n  throw new JudgeParseError('correctness-checker', raw)\n}\n\n/**\n * Production `CorrectnessChecker` — one LLM call per matched artifact,\n * deterministic (temperature 0), structured JSON out. Judges fulfilment\n * only: a plan, a gesture, or a description of what should be done does not\n * fulfil a requirement — the artifact must BE the deliverable.\n */\nexport function createLlmCorrectnessChecker(\n  chat: ChatClient,\n  opts: LlmCorrectnessCheckerOpts = {},\n): CorrectnessChecker {\n  const model = opts.model ?? 'claude-sonnet-4-6'\n  const maxContentChars = opts.maxContentChars ?? 8000\n  const maxAttempts = opts.maxAttempts ?? 2\n  const costLedger = opts.costLedger ?? new CostLedger()\n  const sink = opts.rawSink\n  const record = async (event: RawProviderEvent): Promise<void> => {\n    // Forensic capture is best-effort; the verdict is the system of record.\n    try {\n      await sink?.record(event)\n    } catch {\n      // Intentionally swallowed.\n    }\n  }\n  const checker: CorrectnessChecker = async (requirement, content) => {\n    const request = {\n      model,\n      messages: [\n        {\n          role: 'system' as const,\n          content:\n            'You verify whether a produced work artifact actually fulfils a stated requirement. Judge fulfilment only — is the deliverable substantively present and on-point — not polish. A plan to do it later, a vague gesture, or a description of what should be done does NOT fulfil a requirement; the artifact must BE the deliverable. Respond with a single JSON object: {\"correct\": boolean, \"reason\": string (<= 30 words)}.',\n        },\n        {\n          role: 'user' as const,\n          content: `Requirement: ${requirement.title}\\n${\n            requirement.category ? `Category: ${requirement.category}\\n` : ''\n          }\\nProduced artifact:\\n${content.slice(0, maxContentChars)}`,\n        },\n      ],\n      temperature: 0,\n      maxTokens: 200,\n    } satisfies ChatRequest\n    let lastErr: unknown\n    for (let attempt = 0; attempt < maxAttempts; attempt++) {\n      const started = Date.now()\n      await record({\n        eventId: randomUUID(),\n        provider: chat.transport,\n        model,\n        endpoint: '/chat',\n        baseUrl: '',\n        attemptIndex: attempt,\n        direction: 'request',\n        timestamp: started,\n        requestBody: request,\n        redactedFields: [],\n      })\n      try {\n        const paid = await costLedger.runPaidCall({\n          channel: 'verifier',\n          phase: opts.costPhase ?? 'completion.correctness',\n          actor: 'correctness-checker',\n          model,\n          maximumCharge:\n            chat.maximumAttempts === undefined\n              ? undefined\n              : maximumChargeForLlmRequest(request, { maximumAttempts: chat.maximumAttempts }),\n          tags: {\n            ...opts.costTags,\n            requirementId: requirement.reqId,\n            attempt: String(attempt),\n          },\n          signal: opts.signal,\n          execute: (signal, callId) => chat.chat(request, { signal, idempotencyKey: callId }),\n          receipt: costReceiptFromLlm,\n          receiptFromError: costReceiptFromLlmError,\n        })\n        if (!paid.succeeded) throw paid.error\n        const resp = paid.value\n        // Hold the transport to its own word: a verdict produced by a model\n        // other than the one requested is not this model's verdict. A\n        // transport that echoes no id cannot be made to prove identity here —\n        // callers needing that proof enable it at the client\n        // (`LlmClientOptions.assertServedModel`) or gate on `assertModelsServed`.\n        assertServedModel(model, resp.servedModel, {\n          allowUnreported: true,\n          context: `correctness checker for requirement ${requirement.reqId}`,\n        })\n        const raw = resp.content\n        await record({\n          eventId: randomUUID(),\n          provider: chat.transport,\n          model,\n          endpoint: '/chat',\n          baseUrl: '',\n          attemptIndex: attempt,\n          direction: 'response',\n          timestamp: Date.now(),\n          durationMs: Date.now() - started,\n          responseBody: resp,\n          redactedFields: [],\n        })\n        return parseCorrectnessResponse(raw)\n      } catch (err) {\n        lastErr = err\n        await record({\n          eventId: randomUUID(),\n          provider: chat.transport,\n          model,\n          endpoint: '/chat',\n          baseUrl: '',\n          attemptIndex: attempt,\n          direction: 'error',\n          timestamp: Date.now(),\n          durationMs: Date.now() - started,\n          errorMessage: err instanceof Error ? err.message : String(err),\n          redactedFields: [],\n        })\n        // Substitution is a verdict about identity, not a transient fault:\n        // another attempt reaches the same wrong model and spends more tokens.\n        if (err instanceof ModelSubstitutionError) throw err\n      }\n    }\n    throw lastErr instanceof Error ? lastErr : new Error(String(lastErr))\n  }\n  checker.attestation = {\n    strategy: 'judge',\n    checker: { name: 'llm-correctness-checker', version: model },\n    assumptions: [\n      'served-model identity is accepted when the transport reports none (assertServedModel allowUnreported)',\n    ],\n  }\n  return checker\n}\n\n/** Stopwords for requirement-title tokenization — drops the imperative verbs\n *  ('review', 'update', …) common to deliverable titles so recall keys on the\n *  substantive nouns, not the boilerplate ask. */\nconst TITLE_STOPWORDS = new Set([\n  'the',\n  'a',\n  'an',\n  'and',\n  'or',\n  'for',\n  'to',\n  'of',\n  'in',\n  'on',\n  'with',\n  'review',\n  'update',\n  'new',\n  'proposed',\n])\n\n/**\n * Deterministic `CorrectnessChecker` — the no-LLM counterpart to\n * `createLlmCorrectnessChecker`. A produced item fulfils a requirement when its\n * content is substantive (≥ `minContentLength` chars) AND recalls ≥ `minRecall`\n * of the requirement title's significant tokens. No network.\n *\n * Polarity-blind: token recall credits a negation that contains the\n * requirement's tokens (\"I will NOT produce the comparison\" recalls every token\n * of \"produce the comparison\"). The structural match stage is ALSO lexical, so\n * pairing the two collapses to a single gameable gate. Use this only as an\n * opt-in structural pre-filter or for tasks whose requirements have no polarity\n * to invert; for produced-state grading the correctness checker MUST be semantic\n * (`createLlmCorrectnessChecker`). See the anti-game fixtures in the test suite.\n */\nexport function createTokenRecallChecker(\n  opts: { minRecall?: number; minContentLength?: number } = {},\n): CorrectnessChecker {\n  const minRecall = opts.minRecall ?? 0.5\n  const minLen = opts.minContentLength ?? 120\n  const checker: CorrectnessChecker = async (requirement, content) => {\n    const body = content.trim()\n    if (body.length < minLen)\n      return {\n        correct: false,\n        reason: `content too thin (${body.length} chars) to be the deliverable`,\n      }\n    const titleTokens = requirement.title\n      .toLowerCase()\n      .split(/[^a-z0-9]+/)\n      .filter((t) => t.length > 2 && !TITLE_STOPWORDS.has(t))\n    if (titleTokens.length === 0)\n      return {\n        correct: true,\n        reason: 'requirement title has no significant tokens — structural match accepted',\n      }\n    const lower = body.toLowerCase()\n    const hits = titleTokens.filter((t) => lower.includes(t)).length\n    const recall = hits / titleTokens.length\n    return recall >= minRecall\n      ? {\n          correct: true,\n          reason: `content recalls ${hits}/${titleTokens.length} requirement tokens`,\n        }\n      : {\n          correct: false,\n          reason: `content recalls only ${hits}/${titleTokens.length} requirement tokens`,\n        }\n  }\n  // 'schema': token recall checks shape, not meaning — the member whose\n  // documented failure mode (\"a well-formed wrong answer passes\") is\n  // exactly this checker's polarity blindness.\n  checker.attestation = {\n    strategy: 'schema',\n    checker: { name: 'token-recall-checker', version: packageVersion() },\n    assumptions: ['polarity-blind: a negation that recalls the requirement tokens passes'],\n  }\n  return checker\n}\n","/**\n * Produced-state extraction — normalize a run's runtime event stream into the\n * typed `ProducedState` the completion oracle consumes.\n *\n * `ProducedState` answers \"what did the agent actually produce\" — vault\n * artifacts, proposals, tool calls. The runtime emits these as a stream of\n * events; this module is the single normalization point from that stream to\n * the shape `verifyCompletion` expects.\n *\n * Input is structurally typed (`RuntimeEventLike`) so this module does not\n * depend on agent-runtime — agent-runtime's `RuntimeStreamEvent` satisfies it\n * structurally. The `content` on `ArtifactEventLike` and the whole\n * `proposal_created` variant are the runtime-side enrichments this contract\n * requires; the runtime emits them, this module consumes them.\n */\n\nimport type { Artifact } from './artifact-validator'\nimport type { ProducedProposal, ProducedState } from './completion-verifier'\n\n/** A tool the agent invoked. */\nexport interface ToolCallEventLike {\n  type: 'tool_call'\n  toolName: string\n}\n\n/**\n * An artifact the agent produced. `content` is the enriched field — the\n * runtime's base `artifact` event carries only metadata; the completion\n * oracle needs the body to verify the deliverable, so the runtime emits it.\n */\nexport interface ArtifactEventLike {\n  type: 'artifact'\n  artifactId: string\n  name?: string\n  mimeType?: string\n  uri?: string\n  content?: string\n}\n\n/** A proposal / filing the agent created. */\nexport interface ProposalEventLike {\n  type: 'proposal_created'\n  proposalId: string\n  title: string\n  status?: 'pending' | 'approved' | 'rejected'\n  // body of the proposal (e.g. a submit_proposal `description`). When present,\n  // the completion oracle correctness-checks it like artifact content; absent,\n  // the proposal is graded presence-only.\n  content?: string\n}\n\n/**\n * The subset of runtime stream events `extractProducedState` consumes.\n * agent-runtime's full `RuntimeStreamEvent` union satisfies this structurally;\n * the `{ type: string }` catch-all keeps the input permissive so callers can\n * pass the whole unfiltered telemetry stream — unrecognized events are skipped.\n */\nexport type RuntimeEventLike =\n  | ToolCallEventLike\n  | ArtifactEventLike\n  | ProposalEventLike\n  | { type: string }\n\nfunction artifactKind(mimeType: string | undefined): string {\n  if (!mimeType) return 'file'\n  if (mimeType.includes('json')) return 'json'\n  if (mimeType.startsWith('text/')) return 'text'\n  return 'file'\n}\n\n/**\n * Normalize an emission-ordered stream into its latest observed produced state.\n * Artifacts are keyed by their exact output path (name, then URI, then id);\n * proposals by id. Later observations replace earlier ones, including missing\n * content or a rejected status. An obsolete version cannot prove completion.\n * Distinct paths remain distinct; no path normalization or aliasing is inferred.\n * Results retain first-seen identity order. Tool names describe invocation,\n * not success, and are deduplicated in first-seen order.\n *\n * An artifact without observed content yields empty content, which the\n * completion oracle rejects. This projection does not verify external effects,\n * recover dropped events, or establish freshness beyond the supplied stream.\n */\nexport function extractProducedState(events: readonly RuntimeEventLike[]): ProducedState {\n  const artifacts = new Map<string, Artifact>()\n  const proposals = new Map<string, ProducedProposal>()\n  const toolCalls: string[] = []\n  const seenTools = new Set<string>()\n\n  for (const ev of events) {\n    if (ev.type === 'tool_call') {\n      const name = (ev as ToolCallEventLike).toolName\n      if (name && !seenTools.has(name)) {\n        seenTools.add(name)\n        toolCalls.push(name)\n      }\n    } else if (ev.type === 'artifact') {\n      const a = ev as ArtifactEventLike\n      const path = a.name ?? a.uri ?? a.artifactId\n      artifacts.set(path, {\n        kind: artifactKind(a.mimeType),\n        path,\n        content: a.content ?? '',\n      })\n    } else if (ev.type === 'proposal_created') {\n      const p = ev as ProposalEventLike\n      proposals.set(p.proposalId, {\n        id: p.proposalId,\n        title: p.title,\n        status: p.status ?? 'pending',\n        ...(p.content !