@tangle-network/agent-eval 0.123.7 → 0.124.0

package/dist/{chunk-J3LHTAAB.js → chunk-QOTFXW5L.js}

@@ -8,51 +8,6 @@ import {
   toolSpans
 } from "./chunk-ZET2UAYW.js";
 
-// src/trajectory.ts
-async function buildTrajectory(store, runId) {
-  const spans = await store.spans({ runId });
-  const events = await store.events({ runId });
-  const childrenOf = /* @__PURE__ */ new Map();
-  for (const s of spans) {
-    const arr = childrenOf.get(s.parentSpanId) ?? [];
-    arr.push(s);
-    childrenOf.set(s.parentSpanId, arr);
-  }
-  for (const arr of childrenOf.values()) arr.sort((a, b) => a.startedAt - b.startedAt);
-  const eventsBySpan = /* @__PURE__ */ new Map();
-  for (const e of events) {
-    if (!e.spanId) continue;
-    const arr = eventsBySpan.get(e.spanId) ?? [];
-    arr.push(e);
-    eventsBySpan.set(e.spanId, arr);
-  }
-  const steps = [];
-  const walk = (spanId, depth) => {
-    const kids = childrenOf.get(spanId) ?? [];
-    for (const child of kids) {
-      steps.push({
-        index: steps.length,
-        span: child,
-        depth,
-        events: eventsBySpan.get(child.spanId) ?? []
-      });
-      walk(child.spanId, depth + 1);
-    }
-  };
-  walk(void 0, 0);
-  const llmTurns = steps.filter((s) => s.span.kind === "llm").length;
-  const toolCalls = steps.filter((s) => s.span.kind === "tool").length;
-  const judgeVerdicts = steps.filter((s) => s.span.kind === "judge").length;
-  const retrievals = steps.filter((s) => s.span.kind === "retrieval").length;
-  let totalDurationMs = 0;
-  if (steps.length > 0) {
-    const starts = spans.map((s) => s.startedAt);
-    const ends = spans.map((s) => s.endedAt ?? s.startedAt);
-    totalDurationMs = Math.max(...ends) - Math.min(...starts);
-  }
-  return { runId, steps, llmTurns, toolCalls, judgeVerdicts, retrievals, totalDurationMs };
-}
-
 // src/failure-taxonomy.ts
 var DEFAULT_RULES = [
   // Outcome already named? Respect it.
@@ -611,7 +566,6 @@ function normalCdf(x) {
 }
 
 export {
-  buildTrajectory,
   DEFAULT_RULES,
   classifyFailure,
   computeToolUseMetrics,
@@ -619,4 +573,4 @@ export {
   iqr,
   welchsTTest
 };
-//# sourceMappingURL=chunk-J3LHTAAB.js.map
+//# sourceMappingURL=chunk-QOTFXW5L.js.map

package/dist/chunk-QOTFXW5L.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/failure-taxonomy.ts","../src/tool-use-metrics.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, hasCapturedToolArgs, 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  /** Calls whose arguments were captured and can be compared for duplication. */\n  callsWithCapturedArgs: number\n  byTool: Record<string, ToolStats>\n  errorRate: number\n  /** Ratio of captured-argument calls already seen with the same tool name and arguments. */\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  callsWithCapturedArgs: 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 {\n      runId,\n      totalCalls: 0,\n      callsWithCapturedArgs: 0,\n      byTool: {},\n      errorRate: 0,\n      duplicateRate: 0,\n      retryRate: 0,\n    }\n  }\n\n  const byTool: Record<string, ToolStats> = {}\n  let totalErrors = 0\n  let totalDuplicates = 0\n  let callsWithCapturedArgs = 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] ??= {\n      calls: 0,\n      callsWithCapturedArgs: 0,\n      errors: 0,\n      avgLatencyMs: 0,\n      duplicates: 0,\n    }\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    if (hasCapturedToolArgs(t)) {\n      callsWithCapturedArgs += 1\n      stat.callsWithCapturedArgs += 1\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\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    callsWithCapturedArgs,\n    byTool,\n    errorRate: totalErrors / sortedTools.length,\n    duplicateRate: callsWithCapturedArgs > 0 ? totalDuplicates / callsWithCapturedArgs : 0,\n    retryRate,\n    selectionAccuracy,\n  }\n}\n\nexport type { Span }\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;;;ACpaA,eAAsB,sBACpB,OACA,OACA,UAA0B,CAAC,GACF;AACzB,QAAM,QAAQ,MAAM,UAAU,OAAO,KAAK;AAC1C,MAAI,MAAM,WAAW,GAAG;AACtB,WAAO;AAAA,MACL;AAAA,MACA,YAAY;AAAA,MACZ,uBAAuB;AAAA,MACvB,QAAQ,CAAC;AAAA,MACT,WAAW;AAAA,MACX,eAAe;AAAA,MACf,WAAW;AAAA,IACb;AAAA,EACF;AAEA,QAAM,SAAoC,CAAC;AAC3C,MAAI,cAAc;AAClB,MAAI,kBAAkB;AACtB,MAAI,wBAAwB;AAC5B,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;AAAA,MACrB,OAAO;AAAA,MACP,uBAAuB;AAAA,MACvB,QAAQ;AAAA,MACR,cAAc;AAAA,MACd,YAAY;AAAA,IACd;AACA,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,QAAI,oBAAoB,CAAC,GAAG;AAC1B,+BAAyB;AACzB,WAAK,yBAAyB;AAC9B,YAAM,MAAM,GAAG,EAAE,QAAQ,IAAI,QAAQ,EAAE,IAAI,CAAC;AAC5C,UAAI,eAAe,IAAI,GAAG,GAAG;AAC3B,aAAK,cAAc;AACnB,2BAAmB;AAAA,MACrB;AACA,qBAAe,IAAI,GAAG;AAAA,IACxB;AAAA,EACF;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;AAAA,IACA,WAAW,cAAc,YAAY;AAAA,IACrC,eAAe,wBAAwB,IAAI,kBAAkB,wBAAwB;AAAA,IACrF;AAAA,IACA;AAAA,EACF;AACF;;;AChEO,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-R7ZRE2KV.js

@@ -0,0 +1,138 @@
+// src/rollout/readers/opencode-sqlite.ts
+import { homedir } from "os";
+import { join } from "path";
+var DEFAULT_OPENCODE_DB = join(homedir(), ".local", "share", "opencode", "opencode.db");
+var NODE_SQLITE_SPECIFIER = ["node", "sqlite"].join(":");
+async function openOpencodeDb(path = DEFAULT_OPENCODE_DB) {
+  try {
+    const { DatabaseSync } = await import(
+      /* @vite-ignore */
+      NODE_SQLITE_SPECIFIER
+    );
+    const db = new DatabaseSync(path, { readOnly: true });
+    db.prepare("SELECT id FROM session LIMIT 1").get();
+    return db;
+  } catch {
+    return null;
+  }
+}
+var isRecord = (v) => typeof v === "object" && v !== null && !Array.isArray(v);
+function parseSessionRow(row) {
+  let model = null;
+  if (typeof row.model === "string" && row.model.length > 0) {
+    try {
+      const parsed = JSON.parse(row.model);
+      if (isRecord(parsed)) model = parsed;
+    } catch {
+      model = null;
+    }
+  }
+  return {
+    id: String(row.id),
+    parentId: row.parent_id === null || row.parent_id === void 0 ? null : String(row.parent_id),
+    directory: String(row.directory),
+    agent: row.agent === null || row.agent === void 0 ? null : String(row.agent),
+    model,
+    costUsd: Number(row.cost ?? 0),
+    tokensInput: Number(row.tokens_input ?? 0),
+    tokensOutput: Number(row.tokens_output ?? 0),
+    tokensReasoning: Number(row.tokens_reasoning ?? 0),
+    tokensCacheRead: Number(row.tokens_cache_read ?? 0),
+    tokensCacheWrite: Number(row.tokens_cache_write ?? 0),
+    timeCreated: Number(row.time_created ?? 0),
+    timeUpdated: Number(row.time_updated ?? 0)
+  };
+}
+var SESSION_COLUMNS = "id, parent_id, directory, agent, model, cost, tokens_input, tokens_output, tokens_reasoning, tokens_cache_read, tokens_cache_write, time_created, time_updated";
+function findOpencodeSessionsByDirectory(db, directory) {
+  const rows = db.prepare(`SELECT ${SESSION_COLUMNS} FROM session WHERE directory = ? ORDER BY time_created`).all(directory);
+  return rows.map(parseSessionRow);
+}
+function findOpencodeSessionById(db, sessionId) {
+  const row = db.prepare(`SELECT ${SESSION_COLUMNS} FROM session WHERE id = ?`).get(sessionId);
+  return row === void 0 ? null : parseSessionRow(row);
+}
+function toolResultContent(output) {
+  if (typeof output === "string") return output;
+  if (output === null || output === void 0) return "";
+  return JSON.stringify(output);
+}
+function readOpencodeSessionMessages(db, sessionId) {
+  const messageRows = db.prepare("SELECT id, data FROM message WHERE session_id = ? ORDER BY time_created, id").all(sessionId);
+  const partsStmt = db.prepare("SELECT data FROM part WHERE message_id = ? ORDER BY id");
+  const messages = [];
+  for (const messageRow of messageRows) {
+    let data;
+    try {
+      const parsed = JSON.parse(messageRow.data);
+      if (!isRecord(parsed)) continue;
+      data = parsed;
+    } catch {
+      continue;
+    }
+    const parts = [];
+    for (const row of partsStmt.all(messageRow.id)) {
+      try {
+        const parsed = JSON.parse(row.data);
+        if (isRecord(parsed)) parts.push(parsed);
+      } catch {
+      }
+    }
+    if (data.role === "user") {
+      const text = parts.filter((p) => p.type === "text" && typeof p.text === "string").map((p) => p.text).join("\n");
+      messages.push({ role: "user", content: text });
+      continue;
+    }
+    if (data.role !== "assistant") continue;
+    const steps = [];
+    let current = [];
+    for (const part of parts) {
+      if (part.type === "step-start") {
+        if (current.length > 0) steps.push(current);
+        current = [];
+        continue;
+      }
+      if (part.type === "step-finish" || part.type === "snapshot" || part.type === "patch") continue;
+      current.push(part);
+    }
+    if (current.length > 0) steps.push(current);
+    for (const step of steps) {
+      const reasoning = step.filter((p) => p.type === "reasoning" && typeof p.text === "string" && p.text.length > 0).map((p) => p.text).join("\n");
+      const text = step.filter((p) => p.type === "text" && typeof p.text === "string").map((p) => p.text).join("\n");
+      const toolParts = step.filter((p) => p.type === "tool" && typeof p.callID === "string");
+      const toolCalls = toolParts.map((p) => ({
+        id: p.callID,
+        type: "function",
+        function: {
+          name: p.tool ?? "unknown",
+          arguments: JSON.stringify(p.state?.input ?? {})
+        }
+      }));
+      if (reasoning.length === 0 && text.length === 0 && toolCalls.length === 0) continue;
+      messages.push({
+        role: "assistant",
+        content: text.length > 0 ? text : null,
+        ...reasoning.length > 0 ? { reasoning_content: reasoning } : {},
+        ...toolCalls.length > 0 ? { tool_calls: toolCalls } : {}
+      });
+      for (const p of toolParts) {
+        messages.push({
+          role: "tool",
+          tool_call_id: p.callID,
+          name: p.tool ?? "unknown",
+          content: toolResultContent(p.state?.output)
+        });
+      }
+    }
+  }
+  return messages;
+}
+
+export {
+  DEFAULT_OPENCODE_DB,
+  openOpencodeDb,
+  findOpencodeSessionsByDirectory,
+  findOpencodeSessionById,
+  readOpencodeSessionMessages
+};
+//# sourceMappingURL=chunk-R7ZRE2KV.js.map

