@tangle-network/agent-eval 0.108.1 → 0.110.0

package/dist/chunk-PLOMR3HP.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/failure-taxonomy.ts","../src/tool-use-metrics.ts","../src/trajectory.ts","../src/baseline.ts"],"sourcesContent":["/**\n * Failure taxonomy — canonical classes + a default classifier.\n *\n * Every failed run should end up in a named class. The classifier here\n * is rule-based (fast, deterministic); an LLM fallback can be added by\n * the consumer for novel cases and trained into the rule base over time.\n *\n * Consumers call `classifyFailure(run, spans, events)` and persist the\n * returned class as `Run.outcome.failureClass`.\n */\n\nimport type { FailureClass, Run, Span, TraceEvent } from './trace/schema'\nimport { FAILURE_CLASSES } from './trace/schema'\n\nexport { FAILURE_CLASSES, type FailureClass }\n\nexport interface FailureContext {\n  run: Run\n  spans: Span[]\n  events: TraceEvent[]\n}\n\nexport interface FailureClassification {\n  failureClass: FailureClass\n  reason: string\n  triggerSpanId?: string\n  triggerEventId?: string\n}\n\n/** Ordered rules — first match wins. */\nexport interface FailureRule {\n  id: string\n  match: (ctx: FailureContext) => {\n    failureClass: FailureClass\n    reason: string\n    triggerSpanId?: string\n    triggerEventId?: string\n  } | null\n}\n\nexport const DEFAULT_RULES: FailureRule[] = [\n  // Outcome already named? Respect it.\n  {\n    id: 'explicit-outcome',\n    match: ({ run }) => {\n      const fc = run.outcome?.failureClass\n      if (fc && fc !== 'unknown')\n        return { failureClass: fc, reason: 'outcome.failureClass set explicitly' }\n      return null\n    },\n  },\n  {\n    id: 'knowledge-readiness-blocked',\n    match: ({ events }) => {\n      const event = events.find(\n        (e) =>\n          e.kind === 'custom' &&\n          e.payload.kind === 'readiness_scored' &&\n          e.payload.passed === false,\n      )\n      return event\n        ? {\n            failureClass: 'knowledge_readiness_blocked',\n            reason: 'knowledge readiness report blocked execution',\n            triggerEventId: event.eventId,\n          }\n        : null\n    },\n  },\n  {\n    id: 'bad-integration-manifest',\n    match: ({ events }) => {\n      const event = events.find(\n        (e) =>\n          e.kind === 'custom' &&\n          ((e.payload.kind === 'integration_manifest_validated' && e.payload.valid === false) ||\n            (e.payload.kind === 'integration_invoke_failed' &&\n              e.payload.code === 'manifest_invalid')),\n      )\n      return event\n        ? {\n            failureClass: 'bad_integration_manifest',\n            reason: 'integration manifest validation failed before launch',\n            triggerEventId: event.eventId,\n          }\n        : null\n    },\n  },\n  {\n    id: 'missing-integration-connection',\n    match: ({ events }) => {\n      const event = events.find(\n        (e) =>\n          e.kind === 'custom' &&\n          e.payload.kind === 'integration_manifest_resolved' &&\n          hasResolutionStatus(e.payload, 'missing_connection'),\n      )\n      return event\n        ? {\n            failureClass: 'missing_integration_connection',\n            reason: 'required integration connection was missing',\n            triggerEventId: event.eventId,\n          }\n        : null\n    },\n  },\n  {\n    id: 'missing-integration-scope',\n    match: ({ events }) => {\n      const event = events.find(\n        (e) =>\n          e.kind === 'custom' &&\n          ((e.payload.kind === 'integration_manifest_resolved' && hasMissingScopes(e.payload)) ||\n            (e.payload.kind === 'integration_invoke_failed' && e.payload.code === 'scope_denied')),\n      )\n      return event\n        ? {\n            failureClass: 'missing_integration_scope',\n            reason: 'integration grant or connection lacks required scopes',\n            triggerEventId: event.eventId,\n          }\n        : null\n    },\n  },\n  {\n    id: 'integration-approval-required',\n    match: ({ events }) => {\n      const event = events.find(\n        (e) =>\n          e.kind === 'custom' &&\n          ((e.payload.kind === 'integration_invoke' && e.payload.status === 'approval_required') ||\n            (e.payload.kind === 'integration_invoke_failed' &&\n              e.payload.code === 'approval_required') ||\n            e.payload.kind === 'integration_approval_required'),\n      )\n      return event\n        ? {\n            failureClass: 'integration_approval_required',\n            reason: 'integration write paused for user approval',\n            triggerEventId: event.eventId,\n          }\n        : null\n    },\n  },\n  {\n    id: 'integration-auth-expired',\n    match: ({ events }) => {\n      const event = events.find(\n        (e) =>\n          e.kind === 'custom' &&\n          e.payload.kind === 'integration_invoke_failed' &&\n          (e.payload.code === 'auth_expired' ||\n            e.payload.code === 'connection_not_active' ||\n            e.payload.code === 'capability_expired' ||\n            e.payload.status === 'expired'),\n      )\n      return event\n        ? {\n            failureClass: 'integration_auth_expired',\n            reason: 'integration connection or capability expired',\n            triggerEventId: event.eventId,\n          }\n        : null\n    },\n  },\n  {\n    id: 'unsafe-integration-write-denied',\n    match: ({ events }) => {\n      const event = events.find(\n        (e) =>\n          e.kind === 'custom' &&\n          e.payload.kind === 'integration_invoke_failed' &&\n          (e.payload.code === 'unsafe_write_denied' ||\n            e.payload.code === 'policy_denied' ||\n            e.payload.code === 'action_denied'),\n      )\n      return event\n        ? {\n            failureClass: 'unsafe_integration_write_denied',\n            reason: 'integration write was denied by policy or capability scope',\n            triggerEventId: event.eventId,\n          }\n        : null\n    },\n  },\n  {\n    id: 'integration-provider-failure',\n    match: ({ events }) => {\n      const event = events.find(\n        (e) =>\n          e.kind === 'custom' &&\n          e.payload.kind === 'integration_invoke_failed' &&\n          ![\n            'scope_denied',\n            'approval_required',\n            'auth_expired',\n            