@tangle-network/agent-eval 0.126.6 → 0.127.0

package/dist/{chunk-M7AH34KV.js → chunk-UTA2TGC4.js}

@@ -1,9 +1,15 @@
 import {
   ROLLOUT_SCHEMA
-} from "./chunk-MAX3TN3C.js";
+} from "./chunk-UWZZKKU7.js";
 import {
   buildTrajectory
 } from "./chunk-RZTMDUO7.js";
+import {
+  runTaskScore
+} from "./chunk-NL33OXEP.js";
+import {
+  ValidationError
+} from "./chunk-ONWEPEDO.js";
 
 // src/rollout/mint.ts
 var asText = (v, scrub) => {
@@ -42,15 +48,21 @@ function finalConversation(spans, scrub) {
   }
   return messages;
 }
-function rolloutReward(record) {
+function scoredReward(record) {
+  const score = runTaskScore(record);
+  if (score === void 0) {
+    throw new ValidationError(`Cannot mint rollout for run ${record.runId}: task score is missing`);
+  }
   const gated = record.outcome.realness?.gated === true;
-  const raw = record.outcome.holdoutScore ?? record.outcome.searchScore ?? 0;
-  return { reward: gated ? 0 : raw, gated };
+  return {
+    reward: gated ? 0 : score,
+    gated,
+    source: record.outcome.holdoutScore !== void 0 ? "run-record/holdout-score" : "run-record/search-score"
+  };
 }
-function rewardSource(record) {
-  if (record.outcome.holdoutScore !== void 0) return "run-record/holdout-score";
-  if (record.outcome.searchScore !== void 0) return "run-record/search-score";
-  return "run-record/unscored";
+function rolloutReward(record) {
+  const { reward, gated } = scoredReward(record);
+  return { reward, gated };
 }
 var SPLIT_FROM_TAG = {
   search: "search",
@@ -58,8 +70,12 @@ var SPLIT_FROM_TAG = {
   holdout: "holdout"
 };
 function mintLine(record, steps, messages, options, capturedAt, gap) {
-  const { reward, gated } = rolloutReward(record);
-  const uncaptured = record.costProvenance?.kind === "uncaptured";
+  const { reward, gated, source } = scoredReward(record);
+  const uncaptured = record.costProvenance.kind === "uncaptured";
+  const terminalOutcome = record.terminalOutcome;
+  const isCompleted = terminalOutcome === "succeeded" || terminalOutcome === "failed";
+  const isTruncated = terminalOutcome === "cancelled" || terminalOutcome === "incomplete";
+  const terminalError = terminalOutcome === "failed" || terminalOutcome === "cancelled" || terminalOutcome === "incomplete" ? record.terminalFailureReason ?? `run ended ${terminalOutcome}` : null;
   return {
     schema: ROLLOUT_SCHEMA,
     rollout_id: record.runId,
@@ -72,7 +88,7 @@ function mintLine(record, steps, messages, options, capturedAt, gap) {
     role: options.role ?? "agent",
     task: {
       suite: options.suite ?? record.experimentId,
-      instance_id: record.scenarioId ?? record.experimentId,
+      instance_id: record.scenarioId,
       split: SPLIT_FROM_TAG[record.splitTag],
       seed: record.seed,
       rep: 0
@@ -93,12 +109,12 @@ function mintLine(record, steps, messages, options, capturedAt, gap) {
     ...steps.length > 0 ? { steps } : {},
     outcome: {
       reward,
-      reward_source: rewardSource(record),
+      reward_source: source,
       verdict: null,
       metrics: { ...record.outcome.raw },
-      is_completed: true,
-      is_truncated: false,
-      error: null,
+      is_completed: isCompleted,
+      is_truncated: isTruncated,
+      error: terminalError,
       realness_gated: gated
     },
     cost: {
@@ -152,4 +168,4 @@ export {
   rolloutReward,
   mintRolloutRows
 };
-//# sourceMappingURL=chunk-M7AH34KV.js.map
+//# sourceMappingURL=chunk-UTA2TGC4.js.map

package/dist/chunk-UTA2TGC4.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/rollout/mint.ts"],"sourcesContent":["/**\n * Rollout minting — `tangle.rollout.v1` lines joined from the records the\n * substrate ALREADY keeps. There is no separate rollout store: a rollout\n * is the JOIN of a RunRecord (identity, provenance, cost, outcome) with\n * its trace (spans share `runId`), projected into the canonical line.\n *\n * Composition, not duplication:\n *   - identity/provenance     → `RunRecord` (candidateId, splitTag, agentProfile, hashes)\n *   - step structure          → `buildTrajectory` over the shared TraceStore\n *   - preference-pair export  → `feedbackTrajectoryToOptimizerRow` (feedback-trajectory.ts)\n *   - PRM / reward-model      → `reward-model-export.ts`\n *\n * Anti-Goodhart invariant: a run whose `outcome.realness.gated` is true\n * is never exported with a positive reward — the gate travels into the\n * training data (`reward` forced to 0, `realness_gated: true`), so a\n * fine-tune cannot learn from gamed successes.\n *\n * Records without spans become labeled GAP LINES (messages: [],\n * provenance.gap) — present in the output AND surfaced in\n * `missingTraces`; a capture gap is a finding, never a silent omission.\n */\n\nimport { ValidationError } from '../errors'\nimport { type RunRecord, runTaskScore } from '../run-record'\nimport type { LlmSpan, Message, Span, ToolSpan } from '../trace/schema'\nimport type { TraceStore } from '../trace/store'\nimport { buildTrajectory } from '../trajectory'\nimport {\n  type ChatMessage,\n  ROLLOUT_SCHEMA,\n  type RolloutLine,\n  type RolloutRole,\n  type RolloutSplit,\n  type RolloutStep,\n} from './schema'\n\n/** Redactor applied to every exported string (secrets, PII). Identity by default. */\nexport type RolloutScrubber = (text: string) => string\n\nexport interface MintRolloutOptions {\n  scrub?: RolloutScrubber\n  /** Cap steps per line (longest runs first drop middle steps). Default: no cap. */\n  maxSteps?: number\n  /** Role recorded on every minted line. Default 'agent' (a solo eval run). */\n  role?: RolloutRole\n  /** Task suite label. Default: the record's `experimentId`. */\n  suite?: string\n  /** Injected clock for deterministic output. */\n  now?: () => Date\n}\n\nexport interface MintRolloutResult {\n  rows: RolloutLine[]\n  /** runIds that had a RunRecord but no spans — emitted as gap lines AND listed here. */\n  missingTraces: string[]\n}\n\nconst asText = (v: unknown, scrub: RolloutScrubber): string => {\n  const s = typeof v === 'string' ? v : JSON.stringify(v)\n  return scrub(s ?? '')\n}\n\nfunction projectStep(span: Span, scrub: RolloutScrubber): RolloutStep {\n  const base: RolloutStep = {\n    kind: span.kind,\n    name: scrub(span.name),\n    status: span.status,\n    durationMs: span.endedAt !