package/dist/chunk-R7ZRE2KV.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/rollout/readers/opencode-sqlite.ts"],"sourcesContent":["/**\n * Read-only backfill reader over the opencode sqlite store\n * (~/.local/share/opencode/opencode.db) → canonical chat-with-tools messages.\n *\n * Schema consumed (observed, 2026-07): `session` rows carry directory /\n * parent_id / agent / model / cost / tokens_*; `message` rows carry a JSON\n * `data` blob ({role, modelID, providerID, tokens, cost, finish}); `part`\n * rows carry the actual content ({type: text|reasoning|tool|step-start|\n * step-finish|snapshot…}). Tool parts hold {callID, state:{input, output,\n * status}} — both the call and its result, which we split into an assistant\n * tool_call plus a role:\"tool\" result message.\n *\n * The store is mutable and can be corrupt (a `.corrupt-bak` sibling ships\n * next to it in the wild), so `openOpencodeDb` returns null instead of\n * throwing — callers record a gap line, never crash the backfill.\n */\n\nimport { homedir } from 'node:os'\nimport { join } from 'node:path'\nimport type { DatabaseSync } from 'node:sqlite'\nimport type { ChatMessage, ChatToolCall } from '../schema'\n\nexport const DEFAULT_OPENCODE_DB = join(homedir(), '.local', 'share', 'opencode', 'opencode.db')\n\nexport interface OpencodeSessionRow {\n  id: string\n  parentId: string | null\n  directory: string\n  agent: string | null\n  /** Raw session.model JSON: {id, providerID, variant} where present. */\n  model: { id?: string; providerID?: string } | null\n  costUsd: number\n  tokensInput: number\n  tokensOutput: number\n  tokensReasoning: number\n  tokensCacheRead: number\n  tokensCacheWrite: number\n  timeCreated: number\n  timeUpdated: number\n}\n\n// Opaque specifier: esbuild (bundling) and Vite (tests) both rewrite an\n// analyzable dynamic import and strip the `node:` prefix under an es20xx\n// target, which turns this builtin into a bogus \"sqlite\" package lookup.\n// Composing the string at runtime defeats that analysis in both.\nconst NODE_SQLITE_SPECIFIER = ['node', 'sqlite'].join(':')\n\n/** Open the store read-only; null = unavailable/corrupt (caller records a gap). */\nexport async function openOpencodeDb(\n  path: string = DEFAULT_OPENCODE_DB,\n): Promise<DatabaseSync | null> {\n  try {\n    const { DatabaseSync } = (await import(\n      /* @vite-ignore */ NODE_SQLITE_SPECIFIER\n    )) as typeof import('node:sqlite')\n    const db = new DatabaseSync(path, { readOnly: true })\n    // Probe: a corrupt store can open() fine and fail on first page read.\n    db.prepare('SELECT id FROM session LIMIT 1').get()\n    return db\n  } catch {\n    return null\n  }\n}\n\nconst isRecord = (v: unknown): v is Record<string, unknown> =>\n  typeof v === 'object' && v !== null && !Array.isArray(v)\n\nfunction parseSessionRow(row: Record<string, unknown>): OpencodeSessionRow {\n  let model: OpencodeSessionRow['model'] = null\n  if (typeof row.model === 'string' && row.model.length > 0) {\n    try {\n      const parsed: unknown = JSON.parse(row.model)\n      if (isRecord(parsed)) model = parsed as { id?: string; providerID?: string }\n    } catch {\n      model = null\n    }\n  }\n  return {\n    id: String(row.id),\n    parentId: row.parent_id === null || row.parent_id === undefined ? null : String(row.parent_id),\n    directory: String(row.directory),\n    agent: row.agent === null || row.agent === undefined ? null : String(row.agent),\n    model,\n    costUsd: Number(row.cost ?? 0),\n    tokensInput: Number(row.tokens_input ?? 0),\n    tokensOutput: Number(row.tokens_output ?? 0),\n    tokensReasoning: Number(row.tokens_reasoning ?? 0),\n    tokensCacheRead: Number(row.tokens_cache_read ?? 0),\n    tokensCacheWrite: Number(row.tokens_cache_write ?? 0),\n    timeCreated: Number(row.time_created ?? 0),\n    timeUpdated: Number(row.time_updated ?? 0),\n  }\n}\n\nconst SESSION_COLUMNS =\n  'id, parent_id, directory, agent, model, cost, tokens_input, tokens_output, tokens_reasoning, tokens_cache_read, tokens_cache_write, time_created, time_updated'\n\n/** Sessions whose cwd is `directory` (the worker-clone join key). */\nexport function findOpencodeSessionsByDirectory(\n  db: DatabaseSync,\n  directory: string,\n): OpencodeSessionRow[] {\n  const rows = db\n    .prepare(`SELECT ${SESSION_COLUMNS} FROM session WHERE directory = ? ORDER BY time_created`)\n    .all(directory) as Array<Record<string, unknown>>\n  return rows.map(parseSessionRow)\n}\n\nexport function findOpencodeSessionById(\n  db: DatabaseSync,\n  sessionId: string,\n): OpencodeSessionRow | null {\n  const row = db.prepare(`SELECT ${SESSION_COLUMNS} FROM session WHERE id = ?`).get(sessionId) as\n    | Record<string, unknown>\n    | undefined\n  return row === undefined ? null : parseSessionRow(row)\n}\n\ninterface OpencodePart {\n  type?: string\n  text?: string\n  tool?: string\n  callID?: string\n  state?: { status?: string; input?: unknown; output?: unknown }\n}\n\nfunction toolResultContent(output: unknown): string {\n  if (typeof output === 'string') return output\n  if (output === null || output === undefined) return ''\n  return JSON.stringify(output)\n}\n\n/**\n * Convert one session's message+part rows into canonical messages.\n * An opencode assistant message row spans several model steps; each step's\n * parts (reasoning → text → tool …) become one assistant message followed by\n * the role:\"tool\" results of its calls, preserving order.\n */\nexport function readOpencodeSessionMessages(db: DatabaseSync, sessionId: string): ChatMessage[] {\n  const messageRows = db\n    .prepare('SELECT id, data FROM message WHERE session_id = ? ORDER BY time_created, id')\n    .all(sessionId) as Array<{ id: string; data: string }>\n  const partsStmt = db.prepare('SELECT data FROM part WHERE message_id = ? ORDER BY id')\n\n  const messages: ChatMessage[] = []\n  for (const messageRow of messageRows) {\n    let data: Record<string, unknown>\n    try {\n      const parsed: unknown = JSON.parse(messageRow.data)\n      if (!isRecord(parsed)) continue\n      data = parsed\n    } catch {\n      continue\n    }\n    const parts: OpencodePart[] = []\n    for (const row of partsStmt.all(messageRow.id) as Array<{ data: string }>) {\n      try {\n        const parsed: unknown = JSON.parse(row.data)\n        if (isRecord(parsed)) parts.push(parsed as OpencodePart)\n      } catch {\n        // Malformed part payload: skip the part, keep the message.\n      }\n    }\n\n    if (data.role === 'user') {\n      const text = parts\n        .filter((p) => p.type === 'text' && typeof p.text === 'string')\n        .map((p) => p.text as string)\n        .join('\\n')\n      messages.push({ role: 'user', content: text })\n      continue\n    }\n    if (data.role !== 'assistant') continue\n\n    // Split the row into steps at step-start boundaries; parts before the\n    // first step-start (none observed, but tolerated) form an implicit step.\n    const steps: OpencodePart[][] = []\n    let current: OpencodePart[] = []\n    for (const part of parts) {\n      if (part.type === 'step-start') {\n        if (current.length > 0) steps.push(current)\n        current = []\n        continue\n      }\n      if (part.type === 'step-finish' || part.type === 'snapshot' || part.type === 'patch') continue\n      current.push(part)\n    }\n    if (current.length > 0) steps.push(current)\n\n    for (const step of steps) {\n      const reasoning = step\n        .filter((p) => p.type === 'reasoning' && typeof p.text === 'string' && p.text.length > 0)\n        .map((p) => p.text as string)\n        .join('\\n')\n      const text = step\n        .filter((p) => p.type === 'text' && typeof p.text === 'string')\n        .map((p) => p.text as string)\n        .join('\\n')\n      const toolParts = step.filter((p) => p.type === 'tool' && typeof p.callID === 'string')\n      const toolCalls: ChatToolCall[] = toolParts.map((p) => ({\n        id: p.callID as string,\n        type: 'function',\n        function: {\n          name: p.tool ?? 'unknown',\n          arguments: JSON.stringify(p.state?.input ?? {}),\n        },\n      }))\n      if (reasoning.length === 0 && text.length === 0 && toolCalls.length === 0) continue\n      messages.push({\n        role: 'assistant',\n        content: text.length > 0 ? text : null,\n        ...(reasoning.length > 0 ? { reasoning_content: reasoning } : {}),\n        ...(toolCalls.length > 0 ? { tool_calls: toolCalls } : {}),\n      })\n      for (const p of toolParts) {\n        messages.push({\n          role: 'tool',\n          tool_call_id: p.callID as string,\n          name: p.tool ?? 'unknown',\n          content: toolResultContent(p.state?.output),\n        })\n      }\n    }\n  }\n  return messages\n}\n"],"mappings":";AAiBA,SAAS,eAAe;AACxB,SAAS,YAAY;AAId,IAAM,sBAAsB,KAAK,QAAQ,GAAG,UAAU,SAAS,YAAY,aAAa;AAuB/F,IAAM,wBAAwB,CAAC,QAAQ,QAAQ,EAAE,KAAK,GAAG;AAGzD,eAAsB,eACpB,OAAe,qBACe;AAC9B,MAAI;AACF,UAAM,EAAE,aAAa,IAAK,MAAM;AAAA;AAAA,MACX;AAAA;AAErB,UAAM,KAAK,IAAI,aAAa,MAAM,EAAE,UAAU,KAAK,CAAC;AAEpD,OAAG,QAAQ,gCAAgC,EAAE,IAAI;AACjD,WAAO;AAAA,EACT,QAAQ;AACN,WAAO;AAAA,EACT;AACF;AAEA,IAAM,WAAW,CAAC,MAChB,OAAO,MAAM,YAAY,MAAM,QAAQ,CAAC,MAAM,QAAQ,CAAC;AAEzD,SAAS,gBAAgB,KAAkD;AACzE,MAAI,QAAqC;AACzC,MAAI,OAAO,IAAI,UAAU,YAAY,IAAI,MAAM,SAAS,GAAG;AACzD,QAAI;AACF,YAAM,SAAkB,KAAK,MAAM,IAAI,KAAK;AAC5C,UAAI,SAAS,MAAM,EAAG,SAAQ;AAAA,IAChC,QAAQ;AACN,cAAQ;AAAA,IACV;AAAA,EACF;AACA,SAAO;AAAA,IACL,IAAI,OAAO,IAAI,EAAE;AAAA,IACjB,UAAU,IAAI,cAAc,QAAQ,IAAI,cAAc,SAAY,OAAO,OAAO,IAAI,SAAS;AAAA,IAC7F,WAAW,OAAO,IAAI,SAAS;AAAA,IAC/B,OAAO,IAAI,UAAU,QAAQ,IAAI,UAAU,SAAY,OAAO,OAAO,IAAI,KAAK;AAAA,IAC9E;AAAA,IACA,SAAS,OAAO,IAAI,QAAQ,CAAC;AAAA,IAC7B,aAAa,OAAO,IAAI,gBAAgB,CAAC;AAAA,IACzC,cAAc,OAAO,IAAI,iBAAiB,CAAC;AAAA,IAC3C,iBAAiB,OAAO,IAAI,oBAAoB,CAAC;AAAA,IACjD,iBAAiB,OAAO,IAAI,qBAAqB,CAAC;AAAA,IAClD,kBAAkB,OAAO,IAAI,sBAAsB,CAAC;AAAA,IACpD,aAAa,OAAO,IAAI,gBAAgB,CAAC;AAAA,IACzC,aAAa,OAAO,IAAI,gBAAgB,CAAC;AAAA,EAC3C;AACF;AAEA,IAAM,kBACJ;AAGK,SAAS,gCACd,IACA,WACsB;AACtB,QAAM,OAAO,GACV,QAAQ,UAAU,eAAe,yDAAyD,EAC1F,IAAI,SAAS;AAChB,SAAO,KAAK,IAAI,eAAe;AACjC;AAEO,SAAS,wBACd,IACA,WAC2B;AAC3B,QAAM,MAAM,GAAG,QAAQ,UAAU,eAAe,4BAA4B,EAAE,IAAI,SAAS;AAG3F,SAAO,QAAQ,SAAY,OAAO,gBAAgB,GAAG;AACvD;AAUA,SAAS,kBAAkB,QAAyB;AAClD,MAAI,OAAO,WAAW,SAAU,QAAO;AACvC,MAAI,WAAW,QAAQ,WAAW,OAAW,QAAO;AACpD,SAAO,KAAK,UAAU,MAAM;AAC9B;AAQO,SAAS,4BAA4B,IAAkB,WAAkC;AAC9F,QAAM,cAAc,GACjB,QAAQ,6EAA6E,EACrF,IAAI,SAAS;AAChB,QAAM,YAAY,GAAG,QAAQ,wDAAwD;AAErF,QAAM,WAA0B,CAAC;AACjC,aAAW,cAAc,aAAa;AACpC,QAAI;AACJ,QAAI;AACF,YAAM,SAAkB,KAAK,MAAM,WAAW,IAAI;AAClD,UAAI,CAAC,SAAS,MAAM,EAAG;AACvB,aAAO;AAAA,IACT,QAAQ;AACN;AAAA,IACF;AACA,UAAM,QAAwB,CAAC;AAC/B,eAAW,OAAO,UAAU,IAAI,WAAW,EAAE,GAA8B;AACzE,UAAI;AACF,cAAM,SAAkB,KAAK,MAAM,IAAI,IAAI;AAC3C,YAAI,SAAS,MAAM,EAAG,OAAM,KAAK,MAAsB;AAAA,MACzD,QAAQ;AAAA,MAER;AAAA,IACF;AAEA,QAAI,KAAK,SAAS,QAAQ;AACxB,YAAM,OAAO,MACV,OAAO,CAAC,MAAM,EAAE,SAAS,UAAU,OAAO,EAAE,SAAS,QAAQ,EAC7D,IAAI,CAAC,MAAM,EAAE,IAAc,EAC3B,KAAK,IAAI;AACZ,eAAS,KAAK,EAAE,MAAM,QAAQ,SAAS,KAAK,CAAC;AAC7C;AAAA,IACF;AACA,QAAI,KAAK,SAAS,YAAa;AAI/B,UAAM,QAA0B,CAAC;AACjC,QAAI,UAA0B,CAAC;AAC/B,eAAW,QAAQ,OAAO;AACxB,UAAI,KAAK,SAAS,cAAc;AAC9B,YAAI,QAAQ,SAAS,EAAG,OAAM,KAAK,OAAO;AAC1C,kBAAU,CAAC;AACX;AAAA,MACF;AACA,UAAI,KAAK,SAAS,iBAAiB,KAAK,SAAS,cAAc,KAAK,SAAS,QAAS;AACtF,cAAQ,KAAK,IAAI;AAAA,IACnB;AACA,QAAI,QAAQ,SAAS,EAAG,OAAM,KAAK,OAAO;AAE1C,eAAW,QAAQ,OAAO;AACxB,YAAM,YAAY,KACf,OAAO,CAAC,MAAM,EAAE,SAAS,eAAe,OAAO,EAAE,SAAS,YAAY,EAAE,KAAK,SAAS,CAAC,EACvF,IAAI,CAAC,MAAM,EAAE,IAAc,EAC3B,KAAK,IAAI;AACZ,YAAM,OAAO,KACV,OAAO,CAAC,MAAM,EAAE,SAAS,UAAU,OAAO,EAAE,SAAS,QAAQ,EAC7D,IAAI,CAAC,MAAM,EAAE,IAAc,EAC3B,KAAK,IAAI;AACZ,YAAM,YAAY,KAAK,OAAO,CAAC,MAAM,EAAE,SAAS,UAAU,OAAO,EAAE,WAAW,QAAQ;AACtF,YAAM,YAA4B,UAAU,IAAI,CAAC,OAAO;AAAA,QACtD,IAAI,EAAE;AAAA,QACN,MAAM;AAAA,QACN,UAAU;AAAA,UACR,MAAM,EAAE,QAAQ;AAAA,UAChB,WAAW,KAAK,UAAU,EAAE,OAAO,SAAS,CAAC,CAAC;AAAA,QAChD;AAAA,MACF,EAAE;AACF,UAAI,UAAU,WAAW,KAAK,KAAK,WAAW,KAAK,UAAU,WAAW,EAAG;AAC3E,eAAS,KAAK;AAAA,QACZ,MAAM;AAAA,QACN,SAAS,KAAK,SAAS,IAAI,OAAO;AAAA,QAClC,GAAI,UAAU,SAAS,IAAI,EAAE,mBAAmB,UAAU,IAAI,CAAC;AAAA,QAC/D,GAAI,UAAU,SAAS,IAAI,EAAE,YAAY,UAAU,IAAI,CAAC;AAAA,MAC1D,CAAC;AACD,iBAAW,KAAK,WAAW;AACzB,iBAAS,KAAK;AAAA,UACZ,MAAM;AAAA,UACN,cAAc,EAAE;AAAA,UAChB,MAAM,EAAE,QAAQ;AAAA,UAChB,SAAS,kBAAkB,EAAE,OAAO,MAAM;AAAA,QAC5C,CAAC;AAAA,MACH;AAAA,IACF;AAAA,EACF;AACA,SAAO;AACT;","names":[]}