'connection_not_active',\n            'capability_expired',\n            'unsafe_write_denied',\n            'policy_denied',\n            'action_denied',\n            'manifest_invalid',\n          ].includes(String(e.payload.code)),\n      )\n      return event\n        ? {\n            failureClass: 'integration_provider_failure',\n            reason: 'integration provider invocation failed',\n            triggerEventId: event.eventId,\n          }\n        : null\n    },\n  },\n  {\n    id: 'missing-credentials',\n    match: ({ events }) => {\n      const event = events.find(\n        (e) =>\n          e.kind === 'custom' &&\n          e.payload.kind === 'knowledge_gap' &&\n          e.payload.category === 'credential_or_secret',\n      )\n      return event\n        ? {\n            failureClass: 'missing_credentials',\n            reason: 'required credential or secret was missing',\n            triggerEventId: event.eventId,\n          }\n        : null\n    },\n  },\n  {\n    id: 'bad-retrieval',\n    match: ({ run, spans }) => {\n      if (run.outcome?.pass !== false) return null\n      const retrieval = spans.find(\n        (s) =>\n          s.kind === 'retrieval' && (s.hits.length === 0 || s.hits.every((hit) => hit.score <= 0)),\n      )\n      return retrieval\n        ? {\n            failureClass: 'bad_retrieval',\n            reason: 'retrieval returned no useful hits for a failed run',\n            triggerSpanId: retrieval.spanId,\n          }\n        : null\n    },\n  },\n  {\n    id: 'insufficient-evidence',\n    match: ({ events }) => {\n      const event = events.find(\n        (e) =>\n          e.kind === 'custom' &&\n          e.payload.kind === 'knowledge_gap' &&\n          e.payload.reason === 'insufficient_evidence',\n      )\n      return event\n        ? {\n            failureClass: 'insufficient_evidence',\n            reason: 'task proceeded with insufficient supporting evidence',\n            triggerEventId: event.eventId,\n          }\n        : null\n    },\n  },\n  {\n    id: 'contradictory-evidence',\n    match: ({ events }) => {\n      const event = events.find(\n        (e) =>\n          e.kind === 'custom' &&\n          e.payload.kind === 'knowledge_gap' &&\n          e.payload.reason === 'contradictory_evidence',\n      )\n      return event\n        ? {\n            failureClass: 'contradictory_evidence',\n            reason: 'supporting evidence contradicted itself',\n            triggerEventId: event.eventId,\n          }\n        : null\n    },\n  },\n  // Budget breach events\n  {\n    id: 'budget-breach',\n    match: ({ events }) => {\n      const breach = events.find((e) => e.kind === 'budget_breach')\n      return breach\n        ? {\n            failureClass: 'budget_exceeded',\n            reason: `budget breached on ${breach.payload.dimension ?? 'unknown dimension'}`,\n            triggerEventId: breach.eventId,\n          }\n        : null\n    },\n  },\n  // Policy violations\n  {\n    id: 'policy-violation',\n    match: ({ events }) => {\n      const e = events.find((x) => x.kind === 'policy_violation')\n      return e\n        ? {\n            failureClass: 'policy_violation',\n            reason: 'policy_violation event emitted',\n            triggerEventId: e.eventId,\n          }\n        : null\n    },\n  },\n  // Sandbox non-zero exit code\n  {\n    id: 'sandbox-failure',\n    match: ({ spans }) => {\n      const s = spans.find(\n        (x) => x.kind === 'sandbox' && typeof x.exitCode === 'number' && x.exitCode !== 0,\n      )\n      if (!s) return null\n      return {\n        failureClass: 'sandbox_failure',\n        reason: `sandbox exited ${(s as Extract<Span, { kind: 'sandbox' }>).exitCode}`,\n        triggerSpanId: s.spanId,\n      }\n    },\n  },\n  // Timeout: run aborted by external signal\n  {\n    id: 'timeout',\n    match: ({ run, events }) => {\n      if (run.status !== 'aborted') return null\n      const hasTimeout = events.some(\n        (e) =>\n          e.kind === 'error' &&\n          String(e.payload.reason ?? '')\n            .toLowerCase()\n            .includes('timeout'),\n      )\n      const note = (run.outcome?.notes ?? '').toLowerCase()\n      if (hasTimeout || note.includes('timeout') || note.includes('deadline')) {\n        return { failureClass: 'timeout', reason: 'timeout signal observed' }\n      }\n      return null\n    },\n  },\n  // Tool recovery failure: many consecutive tool errors on the same tool\n  {\n    id: 'tool-recovery-failure',\n    match: ({ spans }) => {\n      const tools = spans.filter((s) => s.kind === 'tool')\n      const byTool = new Map<string, Span[]>()\n      for (const t of tools) {\n        const name = (t as Extract<Span, { kind: 'tool' }>).toolName\n        const arr = byTool.get(name) ?? []\n        arr.push(t)\n        byTool.set(name, arr)\n      }\n      for (const [name, arr] of byTool) {\n        const errs = arr.filter((s) => s.status === 'error')\n        if (errs.length >= 3 && errs.length === arr.length) {\n          return {\n            failureClass: 'tool_recovery_failure',\n            reason: `${errs.length} consecutive errors on tool \"${name}\"`,\n            triggerSpanId: errs[errs.length - 1]!.spanId,\n          }\n        }\n      }\n      return null\n    },\n  },\n  // Tool selection error: the run failed and agent called zero tools despite having them\n  {\n    id: 'tool-selection-error',\n    match: ({ run, spans }) => {\n      if (run.outcome?.pass !== false) return null\n      const hasToolsAvailable = spans.some(\n        (s) =>\n          s.kind === 'agent' &&\n          (s.attributes?.toolsAvailable as number | undefined) !