== undefined ? { content: p.content } : {}),\n      })\n    }\n  }\n\n  return { artifacts: [...artifacts.values()], proposals: [...proposals.values()], toolCalls }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;AAeA,MAAa,mBAA2C;CACtD;CACA;CACA;CACA;AACF;;;;;;AAwBA,MAAa,uBAAuB;;;;;;;;;;;;;;;;;;;AAoBpC,SAAgB,kBAAkB,MAAuC;CACvE,MAAM,YAAY,KAAK,aAAa;CACpC,IAAI,UAAU,WAAW,GAAG,MAAM,IAAI,gBAAgB,0CAA0C;CAChG,MAAM,YAAY,KAAK,KAAK,OAAO;CACnC,MAAM,SAAS,KAAK,WAAW,YAAY,CAAC,SAAS,IAAI,CAAC;CAC1D,IAAI,OAAO,WAAW,GACpB,MAAM,IAAI,gBACR,kGACF;CAEF,MAAM,MAAsB,CAAC;CAC7B,MAAM,uBAAO,IAAI,IAAY;CAC7B,KAAK,MAAM,WAAW,WAAW;EAS/B,MAAM,YAAY,KAAK,mBACnB,SACA,OAAO,QAAQ,UAAU,qBAAqB,SAAS,KAAK,CAAC;EACjE,MAAM,YAAY,UAAU,SAAS,IAAI,YAAY,CAAC,oBAAoB;EAC1E,KAAK,MAAM,SAAS,WAAW;GAC7B,MAAM,UAAwB;IAC5B,GAAG,KAAK;IACR;IACA,MAAM,GAAG,KAAK,KAAK,QAAQ,QAAQ,GAAG,QAAQ,GAAG;IACjD,OAAO;KAAE,GAAG,KAAK,KAAK;KAAO,SAAS;IAAM;IAC5C,UAAU;KAAE,GAAI,KAAK,KAAK,YAAY,CAAC;KAAI;KAAS,cAAc;IAAM;GAC1E;GACA,MAAM,KAAK,eAAe,OAAO;GACjC,IAAI,KAAK,IAAI,EAAE,GAAG;GAClB,KAAK,IAAI,EAAE;GACX,IAAI,KAAK,OAAO;EAClB;CACF;CACA,IAAI,IAAI,WAAW,GAGjB,MAAM,IAAI,gBACR,uDAAuD,UAAU,KAAK,IAAI,EAAE,aAAa,OAAO,KAAK,IAAI,EAAE,IAC7G;CAEF,OAAO;AACT;;;;;;;;AASA,SAAgB,cACd,SACqD;CACrD,MAAM,IAAI,QAAQ;CAClB,MAAM,UAAU,GAAG;CACnB,MAAM,QAAQ,GAAG;CACjB,IAAI,OAAO,YAAY,YAAY,OAAO,UAAU,UAClD,OAAO;EAAW;EAAwB;CAAM;AAGpD;;;;;;;;AASA,SAAgB,eAAe,SAA+B;CAE5D,OAAO,GADO,qBAAqB,yBAAyB,OAAO,CAAC,KAAK,UACzD,GAAG,iBAAiB,OAAO,CAAC,CAAC,MAAM,GAAG,EAAE;AAC1D;;;;;AAMA,SAAgB,oBAAoB,SAA+B;CACjE,MAAM,QAAQ,QAAQ,OAAO,SAAS,KAAK;CAC3C,IAAI,CAAC,OAEH,MAAM,IAAI,gBACR,iBAFY,yBAAyB,OAAO,KAAK,kBAE1B,gDACzB;CAEF,OAAO;AACT;AAEA,SAAS,yBAAyB,SAA2C;CAC3E,OAAO,QAAQ,MAAM,KAAK,KAAK,QAAQ,SAAS,KAAK,KAAK,KAAA;AAC5D;AAEA,SAAS,qBAAqB,OAA+C;CAM3E,OALa,OACT,KAAK,CAAC,CACP,QAAQ,qBAAqB,GAAG,CAAC,CACjC,QAAQ,OAAO,GAAG,CAAC,CACnB,QAAQ,UAAU,EAAE,KACR,KAAA;AACjB;AAEA,SAAS,QAA2C,OAAsB;CACxE,MAAM,MAA+B,CAAC;CACtC,KAAK,MAAM,CAAC,KAAK,UAAU,OAAO,QAAQ,KAAK,GAC7C,IAAI,UAAU,KAAA,GAAW,IAAI,OAAO;CAEtC,OAAO;AACT;;;;;;;;;;;AAYA,SAAgB,iBAAiB,SAA+B;CAC9D,MAAM,QAAQ,oBAAoB,OAAO;CACzC,MAAM,YAAY,QAAQ;EACxB,GAAG;EACH,MAAM,KAAA;EACN,aAAa,KAAA;EACb,MAAM,QAAQ,OAAO,CAAC,GAAG,QAAQ,IAAI,CAAC,CAAC,KAAK,IAAI,KAAA;EAChD,OAAO,QAAQ;GAAE,GAAG,QAAQ;GAAO,SAAS;EAAM,CAAC;CACrD,CAAC;CACD,OAAO,WAAW,QAAQ,CAAC,CAAC,OAAO,gBAAgB,SAAS,CAAC,CAAC,CAAC,OAAO,KAAK;AAC7E;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;ACtEA,SAAgB,kBAAkB,OAMZ;CACpB,IAAI,MAAM,aAAa,WAAW,GAChC,MAAM,IAAI,MACR,4BAA4B,MAAM,OAAO,+DAC3C;CAEF,MAAM,aAAa,MAAM,aAAa,QAAQ,MAAM,CAAC,EAAE,UAAU;CACjE,MAAM,kBAAkB,MAAM,aAAa,SAAS,WAAW;CAC/D,IAAI,WAAW,WAAW,GAGxB,MAAM,IAAI,MACR,4BAA4B,MAAM,OAAO,yCAAyC,MAAM,aAAa,OAAO,8BAA8B,MAAM,aAAa,EAAE,EAAE,oBAAoB,iBAAiB,EACxM;CAEF,MAAM,iBAAiB,WAAW,QAAQ,MAAM,EAAE,SAAS,CAAC,CAAC;CAC7D,MAAM,iBAAiB,iBAAiB,WAAW;CAGnD,MAAM,gBAAgB,oBAAoB,KAAK,mBAAmB,WAAW;CAC7E,OAAO;EACL,QAAQ,MAAM;EACd,cAAc,MAAM;EACpB;EACA;EACA;EACA,OAAO;EACP,OAAO;EACP,GAAI,MAAM,kBAAkB,KAAA,IAAY,CAAC,IAAI,EAAE,eAAe,MAAM,cAAc;CACpF;AACF;AA+BA,MAAM,4BAAY,IAAI,IAAI;CACxB;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF,CAAC;AAUD,MAAM,6CAA6B,IAAI,IAAI;CACzC;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF,CAAC;AAED,MAAM,kBAAkB;AACxB,MAAM,oBAAoB;AAE1B,SAAS,OAAO,GAAW,WAAsC;CAC/D,OAAO,IAAI,IACT,EACG,YAAY,CAAC,CACb,MAAM,YAAY,CAAC,CACnB,QAAQ,MAAM,EAAE,SAAS,KAAK,CAAC,UAAU,IAAI,CAAC,KAAK,CAAC,WAAW,IAAI,CAAC,CAAC,CAC1E;AACF;;;;;;;AAQA,SAAS,YAAY,iBAAyB,eAA+B;CAC3E,MAAM,MAAM,OAAO,iBAAiB,0BAA0B;CAC9D,IAAI,IAAI,SAAS,GAAG,OAAO;CAC3B,MAAM,OAAO,OAAO,aAAa;CACjC,IAAI,MAAM;CACV,KAAK,MAAM,KAAK,KAAK,IAAI,KAAK,IAAI,CAAC,GAAG;CACtC,OAAO,MAAM,IAAI;AACnB;AAYA,SAAS,mBACP,KACA,UACA,WACa;CACb,MAAM,UAAU,GAAG,IAAI,MAAM,GAAG,IAAI,YAAY;CAChD,MAAM,MAAmB,CAAC;CAC1B,UAAU,SAAS,GAAG,MAAM;EAC1B,KAAK,EAAE,WAAW,GAAA,CAAI,KAAK,CAAC,CAAC,SAAS,mBAAmB;EAKzD,IAAI,QAAQ,YACV,SACA,GAAG,EAAE,QAAQ,GAAG,GAAG,EAAE,KAAK,IAAI,EAAE,WAAW,GAAA,CAAI,MAAM,GAAG,GAAI,GAC9D;EACA,IAAI,IAAI,YAAY,EAAE,QAAQ,IAAI,SAAS,YAAY,MAAM,EAAE,KAAK,YAAY,GAC9E,QAAQ,KAAK,IAAI,OAAO,CAAC;EAE3B,IAAI,QAAQ,iBAAiB;EAC7B,IAAI,KAAK;GACP;GACA,SAAS,YAAY;GACrB;GACA,UAAU,aAAa,EAAE,QAAQ,EAAE,KAAK,0BAA0B,MAAM,QAAQ,CAAC,EAAE;GACnF,SAAS,EAAE,WAAW;EACxB,CAAC;CACH,CAAC;CACD,OAAO;AACT;AAEA,SAAS,mBACP,KACA,UACA,WACa;CACb,MAAM,UAAU,GAAG,IAAI,MAAM,GAAG,IAAI,YAAY;CAChD,MAAM,MAAmB,CAAC;CAC1B,KAAK,MAAM,KAAK,WAAW;EAEzB,IAAI,EAAE,WAAW,YAAY;EAM7B,MAAM,QAAQ,EAAE,WAAW,GAAA,CAAI,KAAK;EACpC,IAAI,KAAK,SAAS,mBAAmB;EASrC,MAAM,QAAQ,YAAY,SAAS,GAAG,EAAE,MAAM,GAAG,MAAM;EACvD,IAAI,QAAQ,iBAAiB;EAC7B,IAAI,KAAK;GACP;GACA,SAAS,YAAY,EAAE;GACvB;GACA,UAAU,sBAAsB,EAAE,MAAM,0BAA0B,MAAM,QAAQ,CAAC,EAAE;GACnF,SAAS;EACX,CAAC;CACH;CACA,OAAO;AACT;AAEA,SAAS,mBACP,KACA,UACA,WACa;CACb,MAAM,MAAmB,CAAC;CAC1B,UAAU,SAAS,MAAM,MAAM;EAC7B,MAAM,QAAQ,YAAY,IAAI,OAAO,IAAI;EACzC,IAAI,QAAQ,iBAAiB;EAC7B,IAAI,KAAK;GACP;GACA,SAAS,QAAQ;GACjB;GACA,UAAU,cAAc,KAAK,0BAA0B,MAAM,QAAQ,CAAC,EAAE;GACxE,SAAS;EACX,CAAC;CACH,CAAC;CACD,OAAO;AACT;;;;;;;;AASA,eAAsB,iBACpB,MACA,OACA,kBAC4B;CAC5B,IAAI,KAAK,aAAa,WAAW,GAC/B,MAAM,IAAI,MACR,2BAA2B,KAAK,OAAO,4CACzC;CAMF,MAAM,aAA0B,CAAC;CACjC,KAAK,aAAa,SAAS,KAAK,MAAM;EACpC,MAAM,KAAK,IAAI,eAAe;EAC9B,IAAI,OAAO,cAAc,OAAO,OAC9B,WAAW,KAAK,GAAG,mBAAmB,KAAK,GAAG,MAAM,SAAS,CAAC;EAEhE,IAAI,OAAO,cAAc,OAAO,OAC9B,WAAW,KAAK,GAAG,mBAAmB,KAAK,GAAG,MAAM,SAAS,CAAC;EAEhE,IAAI,OAAO,eAAe,OAAO,OAC/B,WAAW,KAAK,GAAG,mBAAmB,KAAK,GAAG,MAAM,SAAS,CAAC;CAElE,CAAC;CACD,WAAW,MAAM,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK;CAE3C,MAAM,2BAAW,IAAI,IAAuB;CAC5C,MAAM,4BAAY,IAAI,IAAY;CAClC,KAAK,MAAM,KAAK,YAAY;EAC1B,IAAI,SAAS,IAAI,EAAE,QAAQ,KAAK,UAAU,IAAI,EAAE,OAAO,GAAG;EAC1D,SAAS,IAAI,EAAE,UAAU,CAAC;EAC1B,UAAU,IAAI,EAAE,OAAO;CACzB;CAEA,MAAM,eAAmC,CAAC;CAC1C,KAAK,IAAI,IAAI,GAAG,IAAI,KAAK,aAAa,QAAQ,KAAK;EACjD,MAAM,MAAM,KAAK,aAAa;EAC9B,MAAM,QAAQ,SAAS,IAAI,CAAC;EAC5B,MAAM,WAAqB,CAAC;EAC5B,IAAI,UAA0B;EAC9B,IAAI;EAEJ,IAAI,OAAO;GACT,SAAS,KAAK,MAAM,QAAQ;GAC5B,IAAI,MAAM,YAAY,MACpB,IAAI;IACF,MAAM,IAAI,MAAM,iBAAiB,KAAK,MAAM,OAAO;IACnD,UAAU,EAAE;IACZ,SAAS,KAAK,gBAAgB,EAAE,UAAU,SAAS,OAAO,KAAK,EAAE,QAAQ;GAC3E,SAAS,KAAK;IAIZ,mBACE,eAAe,kBACX,6CAA6C,IAAI,IAAI,MAAM,GAAG,GAAG,MACjE,wBAAwB,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;IAC7E,SAAS,KAAK,6BAA6B,kBAAkB;GAC/D;QAEA,SAAS,KAAK,6DAA6D;EAE/E,OAAO;GACL,MAAM,KAAK,IAAI,eAAe;GAC9B,MAAM,OAAO,OAAO,QAAQ,gCAAgC;GAC5D,SAAS,KAAK,eAAe,KAAK,0BAA0B;EAC9D;EAEA,MAAM,sBAAsB,UAAU,KAAA;EACtC,MAAM,aAAa,qBAAqB,KAAA;EACxC,MAAM,YAAY,uBAAuB,CAAC,cAAc,YAAY;EACpE,aAAa,KAAK;GAChB,OAAO,IAAI;GACX,OAAO,IAAI;GACX;GACA;GACA;GACA,GAAI,aAAa;IAAE,YAAY;IAAe;GAAiB,IAAI,CAAC;GACpE;EACF,CAAC;CACH;CAMA,MAAM,cAAc,iBAAiB;CACrC,MAAM,gBAAkD,cACpD;EACE,UAAU,YAAY;EACtB,SAAS,YAAY;EACrB,aAAa,CACX,sHACA,GAAG,YAAY,WACjB;EACA,gBAAgB,4BAA4B;GAAE,QAAQ,KAAK;GAAQ;EAAa,CAAC;CACnF,IACA,KAAA;CAEJ,OAAO,kBAAkB;EACvB,QAAQ,KAAK;EACb;EACA,GAAI,kBAAkB,KAAA,IAAY,CAAC,IAAI,EAAE,cAAc;CACzD,CAAC;AACH;;;;;;;;AAgCA,SAAgB,yBAAyB,KAAmD;CAC1F,MAAM,eAAe,cAAoE;EACvF,IAAI,cAAc,QAAQ,OAAO,cAAc,UAAU,OAAO;EAChE,MAAM,EAAE,SAAS,WAAW;EAC5B,IAAI,OAAO,YAAY,WAAW,OAAO;EACzC,OAAO;GAAE;GAAS,QAAQ,OAAO,WAAW,WAAW,SAAS;EAAG;CACrE;CAEA,MAAM,QAAQ,IAAI,MAAM,aAAa;CACrC,IAAI,OACF,IAAI;EACF,MAAM,SAAS,YAAY,KAAK,MAAM,MAAM,EAAE,CAAC;EAC/C,IAAI,QAAQ,OAAO;CACrB,QAAQ,CAER;CAIF,MAAM,QAAQ,IAAI,QAAQ,GAAG;CAC7B,IAAI,UAAU,IAAI;EAChB,MAAM,YAAY,YAAY,qBAAqB,IAAI,MAAM,KAAK,CAAC,CAAC;EACpE,IAAI,WAAW,OAAO;CACxB;CACA,MAAM,IAAI,gBAAgB,uBAAuB,GAAG;AACtD;;;;;;;AAQA,SAAgB,4BACd,MACA,OAAkC,CAAC,GACf;CACpB,MAAM,QAAQ,KAAK,SAAS;CAC5B,MAAM,kBAAkB,KAAK,mBAAmB;CAChD,MAAM,cAAc,KAAK,eAAe;CACxC,MAAM,aAAa,KAAK,cAAc,IAAI,WAAW;CACrD,MAAM,OAAO,KAAK;CAClB,MAAM,SAAS,OAAO,UAA2C;EAE/D,IAAI;GACF,MAAM,MAAM,OAAO,KAAK;EAC1B,QAAQ,CAER;CACF;CACA,MAAM,UAA8B,OAAO,aAAa,YAAY;EAClE,MAAM,UAAU;GACd;GACA,UAAU,CACR;IACE,MAAM;IACN,SACE;GACJ,GACA;IACE,MAAM;IACN,SAAS,gBAAgB,YAAY,MAAM,IACzC,YAAY,WAAW,aAAa,YAAY,SAAS,MAAM,GAChE,wBAAwB,QAAQ,MAAM,GAAG,eAAe;GAC3D,CACF;GACA,aAAa;GACb,WAAW;EACb;EACA,IAAI;EACJ,KAAK,IAAI,UAAU,GAAG,UAAU,aAAa,WAAW;GACtD,MAAM,UAAU,KAAK,IAAI;GACzB,MAAM,OAAO;IACX,SAAS,WAAW;IACpB,UAAU,KAAK;IACf;IACA,UAAU;IACV,SAAS;IACT,cAAc;IACd,WAAW;IACX,WAAW;IACX,aAAa;IACb,gBAAgB,CAAC;GACnB,CAAC;GACD,IAAI;IACF,MAAM,OAAO,MAAM,WAAW,YAAY;KACxC,SAAS;KACT,OAAO,KAAK,aAAa;KACzB,OAAO;KACP;KACA,eACE,KAAK,oBAAoB,KAAA,IACrB,KAAA,IACA,2BAA2B,SAAS,EAAE,iBAAiB,KAAK,gBAAgB,CAAC;KACnF,MAAM;MACJ,GAAG,KAAK;MACR,eAAe,YAAY;MAC3B,SAAS,OAAO,OAAO;KACzB;KACA,QAAQ,KAAK;KACb,UAAU,QAAQ,WAAW,KAAK,KAAK,SAAS;MAAE;MAAQ,gBAAgB;KAAO,CAAC;KAClF,SAAS;KACT,kBAAkB;IACpB,CAAC;IACD,IAAI,CAAC,KAAK,WAAW,MAAM,KAAK;IAChC,MAAM,OAAO,KAAK;IAMlB,kBAAkB,OAAO,KAAK,aAAa;KACzC,iBAAiB;KACjB,SAAS,uCAAuC,YAAY;IAC9D,CAAC;IACD,MAAM,MAAM,KAAK;IACjB,MAAM,OAAO;KACX,SAAS,WAAW;KACpB,UAAU,KAAK;KACf;KACA,UAAU;KACV,SAAS;KACT,cAAc;KACd,WAAW;KACX,WAAW,KAAK,IAAI;KACpB,YAAY,KAAK,IAAI,IAAI;KACzB,cAAc;KACd,gBAAgB,CAAC;IACnB,CAAC;IACD,OAAO,yBAAyB,GAAG;GACrC,SAAS,KAAK;IACZ,UAAU;IACV,MAAM,OAAO;KACX,SAAS,WAAW;KACpB,UAAU,KAAK;KACf;KACA,UAAU;KACV,SAAS;KACT,cAAc;KACd,WAAW;KACX,WAAW,KAAK,IAAI;KACpB,YAAY,KAAK,IAAI,IAAI;KACzB,cAAc,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;KAC7D,gBAAgB,CAAC;IACnB,CAAC;IAGD,IAAI,eAAe,wBAAwB,MAAM;GACnD;EACF;EACA,MAAM,mBAAmB,QAAQ,UAAU,IAAI,MAAM,OAAO,OAAO,CAAC;CACtE;CACA,QAAQ,cAAc;EACpB,UAAU;EACV,SAAS;GAAE,MAAM;GAA2B,SAAS;EAAM;EAC3D,aAAa,CACX,uGACF;CACF;CACA,OAAO;AACT;;;ACpmBA,SAAS,aAAa,UAAsC;CAC1D,IAAI,CAAC,UAAU,OAAO;CACtB,IAAI,SAAS,SAAS,MAAM,GAAG,OAAO;CACtC,IAAI,SAAS,WAAW,OAAO,GAAG,OAAO;CACzC,OAAO;AACT;;;;;;;;;;;;;;AAeA,SAAgB,qBAAqB,QAAoD;CACvF,MAAM,4BAAY,IAAI,IAAsB;CAC5C,MAAM,4BAAY,IAAI,IAA8B;CACpD,MAAM,YAAsB,CAAC;CAC7B,MAAM,4BAAY,IAAI,IAAY;CAElC,KAAK,MAAM,MAAM,QACf,IAAI,GAAG,SAAS,aAAa;EAC3B,MAAM,OAAQ,GAAyB;EACvC,IAAI,QAAQ,CAAC,UAAU,IAAI,IAAI,GAAG;GAChC,UAAU,IAAI,IAAI;GAClB,UAAU,KAAK,IAAI;EACrB;CACF,OAAO,IAAI,GAAG,SAAS,YAAY;EACjC,MAAM,IAAI;EACV,MAAM,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE;EAClC,UAAU,IAAI,MAAM;GAClB,MAAM,aAAa,EAAE,QAAQ;GAC7B;GACA,SAAS,EAAE,WAAW;EACxB,CAAC;CACH,OAAO,IAAI,GAAG,SAAS,oBAAoB;EACzC,MAAM,IAAI;EACV,UAAU,IAAI,EAAE,YAAY;GAC1B,IAAI,EAAE;GACN,OAAO,EAAE;GACT,QAAQ,EAAE,UAAU;GACpB,GAAI,EAAE,YAAY,KAAA,IAAY,EAAE,SAAS,EAAE,QAAQ,IAAI,CAAC;EAC1D,CAAC;CACH;CAGF,OAAO;EAAE,WAAW,CAAC,GAAG,UAAU,OAAO,CAAC;EAAG,WAAW,CAAC,GAAG,UAAU,OAAO,CAAC;EAAG;CAAU;AAC7F"}
+{"version":3,"file":"produced-state-BXjXJHJT.js","names":[],"sources":["../src/agent-profile.ts","../src/completion-verifier.ts","../src/produced-state.ts"],"sourcesContent":["import { createHash } from 'node:crypto'\nimport type { AgentProfile } from '@tangle-network/agent-interface'\nimport { type HarnessType, harnessSupportsModel } from '@tangle-network/agent-interface'\nimport { ValidationError } from './errors'\nimport { canonicalString } from './ledger-core/canonical'\n\nexport type { AgentProfile, HarnessType } from '@tangle-network/agent-interface'\n\n/**\n * The agentic coding harnesses an eval sweeps by default — the ones we care about\n * ranking. This is the SINGLE source of that list; consumers import it instead of\n * re-declaring their own (a re-declared list is how the fleet drifts). Pass an\n * explicit `harnesses` (e.g. `harnessTypeSchema.options` for literally every known\n * harness) to widen beyond these.\n */\nexport const CODING_HARNESSES: readonly HarnessType[] = [\n  'opencode',\n  'claude-code',\n  'codex',\n  'kimi-code',\n]\n\nexport interface ProfileAxisSpec {\n  /** The domain profile to sweep. Its prompt/tools/skills are held fixed; only the\n   *  harness and model vary. `model.default` is the fallback model. */\n  base: AgentProfile\n  /** Harnesses to cross. Default: {@link CODING_HARNESSES}. */\n  harnesses?: readonly HarnessType[]\n  /** Models to cross. Default: `[base.model.default]` — one model, i.e. today's\n   *  single-model behaviour, so omitting this never changes an existing run. */\n  models?: readonly string[]\n  /** Force every (harness, model) pair verbatim, even ones the harness can't run —\n   *  for deliberately testing failure modes. Default (false): SNAP instead — a\n   *  vendor-locked harness runs only the swept models in its family, or its native\n   *  default when it supports none, so no harness is dropped and none gets a\n   *  guaranteed-failing foreign-model cell. */\n  keepIncompatible?: boolean\n}\n\n/** Model sentinel for a vendor-locked harness that supports none of the swept models:\n *  it carries no provider prefix, so `harnessSupportsModel` accepts it and the harness\n *  resolves it to its own native default model at runtime (e.g. kimi-code → its Kimi\n *  model). Lets `expandProfileAxes` snap-instead-of-drop without a per-harness flagship\n *  table that would rot as router catalogs change. */\nexport const HARNESS_NATIVE_MODEL = 'default'\n\n/**\n * Expand a base profile across the harness × model matrix into the `AgentProfile[]`\n * that `runProfileMatrix` / `selfImprove` score — the ONE place \"which harnesses ×\n * which models do we evaluate\" lives, so no product hand-rolls its own harness list\n * or column→profile mapping (the pattern that let those copies drift and silently\n * break the harness pivot).