package/dist/chunk-RZTMDUO7.js

@@ -0,0 +1,49 @@
+// src/trajectory.ts
+async function buildTrajectory(store, runId) {
+  const spans = await store.spans({ runId });
+  const events = await store.events({ runId });
+  const childrenOf = /* @__PURE__ */ new Map();
+  for (const s of spans) {
+    const arr = childrenOf.get(s.parentSpanId) ?? [];
+    arr.push(s);
+    childrenOf.set(s.parentSpanId, arr);
+  }
+  for (const arr of childrenOf.values()) arr.sort((a, b) => a.startedAt - b.startedAt);
+  const eventsBySpan = /* @__PURE__ */ new Map();
+  for (const e of events) {
+    if (!e.spanId) continue;
+    const arr = eventsBySpan.get(e.spanId) ?? [];
+    arr.push(e);
+    eventsBySpan.set(e.spanId, arr);
+  }
+  const steps = [];
+  const walk = (spanId, depth) => {
+    const kids = childrenOf.get(spanId) ?? [];
+    for (const child of kids) {
+      steps.push({
+        index: steps.length,
+        span: child,
+        depth,
+        events: eventsBySpan.get(child.spanId) ?? []
+      });
+      walk(child.spanId, depth + 1);
+    }
+  };
+  walk(void 0, 0);
+  const llmTurns = steps.filter((s) => s.span.kind === "llm").length;
+  const toolCalls = steps.filter((s) => s.span.kind === "tool").length;
+  const judgeVerdicts = steps.filter((s) => s.span.kind === "judge").length;
+  const retrievals = steps.filter((s) => s.span.kind === "retrieval").length;
+  let totalDurationMs = 0;
+  if (steps.length > 0) {
+    const starts = spans.map((s) => s.startedAt);
+    const ends = spans.map((s) => s.endedAt ?? s.startedAt);
+    totalDurationMs = Math.max(...ends) - Math.min(...starts);
+  }
+  return { runId, steps, llmTurns, toolCalls, judgeVerdicts, retrievals, totalDurationMs };
+}
+
+export {
+  buildTrajectory
+};
+//# sourceMappingURL=chunk-RZTMDUO7.js.map

package/dist/chunk-RZTMDUO7.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/trajectory.ts"],"sourcesContent":["/**\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"],"mappings":";AAgCA,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;","names":[]}

package/dist/{chunk-QNVBALEX.js → chunk-W5B3ZGP3.js}

@@ -3,13 +3,13 @@ import {
   aggregateRunScore,
   clamp01,
   computeFindingId
-} from "./chunk-MHPEGJHC.js";
+} from "./chunk-DT7OXY3C.js";
 import {
   callLlmJson,
   costReceiptFromLlm,
   costReceiptFromLlmError,
   maximumChargeForLlmRequest
-} from "./chunk-MK3Q6GCL.js";
+} from "./chunk-EQUK3RFS.js";
 import {
   CostLedger
 } from "./chunk-GRCDRKII.js";
@@ -747,4 +747,4 @@ export {
   runSemanticConceptJudge,
   createSemanticConceptJudge
 };
-//# sourceMappingURL=chunk-QNVBALEX.js.map
+//# sourceMappingURL=chunk-W5B3ZGP3.js.map

package/dist/cli.js

@@ -5,8 +5,12 @@ import {
   runRpcBatch,
   runRpcOnce,
   startServerAsync
-} from "./chunk-4KO6D2BO.js";
-import "./chunk-MK3Q6GCL.js";
+} from "./chunk-HM6V7F3M.js";
+import {
+  runRolloutReleaseCli
+} from "./chunk-IPYXE555.js";
+import "./chunk-MAX3TN3C.js";
+import "./chunk-EQUK3RFS.js";
 import "./chunk-GRCDRKII.js";
 import "./chunk-VI2UW6B6.js";
 import "./chunk-PC4UYEBM.js";
@@ -79,6 +83,9 @@ Commands:
         Like 'rpc' but JSONL in / JSONL out.
   openapi [--out openapi.json]
         Write the OpenAPI 3.1 spec.
+  rollout-release <ledger.jsonl...> --out <dir> [--formats sft,verifiers,rft,raw] [--include-proposers] [--push <org/name>]
+        Build a HuggingFace-ready dataset dir from tangle.rollout.v1 ledgers:
+        validate, fail-closed split filter, scrub, export formats + card.
   version
         Print server + wire-protocol version JSON.
 
@@ -144,6 +151,9 @@ async function main() {
       console.log(`[agent-eval] wrote OpenAPI 3.1 spec to ${out}`);
       return 0;
     }
+    case "rollout-release": {
+      return await runRolloutReleaseCli(process.argv.slice(3));
+    }
     case "version": {
       process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}
 `);
@@ -181,6 +191,7 @@ var FLAGS_BY_COMMAND = {
   "": /* @__PURE__ */ new Set()
 };
 function assertKnownFlags(command, flags) {
+  if (command === "rollout-release") return;
   const allowed = FLAGS_BY_COMMAND[command];
   if (!allowed) return;
   const unknown = Object.keys(flags).filter((flag) => !allowed.has(flag));

