@tangle-network/agent-eval 0.150.2 → 0.161.1

package/dist/{experiment-tracker-0MhuPArU.d.ts → experiment-tracker-DCO6Cz4s.d.ts}

@@ -1,4 +1,4 @@
-import { p as EvidenceRef } from "./types-B2NsbrNy.js";
+import { p as EvidenceRef } from "./types-D9ssmxKL.js";
 //#region src/statistics/power-and-mde.d.ts
 /**
  * Required N per arm for a two-sample comparison at target effect size,
@@ -234,4 +234,4 @@ declare class ExperimentTracker {
 }
 //#endregion
 export { requiredSampleSize as _, ExperimentVerdict as a, computeExperimentStats as c, inMemoryExperimentStore as d, mulberry32 as f, requiredPairedSampleSize as g, pairedMde as h, ExperimentTracker as i, fileExperimentStore as l, mcnemarRequiredN as m, ExperimentRep as n, ImprovementThresholds as o, mcnemarPower as p, ExperimentStats as r, ImprovementVerdictResult as s, Experiment as t, improvementVerdict as u };
-//# sourceMappingURL=experiment-tracker-0MhuPArU.d.ts.map
+//# sourceMappingURL=experiment-tracker-DCO6Cz4s.d.ts.map

package/dist/{experiment-tracker-0MhuPArU.d.ts.map → experiment-tracker-DCO6Cz4s.d.ts.map}

@@ -1 +1 @@
-{"version":3,"file":"experiment-tracker-0MhuPArU.d.ts","names":[],"sources":["../src/statistics/power-and-mde.ts","../src/statistics/random.ts","../src/experiment-tracker.ts"],"mappings":";;;;;;;;iBAWgB,mBAAmB;EACjC;EACA;EACA;EACA;;;;;;;;;;;iBAsBc,yBAAyB;EACvC;EACA;EACA;EACA;;;;;;;;iBAkBc,UAAU;EACxB;EACA;EACA;EACA;;;;;;;;;;;;;;;;iBAyBc,iBAAiB;EAC/B;EACA;EACA;EACA;EACA;;;;;;;iBAyBc,aAAa;EAC3B;EACA;EACA;EACA;EACA;;;;;;;;iBCrHc,WAAW;;;;;KCyBf;;UAGK;;EAEf;;EAEA;;EAEA;;;;UAKe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA,WAAW;;EAEX;;EAEA,UAAU;;UAGK;EACf;EACA;EACA;EACA;;EAEA;;EAEA;;;EAGA;;EAEA;;EAEA;;UAGe;;EAEf;;EAEA;;EAEA,YAAY;;EAEZ;;EAEA;EACA,MAAM;EACN,OAAO;EACP,SAAS;;EAET;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;;EAGA;;UAGe;EACf,SAAS;;EAET;;EAEA;;;;;;;iBAkDc,uBACd,MAAM,iBACN,aAAa,wBACZ;;;;;;;iBAgDa,mBACd,WAAW,iBACX,QAAQ,wBACR,aAAa,wBACZ;;;KA6CS,yBAAyB,uBAAuB,QAAQ;;;UAInD;EACf,QAAQ,QAAQ;EAChB,KAAK,aAAa,eAAe;;;;iBAoBnB,wBAAwB,UAAS,eAAoB;;iBAarD,oBAAoB,eAAe;UAyBlC;EACf,QAAQ;EACR,mBAAmB;EACnB,aAAa;;EAEb;;UAGe;EACf;EACA;EACA;EACA;;EAEA,aAAa;;;;;;;;;cAUF;mBACM;mBACA;mBACA;mBACA;EAEjB,YAAY,UAAS;EAOf,OAAO,OAAO,wBAAwB,QAAQ;;;EA6B9C,OACJ,sBACA,KAAK,KAAK;IAAwC;IAAc;MAC/D,QAAQ;EAqCL,IAAI,uBAAuB,QAAQ;EAMnC,QAAQ,QAAQ;;EAKhB,WAAW,uBAAuB,QAAQ"}
+{"version":3,"file":"experiment-tracker-DCO6Cz4s.d.ts","names":[],"sources":["../src/statistics/power-and-mde.ts","../src/statistics/random.ts","../src/experiment-tracker.ts"],"mappings":";;;;;;;;iBAWgB,mBAAmB;EACjC;EACA;EACA;EACA;;;;;;;;;;;iBAsBc,yBAAyB;EACvC;EACA;EACA;EACA;;;;;;;;iBAkBc,UAAU;EACxB;EACA;EACA;EACA;;;;;;;;;;;;;;;;iBAyBc,iBAAiB;EAC/B;EACA;EACA;EACA;EACA;;;;;;;iBAyBc,aAAa;EAC3B;EACA;EACA;EACA;EACA;;;;;;;;iBCrHc,WAAW;;;;;KCyBf;;UAGK;;EAEf;;EAEA;;EAEA;;;;UAKe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA,WAAW;;EAEX;;EAEA,UAAU;;UAGK;EACf;EACA;EACA;EACA;;EAEA;;EAEA;;;EAGA;;EAEA;;EAEA;;UAGe;;EAEf;;EAEA;;EAEA,YAAY;;EAEZ;;EAEA;EACA,MAAM;EACN,OAAO;EACP,SAAS;;EAET;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;;EAGA;;UAGe;EACf,SAAS;;EAET;;EAEA;;;;;;;iBAkDc,uBACd,MAAM,iBACN,aAAa,wBACZ;;;;;;;iBAgDa,mBACd,WAAW,iBACX,QAAQ,wBACR,aAAa,wBACZ;;;KA6CS,yBAAyB,uBAAuB,QAAQ;;;UAInD;EACf,QAAQ,QAAQ;EAChB,KAAK,aAAa,eAAe;;;;iBAoBnB,wBAAwB,UAAS,eAAoB;;iBAarD,oBAAoB,eAAe;UAyBlC;EACf,QAAQ;EACR,mBAAmB;EACnB,aAAa;;EAEb;;UAGe;EACf;EACA;EACA;EACA;;EAEA,aAAa;;;;;;;;;cAUF;mBACM;mBACA;mBACA;mBACA;EAEjB,YAAY,UAAS;EAOf,OAAO,OAAO,wBAAwB,QAAQ;;;EA6B9C,OACJ,sBACA,KAAK,KAAK;IAAwC;IAAc;MAC/D,QAAQ;EAqCL,IAAI,uBAAuB,QAAQ;EAMnC,QAAQ,QAAQ;;EAKhB,WAAW,uBAAuB,QAAQ"}

package/dist/{exporters-q9iL-2Jf.js → exporters-Df7TgHFv.js}

@@ -1,4 +1,4 @@
-import { l as assertRewardGate } from "./schema-C6DW4ZHR.js";
+import { s as assertRewardGate } from "./schema-C1aaAxTf.js";
 //#region src/rollout/exporters.ts
 function realnessLabels(line) {
 	return {
@@ -143,6 +143,6 @@ function isTrainingLineEligible(line, options) {
 	return isSplitEligible(line, options);
 }
 //#endregion
-export { toRftItem as a, toVerifiersRolloutOutput as c, toRewardRows as i, toVerifiersRolloutOutputs as l, realnessLabels as n, toRftItems as o, toJsonl as r, toSftRows as s, isSplitEligible as t };
+export { toRftItems as a, toVerifiersRolloutOutputs as c, toRftItem as i, toJsonl as n, toSftRows as o, toRewardRows as r, toVerifiersRolloutOutput as s, isSplitEligible as t };
 
-//# sourceMappingURL=exporters-q9iL-2Jf.js.map
+//# sourceMappingURL=exporters-Df7TgHFv.js.map