== undefined ? span.endedAt - span.startedAt : undefined,\n  }\n  if (span.kind === 'llm') {\n    const llm = span as LlmSpan\n    const last = llm.messages[llm.messages.length - 1]\n    if (last) base.input = scrub(last.content)\n    if (llm.output !== undefined) base.output = scrub(llm.output)\n  } else if (span.kind === 'tool') {\n    const tool = span as ToolSpan\n    base.input = asText(tool.args, scrub)\n    if (tool.result !== undefined) base.output = asText(tool.result, scrub)\n  }\n  return base\n}\n\n/** The final llm span's history + output is the completed conversation. */\nfunction finalConversation(spans: Span[], scrub: RolloutScrubber): ChatMessage[] {\n  const llms = spans.filter((s): s is LlmSpan => s.kind === 'llm')\n  const last = llms[llms.length - 1]\n  if (!last) return []\n  const messages: ChatMessage[] = last.messages.map((m: Message) => ({\n    role: m.role,\n    content: scrub(m.content),\n  }))\n  if (last.output !== undefined && last.output !== '') {\n    messages.push({ role: 'assistant', content: scrub(last.output) })\n  }\n  return messages\n}\n\nfunction scoredReward(record: RunRecord): {\n  reward: number\n  gated: boolean\n  source: 'run-record/holdout-score' | 'run-record/search-score'\n} {\n  const score = runTaskScore(record)\n  if (score === undefined) {\n    throw new ValidationError(`Cannot mint rollout for run ${record.runId}: task score is missing`)\n  }\n  const gated = record.outcome.realness?.gated === true\n  return {\n    reward: gated ? 0 : score,\n    gated,\n    source:\n      record.outcome.holdoutScore !== undefined\n        ? 'run-record/holdout-score'\n        : 'run-record/search-score',\n  }\n}\n\nexport function rolloutReward(record: RunRecord): { reward: number; gated: boolean } {\n  const { reward, gated } = scoredReward(record)\n  return { reward, gated }\n}\n\nconst SPLIT_FROM_TAG: Record<RunRecord['splitTag'], RolloutSplit> = {\n  search: 'search',\n  dev: 'dev',\n  holdout: 'holdout',\n}\n\nfunction mintLine(\n  record: RunRecord,\n  steps: RolloutStep[],\n  messages: ChatMessage[],\n  options: MintRolloutOptions,\n  capturedAt: string,\n  gap?: string,\n): RolloutLine {\n  const { reward, gated, source } = scoredReward(record)\n  const uncaptured = record.costProvenance.kind === 'uncaptured'\n  const terminalOutcome = record.terminalOutcome\n  const isCompleted = terminalOutcome === 'succeeded' || terminalOutcome === 'failed'\n  const isTruncated = terminalOutcome === 'cancelled' || terminalOutcome === 'incomplete'\n  const terminalError =\n    terminalOutcome === 'failed' ||\n    terminalOutcome === 'cancelled' ||\n    terminalOutcome === 'incomplete'\n      ? (record.terminalFailureReason ?? `run ended ${terminalOutcome}`)\n      : null\n  return {\n    schema: ROLLOUT_SCHEMA,\n    rollout_id: record.runId,\n    parent_rollout_id: null,\n    run_id: record.runId,\n    experiment_id: record.experimentId,\n    candidate_id: record.candidateId,\n    generation: null,\n    candidate_index: null,\n    role: options.role ?? 'agent',\n    task: {\n      suite: options.suite ?? record.experimentId,\n      instance_id: record.scenarioId,\n      split: SPLIT_FROM_TAG[record.splitTag],\n      seed: record.seed,\n      rep: 0,\n    },\n    policy: {\n      harness: null,\n      harness_version: null,\n      model: record.model,\n      provider: null,\n      profile_commit: record.commitSha,\n      prompt_hash: record.promptHash,\n      config_hash: record.configHash,\n      agent_profile_cell_id: record.agentProfile?.cellId ?? null,\n      sampling: null,\n    },\n    messages,\n    tool_defs: [],\n    ...(steps.length > 0 ? { steps } : {}),\n    outcome: {\n      reward,\n      reward_source: source,\n      verdict: null,\n      metrics: { ...record.outcome.raw },\n      is_completed: isCompleted,\n      is_truncated: isTruncated,\n      error: terminalError,\n      realness_gated: gated,\n    },\n    cost: {\n      usd: uncaptured ? null : record.costUsd,\n      tokens_in: record.tokenUsage.input,\n      tokens_out: record.tokenUsage.output,\n      tokens_reasoning: record.tokenUsage.reasoning ?? null,\n      cache_read: record.tokenUsage.cached ?? null,\n      cache_write: record.tokenUsage.cacheWrite ?? null,\n      wall_s: Math.round(record.wallMs / 1000),\n    },\n    artifacts: { patch_path: null, run_dir: null, transcript_ref: null },\n    provenance: {\n      captured_at: capturedAt,\n      capture: 'mint',\n      ...(gap !== undefined ? { gap } : {}),\n    },\n  }\n}\n\n/**\n * Join RunRecords with their traces into canonical rollout lines. Records\n * without spans are emitted as labeled gap lines and reported in\n * `missingTraces`. Execution-only records without a task score are rejected\n * because a missing training label is not a zero reward.\n */\nexport async function mintRolloutRows(\n  records: RunRecord[],\n  store: TraceStore,\n  options: MintRolloutOptions = {},\n): Promise<MintRolloutResult> {\n  const scrub = options.scrub ?? ((t) => t)\n  const capturedAt = (options.now?.() ?? new Date()).toISOString()\n  const rows: RolloutLine[] = []\n  const missingTraces: string[] = []\n  for (const record of records) {\n    const trajectory = await buildTrajectory(store, record.runId)\n    if (trajectory.steps.length === 0) {\n      missingTraces.push(record.runId)\n      rows.push(\n        mintLine(record, [], [], options, capturedAt, 'no trace spans recorded for this runId'),\n      )\n      continue\n    }\n    let steps = trajectory.steps.map((s) => projectStep(s.span, scrub))\n    if (options.maxSteps !