package/dist/cli.js.map

@@ -1 +1 @@
-{"version":3,"sources":["../src/cli.ts","../src/cli-config.ts"],"sourcesContent":["#!/usr/bin/env node\n/**\n * agent-eval CLI.\n *\n *   agent-eval serve [--port 5005] [--host 127.0.0.1]\n *   agent-eval rpc <method>          # one request from stdin → one response on stdout\n *   agent-eval rpc-batch <method>    # JSONL stdin → JSONL stdout\n *   agent-eval openapi [--out path]  # write OpenAPI spec\n *   agent-eval version\n *\n * <method> is one of: judge, listRubrics, version. When omitted, the\n * stdin payload must be a full {method, params} envelope.\n */\nimport { writeFileSync } from 'node:fs'\nimport { resolveCliLlmConfig } from './cli-config'\nimport { handleVersion } from './wire/handlers'\nimport { buildOpenApi } from './wire/openapi'\nimport { runRpcBatch, runRpcOnce } from './wire/rpc'\nimport { startServerAsync } from './wire/server'\n\ninterface Args {\n  command: string\n  positional: string[]\n  flags: Record<string, string>\n}\n\nfunction parseArgs(argv: string[]): Args {\n  const [command, ...rest] = argv\n  const positional: string[] = []\n  const flags: Record<string, string> = {}\n  for (let i = 0; i < rest.length; i++) {\n    const tok = rest[i]!\n    if (tok.startsWith('--')) {\n      const raw = tok.slice(2)\n      const equalsAt = raw.indexOf('=')\n      if (equalsAt >= 0) {\n        flags[raw.slice(0, equalsAt)] = raw.slice(equalsAt + 1)\n        continue\n      }\n      const key = raw\n      if (key === 'help') {\n        flags[key] = 'true'\n        continue\n      }\n      const next = rest[i + 1]\n      if (next != null && !next.startsWith('--')) {\n        flags[key] = next\n        i++\n      } else {\n        flags[key] = 'true'\n      }\n    } else if (tok === '-h') {\n      flags.help = 'true'\n    } else {\n      positional.push(tok)\n    }\n  }\n  return { command: command ?? 'help', positional, flags }\n}\n\nconst HELP = `agent-eval: evaluation RPC and HTTP server.\n\nCommands:\n  serve [--port 5005] [--host 127.0.0.1]\n        Start the HTTP server. POST /v1/judge, GET /v1/rubrics, GET /v1/version, GET /openapi.json.\n  rpc <method>\n        Read one JSON object from stdin (the params for <method>), write one\n        JSON object to stdout. Methods: judge, listRubrics, version.\n  rpc-batch <method>\n        Like 'rpc' but JSONL in / JSONL out.\n  openapi [--out openapi.json]\n        Write the OpenAPI 3.1 spec.\n  version\n        Print server + wire-protocol version JSON.\n\nJudge provider:\n  Set AGENT_EVAL_LLM_BASE_URL, AGENT_EVAL_LLM_API_KEY, and AGENT_EVAL_LLM_MODEL.\n  OPENAI_* and TANGLE_* equivalents are also accepted.\n\nWithout arguments, prints this help.`\n\nasync function main(): Promise<number> {\n  const { command, positional, flags } = parseArgs(process.argv.slice(2))\n  assertKnownFlags(command, flags)\n\n  if (flags.help === 'true') {\n    process.stdout.write(`${HELP}\\n`)\n    return 0\n  }\n\n  switch (command) {\n    case 'serve': {\n      const port = parsePort(flags.port ?? '5005')\n      const host = flags.host ?? '127.0.0.1'\n      const llm = resolveCliLlmConfig()\n      const { server } = await startServerAsync({\n        port,\n        host,\n        llm: llm.client,\n        judgeModel: llm.model,\n        llmRouteRequirements: { requireExplicitBaseUrl: true },\n      })\n      // Keep process alive on SIGINT/SIGTERM\n      const shutdown = (sig: string) => {\n        // eslint-disable-next-line no-console\n        console.log(`[agent-eval] received ${sig}, shutting down`)\n        server.close(() => process.exit(0))\n        // Force exit after 5s if close hangs\n        setTimeout(() => process.exit(1), 5000).unref()\n      }\n      process.on('SIGINT', () => shutdown('SIGINT'))\n      process.on('SIGTERM', () => shutdown('SIGTERM'))\n      // Block forever\n      await new Promise(() => {})\n      return 0\n    }\n    case 'rpc': {\n      const [method] = positional\n      const llm = resolveCliLlmConfig()\n      return await runRpcOnce(method, {\n        llm: llm.client,\n        judgeModel: llm.model,\n        llmRouteRequirements: { requireExplicitBaseUrl: true },\n      })\n    }\n    case 'rpc-batch': {\n      const [method] = positional\n      const llm = resolveCliLlmConfig()\n      return await runRpcBatch(method, {\n        llm: llm.client,\n        judgeModel: llm.model,\n        llmRouteRequirements: { requireExplicitBaseUrl: true },\n      })\n    }\n    case 'openapi': {\n      const out = flags.out ?? 'openapi.json'\n      const spec = buildOpenApi(handleVersion().version)\n      writeFileSync(out, `${JSON.stringify(spec, null, 2)}\\n`, 'utf-8')\n      // eslint-disable-next-line no-console\n      console.log(`[agent-eval] wrote OpenAPI 3.1 spec to ${out}`)\n      return 0\n    }\n    case 'version': {\n      process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}\\n`)\n      return 0\n    }\n    case '--version': {\n      process.stdout.write(`${handleVersion().version}\\n`)\n      return 0\n    }\n    case 'help':\n    case '--help':\n    case '-h':\n    case '':\n      process.stdout.write(`${HELP}\\n`)\n      return 0\n    default:\n      process.stderr.write(`unknown command: ${command}\\n${HELP}\\n`)\n      return 1\n  }\n}\n\nconst FLAGS_BY_COMMAND: Record<string, ReadonlySet<string>> = {\n  serve: new Set(['help', 'host', 'port']),\n  rpc: new Set(['help']),\n  'rpc-batch': new Set(['help']),\n  openapi: new Set(['help', 'out']),\n  version: new Set(['help']),\n  help: new Set(),\n  '--help': new Set(),\n  '-h': new Set(),\n  '--version': new Set(),\n  '': new Set(),\n}\n\nfunction assertKnownFlags(command: string, flags: Record<string, string>): void {\n  const allowed = FLAGS_BY_COMMAND[command]\n  if (!allowed) return\n  const unknown = Object.keys(flags).filter((flag) => !allowed.has(flag))\n  if (unknown.length > 0) {\n    throw new Error(`unknown flag for ${command || 'help'}: --${unknown[0]}`)\n  }\n}\n\nfunction parsePort(raw: string): number {\n  const port = Number(raw)\n  if (!Number.isInteger(port) || port < 0 || port > 65_535) {\n    throw new Error(`--port must be an integer from 0 to 65535; received ${JSON.stringify(raw)}`)\n  }\n  return port\n}\n\nmain()\n  .then((code) => process.exit(code))\n  .catch((err) => {\n    // eslint-disable-next-line no-console\n    console.error('[agent-eval] cli error:', err)\n    process.exit(1)\n  })\n","import type { LlmClientOptions } from './llm-client'\n\nexport interface CliLlmConfig {\n  client?: LlmClientOptions\n  model?: string\n}\n\nexport function resolveCliLlmConfig(env: NodeJS.ProcessEnv = process.env): CliLlmConfig {\n  const explicitBaseUrl = nonEmpty(env.AGENT_EVAL_LLM_BASE_URL)\n  const explicitApiKey = nonEmpty(env.AGENT_EVAL_LLM_API_KEY)\n  const openAiApiKey = nonEmpty(env.OPENAI_API_KEY)\n  const tangleApiKey = nonEmpty(env.TANGLE_API_KEY)\n  const baseUrl =\n    explicitBaseUrl ??