== undefined &&\n          (s.attributes?.toolsAvailable as number) > 0,\n      )\n      const tools = spans.filter((s) => s.kind === 'tool')\n      if (hasToolsAvailable && tools.length === 0) {\n        return {\n          failureClass: 'tool_selection_error',\n          reason: 'tools were available but none were called',\n        }\n      }\n      return null\n    },\n  },\n  // Format drift: scored by a judge with dimension='format' below threshold\n  {\n    id: 'format-drift',\n    match: ({ spans }) => {\n      const judge = spans.find(\n        (s) =>\n          s.kind === 'judge' &&\n          (s as Extract<Span, { kind: 'judge' }>).dimension === 'format' &&\n          (s as Extract<Span, { kind: 'judge' }>).score < 0.5,\n      )\n      return judge\n        ? {\n            failureClass: 'format_drift',\n            reason: 'format judge scored below 0.5',\n            triggerSpanId: judge.spanId,\n          }\n        : null\n    },\n  },\n]\n\nfunction hasResolutionStatus(payload: Record<string, unknown>, status: string): boolean {\n  if (status === 'missing_connection' && stringArray(payload.missingConnections).length > 0)\n    return true\n  return resolutionItems(payload).some((item) => item.status === status)\n}\n\nfunction hasMissingScopes(payload: Record<string, unknown>): boolean {\n  if (stringArray(payload.missingScopes).length > 0) return true\n  return resolutionItems(payload).some(\n    (item) => Array.isArray(item.missingScopes) && item.missingScopes.length > 0,\n  )\n}\n\nfunction resolutionItems(payload: Record<string, unknown>): Array<Record<string, unknown>> {\n  return [\n    ...records(payload.missing),\n    ...records(payload.optionalMissing),\n    ...records(payload.ready),\n  ]\n}\n\nfunction records(value: unknown): Array<Record<string, unknown>> {\n  if (!Array.isArray(value)) return []\n  return value.filter(\n    (item): item is Record<string, unknown> =>\n      Boolean(item) && typeof item === 'object' && !Array.isArray(item),\n  )\n}\n\nfunction stringArray(value: unknown): string[] {\n  return Array.isArray(value)\n    ? value.filter((item): item is string => typeof item === 'string')\n    : []\n}\n\n/** Classify the failure mode of a run using an ordered rule list. */\nexport function classifyFailure(\n  ctx: FailureContext,\n  rules: FailureRule[] = DEFAULT_RULES,\n): FailureClassification {\n  if (ctx.run.outcome?.pass !== false && ctx.run.status === 'completed') {\n    return { failureClass: 'success', reason: 'run completed with pass=true (or no explicit fail)' }\n  }\n  for (const rule of rules) {\n    const hit = rule.match(ctx)\n    if (hit) return hit\n  }\n  return { failureClass: 'unknown', reason: 'no rule matched; run failed for unclassified reason' }\n}\n","/**\n * Tool-use metrics — derived purely from trace data.\n *\n * No scoring assumptions: consumers supply optional ground-truth tool\n * selections per turn + optional \"information used downstream\" signals.\n * Without those, we still compute descriptive metrics (error rate,\n * retry rate, duplicate-call rate) that are useful on their own.\n */\n\nimport { argHash, groupBy, toolSpans } from './trace/query'\nimport type { Span } from './trace/schema'\nimport type { TraceStore } from './trace/store'\n\nexport interface ToolUseMetrics {\n  runId: string\n  totalCalls: number\n  byTool: Record<string, ToolStats>\n  errorRate: number\n  /** Ratio of calls with identical (toolName, argHash) already seen earlier in the same run. */\n  duplicateRate: number\n  /** Ratio of error calls followed by ≥1 retry on same tool. */\n  retryRate: number\n  /** Optional: of the calls agent made, fraction the evaluator marked as \"correct selection\". */\n  selectionAccuracy?: number\n}\n\nexport interface ToolStats {\n  calls: number\n  errors: number\n  avgLatencyMs: number\n  duplicates: number\n}\n\nexport interface ToolUseOptions {\n  /** Map of spanId → whether the evaluator judged the tool selection correct. Optional. */\n  selectionLabels?: Record<string, boolean>\n}\n\nexport async function computeToolUseMetrics(\n  store: TraceStore,\n  runId: string,\n  options: ToolUseOptions = {},\n): Promise<ToolUseMetrics> {\n  const tools = await toolSpans(store, runId)\n  if (tools.length === 0) {\n    return { runId, totalCalls: 0, byTool: {}, errorRate: 0, duplicateRate: 0, retryRate: 0 }\n  }\n\n  const byTool: Record<string, ToolStats> = {}\n  let totalErrors = 0\n  let totalDuplicates = 0\n  const sortedTools = [...tools].sort((a, b) => a.startedAt - b.startedAt)\n  const seenSignatures = new Set<string>()\n\n  // duplicate detection + per-tool aggregation\n  for (const t of sortedTools) {\n    byTool[t.toolName] ??= { calls: 0, errors: 0, avgLatencyMs: 0, duplicates: 0 }\n    const stat = byTool[t.toolName]!\n    stat.calls += 1\n    if (t.status === 'error') {\n      stat.errors += 1\n      totalErrors += 1\n    }\n    if (typeof t.latencyMs === 'number') stat.avgLatencyMs += t.latencyMs\n    const sig = `${t.toolName}|${argHash(t.args)}`\n    if (seenSignatures.has(sig)) {\n      stat.duplicates += 1\n      totalDuplicates += 1\n    }\n    seenSignatures.add(sig)\n  }\n\n  for (const stat of Object.values(byTool)) {\n    stat.avgLatencyMs = stat.calls > 0 ? stat.avgLatencyMs / stat.calls : 0\n  }\n\n  // retry detection: per-tool chronological adjacency where error → next same-tool call\n  let retryOpportunities = 0\n  let retriesFollowed = 0\n  for (const [, arr] of groupBy(sortedTools, (t) => t.toolName)) {\n    for (let i = 0; i < arr.length; i++) {\n      if (arr[i]!.status !== 'error') continue\n      retryOpportunities += 1\n      if (arr[i + 1]) retriesFollowed += 1\n    }\n  }\n  const retryRate = retryOpportunities > 0 ? retriesFollowed / retryOpportunities : 0\n\n  let selectionAccuracy: number | undefined\n  if (options.selectionLabels) {\n    const labeled = sortedTools.filter((t) => t.spanId in options.selectionLabels!)\n    if (labeled.length > 0) {\n      selectionAccuracy =\n        labeled.filter((t) => options.selectionLabels![t.spanId]).length / labeled.length\n    }\n  }\n\n  return {\n    runId,\n    totalCalls: sortedTools.length,\n    byTool,\n    errorRate: totalErrors / sortedTools.length,\n    duplicateRate: totalDuplicates / sortedTools.length,\n    retryRate,\n    selectionAccuracy,\n  }\n}\n\nexport type { Span }\n","/**\n * Trajectory — ordered, structured view over a run's spans.\n *\n * A pure function `buildTrajectory(store, runId) → Trajectory` returns\n * a topologically ordered list of `TrajectoryStep` with parent-child\n * grouping collapsed into a single line-of-agent-work. Separate\n * analyzers (stuck-loop detection, waste ratio) live in\n * `pipelines/` and consume the trajectory.