\n *\n * Each cell clones `base`, sets the canonical top-level `harness` and `model.default`,\n * and stamps `metadata.harness` + `metadata.harnessModel` for matrix grouping. Both\n * metadata fields are hash-bearing, so every cell gets a distinct `agentProfileId` row\n * and results join back by harness/model via {@link harnessAxisOf} with no\n * hand-recomputed key. A vendor-locked harness snaps to its family's swept models — or\n * its native default ({@link HARNESS_NATIVE_MODEL}) when it supports none — so every\n * requested harness runs; `keepIncompatible` forces every pair verbatim.\n *\n * Omit `harnesses`/`models` to sweep the full default set — the \"turn it on for\n * everything we care about\" switch, identical in shape whether one harness or all.\n */\nexport function expandProfileAxes(spec: ProfileAxisSpec): AgentProfile[] {\n  const harnesses = spec.harnesses ?? CODING_HARNESSES\n  if (harnesses.length === 0) throw new ValidationError('expandProfileAxes: no harnesses to sweep')\n  const baseModel = spec.base.model?.default\n  const models = spec.models ?? (baseModel ? [baseModel] : [])\n  if (models.length === 0) {\n    throw new ValidationError(\n      'expandProfileAxes: no models to sweep — base profile has no model.default and none were supplied',\n    )\n  }\n  const out: AgentProfile[] = []\n  const seen = new Set<string>()\n  for (const harness of harnesses) {\n    // A universal (router-backed) harness — opencode/pi/claudish — runs every swept\n    // model. A vendor-locked harness — codex/claude-code/kimi-code — runs only the\n    // swept models in its own family; when it supports NONE of them it snaps to its\n    // native default (the `HARNESS_NATIVE_MODEL` sentinel it resolves at runtime)\n    // rather than being dropped, so every requested harness still appears in the\n    // sweep on a model it can actually run — e.g. sweeping `deepseek/x` puts opencode\n    // on deepseek and kimi-code on its own Kimi model, a real head-to-head.\n    // `keepIncompatible` forces every (harness, model) pair verbatim (failure-mode runs).\n    const supported = spec.keepIncompatible\n      ? models\n      : models.filter((model) => harnessSupportsModel(harness, model))\n    const effective = supported.length > 0 ? supported : [HARNESS_NATIVE_MODEL]\n    for (const model of effective) {\n      const profile: AgentProfile = {\n        ...spec.base,\n        harness,\n        name: `${spec.base.name ?? 'agent'}/${harness}/${model}`,\n        model: { ...spec.base.model, default: model },\n        metadata: { ...(spec.base.metadata ?? {}), harness, harnessModel: model },\n      }\n      const id = agentProfileId(profile)\n      if (seen.has(id)) continue\n      seen.add(id)\n      out.push(profile)\n    }\n  }\n  if (out.length === 0) {\n    // Unreachable in normal use — snapping guarantees ≥1 cell per harness — but keep a\n    // fail-closed guard so a future refactor can't silently produce an empty sweep.\n    throw new ValidationError(\n      `expandProfileAxes: produced no profiles (harnesses=[${harnesses.join(', ')}], models=[${models.join(', ')}]).`,\n    )\n  }\n  return out\n}\n\n/**\n * Read the (harness, model) a matrix cell ran under, off a profile or a result row's\n * profile — the join-back for a `byHarness` pivot. Returns undefined when the profile\n * wasn't produced by {@link expandProfileAxes}. Callers group `result.byProfile` by\n * this instead of recomputing an id (recomputing the wrong key is what broke the pivot\n * in the hand-rolled copies).\n */\nexport function harnessAxisOf(\n  profile: Pick<AgentProfile, 'metadata'>,\n): { harness: HarnessType; model: string } | undefined {\n  const m = profile.metadata as Record<string, unknown> | undefined\n  const harness = m?.harness\n  const model = m?.harnessModel\n  if (typeof harness === 'string' && typeof model === 'string') {\n    return { harness: harness as HarnessType, model }\n  }\n  return undefined\n}\n\n/**\n * Collision-resistant, path-safe, human-readable profile id for eval artifacts.\n * Scorecard joins still use `agentProfileHash`; this id is for run ids, matrix\n * keys, and directory names where two profiles must not collapse onto one row.\n * The suffix is the first 64 bits of the behaviour hash, enough for ordinary\n * eval matrices while keeping filenames readable.\n */\nexport function agentProfileId(profile: AgentProfile): string {\n  const label = pathSafeProfileLabel(agentProfileDisplayLabel(profile)) ?? 'profile'\n  return `${label}-${agentProfileHash(profile).slice(0, 16)}`\n}\n\n/**\n * Model snapshot used for `RunRecord.model`. Eval surfaces require a concrete\n * model id because run records reject bare/missing model aliases.\n */\nexport function agentProfileModelId(profile: AgentProfile): string {\n  const model = profile.model?.default?.trim()\n  if (!model) {\n    const label = agentProfileDisplayLabel(profile) ?? 'unnamed profile'\n    throw new ValidationError(\n      `AgentProfile \"${label}\" has no model.default — cannot record eval run`,\n    )\n  }\n  return model\n}\n\nfunction agentProfileDisplayLabel(profile: AgentProfile): string | undefined {\n  return profile.name?.trim() || profile.version?.trim() || undefined\n}\n\nfunction pathSafeProfileLabel(label: string | undefined): string | undefined {\n  const safe = label\n    ?.trim()\n    .replace(/[^A-Za-z0-9._-]+/g, '-')\n    .replace(/-+/g, '-')\n    .replace(/^-|-$/g, '')\n  return safe || undefined\n}\n\nfunction compact<T extends Record<string, unknown>>(input: T): Partial<T> {\n  const out: Record<string, unknown> = {}\n  for (const [key, value] of Object.entries(input)) {\n    if (value !== undefined) out[key] = value\n  }\n  return out as Partial<T>\n}\n\n/**\n * Deterministic behaviour identity for the canonical\n * `@tangle-network/agent-interface` AgentProfile.\n *\n * `name` and `description` are labels and do not affect the hash. Profile\n * `version`, prompt, model hints, tools, resources, hooks, modes, permissions,\n * and extensions do affect the hash. Resource array order is hash-bearing\n * because mount order can change agent behaviour. Undefined fields are treated\n * as absent; explicit `null` fields remain hash-bearing.\n */\nexport function agentProfileHash(profile: AgentProfile): string {\n  const model = agentProfileModelId(profile)\n  const behaviour = compact({\n    ...profile,\n    name: undefined,\n    description: undefined,\n    tags: profile.tags ? [...profile.tags].sort() : undefined,\n    model: compact({ ...profile.model, default: model }),\n  })\n  return createHash('sha256').update(canonicalString(behaviour)).digest('hex')\n}\n","/**\n * Completion verifier — the task-completion oracle.\n *\n * Answers the only eval question that is not a proxy: did the agent actually\n * COMPLETE the task — produce every required deliverable, persisted and\n * correct — rather than describe what should be done. A fluent transcript\n * that never produces the artifact scores zero here.\n *\n * Per requirement, a two-stage check:\n *   1. Structural — a produced item (vault artifact / approved proposal /\n *      tool call) of the right kind is matched against the requirement and\n *      carries non-empty content. Deterministic; no LLM.\n *   2. Correctness — only if structurally present AND the matched item\n *      carries content, one targeted check decides whether that item\n *      actually fulfils the requirement. A hallucinated artifact fails here;\n *      an absent one already failed stage 1.\n *\n * `completionRate` is satisfied / MEASURABLE requirements (unmeasured rows —\n * checker failures — are excluded from the denominator, never scored as\n * zeros). Quality dimensions are meaningless on an incomplete task — callers\n * gate on `fullyComplete` / `completionRate` before scoring quality.\n */\n\nimport { randomUUID } from 'node:crypto'\nimport type { ChatClient, ChatRequest } from './analyst/chat-client'\nimport type { Artifact } from './artifact-validator'\nimport { CostLedger, type CostLedgerHandle } from './cost-ledger'\nimport { assertServedModel, ModelSubstitutionError } from './integrity/served-model'\nimport { recoverTruncatedJson } from './json-recovery'\nimport { JudgeParseError } from './judges'\nimport {\n  costReceiptFromLlm,\n  costReceiptFromLlmError,\n  maximumChargeForLlmRequest,\n} from './llm-client'\nimport { packageVersion } from './package-version'\nimport type { RawProviderEvent, RawProviderSink } from './trace/raw-provider-sink'\nimport {\n  certificationEvidenceDigest,\n  type DefaultVerdict,\n  type VerdictCertification,\n} from './verdict'\nimport type { CheckerIdentity, VerificationStrategySource } from './verification-strategy'\n\n/** What kind of produced state can satisfy a requirement structurally. */\nexport type SatisfiedBy = 'artifact' | 'proposal' | 'tool-call' | 'any'\n\nexport interface CompletionRequirement {\n  /** Stable id from the task gold (e.g. a persona's `expected_requirements[].req_id`). */\n  reqId: string\n  /** Human-readable description of the required deliverable. */\n  title: string\n  /** Optional kind/category hint, matched against a produced item's kind. */\n  category?: string\n  /** What produced state satisfies this requirement. Defaults to 'any'. */\n  satisfiedBy?: SatisfiedBy\n}\n\nexport interface TaskGold {\n  taskId: string\n  requirements: CompletionRequirement[]\n}\n\nexport interface ProducedProposal {\n  id: string\n  title: string\n  status: 'pending' | 'approved' | 'rejected'\n  /** Optional persisted body — when present, enables a correctness check. */\n  content?: string\n}\n\n/** Everything observable about what a run actually produced. */\nexport interface ProducedState {\n  /** Persisted vault artifacts. Reuses the shared `Artifact` shape. */\n  artifacts: Artifact[]\n  /** Proposals / filings the agent created. */\n  proposals: ProducedProposal[]\n  /** Names of tools the agent invoked. */\n  toolCalls: string[]\n}\n\nexport interface RequirementCheck {\n  reqId: string\n  title: string\n  /** A produced item of the right kind matched the requirement, non-empty. */\n  structurallyPresent: boolean\n  /**\n   * Whether the matched item actually fulfils the requirement. `null` when\n   * not structurally present, when the matched item carries no content\n   * to assess, or when the correctness check itself failed (`unmeasured`).\n   */\n  correct: boolean | null\n  /** structurallyPresent && !unmeasured && correct !== false. */\n  satisfied: boolean\n  /**\n   * Set when the correctness check itself errored (LLM call failure or an\n   * unparseable response after retry). The requirement's fulfilment is\n   * UNKNOWN — `correct` stays null, `satisfied` is false, and\n   * `completionVerdict` excludes the row from `completionRate`'s\n   * denominator. Never folded into a zero: a synthetic zero is\n   * indistinguishable from a real failure (see `JudgeParseError`).\n   */\n  unmeasured?: true\n  /** Why the correctness check could not be measured (present iff `unmeasured`). */\n  unmeasuredReason?: string\n  /** Human-readable evidence for the verdict. */\n  evidence: string[]\n}\n\n/** Extends the substrate verdict spine: `valid` = `fullyComplete` and\n *  `score` = `completionRate` — derived in `completionVerdict()`, the one\n *  place those equalities hold by construction. */\nexport interface CompletionVerdict extends DefaultVerdict {\n  taskId: string\n  requirements: RequirementCheck[]\n  /** satisfied / MEASURABLE requirements (unmeasured rows leave the denominator). */\n  completionRate: number\n  /** Every measurable requirement satisfied (false when anything is unmeasured). */\n  fullyComplete: boolean\n  /** Requirements whose correctness check errored — reported, never scored as zero. */\n  unmeasuredCount: number\n}\n\n/**\n * Construct a `CompletionVerdict` from the per-requirement checks, deriving\n * `completionRate` / `fullyComplete` and the spine fields (`valid` =\n * `fullyComplete`, `score` = `completionRate`) in one place. Throws on zero\n * requirements — a verdict over nothing is a misconfiguration, mirroring\n * `verifyCompletion`'s gold-spec guard.\n */\nexport function completionVerdict(input: {\n  taskId: string\n  requirements: RequirementCheck[]\n  /** What certified the correctness stage, when anything did. Omitted =\n   *  an uncertified verdict — the honest default for a bare checker. */\n  certification?: VerdictCertification\n}): CompletionVerdict {\n  if (input.requirements.length === 0) {\n    throw new Error(\n      `completionVerdict: task '${input.taskId}' has no requirement checks — nothing to derive a verdict from`,\n    )\n  }\n  const measurable = input.requirements.filter((r) => !r.unmeasured)\n  const unmeasuredCount = input.requirements.length - measurable.length\n  if (measurable.length === 0) {\n    // Every check errored: this is an infrastructure failure, not a scored\n    // run. A 0-rate verdict here would be a fabricated measurement.