\n    nonEmpty(env.OPENAI_BASE_URL) ??\n    nonEmpty(env.TANGLE_ROUTER_URL) ??\n    (openAiApiKey ? 'https://api.openai.com/v1' : undefined) ??\n    (tangleApiKey ? 'https://router.tangle.tools/v1' : undefined)\n  const apiKey = explicitApiKey ?? openAiApiKey ?? tangleApiKey\n  const model =\n    nonEmpty(env.AGENT_EVAL_LLM_MODEL) ?? nonEmpty(env.OPENAI_MODEL) ?? nonEmpty(env.TANGLE_MODEL)\n\n  const client =\n    baseUrl || apiKey\n      ? { ...(baseUrl ? { baseUrl } : {}), ...(apiKey ? { apiKey } : {}) }\n      : undefined\n  return { ...(client ? { client } : {}), ...(model ? { model } : {}) }\n}\n\nfunction nonEmpty(value: string | undefined): string | undefined {\n  const trimmed = value?.trim()\n  return trimmed ? trimmed : undefined\n}\n"],"mappings":";;;;;;;;;;;;;;;;AAaA,SAAS,qBAAqB;;;ACNvB,SAAS,oBAAoB,MAAyB,QAAQ,KAAmB;AACtF,QAAM,kBAAkB,SAAS,IAAI,uBAAuB;AAC5D,QAAM,iBAAiB,SAAS,IAAI,sBAAsB;AAC1D,QAAM,eAAe,SAAS,IAAI,cAAc;AAChD,QAAM,eAAe,SAAS,IAAI,cAAc;AAChD,QAAM,UACJ,mBACA,SAAS,IAAI,eAAe,KAC5B,SAAS,IAAI,iBAAiB,MAC7B,eAAe,8BAA8B,YAC7C,eAAe,mCAAmC;AACrD,QAAM,SAAS,kBAAkB,gBAAgB;AACjD,QAAM,QACJ,SAAS,IAAI,oBAAoB,KAAK,SAAS,IAAI,YAAY,KAAK,SAAS,IAAI,YAAY;AAE/F,QAAM,SACJ,WAAW,SACP,EAAE,GAAI,UAAU,EAAE,QAAQ,IAAI,CAAC,GAAI,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC,EAAG,IACjE;AACN,SAAO,EAAE,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC,GAAI,GAAI,QAAQ,EAAE,MAAM,IAAI,CAAC,EAAG;AACtE;AAEA,SAAS,SAAS,OAA+C;AAC/D,QAAM,UAAU,OAAO,KAAK;AAC5B,SAAO,UAAU,UAAU;AAC7B;;;ADNA,SAAS,UAAU,MAAsB;AACvC,QAAM,CAAC,SAAS,GAAG,IAAI,IAAI;AAC3B,QAAM,aAAuB,CAAC;AAC9B,QAAM,QAAgC,CAAC;AACvC,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;AACpC,UAAM,MAAM,KAAK,CAAC;AAClB,QAAI,IAAI,WAAW,IAAI,GAAG;AACxB,YAAM,MAAM,IAAI,MAAM,CAAC;AACvB,YAAM,WAAW,IAAI,QAAQ,GAAG;AAChC,UAAI,YAAY,GAAG;AACjB,cAAM,IAAI,MAAM,GAAG,QAAQ,CAAC,IAAI,IAAI,MAAM,WAAW,CAAC;AACtD;AAAA,MACF;AACA,YAAM,MAAM;AACZ,UAAI,QAAQ,QAAQ;AAClB,cAAM,GAAG,IAAI;AACb;AAAA,MACF;AACA,YAAM,OAAO,KAAK,IAAI,CAAC;AACvB,UAAI,QAAQ,QAAQ,CAAC,KAAK,WAAW,IAAI,GAAG;AAC1C,cAAM,GAAG,IAAI;AACb;AAAA,MACF,OAAO;AACL,cAAM,GAAG,IAAI;AAAA,MACf;AAAA,IACF,WAAW,QAAQ,MAAM;AACvB,YAAM,OAAO;AAAA,IACf,OAAO;AACL,iBAAW,KAAK,GAAG;AAAA,IACrB;AAAA,EACF;AACA,SAAO,EAAE,SAAS,WAAW,QAAQ,YAAY,MAAM;AACzD;AAEA,IAAM,OAAO;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAqBb,eAAe,OAAwB;AACrC,QAAM,EAAE,SAAS,YAAY,MAAM,IAAI,UAAU,QAAQ,KAAK,MAAM,CAAC,CAAC;AACtE,mBAAiB,SAAS,KAAK;AAE/B,MAAI,MAAM,SAAS,QAAQ;AACzB,YAAQ,OAAO,MAAM,GAAG,IAAI;AAAA,CAAI;AAChC,WAAO;AAAA,EACT;AAEA,UAAQ,SAAS;AAAA,IACf,KAAK,SAAS;AACZ,YAAM,OAAO,UAAU,MAAM,QAAQ,MAAM;AAC3C,YAAM,OAAO,MAAM,QAAQ;AAC3B,YAAM,MAAM,oBAAoB;AAChC,YAAM,EAAE,OAAO,IAAI,MAAM,iBAAiB;AAAA,QACxC;AAAA,QACA;AAAA,QACA,KAAK,IAAI;AAAA,QACT,YAAY,IAAI;AAAA,QAChB,sBAAsB,EAAE,wBAAwB,KAAK;AAAA,MACvD,CAAC;AAED,YAAM,WAAW,CAAC,QAAgB;AAEhC,gBAAQ,IAAI,yBAAyB,GAAG,iBAAiB;AACzD,eAAO,MAAM,MAAM,QAAQ,KAAK,CAAC,CAAC;AAElC,mBAAW,MAAM,QAAQ,KAAK,CAAC,GAAG,GAAI,EAAE,MAAM;AAAA,MAChD;AACA,cAAQ,GAAG,UAAU,MAAM,SAAS,QAAQ,CAAC;AAC7C,cAAQ,GAAG,WAAW,MAAM,SAAS,SAAS,CAAC;AAE/C,YAAM,IAAI,QAAQ,MAAM;AAAA,MAAC,CAAC;AAC1B,aAAO;AAAA,IACT;AAAA,IACA,KAAK,OAAO;AACV,YAAM,CAAC,MAAM,IAAI;AACjB,YAAM,MAAM,oBAAoB;AAChC,aAAO,MAAM,WAAW,QAAQ;AAAA,QAC9B,KAAK,IAAI;AAAA,QACT,YAAY,IAAI;AAAA,QAChB,sBAAsB,EAAE,wBAAwB,KAAK;AAAA,MACvD,CAAC;AAAA,IACH;AAAA,IACA,KAAK,aAAa;AAChB,YAAM,CAAC,MAAM,IAAI;AACjB,YAAM,MAAM,oBAAoB;AAChC,aAAO,MAAM,YAAY,QAAQ;AAAA,QAC/B,KAAK,IAAI;AAAA,QACT,YAAY,IAAI;AAAA,QAChB,sBAAsB,EAAE,wBAAwB,KAAK;AAAA,MACvD,CAAC;AAAA,IACH;AAAA,IACA,KAAK,WAAW;AACd,YAAM,MAAM,MAAM,OAAO;AACzB,YAAM,OAAO,aAAa,cAAc,EAAE,OAAO;AACjD,oBAAc,KAAK,GAAG,KAAK,UAAU,MAAM,MAAM,CAAC,CAAC;AAAA,GAAM,OAAO;AAEhE,cAAQ,IAAI,0CAA0C,GAAG,EAAE;AAC3D,aAAO;AAAA,IACT;AAAA,IACA,KAAK,WAAW;AACd,cAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,cAAc,GAAG,MAAM,CAAC,CAAC;AAAA,CAAI;AACpE,aAAO;AAAA,IACT;AAAA,IACA,KAAK,aAAa;AAChB,cAAQ,OAAO,MAAM,GAAG,cAAc,EAAE,OAAO;AAAA,CAAI;AACnD,aAAO;AAAA,IACT;AAAA,IACA,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AACH,cAAQ,OAAO,MAAM,GAAG,IAAI;AAAA,CAAI;AAChC,aAAO;AAAA,IACT;AACE,cAAQ,OAAO,MAAM,oBAAoB,OAAO;AAAA,EAAK,IAAI;AAAA,CAAI;AAC7D,aAAO;AAAA,EACX;AACF;AAEA,IAAM,mBAAwD;AAAA,EAC5D,OAAO,oBAAI,IAAI,CAAC,QAAQ,QAAQ,MAAM,CAAC;AAAA,EACvC,KAAK,oBAAI,IAAI,CAAC,MAAM,CAAC;AAAA,EACrB,aAAa,oBAAI,IAAI,CAAC,MAAM,CAAC;AAAA,EAC7B,SAAS,oBAAI,IAAI,CAAC,QAAQ,KAAK,CAAC;AAAA,EAChC,SAAS,oBAAI,IAAI,CAAC,MAAM,CAAC;AAAA,EACzB,MAAM,oBAAI,IAAI;AAAA,EACd,UAAU,oBAAI,IAAI;AAAA,EAClB,MAAM,oBAAI,IAAI;AAAA,EACd,aAAa,oBAAI,IAAI;AAAA,EACrB,IAAI,oBAAI,IAAI;AACd;AAEA,SAAS,iBAAiB,SAAiB,OAAqC;AAC9E,QAAM,UAAU,iBAAiB,OAAO;AACxC,MAAI,CAAC,QAAS;AACd,QAAM,UAAU,OAAO,KAAK,KAAK,EAAE,OAAO,CAAC,SAAS,CAAC,QAAQ,IAAI,IAAI,CAAC;AACtE,MAAI,QAAQ,SAAS,GAAG;AACtB,UAAM,IAAI,MAAM,oBAAoB,WAAW,MAAM,OAAO,QAAQ,CAAC,CAAC,EAAE;AAAA,EAC1E;AACF;AAEA,SAAS,UAAU,KAAqB;AACtC,QAAM,OAAO,OAAO,GAAG;AACvB,MAAI,CAAC,OAAO,UAAU,IAAI,KAAK,OAAO,KAAK,OAAO,OAAQ;AACxD,UAAM,IAAI,MAAM,uDAAuD,KAAK,UAAU,GAAG,CAAC,EAAE;AAAA,EAC9F;AACA,SAAO;AACT;AAEA,KAAK,EACF,KAAK,CAAC,SAAS,QAAQ,KAAK,IAAI,CAAC,EACjC,MAAM,CAAC,QAAQ;AAEd,UAAQ,MAAM,2BAA2B,GAAG;AAC5C,UAAQ,KAAK,CAAC;AAChB,CAAC;","names":[]}
