@tangle-network/agent-eval 0.118.2 → 0.119.0

package/dist/primeintellect/index.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../src/primeintellect/index.ts"],"sourcesContent":["import { mkdir, writeFile } from 'node:fs/promises'\nimport { dirname, join } from 'node:path'\nimport { type RunRecord, validateRunRecord } from '../run-record'\n\nexport type PrimeIntellectJson =\n  | null\n  | boolean\n  | number\n  | string\n  | PrimeIntellectJson[]\n  | { [key: string]: PrimeIntellectJson }\n\nexport interface PrimeIntellectMessage {\n  role: 'system' | 'user' | 'assistant' | 'tool'\n  content: string\n}\n\nexport interface PrimeIntellectScenario {\n  id: string\n  prompt: string | readonly PrimeIntellectMessage[]\n  answer?: string\n  requiredSubstrings?: readonly string[]\n  split?: string\n  info?: Record<string, PrimeIntellectJson>\n}\n\nexport interface PrimeIntellectDatasetRow {\n  id: string\n  prompt: PrimeIntellectMessage[]\n  answer?: string\n  required_substrings?: string[]\n  split?: string\n  info: Record<string, PrimeIntellectJson>\n}\n\nexport type PrimeIntellectScenarioMap =\n  | ReadonlyMap<string, PrimeIntellectScenario>\n  | Record<string, PrimeIntellectScenario>\n  | readonly PrimeIntellectScenario[]\n\nexport interface PrimeIntellectRowsFromRunRecordsOptions {\n  records: readonly RunRecord[]\n  scenarios: PrimeIntellectScenarioMap\n  requireScorable?: boolean\n  includeFailed?: boolean\n}\n\nexport interface PrimeIntellectEnvironmentPackageInput {\n  name: string\n  version?: string\n  description?: string\n  tags?: readonly string[]\n  moduleName?: string\n  rows: readonly PrimeIntellectDatasetRow[]\n  runRecords?: readonly RunRecord[]\n  systemPrompt?: string\n  verifiersVersion?: string\n  dependencies?: readonly string[]\n  readme?: string\n}\n\nexport interface PrimeIntellectPackageFile {\n  path: string\n  content: string\n}\n\nexport interface PrimeIntellectPackageManifest {\n  schemaVersion: 1\n  name: string\n  moduleName: string\n  version: string\n  rowCount: number\n  runRecordCount: number\n  artifactKinds: readonly ['environment', 'dataset', 'run_records']\n}\n\nexport interface PrimeIntellectEnvironmentPackage {\n  files: PrimeIntellectPackageFile[]\n  manifest: PrimeIntellectPackageManifest\n}\n\nexport class PrimeIntellectBridgeError extends Error {\n  constructor(message: string) {\n    super(message)\n    this.name = 'PrimeIntellectBridgeError'\n  }\n}\n\nexport function primeIntellectRowsFromRunRecords(\n  options: PrimeIntellectRowsFromRunRecordsOptions,\n): PrimeIntellectDatasetRow[] {\n  const scenarioById = normalizeScenarios(options.scenarios)\n  const requireScorable = options.requireScorable ?? true\n  const includeFailed = options.includeFailed ?? true\n  const rows: PrimeIntellectDatasetRow[] = []\n\n  for (const record of options.records) {\n    const validRecord = validateRunRecord(record)\n    if (!includeFailed && validRecord.failureMode) continue\n\n    const scenarioId = validRecord.scenarioId\n    if (!scenarioId) {\n      throw new PrimeIntellectBridgeError(\n        `RunRecord ${validRecord.runId} is missing scenarioId; PrimeIntellect rows need stable task ids`,\n      )\n    }\n\n    const scenario = scenarioById.get(scenarioId)\n    if (!scenario) {\n      throw new PrimeIntellectBridgeError(\n        `RunRecord ${validRecord.runId} references unknown scenarioId ${JSON.stringify(scenarioId)}`,\n      )\n    }\n\n    const hasAnswer = typeof scenario.answer === 'string' && scenario.answer.trim().length > 0\n    const hasRequiredSubstrings = (scenario.requiredSubstrings?.length ?? 0) > 0\n    if (requireScorable && !hasAnswer && !hasRequiredSubstrings) {\n      throw new PrimeIntellectBridgeError(\n        `Scenario ${scenario.id} has no answer or requiredSubstrings; generated environments would always score 0`,\n      )\n    }\n\n    const score = validRecord.outcome.holdoutScore ?? validRecord.outcome.searchScore ?? null\n    const split = scenario.split ?? validRecord.splitTag\n    const tangleInfo = {\n      run_id: validRecord.runId,\n      experiment_id: validRecord.experimentId,\n      candidate_id: validRecord.candidateId,\n      scenario_id: scenarioId,\n      split,\n      score,\n      cost_usd: validRecord.costUsd,\n      wall_ms: validRecord.wallMs,\n      failure_mode: validRecord.failureMode ?? null,\n      model: validRecord.model,\n      commit_sha: validRecord.commitSha,\n    } satisfies Record<string, PrimeIntellectJson>\n\n    rows.push({\n      id: `${scenarioId}:${validRecord.runId}`,\n      prompt: normalizePrompt(scenario.prompt),\n      ...(hasAnswer ? { answer: scenario.answer } : {}),\n      ...(hasRequiredSubstrings\n        ? { required_substrings: [...(scenario.requiredSubstrings ?? [])] }\n        : {}),\n      split,\n      info: {\n        ...(scenario.info ?? {}),\n        tangle: tangleInfo,\n      },\n    })\n  }\n\n  return rows\n}\n\nexport function buildPrimeIntellectEnvironmentPackage(\n  input: PrimeIntellectEnvironmentPackageInput,\n): PrimeIntellectEnvironmentPackage {\n  if (input.rows.length === 0) {\n    throw new PrimeIntellectBridgeError('cannot build a PrimeIntellect environment with zero rows')\n  }\n\n  for (const record of input.runRecords ?? []) {\n    validateRunRecord(record)\n  }\n\n  const moduleName = input.moduleName ?? toPythonModuleName(input.name)\n  if (!/^[A-Za-z_][A-Za-z0-9_]*$/.test(moduleName)) {\n    throw new PrimeIntellectBridgeError(\n      `moduleName ${JSON.stringify(moduleName)} is not a valid Python module name`,\n    )\n  }\n\n  const version = input.version ?? '0.1.0'\n  const description =\n    input.description ?? 'PrimeIntellect Verifiers environment exported from Tangle runs.'\n  const verifiersVersion = input.verifiersVersion ?? '>=0.2.0,<0.3.0'\n  const dependencies = unique([\n    'datasets',\n    `verifiers${verifiersVersion}`,\n    ...(input.dependencies ?? []),\n  ])\n  const tags = unique(['tangle', 'agent-eval', ...(input.tags ?? [])])\n  const manifest: PrimeIntellectPackageManifest = {\n    schemaVersion: 1,\n    name: input.name,\n    moduleName,\n    version,\n    rowCount: input.rows.length,\n    runRecordCount: input.runRecords?.length ?? 