\n    throw new Error(\n      `completionVerdict: task '${input.taskId}' has no measurable requirements — all ${input.requirements.length} correctness checks failed (${input.requirements[0]?.unmeasuredReason ?? 'unknown reason'})`,\n    )\n  }\n  const satisfiedCount = measurable.filter((r) => r.satisfied).length\n  const completionRate = satisfiedCount / measurable.length\n  // A run with unmeasured rows can still report a rate over what WAS\n  // measured, but must not claim full completion.\n  const fullyComplete = unmeasuredCount === 0 && satisfiedCount === measurable.length\n  return {\n    taskId: input.taskId,\n    requirements: input.requirements,\n    completionRate,\n    fullyComplete,\n    unmeasuredCount,\n    valid: fullyComplete,\n    score: completionRate,\n    ...(input.certification === undefined ? {} : { certification: input.certification }),\n  }\n}\n\n/**\n * What a correctness checker declares about itself so the completion\n * verdict can carry a certification: which strategy member it discharges,\n * its exact identity, and the steps its answers rest on unverified.\n */\nexport interface CorrectnessCheckerAttestation {\n  strategy: VerificationStrategySource\n  checker: CheckerIdentity\n  assumptions: string[]\n}\n\n/**\n * Decides whether a produced item's content actually fulfils a requirement.\n * Injected so the structural verifier stays pure and unit-testable; the\n * production implementation is `createLlmCorrectnessChecker`.\n *\n * `attestation` is optional metadata on the function value: a checker that\n * carries one yields certified completion verdicts; a bare function yields\n * the same verdict uncertified. A plain arrow function remains a valid\n * checker.\n */\nexport interface CorrectnessChecker {\n  (\n    requirement: CompletionRequirement,\n    content: string,\n  ): Promise<{ correct: boolean; reason: string }>\n  attestation?: CorrectnessCheckerAttestation\n}\n\nconst STOPWORDS = new Set([\n  'the',\n  'a',\n  'an',\n  'of',\n  'for',\n  'and',\n  'or',\n  'to',\n  'in',\n  'on',\n  'with',\n  'by',\n])\n\n// Deliverable-FORM vocabulary — words that name the SHAPE of an output, not its\n// domain content. A correct \"swap-comparison view persisted as a ui/** artifact\"\n// is an OpenUI JSON whose body says nothing about \"artifact\" / \"persisted\" /\n// \"view\"; the discriminative tokens are the domain nouns (swap, comparison).\n// Stripped from the REQUIREMENT side of structural recall so a deliverable is\n// matched on what it IS about, not on the boilerplate describing its form. The\n// correctness checker strips the same class via TITLE_STOPWORDS. Anti-game holds:\n// the distinctive domain tokens remain, so an off-topic item still fails.\nconst REQUIREMENT_FORM_STOPWORDS = new Set([\n  'generated',\n  'generate',\n  'view',\n  'render',\n  'rendered',\n  'persisted',\n  'persist',\n  'artifact',\n  'file',\n  'document',\n  'note',\n  'proposal',\n  'deliverable',\n  'output',\n  'created',\n  'create',\n  'produce',\n  'produced',\n  'flag',\n])\n\nconst MATCH_THRESHOLD = 0.5\nconst MIN_CONTENT_CHARS = 50\n\nfunction tokens(s: string, extraStop?: Set<string>): Set<string> {\n  return new Set(\n    s\n      .toLowerCase()\n      .split(/[^a-z0-9]+/)\n      .filter((t) => t.length > 1 && !STOPWORDS.has(t) && !extraStop?.has(t)),\n  )\n}\n\n/**\n * Recall of the requirement's tokens within a candidate's identifying text.\n * Recall, not Jaccard — a candidate's path/id legitimately carries extra\n * tokens the requirement does not name. The requirement side drops\n * deliverable-FORM vocabulary so recall keys on the distinctive domain tokens.\n */\nfunction tokenRecall(requirementText: string, candidateText: string): number {\n  const req = tokens(requirementText, REQUIREMENT_FORM_STOPWORDS)\n  if (req.size === 0) return 0\n  const cand = tokens(candidateText)\n  let hit = 0\n  for (const t of req) if (cand.has(t)) hit++\n  return hit / req.size\n}\n\ninterface Candidate {\n  reqIndex: number\n  /** Unique key for a produced item — each item satisfies at most one requirement. */\n  itemKey: string\n  score: number\n  evidence: string\n  /** Content to correctness-check, or null when the matched item has none. */\n  content: string | null\n}\n\nfunction artifactCandidates(\n  req: CompletionRequirement,\n  reqIndex: number,\n  artifacts: Artifact[],\n): Candidate[] {\n  const reqText = `${req.title} ${req.category ?? ''}`\n  const out: Candidate[] = []\n  artifacts.forEach((a, i) => {\n    if ((a.content ?? '').trim().length < MIN_CONTENT_CHARS) return\n    // Match against the artifact CONTENT too, not just its path + kind — a\n    // generated view / note whose path is generic still satisfies a requirement\n    // when its body covers it (e.g. an OpenUI comparison grounded in the on-file\n    // figures). Bounded slice keeps the recall text cheap; MATCH_THRESHOLD holds.\n    let score = tokenRecall(\n      reqText,\n      `${a.path ?? ''} ${a.kind} ${(a.content ?? '').slice(0, 4000)}`,\n    )\n    if (req.category && a.kind && req.category.toLowerCase() === a.kind.toLowerCase()) {\n      score = Math.max(score, 1)\n    }\n    if (score < MATCH_THRESHOLD) return\n    out.push({\n      reqIndex,\n      itemKey: `artifact:${i}`,\n      score,\n      evidence: `artifact '${a.path ?? a.kind}' matched (token recall ${score.toFixed(2)})`,\n      content: a.content ?? null,\n    })\n  })\n  return out\n}\n\nfunction proposalCandidates(\n  req: CompletionRequirement,\n  reqIndex: number,\n  proposals: ProducedProposal[],\n): Candidate[] {\n  const reqText = `${req.title} ${req.category ?? ''}`\n  const out: Candidate[] = []\n  for (const p of proposals) {\n    // Pending or rejected work is not a completed deliverable.\n    if (p.status !== 'approved') continue\n    // A proposal needs an assessable BODY to be a deliverable. A bare title is\n    // not completion: correctness cannot be judged on it, so a title-only match\n    // would auto-pass the oracle (structurallyPresent && correct===null →\n    // satisfied) with no verifiable content. Tool calls are the only\n    // legitimately content-less deliverable (`toolCallCandidates`).\n    const body = (p.content ?? '').trim()\n    if (body.length < MIN_CONTENT_CHARS) continue\n    // Match against the body as well as the (often short) title — a refusal /\n    // flag / analysis proposal whose title is a label still satisfies a\n    // descriptively-worded requirement when its content covers it. MATCH_THRESHOLD\n    // + the requirement's distinctive tokens keep an off-topic proposal out;\n    // correctness (a SEMANTIC checker, NOT this lexical pass) then judges\n    // polarity/fulfilment, so a negation that merely contains the tokens fails.\n    // Structural and correctness must use different evidence or the two-stage\n    // check collapses to one lexical gate.\n    const score = tokenRecall(reqText, `${p.title} ${body}`)\n    if (score < MATCH_THRESHOLD) continue\n    out.push({\n      reqIndex,\n      itemKey: `proposal:${p.id}`,\n      score,\n      evidence: `approved proposal '${p.title}' matched (token recall ${score.toFixed(2)})`,\n      content: body,\n    })\n  }\n  return out\n}\n\nfunction toolCallCandidates(\n  req: CompletionRequirement,\n  reqIndex: number,\n  toolCalls: string[],\n): Candidate[] {\n  const out: Candidate[] = []\n  toolCalls.forEach((name, i) => {\n    const score = tokenRecall(req.title, name)\n    if (score < MATCH_THRESHOLD) return\n    out.push({\n      reqIndex,\n      itemKey: `tool:${i}`,\n      score,\n      evidence: `tool call '${name}' matched (token recall ${score.toFixed(2)})`,\n      content: null,\n    })\n  })\n  return out\n}\n\n/**\n * Verify whether a run completed the task. `checkCorrectness` is injected —\n * `createLlmCorrectnessChecker` for production, a deterministic stub in tests.\n *\n * Throws on a gold spec with no requirements: an eval task that requires\n * nothing is a misconfiguration, not a vacuously-complete task.\n */\nexport async function verifyCompletion(\n  gold: TaskGold,\n  state: ProducedState,\n  checkCorrectness: CorrectnessChecker,\n): Promise<CompletionVerdict> {\n  if (gold.requirements.length === 0) {\n    throw new Error(\n      `verifyCompletion: task '${gold.taskId}' has no requirements — malformed gold spec`,\n    )\n  }\n\n  // Collect every above-threshold (requirement, produced-item) candidate, then\n  // assign greedily by descending score: each requirement and each produced\n  // item is used at most once. One deliverable fulfils one requirement.\n  const candidates: Candidate[] = []\n  gold.requirements.forEach((req, i) => {\n    const by = req.satisfiedBy ?? 'any'\n    if (by === 'artifact' || by === 'any') {\n      candidates.push(...artifactCandidates(req, i, state.artifacts))\n    }\n    if (by === 'proposal' || by === 'any') {\n      candidates.push(...proposalCandidates(req, i, state.proposals))\n    }\n    if (by === 'tool-call' || by === 'any') {\n      candidates.push(...toolCallCandidates(req, i, state.toolCalls))\n    }\n  })\n  candidates.sort((a, b) => b.score - a.score)\n\n  const assigned = new Map<number, Candidate>()\n  const itemTaken = new Set<string>()\n  for (const c of candidates) {\n    if (assigned.has(c.reqIndex) || itemTaken.has(c.itemKey)) continue\n    assigned.set(c.reqIndex, c)\n    itemTaken.add(c.itemKey)\n  }\n\n  const requirements: RequirementCheck[] = []\n  for (let i = 0; i < gold.requirements.length; i++) {\n    const req = gold.requirements[i]!\n    const match = assigned.get(i)\n    const evidence: string[] = []\n    let correct: boolean | null = null\n    let unmeasuredReason: string | undefined\n\n    if (match) {\n      evidence.push(match.evidence)\n      if (match.content !== null) {\n        try {\n          const r = await checkCorrectness(req, match.content)\n          correct = r.correct\n          evidence.push(`correctness: ${r.correct ? 'pass' : 'fail'} — ${r.reason}`)\n        } catch (err) {\n          // The CHECKER failed, not the requirement. Recording this as a\n          // zero would fabricate a model failure out of an infrastructure\n          // one; the requirement is unmeasured and leaves the denominator.\n          unmeasuredReason =\n            err instanceof JudgeParseError\n              ? `checker response unparseable after retry: ${err.raw.slice(0, 200)}`\n              : `checker call failed: ${err instanceof Error ? err.message : String(err)}`\n          evidence.push(`correctness: UNMEASURED — ${unmeasuredReason}`)\n        }\n      } else {\n        evidence.push('correctness: not assessed — matched item carries no content')\n      }\n    } else {\n      const by = req.satisfiedBy ?? 'any'\n      const kind = by === 'any' ? 'artifact/proposal/tool-call' : by\n      evidence.push(`no produced ${kind} matched this requirement`)\n    }\n\n    const structurallyPresent = match !== undefined\n    const unmeasured = unmeasuredReason !== undefined\n    const satisfied = structurallyPresent && !unmeasured && correct !== false\n    requirements.push({\n      reqId: req.reqId,\n      title: req.title,\n      structurallyPresent,\n      correct,\n      satisfied,\n      ...(unmeasured ? { unmeasured: true as const, unmeasuredReason } : {}),\n      evidence,\n    })\n  }\n\n  // The verdict blends the deterministic structural stage with the injected\n  // correctness stage, so the certification names the checker's own member\n  // and carries the structural stage's lexical nature as an assumption. A\n  // checker without an attestation yields an uncertified verdict.\n  const attestation = checkCorrectness.attestation\n  const certification: VerdictCertification | undefined = attestation\n    ? {\n        strategy: attestation.strategy,\n        checker: attestation.checker,\n        assumptions: [\n          'structural matching is lexical token recall over produced items — the correctness stage only sees items it matched',\n          ...attestation.assumptions,\n        ],\n        evidenceDigest: certificationEvidenceDigest({ taskId: gold.taskId, requirements }),\n      }\n    : undefined\n\n  return completionVerdict({\n    taskId: gold.taskId,\n    requirements,\n    ...