== undefined && steps.length > options.maxSteps) {\n      // Keep the head and tail — the middle of a long run is the least\n      // informative for outcome attribution.\n      const head = Math.ceil(options.maxSteps / 2)\n      const tail = options.maxSteps - head\n      steps = [...steps.slice(0, head), ...steps.slice(steps.length - tail)]\n    }\n    const conversation = finalConversation(\n      trajectory.steps.map((s) => s.span),\n      scrub,\n    )\n    const gap =\n      conversation.length === 0 ? 'trace has no llm spans — no conversation to inline' : undefined\n    rows.push(mintLine(record, steps, conversation, options, capturedAt, gap))\n  }\n  return { rows, missingTraces }\n}\n"],"mappings":";;;;;;;;;;;;;;AAyDA,IAAM,SAAS,CAAC,GAAY,UAAmC;AAC7D,QAAM,IAAI,OAAO,MAAM,WAAW,IAAI,KAAK,UAAU,CAAC;AACtD,SAAO,MAAM,KAAK,EAAE;AACtB;AAEA,SAAS,YAAY,MAAY,OAAqC;AACpE,QAAM,OAAoB;AAAA,IACxB,MAAM,KAAK;AAAA,IACX,MAAM,MAAM,KAAK,IAAI;AAAA,IACrB,QAAQ,KAAK;AAAA,IACb,YAAY,KAAK,YAAY,SAAY,KAAK,UAAU,KAAK,YAAY;AAAA,EAC3E;AACA,MAAI,KAAK,SAAS,OAAO;AACvB,UAAM,MAAM;AACZ,UAAM,OAAO,IAAI,SAAS,IAAI,SAAS,SAAS,CAAC;AACjD,QAAI,KAAM,MAAK,QAAQ,MAAM,KAAK,OAAO;AACzC,QAAI,IAAI,WAAW,OAAW,MAAK,SAAS,MAAM,IAAI,MAAM;AAAA,EAC9D,WAAW,KAAK,SAAS,QAAQ;AAC/B,UAAM,OAAO;AACb,SAAK,QAAQ,OAAO,KAAK,MAAM,KAAK;AACpC,QAAI,KAAK,WAAW,OAAW,MAAK,SAAS,OAAO,KAAK,QAAQ,KAAK;AAAA,EACxE;AACA,SAAO;AACT;AAGA,SAAS,kBAAkB,OAAe,OAAuC;AAC/E,QAAM,OAAO,MAAM,OAAO,CAAC,MAAoB,EAAE,SAAS,KAAK;AAC/D,QAAM,OAAO,KAAK,KAAK,SAAS,CAAC;AACjC,MAAI,CAAC,KAAM,QAAO,CAAC;AACnB,QAAM,WAA0B,KAAK,SAAS,IAAI,CAAC,OAAgB;AAAA,IACjE,MAAM,EAAE;AAAA,IACR,SAAS,MAAM,EAAE,OAAO;AAAA,EAC1B,EAAE;AACF,MAAI,KAAK,WAAW,UAAa,KAAK,WAAW,IAAI;AACnD,aAAS,KAAK,EAAE,MAAM,aAAa,SAAS,MAAM,KAAK,MAAM,EAAE,CAAC;AAAA,EAClE;AACA,SAAO;AACT;AAEA,SAAS,aAAa,QAIpB;AACA,QAAM,QAAQ,aAAa,MAAM;AACjC,MAAI,UAAU,QAAW;AACvB,UAAM,IAAI,gBAAgB,+BAA+B,OAAO,KAAK,yBAAyB;AAAA,EAChG;AACA,QAAM,QAAQ,OAAO,QAAQ,UAAU,UAAU;AACjD,SAAO;AAAA,IACL,QAAQ,QAAQ,IAAI;AAAA,IACpB;AAAA,IACA,QACE,OAAO,QAAQ,iBAAiB,SAC5B,6BACA;AAAA,EACR;AACF;AAEO,SAAS,cAAc,QAAuD;AACnF,QAAM,EAAE,QAAQ,MAAM,IAAI,aAAa,MAAM;AAC7C,SAAO,EAAE,QAAQ,MAAM;AACzB;AAEA,IAAM,iBAA8D;AAAA,EAClE,QAAQ;AAAA,EACR,KAAK;AAAA,EACL,SAAS;AACX;AAEA,SAAS,SACP,QACA,OACA,UACA,SACA,YACA,KACa;AACb,QAAM,EAAE,QAAQ,OAAO,OAAO,IAAI,aAAa,MAAM;AACrD,QAAM,aAAa,OAAO,eAAe,SAAS;AAClD,QAAM,kBAAkB,OAAO;AAC/B,QAAM,cAAc,oBAAoB,eAAe,oBAAoB;AAC3E,QAAM,cAAc,oBAAoB,eAAe,oBAAoB;AAC3E,QAAM,gBACJ,oBAAoB,YACpB,oBAAoB,eACpB,oBAAoB,eACf,OAAO,yBAAyB,aAAa,eAAe,KAC7D;AACN,SAAO;AAAA,IACL,QAAQ;AAAA,IACR,YAAY,OAAO;AAAA,IACnB,mBAAmB;AAAA,IACnB,QAAQ,OAAO;AAAA,IACf,eAAe,OAAO;AAAA,IACtB,cAAc,OAAO;AAAA,IACrB,YAAY;AAAA,IACZ,iBAAiB;AAAA,IACjB,MAAM,QAAQ,QAAQ;AAAA,IACtB,MAAM;AAAA,MACJ,OAAO,QAAQ,SAAS,OAAO;AAAA,MAC/B,aAAa,OAAO;AAAA,MACpB,OAAO,eAAe,OAAO,QAAQ;AAAA,MACrC,MAAM,OAAO;AAAA,MACb,KAAK;AAAA,IACP;AAAA,IACA,QAAQ;AAAA,MACN,SAAS;AAAA,MACT,iBAAiB;AAAA,MACjB,OAAO,OAAO;AAAA,MACd,UAAU;AAAA,MACV,gBAAgB,OAAO;AAAA,MACvB,aAAa,OAAO;AAAA,MACpB,aAAa,OAAO;AAAA,MACpB,uBAAuB,OAAO,cAAc,UAAU;AAAA,MACtD,UAAU;AAAA,IACZ;AAAA,IACA;AAAA,IACA,WAAW,CAAC;AAAA,IACZ,GAAI,MAAM,SAAS,IAAI,EAAE,MAAM,IAAI,CAAC;AAAA,IACpC,SAAS;AAAA,MACP;AAAA,MACA,eAAe;AAAA,MACf,SAAS;AAAA,MACT,SAAS,EAAE,GAAG,OAAO,QAAQ,IAAI;AAAA,MACjC,cAAc;AAAA,MACd,cAAc;AAAA,MACd,OAAO;AAAA,MACP,gBAAgB;AAAA,IAClB;AAAA,IACA,MAAM;AAAA,MACJ,KAAK,aAAa,OAAO,OAAO;AAAA,MAChC,WAAW,OAAO,WAAW;AAAA,MAC7B,YAAY,OAAO,WAAW;AAAA,MAC9B,kBAAkB,OAAO,WAAW,aAAa;AAAA,MACjD,YAAY,OAAO,WAAW,UAAU;AAAA,MACxC,aAAa,OAAO,WAAW,cAAc;AAAA,MAC7C,QAAQ,KAAK,MAAM,OAAO,SAAS,GAAI;AAAA,IACzC;AAAA,IACA,WAAW,EAAE,YAAY,MAAM,SAAS,MAAM,gBAAgB,KAAK;AAAA,IACnE,YAAY;AAAA,MACV,aAAa;AAAA,MACb,SAAS;AAAA,MACT,GAAI,QAAQ,SAAY,EAAE,IAAI,IAAI,CAAC;AAAA,IACrC;AAAA,EACF;AACF;AAQA,eAAsB,gBACpB,SACA,OACA,UAA8B,CAAC,GACH;AAC5B,QAAM,QAAQ,QAAQ,UAAU,CAAC,MAAM;AACvC,QAAM,cAAc,QAAQ,MAAM,KAAK,oBAAI,KAAK,GAAG,YAAY;AAC/D,QAAM,OAAsB,CAAC;AAC7B,QAAM,gBAA0B,CAAC;AACjC,aAAW,UAAU,SAAS;AAC5B,UAAM,aAAa,MAAM,gBAAgB,OAAO,OAAO,KAAK;AAC5D,QAAI,WAAW,MAAM,WAAW,GAAG;AACjC,oBAAc,KAAK,OAAO,KAAK;AAC/B,WAAK;AAAA,QACH,SAAS,QAAQ,CAAC,GAAG,CAAC,GAAG,SAAS,YAAY,wCAAwC;AAAA,MACxF;AACA;AAAA,IACF;AACA,QAAI,QAAQ,WAAW,MAAM,IAAI,CAAC,MAAM,YAAY,EAAE,MAAM,KAAK,CAAC;AAClE,QAAI,QAAQ,aAAa,UAAa,MAAM,SAAS,QAAQ,UAAU;AAGrE,YAAM,OAAO,KAAK,KAAK,QAAQ,WAAW,CAAC;AAC3C,YAAM,OAAO,QAAQ,WAAW;AAChC,cAAQ,CAAC,GAAG,MAAM,MAAM,GAAG,IAAI,GAAG,GAAG,MAAM,MAAM,MAAM,SAAS,IAAI,CAAC;AAAA,IACvE;AACA,UAAM,eAAe;AAAA,MACnB,WAAW,MAAM,IAAI,CAAC,MAAM,EAAE,IAAI;AAAA,MAClC;AAAA,IACF;AACA,UAAM,MACJ,aAAa,WAAW,IAAI,4DAAuD;AACrF,SAAK,KAAK,SAAS,QAAQ,OAAO,cAAc,SAAS,YAAY,GAAG,CAAC;AAAA,EAC3E;AACA,SAAO,EAAE,MAAM,cAAc;AAC/B;","names":[]}

package/dist/{chunk-MAX3TN3C.js → chunk-UWZZKKU7.js}

@@ -1,6 +1,5 @@
 // src/rollout/schema.ts
 var ROLLOUT_SCHEMA = "tangle.rollout.v1";
-var ROLLOUT_FORMAT = ROLLOUT_SCHEMA;
 var ROLLOUT_ROLES = [
   "agent",
   "supervisor",
@@ -9,14 +8,8 @@ var ROLLOUT_ROLES = [
   "judge",
   "analyst"
 ];
-var ROLLOUT_SPLITS = [
-  "search",
-  "dev",
-  "holdout",
-  "canary",
-  "train"
-];
-var TRAINABLE_SPLITS = ["search", "train"];
+var ROLLOUT_SPLITS = ["search", "dev", "holdout", "canary"];
+var TRAINABLE_SPLITS = ["search"];
 function isTrainableSplit(split) {
   return TRAINABLE_SPLITS.includes(split);
 }
@@ -81,12 +74,8 @@ function validateRolloutLine(value) {
     errors.push("parent_rollout_id: expected string|null");
   if (typeof value.run_id !== "string" || value.run_id.length === 0)
     errors.push("run_id: expected non-empty string");
-  if (value.experiment_id !== void 0 && !isStringOrNull(value.experiment_id)) {
-    errors.push("experiment_id: expected string|null when present");
-  }
-  if (value.candidate_id !== void 0 && !isStringOrNull(value.candidate_id)) {
-    errors.push("candidate_id: expected string|null when present");