+{"version":3,"sources":["../src/cli.ts","../src/cli-config.ts"],"sourcesContent":["#!/usr/bin/env node\n/**\n * agent-eval CLI.\n *\n *   agent-eval serve [--port 5005] [--host 127.0.0.1]\n *   agent-eval rpc <method>          # one request from stdin → one response on stdout\n *   agent-eval rpc-batch <method>    # JSONL stdin → JSONL stdout\n *   agent-eval openapi [--out path]  # write OpenAPI spec\n *   agent-eval version\n *\n * <method> is one of: judge, listRubrics, version. When omitted, the\n * stdin payload must be a full {method, params} envelope.\n */\nimport { writeFileSync } from 'node:fs'\nimport { resolveCliLlmConfig } from './cli-config'\nimport { runRolloutReleaseCli } from './rollout/release/hf-dataset'\nimport { handleVersion } from './wire/handlers'\nimport { buildOpenApi } from './wire/openapi'\nimport { runRpcBatch, runRpcOnce } from './wire/rpc'\nimport { startServerAsync } from './wire/server'\n\ninterface Args {\n  command: string\n  positional: string[]\n  flags: Record<string, string>\n}\n\nfunction parseArgs(argv: string[]): Args {\n  const [command, ...rest] = argv\n  const positional: string[] = []\n  const flags: Record<string, string> = {}\n  for (let i = 0; i < rest.length; i++) {\n    const tok = rest[i]!\n    if (tok.startsWith('--')) {\n      const raw = tok.slice(2)\n      const equalsAt = raw.indexOf('=')\n      if (equalsAt >= 0) {\n        flags[raw.slice(0, equalsAt)] = raw.slice(equalsAt + 1)\n        continue\n      }\n      const key = raw\n      if (key === 'help') {\n        flags[key] = 'true'\n        continue\n      }\n      const next = rest[i + 1]\n      if (next != null && !next.startsWith('--')) {\n        flags[key] = next\n        i++\n      } else {\n        flags[key] = 'true'\n      }\n    } else if (tok === '-h') {\n      flags.help = 'true'\n    } else {\n      positional.push(tok)\n    }\n  }\n  return { command: command ?? 'help', positional, flags }\n}\n\nconst HELP = `agent-eval: evaluation RPC and HTTP server.\n\nCommands:\n  serve [--port 5005] [--host 127.0.0.1]\n        Start the HTTP server. POST /v1/judge, GET /v1/rubrics, GET /v1/version, GET /openapi.json.\n  rpc <method>\n        Read one JSON object from stdin (the params for <method>), write one\n        JSON object to stdout. Methods: judge, listRubrics, version.\n  rpc-batch <method>\n        Like 'rpc' but JSONL in / JSONL out.\n  openapi [--out openapi.json]\n        Write the OpenAPI 3.1 spec.\n  rollout-release <ledger.jsonl...> --out <dir> [--formats sft,verifiers,rft,raw] [--include-proposers] [--push <org/name>]\n        Build a HuggingFace-ready dataset dir from tangle.rollout.v1 ledgers:\n        validate, fail-closed split filter, scrub, export formats + card.\n  version\n        Print server + wire-protocol version JSON.\n\nJudge provider:\n  Set AGENT_EVAL_LLM_BASE_URL, AGENT_EVAL_LLM_API_KEY, and AGENT_EVAL_LLM_MODEL.\n  OPENAI_* and TANGLE_* equivalents are also accepted.\n\nWithout arguments, prints this help.`\n\nasync function main(): Promise<number> {\n  const { command, positional, flags } = parseArgs(process.argv.slice(2))\n  assertKnownFlags(command, flags)\n\n  if (flags.help === 'true') {\n    process.stdout.write(`${HELP}\\n`)\n    return 0\n  }\n\n  switch (command) {\n    case 'serve': {\n      const port = parsePort(flags.port ?? '5005')\n      const host = flags.host ?? '127.0.0.1'\n      const llm = resolveCliLlmConfig()\n      const { server } = await startServerAsync({\n        port,\n        host,\n        llm: llm.client,\n        judgeModel: llm.model,\n        llmRouteRequirements: { requireExplicitBaseUrl: true },\n      })\n      // Keep process alive on SIGINT/SIGTERM\n      const shutdown = (sig: string) => {\n        // eslint-disable-next-line no-console\n        console.log(`[agent-eval] received ${sig}, shutting down`)\n        server.close(() => process.exit(0))\n        // Force exit after 5s if close hangs\n        setTimeout(() => process.exit(1), 5000).unref()\n      }\n      process.on('SIGINT', () => shutdown('SIGINT'))\n      process.on('SIGTERM', () => shutdown('SIGTERM'))\n      // Block forever\n      await new Promise(() => {})\n      return 0\n    }\n    case 'rpc': {\n      const [method] = positional\n      const llm = resolveCliLlmConfig()\n      return await runRpcOnce(method, {\n        llm: llm.client,\n        judgeModel: llm.model,\n        llmRouteRequirements: { requireExplicitBaseUrl: true },\n      })\n    }\n    case 'rpc-batch': {\n      const [method] = positional\n      const llm = resolveCliLlmConfig()\n      return await runRpcBatch(method, {\n        llm: llm.client,\n        judgeModel: llm.model,\n        llmRouteRequirements: { requireExplicitBaseUrl: true },\n      })\n    }\n    case 'openapi': {\n      const out = flags.out ?? 'openapi.json'\n      const spec = buildOpenApi(handleVersion().version)\n      writeFileSync(out, `${JSON.stringify(spec, null, 2)}\\n`, 'utf-8')\n      // eslint-disable-next-line no-console\n      console.log(`[agent-eval] wrote OpenAPI 3.1 spec to ${out}`)\n      return 0\n    }\n    case 'rollout-release': {\n      // The subcommand owns its own flag grammar (multi-value --formats,\n      // boolean --include-proposers) — pass raw argv through untouched.\n      return await runRolloutReleaseCli(process.argv.slice(3))\n    }\n    case 'version': {\n      process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}\\n`)\n      return 0\n    }\n    case '--version': {\n      process.stdout.write(`${handleVersion().version}\\n`)\n      return 0\n    }\n    case 'help':\n    case '--help':\n    case '-h':\n    case '':\n      process.stdout.write(`${HELP}\\n`)\n      return 0\n    default:\n      process.stderr.write(`unknown command: ${command}\\n${HELP}\\n`)\n      return 1\n  }\n}\n\nconst FLAGS_BY_COMMAND: Record<string, ReadonlySet<string>> = {\n  serve: new Set(['help', 'host', 'port']),\n  rpc: new Set(['help']),\n  'rpc-batch': new Set(['help']),\n  openapi: new Set(['help', 'out']),\n  version: new Set(['help']),\n  help: new Set(),\n  '--help': new Set(),\n  '-h': new Set(),\n  '--version': new Set(),\n  '': new Set(),\n}\n\nfunction assertKnownFlags(command: string, flags: Record<string, string>): void {\n  // rollout-release parses its own argv (multi-value flags).