(certification === undefined ? {} : { certification }),\n  })\n}\n\nexport interface LlmCorrectnessCheckerOpts {\n  model?: string\n  /** Optional ledger for direct use. */\n  costLedger?: CostLedgerHandle\n  costPhase?: string\n  costTags?: Record<string, string>\n  signal?: AbortSignal\n  /** Max chars of artifact content sent to the checker. */\n  maxContentChars?: number\n  /**\n   * Checker LLM calls per requirement before giving up (parse failures and\n   * call errors both consume attempts). The failure then surfaces as an\n   * `unmeasured` requirement, never a zero.\n   */\n  maxAttempts?: number\n  /**\n   * Forensic capture of every checker request/response/error — without it a\n   * checker failure is unauditable (the agent-turn raws never contain the\n   * checker's own calls). Same sink contract as `LlmClient`.\n   */\n  rawSink?: RawProviderSink\n}\n\n/**\n * Parse the correctness checker's model response. Tolerates a response\n * truncated mid-JSON (max_tokens cap) by auto-closing the prefix — the\n * verdict boolean usually lands in the first few tokens, so a recovered\n * prefix with a boolean `correct` is a real measurement, not a guess.\n * Fails loud (JudgeParseError) when no boolean verdict is recoverable.\n */\nexport function parseCorrectnessResponse(raw: string): { correct: boolean; reason: string } {\n  const readVerdict = (candidate: unknown): { correct: boolean; reason: string } | null => {\n    if (candidate === null || typeof candidate !== 'object') return null\n    const { correct, reason } = candidate as { correct?: unknown; reason?: unknown }\n    if (typeof correct !== 'boolean') return null\n    return { correct, reason: typeof reason === 'string' ? reason : '' }\n  }\n\n  const match = raw.match(/\\{[\\s\\S]*\\}/)\n  if (match) {\n    try {\n      const strict = readVerdict(JSON.parse(match[0]))\n      if (strict) return strict\n    } catch {\n      // fall through to truncation recovery\n    }\n  }\n  // The strict path needs a closing `}`; a cap-hit response has none. Take\n  // everything from the first `{` and auto-close it.\n  const start = raw.indexOf('{')\n  if (start !== -1) {\n    const recovered = readVerdict(recoverTruncatedJson(raw.slice(start)))\n    if (recovered) return recovered\n  }\n  throw new JudgeParseError('correctness-checker', raw)\n}\n\n/**\n * Production `CorrectnessChecker` — one LLM call per matched artifact,\n * deterministic (temperature 0), structured JSON out. Judges fulfilment\n * only: a plan, a gesture, or a description of what should be done does not\n * fulfil a requirement — the artifact must BE the deliverable.\n */\nexport function createLlmCorrectnessChecker(\n  chat: ChatClient,\n  opts: LlmCorrectnessCheckerOpts = {},\n): CorrectnessChecker {\n  const model = opts.model ?? 'claude-sonnet-4-6'\n  const maxContentChars = opts.maxContentChars ?? 8000\n  const maxAttempts = opts.maxAttempts ?? 2\n  const costLedger = opts.costLedger ?? new CostLedger()\n  const sink = opts.rawSink\n  const record = async (event: RawProviderEvent): Promise<void> => {\n    // Forensic capture is best-effort; the verdict is the system of record.\n    try {\n      await sink?.record(event)\n    } catch {\n      // Intentionally swallowed.\n    }\n  }\n  const checker: CorrectnessChecker = async (requirement, content) => {\n    const request = {\n      model,\n      messages: [\n        {\n          role: 'system' as const,\n          content:\n            'You verify whether a produced work artifact actually fulfils a stated requirement. Judge fulfilment only — is the deliverable substantively present and on-point — not polish. A plan to do it later, a vague gesture, or a description of what should be done does NOT fulfil a requirement; the artifact must BE the deliverable. Respond with a single JSON object: {\"correct\": boolean, \"reason\": string (<= 30 words)}.',\n        },\n        {\n          role: 'user' as const,\n          content: `Requirement: ${requirement.title}\\n${\n            requirement.category ? `Category: ${requirement.category}\\n` : ''\n          }\\nProduced artifact:\\n${content.slice(0, maxContentChars)}`,\n        },\n      ],\n      temperature: 0,\n      maxTokens: 200,\n    } satisfies ChatRequest\n    let lastErr: unknown\n    for (let attempt = 0; attempt < maxAttempts; attempt++) {\n      const started = Date.now()\n      await record({\n        eventId: randomUUID(),\n        provider: chat.transport,\n        model,\n        endpoint: '/chat',\n        baseUrl: '',\n        attemptIndex: attempt,\n        direction: 'request',\n        timestamp: started,\n        requestBody: request,\n        redactedFields: [],\n      })\n      try {\n        const paid = await costLedger.runPaidCall({\n          channel: 'verifier',\n          phase: opts.costPhase ?? 'completion.correctness',\n          actor: 'correctness-checker',\n          model,\n          maximumCharge:\n            chat.maximumAttempts === undefined\n              ? undefined\n              : maximumChargeForLlmRequest(request, { maximumAttempts: chat.maximumAttempts }),\n          tags: {\n            ...opts.costTags,\n            requirementId: requirement.reqId,\n            attempt: String(attempt),\n          },\n          signal: opts.signal,\n          execute: (signal, callId) => chat.chat(request, { signal, idempotencyKey: callId }),\n          receipt: costReceiptFromLlm,\n          receiptFromError: costReceiptFromLlmError,\n        })\n        if (!paid.succeeded) throw paid.error\n        const resp = paid.value\n        // Hold the transport to its own word: a verdict produced by a model\n        // other than the one requested is not this model's verdict. A\n        // transport that echoes no id cannot be made to prove identity here —\n        // callers needing that proof enable it at the client\n        // (`LlmClientOptions.assertServedModel`) or gate on `assertModelsServed`.\n        assertServedModel(model, resp.servedModel, {\n          allowUnreported: true,\n          context: `correctness checker for requirement ${requirement.reqId}`,\n        })\n        const raw = resp.content\n        await record({\n          eventId: randomUUID(),\n          provider: chat.transport,\n          model,\n          endpoint: '/chat',\n          baseUrl: '',\n          attemptIndex: attempt,\n          direction: 'response',\n          timestamp: Date.now(),\n          durationMs: Date.now() - started,\n          responseBody: resp,\n          redactedFields: [],\n        })\n        return parseCorrectnessResponse(raw)\n      } catch (err) {\n        lastErr = err\n        await record({\n          eventId: randomUUID(),\n          provider: chat.transport,\n          model,\n          endpoint: '/chat',\n          baseUrl: '',\n          attemptIndex: attempt,\n          direction: 'error',\n          timestamp: Date.now(),\n          durationMs: Date.now() - started,\n          errorMessage: err instanceof Error ? err.message : String(err),\n          redactedFields: [],\n        })\n        // Substitution is a verdict about identity, not a transient fault:\n        // another attempt reaches the same wrong model and spends more tokens.\n        if (err instanceof ModelSubstitutionError) throw err\n      }\n    }\n    throw lastErr instanceof Error ? lastErr : new Error(String(lastErr))\n  }\n  checker.attestation = {\n    strategy: 'judge',\n    checker: { name: 'llm-correctness-checker', version: model },\n    assumptions: [\n      'served-model identity is accepted when the transport reports none (assertServedModel allowUnreported)',\n    ],\n  }\n  return checker\n}\n\n/** Stopwords for requirement-title tokenization — drops the imperative verbs\n *  ('review', 'update', …) common to deliverable titles so recall keys on the\n *  substantive nouns, not the boilerplate ask. */\nconst TITLE_STOPWORDS = new Set([\n  'the',\n  'a',\n  'an',\n  'and',\n  'or',\n  'for',\n  'to',\n  'of',\n  'in',\n  'on',\n  'with',\n  'review',\n  'update',\n  'new',\n  'proposed',\n])\n\n/**\n * Deterministic `CorrectnessChecker` — the no-LLM counterpart to\n * `createLlmCorrectnessChecker`. A produced item fulfils a requirement when its\n * content is substantive (≥ `minContentLength` chars) AND recalls ≥ `minRecall`\n * of the requirement title's significant tokens. No network.\n *\n * Polarity-blind: token recall credits a negation that contains the\n * requirement's tokens (\"I will NOT produce the comparison\" recalls every token\n * of \"produce the comparison\"). The structural match stage is ALSO lexical, so\n * pairing the two collapses to a single gameable gate. Use this only as an\n * opt-in structural pre-filter or for tasks whose requirements have no polarity\n * to invert; for produced-state grading the correctness checker MUST be semantic\n * (`createLlmCorrectnessChecker`). See the anti-game fixtures in the test suite.\n */\nexport function createTokenRecallChecker(\n  opts: { minRecall?: number; minContentLength?: number } = {},\n): CorrectnessChecker {\n  const minRecall = opts.minRecall ?? 0.5\n  const minLen = opts.minContentLength ?? 120\n  const checker: CorrectnessChecker = async (requirement, content) => {\n    const body = content.trim()\n    if (body.length < minLen)\n      return {\n        correct: false,\n        reason: `content too thin (${body.length} chars) to be the deliverable`,\n      }\n    const titleTokens = requirement.title\n      .toLowerCase()\n      .split(/[^a-z0-9]+/)\n      .filter((t) => t.length > 2 && !TITLE_STOPWORDS.has(t))\n    if (titleTokens.length === 0)\n      return {\n        correct: true,\n        reason: 'requirement title has no significant tokens — structural match accepted',\n      }\n    const lower = body.toLowerCase()\n    const hits = titleTokens.filter((t) => lower.includes(t)).length\n    const recall = hits / titleTokens.length\n    return recall >= minRecall\n      ? {\n          correct: true,\n          reason: `content recalls ${hits}/${titleTokens.length} requirement tokens`,\n        }\n      : {\n          correct: false,\n          reason: `content recalls only ${hits}/${titleTokens.length} requirement tokens`,\n        }\n  }\n  // 'schema': token recall checks shape, not meaning — the member whose\n  // documented failure mode (\"a well-formed wrong answer passes\") is\n  // exactly this checker's polarity blindness.\n  checker.attestation = {\n    strategy: 'schema',\n    checker: { name: 'token-recall-checker', version: packageVersion() },\n    assumptions: ['polarity-blind: a negation that recalls the requirement tokens passes'],\n  }\n  return checker\n}\n","/**\n * Produced-state extraction — normalize a run's runtime event stream into the\n * typed `ProducedState` the completion oracle consumes.\n *\n * `ProducedState` answers \"what did the agent actually produce\" — vault\n * artifacts, proposals, tool calls. The runtime emits these as a stream of\n * events; this module is the single normalization point from that stream to\n * the shape `verifyCompletion` expects.\n *\n * Input is structurally typed (`RuntimeEventLike`) so this module does not\n * depend on agent-runtime — agent-runtime's `RuntimeStreamEvent` satisfies it\n * structurally. The `content` on `ArtifactEventLike` and the whole\n * `proposal_created` variant are the runtime-side enrichments this contract\n * requires; the runtime emits them, this module consumes them.\n */\n\nimport type { Artifact } from './artifact-validator'\nimport type { ProducedProposal, ProducedState } from './completion-verifier'\n\n/** A tool the agent invoked. */\nexport interface ToolCallEventLike {\n  type: 'tool_call'\n  toolName: string\n}\n\n/**\n * An artifact the agent produced. `content` is the enriched field — the\n * runtime's base `artifact` event carries only metadata; the completion\n * oracle needs the body to verify the deliverable, so the runtime emits it.\n */\nexport interface ArtifactEventLike {\n  type: 'artifact'\n  artifactId: string\n  name?: string\n  mimeType?: string\n  uri?: string\n  content?: string\n}\n\n/** A proposal / filing the agent created. */\nexport interface ProposalEventLike {\n  type: 'proposal_created'\n  proposalId: string\n  title: string\n  status?: 'pending' | 'approved' | 'rejected'\n  // body of the proposal (e.g. a submit_proposal `description`). When present,\n  // the completion oracle correctness-checks it like artifact content; absent,\n  // the proposal is graded presence-only.\n  content?: string\n}\n\n/**\n * The subset of runtime stream events `extractProducedState` consumes.\n * agent-runtime's full `RuntimeStreamEvent` union satisfies this structurally;\n * the `{ type: string }` catch-all keeps the input permissive so callers can\n * pass the whole unfiltered telemetry stream — unrecognized events are skipped.