-  }
+  if (!isStringOrNull(value.experiment_id)) errors.push("experiment_id: expected string|null");
+  if (!isStringOrNull(value.candidate_id)) errors.push("candidate_id: expected string|null");
   if (!isIntegerOrNull(value.generation)) errors.push("generation: expected integer|null");
   if (!isIntegerOrNull(value.candidate_index)) errors.push("candidate_index: expected integer|null");
   if (!ROLLOUT_ROLES.includes(value.role))
@@ -167,8 +156,8 @@ function validateRolloutLine(value) {
   );
   if (isRecord(value.outcome)) {
     if (!("verdict" in value.outcome)) errors.push("outcome.verdict: field required (may be null)");
-    if (value.outcome.realness_gated !== void 0 && typeof value.outcome.realness_gated !== "boolean") {
-      errors.push("outcome.realness_gated: expected boolean when present");
+    if (typeof value.outcome.realness_gated !== "boolean") {
+      errors.push("outcome.realness_gated: expected boolean");
     }
   }
   validateSection(
@@ -235,7 +224,6 @@ function isRolloutLine(value) {
 
 export {
   ROLLOUT_SCHEMA,
-  ROLLOUT_FORMAT,
   ROLLOUT_ROLES,
   ROLLOUT_SPLITS,
   TRAINABLE_SPLITS,
@@ -246,4 +234,4 @@ export {
   assertRolloutLine,
   isRolloutLine
 };
-//# sourceMappingURL=chunk-MAX3TN3C.js.map
+//# sourceMappingURL=chunk-UWZZKKU7.js.map

package/dist/chunk-UWZZKKU7.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/rollout/schema.ts"],"sourcesContent":["/**\n * `tangle.rollout.v1` — THE canonical rollout serialization, owned by\n * agent-eval. One JSONL line per agent invocation (a solo eval run, a\n * supervisor episode, a worker session, a proposer shot, a judge call, an\n * analyst pass), labeled with its task/split coordinates and a single\n * scalar reward, carrying the FULL message transcript inline.\n *\n * This schema is the reconciliation of two prior producers:\n *   - agent-eval's RunRecord-joined rollout rows (PR #410): identity,\n *     provenance hashes, the realness gate travelling into the reward,\n *     trace-derived steps.\n *   - the bench rollout-ledger (agent-runtime PR #591): the wire shape —\n *     role, task.split/rep, parent_rollout_id, policy provenance, capture\n *     provenance, inline canonical chat-with-tools messages.\n * Where the two conflicted, RunRecord-derived semantics won; the wire\n * field names follow the ledger (snake_case). See `docs/rollout.md` for\n * the field-by-field decision table.\n *\n * Messages are inlined — never referenced — because every harness store a\n * rollout can be recovered from is mutable or garbage-collected. A line\n * must stay a complete training/eval example on its own.\n *\n * `outcome.reward` is THE single scalar (null = no verdict exists — a\n * labeled gap, never 0). `outcome.realness_gated` is the anti-Goodhart\n * flag: a gated line must never export as a positive training example.\n */\n\nexport const ROLLOUT_SCHEMA = 'tangle.rollout.v1'\n\n/** `agent` = a solo evaluation run (no multi-agent topology). */\nexport type RolloutRole = 'agent' | 'supervisor' | 'worker' | 'proposer' | 'judge' | 'analyst'\nexport const ROLLOUT_ROLES: readonly RolloutRole[] = [\n  'agent',\n  'supervisor',\n  'worker',\n  'proposer',\n  'judge',\n  'analyst',\n]\n\n/** Split vocabulary follows `RunRecord.splitTag`, extended with `canary`. */\nexport type RolloutSplit = 'search' | 'dev' | 'holdout' | 'canary'\nexport const ROLLOUT_SPLITS: readonly RolloutSplit[] = ['search', 'dev', 'holdout', 'canary']\n/** Splits that may ship in training exports. Everything else is fail-closed excluded. */\nexport const TRAINABLE_SPLITS: readonly RolloutSplit[] = ['search']\n\nexport function isTrainableSplit(split: RolloutSplit): boolean {\n  return TRAINABLE_SPLITS.includes(split)\n}\n\n/** 'mint' = joined live from RunRecord + trace by `mintRolloutRows`. */\nexport type RolloutCapture = 'mint' | 'settle-time' | 'backfill'\nexport const ROLLOUT_CAPTURES: readonly RolloutCapture[] = ['mint', 'settle-time', 'backfill']\n\n// ---------------------------------------------------------------------------\n// Canonical message format — OpenAI chat-with-tools, full fidelity.\n// ---------------------------------------------------------------------------\n\nexport type ChatRole = 'system' | 'user' | 'assistant' | 'tool'\nexport const CHAT_ROLES: readonly ChatRole[] = ['system', 'user', 'assistant', 'tool']\n\nexport interface ChatToolCall {\n  id: string\n  type: 'function'\n  function: {\n    name: string\n    /** JSON-encoded argument object, exactly as the model emitted it. */\n    arguments: string\n  }\n}\n\nexport interface ChatMessage {\n  role: ChatRole\n  content: string | null\n  /** Reasoning/thinking channel where the harness captured it (full fidelity). */\n  reasoning_content?: string\n  tool_calls?: ChatToolCall[]\n  /** Required on role:\"tool\" — the ChatToolCall this result answers. */\n  tool_call_id?: string\n  name?: string\n}\n\nexport interface ToolDef {\n  type: 'function'\n  function: {\n    name: string\n    description?: string\n    parameters?: Record<string, unknown>\n  }\n}\n\n/**\n * Compact trace-span projection (llm/tool step) carried alongside the\n * conversation when the line was minted from a trace. Optional: lines\n * recovered from harness stores have no span structure.\n */\nexport interface RolloutStep {\n  kind: string\n  name: string\n  /** llm: last-message summary · tool: stringified args. Scrubbed. */\n  input?: string\n  /** llm: output text · tool: stringified result. Scrubbed. */\n  output?: string\n  status?: 'ok' | 'error'\n  durationMs?: number\n}\n\n// ---------------------------------------------------------------------------\n// Ledger line sections.