\n */\n\nimport type { Span, TraceEvent } from './trace/schema'\nimport type { TraceStore } from './trace/store'\n\nexport interface TrajectoryStep {\n  index: number\n  span: Span\n  /** Depth in the span tree from the root. 0 = top-level. */\n  depth: number\n  /** Events attached to this span. */\n  events: TraceEvent[]\n}\n\nexport interface Trajectory {\n  runId: string\n  steps: TrajectoryStep[]\n  llmTurns: number\n  toolCalls: number\n  judgeVerdicts: number\n  retrievals: number\n  totalDurationMs: number\n}\n\nexport async function buildTrajectory(store: TraceStore, runId: string): Promise<Trajectory> {\n  const spans = await store.spans({ runId })\n  const events = await store.events({ runId })\n  const childrenOf = new Map<string | undefined, Span[]>()\n  for (const s of spans) {\n    const arr = childrenOf.get(s.parentSpanId) ?? []\n    arr.push(s)\n    childrenOf.set(s.parentSpanId, arr)\n  }\n  // Sort children by startedAt so DFS yields chronological order within siblings.\n  for (const arr of childrenOf.values()) arr.sort((a, b) => a.startedAt - b.startedAt)\n\n  const eventsBySpan = new Map<string, TraceEvent[]>()\n  for (const e of events) {\n    if (!e.spanId) continue\n    const arr = eventsBySpan.get(e.spanId) ?? []\n    arr.push(e)\n    eventsBySpan.set(e.spanId, arr)\n  }\n\n  const steps: TrajectoryStep[] = []\n  const walk = (spanId: string | undefined, depth: number): void => {\n    const kids = childrenOf.get(spanId) ?? []\n    for (const child of kids) {\n      steps.push({\n        index: steps.length,\n        span: child,\n        depth,\n        events: eventsBySpan.get(child.spanId) ?? [],\n      })\n      walk(child.spanId, depth + 1)\n    }\n  }\n  walk(undefined, 0)\n\n  const llmTurns = steps.filter((s) => s.span.kind === 'llm').length\n  const toolCalls = steps.filter((s) => s.span.kind === 'tool').length\n  const judgeVerdicts = steps.filter((s) => s.span.kind === 'judge').length\n  const retrievals = steps.filter((s) => s.span.kind === 'retrieval').length\n\n  let totalDurationMs = 0\n  if (steps.length > 0) {\n    const starts = spans.map((s) => s.startedAt)\n    const ends = spans.map((s) => s.endedAt ?? s.startedAt)\n    totalDurationMs = Math.max(...ends) - Math.min(...starts)\n  }\n\n  return { runId, steps, llmTurns, toolCalls, judgeVerdicts, retrievals, totalDurationMs }\n}\n\n// Re-export core types for convenience so consumers don't import from two paths.\nexport type { Span, TraceEvent } from './trace/schema'\nexport type { TraceStore } from './trace/store'\n","/**\n * Baseline regression detection.\n *\n * Lifted from ADC baseline.ts. Every promotion-blocking signal boils down\n * to: \"is this run measurably worse than baseline?\" — with enough\n * statistical rigor to distinguish noise from drift.\n *\n * Uses:\n *   - Welch's t-test (unequal variance) for per-metric mean comparison\n *   - Cohen's d for effect size magnitude\n *   - IQR for stability flag (unstable samples can't be trusted for comparisons)\n *\n * Returns a structured verdict: improved | regressed | stable | unstable.\n */\n\nimport { cohensD } from './statistics'\n\nexport interface MetricSamples {\n  /** Stable metric key (e.g. \"overallScore\", \"firstTokenMs\"). */\n  metric: string\n  /** Whether higher values are better. */\n  higherIsBetter: boolean\n  baseline: number[]\n  candidate: number[]\n}\n\nexport interface MetricVerdict {\n  metric: string\n  baselineMean: number\n  candidateMean: number\n  delta: number\n  cohensD: number\n  welchT: number\n  welchDf: number\n  welchP: number\n  stable: boolean\n  /** IQR of the combined samples — used as a rough stability indicator. */\n  iqr: number\n  verdict: 'improved' | 'regressed' | 'stable' | 'unstable'\n}\n\nexport interface BaselineReport {\n  metrics: MetricVerdict[]\n  /** True if any critical metric regressed. */\n  hasRegression: boolean\n  /** True if any metric is unstable (too noisy to judge). */\n  hasUnstable: boolean\n}\n\nexport interface BaselineOptions {\n  /** Effect size threshold for meaningful delta (default 0.5 — medium effect). */\n  effectThreshold?: number\n  /** p-value threshold for statistical significance (default 0.05). */\n  alpha?: number\n  /** IQR/mean ratio above which samples are flagged unstable (default 0.30). */\n  unstableCvThreshold?: number\n}\n\n/**\n * Compare candidate samples against baseline per metric. Verdict logic:\n *   - unstable: IQR/|mean| > threshold on either set — not enough signal\n *   - improved: meaningful effect in the \"better\" direction AND p < alpha\n *   - regressed: meaningful effect in the \"worse\" direction AND p < alpha\n *   - stable: otherwise (no significant change)\n */\nexport function compareToBaseline(\n  samples: MetricSamples[],\n  options: BaselineOptions = {},\n): BaselineReport {\n  const effectThreshold = options.effectThreshold ?? 0.5\n  const alpha = options.alpha ?? 0.05\n  const cvThreshold = options.unstableCvThreshold ?? 0.3\n\n  const metrics: MetricVerdict[] = samples.map((s) => {\n    if (s.baseline.length < 2 || s.candidate.length < 2) {\n      throw new Error(`compareToBaseline: need ≥2 samples per side for \"${s.metric}\"`)\n    }\n    const bMean = mean(s.baseline)\n    const cMean = mean(s.candidate)\n    const delta = cMean - bMean\n    const d = cohensD(s.baseline, s.candidate) // positive = candidate higher\n    const { t, df, p } = welchsTTest(s.baseline, s.candidate)\n    // Stability is per-side: a comparison is trustworthy only when BOTH\n    // samples are internally consistent. Combining the sides would flag\n    // large-but-real deltas as \"unstable\" which is exactly what we want\n    // to detect.\n    const baselineIqr = iqr(s.baseline)\n    const candidateIqr = iqr(s.candidate)\n    const baselineStable = baselineIqr / Math.max(Math.abs(bMean), 1e-9) <= cvThreshold\n    const candidateStable = candidateIqr / Math.max(Math.abs(cMean), 1e-9) <= cvThreshold\n    const stable = baselineStable && candidateStable\n    const reportedIqr = Math.max(baselineIqr, candidateIqr)\n\n    let verdict: MetricVerdict['verdict']\n    if (!stable) {\n      verdict = 'unstable'\n    } else if (p < alpha && Math.abs(d) >= effectThreshold) {\n      const candidateIsBetter = s.higherIsBetter ? delta > 0 : delta < 0\n      verdict = candidateIsBetter ? 