package/dist/exporters-Df7TgHFv.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"exporters-Df7TgHFv.js","names":[],"sources":["../src/rollout/exporters.ts"],"sourcesContent":["/**\n * Pure exporters over `tangle.rollout.v1` lines → the training-data shapes\n * the improvement loops feed:\n *   - SFT chat JSONL           (clean trainable successes, {messages, metadata})\n *   - reward rows              (every scored line, success or failure, with steps)\n *   - Prime Intellect verifiers RolloutOutput (prompt/completion split + reward)\n *   - OpenAI RFT items         (prompt turns + verdict reference fields)\n *\n * All exporters are pure functions of the lines — filtering (never train on\n * holdout, reward thresholds, the realness gate) happens HERE, on inline\n * labels, no joins.\n *\n * Every exporter takes `MintedRolloutLine[]`, not `RolloutLine[]`: the reward\n * on a minted line has been checked against the anti-Goodhart invariant, and\n * the brand is what stops a hand-built object literal claiming a positive\n * reward on a gamed run from being handed to an exporter that copies it\n * verbatim into training data.\n */\n\nimport type {\n  ChatMessage,\n  MintedRolloutLine,\n  RolloutLine,\n  RolloutSplit,\n  RolloutStep,\n  ToolDef,\n} from './schema'\nimport { assertRewardGate } from './schema'\n\n// ---------------------------------------------------------------------------\n// The realness claims every emitted row carries.\n// ---------------------------------------------------------------------------\n\n/**\n * The gate's two claims, which travel TOGETHER on every emitted row.\n *\n * `realness_gated` alone is ambiguous, and the ambiguity is exploitable:\n * `false` reads as \"we screened it and nothing fired\", so a producer that has no\n * screen at all emitted rows indistinguishable from screened-clean ones, and\n * every consumer of the published dataset read them as clean. The second field\n * is what separates the two claims, and it only removes the ambiguity if it\n * reaches the WIRE — for a round it existed on `RolloutOutcome` and on no\n * exported row shape at all, which left the published rows exactly as ambiguous\n * as before.\n *\n * So there is one helper and every row shape spreads it. A row that states one\n * claim without the other is not constructible by copying the pattern, and\n * `exporters.test.ts` walks every emitted shape to prove none does.\n */\nexport interface RealnessLabels {\n  /** The screen's VERDICT: the run faked its success signal. */\n  realness_gated: boolean\n  /**\n   * Whether a screen RAN at all. `true` = it ran, so `realness_gated` is its\n   * verdict. `false` = the producer declares it has none. `null` = not stated\n   * (pre-unification producers), which is \"unknown\" and never \"clean\".\n   */\n  realness_screened: boolean | null\n}\n\nfunction realnessLabels(line: MintedRolloutLine): RealnessLabels {\n  return {\n    realness_gated: line.outcome.realness_gated === true,\n    // `?? null` rather than `?? false`: absent means the producer never stated\n    // it, and `false` is a load-bearing claim (\"no screen exists\") that would be\n    // an invention here.\n    realness_screened: line.outcome.realness_screened ?? null,\n  }\n}\n\n// ---------------------------------------------------------------------------\n// (a) SFT chat JSONL\n// ---------------------------------------------------------------------------\n\nexport interface TrainingExportOptions {\n  /** Include held-out evaluation data in training output. Default false. */\n  allowHeldOutTrainingData?: boolean\n  /** Require reward to be strictly greater than this value. Default 0. */\n  minimumQualityExclusive?: number\n}\n\n/**\n * What a signed-signal exporter (verifiers, RFT) does with lines that are not\n * clean trainable successes — realness-gated lines above all.\n *\n *   - 'exclude'       — the default, the same fail-closed policy as every\n *                       other training export: positive, completed,\n *                       non-gated rows on a trainable split.\n *   - 'zero-and-flag' — keep them, at their non-positive (or null) reward,\n *                       with `RealnessLabels` on the row. The dataset release\n *                       sets this per `FORMAT_GATE_DISPOSITION`: in these\n *                       formats the reward is a signed learning signal, so a\n *                       gamed trajectory at reward 0 is a correct negative,\n *                       and dropping it would bias the negative population\n *                       toward honest failures and leave a trainer no example\n *                       of gaming being penalized. The split policy is NOT\n *                       relaxed: held-out lines still need the named opt-in.\n *\n * SFT deliberately has no such option — an SFT row is an imitation target and\n * a gamed trajectory must never appear in one at any weight.\n */\nexport type GatedLineDisposition = 'exclude' | 'zero-and-flag'\n\nexport interface SignedSignalExportOptions extends TrainingExportOptions {\n  /** Disposition for non-trainable lines. Default 'exclude'. */\n  gatedLines?: GatedLineDisposition\n}\n\nexport type SftExportOptions = TrainingExportOptions\n\nexport interface SftRow {\n  messages: ChatMessage[]\n  metadata: {\n    rollout_id: string\n    run_id: string\n    candidate_id: string | null\n    instance_id: string\n    reward: number\n  } & RealnessLabels\n}\n\n/**\n * Supervised fine-tune rows: the completed conversation of each qualifying\n * line. Fail-closed filters: trainable split only (never holdout/canary),\n * reward strictly above `minimumQualityExclusive` (default 0), realness-gated\n * lines never qualify, gap lines carry no trainable content, and\n * copied-context turns are dropped from the transcript (Harbor ATIF RFC 0001\n * rule 7 — see `ChatMessage.is_copied_context`).\n *\n * `realness_gated` is therefore always `false` on an emitted row. It is carried\n * anyway: an SFT row is a pure imitation target, so the row states its realness\n * claims instead of making the reader know the format's policy, and carrying\n * both flags on all four shapes is what lets the release accounting measure\n * every config with one rule rather than skipping the one whose row shape\n * happened to omit the field.\n */\nexport function toSftRows(lines: MintedRolloutLine[], options: SftExportOptions = {}): SftRow[] {\n  for (const line of lines) assertRewardGate(line, 'SFT export')\n  return lines\n    .filter((line) => isTrainingLineEligible(line, options) && line.messages.length > 0)\n    .map((line) => ({\n      // Dropped, not kept-and-masked: a copied-context turn was authored by\n      // another agent, and an SFT trainer has no notion of \"present but not a\n      // target\" — every message it is handed is something to imitate.\n      messages: line.messages.filter((message) => message.is_copied_context !== true),\n      metadata: {\n        rollout_id: line.rollout_id,\n        run_id: line.run_id,\n        candidate_id: line.candidate_id ?? null,\n        instance_id: line.task.instance_id,\n        reward: line.outcome.reward as number,\n        ...realnessLabels(line),\n      },\n    }))\n    .filter((row) => row.messages.length > 0)\n}\n\n// ---------------------------------------------------------------------------\n// (b) Reward rows\n// ---------------------------------------------------------------------------\n\nexport interface RewardRow {\n  /** First user turn — the task prompt. */\n  prompt: string\n  steps: RolloutStep[]\n  reward: number\n  metadata: {\n    rollout_id: string\n    run_id: string\n    candidate_id: string | null\n    instance_id: string\n    split: RolloutSplit\n    /**\n     * The anti-Goodhart verdict AND whether a screen produced it, carried on the\n     * row rather than left implicit in a reward of 0. Zeroing alone is lossy: it\n     * makes a run that FAKED its success indistinguishable from one that\n     * honestly failed, so a consumer can neither drop the gamed population nor\n     * mine it (a labeled gamed trajectory is the training signal for a gaming\n     * detector). `realness_gated` alone is lossy the same way in the other\n     * direction — see `RealnessLabels`.\n     */\n  } & RealnessLabels\n}\n\n/**\n * Reward-labeled rows for completed, positive-quality training runs.\n */\nexport function toRewardRows(\n  lines: MintedRolloutLine[],\n  options: TrainingExportOptions = {},\n): RewardRow[] {\n  for (const line of lines) assertRewardGate(line, 'reward-row export')\n  return lines\n    .filter(\n      (line) =>\n        isTrainingLineEligible(line, options) &&\n        line.messages.some(\n          (message) =>\n            message.role === 'user' &&\n            typeof message.content === 'string' &&\n            message.content.length > 0,\n        ),\n    )\n    .map((line) => ({\n      prompt: line.messages.find((m) => m.role === 'user')?.content ?? '',\n      steps: line.steps ?? [],\n      reward: line.outcome.reward as number,\n      metadata: {\n        rollout_id: line.rollout_id,\n        run_id: line.run_id,\n        candidate_id: line.candidate_id ?? null,\n        instance_id: line.task.instance_id,\n        split: line.task.split,\n        ...realnessLabels(line),\n      },\n    }))\n}\n\n// ---------------------------------------------------------------------------\n// (c) Prime Intellect verifiers RolloutOutput\n// ---------------------------------------------------------------------------\n\nexport interface VerifiersTokenUsage {\n  input_tokens: number | null\n  output_tokens: number | null\n  reasoning_tokens: number | null\n  cache_read_tokens: number | null\n  cache_write_tokens: number | null\n}\n\nexport interface VerifiersRolloutOutput {\n  /** Messages through the last turn BEFORE the first assistant turn. */\n  prompt: ChatMessage[]\n  /** The first assistant turn onward — what the policy produced. */\n  completion: ChatMessage[]\n  reward: number | null\n  metrics: Record<string, unknown>\n  tool_defs: ToolDef[]\n  token_usage: VerifiersTokenUsage\n  info: {\n    task: RolloutLine['task']\n    policy: RolloutLine['policy']\n    rollout_id: string\n    run_id: string\n    experiment_id: string | null\n    candidate_id: string | null\n    generation: number | null\n    candidate_index: number | null\n    role: RolloutLine['role']\n  } & RealnessLabels\n}\n\n/** Index of the first assistant turn; messages.length when none exists. */\nfunction firstAssistantIndex(messages: ChatMessage[]): number {\n  const index = messages.findIndex((m) => m.role === 'assistant')\n  return index === -1 ? messages.length : index\n}\n\nexport function toVerifiersRolloutOutput(line: MintedRolloutLine): VerifiersRolloutOutput {\n  assertRewardGate(line, 'verifiers export')\n  const split = firstAssistantIndex(line.messages)\n  return {\n    prompt: line.messages.slice(0, split),\n    completion: line.messages.slice(split),\n    reward: line.outcome.reward,\n    metrics: line.outcome.metrics,\n    tool_defs: line.tool_defs,\n    token_usage: {\n      input_tokens: line.cost.tokens_in,\n      output_tokens: line.cost.tokens_out,\n      reasoning_tokens: line.cost.tokens_reasoning,\n      cache_read_tokens: line.cost.cache_read,\n      cache_write_tokens: line.cost.cache_write,\n    },\n    info: {\n      task: line.task,\n      policy: line.policy,\n      rollout_id: line.rollout_id,\n      run_id: line.run_id,\n      experiment_id: line.experiment_id ?? null,\n      candidate_id: line.candidate_id ?? null,\n      generation: line.generation,\n      candidate_index: line.candidate_index,\n      role: line.role,\n      ...realnessLabels(line),\n    },\n  }\n}\n\nexport function toVerifiersRolloutOutputs(\n  lines: MintedRolloutLine[],\n  options: SignedSignalExportOptions = {},\n): VerifiersRolloutOutput[] {\n  if (options.gatedLines === 'zero-and-flag') {\n    return lines\n      .filter((line) => isSplitEligible(line, options) && line.messages.length > 0)\n      .map(toVerifiersRolloutOutput)\n  }\n  return lines\n    .filter(\n      (line) =>\n        isTrainingLineEligible(line, options) &&\n        firstAssistantIndex(line.messages) > 0 &&\n        firstAssistantIndex(line.messages) < line.messages.length,\n    )\n    .map(toVerifiersRolloutOutput)\n}\n\n// ---------------------------------------------------------------------------\n// (d) OpenAI RFT items\n// ---------------------------------------------------------------------------\n\nexport interface RftItem {\n  /** Prompt turns only — the graded completion is re-sampled during RFT. */\n  messages: ChatMessage[]\n  /** Verdict/label fields the grader references as item.reference.* */\n  reference: {\n    reward: number | null\n    reward_source: string | null\n    verdict: unknown\n    instance_id: string\n    suite: string\n    split: RolloutSplit\n    rollout_id: string\n  } & RealnessLabels\n}\n\nexport function toRftItem(line: MintedRolloutLine): RftItem {\n  assertRewardGate(line, 'RFT export')\n  const split = firstAssistantIndex(line.messages)\n  return {\n    messages: line.messages.slice(0, split),\n    reference: {\n      reward: line.outcome.reward,\n      reward_source: line.outcome.reward_source,\n      verdict: line.outcome.verdict,\n      instance_id: line.task.instance_id,\n      suite: line.task.suite,\n      split: line.task.split,\n      rollout_id: line.rollout_id,\n      ...realnessLabels(line),\n    },\n  }\n}\n\n/** RFT needs a real prompt: lines whose transcript starts with prompt turns. */\nexport function toRftItems(\n  lines: MintedRolloutLine[],\n  options: SignedSignalExportOptions = {},\n): RftItem[] {\n  return lines\n    .filter(\n      (line) =>\n        (options.gatedLines === 'zero-and-flag'\n          ? isSplitEligible(line, options)\n          : isTrainingLineEligible(line, options)) &&\n        line.messages.length > 0 &&\n        firstAssistantIndex(line.messages) > 0,\n    )\n    .map(toRftItem)\n}\n\n// ---------------------------------------------------------------------------\n// Serialization — one JSON object per line, the interchange format for\n// every export. `tangle.rollout.v1` lines and export rows alike.\n// ---------------------------------------------------------------------------\n\nexport function toJsonl(rows: ReadonlyArray<unknown>): string {\n  return rows.map((r) => JSON.stringify(r)).join('\\n') + (rows.length ? '\\n' : '')\n}\n\n/**\n * The split half of the training policy alone: `search` is trainable, held-out\n * needs the named opt-in, `dev` and `canary` never ship. This is the ONE check\n * `'zero-and-flag'` does not relax — a gated line is shipped as a labeled\n * negative, not as a licence to train on evaluation data.\n *\n * Exported as the single implementation of that rule: `rl/exporters` applies\n * it on its line paths too, so the two waists cannot drift on which splits are\n * trainable.\n */\nexport function isSplitEligible(\n  line: RolloutLine,\n  options: Pick<TrainingExportOptions, 'allowHeldOutTrainingData'>,\n): boolean {\n  if (line.task.split === 'search') return true\n  return line.task.split === 'holdout' && options.allowHeldOutTrainingData === true\n}\n\nfunction isTrainingLineEligible(\n  line: RolloutLine,\n  options: TrainingExportOptions,\n): line is RolloutLine & { outcome: RolloutLine['outcome'] & { reward: number } } {\n  const minimumQualityExclusive = options.minimumQualityExclusive ?? 0\n  if (!Number.isFinite(minimumQualityExclusive)) {\n    throw new Error('minimumQualityExclusive must be finite')\n  }\n\n  const reward = line.outcome.reward\n  if (reward === null) return false\n  if (!Number.isFinite(reward)) {\n    throw new Error(`training reward for rollout \"${line.rollout_id}\" must be finite`)\n  }\n  if (reward <= minimumQualityExclusive) return false\n  if (!line.outcome.is_completed || line.outcome.is_truncated || line.outcome.error !== null) {\n    return false\n  }\n  if (line.outcome.realness_gated === true) return false\n  return isSplitEligible(line, options)\n}\n"],"mappings":";;AA4DA,SAAS,eAAe,MAAyC;CAC/D,OAAO;EACL,gBAAgB,KAAK,QAAQ,mBAAmB;EAIhD,mBAAmB,KAAK,QAAQ,qBAAqB;CACvD;AACF;;;;;;;;;;;;;;;;AAoEA,SAAgB,UAAU,OAA4B,UAA4B,CAAC,GAAa;CAC9F,KAAK,MAAM,QAAQ,OAAO,iBAAiB,MAAM,YAAY;CAC7D,OAAO,MACJ,QAAQ,SAAS,uBAAuB,MAAM,OAAO,KAAK,KAAK,SAAS,SAAS,CAAC,CAAC,CACnF,KAAK,UAAU;EAId,UAAU,KAAK,SAAS,QAAQ,YAAY,QAAQ,sBAAsB,IAAI;EAC9E,UAAU;GACR,YAAY,KAAK;GACjB,QAAQ,KAAK;GACb,cAAc,KAAK,gBAAgB;GACnC,aAAa,KAAK,KAAK;GACvB,QAAQ,KAAK,QAAQ;GACrB,GAAG,eAAe,IAAI;EACxB;CACF,EAAE,CAAC,CACF,QAAQ,QAAQ,IAAI,SAAS,SAAS,CAAC;AAC5C;;;;AAgCA,SAAgB,aACd,OACA,UAAiC,CAAC,GACrB;CACb,KAAK,MAAM,QAAQ,OAAO,iBAAiB,MAAM,mBAAmB;CACpE,OAAO,MACJ,QACE,SACC,uBAAuB,MAAM,OAAO,KACpC,KAAK,SAAS,MACX,YACC,QAAQ,SAAS,UACjB,OAAO,QAAQ,YAAY,YAC3B,QAAQ,QAAQ,SAAS,CAC7B,CACJ,CAAC,CACA,KAAK,UAAU;EACd,QAAQ,KAAK,SAAS,MAAM,MAAM,EAAE,SAAS,MAAM,CAAC,EAAE,WAAW;EACjE,OAAO,KAAK,SAAS,CAAC;EACtB,QAAQ,KAAK,QAAQ;EACrB,UAAU;GACR,YAAY,KAAK;GACjB,QAAQ,KAAK;GACb,cAAc,KAAK,gBAAgB;GACnC,aAAa,KAAK,KAAK;GACvB,OAAO,KAAK,KAAK;GACjB,GAAG,eAAe,IAAI;EACxB;CACF,EAAE;AACN;;AAqCA,SAAS,oBAAoB,UAAiC;CAC5D,MAAM,QAAQ,SAAS,WAAW,MAAM,EAAE,SAAS,WAAW;CAC9D,OAAO,UAAU,KAAK,SAAS,SAAS;AAC1C;AAEA,SAAgB,yBAAyB,MAAiD;CACxF,iBAAiB,MAAM,kBAAkB;CACzC,MAAM,QAAQ,oBAAoB,KAAK,QAAQ;CAC/C,OAAO;EACL,QAAQ,KAAK,SAAS,MAAM,GAAG,KAAK;EACpC,YAAY,KAAK,SAAS,MAAM,KAAK;EACrC,QAAQ,KAAK,QAAQ;EACrB,SAAS,KAAK,QAAQ;EACtB,WAAW,KAAK;EAChB,aAAa;GACX,cAAc,KAAK,KAAK;GACxB,eAAe,KAAK,KAAK;GACzB,kBAAkB,KAAK,KAAK;GAC5B,mBAAmB,KAAK,KAAK;GAC7B,oBAAoB,KAAK,KAAK;EAChC;EACA,MAAM;GACJ,MAAM,KAAK;GACX,QAAQ,KAAK;GACb,YAAY,KAAK;GACjB,QAAQ,KAAK;GACb,eAAe,KAAK,iBAAiB;GACrC,cAAc,KAAK,gBAAgB;GACnC,YAAY,KAAK;GACjB,iBAAiB,KAAK;GACtB,MAAM,KAAK;GACX,GAAG,eAAe,IAAI;EACxB;CACF;AACF;AAEA,SAAgB,0BACd,OACA,UAAqC,CAAC,GACZ;CAC1B,IAAI,QAAQ,eAAe,iBACzB,OAAO,MACJ,QAAQ,SAAS,gBAAgB,MAAM,OAAO,KAAK,KAAK,SAAS,SAAS,CAAC,CAAC,CAC5E,IAAI,wBAAwB;CAEjC,OAAO,MACJ,QACE,SACC,uBAAuB,MAAM,OAAO,KACpC,oBAAoB,KAAK,QAAQ,IAAI,KACrC,oBAAoB,KAAK,QAAQ,IAAI,KAAK,SAAS,MACvD,CAAC,CACA,IAAI,wBAAwB;AACjC;AAqBA,SAAgB,UAAU,MAAkC;CAC1D,iBAAiB,MAAM,YAAY;CACnC,MAAM,QAAQ,oBAAoB,KAAK,QAAQ;CAC/C,OAAO;EACL,UAAU,KAAK,SAAS,MAAM,GAAG,KAAK;EACtC,WAAW;GACT,QAAQ,KAAK,QAAQ;GACrB,eAAe,KAAK,QAAQ;GAC5B,SAAS,KAAK,QAAQ;GACtB,aAAa,KAAK,KAAK;GACvB,OAAO,KAAK,KAAK;GACjB,OAAO,KAAK,KAAK;GACjB,YAAY,KAAK;GACjB,GAAG,eAAe,IAAI;EACxB;CACF;AACF;;AAGA,SAAgB,WACd,OACA,UAAqC,CAAC,GAC3B;CACX,OAAO,MACJ,QACE,UACE,QAAQ,eAAe,kBACpB,gBAAgB,MAAM,OAAO,IAC7B,uBAAuB,MAAM,OAAO,MACxC,KAAK,SAAS,SAAS,KACvB,oBAAoB,KAAK,QAAQ,IAAI,CACzC,CAAC,CACA,IAAI,SAAS;AAClB;AAOA,SAAgB,QAAQ,MAAsC;CAC5D,OAAO,KAAK,KAAK,MAAM,KAAK,UAAU,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,KAAK,KAAK,SAAS,OAAO;AAC/E;;;;;;;;;;;AAYA,SAAgB,gBACd,MACA,SACS;CACT,IAAI,KAAK,KAAK,UAAU,UAAU,OAAO;CACzC,OAAO,KAAK,KAAK,UAAU,aAAa,QAAQ,6BAA6B;AAC/E;AAEA,SAAS,uBACP,MACA,SACgF;CAChF,MAAM,0BAA0B,QAAQ,2BAA2B;CACnE,IAAI,CAAC,OAAO,SAAS,uBAAuB,GAC1C,MAAM,IAAI,MAAM,wCAAwC;CAG1D,MAAM,SAAS,KAAK,QAAQ;CAC5B,IAAI,WAAW,MAAM,OAAO;CAC5B,IAAI,CAAC,OAAO,SAAS,MAAM,GACzB,MAAM,IAAI,MAAM,gCAAgC,KAAK,WAAW,iBAAiB;CAEnF,IAAI,UAAU,yBAAyB,OAAO;CAC9C,IAAI,CAAC,KAAK,QAAQ,gBAAgB,KAAK,QAAQ,gBAAgB,KAAK,QAAQ,UAAU,MACpF,OAAO;CAET,IAAI,KAAK,QAAQ,mBAAmB,MAAM,OAAO;CACjD,OAAO,gBAAgB,MAAM,OAAO;AACtC"}