\n// ---------------------------------------------------------------------------\n\nexport interface RolloutTask {\n  /** Benchmark/suite id (e.g. \"swe-bench-verified\") or the experiment id. */\n  suite: string\n  instance_id: string\n  split: RolloutSplit\n  /** Sampling seed the campaign pinned; null = not recorded. */\n  seed: number | null\n  /** Replicate index (0-based). */\n  rep: number\n}\n\nexport interface RolloutPolicy {\n  /** Harness that drove the invocation (e.g. \"opencode\", \"claude\", \"pi-loops\"). */\n  harness: string | null\n  harness_version: string | null\n  model: string | null\n  provider: string | null\n  /** Commit of the agent profile / candidate under evaluation. */\n  profile_commit: string | null\n  /** sha256 of the effective prompt (post-steering), when recorded. */\n  prompt_hash?: string | null\n  /** sha256 of the effective run config, when recorded. */\n  config_hash?: string | null\n  /** Canonical agent-profile cell identity, when the run carries one. */\n  agent_profile_cell_id?: string | null\n  /** Sampling params (temperature, top_p, max_tokens…); null = not recorded. */\n  sampling: Record<string, unknown> | null\n}\n\nexport interface RolloutOutcome {\n  /**\n   * THE single scalar training signal — the official verdict.\n   * null = no verdict exists for this invocation (a labeled gap, never 0).\n   */\n  reward: number | null\n  /** Where the reward came from (judge id; \"/inherited\" = parent episode's). */\n  reward_source: string | null\n  /** Raw judge verdict record, verbatim. */\n  verdict: unknown\n  /** Everything that is NOT the scalar reward. */\n  metrics: Record<string, unknown>\n  is_completed: boolean\n  is_truncated: boolean\n  error: string | null\n  /**\n   * Anti-Goodhart flag from `RunRecord.outcome.realness.gated`: the run\n   * faked its success signal. Reward is forced to 0 at mint time and the\n   * line never qualifies for SFT.\n   */\n  realness_gated: boolean\n}\n\nexport interface RolloutCostBlock {\n  usd: number | null\n  tokens_in: number | null\n  tokens_out: number | null\n  tokens_reasoning: number | null\n  cache_read: number | null\n  cache_write: number | null\n  wall_s: number | null\n}\n\nexport interface RolloutArtifacts {\n  patch_path: string | null\n  run_dir: string | null\n  /** Source-of-truth transcript pointer (session id / jsonl path) for audit. */\n  transcript_ref: string | null\n}\n\nexport interface RolloutProvenance {\n  captured_at: string\n  capture: RolloutCapture\n  /** Present on gap lines: why `messages` could not be recovered. */\n  gap?: string\n}\n\nexport interface RolloutLine {\n  schema: typeof ROLLOUT_SCHEMA\n  rollout_id: string\n  /** Spawning invocation within the same episode (worker → supervisor). */\n  parent_rollout_id: string | null\n  run_id: string\n  /** Logical experiment grouping from `RunRecord.experimentId`; null = not recorded. */\n  experiment_id: string | null\n  /** Stable candidate identity from `RunRecord.candidateId`; null = not recorded. */\n  candidate_id: string | null\n  /** Improvement-loop generation (-1 = baseline); null = not an improvement loop. */\n  generation: number | null\n  /** Improvement-loop candidate index (-1 = baseline); null = not an improvement loop. */\n  candidate_index: number | null\n  role: RolloutRole\n  task: RolloutTask\n  policy: RolloutPolicy\n  /** Full transcript, inline. [] = gap line (see provenance.gap). */\n  messages: ChatMessage[]\n  tool_defs: ToolDef[]\n  /** Trace-span projections, when minted from a trace. */\n  steps?: RolloutStep[]\n  outcome: RolloutOutcome\n  cost: RolloutCostBlock\n  artifacts: RolloutArtifacts\n  provenance: RolloutProvenance\n}\n\n// ---------------------------------------------------------------------------\n// Validation — pure TS, no runtime schema dependency, mirroring the\n// run-record validator's fail-loud discipline. Returns [] when the value\n// is a valid RolloutLine; otherwise one dotted-path error per defect.\n// ---------------------------------------------------------------------------\n\nconst isRecord = (v: unknown): v is Record<string, unknown> =>\n  typeof v === 'object' && v !== null && !Array.isArray(v)\n\nconst isNumberOrNull = (v: unknown): boolean => v === null || typeof v === 'number'\nconst isStringOrNull = (v: unknown): boolean => v === null || typeof v === 'string'\nconst isIntegerOrNull = (v: unknown): boolean => v === null || Number.isInteger(v)\n\nfunction validateChatMessage(value: unknown, path: string, errors: string[]): void {\n  if (!isRecord(value)) {\n    errors.push(`${path}: not an object`)\n    return\n  }\n  if (!CHAT_ROLES.includes(value.role as ChatRole))\n    errors.push(`${path}.role: invalid role ${String(value.role)}`)\n  if (!isStringOrNull(value.content)) errors.push(`${path}.content: must be string|null`)\n  if (value.reasoning_content !== undefined && typeof value.reasoning_content !== 'string') {\n    errors.push(`${path}.reasoning_content: must be string when present`)\n  }\n  if (value.tool_call_id !== undefined && typeof value.tool_call_id !== 'string') {\n    errors.push(`${path}.tool_call_id: must be string when present`)\n  }\n  if (value.role === 'tool' && typeof value.tool_call_id !== 'string') {\n    errors.push(`${path}.tool_call_id: required on role:\"tool\"`)\n  }\n  if (value.tool_calls !== undefined) {\n    if (!Array.isArray(value.tool_calls)) {\n      errors.push(`${path}.tool_calls: must be an array when present`)\n    } else {\n      value.tool_calls.forEach((call, i) => {\n        if (!isRecord(call) || typeof call.id !== 'string' || call.type !== 'function') {\n          errors.push(`${path}.tool_calls[${i}]: must be {id, type:\"function\", function}`)\n          return\n        }\n        const fn = call.function\n        if (!isRecord(fn) || typeof fn.name !== 'string' || typeof fn.arguments !