'improved' : 'regressed'\n    } else {\n      verdict = 'stable'\n    }\n\n    return {\n      metric: s.metric,\n      baselineMean: bMean,\n      candidateMean: cMean,\n      delta,\n      cohensD: d,\n      welchT: t,\n      welchDf: df,\n      welchP: p,\n      stable,\n      iqr: reportedIqr,\n      verdict,\n    }\n  })\n\n  return {\n    metrics,\n    hasRegression: metrics.some((m) => m.verdict === 'regressed'),\n    hasUnstable: metrics.some((m) => m.verdict === 'unstable'),\n  }\n}\n\nfunction mean(xs: number[]): number {\n  return xs.reduce((a, b) => a + b, 0) / xs.length\n}\n\n/** Inter-quartile range; 0 when the sample has no spread. */\nexport function iqr(xs: number[]): number {\n  if (xs.length === 0) return 0\n  const sorted = [...xs].sort((a, b) => a - b)\n  const q = (p: number) => {\n    const idx = p * (sorted.length - 1)\n    const lo = Math.floor(idx)\n    const hi = Math.ceil(idx)\n    return sorted[lo]! + (sorted[hi]! - sorted[lo]!) * (idx - lo)\n  }\n  return q(0.75) - q(0.25)\n}\n\n/**\n * Welch's t-test — unequal-variance two-sample t. Uses the same Student-t\n * CDF as `pairedTTest` (via incomplete beta); falls back to normal tail\n * when df is large.\n */\nexport function welchsTTest(a: number[], b: number[]): { t: number; df: number; p: number } {\n  if (a.length < 2 || b.length < 2) return { t: 0, df: 0, p: 1 }\n  const mA = mean(a)\n  const mB = mean(b)\n  const vA = variance(a, mA)\n  const vB = variance(b, mB)\n  const seSquared = vA / a.length + vB / b.length\n  if (seSquared === 0) return { t: mA === mB ? 0 : Infinity, df: 0, p: mA === mB ? 1 : 0 }\n  const t = (mB - mA) / Math.sqrt(seSquared)\n  const df =\n    (seSquared * seSquared) /\n    ((vA / a.length) ** 2 / (a.length - 1) + (vB / b.length) ** 2 / (b.length - 1))\n  const p = 2 * (1 - studentTCdf(Math.abs(t), df))\n  return { t, df, p }\n}\n\nfunction variance(xs: number[], m: number): number {\n  return xs.reduce((acc, x) => acc + (x - m) ** 2, 0) / (xs.length - 1)\n}\n\n// Re-used from statistics.ts via small local copy to avoid exporting internals.\nfunction studentTCdf(t: number, df: number): number {\n  if (df <= 0) return 0.5\n  if (df > 100) return normalCdf(t)\n  const x = df / (df + t * t)\n  const ib = incompleteBeta(x, df / 2, 0.5)\n  return t >= 0 ? 1 - 0.5 * ib : 0.5 * ib\n}\n\nfunction incompleteBeta(x: number, a: number, b: number): number {\n  if (x <= 0) return 0\n  if (x >= 1) return 1\n  const lnBeta = lnGamma(a) + lnGamma(b) - lnGamma(a + b)\n  const front = Math.exp(Math.log(x) * a + Math.log(1 - x) * b - lnBeta) / a\n  let c = 1\n  let d = 1 - ((a + b) * x) / (a + 1)\n  if (Math.abs(d) < 1e-30) d = 1e-30\n  d = 1 / d\n  let f = d\n  for (let m = 1; m <= 200; m++) {\n    const m2 = 2 * m\n    let num = (m * (b - m) * x) / ((a + m2 - 1) * (a + m2))\n    d = 1 + num * d\n    if (Math.abs(d) < 1e-30) d = 1e-30\n    c = 1 + num / c\n    if (Math.abs(c) < 1e-30) c = 1e-30\n    d = 1 / d\n    f *= d * c\n    num = -((a + m) * (a + b + m) * x) / ((a + m2) * (a + m2 + 1))\n    d = 1 + num * d\n    if (Math.abs(d) < 1e-30) d = 1e-30\n    c = 1 + num / c\n    if (Math.abs(c) < 1e-30) c = 1e-30\n    d = 1 / d\n    const delta = d * c\n    f *= delta\n    if (Math.abs(delta - 1) < 3e-7) break\n  }\n  return front * f\n}\n\nfunction lnGamma(z: number): number {\n  const coefs = [\n    0.99999999999980993, 676.5203681218851, -1259.1392167224028, 771.32342877765313,\n    -176.61502916214059, 12.507343278686905, -0.13857109526572012, 9.9843695780195716e-6,\n    1.5056327351493116e-7,\n  ]\n  if (z < 0.5) return Math.log(Math.PI / Math.sin(Math.PI * z)) - lnGamma(1 - z)\n  z -= 1\n  let x = coefs[0]!\n  for (let i = 1; i < 9; i++) x += coefs[i]! / (z + i)\n  const t = z + 7.5\n  return 0.5 * Math.log(2 * Math.PI) + (z + 0.5) * Math.log(t) - t + Math.log(x)\n}\n\nfunction normalCdf(x: number): number {\n  const a1 = 0.254829592\n  const a2 = -0.284496736\n  const a3 = 1.421413741\n  const a4 = -1.453152027\n  const a5 = 1.061405429\n  const p = 0.3275911\n  const sign = x < 0 ? -1 : 1\n  const absX = Math.abs(x)\n  const t = 1 / (1 + p * absX)\n  const y = 1 - ((((a5 * t + a4) * t + a3) * t + a2) * t + a1) * t * Math.exp((-absX * absX) / 2)\n  return 0.5 * (1 + sign * y)\n}\n"],"mappings":";;;;;;;;;;AAwCO,IAAM,gBAA+B;AAAA;AAAA,EAE1C;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,IAAI,MAAM;AAClB,YAAM,KAAK,IAAI,SAAS;AACxB,UAAI,MAAM,OAAO;AACf,eAAO,EAAE,cAAc,IAAI,QAAQ,sCAAsC;AAC3E,aAAO;AAAA,IACT;AAAA,EACF;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,OAAO,MAAM;AACrB,YAAM,QAAQ,OAAO;AAAA,QACnB,CAAC,MACC,EAAE,SAAS,YACX,EAAE,QAAQ,SAAS,sBACnB,EAAE,QAAQ,WAAW;AAAA,MACzB;AACA,aAAO,QACH;AAAA,QACE,cAAc;AAAA,QACd,QAAQ;AAAA,QACR,gBAAgB,MAAM;AAAA,MACxB,IACA;AAAA,IACN;AAAA,EACF;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,OAAO,MAAM;AACrB,YAAM,QAAQ,OAAO;AAAA,QACnB,CAAC,MACC,EAAE,SAAS,aACT,EAAE,QAAQ,SAAS,oCAAoC,EAAE,QAAQ,UAAU,SAC1E,EAAE,QAAQ,SAAS,+BAClB,EAAE,QAAQ,SAAS;AAAA,MAC3B;AACA,aAAO,QACH;AAAA,QACE,cAAc;AAAA,QACd,QAAQ;AAAA,QACR,gBAAgB,MAAM;AAAA,MACxB,IACA;AAAA,IACN;AAAA,EACF;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,OAAO,MAAM;AACrB,YAAM,QAAQ,OAAO;AAAA,QACnB,CAAC,MACC,EAAE,SAAS,YACX,EAAE,QAAQ,SAAS,mCACnB,oBAAoB,EAAE,SAAS,oBAAoB;AAAA,MACvD;AACA,aAAO,QACH;AAAA,QACE,cAAc;AAAA,QACd,QAAQ;AAAA,QACR,gBAAgB,MAAM;AAAA,MACxB,IACA;AAAA,IACN;AAAA,EACF;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,OAAO,MAAM;AACrB,YAAM,QAAQ,OAAO;AAAA,QACnB,CAAC,MACC,EAAE,SAAS,aACT,EAAE,QAAQ,SAAS,mCAAmC,iBAAiB,EAAE,OAAO,KAC/E,EAAE,QAAQ,SAAS,+BAA+B,EAAE,QAAQ,SAAS;AAAA,MAC5E;AACA,aAAO,QACH;AAAA,QACE,cAAc;AAAA,QACd,QAAQ;AAAA,QACR,gBAAgB,MAAM;AAAA,MACxB,IACA;AAAA,IACN;AAAA,EACF;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,OAAO,MAAM;AACrB,YAAM,QAAQ,OAAO;AAAA,QACnB,CAAC,MACC,EAAE,SAAS,aACT,EAAE,QAAQ,SAAS,wBAAwB,EAAE,QAAQ,WAAW,uBAC/D,EAAE,QAAQ,SAAS,+BAClB,EAAE,QAAQ,SAAS,uBACrB,EAAE,QAAQ,SAAS;AAAA,MACzB;AACA,aAAO,QACH;AAAA,QACE,cAAc;AAAA,QACd,QAAQ;AAAA,QACR,gBAAgB,MAAM;AAAA,MACxB,IACA;AAAA,IACN;AAAA,EACF;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,OAAO,MAAM;AACrB,YAAM,QAAQ,OAAO;AAAA,QACnB,CAAC,MACC,EAAE,SAAS,YACX,EAAE,QAAQ,SAAS,gCAClB,EAAE,QAAQ,SAAS,kBAClB,EAAE,QAAQ,SAAS,2BACnB,EAAE,QAAQ,SAAS,wBACnB,EAAE,QAAQ,WAAW;AAAA,MAC3B;AACA,aAAO,QACH;AAAA,QACE,cAAc;AAAA,QACd,QAAQ;AAAA,QACR,gBAAgB,MAAM;AAAA,MACxB,IACA;AAAA,IACN;AAAA,EACF;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,OAAO,MAAM;AACrB,YAAM,QAAQ,OAAO;AAAA,QACnB,CAAC,MACC,EAAE,SAAS,YACX,EAAE,QAAQ,SAAS,gCAClB,EAAE,QAAQ,SAAS,yBAClB,EAAE,QAAQ,SAAS,mBACnB,EAAE,QAAQ,SAAS;AAAA,MACzB;AACA,aAAO,QACH;AAAA,QACE,cAAc;AAAA,QACd,QAAQ;AAAA,QACR,gBAAgB,MAAM;AAAA,MACxB,IACA;AAAA,IACN;AAAA,EACF;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,OAAO,MAAM;AACrB,YAAM,QAAQ,OAAO;AAAA,QACnB,CAAC,MACC,EAAE,SAAS,YACX,EAAE,QAAQ,SAAS,+BACnB,CAAC;AAAA,UACC;AAAA,UACA;AAAA,UACA;AAAA,UACA;AAAA,UACA;AAAA,UACA;AAAA,UACA;AAAA,UACA;AAAA,UACA;AAAA,QACF,EAAE,SAAS,OAAO,EAAE,QAAQ,IAAI,CAAC;AAAA,MACrC;AACA,aAAO,QACH;AAAA,QACE,cAAc;AAAA,QACd,QAAQ;AAAA,QACR,gBAAgB,MAAM;AAAA,MACxB,IACA;AAAA,IACN;AAAA,EACF;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,OAAO,MAAM;AACrB,YAAM,QAAQ,OAAO;AAAA,QACnB,CAAC,MACC,EAAE,SAAS,YACX,EAAE,QAAQ,SAAS,mBACnB,EAAE,QAAQ,aAAa;AAAA,MAC3B;AACA,aAAO,QACH;AAAA,QACE,cAAc;AAAA,QACd,QAAQ;AAAA,QACR,gBAAgB,MAAM;AAAA,MACxB,IACA;AAAA,IACN;AAAA,EACF;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,KAAK,MAAM,MAAM;AACzB,UAAI,IAAI,SAAS,SAAS,MAAO,QAAO;AACxC,YAAM,YAAY,MAAM;AAAA,QACtB,CAAC,MACC,EAAE,SAAS,gBAAgB,EAAE,KAAK,WAAW,KAAK,EAAE,KAAK,MAAM,CAAC,QAAQ,IAAI,SAAS,CAAC;AAAA,MAC1F;AACA,aAAO,YACH;AAAA,QACE,cAAc;AAAA,QACd,QAAQ;AAAA,QACR,eAAe,UAAU;AAAA,MAC3B,IACA;AAAA,IACN;AAAA,EACF;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,OAAO,MAAM;AACrB,YAAM,QAAQ,OAAO;AAAA,QACnB,CAAC,MACC,EAAE,SAAS,YACX,EAAE,QAAQ,SAAS,mBACnB,EAAE,QAAQ,WAAW;AAAA,MACzB;AACA,aAAO,QACH;AAAA,QACE,cAAc;AAAA,QACd,QAAQ;AAAA,QACR,gBAAgB,MAAM;AAAA,MACxB,IACA;AAAA,IACN;AAAA,EACF;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,OAAO,MAAM;AACrB,YAAM,QAAQ,OAAO;AAAA,QACnB,CAAC,MACC,EAAE,SAAS,YACX,EAAE,QAAQ,SAAS,mBACnB,EAAE,QAAQ,WAAW;AAAA,MACzB;AACA,aAAO,QACH;AAAA,QACE,cAAc;AAAA,QACd,QAAQ;AAAA,QACR,gBAAgB,MAAM;AAAA,MACxB,IACA;AAAA,IACN;AAAA,EACF;AAAA;AAAA,EAEA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,OAAO,MAAM;AACrB,YAAM,SAAS,OAAO,KAAK,CAAC,MAAM,EAAE,SAAS,eAAe;AAC5D,aAAO,SACH;AAAA,QACE,cAAc;AAAA,QACd,QAAQ,sBAAsB,OAAO,QAAQ,aAAa,mBAAmB;AAAA,QAC7E,gBAAgB,OAAO;AAAA,MACzB,IACA;AAAA,IACN;AAAA,EACF;AAAA;AAAA,EAEA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,OAAO,MAAM;AACrB,YAAM,IAAI,OAAO,KAAK,CAAC,MAAM,EAAE,SAAS,kBAAkB;AAC1D,aAAO,IACH;AAAA,QACE,cAAc;AAAA,QACd,QAAQ;AAAA,QACR,gBAAgB,EAAE;AAAA,MACpB,IACA;AAAA,IACN;AAAA,EACF;AAAA;AAAA,EAEA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,MAAM,MAAM;AACpB,YAAM,IAAI,MAAM;AAAA,QACd,CAAC,MAAM,EAAE,SAAS,aAAa,OAAO,EAAE,aAAa,YAAY,EAAE,aAAa;AAAA,MAClF;AACA,UAAI,CAAC,EAAG,QAAO;AACf,aAAO;AAAA,QACL,cAAc;AAAA,QACd,QAAQ,kBAAmB,EAAyC,QAAQ;AAAA,QAC5E,eAAe,EAAE;AAAA,MACnB;AAAA,IACF;AAAA,EACF;AAAA;AAAA,EAEA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,KAAK,OAAO,MAAM;AAC1B,UAAI,IAAI,WAAW,UAAW,QAAO;AACrC,YAAM,aAAa,OAAO;AAAA,QACxB,CAAC,MACC,EAAE,SAAS,WACX,OAAO,EAAE,QAAQ,UAAU,EAAE,EAC1B,YAAY,EACZ,SAAS,SAAS;AAAA,MACzB;AACA,YAAM,QAAQ,IAAI,SAAS,SAAS,IAAI,YAAY;AACpD,UAAI,cAAc,KAAK,SAAS,SAAS,KAAK,KAAK,SAAS,UAAU,GAAG;AACvE,eAAO,EAAE,cAAc,WAAW,QAAQ,0BAA0B;AAAA,MACtE;AACA,aAAO;AAAA,IACT;AAAA,EACF;AAAA;AAAA,EAEA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,MAAM,MAAM;AACpB,YAAM,QAAQ,MAAM,OAAO,CAAC,MAAM,EAAE,SAAS,MAAM;AACnD,YAAM,SAAS,oBAAI,IAAoB;AACvC,iBAAW,KAAK,OAAO;AACrB,cAAM,OAAQ,EAAsC;AACpD,cAAM,MAAM,OAAO,IAAI,IAAI,KAAK,CAAC;AACjC,YAAI,KAAK,CAAC;AACV,eAAO,IAAI,MAAM,GAAG;AAAA,MACtB;AACA,iBAAW,CAAC,MAAM,GAAG,KAAK,QAAQ;AAChC,cAAM,OAAO,IAAI,OAAO,CAAC,MAAM,EAAE,WAAW,OAAO;AACnD,YAAI,KAAK,UAAU,KAAK,KAAK,WAAW,IAAI,QAAQ;AAClD,iBAAO;AAAA,YACL,cAAc;AAAA,YACd,QAAQ,GAAG,KAAK,MAAM,gCAAgC,IAAI;AAAA,YAC1D,eAAe,KAAK,KAAK,SAAS,CAAC,EAAG;AAAA,UACxC;AAAA,QACF;AAAA,MACF;AACA,aAAO;AAAA,IACT;AAAA,EACF;AAAA;AAAA,EAEA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,KAAK,MAAM,MAAM;AACzB,UAAI,IAAI,SAAS,SAAS,MAAO,QAAO;AACxC,YAAM,oBAAoB,MAAM;AAAA,QAC9B,CAAC,MACC,EAAE,SAAS,WACV,EAAE,YAAY,mBAA0C,UACxD,EAAE,YAAY,iBAA4B;AAAA,MAC/C;AACA,YAAM,QAAQ,MAAM,OAAO,CAAC,MAAM,EAAE,SAAS,MAAM;AACnD,UAAI,qBAAqB,MAAM,WAAW,GAAG;AAC3C,eAAO;AAAA,UACL,cAAc;AAAA,UACd,QAAQ;AAAA,QACV;AAAA,MACF;AACA,aAAO;AAAA,IACT;AAAA,EACF;AAAA;AAAA,EAEA;AAAA,IACE,IAAI;AAAA,IACJ,OAAO,CAAC,EAAE,MAAM,MAAM;AACpB,YAAM,QAAQ,MAAM;AAAA,QAClB,CAAC,MACC,EAAE,SAAS,WACV,EAAuC,cAAc,YACrD,EAAuC,QAAQ;AAAA,MACpD;AACA,aAAO,QACH;AAAA,QACE,cAAc;AAAA,QACd,QAAQ;AAAA,QACR,eAAe,MAAM;AAAA,MACvB,IACA;AAAA,IACN;AAAA,EACF;AACF;AAEA,SAAS,oBAAoB,SAAkC,QAAyB;AACtF,MAAI,WAAW,wBAAwB,YAAY,QAAQ,kBAAkB,EAAE,SAAS;AACtF,WAAO;AACT,SAAO,gBAAgB,OAAO,EAAE,KAAK,CAAC,SAAS,KAAK,WAAW,MAAM;AACvE;AAEA,SAAS,iBAAiB,SAA2C;AACnE,MAAI,YAAY,QAAQ,aAAa,EAAE,SAAS,EAAG,QAAO;AAC1D,SAAO,gBAAgB,OAAO,EAAE;AAAA,IAC9B,CAAC,SAAS,MAAM,QAAQ,KAAK,aAAa,KAAK,KAAK,cAAc,SAAS;AAAA,EAC7E;AACF;AAEA,SAAS,gBAAgB,SAAkE;AACzF,SAAO;AAAA,IACL,GAAG,QAAQ,QAAQ,OAAO;AAAA,IAC1B,GAAG,QAAQ,QAAQ,eAAe;AAAA,IAClC,GAAG,QAAQ,QAAQ,KAAK;AAAA,EAC1B;AACF;AAEA,SAAS,QAAQ,OAAgD;AAC/D,MAAI,CAAC,MAAM,QAAQ,KAAK,EAAG,QAAO,CAAC;AACnC,SAAO,MAAM;AAAA,IACX,CAAC,SACC,QAAQ,IAAI,KAAK,OAAO,SAAS,YAAY,CAAC,MAAM,QAAQ,IAAI;AAAA,EACpE;AACF;AAEA,SAAS,YAAY,OAA0B;AAC7C,SAAO,MAAM,QAAQ,KAAK,IACtB,MAAM,OAAO,CAAC,SAAyB,OAAO,SAAS,QAAQ,IAC/D,CAAC;AACP;AAGO,SAAS,gBACd,KACA,QAAuB,eACA;AACvB,MAAI,IAAI,IAAI,SAAS,SAAS,SAAS,IAAI,IAAI,WAAW,aAAa;AACrE,WAAO,EAAE,cAAc,WAAW,QAAQ,qDAAqD;AAAA,EACjG;AACA,aAAW,QAAQ,OAAO;AACxB,UAAM,MAAM,KAAK,MAAM,GAAG;AAC1B,QAAI,IAAK,QAAO;AAAA,EAClB;AACA,SAAO,EAAE,cAAc,WAAW,QAAQ,sDAAsD;AAClG;;;ACvaA,eAAsB,sBACpB,OACA,OACA,UAA0B,CAAC,GACF;AACzB,QAAM,QAAQ,MAAM,UAAU,OAAO,KAAK;AAC1C,MAAI,MAAM,WAAW,GAAG;AACtB,WAAO,EAAE,OAAO,YAAY,GAAG,QAAQ,CAAC,GAAG,WAAW,GAAG,eAAe,GAAG,WAAW,EAAE;AAAA,EAC1F;AAEA,QAAM,SAAoC,CAAC;AAC3C,MAAI,cAAc;AAClB,MAAI,kBAAkB;AACtB,QAAM,cAAc,CAAC,GAAG,KAAK,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;AACvE,QAAM,iBAAiB,oBAAI,IAAY;AAGvC,aAAW,KAAK,aAAa;AAC3B,WAAO,EAAE,QAAQ,MAAM,EAAE,OAAO,GAAG,QAAQ,GAAG,cAAc,GAAG,YAAY,EAAE;AAC7E,UAAM,OAAO,OAAO,EAAE,QAAQ;AAC9B,SAAK,SAAS;AACd,QAAI,EAAE,WAAW,SAAS;AACxB,WAAK,UAAU;AACf,qBAAe;AAAA,IACjB;AACA,QAAI,OAAO,EAAE,cAAc,SAAU,MAAK,gBAAgB,EAAE;AAC5D,UAAM,MAAM,GAAG,EAAE,QAAQ,IAAI,QAAQ,EAAE,IAAI,CAAC;AAC5C,QAAI,eAAe,IAAI,GAAG,GAAG;AAC3B,WAAK,cAAc;AACnB,yBAAmB;AAAA,IACrB;AACA,mBAAe,IAAI,GAAG;AAAA,EACxB;AAEA,aAAW,QAAQ,OAAO,OAAO,MAAM,GAAG;AACxC,SAAK,eAAe,KAAK,QAAQ,IAAI,KAAK,eAAe,KAAK,QAAQ;AAAA,EACxE;AAGA,MAAI,qBAAqB;AACzB,MAAI,kBAAkB;AACtB,aAAW,CAAC,EAAE,GAAG,KAAK,QAAQ,aAAa,CAAC,MAAM,EAAE,QAAQ,GAAG;AAC7D,aAAS,IAAI,GAAG,IAAI,IAAI,QAAQ,KAAK;AACnC,UAAI,IAAI,CAAC,EAAG,WAAW,QAAS;AAChC,4BAAsB;AACtB,UAAI,IAAI,IAAI,CAAC,EAAG,oBAAmB;AAAA,IACrC;AAAA,EACF;AACA,QAAM,YAAY,qBAAqB,IAAI,kBAAkB,qBAAqB;AAElF,MAAI;AACJ,MAAI,QAAQ,iBAAiB;AAC3B,UAAM,UAAU,YAAY,OAAO,CAAC,MAAM,EAAE,UAAU,QAAQ,eAAgB;AAC9E,QAAI,QAAQ,SAAS,GAAG;AACtB,0BACE,QAAQ,OAAO,CAAC,MAAM,QAAQ,gBAAiB,EAAE,MAAM,CAAC,EAAE,SAAS,QAAQ;AAAA,IAC/E;AAAA,EACF;AAEA,SAAO;AAAA,IACL;AAAA,IACA,YAAY,YAAY;AAAA,IACxB;AAAA,IACA,WAAW,cAAc,YAAY;AAAA,IACrC,eAAe,kBAAkB,YAAY;AAAA,IAC7C;AAAA,IACA;AAAA,EACF;AACF;;;AC1EA,eAAsB,gBAAgB,OAAmB,OAAoC;AAC3F,QAAM,QAAQ,MAAM,MAAM,MAAM,EAAE,MAAM,CAAC;AACzC,QAAM,SAAS,MAAM,MAAM,OAAO,EAAE,MAAM,CAAC;AAC3C,QAAM,aAAa,oBAAI,IAAgC;AACvD,aAAW,KAAK,OAAO;AACrB,UAAM,MAAM,WAAW,IAAI,EAAE,YAAY,KAAK,CAAC;AAC/C,QAAI,KAAK,CAAC;AACV,eAAW,IAAI,EAAE,cAAc,GAAG;AAAA,EACpC;AAEA,aAAW,OAAO,WAAW,OAAO,EAAG,KAAI,KAAK,CAAC,GAAG,MAAM,EAAE,YAAY,EAAE,SAAS;AAEnF,QAAM,eAAe,oBAAI,IAA0B;AACnD,aAAW,KAAK,QAAQ;AACtB,QAAI,CAAC,EAAE,OAAQ;AACf,UAAM,MAAM,aAAa,IAAI,EAAE,MAAM,KAAK,CAAC;AAC3C,QAAI,KAAK,CAAC;AACV,iBAAa,IAAI,EAAE,QAAQ,GAAG;AAAA,EAChC;AAEA,QAAM,QAA0B,CAAC;AACjC,QAAM,OAAO,CAAC,QAA4B,UAAwB;AAChE,UAAM,OAAO,WAAW,IAAI,MAAM,KAAK,CAAC;AACxC,eAAW,SAAS,MAAM;AACxB,YAAM,KAAK;AAAA,QACT,OAAO,MAAM;AAAA,QACb,MAAM;AAAA,QACN;AAAA,QACA,QAAQ,aAAa,IAAI,MAAM,MAAM,KAAK,CAAC;AAAA,MAC7C,CAAC;AACD,WAAK,MAAM,QAAQ,QAAQ,CAAC;AAAA,IAC9B;AAAA,EACF;AACA,OAAK,QAAW,CAAC;AAEjB,QAAM,WAAW,MAAM,OAAO,CAAC,MAAM,EAAE,KAAK,SAAS,KAAK,EAAE;AAC5D,QAAM,YAAY,MAAM,OAAO,CAAC,MAAM,EAAE,KAAK,SAAS,MAAM,EAAE;AAC9D,QAAM,gBAAgB,MAAM,OAAO,CAAC,MAAM,EAAE,KAAK,SAAS,OAAO,EAAE;AACnE,QAAM,aAAa,MAAM,OAAO,CAAC,MAAM,EAAE,KAAK,SAAS,WAAW,EAAE;AAEpE,MAAI,kBAAkB;AACtB,MAAI,MAAM,SAAS,GAAG;AACpB,UAAM,SAAS,MAAM,IAAI,CAAC,MAAM,EAAE,SAAS;AAC3C,UAAM,OAAO,MAAM,IAAI,CAAC,MAAM,EAAE,WAAW,EAAE,SAAS;AACtD,sBAAkB,KAAK,IAAI,GAAG,IAAI,IAAI,KAAK,IAAI,GAAG,MAAM;AAAA,EAC1D;AAEA,SAAO,EAAE,OAAO,OAAO,UAAU,WAAW,eAAe,YAAY,gBAAgB;AACzF;;;ACfO,SAAS,kBACd,SACA,UAA2B,CAAC,GACZ;AAChB,QAAM,kBAAkB,QAAQ,mBAAmB;AACnD,QAAM,QAAQ,QAAQ,SAAS;AAC/B,QAAM,cAAc,QAAQ,uBAAuB;AAEnD,QAAM,UAA2B,QAAQ,IAAI,CAAC,MAAM;AAClD,QAAI,EAAE,SAAS,SAAS,KAAK,EAAE,UAAU,SAAS,GAAG;AACnD,YAAM,IAAI,MAAM,yDAAoD,EAAE,MAAM,GAAG;AAAA,IACjF;AACA,UAAM,QAAQ,KAAK,EAAE,QAAQ;AAC7B,UAAM,QAAQ,KAAK,EAAE,SAAS;AAC9B,UAAM,QAAQ,QAAQ;AACtB,UAAM,IAAI,QAAQ,EAAE,UAAU,EAAE,SAAS;AACzC,UAAM,EAAE,GAAG,IAAI,EAAE,IAAI,YAAY,EAAE,UAAU,EAAE,SAAS;AAKxD,UAAM,cAAc,IAAI,EAAE,QAAQ;AAClC,UAAM,eAAe,IAAI,EAAE,SAAS;AACpC,UAAM,iBAAiB,cAAc,KAAK,IAAI,KAAK,IAAI,KAAK,GAAG,IAAI,KAAK;AACxE,UAAM,kBAAkB,eAAe,KAAK,IAAI,KAAK,IAAI,KAAK,GAAG,IAAI,KAAK;AAC1E,UAAM,SAAS,kBAAkB;AACjC,UAAM,cAAc,KAAK,IAAI,aAAa,YAAY;AAEtD,QAAI;AACJ,QAAI,CAAC,QAAQ;AACX,gBAAU;AAAA,IACZ,WAAW,IAAI,SAAS,KAAK,IAAI,CAAC,KAAK,iBAAiB;AACtD,YAAM,oBAAoB,EAAE,iBAAiB,QAAQ,IAAI,QAAQ;AACjE,gBAAU,oBAAoB,aAAa;AAAA,IAC7C,OAAO;AACL,gBAAU;AAAA,IACZ;AAEA,WAAO;AAAA,MACL,QAAQ,EAAE;AAAA,MACV,cAAc;AAAA,MACd,eAAe;AAAA,MACf;AAAA,MACA,SAAS;AAAA,MACT,QAAQ;AAAA,MACR,SAAS;AAAA,MACT,QAAQ;AAAA,MACR;AAAA,MACA,KAAK;AAAA,MACL;AAAA,IACF;AAAA,EACF,CAAC;AAED,SAAO;AAAA,IACL;AAAA,IACA,eAAe,QAAQ,KAAK,CAAC,MAAM,EAAE,YAAY,WAAW;AAAA,IAC5D,aAAa,QAAQ,KAAK,CAAC,MAAM,EAAE,YAAY,UAAU;AAAA,EAC3D;AACF;AAEA,SAAS,KAAK,IAAsB;AAClC,SAAO,GAAG,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,GAAG;AAC5C;AAGO,SAAS,IAAI,IAAsB;AACxC,MAAI,GAAG,WAAW,EAAG,QAAO;AAC5B,QAAM,SAAS,CAAC,GAAG,EAAE,EAAE,KAAK,CAAC,GAAG,MAAM,IAAI,CAAC;AAC3C,QAAM,IAAI,CAAC,MAAc;AACvB,UAAM,MAAM,KAAK,OAAO,SAAS;AACjC,UAAM,KAAK,KAAK,MAAM,GAAG;AACzB,UAAM,KAAK,KAAK,KAAK,GAAG;AACxB,WAAO,OAAO,EAAE,KAAM,OAAO,EAAE,IAAK,OAAO,EAAE,MAAO,MAAM;AAAA,EAC5D;AACA,SAAO,EAAE,IAAI,IAAI,EAAE,IAAI;AACzB;AAOO,SAAS,YAAY,GAAa,GAAmD;AAC1F,MAAI,EAAE,SAAS,KAAK,EAAE,SAAS,EAAG,QAAO,EAAE,GAAG,GAAG,IAAI,GAAG,GAAG,EAAE;AAC7D,QAAM,KAAK,KAAK,CAAC;AACjB,QAAM,KAAK,KAAK,CAAC;AACjB,QAAM,KAAK,SAAS,GAAG,EAAE;AACzB,QAAM,KAAK,SAAS,GAAG,EAAE;AACzB,QAAM,YAAY,KAAK,EAAE,SAAS,KAAK,EAAE;AACzC,MAAI,cAAc,EAAG,QAAO,EAAE,GAAG,OAAO,KAAK,IAAI,UAAU,IAAI,GAAG,GAAG,OAAO,KAAK,IAAI,EAAE;AACvF,QAAM,KAAK,KAAK,MAAM,KAAK,KAAK,SAAS;AACzC,QAAM,KACH,YAAY,cACX,KAAK,EAAE,WAAW,KAAK,EAAE,SAAS,MAAM,KAAK,EAAE,WAAW,KAAK,EAAE,SAAS;AAC9E,QAAM,IAAI,KAAK,IAAI,YAAY,KAAK,IAAI,CAAC,GAAG,EAAE;AAC9C,SAAO,EAAE,GAAG,IAAI,EAAE;AACpB;AAEA,SAAS,SAAS,IAAc,GAAmB;AACjD,SAAO,GAAG,OAAO,CAAC,KAAK,MAAM,OAAO,IAAI,MAAM,GAAG,CAAC,KAAK,GAAG,SAAS;AACrE;AAGA,SAAS,YAAY,GAAW,IAAoB;AAClD,MAAI,MAAM,EAAG,QAAO;AACpB,MAAI,KAAK,IAAK,QAAO,UAAU,CAAC;AAChC,QAAM,IAAI,MAAM,KAAK,IAAI;AACzB,QAAM,KAAK,eAAe,GAAG,KAAK,GAAG,GAAG;AACxC,SAAO,KAAK,IAAI,IAAI,MAAM,KAAK,MAAM;AACvC;AAEA,SAAS,eAAe,GAAW,GAAW,GAAmB;AAC/D,MAAI,KAAK,EAAG,QAAO;AACnB,MAAI,KAAK,EAAG,QAAO;AACnB,QAAM,SAAS,QAAQ,CAAC,IAAI,QAAQ,CAAC,IAAI,QAAQ,IAAI,CAAC;AACtD,QAAM,QAAQ,KAAK,IAAI,KAAK,IAAI,CAAC,IAAI,IAAI,KAAK,IAAI,IAAI,CAAC,IAAI,IAAI,MAAM,IAAI;AACzE,MAAI,IAAI;AACR,MAAI,IAAI,KAAM,IAAI,KAAK,KAAM,IAAI;AACjC,MAAI,KAAK,IAAI,CAAC,IAAI,MAAO,KAAI;AAC7B,MAAI,IAAI;AACR,MAAI,IAAI;AACR,WAAS,IAAI,GAAG,KAAK,KAAK,KAAK;AAC7B,UAAM,KAAK,IAAI;AACf,QAAI,MAAO,KAAK,IAAI,KAAK,MAAO,IAAI,KAAK,MAAM,IAAI;AACnD,QAAI,IAAI,MAAM;AACd,QAAI,KAAK,IAAI,CAAC,IAAI,MAAO,KAAI;AAC7B,QAAI,IAAI,MAAM;AACd,QAAI,KAAK,IAAI,CAAC,IAAI,MAAO,KAAI;AAC7B,QAAI,IAAI;AACR,SAAK,IAAI;AACT,UAAM,GAAG,IAAI,MAAM,IAAI,IAAI,KAAK,OAAO,IAAI,OAAO,IAAI,KAAK;AAC3D,QAAI,IAAI,MAAM;AACd,QAAI,KAAK,IAAI,CAAC,IAAI,MAAO,KAAI;AAC7B,QAAI,IAAI,MAAM;AACd,QAAI,KAAK,IAAI,CAAC,IAAI,MAAO,KAAI;AAC7B,QAAI,IAAI;AACR,UAAM,QAAQ,IAAI;AAClB,SAAK;AACL,QAAI,KAAK,IAAI,QAAQ,CAAC,IAAI,KAAM;AAAA,EAClC;AACA,SAAO,QAAQ;AACjB;AAEA,SAAS,QAAQ,GAAmB;AAClC,QAAM,QAAQ;AAAA,IACZ;AAAA,IAAqB;AAAA,IAAmB;AAAA,IAAqB;AAAA,IAC7D;AAAA,IAAqB;AAAA,IAAoB;AAAA,IAAsB;AAAA,IAC/D;AAAA,EACF;AACA,MAAI,IAAI,IAAK,QAAO,KAAK,IAAI,KAAK,KAAK,KAAK,IAAI,KAAK,KAAK,CAAC,CAAC,IAAI,QAAQ,IAAI,CAAC;AAC7E,OAAK;AACL,MAAI,IAAI,MAAM,CAAC;AACf,WAAS,IAAI,GAAG,IAAI,GAAG,IAAK,MAAK,MAAM,CAAC,KAAM,IAAI;AAClD,QAAM,IAAI,IAAI;AACd,SAAO,MAAM,KAAK,IAAI,IAAI,KAAK,EAAE,KAAK,IAAI,OAAO,KAAK,IAAI,CAAC,IAAI,IAAI,KAAK,IAAI,CAAC;AAC/E;AAEA,SAAS,UAAU,GAAmB;AACpC,QAAM,KAAK;AACX,QAAM,KAAK;AACX,QAAM,KAAK;AACX,QAAM,KAAK;AACX,QAAM,KAAK;AACX,QAAM,IAAI;AACV,QAAM,OAAO,IAAI,IAAI,KAAK;AAC1B,QAAM,OAAO,KAAK,IAAI,CAAC;AACvB,QAAM,IAAI,KAAK,IAAI,IAAI;AACvB,QAAM,IAAI,QAAQ,KAAK,IAAI,MAAM,IAAI,MAAM,IAAI,MAAM,IAAI,MAAM,IAAI,KAAK,IAAK,CAAC,OAAO,OAAQ,CAAC;AAC9F,SAAO,OAAO,IAAI,OAAO;AAC3B;","names":[]}

package/dist/{chunk-GDZAWO2I.js → chunk-QFGTU7MT.js}

@@ -15,7 +15,7 @@ import {
 } from "./chunk-XMSYF4A7.js";
 import {
   llmSpans
-} from "./chunk-JZXGWLK5.js";
+} from "./chunk-MHNQWM4I.js";
 
 // src/contamination-guard.ts
 function checkCanaries(output, scenarios) {
@@ -822,4 +822,4 @@ export {
   HoldoutAuditor,
   analyzeRuns
 };
-//# sourceMappingURL=chunk-GDZAWO2I.js.map
+//# sourceMappingURL=chunk-QFGTU7MT.js.map

package/dist/{chunk-6SKVFBTR.js → chunk-RNB2NICW.js}

@@ -10,7 +10,11 @@ import {
   renderAnalystEvidence,
   runImprovementLoop,
   surfaceHash
-} from "./chunk-OVPVM4JC.js";