0,\n    artifactKinds: ['environment', 'dataset', 'run_records'],\n  }\n\n  const files: PrimeIntellectPackageFile[] = [\n    {\n      path: 'pyproject.toml',\n      content: renderPyproject({\n        name: input.name,\n        version,\n        description,\n        dependencies,\n        tags,\n      }),\n    },\n    {\n      path: 'README.md',\n      content: input.readme ?? renderReadme({ name: input.name, description }),\n    },\n    {\n      path: `${moduleName}.py`,\n      content: renderEnvironmentModule({\n        systemPrompt: input.systemPrompt,\n      }),\n    },\n    {\n      path: 'data/dataset.jsonl',\n      content: toJsonl(input.rows),\n    },\n    {\n      path: 'data/run_records.jsonl',\n      content: toJsonl(input.runRecords ?? []),\n    },\n    {\n      path: 'tangle-primeintellect-manifest.json',\n      content: `${stableJson(manifest)}\\n`,\n    },\n  ]\n\n  return { files, manifest }\n}\n\nexport async function writePrimeIntellectEnvironmentPackage(\n  root: string,\n  input: PrimeIntellectEnvironmentPackageInput,\n): Promise<PrimeIntellectEnvironmentPackage> {\n  const pkg = buildPrimeIntellectEnvironmentPackage(input)\n  await Promise.all(\n    pkg.files.map(async (file) => {\n      const target = join(root, file.path)\n      await mkdir(dirname(target), { recursive: true })\n      await writeFile(target, file.content, 'utf8')\n    }),\n  )\n  return pkg\n}\n\nfunction normalizeScenarios(\n  scenarios: PrimeIntellectScenarioMap,\n): Map<string, PrimeIntellectScenario> {\n  const map = new Map<string, PrimeIntellectScenario>()\n  if (scenarios instanceof Map) {\n    for (const [id, scenario] of scenarios.entries()) {\n      map.set(id, { ...scenario, id: scenario.id ?? id })\n    }\n    return map\n  }\n\n  const values = Array.isArray(scenarios) ? scenarios : Object.values(scenarios)\n  for (const scenario of values) {\n    if (!scenario.id) {\n      throw new PrimeIntellectBridgeError('every PrimeIntellect scenario needs an id')\n    }\n    if (map.has(scenario.id)) {\n      throw new PrimeIntellectBridgeError(`duplicate PrimeIntellect scenario id ${scenario.id}`)\n    }\n    map.set(scenario.id, scenario)\n  }\n  return map\n}\n\nfunction normalizePrompt(\n  prompt: string | readonly PrimeIntellectMessage[],\n): PrimeIntellectMessage[] {\n  if (typeof prompt === 'string') {\n    return [{ role: 'user', content: prompt }]\n  }\n  return prompt.map((message) => ({ role: message.role, content: message.content }))\n}\n\nfunction toPythonModuleName(name: string): string {\n  return name\n    .trim()\n    .toLowerCase()\n    .replace(/[^a-z0-9_]+/g, '_')\n    .replace(/^_+|_+$/g, '')\n    .replace(/^[0-9]/, '_$&')\n}\n\nfunction unique(values: readonly string[]): string[] {\n  return [...new Set(values)]\n}\n\nfunction toJsonl(values: readonly unknown[]): string {\n  if (values.length === 0) return ''\n  return `${values.map((value) => stableJson(value)).join('\\n')}\\n`\n}\n\nfunction stableJson(value: unknown): string {\n  return JSON.stringify(sortJson(value))\n}\n\nfunction sortJson(value: unknown): unknown {\n  if (Array.isArray(value)) return value.map(sortJson)\n  if (value === null || typeof value !