\n  if (command === 'rollout-release') return\n  const allowed = FLAGS_BY_COMMAND[command]\n  if (!allowed) return\n  const unknown = Object.keys(flags).filter((flag) => !allowed.has(flag))\n  if (unknown.length > 0) {\n    throw new Error(`unknown flag for ${command || 'help'}: --${unknown[0]}`)\n  }\n}\n\nfunction parsePort(raw: string): number {\n  const port = Number(raw)\n  if (!Number.isInteger(port) || port < 0 || port > 65_535) {\n    throw new Error(`--port must be an integer from 0 to 65535; received ${JSON.stringify(raw)}`)\n  }\n  return port\n}\n\nmain()\n  .then((code) => process.exit(code))\n  .catch((err) => {\n    // eslint-disable-next-line no-console\n    console.error('[agent-eval] cli error:', err)\n    process.exit(1)\n  })\n","import type { LlmClientOptions } from './llm-client'\n\nexport interface CliLlmConfig {\n  client?: LlmClientOptions\n  model?: string\n}\n\nexport function resolveCliLlmConfig(env: NodeJS.ProcessEnv = process.env): CliLlmConfig {\n  const explicitBaseUrl = nonEmpty(env.AGENT_EVAL_LLM_BASE_URL)\n  const explicitApiKey = nonEmpty(env.AGENT_EVAL_LLM_API_KEY)\n  const openAiApiKey = nonEmpty(env.OPENAI_API_KEY)\n  const tangleApiKey = nonEmpty(env.TANGLE_API_KEY)\n  const baseUrl =\n    explicitBaseUrl ??\n    nonEmpty(env.OPENAI_BASE_URL) ??\n    nonEmpty(env.TANGLE_ROUTER_URL) ??\n    (openAiApiKey ? 'https://api.openai.com/v1' : undefined) ??\n    (tangleApiKey ? 'https://router.tangle.tools/v1' : undefined)\n  const apiKey = explicitApiKey ?? openAiApiKey ?? tangleApiKey\n  const model =\n    nonEmpty(env.AGENT_EVAL_LLM_MODEL) ?? nonEmpty(env.OPENAI_MODEL) ?? nonEmpty(env.TANGLE_MODEL)\n\n  const client =\n    baseUrl || apiKey\n      ? { ...(baseUrl ? { baseUrl } : {}), ...(apiKey ? { apiKey } : {}) }\n      : undefined\n  return { ...(client ? { client } : {}), ...(model ? { model } : {}) }\n}\n\nfunction nonEmpty(value: string | undefined): string | undefined {\n  const trimmed = value?.trim()\n  return trimmed ? trimmed : undefined\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAaA,SAAS,qBAAqB;;;ACNvB,SAAS,oBAAoB,MAAyB,QAAQ,KAAmB;AACtF,QAAM,kBAAkB,SAAS,IAAI,uBAAuB;AAC5D,QAAM,iBAAiB,SAAS,IAAI,sBAAsB;AAC1D,QAAM,eAAe,SAAS,IAAI,cAAc;AAChD,QAAM,eAAe,SAAS,IAAI,cAAc;AAChD,QAAM,UACJ,mBACA,SAAS,IAAI,eAAe,KAC5B,SAAS,IAAI,iBAAiB,MAC7B,eAAe,8BAA8B,YAC7C,eAAe,mCAAmC;AACrD,QAAM,SAAS,kBAAkB,gBAAgB;AACjD,QAAM,QACJ,SAAS,IAAI,oBAAoB,KAAK,SAAS,IAAI,YAAY,KAAK,SAAS,IAAI,YAAY;AAE/F,QAAM,SACJ,WAAW,SACP,EAAE,GAAI,UAAU,EAAE,QAAQ,IAAI,CAAC,GAAI,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC,EAAG,IACjE;AACN,SAAO,EAAE,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC,GAAI,GAAI,QAAQ,EAAE,MAAM,IAAI,CAAC,EAAG;AACtE;AAEA,SAAS,SAAS,OAA+C;AAC/D,QAAM,UAAU,OAAO,KAAK;AAC5B,SAAO,UAAU,UAAU;AAC7B;;;ADLA,SAAS,UAAU,MAAsB;AACvC,QAAM,CAAC,SAAS,GAAG,IAAI,IAAI;AAC3B,QAAM,aAAuB,CAAC;AAC9B,QAAM,QAAgC,CAAC;AACvC,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;AACpC,UAAM,MAAM,KAAK,CAAC;AAClB,QAAI,IAAI,WAAW,IAAI,GAAG;AACxB,YAAM,MAAM,IAAI,MAAM,CAAC;AACvB,YAAM,WAAW,IAAI,QAAQ,GAAG;AAChC,UAAI,YAAY,GAAG;AACjB,cAAM,IAAI,MAAM,GAAG,QAAQ,CAAC,IAAI,IAAI,MAAM,WAAW,CAAC;AACtD;AAAA,MACF;AACA,YAAM,MAAM;AACZ,UAAI,QAAQ,QAAQ;AAClB,cAAM,GAAG,IAAI;AACb;AAAA,MACF;AACA,YAAM,OAAO,KAAK,IAAI,CAAC;AACvB,UAAI,QAAQ,QAAQ,CAAC,KAAK,WAAW,IAAI,GAAG;AAC1C,cAAM,GAAG,IAAI;AACb;AAAA,MACF,OAAO;AACL,cAAM,GAAG,IAAI;AAAA,MACf;AAAA,IACF,WAAW,QAAQ,MAAM;AACvB,YAAM,OAAO;AAAA,IACf,OAAO;AACL,iBAAW,KAAK,GAAG;AAAA,IACrB;AAAA,EACF;AACA,SAAO,EAAE,SAAS,WAAW,QAAQ,YAAY,MAAM;AACzD;AAEA,IAAM,OAAO;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAwBb,eAAe,OAAwB;AACrC,QAAM,EAAE,SAAS,YAAY,MAAM,IAAI,UAAU,QAAQ,KAAK,MAAM,CAAC,CAAC;AACtE,mBAAiB,SAAS,KAAK;AAE/B,MAAI,MAAM,SAAS,QAAQ;AACzB,YAAQ,OAAO,MAAM,GAAG,IAAI;AAAA,CAAI;AAChC,WAAO;AAAA,EACT;AAEA,UAAQ,SAAS;AAAA,IACf,KAAK,SAAS;AACZ,YAAM,OAAO,UAAU,MAAM,QAAQ,MAAM;AAC3C,YAAM,OAAO,MAAM,QAAQ;AAC3B,YAAM,MAAM,oBAAoB;AAChC,YAAM,EAAE,OAAO,IAAI,MAAM,iBAAiB;AAAA,QACxC;AAAA,QACA;AAAA,QACA,KAAK,IAAI;AAAA,QACT,YAAY,IAAI;AAAA,QAChB,sBAAsB,EAAE,wBAAwB,KAAK;AAAA,MACvD,CAAC;AAED,YAAM,WAAW,CAAC,QAAgB;AAEhC,gBAAQ,IAAI,yBAAyB,GAAG,iBAAiB;AACzD,eAAO,MAAM,MAAM,QAAQ,KAAK,CAAC,CAAC;AAElC,mBAAW,MAAM,QAAQ,KAAK,CAAC,GAAG,GAAI,EAAE,MAAM;AAAA,MAChD;AACA,cAAQ,GAAG,UAAU,MAAM,SAAS,QAAQ,CAAC;AAC7C,cAAQ,GAAG,WAAW,MAAM,SAAS,SAAS,CAAC;AAE/C,YAAM,IAAI,QAAQ,MAAM;AAAA,MAAC,CAAC;AAC1B,aAAO;AAAA,IACT;AAAA,IACA,KAAK,OAAO;AACV,YAAM,CAAC,MAAM,IAAI;AACjB,YAAM,MAAM,oBAAoB;AAChC,aAAO,MAAM,WAAW,QAAQ;AAAA,QAC9B,KAAK,IAAI;AAAA,QACT,YAAY,IAAI;AAAA,QAChB,sBAAsB,EAAE,wBAAwB,KAAK;AAAA,MACvD,CAAC;AAAA,IACH;AAAA,IACA,KAAK,aAAa;AAChB,YAAM,CAAC,MAAM,IAAI;AACjB,YAAM,MAAM,oBAAoB;AAChC,aAAO,MAAM,YAAY,QAAQ;AAAA,QAC/B,KAAK,IAAI;AAAA,QACT,YAAY,IAAI;AAAA,QAChB,sBAAsB,EAAE,wBAAwB,KAAK;AAAA,MACvD,CAAC;AAAA,IACH;AAAA,IACA,KAAK,WAAW;AACd,YAAM,MAAM,MAAM,OAAO;AACzB,YAAM,OAAO,aAAa,cAAc,EAAE,OAAO;AACjD,oBAAc,KAAK,GAAG,KAAK,UAAU,MAAM,MAAM,CAAC,CAAC;AAAA,GAAM,OAAO;AAEhE,cAAQ,IAAI,0CAA0C,GAAG,EAAE;AAC3D,aAAO;AAAA,IACT;AAAA,IACA,KAAK,mBAAmB;AAGtB,aAAO,MAAM,qBAAqB,QAAQ,KAAK,MAAM,CAAC,CAAC;AAAA,IACzD;AAAA,IACA,KAAK,WAAW;AACd,cAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,cAAc,GAAG,MAAM,CAAC,CAAC;AAAA,CAAI;AACpE,aAAO;AAAA,IACT;AAAA,IACA,KAAK,aAAa;AAChB,cAAQ,OAAO,MAAM,GAAG,cAAc,EAAE,OAAO;AAAA,CAAI;AACnD,aAAO;AAAA,IACT;AAAA,IACA,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AACH,cAAQ,OAAO,MAAM,GAAG,IAAI;AAAA,CAAI;AAChC,aAAO;AAAA,IACT;AACE,cAAQ,OAAO,MAAM,oBAAoB,OAAO;AAAA,EAAK,IAAI;AAAA,CAAI;AAC7D,aAAO;AAAA,EACX;AACF;AAEA,IAAM,mBAAwD;AAAA,EAC5D,OAAO,oBAAI,IAAI,CAAC,QAAQ,QAAQ,MAAM,CAAC;AAAA,EACvC,KAAK,oBAAI,IAAI,CAAC,MAAM,CAAC;AAAA,EACrB,aAAa,oBAAI,IAAI,CAAC,MAAM,CAAC;AAAA,EAC7B,SAAS,oBAAI,IAAI,CAAC,QAAQ,KAAK,CAAC;AAAA,EAChC,SAAS,oBAAI,IAAI,CAAC,MAAM,CAAC;AAAA,EACzB,MAAM,oBAAI,IAAI;AAAA,EACd,UAAU,oBAAI,IAAI;AAAA,EAClB,MAAM,oBAAI,IAAI;AAAA,EACd,aAAa,oBAAI,IAAI;AAAA,EACrB,IAAI,oBAAI,IAAI;AACd;AAEA,SAAS,iBAAiB,SAAiB,OAAqC;AAE9E,MAAI,YAAY,kBAAmB;AACnC,QAAM,UAAU,iBAAiB,OAAO;AACxC,MAAI,CAAC,QAAS;AACd,QAAM,UAAU,OAAO,KAAK,KAAK,EAAE,OAAO,CAAC,SAAS,CAAC,QAAQ,IAAI,IAAI,CAAC;AACtE,MAAI,QAAQ,SAAS,GAAG;AACtB,UAAM,IAAI,MAAM,oBAAoB,WAAW,MAAM,OAAO,QAAQ,CAAC,CAAC,EAAE;AAAA,EAC1E;AACF;AAEA,SAAS,UAAU,KAAqB;AACtC,QAAM,OAAO,OAAO,GAAG;AACvB,MAAI,CAAC,OAAO,UAAU,IAAI,KAAK,OAAO,KAAK,OAAO,OAAQ;AACxD,UAAM,IAAI,MAAM,uDAAuD,KAAK,UAAU,GAAG,CAAC,EAAE;AAAA,EAC9F;AACA,SAAO;AACT;AAEA,KAAK,EACF,KAAK,CAAC,SAAS,QAAQ,KAAK,IAAI,CAAC,EACjC,MAAM,CAAC,QAAQ;AAEd,UAAQ,MAAM,2BAA2B,GAAG;AAC5C,UAAQ,KAAK,CAAC;AAChB,CAAC;","names":[]}