package/dist/{external-optimizer-contracts-DbLsm4Po.d.ts → external-optimizer-contracts-szBJ_1vh.d.ts}

@@ -1,5 +1,5 @@
 import { b as CustomTokenPricing, g as CostReceiptInput } from "./cost-ledger-DbQdN3nO.js";
-import { h as ChatResponse, m as ChatRequest } from "./types-jUBXJ7Iz.js";
+import { h as ChatResponse, m as ChatRequest } from "./types-Bfk0uxRj.js";
 //#region src/campaign/external-optimizer-contracts.d.ts
 interface ExternalOptimizerProcessLimits {
   /** Maximum serialized input written for the child process. */
@@ -169,4 +169,4 @@ interface ExternalOptimizerWireCounts {
 }
 //#endregion
 export { ExternalTextCandidate as _, ExternalOptimizerEndpointFormat as a, resolveExternalOptimizerProcessLimits as b, ExternalOptimizerModelBudget as c, ExternalOptimizerModelCallResult as d, ExternalOptimizerModelExecutionObservation as f, ExternalOptimizerWireCounts as g, ExternalOptimizerRunnerCommand as h, ExternalOptimizerChatRequest as i, ExternalOptimizerModelCall as l, ExternalOptimizerResumeMode as m, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS as n, ExternalOptimizerEvaluationObservation as o, ExternalOptimizerProcessLimits as p, ExternalOptimizerCallbackLimits as r, ExternalOptimizerEvaluationRefusalReason as s, DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS as t, ExternalOptimizerModelCallRequest as u, ExternalTextEvaluationRequest as v, resolveExternalOptimizerCallbackLimits as y };
-//# sourceMappingURL=external-optimizer-contracts-DbLsm4Po.d.ts.map
+//# sourceMappingURL=external-optimizer-contracts-szBJ_1vh.d.ts.map

package/dist/{external-optimizer-contracts-DbLsm4Po.d.ts.map → external-optimizer-contracts-szBJ_1vh.d.ts.map}

@@ -1 +1 @@
-{"version":3,"file":"external-optimizer-contracts-DbLsm4Po.d.ts","names":[],"sources":["../src/campaign/external-optimizer-contracts.ts"],"mappings":";;;UAWiB;;EAEf;;EAEA;;EAEA;;cAGW,2CAA2C,SAAS;UAOhD;;EAEf;;EAEA;;cAGW,4CAA4C,SAAS;UAMjD;EACf;EACA;EACA,MAAM,OAAO;;EAEb,SAAS,QAAQ;;KAGP;KAEA,iCAAiC;UAE5B;EACf,WAAW;EACX;;iBAUc,sCACd,OAAO,QAAQ,6CACf,iBACC;iBAmBa,uCACd,OAAO,QAAQ,8CACf,iBACC;KAkBE,aAAa,KAAK,eAAc,6BACjC,IACA,0BAA0B,gBACf,aAAa,OACtB,+BACc,WAAW,IAAI,aAAa,EAAE,SAC1C;;KAGI,+BAA+B,aACzC,KAAK;EAA0B;;KAGrB;;UAMK;;WAEN;;WAEA,SAAS;;WAET,iBAAiB;WACjB,QAAQ;;;KAIP;WAEG;;WAEA,UAAU;;;;;;;WAOV,SAPU;;WASV;;WAGA;;WAEA;;WAEA,SATkC;;WAWlC;;;;;;;;;;;;KAaH,8BACV,SAAS,sCACN,QAAQ;;KAGD;WAEG;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;WAGA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;KAGH;;;;;;KAUA;WAEG;WACA;WACA,WAAW;WACX;;WAGA;WACA;WACA,WAAW;WACX;WACA;;WAEA;WACA;;WAGA;WACA;WACA,QAAQ;WACR,YAAY;WACZ;WACA;;UAGE;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;EAUA;;;;;;EAMA,UAAU;;EAEV;;;UAIe;;EAEf;;EAEA"}
+{"version":3,"file":"external-optimizer-contracts-szBJ_1vh.d.ts","names":[],"sources":["../src/campaign/external-optimizer-contracts.ts"],"mappings":";;;UAWiB;;EAEf;;EAEA;;EAEA;;cAGW,2CAA2C,SAAS;UAOhD;;EAEf;;EAEA;;cAGW,4CAA4C,SAAS;UAMjD;EACf;EACA;EACA,MAAM,OAAO;;EAEb,SAAS,QAAQ;;KAGP;KAEA,iCAAiC;UAE5B;EACf,WAAW;EACX;;iBAUc,sCACd,OAAO,QAAQ,6CACf,iBACC;iBAmBa,uCACd,OAAO,QAAQ,8CACf,iBACC;KAkBE,aAAa,KAAK,eAAc,6BACjC,IACA,0BAA0B,gBACf,aAAa,OACtB,+BACc,WAAW,IAAI,aAAa,EAAE,SAC1C;;KAGI,+BAA+B,aACzC,KAAK;EAA0B;;KAGrB;;UAMK;;WAEN;;WAEA,SAAS;;WAET,iBAAiB;WACjB,QAAQ;;;KAIP;WAEG;;WAEA,UAAU;;;;;;;WAOV,SAPU;;WASV;;WAGA;;WAEA;;WAEA,SATkC;;WAWlC;;;;;;;;;;;;KAaH,8BACV,SAAS,sCACN,QAAQ;;KAGD;WAEG;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;WAGA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;KAGH;;;;;;KAUA;WAEG;WACA;WACA,WAAW;WACX;;WAGA;WACA;WACA,WAAW;WACX;WACA;;WAEA;WACA;;WAGA;WACA;WACA,QAAQ;WACR,YAAY;WACZ;WACA;;UAGE;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;EAUA;;;;;;EAMA,UAAU;;EAEV;;;UAIe;;EAEf;;EAEA"}

package/dist/{external-optimizer-process-x9oEXKsU.js → external-optimizer-process-WosTBChy.js}