== 'object') return value\n  return Object.fromEntries(\n    Object.entries(value as Record<string, unknown>)\n      .sort(([a], [b]) => a.localeCompare(b))\n      .map(([key, nested]) => [key, sortJson(nested)]),\n  )\n}\n\nfunction renderPyproject(input: {\n  name: string\n  version: string\n  description: string\n  dependencies: readonly string[]\n  tags: readonly string[]\n}): string {\n  return `[project]\nname = ${tomlString(input.name)}\nversion = ${tomlString(input.version)}\ndescription = ${tomlString(input.description)}\nreadme = \"README.md\"\nrequires-python = \">=3.11\"\nlicense = \"MIT\"\nkeywords = ${tomlArray(input.tags)}\ndependencies = ${tomlArray(input.dependencies)}\n\n[build-system]\nrequires = [\"hatchling\"]\nbuild-backend = \"hatchling.build\"\n\n[tool.hatch.build]\ninclude = [\"*.py\", \"README.md\", \"data/*.jsonl\", \"tangle-primeintellect-manifest.json\"]\n\n[tool.uv]\nprerelease = \"allow\"\n\n[tool.verifiers.eval]\nnum_examples = 5\nrollouts_per_example = 3\n`\n}\n\nfunction renderReadme(input: { name: string; description: string }): string {\n  return `# ${input.name}\n\n${input.description}\n\nGenerated by \\`@tangle-network/agent-eval/primeintellect\\` from validated Tangle \\`RunRecord\\` rows.\n\n## Files\n\n- \\`data/dataset.jsonl\\`: prompts and scoring references for PrimeIntellect Verifiers.\n- \\`data/run_records.jsonl\\`: original Tangle run rows for provenance and analysis.\n- \\`tangle-primeintellect-manifest.json\\`: export metadata.\n\n## Local check\n\n\\`\\`\\`sh\nuv pip install --prerelease=allow -e .\nuv run vf-eval ${input.name}\n\\`\\`\\`\n\n## Upload\n\n\\`\\`\\`sh\nprime login\nprime env push\n\\`\\`\\`\n`\n}\n\nfunction renderEnvironmentModule(input: { systemPrompt?: string }): string {\n  const systemPrompt = input.systemPrompt === undefined ? 'None' : pyString(input.systemPrompt)\n  return `import json\nimport re\nfrom pathlib import Path\n\nfrom datasets import Dataset\nimport verifiers as vf\n\n\nDATA_PATH = Path(__file__).parent / \"data\" / \"dataset.jsonl\"\nDEFAULT_SYSTEM_PROMPT = ${systemPrompt}\n\n\ndef _load_rows(dataset_path=None, dataset_split=None):\n    path = Path(dataset_path) if dataset_path is not None else DATA_PATH\n    rows = []\n    with path.open(\"r\", encoding=\"utf-8\") as handle:\n        for line in handle:\n            row = json.loads(line)\n            if dataset_split is None or row.get(\"split\") == dataset_split:\n                rows.append(row)\n    if not rows:\n        raise ValueError(f\"No dataset rows found in {path} for split={dataset_split!r}\")\n    return rows\n\n\ndef _message_text(message):\n    if isinstance(message, dict):\n        content = message.get(\"content\", \"\")\n    else:\n        content = getattr(message, \"content\", \"\")\n    if isinstance(content, list):\n        return \" \".join(\n            