package/dist/contract/index.d.ts

@@ -838,6 +838,7 @@ interface LlmMessage {
         };
     }>;
 }
+type LlmThinkingMode = 'enabled' | 'disabled';
 interface LlmCallRequest {
     model: string;
     messages: LlmMessage[];
@@ -850,6 +851,8 @@ interface LlmCallRequest {
     };
     temperature?: number;
     maxTokens?: number;
+    /** OpenAI-compatible reasoning mode. Omitted when the provider default should apply. */
+    thinking?: LlmThinkingMode;
     /** Per-call timeout, default 300s. */
     timeoutMs?: number;
 }
@@ -859,6 +862,8 @@ interface LlmUsage {
     totalTokens: number;
     /** False when the provider omitted or malformed prompt/completion usage. */
     captured?: boolean;
+    /** Reasoning-token subset of completionTokens, when reported. */
+    reasoningTokens?: number;
     /** Proxies populate this when prompt caching is on. */
     cachedPromptTokens?: number;
 }
@@ -942,6 +947,8 @@ interface LlmClientOptions {
      * Default: `extract`.
      */
     jsonPayloadMode?: 'extract' | 'exact';
+    /** Default provider reasoning mode. A per-call request value takes precedence. */
+    thinking?: LlmThinkingMode;
     /** Fetch implementation — defaults to global `fetch`. Override for custom transport (e.g. tests). */
     fetch?: typeof fetch;
     /**
@@ -2662,6 +2669,8 @@ interface GepaProposerOptions {
     temperature?: number;
     /** Reflection max tokens. Default 6000. */
     maxTokens?: number;
+    /** Provider reasoning mode for each reflection call. Omitted to preserve the provider default. */
+    thinking?: LlmCallRequest['thinking'];
     /** Structured-doc constraints. Candidates violating any are rejected
      *  post-parse and dropped from the returned population. */
     constraints?: GepaProposerConstraints;

package/dist/contract/index.js

@@ -37,7 +37,7 @@ import {
   runReferenceEquivalenceJudge,
   surfaceContentHash,
   surfaceHash
-} from "../chunk-KKPPFIDS.js";
+} from "../chunk-PMITBABE.js";
 import {
   campaignSplitDigest,
   createRunCostLedger,
@@ -49,15 +49,15 @@ import {
 import {
   buildDefaultAnalystRegistry,
   createChatClient
-} from "../chunk-WP3GEXKI.js";
+} from "../chunk-A6GT67HT.js";
 import "../chunk-HHWE3POT.js";
-import "../chunk-MHPEGJHC.js";
+import "../chunk-DT7OXY3C.js";
 import {
   FileSystemOutcomeStore,
   InMemoryOutcomeStore
 } from "../chunk-3RF76KTD.js";
 import "../chunk-ARU2PZFM.js";
-import "../chunk-MK3Q6GCL.js";
+import "../chunk-EQUK3RFS.js";
 import "../chunk-DPZAEKA6.js";
 import {
   pairedBootstrap