@@ -1,6 +1,6 @@
-import { n as contentHash } from "./verdict-cache-mZf5FEiY.js";
-import { m as tryAcquireAtomicFileLock, p as probeAtomicFileLock } from "./ledger-core-DTae9rv_.js";
-import { _ as resolveExternalOptimizerCallbackLimits, a as closeServer, h as isRecord, m as isExternalTextCandidate, o as listenLocal, s as sendJson } from "./external-optimizer-subprocess-CKNb42oM.js";
+import { n as contentHash } from "./verdict-cache-CdVVTVmn.js";
+import { m as tryAcquireAtomicFileLock, p as probeAtomicFileLock } from "./ledger-core-BOzlRygb.js";
+import { _ as resolveExternalOptimizerCallbackLimits, a as closeServer, h as isRecord, m as isExternalTextCandidate, o as listenLocal, s as sendJson } from "./external-optimizer-subprocess-BIWbHpgD.js";
 import { createServer } from "node:http";
 //#region src/campaign/single-run-lock.ts
 /**
@@ -298,4 +298,4 @@ function cloneCandidate(candidate) {
 //#endregion
 export { acquireSingleRunLock as n, startExternalOptimizerCallback as t };
 
-//# sourceMappingURL=external-optimizer-process-x9oEXKsU.js.map
+//# sourceMappingURL=external-optimizer-process-WosTBChy.js.map

package/dist/{external-optimizer-process-x9oEXKsU.js.map → external-optimizer-process-WosTBChy.js.map}