+} from "./chunk-GTERJI6Q.js";
+import {
+  estimateCost,
+  isModelPriced
+} from "./chunk-VI2UW6B6.js";
 import {
   assertRealBackend,
   contentHash,
@@ -18,25 +22,18 @@ import {
   runCampaign,
   summarizeBackendIntegrity
 } from "./chunk-3PFZBGMR.js";
-import {
-  estimateCost,
-  isModelPriced
-} from "./chunk-VI2UW6B6.js";
 import {
   admitPolicyEdit,
   applyPolicyEditToSurface,
   clamp01,
   isPolicyEdit,
   policyEditsFromFindings
-} from "./chunk-LOJ2QVCE.js";
+} from "./chunk-2IY4ILP4.js";
 import {
   AnalystRegistry,
   DEFAULT_TRACE_ANALYST_KINDS,
   createTraceAnalystKind
-} from "./chunk-2OGPXHOB.js";
-import {
-  callLlm
-} from "./chunk-FUCQVFMU.js";
+} from "./chunk-7NX6ZSBG.js";
 import {
   eProcess,
   mulberry32,
@@ -59,6 +56,9 @@ import {
 import {
   canonicalize
 } from "./chunk-VSMTAMNK.js";
+import {
+  callLlm
+} from "./chunk-FUCQVFMU.js";
 import {
   AgentEvalError,
   JudgeError,
@@ -1378,6 +1378,57 @@ function sequentialDecide(options = {}) {
   return decide;
 }
 
+// src/campaign/grounded-reflection.ts
+function rolloutArgumentDiff(rollouts, opts = {}) {
+  const passThreshold = opts.passThreshold ?? 1;
+  const maxValues = opts.maxValuesPerField ?? 4;
+  const collect = (pass) => {
+    const byField = /* @__PURE__ */ new Map();
+    for (const r of rollouts) {
+      if (pass !== r.score >= passThreshold) continue;
+      for (const c of r.calls) {
+        for (const [k, v] of Object.entries(c.args)) {
+          const set = byField.get(k) ?? /* @__PURE__ */ new Set();
+          set.add(String(v));
+          byField.set(k, set);
+        }
+      }
+    }
+    return byField;
+  };
+  const passing = collect(true);
+  const failing = collect(false);
+  const lines = [];
+  for (const field of /* @__PURE__ */ new Set([...passing.keys(), ...failing.keys()])) {
+    const pv = [...passing.get(field) ?? []].slice(0, maxValues);
+    const fv = [...failing.get(field) ?? []].slice(0, maxValues);
+    const render = (vals) => vals.length ? JSON.stringify(vals) : "NOT SET (omitted)";
+    lines.push(`  ${field}: passing runs -> ${render(pv)} | failing runs -> ${render(fv)}`);
+  }
+  const lower = (m) => {
+    const out = /* @__PURE__ */ new Set();
+    for (const vals of m.values()) for (const v of vals) out.add(v.toLowerCase());
+    return out;
+  };
+  return {
+    text: lines.join("\n") || "  (no calls observed)",
+    passingValues: lower(passing),
+    failingValues: lower(failing)
+  };
+}
+function classifyUngroundedLiterals(text, diff) {
+  const ungrounded = /* @__PURE__ */ new Set();
+  for (const m of text.matchAll(/'([a-z][a-z_-]{1,19})'/gi)) {
+    const w = m[1].toLowerCase();
+    if (!diff.passingValues.has(w)) ungrounded.add(w);
+  }
+  const all = [...ungrounded];
+  return {
+    ungrounded: all,
+    harmful: all.filter((w) => diff.failingValues.has(w))
+  };
+}
+
 // src/campaign/labeled-store/fs-adapter.ts
 import { createHash as createHash4 } from "crypto";
 import { existsSync as existsSync2, mkdirSync, readFileSync as readFileSync2, writeFileSync } from "fs";
@@ -4169,9 +4220,56 @@ function selectDiscriminative(signals, k, opts) {
   return [...nonTied, ...fill].map((s) => s.scenarioId);
 }
 
+// src/campaign/single-run-lock.ts
+import { existsSync as existsSync3, readFileSync as readFileSync3, unlinkSync, writeFileSync as writeFileSync3 } from "fs";
+function liveHolder(path) {
+  if (!existsSync3(path)) return null;
+  const holder = Number(readFileSync3(path, "utf8").trim());
+  if (!Number.isFinite(holder) || holder <= 0) return null;
+  try {
+    process.kill(holder, 0);
+    return holder;
+  } catch {
+    return null;
+  }
+}
+function acquireSingleRunLock(opts) {
+  const pid = opts.pid ?? process.pid;
+  for (const path of [opts.lockPath, ...opts.alsoCheck ?? []]) {
+    const holder = liveHolder(path);
+    if (holder !== null && holder !== pid) {
+      throw new Error(
+        `single-run lock held by live pid ${holder} (${path}); refusing a concurrent run on the shared resource`
+      );
+    }
+  }
+  writeFileSync3(opts.lockPath, String(pid));
+  const release = () => {
+    try {
+      if (existsSync3(opts.lockPath) && readFileSync3(opts.lockPath, "utf8").trim() === String(pid)) {
+        unlinkSync(opts.lockPath);
+      }
+    } catch {
+    }
+  };
+  if (opts.releaseOnExit ?? true) process.on("exit", release);
+  return { release };
+}
+
+// src/campaign/transient-failure.ts
+var BASE_TRANSIENT = /\b50[234]\b|no stream output|produced no stream|admission timed out|admission_rejected|queue_timeout|fetch failed|ECONNRESET|This operation was aborted/i;
+var TIMEOUT_PATTERN = /timeout after \d+ ?ms|cli-bridge timeout/i;
+function isTransientTransportFailure(message, opts = {}) {
+  if (!message) return false;
+  if (BASE_TRANSIENT.test(message)) return true;
+  if ((opts.retryFullDurationTimeouts ?? false) && TIMEOUT_PATTERN.test(message)) return true;
+  for (const p of opts.extraPatterns ?? []) if (p.test(message)) return true;
+  return false;
+}
+
 // src/campaign/worktree/index.ts
 import { execFileSync } from "child_process";
-import { existsSync as existsSync3 } from "fs";
+import { existsSync as existsSync4 } from "fs";
 import { basename, isAbsolute as isAbsolute2, join as join5 } from "path";
 var WorktreeAdapterError = class extends Error {
   constructor(message, cause) {
@@ -4224,7 +4322,7 @@ function gitWorktreeAdapter(opts) {
   };
 }
 function resolveWorktreePath(surface, worktreeDir) {
-  if (isAbsolute2(surface.worktreeRef) && existsSync3(surface.worktreeRef)) return surface.worktreeRef;
+  if (isAbsolute2(surface.worktreeRef) && existsSync4(surface.worktreeRef)) return surface.worktreeRef;
   if (worktreeDir) return join5(worktreeDir, basename(surface.worktreeRef));
   return surface.worktreeRef;
 }
@@ -4267,6 +4365,8 @@ export {
   neutralizationGate,
   sequentialPairedGate,
   sequentialDecide,
+  rolloutArgumentDiff,
+  classifyUngroundedLiterals,
   LabeledScenarioStoreError,
   FsLabeledScenarioStore,
   neutralizeText,
@@ -4299,8 +4399,10 @@ export {
   traceAnalystProposer,
   scoreDiscrimination,
   selectDiscriminative,
+  acquireSingleRunLock,
+  isTransientTransportFailure,
   WorktreeAdapterError,
   gitWorktreeAdapter,
   resolveWorktreePath
 };
-//# sourceMappingURL=chunk-6SKVFBTR.js.map
+//# sourceMappingURL=chunk-RNB2NICW.js.map