str(part.get(\"text\", part)) if isinstance(part, dict) else str(part)\n            for part in content\n        )\n    return str(content)\n\n\ndef _completion_text(completion):\n    if isinstance(completion, str):\n        return completion\n    if isinstance(completion, list):\n        return \"\\\\n\".join(_message_text(message) for message in completion)\n    return _message_text(completion)\n\n\ndef _normalize(text):\n    return re.sub(r\"\\\\s+\", \" \", str(text).strip().lower())\n\n\nasync def tangle_reward(completion, answer=None, required_substrings=None, **kwargs):\n    response = _normalize(_completion_text(completion))\n    scores = []\n    if answer:\n        expected = _normalize(answer)\n        scores.append(1.0 if expected == response or expected in response else 0.0)\n    if required_substrings:\n        required = [_normalize(item) for item in required_substrings]\n        hits = sum(1 for item in required if item and item in response)\n        scores.append(hits / len(required))\n    return max(scores) if scores else 0.0\n\n\ndef build_dataset(dataset_path=None, dataset_split=None):\n    return Dataset.from_list(_load_rows(dataset_path=dataset_path, dataset_split=dataset_split))\n\n\ndef load_environment(dataset_path=None, dataset_split=None, system_prompt=DEFAULT_SYSTEM_PROMPT):\n    rubric = vf.Rubric(funcs=[tangle_reward])\n    return vf.SingleTurnEnv(\n        dataset=lambda: build_dataset(dataset_path=dataset_path, dataset_split=dataset_split),\n        system_prompt=system_prompt,\n        rubric=rubric,\n    )\n`\n}\n\nfunction tomlString(value: string): string {\n  return JSON.stringify(value)\n}\n\nfunction tomlArray(values: readonly string[]): string {\n  return `[${values.map(tomlString).join(', ')}]`\n}\n\nfunction pyString(value: string): string {\n  return JSON.stringify(value)\n}\n"],"mappings":";;;;;;;;;;AAAA,SAAS,OAAO,iBAAiB;AACjC,SAAS,SAAS,YAAY;AAgFvB,IAAM,4BAAN,cAAwC,MAAM;AAAA,EACnD,YAAY,SAAiB;AAC3B,UAAM,OAAO;AACb,SAAK,OAAO;AAAA,EACd;AACF;AAEO,SAAS,iCACd,SAC4B;AAC5B,QAAM,eAAe,mBAAmB,QAAQ,SAAS;AACzD,QAAM,kBAAkB,QAAQ,mBAAmB;AACnD,QAAM,gBAAgB,QAAQ,iBAAiB;AAC/C,QAAM,OAAmC,CAAC;AAE1C,aAAW,UAAU,QAAQ,SAAS;AACpC,UAAM,cAAc,kBAAkB,MAAM;AAC5C,QAAI,CAAC,iBAAiB,YAAY,YAAa;AAE/C,UAAM,aAAa,YAAY;AAC/B,QAAI,CAAC,YAAY;AACf,YAAM,IAAI;AAAA,QACR,aAAa,YAAY,KAAK;AAAA,MAChC;AAAA,IACF;AAEA,UAAM,WAAW,aAAa,IAAI,UAAU;AAC5C,QAAI,CAAC,UAAU;AACb,YAAM,IAAI;AAAA,QACR,aAAa,YAAY,KAAK,kCAAkC,KAAK,UAAU,UAAU,CAAC;AAAA,MAC5F;AAAA,IACF;AAEA,UAAM,YAAY,OAAO,SAAS,WAAW,YAAY,SAAS,OAAO,KAAK,EAAE,SAAS;AACzF,UAAM,yBAAyB,SAAS,oBAAoB,UAAU,KAAK;AAC3E,QAAI,mBAAmB,CAAC,aAAa,CAAC,uBAAuB;AAC3D,YAAM,IAAI;AAAA,QACR,YAAY,SAAS,EAAE;AAAA,MACzB;AAAA,IACF;AAEA,UAAM,QAAQ,YAAY,QAAQ,gBAAgB,YAAY,QAAQ,eAAe;AACrF,UAAM,QAAQ,SAAS,SAAS,YAAY;AAC5C,UAAM,aAAa;AAAA,MACjB,QAAQ,YAAY;AAAA,MACpB,eAAe,YAAY;AAAA,MAC3B,cAAc,YAAY;AAAA,MAC1B,aAAa;AAAA,MACb;AAAA,MACA;AAAA,MACA,UAAU,YAAY;AAAA,MACtB,SAAS,YAAY;AAAA,MACrB,cAAc,YAAY,eAAe;AAAA,MACzC,OAAO,YAAY;AAAA,MACnB,YAAY,YAAY;AAAA,IAC1B;AAEA,SAAK,KAAK;AAAA,MACR,IAAI,GAAG,UAAU,IAAI,YAAY,KAAK;AAAA,MACtC,QAAQ,gBAAgB,SAAS,MAAM;AAAA,MACvC,GAAI,YAAY,EAAE,QAAQ,SAAS,OAAO,IAAI,CAAC;AAAA,MAC/C,GAAI,wBACA,EAAE,qBAAqB,CAAC,GAAI,SAAS,sBAAsB,CAAC,CAAE,EAAE,IAChE,CAAC;AAAA,MACL;AAAA,MACA,MAAM;AAAA,QACJ,GAAI,SAAS,QAAQ,CAAC;AAAA,QACtB,QAAQ;AAAA,MACV;AAAA,IACF,CAAC;AAAA,EACH;AAEA,SAAO;AACT;AAEO,SAAS,sCACd,OACkC;AAClC,MAAI,MAAM,KAAK,WAAW,GAAG;AAC3B,UAAM,IAAI,0BAA0B,0DAA0D;AAAA,EAChG;AAEA,aAAW,UAAU,MAAM,cAAc,CAAC,GAAG;AAC3C,sBAAkB,MAAM;AAAA,EAC1B;AAEA,QAAM,aAAa,MAAM,cAAc,mBAAmB,MAAM,IAAI;AACpE,MAAI,CAAC,2BAA2B,KAAK,UAAU,GAAG;AAChD,UAAM,IAAI;AAAA,MACR,cAAc,KAAK,UAAU,UAAU,CAAC;AAAA,IAC1C;AAAA,EACF;AAEA,QAAM,UAAU,MAAM,WAAW;AACjC,QAAM,cACJ,MAAM,eAAe;AACvB,QAAM,mBAAmB,MAAM,oBAAoB;AACnD,QAAM,eAAe,OAAO;AAAA,IAC1B;AAAA,IACA,YAAY,gBAAgB;AAAA,IAC5B,GAAI,MAAM,gBAAgB,CAAC;AAAA,EAC7B,CAAC;AACD,QAAM,OAAO,OAAO,CAAC,UAAU,cAAc,GAAI,MAAM,QAAQ,CAAC,CAAE,CAAC;AACnE,QAAM,WAA0C;AAAA,IAC9C,eAAe;AAAA,IACf,MAAM,MAAM;AAAA,IACZ;AAAA,IACA;AAAA,IACA,UAAU,MAAM,KAAK;AAAA,IACrB,gBAAgB,MAAM,YAAY,UAAU;AAAA,IAC5C,eAAe,CAAC,eAAe,WAAW,aAAa;AAAA,EACzD;AAEA,QAAM,QAAqC;AAAA,IACzC;AAAA,MACE,MAAM;AAAA,MACN,SAAS,gBAAgB;AAAA,QACvB,MAAM,MAAM;AAAA,QACZ;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,MACF,CAAC;AAAA,IACH;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,SAAS,MAAM,UAAU,aAAa,EAAE,MAAM,MAAM,MAAM,YAAY,CAAC;AAAA,IACzE;AAAA,IACA;AAAA,MACE,MAAM,GAAG,UAAU;AAAA,MACnB,SAAS,wBAAwB;AAAA,QAC/B,cAAc,MAAM;AAAA,MACtB,CAAC;AAAA,IACH;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,SAAS,QAAQ,MAAM,IAAI;AAAA,IAC7B;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,SAAS,QAAQ,MAAM,cAAc,CAAC,CAAC;AAAA,IACzC;AAAA,IACA;AAAA,MACE,MAAM;AAAA,MACN,SAAS,GAAG,WAAW,QAAQ,CAAC;AAAA;AAAA,IAClC;AAAA,EACF;AAEA,SAAO,EAAE,OAAO,SAAS;AAC3B;AAEA,eAAsB,sCACpB,MACA,OAC2C;AAC3C,QAAM,MAAM,sCAAsC,KAAK;AACvD,QAAM,QAAQ;AAAA,IACZ,IAAI,MAAM,IAAI,OAAO,SAAS;AAC5B,YAAM,SAAS,KAAK,MAAM,KAAK,IAAI;AACnC,YAAM,MAAM,QAAQ,MAAM,GAAG,EAAE,WAAW,KAAK,CAAC;AAChD,YAAM,UAAU,QAAQ,KAAK,SAAS,MAAM;AAAA,IAC9C,CAAC;AAAA,EACH;AACA,SAAO;AACT;AAEA,SAAS,mBACP,WACqC;AACrC,QAAM,MAAM,oBAAI,IAAoC;AACpD,MAAI,qBAAqB,KAAK;AAC5B,eAAW,CAAC,IAAI,QAAQ,KAAK,UAAU,QAAQ,GAAG;AAChD,UAAI,IAAI,IAAI,EAAE,GAAG,UAAU,IAAI,SAAS,MAAM,GAAG,CAAC;AAAA,IACpD;AACA,WAAO;AAAA,EACT;AAEA,QAAM,SAAS,MAAM,QAAQ,SAAS,IAAI,YAAY,OAAO,OAAO,SAAS;AAC7E,aAAW,YAAY,QAAQ;AAC7B,QAAI,CAAC,SAAS,IAAI;AAChB,YAAM,IAAI,0BAA0B,2CAA2C;AAAA,IACjF;AACA,QAAI,IAAI,IAAI,SAAS,EAAE,GAAG;AACxB,YAAM,IAAI,0BAA0B,wCAAwC,SAAS,EAAE,EAAE;AAAA,IAC3F;AACA,QAAI,IAAI,SAAS,IAAI,QAAQ;AAAA,EAC/B;AACA,SAAO;AACT;AAEA,SAAS,gBACP,QACyB;AACzB,MAAI,OAAO,WAAW,UAAU;AAC9B,WAAO,CAAC,EAAE,MAAM,QAAQ,SAAS,OAAO,CAAC;AAAA,EAC3C;AACA,SAAO,OAAO,IAAI,CAAC,aAAa,EAAE,MAAM,QAAQ,MAAM,SAAS,QAAQ,QAAQ,EAAE;AACnF;AAEA,SAAS,mBAAmB,MAAsB;AAChD,SAAO,KACJ,KAAK,EACL,YAAY,EACZ,QAAQ,gBAAgB,GAAG,EAC3B,QAAQ,YAAY,EAAE,EACtB,QAAQ,UAAU,KAAK;AAC5B;AAEA,SAAS,OAAO,QAAqC;AACnD,SAAO,CAAC,GAAG,IAAI,IAAI,MAAM,CAAC;AAC5B;AAEA,SAAS,QAAQ,QAAoC;AACnD,MAAI,OAAO,WAAW,EAAG,QAAO;AAChC,SAAO,GAAG,OAAO,IAAI,CAAC,UAAU,WAAW,KAAK,CAAC,EAAE,KAAK,IAAI,CAAC;AAAA;AAC/D;AAEA,SAAS,WAAW,OAAwB;AAC1C,SAAO,KAAK,UAAU,SAAS,KAAK,CAAC;AACvC;AAEA,SAAS,SAAS,OAAyB;AACzC,MAAI,MAAM,QAAQ,KAAK,