@@ -1 +1 @@
-{"version":3,"file":"external-optimizer-process-x9oEXKsU.js","names":[],"sources":["../src/campaign/single-run-lock.ts","../src/campaign/external-optimizer-callback.ts"],"sourcesContent":["/**\n * Single-run lock for evaluations that share one mutable environment.\n *\n * Two concurrent runs against a shared stateful gym silently corrupt each\n * other: each resets/mutates environment state mid-cell of the other, and\n * every score from both becomes garbage that LOOKS like worker variance\n * (agent-lab R357 burned hours on flip-flopping scores before tracing them\n * to exactly this). The fix is a pid lockfile: refuse to start while a live\n * holder exists, reclaim stale locks whose pid is gone, release only if the\n * lock is still ours.\n *\n * `alsoCheck` exists because independent runners can guard the same shared\n * resource with differently named lockfiles; a runner must respect all of\n * them even though it writes only its own.\n */\n\nimport {\n  type AtomicFileLockUnavailable,\n  probeAtomicFileLock,\n  tryAcquireAtomicFileLock,\n} from '../ledger-core/atomic-file-lock'\n\nexport interface SingleRunLockOptions {\n  /** Lockfile this runner writes (and checks). */\n  readonly lockPath: string\n  /** Other runners' lockfiles guarding the same resource; checked, never written. */\n  readonly alsoCheck?: readonly string[]\n  /** Install a process 'exit' hook that releases the lock. Default true. */\n  readonly releaseOnExit?: boolean\n  /** Owner pid recorded in the lockfile metadata. Default process.pid. */\n  readonly pid?: number\n}\n\nexport interface SingleRunLock {\n  /** Remove the lockfile if this process still owns it. Idempotent. */\n  release(): void\n}\n\nfunction assertAvailable(path: string): void {\n  const unavailable = probeAtomicFileLock({ lockPath: path })\n  if (unavailable) throw unavailableError(path, unavailable)\n}\n\nfunction unavailableError(path: string, unavailable: AtomicFileLockUnavailable): Error {\n  if (unavailable.reason === 'recovery') {\n    return new Error(\n      `single-run lock recovery is already in progress (${path}); refusing a concurrent run`,\n    )\n  }\n  return new Error(\n    `single-run lock held by live pid ${unavailable.holder.pid} (${path}); refusing a concurrent run on the shared resource`,\n  )\n}\n\n/**\n * Acquire the lock or throw naming the live holder. A stale lock (holder pid\n * no longer running) is reclaimed by one contender. An interrupted reclaim\n * leaves a marker that fails closed instead of admitting overlapping runs.\n */\nexport function acquireSingleRunLock(opts: SingleRunLockOptions): SingleRunLock {\n  const pid = opts.pid ?? process.pid\n  if (!Number.isSafeInteger(pid) || pid <= 0)\n    throw new Error('single-run lock pid must be a positive integer')\n  for (const path of opts.alsoCheck ?? []) assertAvailable(path)\n  const acquisition = tryAcquireAtomicFileLock({\n    lockPath: opts.lockPath,\n    pid,\n  })\n  if (!acquisition.acquired) throw unavailableError(opts.lockPath, acquisition)\n  const releaseOnExit = opts.releaseOnExit ?? true\n  let released = false\n  const release = (): void => {\n    if (released) return\n    released = true\n    if (releaseOnExit) process.off('exit', release)\n    try {\n      acquisition.lock.release()\n    } catch {\n      // release is best-effort; a leftover stale lock is reclaimed on next acquire\n    }\n  }\n  try {\n    for (const path of opts.alsoCheck ?? []) assertAvailable(path)\n  } catch (error) {\n    release()\n    throw error\n  }\n  if (releaseOnExit) process.on('exit', release)\n  return { release }\n}\n","import { createServer, type IncomingMessage, type ServerResponse } from 'node:http'\nimport { contentHash } from '../verdict-cache'\nimport {\n  type ExternalOptimizerCallback,\n  type ExternalOptimizerCallbackLimits,\n  type ExternalOptimizerEvaluationObservation,\n  type ExternalTextCandidate,\n  type ExternalTextEvaluationRequest,\n  isExternalTextCandidate,\n  isRecord,\n  resolveExternalOptimizerCallbackLimits,\n} from './external-optimizer-contracts'\nimport { closeServer, listenLocal, sendJson } from './external-optimizer-http'\n\ntype UnsequencedObservation = ExternalOptimizerEvaluationObservation extends infer T\n  ? T extends ExternalOptimizerEvaluationObservation\n    ? Omit<T, 'sequence'>\n    : never\n  : never\n\nexport async function startExternalOptimizerCallback<TResponse>(args: {\n  token: string\n  maxEvaluations: number\n  acceptEvaluation?: () => number | undefined\n  evaluate: (request: ExternalTextEvaluationRequest, signal: AbortSignal) => Promise<TResponse>\n  observe?: (observation: ExternalOptimizerEvaluationObservation) => void\n  /** Loopback JSON byte limits. Omitted fields use finite defaults. */\n  limits?: Partial<ExternalOptimizerCallbackLimits>\n  signal?: AbortSignal\n}): Promise<ExternalOptimizerCallback> {\n  assertCallbackConfig(args)\n  const limits = resolveExternalOptimizerCallbackLimits(args.limits)\n  args.signal?.throwIfAborted()\n  let evaluations = 0\n  let accepting = true\n  let closePromise: Promise<void> | undefined\n  const activeControllers = new Set<AbortController>()\n  const activeHandlers = new Set<Promise<void>>()\n  const proposedCandidateHashes = new Set<string>()\n  let observationSequence = 0\n  const server = createServer((request, response) => {\n    if (!accepting) {\n      sendJsonIfOpen(response, 503, { error: 'external optimizer callback is closing' })\n      return\n    }\n\n    const controller = new AbortController()\n    const abortRequest = (): void => {\n      request.destroy()\n      response.destroy()\n    }\n    activeControllers.add(controller)\n    controller.signal.addEventListener('abort', abortRequest, { once: true })\n\n    let handler!: Promise<void>\n    handler = handleCallback(\n      request,\n      response,\n      controller.signal,\n      args,\n      limits,\n      () => {\n        const accepted = args.acceptEvaluation ? args.acceptEvaluation() : evaluations + 1\n        if (accepted === undefined) return undefined\n        if (!Number.isSafeInteger(accepted) || accepted <= 0) {\n          throw new Error('external optimizer callback: invalid accepted evaluation count')\n        }\n        if (accepted > args.maxEvaluations) return undefined\n        evaluations += 1\n        return accepted\n      },\n      (observation) => {\n        if (!args.observe) return\n        args.observe({ ...observation, sequence: ++observationSequence })\n      },\n      proposedCandidateHashes,\n    ).finally(() => {\n      controller.signal.removeEventListener('abort', abortRequest)\n      activeControllers.delete(controller)\n      activeHandlers.delete(handler)\n    })\n    activeHandlers.add(handler)\n    void handler.catch(() => undefined)\n  })\n  const port = await listenLocal(server)\n  const close = (): Promise<void> => {\n    closePromise ??= closeCallbackServer()\n    return closePromise\n  }\n  const onAbort = (): void => {\n    void close().catch(() => undefined)\n  }\n  args.signal?.addEventListener('abort', onAbort, { once: true })\n  if (args.signal?.aborted) onAbort()\n  return {\n    url: `http://127.0.0.1:${port}/evaluate`,\n    token: args.token,\n    evaluations: () => evaluations,\n    close,\n  }\n\n  async function closeCallbackServer(): Promise<void> {\n    args.signal?.removeEventListener('abort', onAbort)\n    accepting = false\n    const closingServer = closeServer(server)\n    server.closeIdleConnections?.()\n    for (const controller of activeControllers) controller.abort()\n    const [serverResult] = await Promise.allSettled([\n      closingServer,\n      waitForActiveHandlers(activeHandlers),\n    ])\n    if (activeControllers.size !== 0 || activeHandlers.size !== 0) {\n      throw new Error('external optimizer callback closed with active request work')\n    }\n    if (serverResult?.status === 'rejected') throw serverResult.reason\n  }\n}\n\nasync function handleCallback<TResponse>(\n  request: IncomingMessage,\n  response: ServerResponse,\n  signal: AbortSignal,\n  args: {\n    token: string\n    maxEvaluations: number\n    acceptEvaluation?: () => number | undefined\n    evaluate: (request: ExternalTextEvaluationRequest, signal: AbortSignal) => Promise<TResponse>\n    observe?: (observation: ExternalOptimizerEvaluationObservation) => void\n  },\n  limits: ExternalOptimizerCallbackLimits,\n  nextEvaluation: () => number | undefined,\n  observe: (observation: UnsequencedObservation) => void,\n  proposedCandidateHashes: Set<string>,\n): Promise<void> {\n  try {\n    if (request.method !== 'POST' || request.url !== '/evaluate') {\n      sendJsonIfOpen(response, 404, { error: 'not found' })\n      return\n    }\n    if (request.headers.authorization !== `Bearer ${args.token}`) {\n      sendJsonIfOpen(response, 401, { error: 'unauthorized' })\n      return\n    }\n    let body: unknown\n    try {\n      body = await readJson(request, limits.maxRequestBytes)\n    } catch {\n      observe({ kind: 'refusal', reason: 'invalid-request' })\n      sendJsonIfOpen(response, 400, { error: 'request body must be bounded valid JSON' })\n      return\n    }\n    if (\n      !isRecord(body) ||\n      !isExternalTextCandidate(body.candidate) ||\n      typeof body.exampleId !== 'string'\n    ) {\n      observe({ kind: 'refusal', reason: 'invalid-request' })\n      sendJsonIfOpen(response, 400, { error: 'candidate and exampleId are required strings' })\n      return\n    }\n    const candidate = cloneCandidate(body.candidate)\n    const candidateHash = contentHash({ kind: 'external-text-candidate', candidate })\n    if (!proposedCandidateHashes.has(candidateHash)) {\n      observe({ kind: 'proposal', candidate, candidateHash })\n      proposedCandidateHashes.add(candidateHash)\n    }\n    const count = nextEvaluation()\n    if (count === undefined) {\n      observe({\n        kind: 'refusal',\n        reason: 'evaluation-limit',\n        candidate,\n        candidateHash,\n        exampleId: body.exampleId,\n      })\n      sendJsonIfOpen(response, 429, { error: 'evaluation limit reached' })\n      return\n    }\n    let result: TResponse\n    try {\n      result = await args.evaluate({ candidate, exampleId: body.exampleId }, signal)\n    } catch (error) {\n      observe({\n        kind: 'refusal',\n        reason: 'evaluation-failed',\n        candidate,\n        candidateHash,\n        exampleId: body.exampleId,\n      })\n      // The thrown detail is the only diagnostic for a failed evaluation —\n      // an opaque 500 costs the caller a blind debugging round.\n      sendJsonIfOpen(response, 500, {\n        error: `evaluation failed: ${String(error).slice(0, 400)}`,\n      })\n      return\n    }\n    let encoded: string\n    try {\n      encoded = encodeJsonResponse(result)\n    } catch {\n      observe({\n        kind: 'refusal',\n        reason: 'evaluation-failed',\n        candidate,\n        candidateHash,\n        exampleId: body.exampleId,\n      })\n      sendJsonIfOpen(response, 500, { error: 'evaluation response must be valid JSON' })\n      return\n    }\n    if (Buffer.byteLength(encoded) > limits.maxResponseBytes) {\n      observe({\n        kind: 'refusal',\n        reason: 'evaluation-failed',\n        candidate,\n        candidateHash,\n        exampleId: body.exampleId,\n      })\n      sendJsonIfOpen(response, 413, { error: 'evaluation response too large' })\n      return\n    }\n    observe({\n      kind: 'evaluation',\n      candidate,\n      candidateHash,\n      exampleId: body.exampleId,\n      evaluationNumber: count,\n      response: structuredClone(result),\n    })\n    sendEncodedJsonIfOpen(response, 200, encoded)\n  } catch {\n    sendJsonIfOpen(response, 500, { error: 'evaluation failed' })\n  }\n}\n\nasync function waitForActiveHandlers(activeHandlers: Set<Promise<void>>): Promise<void> {\n  while (activeHandlers.size > 0) {\n    await Promise.allSettled([...activeHandlers])\n  }\n}\n\nfunction sendJsonIfOpen(response: ServerResponse, status: number, body: unknown): void {\n  if (response.destroyed || response.writableEnded) return\n  sendJson(response, status, body)\n}\n\nfunction readJson(request: IncomingMessage, maxRequestBytes: number): Promise<unknown> {\n  return new Promise((resolvePromise, reject) => {\n    let size = 0\n    const chunks: Buffer[] = []\n    request.on('data', (chunk: Buffer) => {\n      size += chunk.length\n      if (size > maxRequestBytes) {\n        reject(new Error('callback body too large'))\n        request.destroy()\n        return\n      }\n      chunks.push(chunk)\n    })\n    request.on('error', reject)\n    request.on('end', () => {\n      try {\n        resolvePromise(JSON.parse(Buffer.concat(chunks).toString('utf8')))\n      } catch (error) {\n        reject(error)\n      }\n    })\n  })\n}\n\nfunction encodeJsonResponse(value: unknown): string {\n  const encoded = JSON.stringify(value)\n  if (encoded === undefined) throw new Error('evaluation response must be JSON-serializable')\n  return encoded\n}\n\nfunction sendEncodedJsonIfOpen(response: ServerResponse, status: number, encoded: string): void {\n  if (response.destroyed || response.writableEnded) return\n  response.writeHead(status, {\n    'content-type': 'application/json; charset=utf-8',\n    'content-length': String(Buffer.byteLength(encoded)),\n  })\n  response.end(encoded)\n}\n\nfunction assertCallbackConfig(args: {\n  token: string\n  maxEvaluations: number\n  acceptEvaluation?: () => number | undefined\n  evaluate: (request: ExternalTextEvaluationRequest, signal: AbortSignal) => Promise<unknown>\n  observe?: (observation: ExternalOptimizerEvaluationObservation) => void\n  limits?: Partial<ExternalOptimizerCallbackLimits>\n  signal?: AbortSignal\n}): void {\n  if (typeof args.token !