== 'string') {\n          errors.push(\n            `${path}.tool_calls[${i}].function: must be {name: string, arguments: string}`,\n          )\n        }\n      })\n    }\n  }\n}\n\nfunction validateSection(\n  value: unknown,\n  path: string,\n  fields: Array<[name: string, check: (v: unknown) => boolean, expect: string]>,\n  errors: string[],\n): void {\n  if (!isRecord(value)) {\n    errors.push(`${path}: not an object`)\n    return\n  }\n  for (const [name, check, expect] of fields) {\n    if (!check(value[name])) errors.push(`${path}.${name}: expected ${expect}`)\n  }\n}\n\nexport function validateRolloutLine(value: unknown): string[] {\n  const errors: string[] = []\n  if (!isRecord(value)) return ['line: not an object']\n\n  if (value.schema !== ROLLOUT_SCHEMA) errors.push(`schema: expected \"${ROLLOUT_SCHEMA}\"`)\n  if (typeof value.rollout_id !== 'string' || value.rollout_id.length === 0)\n    errors.push('rollout_id: expected non-empty string')\n  if (!isStringOrNull(value.parent_rollout_id))\n    errors.push('parent_rollout_id: expected string|null')\n  if (typeof value.run_id !== 'string' || value.run_id.length === 0)\n    errors.push('run_id: expected non-empty string')\n  if (!isStringOrNull(value.experiment_id)) errors.push('experiment_id: expected string|null')\n  if (!isStringOrNull(value.candidate_id)) errors.push('candidate_id: expected string|null')\n  if (!isIntegerOrNull(value.generation)) errors.push('generation: expected integer|null')\n  if (!isIntegerOrNull(value.candidate_index)) errors.push('candidate_index: expected integer|null')\n  if (!ROLLOUT_ROLES.includes(value.role as RolloutRole))\n    errors.push(`role: invalid role ${String(value.role)}`)\n\n  validateSection(\n    value.task,\n    'task',\n    [\n      ['suite', (v) => typeof v === 'string' && v.length > 0, 'non-empty string'],\n      ['instance_id', (v) => typeof v === 'string' && v.length > 0, 'non-empty string'],\n      [\n        'split',\n        (v) => ROLLOUT_SPLITS.includes(v as RolloutSplit),\n        `one of ${ROLLOUT_SPLITS.join('|')}`,\n      ],\n      ['seed', isNumberOrNull, 'number|null'],\n      ['rep', (v) => Number.isInteger(v), 'integer'],\n    ],\n    errors,\n  )\n\n  validateSection(\n    value.policy,\n    'policy',\n    [\n      ['harness', isStringOrNull, 'string|null'],\n      ['harness_version', isStringOrNull, 'string|null'],\n      ['model', isStringOrNull, 'string|null'],\n      ['provider', isStringOrNull, 'string|null'],\n      ['profile_commit', isStringOrNull, 'string|null'],\n      ['sampling', (v) => v === null || isRecord(v), 'object|null'],\n    ],\n    errors,\n  )\n  if (isRecord(value.policy)) {\n    for (const key of ['prompt_hash', 'config_hash', 'agent_profile_cell_id'] as const) {\n      if (value.policy[key] !== undefined && !isStringOrNull(value.policy[key])) {\n        errors.push(`policy.${key}: expected string|null when present`)\n      }\n    }\n  }\n\n  if (!Array.isArray(value.messages)) {\n    errors.push('messages: expected array')\n  } else {\n    for (const [i, m] of value.messages.entries()) validateChatMessage(m, `messages[${i}]`, errors)\n  }\n\n  if (!Array.isArray(value.tool_defs)) {\n    errors.push('tool_defs: expected array')\n  } else {\n    value.tool_defs.forEach((d, i) => {\n      if (\n        !isRecord(d) ||\n        d.type !== 'function' ||\n        !isRecord(d.function) ||\n        typeof d.function.name !== 'string'\n      ) {\n        errors.push(`tool_defs[${i}]: must be {type:\"function\", function:{name}}`)\n      }\n    })\n  }\n\n  if (value.steps !== undefined) {\n    if (!Array.isArray(value.steps)) {\n      errors.push('steps: expected array when present')\n    } else {\n      value.steps.forEach((s, i) => {\n        if (!isRecord(s) || typeof s.kind !== 'string' || typeof s.name !== 'string') {\n          errors.push(`steps[${i}]: must be {kind: string, name: string, …}`)\n        }\n      })\n    }\n  }\n\n  validateSection(\n    value.outcome,\n    'outcome',\n    [\n      ['reward', isNumberOrNull, 'number|null'],\n      ['reward_source', isStringOrNull, 'string|null'],\n      ['metrics', isRecord, 'object'],\n      ['is_completed', (v) => typeof v === 'boolean', 'boolean'],\n      ['is_truncated', (v) => typeof v === 'boolean', 'boolean'],\n      ['error', isStringOrNull, 'string|null'],\n    ],\n    errors,\n  )\n  if (isRecord(value.outcome)) {\n    if (!('verdict' in value.outcome)) errors.push('outcome.verdict: field required (may be null)')\n    if (typeof value.outcome.realness_gated !== 'boolean') {\n      errors.push('outcome.realness_gated: expected boolean')\n    }\n  }\n\n  validateSection(\n    value.cost,\n    'cost',\n    [\n      ['usd', isNumberOrNull, 'number|null'],\n      ['tokens_in', isNumberOrNull, 'number|null'],\n      ['tokens_out', isNumberOrNull, 'number|null'],\n      ['tokens_reasoning', isNumberOrNull, 'number|null'],\n      ['cache_read', isNumberOrNull, 'number|null'],\n      ['cache_write', isNumberOrNull, 'number|null'],\n      ['wall_s', isNumberOrNull, 'number|null'],\n    ],\n    errors,\n  )\n\n  validateSection(\n    value.artifacts,\n    'artifacts',\n    [\n      ['patch_path', isStringOrNull, 'string|null'],\n      ['run_dir', isStringOrNull, 'string|null'],\n      ['transcript_ref', isStringOrNull, 'string|null'],\n    ],\n    errors,\n  )\n\n  validateSection(\n    value.provenance,\n    'provenance',\n    [\n      [\n        'captured_at',\n        (v) => typeof v === 'string' && !Number.isNaN(Date.parse(v)),\n        'ISO-8601 timestamp',\n      ],\n      [\n        'capture',\n        (v) => ROLLOUT_CAPTURES.includes(v as RolloutCapture),\n        `one of ${ROLLOUT_CAPTURES.join('|')}`,\n      ],\n    ],\n    errors,\n  )\n  if (\n    isRecord(value.provenance) &&\n    value.provenance.gap !== undefined &&\n    typeof value.provenance.gap !== 'string'\n  ) {\n    errors.push('provenance.gap: must be string when present')\n  }\n\n  // A gap line must say WHY it is a gap; a full line must not carry a gap note.