\n */\nexport type RuntimeEventLike =\n  | ToolCallEventLike\n  | ArtifactEventLike\n  | ProposalEventLike\n  | { type: string }\n\nfunction artifactKind(mimeType: string | undefined): string {\n  if (!mimeType) return 'file'\n  if (mimeType.includes('json')) return 'json'\n  if (mimeType.startsWith('text/')) return 'text'\n  return 'file'\n}\n\n/**\n * Normalize an emission-ordered stream into its latest observed produced state.\n * Artifacts are keyed by their exact output path (name, then URI, then id);\n * proposals by id. Later observations replace earlier ones, including missing\n * content or a rejected status. An obsolete version cannot prove completion.\n * Distinct paths remain distinct; no path normalization or aliasing is inferred.\n * Results retain first-seen identity order. Tool names describe invocation,\n * not success, and are deduplicated in first-seen order.\n *\n * An artifact without observed content yields empty content, which the\n * completion oracle rejects. This projection does not verify external effects,\n * recover dropped events, or establish freshness beyond the supplied stream.\n */\nexport function extractProducedState(events: readonly RuntimeEventLike[]): ProducedState {\n  const artifacts = new Map<string, Artifact>()\n  const proposals = new Map<string, ProducedProposal>()\n  const toolCalls: string[] = []\n  const seenTools = new Set<string>()\n\n  for (const ev of events) {\n    if (ev.type === 'tool_call') {\n      const name = (ev as ToolCallEventLike).toolName\n      if (name && !seenTools.has(name)) {\n        seenTools.add(name)\n        toolCalls.push(name)\n      }\n    } else if (ev.type === 'artifact') {\n      const a = ev as ArtifactEventLike\n      const path = a.name ?? a.uri ?? a.artifactId\n      artifacts.set(path, {\n        kind: artifactKind(a.mimeType),\n        path,\n        content: a.content ?? '',\n      })\n    } else if (ev.type === 'proposal_created') {\n      const p = ev as ProposalEventLike\n      proposals.set(p.proposalId, {\n        id: p.proposalId,\n        title: p.title,\n        status: p.status ?? 'pending',\n        ...(p.content !== undefined ? { content: p.content } : {}),\n      })\n    }\n  }\n\n  return { artifacts: [...artifacts.values()], proposals: [...proposals.values()], toolCalls }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;AAeA,MAAa,mBAA2C;CACtD;CACA;CACA;CACA;AACF;;;;;;AAwBA,MAAa,uBAAuB;;;;;;;;;;;;;;;;;;;AAoBpC,SAAgB,kBAAkB,MAAuC;CACvE,MAAM,YAAY,KAAK,aAAa;CACpC,IAAI,UAAU,WAAW,GAAG,MAAM,IAAI,gBAAgB,0CAA0C;CAChG,MAAM,YAAY,KAAK,KAAK,OAAO;CACnC,MAAM,SAAS,KAAK,WAAW,YAAY,CAAC,SAAS,IAAI,CAAC;CAC1D,IAAI,OAAO,WAAW,GACpB,MAAM,IAAI,gBACR,kGACF;CAEF,MAAM,MAAsB,CAAC;CAC7B,MAAM,uBAAO,IAAI,IAAY;CAC7B,KAAK,MAAM,WAAW,WAAW;EAS/B,MAAM,YAAY,KAAK,mBACnB,SACA,OAAO,QAAQ,UAAU,qBAAqB,SAAS,KAAK,CAAC;EACjE,MAAM,YAAY,UAAU,SAAS,IAAI,YAAY,CAAC,oBAAoB;EAC1E,KAAK,MAAM,SAAS,WAAW;GAC7B,MAAM,UAAwB;IAC5B,GAAG,KAAK;IACR;IACA,MAAM,GAAG,KAAK,KAAK,QAAQ,QAAQ,GAAG,QAAQ,GAAG;IACjD,OAAO;KAAE,GAAG,KAAK,KAAK;KAAO,SAAS;IAAM;IAC5C,UAAU;KAAE,GAAI,KAAK,KAAK,YAAY,CAAC;KAAI;KAAS,cAAc;IAAM;GAC1E;GACA,MAAM,KAAK,eAAe,OAAO;GACjC,IAAI,KAAK,IAAI,EAAE,GAAG;GAClB,KAAK,IAAI,EAAE;GACX,IAAI,KAAK,OAAO;EAClB;CACF;CACA,IAAI,IAAI,WAAW,GAGjB,MAAM,IAAI,gBACR,uDAAuD,UAAU,KAAK,IAAI,EAAE,aAAa,OAAO,KAAK,IAAI,EAAE,IAC7G;CAEF,OAAO;AACT;;;;;;;;AASA,SAAgB,cACd,SACqD;CACrD,MAAM,IAAI,QAAQ;CAClB,MAAM,UAAU,GAAG;CACnB,MAAM,QAAQ,GAAG;CACjB,IAAI,OAAO,YAAY,YAAY,OAAO,UAAU,UAClD,OAAO;EAAW;EAAwB;CAAM;AAGpD;;;;;;;;AASA,SAAgB,eAAe,SAA+B;CAE5D,OAAO,GADO,qBAAqB,yBAAyB,OAAO,CAAC,KAAK,UACzD,GAAG,iBAAiB,OAAO,CAAC,CAAC,MAAM,GAAG,EAAE;AAC1D;;;;;AAMA,SAAgB,oBAAoB,SAA+B;CACjE,MAAM,QAAQ,QAAQ,OAAO,SAAS,KAAK;CAC3C,IAAI,CAAC,OAEH,MAAM,IAAI,gBACR,iBAFY,yBAAyB,OAAO,KAAK,kBAE1B,gDACzB;CAEF,OAAO;AACT;AAEA,SAAS,yBAAyB,SAA2C;CAC3E,OAAO,QAAQ,MAAM,KAAK,KAAK,QAAQ,SAAS,KAAK,KAAK,KAAA;AAC5D;AAEA,SAAS,qBAAqB,OAA+C;CAM3E,OALa,OACT,KAAK,CAAC,CACP,QAAQ,qBAAqB,GAAG,CAAC,CACjC,QAAQ,OAAO,GAAG,CAAC,CACnB,QAAQ,UAAU,EAAE,KACR,KAAA;AACjB;AAEA,SAAS,QAA2C,OAAsB;CACxE,MAAM,MAA+B,CAAC;CACtC,KAAK,MAAM,CAAC,KAAK,UAAU,OAAO,QAAQ,KAAK,GAC7C,IAAI,UAAU,KAAA,GAAW,IAAI,OAAO;CAEtC,OAAO;AACT;;;;;;;;;;;AAYA,SAAgB,iBAAiB,SAA+B;CAC9D,MAAM,QAAQ,oBAAoB,OAAO;CACzC,MAAM,YAAY,QAAQ;EACxB,GAAG;EACH,MAAM,KAAA;EACN,aAAa,KAAA;EACb,MAAM,QAAQ,OAAO,CAAC,GAAG,QAAQ,IAAI,CAAC,CAAC,KAAK,IAAI,KAAA;EAChD,OAAO,QAAQ;GAAE,GAAG,QAAQ;GAAO,SAAS;EAAM,CAAC;CACrD,CAAC;CACD,OAAO,WAAW,QAAQ,CAAC,CAAC,OAAO,gBAAgB,SAAS,CAAC,CAAC,CAAC,OAAO,KAAK;AAC7E;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;ACtEA,SAAgB,kBAAkB,OAMZ;CACpB,IAAI,MAAM,aAAa,WAAW,GAChC,MAAM,IAAI,MACR,4BAA4B,MAAM,OAAO,+DAC3C;CAEF,MAAM,aAAa,MAAM,aAAa,QAAQ,MAAM,CAAC,EAAE,UAAU;CACjE,MAAM,kBAAkB,MAAM,aAAa,SAAS,WAAW;CAC/D,IAAI,WAAW,WAAW,GAGxB,MAAM,IAAI,MACR,4BAA4B,MAAM,OAAO,yCAAyC,MAAM,aAAa,OAAO,8BAA8B,MAAM,aAAa,EAAE,EAAE,oBAAoB,iBAAiB,EACxM;CAEF,MAAM,iBAAiB,WAAW,QAAQ,MAAM,EAAE,SAAS,CAAC,CAAC;CAC7D,MAAM,iBAAiB,iBAAiB,WAAW;CAGnD,MAAM,gBAAgB,oBAAoB,KAAK,mBAAmB,WAAW;CAC7E,OAAO;EACL,QAAQ,MAAM;EACd,cAAc,MAAM;EACpB;EACA;EACA;EACA,OAAO;EACP,OAAO;EACP,GAAI,MAAM,kBAAkB,KAAA,IAAY,CAAC,IAAI,EAAE,eAAe,MAAM,cAAc;CACpF;AACF;AA+BA,MAAM,4BAAY,IAAI,IAAI;CACxB;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF,CAAC;AAUD,MAAM,6CAA6B,IAAI,IAAI;CACzC;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF,CAAC;AAED,MAAM,kBAAkB;AACxB,MAAM,oBAAoB;AAE1B,SAAS,OAAO,GAAW,WAAsC;CAC/D,OAAO,IAAI,IACT,EACG,YAAY,CAAC,CACb,MAAM,YAAY,CAAC,CACnB,QAAQ,MAAM,EAAE,SAAS,KAAK,CAAC,UAAU,IAAI,CAAC,KAAK,CAAC,WAAW,IAAI,CAAC,CAAC,CAC1E;AACF;;;;;;;AAQA,SAAS,YAAY,iBAAyB,eAA+B;CAC3E,MAAM,MAAM,OAAO,iBAAiB,0BAA0B;CAC9D,IAAI,IAAI,SAAS,GAAG,OAAO;CAC3B,MAAM,OAAO,OAAO,aAAa;CACjC,IAAI,MAAM;CACV,KAAK,MAAM,KAAK,KAAK,IAAI,KAAK,IAAI,CAAC,GAAG;CACtC,OAAO,MAAM,IAAI;AACnB;AAYA,SAAS,mBACP,KACA,UACA,WACa;CACb,MAAM,UAAU,GAAG,IAAI,MAAM,GAAG,IAAI,YAAY;CAChD,MAAM,MAAmB,CAAC;CAC1B,UAAU,SAAS,GAAG,MAAM;EAC1B,KAAK,EAAE,WAAW,GAAA,CAAI,KAAK,CAAC,CAAC,SAAS,mBAAmB;EAKzD,IAAI,QAAQ,YACV,SACA,GAAG,EAAE,QAAQ,GAAG,GAAG,EAAE,KAAK,IAAI,EAAE,WAAW,GAAA,CAAI,MAAM,GAAG,GAAI,GAC9D;EACA,IAAI,IAAI,YAAY,EAAE,QAAQ,IAAI,SAAS,YAAY,MAAM,EAAE,KAAK,YAAY,GAC9E,QAAQ,KAAK,IAAI,OAAO,CAAC;EAE3B,IAAI,QAAQ,iBAAiB;EAC7B,IAAI,KAAK;GACP;GACA,SAAS,YAAY;GACrB;GACA,UAAU,aAAa,EAAE,QAAQ,EAAE,KAAK,0BAA0B,MAAM,QAAQ,CAAC,EAAE;GACnF,SAAS,EAAE,WAAW;EACxB,CAAC;CACH,CAAC;CACD,OAAO;AACT;AAEA,SAAS,mBACP,KACA,UACA,WACa;CACb,MAAM,UAAU,GAAG,IAAI,MAAM,GAAG,IAAI,YAAY;CAChD,MAAM,MAAmB,CAAC;CAC1B,KAAK,MAAM,KAAK,WAAW;EAEzB,IAAI,EAAE,WAAW,YAAY;EAM7B,MAAM,QAAQ,EAAE,WAAW,GAAA,CAAI,KAAK;EACpC,IAAI,KAAK,SAAS,mBAAmB;EASrC,MAAM,QAAQ,YAAY,SAAS,GAAG,EAAE,MAAM,GAAG,MAAM;EACvD,IAAI,QAAQ,iBAAiB;EAC7B,IAAI,KAAK;GACP;GACA,SAAS,YAAY,EAAE;GACvB;GACA,UAAU,sBAAsB,EAAE,MAAM,0BAA0B,MAAM,QAAQ,CAAC,EAAE;GACnF,SAAS;EACX,CAAC;CACH;CACA,OAAO;AACT;AAEA,SAAS,mBACP,KACA,UACA,WACa;CACb,MAAM,MAAmB,CAAC;CAC1B,UAAU,SAAS,MAAM,MAAM;EAC7B,MAAM,QAAQ,YAAY,IAAI,OAAO,IAAI;EACzC,IAAI,QAAQ,iBAAiB;EAC7B,IAAI,KAAK;GACP;GACA,SAAS,QAAQ;GACjB;GACA,UAAU,cAAc,KAAK,0BAA0B,MAAM,QAAQ,CAAC,EAAE;GACxE,SAAS;EACX,CAAC;CACH,CAAC;CACD,OAAO;AACT;;;;;;;;AASA,eAAsB,iBACpB,MACA,OACA,kBAC4B;CAC5B,IAAI,KAAK,aAAa,WAAW,GAC/B,MAAM,IAAI,MACR,2BAA2B,KAAK,OAAO,4CACzC;CAMF,MAAM,aAA0B,CAAC;CACjC,KAAK,aAAa,SAAS,KAAK,MAAM;EACpC,MAAM,KAAK,IAAI,eAAe;EAC9B,IAAI,OAAO,cAAc,OAAO,OAC9B,WAAW,KAAK,GAAG,mBAAmB,KAAK,GAAG,MAAM,SAAS,CAAC;EAEhE,IAAI,OAAO,cAAc,OAAO,OAC9B,WAAW,KAAK,GAAG,mBAAmB,KAAK,GAAG,MAAM,SAAS,CAAC;EAEhE,IAAI,OAAO,eAAe,OAAO,OAC/B,WAAW,KAAK,GAAG,mBAAmB,KAAK,GAAG,MAAM,SAAS,CAAC;CAElE,CAAC;CACD,WAAW,MAAM,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK;CAE3C,MAAM,2BAAW,IAAI,IAAuB;CAC5C,MAAM,4BAAY,IAAI,IAAY;CAClC,KAAK,MAAM,KAAK,YAAY;EAC1B,IAAI,SAAS,IAAI,EAAE,QAAQ,KAAK,UAAU,IAAI,EAAE,OAAO,GAAG;EAC1D,SAAS,IAAI,EAAE,UAAU,CAAC;EAC1B,UAAU,IAAI,EAAE,OAAO;CACzB;CAEA,MAAM,eAAmC,CAAC;CAC1C,KAAK,IAAI,IAAI,GAAG,IAAI,KAAK,aAAa,QAAQ,KAAK;EACjD,MAAM,MAAM,KAAK,aAAa;EAC9B,MAAM,QAAQ,SAAS,IAAI,CAAC;EAC5B,MAAM,WAAqB,CAAC;EAC5B,IAAI,UAA0B;EAC9B,IAAI;EAEJ,IAAI,OAAO;GACT,SAAS,KAAK,MAAM,QAAQ;GAC5B,IAAI,MAAM,YAAY,MACpB,IAAI;IACF,MAAM,IAAI,MAAM,iBAAiB,KAAK,MAAM,OAAO;IACnD,UAAU,EAAE;IACZ,SAAS,KAAK,gBAAgB,EAAE,UAAU,SAAS,OAAO,KAAK,EAAE,QAAQ;GAC3E,SAAS,KAAK;IAIZ,mBACE,eAAe,kBACX,6CAA6C,IAAI,IAAI,MAAM,GAAG,GAAG,MACjE,wBAAwB,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;IAC7E,SAAS,KAAK,6BAA6B,kBAAkB;GAC/D;QAEA,SAAS,KAAK,6DAA6D;EAE/E,OAAO;GACL,MAAM,KAAK,IAAI,eAAe;GAC9B,MAAM,OAAO,OAAO,QAAQ,gCAAgC;GAC5D,SAAS,KAAK,eAAe,KAAK,0BAA0B;EAC9D;EAEA,MAAM,sBAAsB,UAAU,KAAA;EACtC,MAAM,aAAa,qBAAqB,KAAA;EACxC,MAAM,YAAY,uBAAuB,CAAC,cAAc,YAAY;EACpE,aAAa,KAAK;GAChB,OAAO,IAAI;GACX,OAAO,IAAI;GACX;GACA;GACA;GACA,GAAI,aAAa;IAAE,YAAY;IAAe;GAAiB,IAAI,CAAC;GACpE;EACF,CAAC;CACH;CAMA,MAAM,cAAc,iBAAiB;CACrC,MAAM,gBAAkD,cACpD;EACE,UAAU,YAAY;EACtB,SAAS,YAAY;EACrB,aAAa,CACX,sHACA,GAAG,YAAY,WACjB;EACA,gBAAgB,4BAA4B;GAAE,QAAQ,KAAK;GAAQ;EAAa,CAAC;CACnF,IACA,KAAA;CAEJ,OAAO,kBAAkB;EACvB,QAAQ,KAAK;EACb;EACA,GAAI,kBAAkB,KAAA,IAAY,CAAC,IAAI,EAAE,cAAc;CACzD,CAAC;AACH;;;;;;;;AAgCA,SAAgB,yBAAyB,KAAmD;CAC1F,MAAM,eAAe,cAAoE;EACvF,IAAI,cAAc,QAAQ,OAAO,cAAc,UAAU,OAAO;EAChE,MAAM,EAAE,SAAS,WAAW;EAC5B,IAAI,OAAO,YAAY,WAAW,OAAO;EACzC,OAAO;GAAE;GAAS,QAAQ,OAAO,WAAW,WAAW,SAAS;EAAG;CACrE;CAEA,MAAM,QAAQ,IAAI,MAAM,aAAa;CACrC,IAAI,OACF,IAAI;EACF,MAAM,SAAS,YAAY,KAAK,MAAM,MAAM,EAAE,CAAC;EAC/C,IAAI,QAAQ,OAAO;CACrB,QAAQ,CAER;CAIF,MAAM,QAAQ,IAAI,QAAQ,GAAG;CAC7B,IAAI,UAAU,IAAI;EAChB,MAAM,YAAY,YAAY,qBAAqB,IAAI,MAAM,KAAK,CAAC,CAAC;EACpE,IAAI,WAAW,OAAO;CACxB;CACA,MAAM,IAAI,gBAAgB,uBAAuB,GAAG;AACtD;;;;;;;AAQA,SAAgB,4BACd,MACA,OAAkC,CAAC,GACf;CACpB,MAAM,QAAQ,KAAK,SAAS;CAC5B,MAAM,kBAAkB,KAAK,mBAAmB;CAChD,MAAM,cAAc,KAAK,eAAe;CACxC,MAAM,aAAa,KAAK,cAAc,IAAI,WAAW;CACrD,MAAM,OAAO,KAAK;CAClB,MAAM,SAAS,OAAO,UAA2C;EAE/D,IAAI;GACF,MAAM,MAAM,OAAO,KAAK;EAC1B,QAAQ,CAER;CACF;CACA,MAAM,UAA8B,OAAO,aAAa,YAAY;EAClE,MAAM,UAAU;GACd;GACA,UAAU,CACR;IACE,MAAM;IACN,SACE;GACJ,GACA;IACE,MAAM;IACN,SAAS,gBAAgB,YAAY,MAAM,IACzC,YAAY,WAAW,aAAa,YAAY,SAAS,MAAM,GAChE,wBAAwB,QAAQ,MAAM,GAAG,eAAe;GAC3D,CACF;GACA,aAAa;GACb,WAAW;EACb;EACA,IAAI;EACJ,KAAK,IAAI,UAAU,GAAG,UAAU,aAAa,WAAW;GACtD,MAAM,UAAU,KAAK,IAAI;GACzB,MAAM,OAAO;IACX,SAAS,WAAW;IACpB,UAAU,KAAK;IACf;IACA,UAAU;IACV,SAAS;IACT,cAAc;IACd,WAAW;IACX,WAAW;IACX,aAAa;IACb,gBAAgB,CAAC;GACnB,CAAC;GACD,IAAI;IACF,MAAM,OAAO,MAAM,WAAW,YAAY;KACxC,SAAS;KACT,OAAO,KAAK,aAAa;KACzB,OAAO;KACP;KACA,eACE,KAAK,oBAAoB,KAAA,IACrB,KAAA,IACA,2BAA2B,SAAS,EAAE,iBAAiB,KAAK,gBAAgB,CAAC;KACnF,MAAM;MACJ,GAAG,KAAK;MACR,eAAe,YAAY;MAC3B,SAAS,OAAO,OAAO;KACzB;KACA,QAAQ,KAAK;KACb,UAAU,QAAQ,WAAW,KAAK,KAAK,SAAS;MAAE;MAAQ,gBAAgB;KAAO,CAAC;KAClF,SAAS;KACT,kBAAkB;IACpB,CAAC;IACD,IAAI,CAAC,KAAK,WAAW,MAAM,KAAK;IAChC,MAAM,OAAO,KAAK;IAMlB,kBAAkB,OAAO,KAAK,aAAa;KACzC,iBAAiB;KACjB,SAAS,uCAAuC,YAAY;IAC9D,CAAC;IACD,MAAM,MAAM,KAAK;IACjB,MAAM,OAAO;KACX,SAAS,WAAW;KACpB,UAAU,KAAK;KACf;KACA,UAAU;KACV,SAAS;KACT,cAAc;KACd,WAAW;KACX,WAAW,KAAK,IAAI;KACpB,YAAY,KAAK,IAAI,IAAI;KACzB,cAAc;KACd,gBAAgB,CAAC;IACnB,CAAC;IACD,OAAO,yBAAyB,GAAG;GACrC,SAAS,KAAK;IACZ,UAAU;IACV,MAAM,OAAO;KACX,SAAS,WAAW;KACpB,UAAU,KAAK;KACf;KACA,UAAU;KACV,SAAS;KACT,cAAc;KACd,WAAW;KACX,WAAW,KAAK,IAAI;KACpB,YAAY,KAAK,IAAI,IAAI;KACzB,cAAc,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;KAC7D,gBAAgB,CAAC;IACnB,CAAC;IAGD,IAAI,eAAe,wBAAwB,MAAM;GACnD;EACF;EACA,MAAM,mBAAmB,QAAQ,UAAU,IAAI,MAAM,OAAO,OAAO,CAAC;CACtE;CACA,QAAQ,cAAc;EACpB,UAAU;EACV,SAAS;GAAE,MAAM;GAA2B,SAAS;EAAM;EAC3D,aAAa,CACX,uGACF;CACF;CACA,OAAO;AACT;;;ACpmBA,SAAS,aAAa,UAAsC;CAC1D,IAAI,CAAC,UAAU,OAAO;CACtB,IAAI,SAAS,SAAS,MAAM,GAAG,OAAO;CACtC,IAAI,SAAS,WAAW,OAAO,GAAG,OAAO;CACzC,OAAO;AACT;;;;;;;;;;;;;;AAeA,SAAgB,qBAAqB,QAAoD;CACvF,MAAM,4BAAY,IAAI,IAAsB;CAC5C,MAAM,4BAAY,IAAI,IAA8B;CACpD,MAAM,YAAsB,CAAC;CAC7B,MAAM,4BAAY,IAAI,IAAY;CAElC,KAAK,MAAM,MAAM,QACf,IAAI,GAAG,SAAS,aAAa;EAC3B,MAAM,OAAQ,GAAyB;EACvC,IAAI,QAAQ,CAAC,UAAU,IAAI,IAAI,GAAG;GAChC,UAAU,IAAI,IAAI;GAClB,UAAU,KAAK,IAAI;EACrB;CACF,OAAO,IAAI,GAAG,SAAS,YAAY;EACjC,MAAM,IAAI;EACV,MAAM,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE;EAClC,UAAU,IAAI,MAAM;GAClB,MAAM,aAAa,EAAE,QAAQ;GAC7B;GACA,SAAS,EAAE,WAAW;EACxB,CAAC;CACH,OAAO,IAAI,GAAG,SAAS,oBAAoB;EACzC,MAAM,IAAI;EACV,UAAU,IAAI,EAAE,YAAY;GAC1B,IAAI,EAAE;GACN,OAAO,EAAE;GACT,QAAQ,EAAE,UAAU;GACpB,GAAI,EAAE,YAAY,KAAA,IAAY,EAAE,SAAS,EAAE,QAAQ,IAAI,CAAC;EAC1D,CAAC;CACH;CAGF,OAAO;EAAE,WAAW,CAAC,GAAG,UAAU,OAAO,CAAC;EAAG,WAAW,CAAC,GAAG,UAAU,OAAO,CAAC;EAAG;CAAU;AAC7F"}

package/dist/{raw-provider-sink-DIRCZjP_.js → raw-provider-sink-DYh4Bszw.js}

@@ -1,4 +1,4 @@
-import { u as redact } from "./redact-8fX8-uKr.js";
+import { u as redact } from "./redact-B7rt9Zot.js";
 import { promises } from "node:fs";
 import * as path from "node:path";
 //#region src/trace/raw-provider-sink.ts
@@ -161,4 +161,4 @@ function providerFromBaseUrl(baseUrl) {
 //#endregion
 export { providerFromBaseUrl as a, defaultProviderRedactor as i, InMemoryRawProviderSink as n, NoopRawProviderSink as r, FileSystemRawProviderSink as t };
 
-//# sourceMappingURL=raw-provider-sink-DIRCZjP_.js.map
+//# sourceMappingURL=raw-provider-sink-DYh4Bszw.js.map