== 'string' || !args.token.trim()) {\n    throw new Error('external optimizer callback: token must be non-empty')\n  }\n  if (!Number.isSafeInteger(args.maxEvaluations) || args.maxEvaluations <= 0) {\n    throw new Error('external optimizer callback: maxEvaluations must be a positive safe integer')\n  }\n  if (args.acceptEvaluation !== undefined && typeof args.acceptEvaluation !== 'function') {\n    throw new Error('external optimizer callback: acceptEvaluation must be a function')\n  }\n  if (typeof args.evaluate !== 'function') {\n    throw new Error('external optimizer callback: evaluate must be a function')\n  }\n  if (args.observe !== undefined && typeof args.observe !== 'function') {\n    throw new Error('external optimizer callback: observe must be a function')\n  }\n}\n\nfunction cloneCandidate(candidate: ExternalTextCandidate): ExternalTextCandidate {\n  return typeof candidate === 'string' ? candidate : { ...candidate }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAsCA,SAAS,gBAAgB,MAAoB;CAC3C,MAAM,cAAc,oBAAoB,EAAE,UAAU,KAAK,CAAC;CAC1D,IAAI,aAAa,MAAM,iBAAiB,MAAM,WAAW;AAC3D;AAEA,SAAS,iBAAiB,MAAc,aAA+C;CACrF,IAAI,YAAY,WAAW,YACzB,uBAAO,IAAI,MACT,oDAAoD,KAAK,6BAC3D;CAEF,uBAAO,IAAI,MACT,oCAAoC,YAAY,OAAO,IAAI,IAAI,KAAK,oDACtE;AACF;;;;;;AAOA,SAAgB,qBAAqB,MAA2C;CAC9E,MAAM,MAAM,KAAK,OAAO,QAAQ;CAChC,IAAI,CAAC,OAAO,cAAc,GAAG,KAAK,OAAO,GACvC,MAAM,IAAI,MAAM,gDAAgD;CAClE,KAAK,MAAM,QAAQ,KAAK,aAAa,CAAC,GAAG,gBAAgB,IAAI;CAC7D,MAAM,cAAc,yBAAyB;EAC3C,UAAU,KAAK;EACf;CACF,CAAC;CACD,IAAI,CAAC,YAAY,UAAU,MAAM,iBAAiB,KAAK,UAAU,WAAW;CAC5E,MAAM,gBAAgB,KAAK,iBAAiB;CAC5C,IAAI,WAAW;CACf,MAAM,gBAAsB;EAC1B,IAAI,UAAU;EACd,WAAW;EACX,IAAI,eAAe,QAAQ,IAAI,QAAQ,OAAO;EAC9C,IAAI;GACF,YAAY,KAAK,QAAQ;EAC3B,QAAQ,CAER;CACF;CACA,IAAI;EACF,KAAK,MAAM,QAAQ,KAAK,aAAa,CAAC,GAAG,gBAAgB,IAAI;CAC/D,SAAS,OAAO;EACd,QAAQ;EACR,MAAM;CACR;CACA,IAAI,eAAe,QAAQ,GAAG,QAAQ,OAAO;CAC7C,OAAO,EAAE,QAAQ;AACnB;;;ACrEA,eAAsB,+BAA0C,MASzB;CACrC,qBAAqB,IAAI;CACzB,MAAM,SAAS,uCAAuC,KAAK,MAAM;CACjE,KAAK,QAAQ,eAAe;CAC5B,IAAI,cAAc;CAClB,IAAI,YAAY;CAChB,IAAI;CACJ,MAAM,oCAAoB,IAAI,IAAqB;CACnD,MAAM,iCAAiB,IAAI,IAAmB;CAC9C,MAAM,0CAA0B,IAAI,IAAY;CAChD,IAAI,sBAAsB;CAC1B,MAAM,SAAS,cAAc,SAAS,aAAa;EACjD,IAAI,CAAC,WAAW;GACd,eAAe,UAAU,KAAK,EAAE,OAAO,yCAAyC,CAAC;GACjF;EACF;EAEA,MAAM,aAAa,IAAI,gBAAgB;EACvC,MAAM,qBAA2B;GAC/B,QAAQ,QAAQ;GAChB,SAAS,QAAQ;EACnB;EACA,kBAAkB,IAAI,UAAU;EAChC,WAAW,OAAO,iBAAiB,SAAS,cAAc,EAAE,MAAM,KAAK,CAAC;EAExE,IAAI;EACJ,UAAU,eACR,SACA,UACA,WAAW,QACX,MACA,cACM;GACJ,MAAM,WAAW,KAAK,mBAAmB,KAAK,iBAAiB,IAAI,cAAc;GACjF,IAAI,aAAa,KAAA,GAAW,OAAO,KAAA;GACnC,IAAI,CAAC,OAAO,cAAc,QAAQ,KAAK,YAAY,GACjD,MAAM,IAAI,MAAM,gEAAgE;GAElF,IAAI,WAAW,KAAK,gBAAgB,OAAO,KAAA;GAC3C,eAAe;GACf,OAAO;EACT,IACC,gBAAgB;GACf,IAAI,CAAC,KAAK,SAAS;GACnB,KAAK,QAAQ;IAAE,GAAG;IAAa,UAAU,EAAE;GAAoB,CAAC;EAClE,GACA,uBACF,CAAC,CAAC,cAAc;GACd,WAAW,OAAO,oBAAoB,SAAS,YAAY;GAC3D,kBAAkB,OAAO,UAAU;GACnC,eAAe,OAAO,OAAO;EAC/B,CAAC;EACD,eAAe,IAAI,OAAO;EAC1B,QAAa,YAAY,KAAA,CAAS;CACpC,CAAC;CACD,MAAM,OAAO,MAAM,YAAY,MAAM;CACrC,MAAM,cAA6B;EACjC,iBAAiB,oBAAoB;EACrC,OAAO;CACT;CACA,MAAM,gBAAsB;EAC1B,MAAW,CAAC,CAAC,YAAY,KAAA,CAAS;CACpC;CACA,KAAK,QAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM,KAAK,CAAC;CAC9D,IAAI,KAAK,QAAQ,SAAS,QAAQ;CAClC,OAAO;EACL,KAAK,oBAAoB,KAAK;EAC9B,OAAO,KAAK;EACZ,mBAAmB;EACnB;CACF;CAEA,eAAe,sBAAqC;EAClD,KAAK,QAAQ,oBAAoB,SAAS,OAAO;EACjD,YAAY;EACZ,MAAM,gBAAgB,YAAY,MAAM;EACxC,OAAO,uBAAuB;EAC9B,KAAK,MAAM,cAAc,mBAAmB,WAAW,MAAM;EAC7D,MAAM,CAAC,gBAAgB,MAAM,QAAQ,WAAW,CAC9C,eACA,sBAAsB,cAAc,CACtC,CAAC;EACD,IAAI,kBAAkB,SAAS,KAAK,eAAe,SAAS,GAC1D,MAAM,IAAI,MAAM,6DAA6D;EAE/E,IAAI,cAAc,WAAW,YAAY,MAAM,aAAa;CAC9D;AACF;AAEA,eAAe,eACb,SACA,UACA,QACA,MAOA,QACA,gBACA,SACA,yBACe;CACf,IAAI;EACF,IAAI,QAAQ,WAAW,UAAU,QAAQ,QAAQ,aAAa;GAC5D,eAAe,UAAU,KAAK,EAAE,OAAO,YAAY,CAAC;GACpD;EACF;EACA,IAAI,QAAQ,QAAQ,kBAAkB,UAAU,KAAK,SAAS;GAC5D,eAAe,UAAU,KAAK,EAAE,OAAO,eAAe,CAAC;GACvD;EACF;EACA,IAAI;EACJ,IAAI;GACF,OAAO,MAAM,SAAS,SAAS,OAAO,eAAe;EACvD,QAAQ;GACN,QAAQ;IAAE,MAAM;IAAW,QAAQ;GAAkB,CAAC;GACtD,eAAe,UAAU,KAAK,EAAE,OAAO,0CAA0C,CAAC;GAClF;EACF;EACA,IACE,CAAC,SAAS,IAAI,KACd,CAAC,wBAAwB,KAAK,SAAS,KACvC,OAAO,KAAK,cAAc,UAC1B;GACA,QAAQ;IAAE,MAAM;IAAW,QAAQ;GAAkB,CAAC;GACtD,eAAe,UAAU,KAAK,EAAE,OAAO,+CAA+C,CAAC;GACvF;EACF;EACA,MAAM,YAAY,eAAe,KAAK,SAAS;EAC/C,MAAM,gBAAgB,YAAY;GAAE,MAAM;GAA2B;EAAU,CAAC;EAChF,IAAI,CAAC,wBAAwB,IAAI,aAAa,GAAG;GAC/C,QAAQ;IAAE,MAAM;IAAY;IAAW;GAAc,CAAC;GACtD,wBAAwB,IAAI,aAAa;EAC3C;EACA,MAAM,QAAQ,eAAe;EAC7B,IAAI,UAAU,KAAA,GAAW;GACvB,QAAQ;IACN,MAAM;IACN,QAAQ;IACR;IACA;IACA,WAAW,KAAK;GAClB,CAAC;GACD,eAAe,UAAU,KAAK,EAAE,OAAO,2BAA2B,CAAC;GACnE;EACF;EACA,IAAI;EACJ,IAAI;GACF,SAAS,MAAM,KAAK,SAAS;IAAE;IAAW,WAAW,KAAK;GAAU,GAAG,MAAM;EAC/E,SAAS,OAAO;GACd,QAAQ;IACN,MAAM;IACN,QAAQ;IACR;IACA;IACA,WAAW,KAAK;GAClB,CAAC;GAGD,eAAe,UAAU,KAAK,EAC5B,OAAO,sBAAsB,OAAO,KAAK,CAAC,CAAC,MAAM,GAAG,GAAG,IACzD,CAAC;GACD;EACF;EACA,IAAI;EACJ,IAAI;GACF,UAAU,mBAAmB,MAAM;EACrC,QAAQ;GACN,QAAQ;IACN,MAAM;IACN,QAAQ;IACR;IACA;IACA,WAAW,KAAK;GAClB,CAAC;GACD,eAAe,UAAU,KAAK,EAAE,OAAO,yCAAyC,CAAC;GACjF;EACF;EACA,IAAI,OAAO,WAAW,OAAO,IAAI,OAAO,kBAAkB;GACxD,QAAQ;IACN,MAAM;IACN,QAAQ;IACR;IACA;IACA,WAAW,KAAK;GAClB,CAAC;GACD,eAAe,UAAU,KAAK,EAAE,OAAO,gCAAgC,CAAC;GACxE;EACF;EACA,QAAQ;GACN,MAAM;GACN;GACA;GACA,WAAW,KAAK;GAChB,kBAAkB;GAClB,UAAU,gBAAgB,MAAM;EAClC,CAAC;EACD,sBAAsB,UAAU,KAAK,OAAO;CAC9C,QAAQ;EACN,eAAe,UAAU,KAAK,EAAE,OAAO,oBAAoB,CAAC;CAC9D;AACF;AAEA,eAAe,sBAAsB,gBAAmD;CACtF,OAAO,eAAe,OAAO,GAC3B,MAAM,QAAQ,WAAW,CAAC,GAAG,cAAc,CAAC;AAEhD;AAEA,SAAS,eAAe,UAA0B,QAAgB,MAAqB;CACrF,IAAI,SAAS,aAAa,SAAS,eAAe;CAClD,SAAS,UAAU,QAAQ,IAAI;AACjC;AAEA,SAAS,SAAS,SAA0B,iBAA2C;CACrF,OAAO,IAAI,SAAS,gBAAgB,WAAW;EAC7C,IAAI,OAAO;EACX,MAAM,SAAmB,CAAC;EAC1B,QAAQ,GAAG,SAAS,UAAkB;GACpC,QAAQ,MAAM;GACd,IAAI,OAAO,iBAAiB;IAC1B,uBAAO,IAAI,MAAM,yBAAyB,CAAC;IAC3C,QAAQ,QAAQ;IAChB;GACF;GACA,OAAO,KAAK,KAAK;EACnB,CAAC;EACD,QAAQ,GAAG,SAAS,MAAM;EAC1B,QAAQ,GAAG,aAAa;GACtB,IAAI;IACF,eAAe,KAAK,MAAM,OAAO,OAAO,MAAM,CAAC,CAAC,SAAS,MAAM,CAAC,CAAC;GACnE,SAAS,OAAO;IACd,OAAO,KAAK;GACd;EACF,CAAC;CACH,CAAC;AACH;AAEA,SAAS,mBAAmB,OAAwB;CAClD,MAAM,UAAU,KAAK,UAAU,KAAK;CACpC,IAAI,YAAY,KAAA,GAAW,MAAM,IAAI,MAAM,+CAA+C;CAC1F,OAAO;AACT;AAEA,SAAS,sBAAsB,UAA0B,QAAgB,SAAuB;CAC9F,IAAI,SAAS,aAAa,SAAS,eAAe;CAClD,SAAS,UAAU,QAAQ;EACzB,gBAAgB;EAChB,kBAAkB,OAAO,OAAO,WAAW,OAAO,CAAC;CACrD,CAAC;CACD,SAAS,IAAI,OAAO;AACtB;AAEA,SAAS,qBAAqB,MAQrB;CACP,IAAI,OAAO,KAAK,UAAU,YAAY,CAAC,KAAK,MAAM,KAAK,GACrD,MAAM,IAAI,MAAM,sDAAsD;CAExE,IAAI,CAAC,OAAO,cAAc,KAAK,cAAc,KAAK,KAAK,kBAAkB,GACvE,MAAM,IAAI,MAAM,6EAA6E;CAE/F,IAAI,KAAK,qBAAqB,KAAA,KAAa,OAAO,KAAK,qBAAqB,YAC1E,MAAM,IAAI,MAAM,kEAAkE;CAEpF,IAAI,OAAO,KAAK,aAAa,YAC3B,MAAM,IAAI,MAAM,0DAA0D;CAE5E,IAAI,KAAK,YAAY,KAAA,KAAa,OAAO,KAAK,YAAY,YACxD,MAAM,IAAI,MAAM,yDAAyD;AAE7E;AAEA,SAAS,eAAe,WAAyD;CAC/E,OAAO,OAAO,cAAc,WAAW,YAAY,EAAE,GAAG,UAAU;AACpE"}
+{"version":3,"file":"external-optimizer-process-WosTBChy.js","names":[],"sources":["../src/campaign/single-run-lock.ts","../src/campaign/external-optimizer-callback.ts"],"sourcesContent":["/**\n * Single-run lock for evaluations that share one mutable environment.\n *\n * Two concurrent runs against a shared stateful gym silently corrupt each\n * other: each resets/mutates environment state mid-cell of the other, and\n * every score from both becomes garbage that LOOKS like worker variance\n * (agent-lab R357 burned hours on flip-flopping scores before tracing them\n * to exactly this). The fix is a pid lockfile: refuse to start while a live\n * holder exists, reclaim stale locks whose pid is gone, release only if the\n * lock is still ours.\n *\n * `alsoCheck` exists because independent runners can guard the same shared\n * resource with differently named lockfiles; a runner must respect all of\n * them even though it writes only its own.\n */\n\nimport {\n  type AtomicFileLockUnavailable,\n  probeAtomicFileLock,\n  tryAcquireAtomicFileLock,\n} from '../ledger-core/atomic-file-lock'\n\nexport interface SingleRunLockOptions {\n  /** Lockfile this runner writes (and checks). */\n  readonly lockPath: string\n  /** Other runners' lockfiles guarding the same resource; checked, never written. */\n  readonly alsoCheck?: readonly string[]\n  /** Install a process 'exit' hook that releases the lock. Default true. */\n  readonly releaseOnExit?: boolean\n  /** Owner pid recorded in the lockfile metadata. Default process.pid. */\n  readonly pid?: number\n}\n\nexport interface SingleRunLock {\n  /** Remove the lockfile if this process still owns it. Idempotent. */\n  release(): void\n}\n\nfunction assertAvailable(path: string): void {\n  const unavailable = probeAtomicFileLock({ lockPath: path })\n  if (unavailable) throw unavailableError(path, unavailable)\n}\n\nfunction unavailableError(path: string, unavailable: AtomicFileLockUnavailable): Error {\n  if (unavailable.reason === 'recovery') {\n    return new Error(\n      `single-run lock recovery is already in progress (${path}); refusing a concurrent run`,\n    )\n  }\n  return new Error(\n    `single-run lock held by live pid ${unavailable.holder.pid} (${path}); refusing a concurrent run on the shared resource`,\n  )\n}\n\n/**\n * Acquire the lock or throw naming the live holder. A stale lock (holder pid\n * no longer running) is reclaimed by one contender. An interrupted reclaim\n * leaves a marker that fails closed instead of admitting overlapping runs.\n */\nexport function acquireSingleRunLock(opts: SingleRunLockOptions): SingleRunLock {\n  const pid = opts.pid ?? process.pid\n  if (!Number.isSafeInteger(pid) || pid <= 0)\n    throw new Error('single-run lock pid must be a positive integer')\n  for (const path of opts.alsoCheck ?? []) assertAvailable(path)\n  const acquisition = tryAcquireAtomicFileLock({\n    lockPath: opts.lockPath,\n    pid,\n  })\n  if (!acquisition.acquired) throw unavailableError(opts.lockPath, acquisition)\n  const releaseOnExit = opts.releaseOnExit ?? true\n  let released = false\n  const release = (): void => {\n    if (released) return\n    released = true\n    if (releaseOnExit) process.off('exit', release)\n    try {\n      acquisition.lock.release()\n    } catch {\n      // release is best-effort; a leftover stale lock is reclaimed on next acquire\n    }\n  }\n  try {\n    for (const path of opts.alsoCheck ?? []) assertAvailable(path)\n  } catch (error) {\n    release()\n    throw error\n  }\n  if (releaseOnExit) process.on('exit', release)\n  return { release }\n}\n","import { createServer, type IncomingMessage, type ServerResponse } from 'node:http'\nimport { contentHash } from '../verdict-cache'\nimport {\n  type ExternalOptimizerCallback,\n  type ExternalOptimizerCallbackLimits,\n  type ExternalOptimizerEvaluationObservation,\n  type ExternalTextCandidate,\n  type ExternalTextEvaluationRequest,\n  isExternalTextCandidate,\n  isRecord,\n  resolveExternalOptimizerCallbackLimits,\n} from './external-optimizer-contracts'\nimport { closeServer, listenLocal, sendJson } from './external-optimizer-http'\n\ntype UnsequencedObservation = ExternalOptimizerEvaluationObservation extends infer T\n  ? T extends ExternalOptimizerEvaluationObservation\n    ? Omit<T, 'sequence'>\n    : never\n  : never\n\nexport async function startExternalOptimizerCallback<TResponse>(args: {\n  token: string\n  maxEvaluations: number\n  acceptEvaluation?: () => number | undefined\n  evaluate: (request: ExternalTextEvaluationRequest, signal: AbortSignal) => Promise<TResponse>\n  observe?: (observation: ExternalOptimizerEvaluationObservation) => void\n  /** Loopback JSON byte limits. Omitted fields use finite defaults. */\n  limits?: Partial<ExternalOptimizerCallbackLimits>\n  signal?: AbortSignal\n}): Promise<ExternalOptimizerCallback> {\n  assertCallbackConfig(args)\n  const limits = resolveExternalOptimizerCallbackLimits(args.limits)\n  args.signal?.throwIfAborted()\n  let evaluations = 0\n  let accepting = true\n  let closePromise: Promise<void> | undefined\n  const activeControllers = new Set<AbortController>()\n  const activeHandlers = new Set<Promise<void>>()\n  const proposedCandidateHashes = new Set<string>()\n  let observationSequence = 0\n  const server = createServer((request, response) => {\n    if (!accepting) {\n      sendJsonIfOpen(response, 503, { error: 'external optimizer callback is closing' })\n      return\n    }\n\n    const controller = new AbortController()\n    const abortRequest = (): void => {\n      request.destroy()\n      response.destroy()\n    }\n    activeControllers.add(controller)\n    controller.signal.addEventListener('abort', abortRequest, { once: true })\n\n    let handler!: Promise<void>\n    handler = handleCallback(\n      request,\n      response,\n      controller.signal,\n      args,\n      limits,\n      () => {\n        const accepted = args.acceptEvaluation ? args.acceptEvaluation() : evaluations + 1\n        if (accepted === undefined) return undefined\n        if (!Number.isSafeInteger(accepted) || accepted <= 0) {\n          throw new Error('external optimizer callback: invalid accepted evaluation count')\n        }\n        if (accepted > args.maxEvaluations) return undefined\n        evaluations += 1\n        return accepted\n      },\n      (observation) => {\n        if (!args.observe) return\n        args.observe({ ...observation, sequence: ++observationSequence })\n      },\n      proposedCandidateHashes,\n    ).finally(() => {\n      controller.signal.removeEventListener('abort', abortRequest)\n      activeControllers.delete(controller)\n      activeHandlers.delete(handler)\n    })\n    activeHandlers.add(handler)\n    void handler.catch(() => undefined)\n  })\n  const port = await listenLocal(server)\n  const close = (): Promise<void> => {\n    closePromise ??