EAAG,QAAO,MAAM,IAAI,QAAQ;AACnD,MAAI,UAAU,QAAQ,OAAO,UAAU,SAAU,QAAO;AACxD,SAAO,OAAO;AAAA,IACZ,OAAO,QAAQ,KAAgC,EAC5C,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC,EACrC,IAAI,CAAC,CAAC,KAAK,MAAM,MAAM,CAAC,KAAK,SAAS,MAAM,CAAC,CAAC;AAAA,EACnD;AACF;AAEA,SAAS,gBAAgB,OAMd;AACT,SAAO;AAAA,SACA,WAAW,MAAM,IAAI,CAAC;AAAA,YACnB,WAAW,MAAM,OAAO,CAAC;AAAA,gBACrB,WAAW,MAAM,WAAW,CAAC;AAAA;AAAA;AAAA;AAAA,aAIhC,UAAU,MAAM,IAAI,CAAC;AAAA,iBACjB,UAAU,MAAM,YAAY,CAAC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAgB9C;AAEA,SAAS,aAAa,OAAsD;AAC1E,SAAO,KAAK,MAAM,IAAI;AAAA;AAAA,EAEtB,MAAM,WAAW;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,iBAcF,MAAM,IAAI;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAU3B;AAEA,SAAS,wBAAwB,OAA0C;AACzE,QAAM,eAAe,MAAM,iBAAiB,SAAY,SAAS,SAAS,MAAM,YAAY;AAC5F,SAAO;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,0BASiB,YAAY;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAkEtC;AAEA,SAAS,WAAW,OAAuB;AACzC,SAAO,KAAK,UAAU,KAAK;AAC7B;AAEA,SAAS,UAAU,QAAmC;AACpD,SAAO,IAAI,OAAO,IAAI,UAAU,EAAE,KAAK,IAAI,CAAC;AAC9C;AAEA,SAAS,SAAS,OAAuB;AACvC,SAAO,KAAK,UAAU,KAAK;AAC7B;","names":[]}

package/dist/rl.d.ts

@@ -2384,7 +2384,9 @@ interface ChannelRollup {
     calls: number;
     inputTokens: number;
     outputTokens: number;
+    reasoningTokens?: number;
     cachedTokens: number;
+    cacheWriteTokens?: number;
     costUsd: number;
     unpricedCalls: number;
     unknownUsageCalls: number;
@@ -2396,7 +2398,9 @@ interface CostLedgerSummary {
     reservedCostUsd: number;
     inputTokens: number;
     outputTokens: number;
+    reasoningTokens?: number;
     cachedTokens: number;
+    cacheWriteTokens?: number;
     totalCostUsd: number;
     byChannel: ChannelRollup[];
     unpricedModels: string[];

package/dist/{run-campaign-DWC67KJP.js → run-campaign-OBXSN5WK.js}

@@ -1,9 +1,9 @@
 import {
   planCampaignRun,
   runCampaign
-} from "./chunk-EGNRE7VA.js";
+} from "./chunk-PAHNGS65.js";
 import "./chunk-PJQFMIOX.js";
-import "./chunk-MLHRFWQK.js";
+import "./chunk-JHOJHHU7.js";
 import "./chunk-VI2UW6B6.js";
 import "./chunk-ONWEPEDO.js";
 import "./chunk-PZ5AY32C.js";
@@ -11,4 +11,4 @@ export {
   planCampaignRun,
   runCampaign
 };
-//# sourceMappingURL=run-campaign-DWC67KJP.js.map
+//# sourceMappingURL=run-campaign-OBXSN5WK.js.map

package/dist/trace-attributes.d.ts

@@ -14,6 +14,22 @@ declare const TOOL_ARGS_CAPTURED = "tool.args_captured";
 declare const TOOL_LATENCY_MS = "tool.latency_ms";
 declare const INPUT_VALUE = "input.value";
 declare const OUTPUT_VALUE = "output.value";
+declare const SPAN_KIND_ATTR_KEYS: readonly ["openinference.span.kind", "inference.observation_kind"];
+declare const LLM_MODEL_ATTR_KEYS: readonly ["llm.model_name", "inference.llm.model_name", "llm.model", ...string[], "tangle.model"];
+declare const LLM_INPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt", "inference.llm.input_tokens", "llm.input_tokens", ...string[], "tangle.tokens.in", "tokens.in"];
+declare const LLM_OUTPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.completion", "inference.llm.output_tokens", "llm.output_tokens", ...string[], "tangle.tokens.out", "tokens.out"];