\n  if (Array.isArray(value.messages) && isRecord(value.provenance)) {\n    if (value.messages.length === 0 && typeof value.provenance.gap !== 'string') {\n      errors.push('provenance.gap: required when messages is empty')\n    }\n  }\n\n  return errors\n}\n\nexport function assertRolloutLine(\n  value: unknown,\n  context = 'rollout line',\n): asserts value is RolloutLine {\n  const errors = validateRolloutLine(value)\n  if (errors.length > 0) {\n    throw new Error(`invalid ${context}:\\n  ${errors.join('\\n  ')}`)\n  }\n}\n\nexport function isRolloutLine(value: unknown): value is RolloutLine {\n  return validateRolloutLine(value).length === 0\n}\n"],"mappings":";AA2BO,IAAM,iBAAiB;AAIvB,IAAM,gBAAwC;AAAA,EACnD;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF;AAIO,IAAM,iBAA0C,CAAC,UAAU,OAAO,WAAW,QAAQ;AAErF,IAAM,mBAA4C,CAAC,QAAQ;AAE3D,SAAS,iBAAiB,OAA8B;AAC7D,SAAO,iBAAiB,SAAS,KAAK;AACxC;AAIO,IAAM,mBAA8C,CAAC,QAAQ,eAAe,UAAU;AAOtF,IAAM,aAAkC,CAAC,UAAU,QAAQ,aAAa,MAAM;AAkKrF,IAAM,WAAW,CAAC,MAChB,OAAO,MAAM,YAAY,MAAM,QAAQ,CAAC,MAAM,QAAQ,CAAC;AAEzD,IAAM,iBAAiB,CAAC,MAAwB,MAAM,QAAQ,OAAO,MAAM;AAC3E,IAAM,iBAAiB,CAAC,MAAwB,MAAM,QAAQ,OAAO,MAAM;AAC3E,IAAM,kBAAkB,CAAC,MAAwB,MAAM,QAAQ,OAAO,UAAU,CAAC;AAEjF,SAAS,oBAAoB,OAAgB,MAAc,QAAwB;AACjF,MAAI,CAAC,SAAS,KAAK,GAAG;AACpB,WAAO,KAAK,GAAG,IAAI,iBAAiB;AACpC;AAAA,EACF;AACA,MAAI,CAAC,WAAW,SAAS,MAAM,IAAgB;AAC7C,WAAO,KAAK,GAAG,IAAI,uBAAuB,OAAO,MAAM,IAAI,CAAC,EAAE;AAChE,MAAI,CAAC,eAAe,MAAM,OAAO,EAAG,QAAO,KAAK,GAAG,IAAI,+BAA+B;AACtF,MAAI,MAAM,sBAAsB,UAAa,OAAO,MAAM,sBAAsB,UAAU;AACxF,WAAO,KAAK,GAAG,IAAI,iDAAiD;AAAA,EACtE;AACA,MAAI,MAAM,iBAAiB,UAAa,OAAO,MAAM,iBAAiB,UAAU;AAC9E,WAAO,KAAK,GAAG,IAAI,4CAA4C;AAAA,EACjE;AACA,MAAI,MAAM,SAAS,UAAU,OAAO,MAAM,iBAAiB,UAAU;AACnE,WAAO,KAAK,GAAG,IAAI,wCAAwC;AAAA,EAC7D;AACA,MAAI,MAAM,eAAe,QAAW;AAClC,QAAI,CAAC,MAAM,QAAQ,MAAM,UAAU,GAAG;AACpC,aAAO,KAAK,GAAG,IAAI,4CAA4C;AAAA,IACjE,OAAO;AACL,YAAM,WAAW,QAAQ,CAAC,MAAM,MAAM;AACpC,YAAI,CAAC,SAAS,IAAI,KAAK,OAAO,KAAK,OAAO,YAAY,KAAK,SAAS,YAAY;AAC9E,iBAAO,KAAK,GAAG,IAAI,eAAe,CAAC,4CAA4C;AAC/E;AAAA,QACF;AACA,cAAM,KAAK,KAAK;AAChB,YAAI,CAAC,SAAS,EAAE,KAAK,OAAO,GAAG,SAAS,YAAY,OAAO,GAAG,cAAc,UAAU;AACpF,iBAAO;AAAA,YACL,GAAG,IAAI,eAAe,CAAC;AAAA,UACzB;AAAA,QACF;AAAA,MACF,CAAC;AAAA,IACH;AAAA,EACF;AACF;AAEA,SAAS,gBACP,OACA,MACA,QACA,QACM;AACN,MAAI,CAAC,SAAS,KAAK,GAAG;AACpB,WAAO,KAAK,GAAG,IAAI,iBAAiB;AACpC;AAAA,EACF;AACA,aAAW,CAAC,MAAM,OAAO,MAAM,KAAK,QAAQ;AAC1C,QAAI,CAAC,MAAM,MAAM,IAAI,CAAC,EAAG,QAAO,KAAK,GAAG,IAAI,IAAI,IAAI,cAAc,MAAM,EAAE;AAAA,EAC5E;AACF;AAEO,SAAS,oBAAoB,OAA0B;AAC5D,QAAM,SAAmB,CAAC;AAC1B,MAAI,CAAC,SAAS,KAAK,EAAG,QAAO,CAAC,qBAAqB;AAEnD,MAAI,MAAM,WAAW,eAAgB,QAAO,KAAK,qBAAqB,cAAc,GAAG;AACvF,MAAI,OAAO,MAAM,eAAe,YAAY,MAAM,WAAW,WAAW;AACtE,WAAO,KAAK,uCAAuC;AACrD,MAAI,CAAC,eAAe,MAAM,iBAAiB;AACzC,WAAO,KAAK,yCAAyC;AACvD,MAAI,OAAO,MAAM,WAAW,YAAY,MAAM,OAAO,WAAW;AAC9D,WAAO,KAAK,mCAAmC;AACjD,MAAI,CAAC,eAAe,MAAM,aAAa,EAAG,QAAO,KAAK,qCAAqC;AAC3F,MAAI,CAAC,eAAe,MAAM,YAAY,EAAG,QAAO,KAAK,oCAAoC;AACzF,MAAI,CAAC,gBAAgB,MAAM,UAAU,EAAG,QAAO,KAAK,mCAAmC;AACvF,MAAI,CAAC,gBAAgB,MAAM,eAAe,EAAG,QAAO,KAAK,wCAAwC;AACjG,MAAI,CAAC,cAAc,SAAS,MAAM,IAAmB;AACnD,WAAO,KAAK,sBAAsB,OAAO,MAAM,IAAI,CAAC,EAAE;AAExD;AAAA,IACE,MAAM;AAAA,IACN;AAAA,IACA;AAAA,MACE,CAAC,SAAS,CAAC,MAAM,OAAO,MAAM,YAAY,EAAE,SAAS,GAAG,kBAAkB;AAAA,MAC1E,CAAC,eAAe,CAAC,MAAM,OAAO,MAAM,YAAY,EAAE,SAAS,GAAG,kBAAkB;AAAA,MAChF;AAAA,QACE;AAAA,QACA,CAAC,MAAM,eAAe,SAAS,CAAiB;AAAA,QAChD,UAAU,eAAe,KAAK,GAAG,CAAC;AAAA,MACpC;AAAA,MACA,CAAC,QAAQ,gBAAgB,aAAa;AAAA,MACtC,CAAC,OAAO,CAAC,MAAM,OAAO,UAAU,CAAC,GAAG,SAAS;AAAA,IAC/C;AAAA,IACA;AAAA,EACF;AAEA;AAAA,IACE,MAAM;AAAA,IACN;AAAA,IACA;AAAA,MACE,CAAC,WAAW,gBAAgB,aAAa;AAAA,MACzC,CAAC,mBAAmB,gBAAgB,aAAa;AAAA,MACjD,CAAC,SAAS,gBAAgB,aAAa;AAAA,MACvC,CAAC,YAAY,gBAAgB,aAAa;AAAA,MAC1C,CAAC,kBAAkB,gBAAgB,aAAa;AAAA,MAChD,CAAC,YAAY,CAAC,MAAM,MAAM,QAAQ,SAAS,CAAC,GAAG,aAAa;AAAA,IAC9D;AAAA,IACA;AAAA,EACF;AACA,MAAI,SAAS,MAAM,MAAM,GAAG;AAC1B,eAAW,OAAO,CAAC,eAAe,eAAe,uBAAuB,GAAY;AAClF,UAAI,MAAM,OAAO,GAAG,MAAM,UAAa,CAAC,eAAe,MAAM,OAAO,GAAG,CAAC,GAAG;AACzE,eAAO,KAAK,UAAU,GAAG,qCAAqC;AAAA,MAChE;AAAA,IACF;AAAA,EACF;AAEA,MAAI,CAAC,MAAM,QAAQ,MAAM,QAAQ,GAAG;AAClC,WAAO,KAAK,0BAA0B;AAAA,EACxC,OAAO;AACL,eAAW,CAAC,GAAG,CAAC,KAAK,MAAM,SAAS,QAAQ,EAAG,qBAAoB,GAAG,YAAY,CAAC,KAAK,MAAM;AAAA,EAChG;AAEA,MAAI,CAAC,MAAM,QAAQ,MAAM,SAAS,GAAG;AACnC,WAAO,KAAK,2BAA2B;AAAA,EACzC,OAAO;AACL,UAAM,UAAU,QAAQ,CAAC,GAAG,MAAM;AAChC,UACE,CAAC,SAAS,CAAC,KACX,EAAE,SAAS,cACX,CAAC,SAAS,EAAE,QAAQ,KACpB,OAAO,EAAE,SAAS,SAAS,UAC3B;AACA,eAAO,KAAK,aAAa,CAAC,+CAA+C;AAAA,MAC3E;AAAA,IACF,CAAC;AAAA,EACH;AAEA,MAAI,MAAM,UAAU,QAAW;AAC7B,QAAI,CAAC,MAAM,QAAQ,MAAM,KAAK,GAAG;AAC/B,aAAO,KAAK,oCAAoC;AAAA,IAClD,OAAO;AACL,YAAM,MAAM,QAAQ,CAAC,GAAG,MAAM;AAC5B,YAAI,CAAC,SAAS,CAAC,KAAK,OAAO,EAAE,SAAS,YAAY,OAAO,EAAE,SAAS,UAAU;AAC5E,iBAAO,KAAK,SAAS,CAAC,iDAA4C;AAAA,QACpE;AAAA,MACF,CAAC;AAAA,IACH;AAAA,EACF;AAEA;AAAA,IACE,MAAM;AAAA,IACN;AAAA,IACA;AAAA,MACE,CAAC,UAAU,gBAAgB,aAAa;AAAA,MACxC,CAAC,iBAAiB,gBAAgB,aAAa;AAAA,MAC/C,CAAC,WAAW,UAAU,QAAQ;AAAA,MAC9B,CAAC,gBAAgB,CAAC,MAAM,OAAO,MAAM,WAAW,SAAS;AAAA,MACzD,CAAC,gBAAgB,CAAC,MAAM,OAAO,MAAM,WAAW,SAAS;AAAA,MACzD,CAAC,SAAS,gBAAgB,aAAa;AAAA,IACzC;AAAA,IACA;AAAA,EACF;AACA,MAAI,SAAS,MAAM,OAAO,GAAG;AAC3B,QAAI,EAAE,aAAa,MAAM,SAAU,QAAO,KAAK,+CAA+C;AAC9F,QAAI,OAAO,MAAM,QAAQ,mBAAmB,WAAW;AACrD,aAAO,KAAK,0CAA0C;AAAA,IACxD;AAAA,EACF;AAEA;AAAA,IACE,MAAM;AAAA,IACN;AAAA,IACA;AAAA,MACE,CAAC,OAAO,gBAAgB,aAAa;AAAA,MACrC,CAAC,aAAa,gBAAgB,aAAa;AAAA,MAC3C,CAAC,cAAc,gBAAgB,aAAa;AAAA,MAC5C,CAAC,oBAAoB,gBAAgB,aAAa;AAAA,MAClD,CAAC,cAAc,gBAAgB,aAAa;AAAA,MAC5C,CAAC,eAAe,gBAAgB,aAAa;AAAA,MAC7C,CAAC,UAAU,gBAAgB,aAAa;AAAA,IAC1C;AAAA,IACA;AAAA,EACF;AAEA;AAAA,IACE,MAAM;AAAA,IACN;AAAA,IACA;AAAA,MACE,CAAC,cAAc,gBAAgB,aAAa;AAAA,MAC5C,CAAC,WAAW,gBAAgB,aAAa;AAAA,MACzC,CAAC,kBAAkB,gBAAgB,aAAa;AAAA,IAClD;AAAA,IACA;AAAA,EACF;AAEA;AAAA,IACE,MAAM;AAAA,IACN;AAAA,IACA;AAAA,MACE;AAAA,QACE;AAAA,QACA,CAAC,MAAM,OAAO,MAAM,YAAY,CAAC,OAAO,MAAM,KAAK,MAAM,CAAC,CAAC;AAAA,QAC3D;AAAA,MACF;AAAA,MACA;AAAA,QACE;AAAA,QACA,CAAC,MAAM,iBAAiB,SAAS,CAAmB;AAAA,QACpD,UAAU,iBAAiB,KAAK,GAAG,CAAC;AAAA,MACtC;AAAA,IACF;AAAA,IACA;AAAA,EACF;AACA,MACE,SAAS,MAAM,UAAU,KACzB,MAAM,WAAW,QAAQ,UACzB,OAAO,MAAM,WAAW,QAAQ,UAChC;AACA,WAAO,KAAK,6CAA6C;AAAA,EAC3D;AAGA,MAAI,MAAM,QAAQ,MAAM,QAAQ,KAAK,SAAS,MAAM,UAAU,GAAG;AAC/D,QAAI,MAAM,SAAS,WAAW,KAAK,OAAO,MAAM,WAAW,QAAQ,UAAU;AAC3E,aAAO,KAAK,iDAAiD;AAAA,IAC/D;AAAA,EACF;AAEA,SAAO;AACT;AAEO,SAAS,kBACd,OACA,UAAU,gBACoB;AAC9B,QAAM,SAAS,oBAAoB,KAAK;AACxC,MAAI,OAAO,SAAS,GAAG;AACrB,UAAM,IAAI,MAAM,WAAW,OAAO;AAAA,IAAQ,OAAO,KAAK,MAAM,CAAC,EAAE;AAAA,EACjE;AACF;AAEO,SAAS,cAAc,OAAsC;AAClE,SAAO,oBAAoB,KAAK,EAAE,WAAW;AAC/C;","names":[]}