= closeCallbackServer()\n    return closePromise\n  }\n  const onAbort = (): void => {\n    void close().catch(() => undefined)\n  }\n  args.signal?.addEventListener('abort', onAbort, { once: true })\n  if (args.signal?.aborted) onAbort()\n  return {\n    url: `http://127.0.0.1:${port}/evaluate`,\n    token: args.token,\n    evaluations: () => evaluations,\n    close,\n  }\n\n  async function closeCallbackServer(): Promise<void> {\n    args.signal?.removeEventListener('abort', onAbort)\n    accepting = false\n    const closingServer = closeServer(server)\n    server.closeIdleConnections?.()\n    for (const controller of activeControllers) controller.abort()\n    const [serverResult] = await Promise.allSettled([\n      closingServer,\n      waitForActiveHandlers(activeHandlers),\n    ])\n    if (activeControllers.size !== 0 || activeHandlers.size !== 0) {\n      throw new Error('external optimizer callback closed with active request work')\n    }\n    if (serverResult?.status === 'rejected') throw serverResult.reason\n  }\n}\n\nasync function handleCallback<TResponse>(\n  request: IncomingMessage,\n  response: ServerResponse,\n  signal: AbortSignal,\n  args: {\n    token: string\n    maxEvaluations: number\n    acceptEvaluation?: () => number | undefined\n    evaluate: (request: ExternalTextEvaluationRequest, signal: AbortSignal) => Promise<TResponse>\n    observe?: (observation: ExternalOptimizerEvaluationObservation) => void\n  },\n  limits: ExternalOptimizerCallbackLimits,\n  nextEvaluation: () => number | undefined,\n  observe: (observation: UnsequencedObservation) => void,\n  proposedCandidateHashes: Set<string>,\n): Promise<void> {\n  try {\n    if (request.method !== 'POST' || request.url !== '/evaluate') {\n      sendJsonIfOpen(response, 404, { error: 'not found' })\n      return\n    }\n    if (request.headers.authorization !== `Bearer ${args.token}`) {\n      sendJsonIfOpen(response, 401, { error: 'unauthorized' })\n      return\n    }\n    let body: unknown\n    try {\n      body = await readJson(request, limits.maxRequestBytes)\n    } catch {\n      observe({ kind: 'refusal', reason: 'invalid-request' })\n      sendJsonIfOpen(response, 400, { error: 'request body must be bounded valid JSON' })\n      return\n    }\n    if (\n      !isRecord(body) ||\n      !isExternalTextCandidate(body.candidate) ||\n      typeof body.exampleId !== 'string'\n    ) {\n      observe({ kind: 'refusal', reason: 'invalid-request' })\n      sendJsonIfOpen(response, 400, { error: 'candidate and exampleId are required strings' })\n      return\n    }\n    const candidate = cloneCandidate(body.candidate)\n    const candidateHash = contentHash({ kind: 'external-text-candidate', candidate })\n    if (!proposedCandidateHashes.has(candidateHash)) {\n      observe({ kind: 'proposal', candidate, candidateHash })\n      proposedCandidateHashes.add(candidateHash)\n    }\n    const count = nextEvaluation()\n    if (count === undefined) {\n      observe({\n        kind: 'refusal',\n        reason: 'evaluation-limit',\n        candidate,\n        candidateHash,\n        exampleId: body.exampleId,\n      })\n      sendJsonIfOpen(response, 429, { error: 'evaluation limit reached' })\n      return\n    }\n    let result: TResponse\n    try {\n      result = await args.evaluate({ candidate, exampleId: body.exampleId }, signal)\n    } catch (error) {\n      observe({\n        kind: 'refusal',\n        reason: 'evaluation-failed',\n        candidate,\n        candidateHash,\n        exampleId: body.exampleId,\n      })\n      // The thrown detail is the only diagnostic for a failed evaluation —\n      // an opaque 500 costs the caller a blind debugging round.\n      sendJsonIfOpen(response, 500, {\n        error: `evaluation failed: ${String(error).slice(0, 400)}`,\n      })\n      return\n    }\n    let encoded: string\n    try {\n      encoded = encodeJsonResponse(result)\n    } catch {\n      observe({\n        kind: 'refusal',\n        reason: 'evaluation-failed',\n        candidate,\n        candidateHash,\n        exampleId: body.exampleId,\n      })\n      sendJsonIfOpen(response, 500, { error: 'evaluation response must be valid JSON' })\n      return\n    }\n    if (Buffer.byteLength(encoded) > limits.maxResponseBytes) {\n      observe({\n        kind: 'refusal',\n        reason: 'evaluation-failed',\n        candidate,\n        candidateHash,\n        exampleId: body.exampleId,\n      })\n      sendJsonIfOpen(response, 413, { error: 'evaluation response too large' })\n      return\n    }\n    observe({\n      kind: 'evaluation',\n      candidate,\n      candidateHash,\n      exampleId: body.exampleId,\n      evaluationNumber: count,\n      response: structuredClone(result),\n    })\n    sendEncodedJsonIfOpen(response, 200, encoded)\n  } catch {\n    sendJsonIfOpen(response, 500, { error: 'evaluation failed' })\n  }\n}\n\nasync function waitForActiveHandlers(activeHandlers: Set<Promise<void>>): Promise<void> {\n  while (activeHandlers.size > 0) {\n    await Promise.allSettled([...activeHandlers])\n  }\n}\n\nfunction sendJsonIfOpen(response: ServerResponse, status: number, body: unknown): void {\n  if (response.destroyed || response.writableEnded) return\n  sendJson(response, status, body)\n}\n\nfunction readJson(request: IncomingMessage, maxRequestBytes: number): Promise<unknown> {\n  return new Promise((resolvePromise, reject) => {\n    let size = 0\n    const chunks: Buffer[] = []\n    request.on('data', (chunk: Buffer) => {\n      size += chunk.length\n      if (size > maxRequestBytes) {\n        reject(new Error('callback body too large'))\n        request.destroy()\n        return\n      }\n      chunks.push(chunk)\n    })\n    request.on('error', reject)\n    request.on('end', () => {\n      try {\n        resolvePromise(JSON.parse(Buffer.concat(chunks).toString('utf8')))\n      } catch (error) {\n        reject(error)\n      }\n    })\n  })\n}\n\nfunction encodeJsonResponse(value: unknown): string {\n  const encoded = JSON.stringify(value)\n  if (encoded === undefined) throw new Error('evaluation response must be JSON-serializable')\n  return encoded\n}\n\nfunction sendEncodedJsonIfOpen(response: ServerResponse, status: number, encoded: string): void {\n  if (response.destroyed || response.writableEnded) return\n  response.writeHead(status, {\n    'content-type': 'application/json; charset=utf-8',\n    'content-length': String(Buffer.byteLength(encoded)),\n  })\n  response.end(encoded)\n}\n\nfunction assertCallbackConfig(args: {\n  token: string\n  maxEvaluations: number\n  acceptEvaluation?: () => number | undefined\n  evaluate: (request: ExternalTextEvaluationRequest, signal: AbortSignal) => Promise<unknown>\n  observe?: (observation: ExternalOptimizerEvaluationObservation) => void\n  limits?: Partial<ExternalOptimizerCallbackLimits>\n  signal?: AbortSignal\n}): void {\n  if (typeof args.token !== 'string' || !args.token.trim()) {\n    throw new Error('external optimizer callback: token must be non-empty')\n  }\n  if (!Number.isSafeInteger(args.maxEvaluations) || args.maxEvaluations <= 0) {\n    throw new Error('external optimizer callback: maxEvaluations must be a positive safe integer')\n  }\n  if (args.acceptEvaluation !== undefined && typeof args.acceptEvaluation !== 'function') {\n    throw new Error('external optimizer callback: acceptEvaluation must be a function')\n  }\n  if (typeof args.evaluate !== 'function') {\n    throw new Error('external optimizer callback: evaluate must be a function')\n  }\n  if (args.observe !== undefined && typeof args.observe !== 'function') {\n    throw new Error('external optimizer callback: observe must be a function')\n  }\n}\n\nfunction cloneCandidate(candidate: ExternalTextCandidate): ExternalTextCandidate {\n  return typeof candidate === 'string' ? candidate : { ...candidate }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAsCA,SAAS,gBAAgB,MAAoB;CAC3C,MAAM,cAAc,oBAAoB,EAAE,UAAU,KAAK,CAAC;CAC1D,IAAI,aAAa,MAAM,iBAAiB,MAAM,WAAW;AAC3D;AAEA,SAAS,iBAAiB,MAAc,aAA+C;CACrF,IAAI,YAAY,WAAW,YACzB,uBAAO,IAAI,MACT,oDAAoD,KAAK,6BAC3D;CAEF,uBAAO,IAAI,MACT,oCAAoC,YAAY,OAAO,IAAI,IAAI,KAAK,oDACtE;AACF;;;;;;AAOA,SAAgB,qBAAqB,MAA2C;CAC9E,MAAM,MAAM,KAAK,OAAO,QAAQ;CAChC,IAAI,CAAC,OAAO,cAAc,GAAG,KAAK,OAAO,GACvC,MAAM,IAAI,MAAM,gDAAgD;CAClE,KAAK,MAAM,QAAQ,KAAK,aAAa,CAAC,GAAG,gBAAgB,IAAI;CAC7D,MAAM,cAAc,yBAAyB;EAC3C,UAAU,KAAK;EACf;CACF,CAAC;CACD,IAAI,CAAC,YAAY,UAAU,MAAM,iBAAiB,KAAK,UAAU,WAAW;CAC5E,MAAM,gBAAgB,KAAK,iBAAiB;CAC5C,IAAI,WAAW;CACf,MAAM,gBAAsB;EAC1B,IAAI,UAAU;EACd,WAAW;EACX,IAAI,eAAe,QAAQ,IAAI,QAAQ,OAAO;EAC9C,IAAI;GACF,YAAY,KAAK,QAAQ;EAC3B,QAAQ,CAER;CACF;CACA,IAAI;EACF,KAAK,MAAM,QAAQ,KAAK,aAAa,CAAC,GAAG,gBAAgB,IAAI;CAC/D,SAAS,OAAO;EACd,QAAQ;EACR,MAAM;CACR;CACA,IAAI,eAAe,QAAQ,GAAG,QAAQ,OAAO;CAC7C,OAAO,EAAE,QAAQ;AACnB;;;ACrEA,eAAsB,+BAA0C,MASzB;CACrC,qBAAqB,IAAI;CACzB,MAAM,SAAS,uCAAuC,KAAK,MAAM;CACjE,KAAK,QAAQ,eAAe;CAC5B,IAAI,cAAc;CAClB,IAAI,YAAY;CAChB,IAAI;CACJ,MAAM,oCAAoB,IAAI,IAAqB;CACnD,MAAM,iCAAiB,IAAI,IAAmB;CAC9C,MAAM,0CAA0B,IAAI,IAAY;CAChD,IAAI,sBAAsB;CAC1B,MAAM,SAAS,cAAc,SAAS,aAAa;EACjD,IAAI,CAAC,WAAW;GACd,eAAe,UAAU,KAAK,EAAE,OAAO,yCAAyC,CAAC;GACjF;EACF;EAEA,MAAM,aAAa,IAAI,gBAAgB;EACvC,MAAM,qBAA2B;GAC/B,QAAQ,QAAQ;GAChB,SAAS,QAAQ;EACnB;EACA,kBAAkB,IAAI,UAAU;EAChC,WAAW,OAAO,iBAAiB,SAAS,cAAc,EAAE,MAAM,KAAK,CAAC;EAExE,IAAI;EACJ,UAAU,eACR,SACA,UACA,WAAW,QACX,MACA,cACM;GACJ,MAAM,WAAW,KAAK,mBAAmB,KAAK,iBAAiB,IAAI,cAAc;GACjF,IAAI,aAAa,KAAA,GAAW,OAAO,KAAA;GACnC,IAAI,CAAC,OAAO,cAAc,QAAQ,KAAK,YAAY,GACjD,MAAM,IAAI,MAAM,gEAAgE;GAElF,IAAI,WAAW,KAAK,gBAAgB,OAAO,KAAA;GAC3C,eAAe;GACf,OAAO;EACT,IACC,gBAAgB;GACf,IAAI,CAAC,KAAK,SAAS;GACnB,KAAK,QAAQ;IAAE,GAAG;IAAa,UAAU,EAAE;GAAoB,CAAC;EAClE,GACA,uBACF,CAAC,CAAC,cAAc;GACd,WAAW,OAAO,oBAAoB,SAAS,YAAY;GAC3D,kBAAkB,OAAO,UAAU;GACnC,eAAe,OAAO,OAAO;EAC/B,CAAC;EACD,eAAe,IAAI,OAAO;EAC1B,QAAa,YAAY,KAAA,CAAS;CACpC,CAAC;CACD,MAAM,OAAO,MAAM,YAAY,MAAM;CACrC,MAAM,cAA6B;EACjC,iBAAiB,oBAAoB;EACrC,OAAO;CACT;CACA,MAAM,gBAAsB;EAC1B,MAAW,CAAC,CAAC,YAAY,KAAA,CAAS;CACpC;CACA,KAAK,QAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM,KAAK,CAAC;CAC9D,IAAI,KAAK,QAAQ,SAAS,QAAQ;CAClC,OAAO;EACL,KAAK,oBAAoB,KAAK;EAC9B,OAAO,KAAK;EACZ,mBAAmB;EACnB;CACF;CAEA,eAAe,sBAAqC;EAClD,KAAK,QAAQ,oBAAoB,SAAS,OAAO;EACjD,YAAY;EACZ,MAAM,gBAAgB,YAAY,MAAM;EACxC,OAAO,uBAAuB;EAC9B,KAAK,MAAM,cAAc,mBAAmB,WAAW,MAAM;EAC7D,MAAM,CAAC,gBAAgB,MAAM,QAAQ,WAAW,CAC9C,eACA,sBAAsB,cAAc,CACtC,CAAC;EACD,IAAI,kBAAkB,SAAS,KAAK,eAAe,SAAS,GAC1D,MAAM,IAAI,MAAM,6DAA6D;EAE/E,IAAI,cAAc,WAAW,YAAY,MAAM,aAAa;CAC9D;AACF;AAEA,eAAe,eACb,SACA,UACA,QACA,MAOA,QACA,gBACA,SACA,yBACe;CACf,IAAI;EACF,IAAI,QAAQ,WAAW,UAAU,QAAQ,QAAQ,aAAa;GAC5D,eAAe,UAAU,KAAK,EAAE,OAAO,YAAY,CAAC;GACpD;EACF;EACA,IAAI,QAAQ,QAAQ,kBAAkB,UAAU,KAAK,SAAS;GAC5D,eAAe,UAAU,KAAK,EAAE,OAAO,eAAe,CAAC;GACvD;EACF;EACA,IAAI;EACJ,IAAI;GACF,OAAO,MAAM,SAAS,SAAS,OAAO,eAAe;EACvD,QAAQ;GACN,QAAQ;IAAE,MAAM;IAAW,QAAQ;GAAkB,CAAC;GACtD,eAAe,UAAU,KAAK,EAAE,OAAO,0CAA0C,CAAC;GAClF;EACF;EACA,IACE,CAAC,SAAS,IAAI,KACd,CAAC,wBAAwB,KAAK,SAAS,KACvC,OAAO,KAAK,cAAc,UAC1B;GACA,QAAQ;IAAE,MAAM;IAAW,QAAQ;GAAkB,CAAC;GACtD,eAAe,UAAU,KAAK,EAAE,OAAO,+CAA+C,CAAC;GACvF;EACF;EACA,MAAM,YAAY,eAAe,KAAK,SAAS;EAC/C,MAAM,gBAAgB,YAAY;GAAE,MAAM;GAA2B;EAAU,CAAC;EAChF,IAAI,CAAC,wBAAwB,IAAI,aAAa,GAAG;GAC/C,QAAQ;IAAE,MAAM;IAAY;IAAW;GAAc,CAAC;GACtD,wBAAwB,IAAI,aAAa;EAC3C;EACA,MAAM,QAAQ,eAAe;EAC7B,IAAI,UAAU,KAAA,GAAW;GACvB,QAAQ;IACN,MAAM;IACN,QAAQ;IACR;IACA;IACA,WAAW,KAAK;GAClB,CAAC;GACD,eAAe,UAAU,KAAK,EAAE,OAAO,2BAA2B,CAAC;GACnE;EACF;EACA,IAAI;EACJ,IAAI;GACF,SAAS,MAAM,KAAK,SAAS;IAAE;IAAW,WAAW,KAAK;GAAU,GAAG,MAAM;EAC/E,SAAS,OAAO;GACd,QAAQ;IACN,MAAM;IACN,QAAQ;IACR;IACA;IACA,WAAW,KAAK;GAClB,CAAC;GAGD,eAAe,UAAU,KAAK,EAC5B,OAAO,sBAAsB,OAAO,KAAK,CAAC,CAAC,MAAM,GAAG,GAAG,IACzD,CAAC;GACD;EACF;EACA,IAAI;EACJ,IAAI;GACF,UAAU,mBAAmB,MAAM;EACrC,QAAQ;GACN,QAAQ;IACN,MAAM;IACN,QAAQ;IACR;IACA;IACA,WAAW,KAAK;GAClB,CAAC;GACD,eAAe,UAAU,KAAK,EAAE,OAAO,yCAAyC,CAAC;GACjF;EACF;EACA,IAAI,OAAO,WAAW,OAAO,IAAI,OAAO,kBAAkB;GACxD,QAAQ;IACN,MAAM;IACN,QAAQ;IACR;IACA;IACA,WAAW,KAAK;GAClB,CAAC;GACD,eAAe,UAAU,KAAK,EAAE,OAAO,gCAAgC,CAAC;GACxE;EACF;EACA,QAAQ;GACN,MAAM;GACN;GACA;GACA,WAAW,KAAK;GAChB,kBAAkB;GAClB,UAAU,gBAAgB,MAAM;EAClC,CAAC;EACD,sBAAsB,UAAU,KAAK,OAAO;CAC9C,QAAQ;EACN,eAAe,UAAU,KAAK,EAAE,OAAO,oBAAoB,CAAC;CAC9D;AACF;AAEA,eAAe,sBAAsB,gBAAmD;CACtF,OAAO,eAAe,OAAO,GAC3B,MAAM,QAAQ,WAAW,CAAC,GAAG,cAAc,CAAC;AAEhD;AAEA,SAAS,eAAe,UAA0B,QAAgB,MAAqB;CACrF,IAAI,SAAS,aAAa,SAAS,eAAe;CAClD,SAAS,UAAU,QAAQ,IAAI;AACjC;AAEA,SAAS,SAAS,SAA0B,iBAA2C;CACrF,OAAO,IAAI,SAAS,gBAAgB,WAAW;EAC7C,IAAI,OAAO;EACX,MAAM,SAAmB,CAAC;EAC1B,QAAQ,GAAG,SAAS,UAAkB;GACpC,QAAQ,MAAM;GACd,IAAI,OAAO,iBAAiB;IAC1B,uBAAO,IAAI,MAAM,yBAAyB,CAAC;IAC3C,QAAQ,QAAQ;IAChB;GACF;GACA,OAAO,KAAK,KAAK;EACnB,CAAC;EACD,QAAQ,GAAG,SAAS,MAAM;EAC1B,QAAQ,GAAG,aAAa;GACtB,IAAI;IACF,eAAe,KAAK,MAAM,OAAO,OAAO,MAAM,CAAC,CAAC,SAAS,MAAM,CAAC,CAAC;GACnE,SAAS,OAAO;IACd,OAAO,KAAK;GACd;EACF,CAAC;CACH,CAAC;AACH;AAEA,SAAS,mBAAmB,OAAwB;CAClD,MAAM,UAAU,KAAK,UAAU,KAAK;CACpC,IAAI,YAAY,KAAA,GAAW,MAAM,IAAI,MAAM,+CAA+C;CAC1F,OAAO;AACT;AAEA,SAAS,sBAAsB,UAA0B,QAAgB,SAAuB;CAC9F,IAAI,SAAS,aAAa,SAAS,eAAe;CAClD,SAAS,UAAU,QAAQ;EACzB,gBAAgB;EAChB,kBAAkB,OAAO,OAAO,WAAW,OAAO,CAAC;CACrD,CAAC;CACD,SAAS,IAAI,OAAO;AACtB;AAEA,SAAS,qBAAqB,MAQrB;CACP,IAAI,OAAO,KAAK,UAAU,YAAY,CAAC,KAAK,MAAM,KAAK,GACrD,MAAM,IAAI,MAAM,sDAAsD;CAExE,IAAI,CAAC,OAAO,cAAc,KAAK,cAAc,KAAK,KAAK,kBAAkB,GACvE,MAAM,IAAI,MAAM,6EAA6E;CAE/F,IAAI,KAAK,qBAAqB,KAAA,KAAa,OAAO,KAAK,qBAAqB,YAC1E,MAAM,IAAI,MAAM,kEAAkE;CAEpF,IAAI,OAAO,KAAK,aAAa,YAC3B,MAAM,IAAI,MAAM,0DAA0D;CAE5E,IAAI,KAAK,YAAY,KAAA,KAAa,OAAO,KAAK,YAAY,YACxD,MAAM,IAAI,MAAM,yDAAyD;AAE7E;AAEA,SAAS,eAAe,WAAyD;CAC/E,OAAO,OAAO,cAAc,WAAW,YAAY,EAAE,GAAG,UAAU;AACpE"}