package/dist/{raw-provider-sink-DIRCZjP_.js.map → raw-provider-sink-DYh4Bszw.js.map}

@@ -1 +1 @@
-{"version":3,"file":"raw-provider-sink-DIRCZjP_.js","names":["fs"],"sources":["../src/trace/raw-provider-sink.ts"],"sourcesContent":["/**\n * RawProviderSink — first-class persistence for the actual HTTP-level\n * request/response bodies of every LLM provider call.\n *\n * Why this is a separate sink from the structured `LlmSpan`:\n *\n *   - `LlmSpan` records the *intent* — model name, messages, output text,\n *     usage. It's what dashboards read; it's NOT enough for forensics.\n *   - When a downstream consumer reports \"the verifier used the wrong route\"\n *     or \"tokens look right but reasoning was missing,\" the only way to\n *     answer is the raw HTTP body. Span fields can lie (a proxy can echo\n *     a different `model` value than what actually answered); the raw\n *     response is ground truth.\n *\n * Default behaviour: opt-in. Pass `rawSink` to `LlmClientOptions` (or the\n * matrix runner / BuilderSession sets it up automatically) and every\n * request, response, and error is recorded — including retries, with the\n * attempt index attached so a flaky call's full event chain is recoverable.\n *\n * Redaction is enforced at sink time. The default redactor runs the redaction\n * core (`./redact`, default profile) over headers, bodies and the error\n * message: credential headers and fields are replaced by name, credential and\n * personal-data values by shape. Override via the sink constructor or the\n * per-call `redactor`. The `redactedFields` array on the persisted event lets\n * a reviewer see what was replaced without exposing the values.\n */\n\nimport { promises as fs } from 'node:fs'\nimport * as path from 'node:path'\nimport { redact } from './redact'\n\nexport type RawProviderDirection = 'request' | 'response' | 'error'\n\nexport interface RawProviderEvent {\n  /** Stable id. Generated by the sink if omitted. */\n  eventId: string\n  /** Trace context populated by `LlmClient` when the call is wrapped in a span. */\n  runId?: string\n  spanId?: string\n  /**\n   * Logical provider name. Free-form so callers can use whatever id matches\n   * their topology (`'openai'`, `'anthropic'`, `'tangle-router'`, …). When\n   * omitted, derived from `baseUrl` in `LlmClientOptions`.\n   */\n  provider: string\n  model: string\n  /** Endpoint path, e.g. `'/v1/chat/completions'`. */\n  endpoint: string\n  /** Base URL used for the call (already-normalised — no trailing slash). */\n  baseUrl: string\n  /** 0-indexed retry attempt. The first attempt is 0; a retried call gets 1, 2, … */\n  attemptIndex: number\n  direction: RawProviderDirection\n  /** Unix ms. */\n  timestamp: number\n  /** Wall-clock duration of the call leg. Set on `response` and `error` events; null on `request`. */\n  durationMs?: number\n  statusCode?: number\n  requestHeaders?: Record<string, string>\n  requestBody?: unknown\n  responseHeaders?: Record<string, string>\n  responseBody?: unknown\n  /** Set on `direction: 'error'` events. */\n  errorMessage?: string\n  /** JSON Pointers of the fields the redactor replaced ('/requestHeaders/Authorization', '/requestBody/apiKey', …). */\n  redactedFields: string[]\n}\n\nexport interface RawProviderSinkFilter {\n  runId?: string\n  spanId?: string\n  direction?: RawProviderDirection\n  attemptIndex?: number\n}\n\nexport interface RawProviderSink {\n  record(event: RawProviderEvent): Promise<void>\n  /** Optional listing — implementations that durably persist (file, db) should support this. */\n  list?(filter?: RawProviderSinkFilter): Promise<RawProviderEvent[]>\n  /** Optional teardown for backed implementations. */\n  close?(): Promise<void>\n}\n\nexport type ProviderRedactor = (event: RawProviderEvent) => RawProviderEvent\n\n/**\n * The default redactor: the redaction core over the parts of an event that\n * carry provider traffic. Every replaced path is added to `redactedFields`.\n */\nexport function defaultProviderRedactor(event: RawProviderEvent): RawProviderEvent {\n  const { requestHeaders, requestBody, responseHeaders, responseBody, errorMessage } = event\n  const { value, report } = redact({\n    requestHeaders,\n    requestBody,\n    responseHeaders,\n    responseBody,\n    errorMessage,\n  })\n  return {\n    ...event,\n    ...value,\n    redactedFields: [\n      ...(event.redactedFields ?? []),\n      ...report.findings.map((finding) => finding.path),\n    ],\n  }\n}\n\n// ── In-memory ────────────────────────────────────────────────────────────\n\nexport interface InMemoryRawProviderSinkOptions {\n  redactor?: ProviderRedactor\n}\n\nexport class InMemoryRawProviderSink implements RawProviderSink {\n  private events: RawProviderEvent[] = []\n  private redactor: ProviderRedactor\n\n  constructor(opts: InMemoryRawProviderSinkOptions = {}) {\n    this.redactor = opts.redactor ?? defaultProviderRedactor\n  }\n\n  async record(event: RawProviderEvent): Promise<void> {\n    this.events.push(this.redactor({ ...event, redactedFields: event.redactedFields ?? [] }))\n  }\n\n  async list(filter: RawProviderSinkFilter = {}): Promise<RawProviderEvent[]> {\n    return this.events.filter(\n      (e) =>\n        (filter.runId === undefined || e.runId === filter.runId) &&\n        (filter.spanId === undefined || e.spanId === filter.spanId) &&\n        (filter.direction === undefined || e.direction === filter.direction) &&\n        (filter.attemptIndex === undefined || e.attemptIndex === filter.attemptIndex),\n    )\n  }\n\n  size(): number {\n    return this.events.length\n  }\n}\n\nexport class NoopRawProviderSink implements RawProviderSink {\n  async record(): Promise<void> {\n    /* no-op */\n  }\n  /**\n   * Returns an empty array. Implemented so `assertRunCaptured` does not\n   * trip the `no_raw_sink` issue when a caller explicitly opts out of\n   * capture by passing this sink — opt-out is a deliberate choice, not a\n   * misconfiguration.\n   */\n  async list(): Promise<RawProviderEvent[]> {\n    return []\n  }\n}\n\n// ── Filesystem (NDJSON) ──────────────────────────────────────────────────\n\nexport interface FileSystemRawProviderSinkOptions {\n  /** Directory the NDJSON file is written into. Created if missing. */\n  dir: string\n  /** File name; default `'raw-provider-events.ndjson'`. */\n  fileName?: string\n  /** Bytes after which the writer rolls over to a new file (default 32 MiB). */\n  rollAtBytes?: number\n  redactor?: ProviderRedactor\n}\n\nexport class FileSystemRawProviderSink implements RawProviderSink {\n  private dir: string\n  private fileName: string\n  private rollAtBytes: number\n  private redactor: ProviderRedactor\n  private bytesWritten = 0\n  private rollIndex = 0\n  private initPromise: Promise<void> | null = null\n\n  constructor(opts: FileSystemRawProviderSinkOptions) {\n    this.dir = opts.dir\n    this.fileName = opts.fileName ?? 'raw-provider-events.ndjson'\n    this.rollAtBytes = opts.rollAtBytes ?? 32 * 1024 * 1024\n    this.redactor = opts.redactor ?? defaultProviderRedactor\n  }\n\n  private async ensureInit(): Promise<void> {\n    if (!this.initPromise) {\n      this.initPromise = fs.mkdir(this.dir, { recursive: true }).then(() => undefined)\n    }\n    await this.initPromise\n  }\n\n  private currentPath(): string {\n    if (this.rollIndex === 0) return path.join(this.dir, this.fileName)\n    return path.join(this.dir, `${this.fileName}.${this.rollIndex}`)\n  }\n\n  async record(event: RawProviderEvent): Promise<void> {\n    await this.ensureInit()\n    const redacted = this.redactor({ ...event, redactedFields: event.redactedFields ?? [] })\n    const line = `${JSON.stringify(redacted)}\\n`\n    if (this.bytesWritten + line.length > this.rollAtBytes && this.bytesWritten > 0) {\n      this.rollIndex += 1\n      this.bytesWritten = 0\n    }\n    await fs.appendFile(this.currentPath(), line, 'utf8')\n    this.bytesWritten += line.length\n  }\n\n  async list(filter: RawProviderSinkFilter = {}): Promise<RawProviderEvent[]> {\n    await this.ensureInit()\n    const out: RawProviderEvent[] = []\n    for (let i = 0; i <= this.rollIndex; i++) {\n      const file =\n        i === 0 ? path.join(this.dir, this.fileName) : path.join(this.dir, `${this.fileName}.${i}`)\n      let body: string\n      try {\n        body = await fs.readFile(file, 'utf8')\n      } catch (err) {\n        if ((err as NodeJS.ErrnoException).code === 'ENOENT') continue\n        throw err\n      }\n      for (const line of body.split('\\n')) {\n        if (!line) continue\n        const event = JSON.parse(line) as RawProviderEvent\n        if (filter.runId !== undefined && event.runId !== filter.runId) continue\n        if (filter.spanId !== undefined && event.spanId !== filter.spanId) continue\n        if (filter.direction !== undefined && event.direction !== filter.direction) continue\n        if (filter.attemptIndex !== undefined && event.attemptIndex !== filter.attemptIndex)\n          continue\n        out.push(event)\n      }\n    }\n    return out\n  }\n}\n\n// ── Helpers ──────────────────────────────────────────────────────────────\n\n/**\n * Best-effort provider id from a base URL. Falls back to the URL host when\n * none of the well-known patterns match.\n */\nexport function providerFromBaseUrl(baseUrl: string): string {\n  const lower = baseUrl.toLowerCase()\n  if (lower.includes('api.openai.com')) return 'openai'\n  if (lower.includes('api.anthropic.com')) return 'anthropic'\n  if (lower.includes('generativelanguage.googleapis.com')) return 'google'\n  if (lower.includes('api.together.ai') || lower.includes('api.together.xyz')) return 'together'\n  if (lower.includes('api.deepseek.com')) return 'deepseek'\n  if (lower.includes('router.tangle.tools')) return 'tangle-router'\n  if (lower.includes('api.litellm') || lower.includes('litellm')) return 'litellm'\n  try {\n    return new URL(baseUrl).host\n  } catch {\n    return baseUrl\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAyFA,SAAgB,wBAAwB,OAA2C;CACjF,MAAM,EAAE,gBAAgB,aAAa,iBAAiB,cAAc,iBAAiB;CACrF,MAAM,EAAE,OAAO,WAAW,OAAO;EAC/B;EACA;EACA;EACA;EACA;CACF,CAAC;CACD,OAAO;EACL,GAAG;EACH,GAAG;EACH,gBAAgB,CACd,GAAI,MAAM,kBAAkB,CAAC,GAC7B,GAAG,OAAO,SAAS,KAAK,YAAY,QAAQ,IAAI,CAClD;CACF;AACF;AAQA,IAAa,0BAAb,MAAgE;CAC9D,SAAqC,CAAC;CACtC;CAEA,YAAY,OAAuC,CAAC,GAAG;EACrD,KAAK,WAAW,KAAK,YAAY;CACnC;CAEA,MAAM,OAAO,OAAwC;EACnD,KAAK,OAAO,KAAK,KAAK,SAAS;GAAE,GAAG;GAAO,gBAAgB,MAAM,kBAAkB,CAAC;EAAE,CAAC,CAAC;CAC1F;CAEA,MAAM,KAAK,SAAgC,CAAC,GAAgC;EAC1E,OAAO,KAAK,OAAO,QAChB,OACE,OAAO,UAAU,KAAA,KAAa,EAAE,UAAU,OAAO,WACjD,OAAO,WAAW,KAAA,KAAa,EAAE,WAAW,OAAO,YACnD,OAAO,cAAc,KAAA,KAAa,EAAE,cAAc,OAAO,eACzD,OAAO,iBAAiB,KAAA,KAAa,EAAE,iBAAiB,OAAO,aACpE;CACF;CAEA,OAAe;EACb,OAAO,KAAK,OAAO;CACrB;AACF;AAEA,IAAa,sBAAb,MAA4D;CAC1D,MAAM,SAAwB,CAE9B;;;;;;;CAOA,MAAM,OAAoC;EACxC,OAAO,CAAC;CACV;AACF;AAcA,IAAa,4BAAb,MAAkE;CAChE;CACA;CACA;CACA;CACA,eAAuB;CACvB,YAAoB;CACpB,cAA4C;CAE5C,YAAY,MAAwC;EAClD,KAAK,MAAM,KAAK;EAChB,KAAK,WAAW,KAAK,YAAY;EACjC,KAAK,cAAc,KAAK,eAAe,KAAK,OAAO;EACnD,KAAK,WAAW,KAAK,YAAY;CACnC;CAEA,MAAc,aAA4B;EACxC,IAAI,CAAC,KAAK,aACR,KAAK,cAAcA,SAAG,MAAM,KAAK,KAAK,EAAE,WAAW,KAAK,CAAC,CAAC,CAAC,WAAW,KAAA,CAAS;EAEjF,MAAM,KAAK;CACb;CAEA,cAA8B;EAC5B,IAAI,KAAK,cAAc,GAAG,OAAO,KAAK,KAAK,KAAK,KAAK,KAAK,QAAQ;EAClE,OAAO,KAAK,KAAK,KAAK,KAAK,GAAG,KAAK,SAAS,GAAG,KAAK,WAAW;CACjE;CAEA,MAAM,OAAO,OAAwC;EACnD,MAAM,KAAK,WAAW;EACtB,MAAM,WAAW,KAAK,SAAS;GAAE,GAAG;GAAO,gBAAgB,MAAM,kBAAkB,CAAC;EAAE,CAAC;EACvF,MAAM,OAAO,GAAG,KAAK,UAAU,QAAQ,EAAE;EACzC,IAAI,KAAK,eAAe,KAAK,SAAS,KAAK,eAAe,KAAK,eAAe,GAAG;GAC/E,KAAK,aAAa;GAClB,KAAK,eAAe;EACtB;EACA,MAAMA,SAAG,WAAW,KAAK,YAAY,GAAG,MAAM,MAAM;EACpD,KAAK,gBAAgB,KAAK;CAC5B;CAEA,MAAM,KAAK,SAAgC,CAAC,GAAgC;EAC1E,MAAM,KAAK,WAAW;EACtB,MAAM,MAA0B,CAAC;EACjC,KAAK,IAAI,IAAI,GAAG,KAAK,KAAK,WAAW,KAAK;GACxC,MAAM,OACJ,MAAM,IAAI,KAAK,KAAK,KAAK,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,KAAK,KAAK,GAAG,KAAK,SAAS,GAAG,GAAG;GAC5F,IAAI;GACJ,IAAI;IACF,OAAO,MAAMA,SAAG,SAAS,MAAM,MAAM;GACvC,SAAS,KAAK;IACZ,IAAK,IAA8B,SAAS,UAAU;IACtD,MAAM;GACR;GACA,KAAK,MAAM,QAAQ,KAAK,MAAM,IAAI,GAAG;IACnC,IAAI,CAAC,MAAM;IACX,MAAM,QAAQ,KAAK,MAAM,IAAI;IAC7B,IAAI,OAAO,UAAU,KAAA,KAAa,MAAM,UAAU,OAAO,OAAO;IAChE,IAAI,OAAO,WAAW,KAAA,KAAa,MAAM,WAAW,OAAO,QAAQ;IACnE,IAAI,OAAO,cAAc,KAAA,KAAa,MAAM,cAAc,OAAO,WAAW;IAC5E,IAAI,OAAO,iBAAiB,KAAA,KAAa,MAAM,iBAAiB,OAAO,cACrE;IACF,IAAI,KAAK,KAAK;GAChB;EACF;EACA,OAAO;CACT;AACF;;;;;AAQA,SAAgB,oBAAoB,SAAyB;CAC3D,MAAM,QAAQ,QAAQ,YAAY;CAClC,IAAI,MAAM,SAAS,gBAAgB,GAAG,OAAO;CAC7C,IAAI,MAAM,SAAS,mBAAmB,GAAG,OAAO;CAChD,IAAI,MAAM,SAAS,mCAAmC,GAAG,OAAO;CAChE,IAAI,MAAM,SAAS,iBAAiB,KAAK,MAAM,SAAS,kBAAkB,GAAG,OAAO;CACpF,IAAI,MAAM,SAAS,kBAAkB,GAAG,OAAO;CAC/C,IAAI,MAAM,SAAS,qBAAqB,GAAG,OAAO;CAClD,IAAI,MAAM,SAAS,aAAa,KAAK,MAAM,SAAS,SAAS,GAAG,OAAO;CACvE,IAAI;EACF,OAAO,IAAI,IAAI,OAAO,CAAC,CAAC;CAC1B,QAAQ;EACN,OAAO;CACT;AACF"}