package/dist/{chunk-ZVCHKKOP.js → chunk-WED5KLZQ.js}

@@ -1,6 +1,9 @@
+import {
+  projectCampaignCellQuality
+} from "./chunk-ZPUEEM5U.js";
 import {
   confidenceInterval
-} from "./chunk-PJQFMIOX.js";
+} from "./chunk-MHELPNRP.js";
 import {
   CostAccountingIncompleteError,
   CostLedger
@@ -50,10 +53,10 @@ function summarizeBackendUsage(records) {
   for (const rec of records) {
     totalInputTokens += rec.inputTokens;
     totalOutputTokens += rec.outputTokens;
-    totalCostUsd += rec.costUsd;
+    totalCostUsd += rec.costUsd ?? 0;
     if (rec.inputTokens === 0 && rec.outputTokens === 0) stubRecords++;
     else realRecords++;
-    if (rec.outputTokens > 0 && rec.costUsd === 0) uncostedRecords++;
+    if (rec.outputTokens > 0 && (rec.costUsd === null || rec.costUsd === 0)) uncostedRecords++;
   }
   const verdict = totalRecords === 0 ? "stub" : stubRecords === totalRecords ? "stub" : stubRecords === 0 ? "real" : "mixed";
   const diagnosis = buildDiagnosis({
@@ -352,6 +355,12 @@ function campaignCoverage(cells, scenarios, reps, requireJudgeScore) {
     if (scoreEntries.some(([, score]) => score.failed === true)) {
       reasons.push("judge score marked failed");
     }
+    const failedPanelJudges = [
+      ...new Set(scoreEntries.flatMap(([, score]) => score.failedJudges ?? []))
+    ].sort();
+    if (failedPanelJudges.length > 0) {
+      reasons.push(`judge panel incomplete: ${failedPanelJudges.join(", ")}`);
+    }
     if (nonFiniteScores.length > 0) {
       reasons.push(
         `non-finite judge score: ${nonFiniteScores.map(([name]) => name).sort().join(", ")}`
@@ -1154,6 +1163,8 @@ async function executeCell(args) {
     durationMs: Date.now() - startMs,
     seed: args.slot.cellSeed,
     cached: false,
+    ...failure ? { errorStage: failure.stage } : {},
+    ...failure?.judge ? { errorJudge: failure.judge } : {},
     error: errorMessage
   };
   if (failure) {
@@ -1455,43 +1466,56 @@ function judgeVersionFor(judge) {
   });
 }
 function computeAggregates(cells, judges, seed, cost) {
+  const cellQuality = cells.map((cell) => ({
+    cell,
+    quality: projectCampaignCellQuality(cell)
+  }));
   const byJudge = {};
   for (const judge of judges) {
     const scores = [];
-    for (const cell of cells) {
-      const s = cell.judgeScores[judge.name];
+    for (const { quality } of cellQuality) {
+      const s = quality.successfulJudgeScores[judge.name];
       if (s !== void 0) scores.push(s.composite);
     }
-    byJudge[judge.name] = aggregate(scores, seed);
+    if (scores.length > 0) byJudge[judge.name] = aggregate(scores, seed);
   }
   const byScenario = {};
   const scenarioGroups = /* @__PURE__ */ new Map();
-  for (const cell of cells) {
-    const composites = Object.values(cell.judgeScores).map((s) => s.composite);
-    if (composites.length === 0) continue;
-    const mean = composites.reduce((a, b) => a + b, 0) / composites.length;
+  for (const { cell, quality } of cellQuality) {
+    const score = quality.score;
+    if (score === void 0) continue;
     const arr = scenarioGroups.get(cell.scenarioId) ?? [];
-    arr.push(mean);
+    arr.push(score);
     scenarioGroups.set(cell.scenarioId, arr);
   }
   for (const [scenarioId, samples] of scenarioGroups) {
     const ag = aggregate(samples, seed);
     byScenario[scenarioId] = { meanComposite: ag.mean, ci95: ag.ci95, n: ag.n };
   }
+  const dispatchFailures = cells.filter((cell) => cell.errorStage === "dispatch");
+  const judgeFailures = cellQuality.filter(({ cell, quality }) => cell.errorStage === "judge" || quality.failedJudges.length > 0).map(({ cell }) => cell);
+  const unclassifiedFailures = cells.filter(
+    (cell) => Boolean(cell.error) && !cell.error?.startsWith("skipped:") && cell.errorStage === void 0
+  );
   return {
     byJudge,
     byScenario,
     cost,
     totalCostUsd: cost.totalCostUsd,
-    cellsExecuted: cells.filter((c) => !c.error).length,
+    cellsExecuted: cells.filter(
+      (cell) => !cell.error?.startsWith("skipped:") && cell.errorStage !== "dispatch" && !(cell.error && cell.errorStage === void 0)
+    ).length,
     cellsSkipped: cells.filter((c) => c.error?.startsWith("skipped:")).length,
     cellsCached: cells.filter((c) => c.cached).length,
-    cellsFailed: cells.filter((c) => c.error && !c.error.startsWith("skipped:")).length
+    cellsFailed: (/* @__PURE__ */ new Set([...dispatchFailures, ...judgeFailures, ...unclassifiedFailures])).size,
+    cellsDispatchFailed: dispatchFailures.length,
+    cellsJudgeFailed: judgeFailures.length,
+    cellsUnclassifiedFailed: unclassifiedFailures.length
   };
 }
 function aggregate(samples, seed) {
   const n = samples.length;
-  if (n === 0) return { mean: 0, stdev: 0, ci95: [0, 0], n: 0 };
+  if (n === 0) throw new Error("aggregate requires at least one finite score");
   const mean = samples.reduce((a, b) => a + b, 0) / n;
   const variance = samples.reduce((a, b) => a + (b - mean) ** 2, 0) / Math.max(1, n - 1);
   const stdev = Math.sqrt(variance);
@@ -1532,4 +1556,4 @@ export {
   runCampaign,
   planCampaignRun
 };
-//# sourceMappingURL=chunk-ZVCHKKOP.js.map
+//# sourceMappingURL=chunk-WED5KLZQ.js.map