package/dist/{external-optimizer-subprocess-CKNb42oM.js → external-optimizer-subprocess-BIWbHpgD.js}

@@ -1,8 +1,8 @@
 import { s as ValidationError } from "./errors-Dngq5h35.js";
-import { t as canonicalJson } from "./verdict-cache-mZf5FEiY.js";
+import { t as canonicalJson } from "./verdict-cache-CdVVTVmn.js";
 import { c as costForTokenPricing, i as CostLedger } from "./cost-ledger-B1qx30B4.js";
-import { a as appendLedgerLine, o as tryWithLedgerFileLock } from "./ledger-core-DTae9rv_.js";
-import { h as ModelSubstitutionError, v as assertServedModel, y as assertServedModelPolicy } from "./llm-client-Bg32RW0j.js";
+import { a as appendLedgerLine, o as tryWithLedgerFileLock } from "./ledger-core-BOzlRygb.js";
+import { f as ModelSubstitutionError, g as assertServedModelPolicy, h as assertServedModel } from "./llm-client-hgDieDNN.js";
 import { createRequire } from "node:module";
 import { randomBytes } from "node:crypto";
 import { isAbsolute, join, resolve } from "node:path";
@@ -1901,4 +1901,4 @@ function summarizeProcessOutput(capture, max = 4e3) {
 //#endregion
 export { SEARCH_LEDGER_FILE_CONTEXT as C, SearchLedgerIntegrityError as E, inMemoryCampaignStorage as S, SearchLedgerError as T, resolveExternalOptimizerCallbackLimits as _, closeServer as a, createRunCostLedger as b, DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS as c, assertJsonValue as d, assertNoCredentialValues as f, removeCredentialEnvironment as g, isRecord as h, startExternalOptimizerModelProxy as i, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS as l, isExternalTextCandidate as m, closeExternalOptimizerResources as n, listenLocal as o, isCandidateText as p, runWithCleanup as r, sendJson as s, runExternalOptimizerProcess as t, assertExternalOptimizerModelBudget as u, resolveExternalOptimizerProcessLimits as v, SearchLedgerConflictError as w, fsCampaignStorage as x, safePathComponent as y };
 
-//# sourceMappingURL=external-optimizer-subprocess-CKNb42oM.js.map
+//# sourceMappingURL=external-optimizer-subprocess-BIWbHpgD.js.map