+{"version":3,"file":"raw-provider-sink-DYh4Bszw.js","names":["fs"],"sources":["../src/trace/raw-provider-sink.ts"],"sourcesContent":["/**\n * RawProviderSink — first-class persistence for the actual HTTP-level\n * request/response bodies of every LLM provider call.\n *\n * Why this is a separate sink from the structured `LlmSpan`:\n *\n *   - `LlmSpan` records the *intent* — model name, messages, output text,\n *     usage. It's what dashboards read; it's NOT enough for forensics.\n *   - When a downstream consumer reports \"the verifier used the wrong route\"\n *     or \"tokens look right but reasoning was missing,\" the only way to\n *     answer is the raw HTTP body. Span fields can lie (a proxy can echo\n *     a different `model` value than what actually answered); the raw\n *     response is ground truth.\n *\n * Default behaviour: opt-in. Pass `rawSink` to `LlmClientOptions` (or the\n * matrix runner / BuilderSession sets it up automatically) and every\n * request, response, and error is recorded — including retries, with the\n * attempt index attached so a flaky call's full event chain is recoverable.\n *\n * Redaction is enforced at sink time. The default redactor runs the redaction\n * core (`./redact`, default profile) over headers, bodies and the error\n * message: credential headers and fields are replaced by name, credential and\n * personal-data values by shape. Override via the sink constructor or the\n * per-call `redactor`. The `redactedFields` array on the persisted event lets\n * a reviewer see what was replaced without exposing the values.\n */\n\nimport { promises as fs } from 'node:fs'\nimport * as path from 'node:path'\nimport { redact } from './redact'\n\nexport type RawProviderDirection = 'request' | 'response' | 'error'\n\nexport interface RawProviderEvent {\n  /** Stable id. Generated by the sink if omitted. */\n  eventId: string\n  /** Trace context populated by `LlmClient` when the call is wrapped in a span. */\n  runId?: string\n  spanId?: string\n  /**\n   * Logical provider name. Free-form so callers can use whatever id matches\n   * their topology (`'openai'`, `'anthropic'`, `'tangle-router'`, …). When\n   * omitted, derived from `baseUrl` in `LlmClientOptions`.\n   */\n  provider: string\n  model: string\n  /** Endpoint path, e.g. `'/v1/chat/completions'`. */\n  endpoint: string\n  /** Base URL used for the call (already-normalised — no trailing slash). */\n  baseUrl: string\n  /** 0-indexed retry attempt. The first attempt is 0; a retried call gets 1, 2, … */\n  attemptIndex: number\n  direction: RawProviderDirection\n  /** Unix ms. */\n  timestamp: number\n  /** Wall-clock duration of the call leg. Set on `response` and `error` events; null on `request`. */\n  durationMs?: number\n  statusCode?: number\n  requestHeaders?: Record<string, string>\n  requestBody?: unknown\n  responseHeaders?: Record<string, string>\n  responseBody?: unknown\n  /** Set on `direction: 'error'` events. */\n  errorMessage?: string\n  /** JSON Pointers of the fields the redactor replaced ('/requestHeaders/Authorization', '/requestBody/apiKey', …). */\n  redactedFields: string[]\n}\n\nexport interface RawProviderSinkFilter {\n  runId?: string\n  spanId?: string\n  direction?: RawProviderDirection\n  attemptIndex?: number\n}\n\nexport interface RawProviderSink {\n  record(event: RawProviderEvent): Promise<void>\n  /** Optional listing — implementations that durably persist (file, db) should support this. */\n  list?(filter?: RawProviderSinkFilter): Promise<RawProviderEvent[]>\n  /** Optional teardown for backed implementations. */\n  close?(): Promise<void>\n}\n\nexport type ProviderRedactor = (event: RawProviderEvent) => RawProviderEvent\n\n/**\n * The default redactor: the redaction core over the parts of an event that\n * carry provider traffic. Every replaced path is added to `redactedFields`.\n */\nexport function defaultProviderRedactor(event: RawProviderEvent): RawProviderEvent {\n  const { requestHeaders, requestBody, responseHeaders, responseBody, errorMessage } = event\n  const { value, report } = redact({\n    requestHeaders,\n    requestBody,\n    responseHeaders,\n    responseBody,\n    errorMessage,\n  })\n  return {\n    ...event,\n    ...value,\n    redactedFields: [\n      ...(event.redactedFields ?? []),\n      ...report.findings.map((finding) => finding.path),\n    ],\n  }\n}\n\n// ── In-memory ────────────────────────────────────────────────────────────\n\nexport interface InMemoryRawProviderSinkOptions {\n  redactor?: ProviderRedactor\n}\n\nexport class InMemoryRawProviderSink implements RawProviderSink {\n  private events: RawProviderEvent[] = []\n  private redactor: ProviderRedactor\n\n  constructor(opts: InMemoryRawProviderSinkOptions = {}) {\n    this.redactor = opts.redactor ?? defaultProviderRedactor\n  }\n\n  async record(event: RawProviderEvent): Promise<void> {\n    this.events.push(this.redactor({ ...event, redactedFields: event.redactedFields ?? [] }))\n  }\n\n  async list(filter: RawProviderSinkFilter = {}): Promise<RawProviderEvent[]> {\n    return this.events.filter(\n      (e) =>\n        (filter.runId === undefined || e.runId === filter.runId) &&\n        (filter.spanId === undefined || e.spanId === filter.spanId) &&\n        (filter.direction === undefined || e.direction === filter.direction) &&\n        (filter.attemptIndex === undefined || e.attemptIndex === filter.attemptIndex),\n    )\n  }\n\n  size(): number {\n    return this.events.length\n  }\n}\n\nexport class NoopRawProviderSink implements RawProviderSink {\n  async record(): Promise<void> {\n    /* no-op */\n  }\n  /**\n   * Returns an empty array. Implemented so `assertRunCaptured` does not\n   * trip the `no_raw_sink` issue when a caller explicitly opts out of\n   * capture by passing this sink — opt-out is a deliberate choice, not a\n   * misconfiguration.\n   */\n  async list(): Promise<RawProviderEvent[]> {\n    return []\n  }\n}\n\n// ── Filesystem (NDJSON) ──────────────────────────────────────────────────\n\nexport interface FileSystemRawProviderSinkOptions {\n  /** Directory the NDJSON file is written into. Created if missing. */\n  dir: string\n  /** File name; default `'raw-provider-events.ndjson'`. */\n  fileName?: string\n  /** Bytes after which the writer rolls over to a new file (default 32 MiB). */\n  rollAtBytes?: number\n  redactor?: ProviderRedactor\n}\n\nexport class FileSystemRawProviderSink implements RawProviderSink {\n  private dir: string\n  private fileName: string\n  private rollAtBytes: number\n  private redactor: ProviderRedactor\n  private bytesWritten = 0\n  private rollIndex = 0\n  private initPromise: Promise<void> | null = null\n\n  constructor(opts: FileSystemRawProviderSinkOptions) {\n    this.dir = opts.dir\n    this.fileName = opts.fileName ?? 'raw-provider-events.ndjson'\n    this.rollAtBytes = opts.rollAtBytes ?? 32 * 1024 * 1024\n    this.redactor = opts.redactor ?? defaultProviderRedactor\n  }\n\n  private async ensureInit(): Promise<void> {\n    if (!this.initPromise) {\n      this.initPromise = fs.mkdir(this.dir, { recursive: true }).then(() => undefined)\n    }\n    await this.initPromise\n  }\n\n  private currentPath(): string {\n    if (this.rollIndex === 0) return path.join(this.dir, this.fileName)\n    return path.join(this.dir, `${this.fileName}.${this.rollIndex}`)\n  }\n\n  async record(event: RawProviderEvent): Promise<void> {\n    await this.ensureInit()\n    const redacted = this.redactor({ ...event, redactedFields: event.redactedFields ?? [] })\n    const line = `${JSON.stringify(redacted)}\\n`\n    if (this.bytesWritten + line.length > this.rollAtBytes && this.bytesWritten > 0) {\n      this.rollIndex += 1\n      this.bytesWritten = 0\n    }\n    await fs.appendFile(this.currentPath(), line, 'utf8')\n    this.bytesWritten += line.length\n  }\n\n  async list(filter: RawProviderSinkFilter = {}): Promise<RawProviderEvent[]> {\n    await this.ensureInit()\n    const out: RawProviderEvent[] = []\n    for (let i = 0; i <= this.rollIndex; i++) {\n      const file =\n        i === 0 ? path.join(this.dir, this.fileName) : path.join(this.dir, `${this.fileName}.${i}`)\n      let body: string\n      try {\n        body = await fs.readFile(file, 'utf8')\n      } catch (err) {\n        if ((err as NodeJS.ErrnoException).code === 'ENOENT') continue\n        throw err\n      }\n      for (const line of body.split('\\n')) {\n        if (!line) continue\n        const event = JSON.parse(line) as RawProviderEvent\n        if (filter.runId !== undefined && event.runId !== filter.runId) continue\n        if (filter.spanId !== undefined && event.spanId !== filter.spanId) continue\n        if (filter.direction !== undefined && event.direction !== filter.direction) continue\n        if (filter.attemptIndex !== undefined && event.attemptIndex !== filter.attemptIndex)\n          continue\n        out.push(event)\n      }\n    }\n    return out\n  }\n}\n\n// ── Helpers ──────────────────────────────────────────────────────────────\n\n/**\n * Best-effort provider id from a base URL. Falls back to the URL host when\n * none of the well-known patterns match.\n */\nexport function providerFromBaseUrl(baseUrl: string): string {\n  const lower = baseUrl.toLowerCase()\n  if (lower.includes('api.openai.com')) return 'openai'\n  if (lower.includes('api.anthropic.com')) return 'anthropic'\n  if (lower.includes('generativelanguage.googleapis.com')) return 'google'\n  if (lower.includes('api.together.ai') || lower.includes('api.together.xyz')) return 'together'\n  if (lower.includes('api.deepseek.com')) return 'deepseek'\n  if (lower.includes('router.tangle.tools')) return 'tangle-router'\n  if (lower.includes('api.litellm') || lower.includes('litellm')) return 'litellm'\n  try {\n    return new URL(baseUrl).host\n  } catch {\n    return baseUrl\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAyFA,SAAgB,wBAAwB,OAA2C;CACjF,MAAM,EAAE,gBAAgB,aAAa,iBAAiB,cAAc,iBAAiB;CACrF,MAAM,EAAE,OAAO,WAAW,OAAO;EAC/B;EACA;EACA;EACA;EACA;CACF,CAAC;CACD,OAAO;EACL,GAAG;EACH,GAAG;EACH,gBAAgB,CACd,GAAI,MAAM,kBAAkB,CAAC,GAC7B,GAAG,OAAO,SAAS,KAAK,YAAY,QAAQ,IAAI,CAClD;CACF;AACF;AAQA,IAAa,0BAAb,MAAgE;CAC9D,SAAqC,CAAC;CACtC;CAEA,YAAY,OAAuC,CAAC,GAAG;EACrD,KAAK,WAAW,KAAK,YAAY;CACnC;CAEA,MAAM,OAAO,OAAwC;EACnD,KAAK,OAAO,KAAK,KAAK,SAAS;GAAE,GAAG;GAAO,gBAAgB,MAAM,kBAAkB,CAAC;EAAE,CAAC,CAAC;CAC1F;CAEA,MAAM,KAAK,SAAgC,CAAC,GAAgC;EAC1E,OAAO,KAAK,OAAO,QAChB,OACE,OAAO,UAAU,KAAA,KAAa,EAAE,UAAU,OAAO,WACjD,OAAO,WAAW,KAAA,KAAa,EAAE,WAAW,OAAO,YACnD,OAAO,cAAc,KAAA,KAAa,EAAE,cAAc,OAAO,eACzD,OAAO,iBAAiB,KAAA,KAAa,EAAE,iBAAiB,OAAO,aACpE;CACF;CAEA,OAAe;EACb,OAAO,KAAK,OAAO;CACrB;AACF;AAEA,IAAa,sBAAb,MAA4D;CAC1D,MAAM,SAAwB,CAE9B;;;;;;;CAOA,MAAM,OAAoC;EACxC,OAAO,CAAC;CACV;AACF;AAcA,IAAa,4BAAb,MAAkE;CAChE;CACA;CACA;CACA;CACA,eAAuB;CACvB,YAAoB;CACpB,cAA4C;CAE5C,YAAY,MAAwC;EAClD,KAAK,MAAM,KAAK;EAChB,KAAK,WAAW,KAAK,YAAY;EACjC,KAAK,cAAc,KAAK,eAAe,KAAK,OAAO;EACnD,KAAK,WAAW,KAAK,YAAY;CACnC;CAEA,MAAc,aAA4B;EACxC,IAAI,CAAC,KAAK,aACR,KAAK,cAAcA,SAAG,MAAM,KAAK,KAAK,EAAE,WAAW,KAAK,CAAC,CAAC,CAAC,WAAW,KAAA,CAAS;EAEjF,MAAM,KAAK;CACb;CAEA,cAA8B;EAC5B,IAAI,KAAK,cAAc,GAAG,OAAO,KAAK,KAAK,KAAK,KAAK,KAAK,QAAQ;EAClE,OAAO,KAAK,KAAK,KAAK,KAAK,GAAG,KAAK,SAAS,GAAG,KAAK,WAAW;CACjE;CAEA,MAAM,OAAO,OAAwC;EACnD,MAAM,KAAK,WAAW;EACtB,MAAM,WAAW,KAAK,SAAS;GAAE,GAAG;GAAO,gBAAgB,MAAM,kBAAkB,CAAC;EAAE,CAAC;EACvF,MAAM,OAAO,GAAG,KAAK,UAAU,QAAQ,EAAE;EACzC,IAAI,KAAK,eAAe,KAAK,SAAS,KAAK,eAAe,KAAK,eAAe,GAAG;GAC/E,KAAK,aAAa;GAClB,KAAK,eAAe;EACtB;EACA,MAAMA,SAAG,WAAW,KAAK,YAAY,GAAG,MAAM,MAAM;EACpD,KAAK,gBAAgB,KAAK;CAC5B;CAEA,MAAM,KAAK,SAAgC,CAAC,GAAgC;EAC1E,MAAM,KAAK,WAAW;EACtB,MAAM,MAA0B,CAAC;EACjC,KAAK,IAAI,IAAI,GAAG,KAAK,KAAK,WAAW,KAAK;GACxC,MAAM,OACJ,MAAM,IAAI,KAAK,KAAK,KAAK,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,KAAK,KAAK,GAAG,KAAK,SAAS,GAAG,GAAG;GAC5F,IAAI;GACJ,IAAI;IACF,OAAO,MAAMA,SAAG,SAAS,MAAM,MAAM;GACvC,SAAS,KAAK;IACZ,IAAK,IAA8B,SAAS,UAAU;IACtD,MAAM;GACR;GACA,KAAK,MAAM,QAAQ,KAAK,MAAM,IAAI,GAAG;IACnC,IAAI,CAAC,MAAM;IACX,MAAM,QAAQ,KAAK,MAAM,IAAI;IAC7B,IAAI,OAAO,UAAU,KAAA,KAAa,MAAM,UAAU,OAAO,OAAO;IAChE,IAAI,OAAO,WAAW,KAAA,KAAa,MAAM,WAAW,OAAO,QAAQ;IACnE,IAAI,OAAO,cAAc,KAAA,KAAa,MAAM,cAAc,OAAO,WAAW;IAC5E,IAAI,OAAO,iBAAiB,KAAA,KAAa,MAAM,iBAAiB,OAAO,cACrE;IACF,IAAI,KAAK,KAAK;GAChB;EACF;EACA,OAAO;CACT;AACF;;;;;AAQA,SAAgB,oBAAoB,SAAyB;CAC3D,MAAM,QAAQ,QAAQ,YAAY;CAClC,IAAI,MAAM,SAAS,gBAAgB,GAAG,OAAO;CAC7C,IAAI,MAAM,SAAS,mBAAmB,GAAG,OAAO;CAChD,IAAI,MAAM,SAAS,mCAAmC,GAAG,OAAO;CAChE,IAAI,MAAM,SAAS,iBAAiB,KAAK,MAAM,SAAS,kBAAkB,GAAG,OAAO;CACpF,IAAI,MAAM,SAAS,kBAAkB,GAAG,OAAO;CAC/C,IAAI,MAAM,SAAS,qBAAqB,GAAG,OAAO;CAClD,IAAI,MAAM,SAAS,aAAa,KAAK,MAAM,SAAS,SAAS,GAAG,OAAO;CACvE,IAAI;EACF,OAAO,IAAI,IAAI,OAAO,CAAC,CAAC;CAC1B,QAAQ;EACN,OAAO;CACT;AACF"}