+/** Reasoning-token subset of output, when a producer exposes it separately. */
+declare const LLM_REASONING_TOKEN_ATTR_KEYS: readonly ["llm.token_count.reasoning", "inference.llm.reasoning_tokens", "llm.reasoning_tokens", "gen_ai.usage.reasoning_tokens", "gen_ai.usage.reasoning_output_tokens", "reasoning_tokens", "reasoning_output_tokens", "tangle.tokens.reasoning", "gen_ai.usage.output_tokens_details.reasoning_tokens", "gen_ai.usage.completion_tokens_details.reasoning_tokens"];
+declare const LLM_CACHED_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt_cache_hit", "inference.llm.cached_tokens", "llm.cached_tokens", "gen_ai.usage.cached_tokens", "gen_ai.usage.prompt_tokens_details.cached_tokens", "gen_ai.usage.input_tokens_details.cached_tokens", "gen_ai.usage.cache_read_tokens", "gen_ai.usage.cache_read_input_tokens", "cache_read_tokens", "cache_read_input_tokens", "input_cache_read", "tangle.tokens.cached"];
+declare const LLM_CACHE_WRITE_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt_cache_write", "inference.llm.cache_write_tokens", "llm.cache_write_tokens", "gen_ai.usage.cache_creation_tokens", "gen_ai.usage.cache_creation_input_tokens", "cache_creation_tokens", "cache_creation_input_tokens", "input_cache_creation", "tangle.tokens.cache_write"];
+declare const LLM_COST_ATTR_KEYS: readonly ["llm.cost_usd", "inference.llm.cost.total", "llm.cost.total", "gen_ai.usage.cost", "gen_ai.usage.cost_usd", "tangle.cost.usd", "cost.usd", "cost"];
+/** Explicit run-total cost keys safe to preserve on an untyped span. */
+declare const RUN_COST_ATTR_KEYS: readonly ["tangle.cost.usd", "cost.usd"];
+declare const TOOL_NAME_ATTR_KEYS: readonly ["tool.name", "inference.tool.name"];
+/** Read a numeric attribute, tolerating numeric strings; `null` if absent or invalid. */
+declare function asNumber(value: unknown): number | null;
+/** First finite numeric value across a list of candidate attribute keys. */
+declare function firstNumberAttr(attributes: Record<string, unknown>, keys: readonly string[]): number | null;
 /** Sum producer-supplied, mutually exclusive prompt-token categories. */
 declare function contextInputTokens(usage: {
     inputTokens?: number | null;
@@ -33,4 +49,4 @@ interface LlmSpanOtlpInput {
 /** Write the canonical LLM attributes shared by trace producers. */
 declare function applyLlmSpanOtlpAttributes(attributes: Record<string, unknown>, span: LlmSpanOtlpInput): void;
 
-export { INPUT_VALUE, LLM_CACHED_TOKENS, LLM_CACHE_WRITE_TOKENS, LLM_CONTEXT_TOKENS, LLM_COST_USD, LLM_INPUT_TOKENS, LLM_MODEL_NAME, LLM_OUTPUT_TOKENS, LLM_REASONING_TOKENS, type LlmSpanOtlpInput, OPENINFERENCE_SPAN_KIND, OUTPUT_VALUE, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, applyLlmSpanOtlpAttributes, contextInputTokens };
+export { INPUT_VALUE, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_CACHE_WRITE_TOKENS, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, LLM_CONTEXT_TOKENS, LLM_COST_ATTR_KEYS, LLM_COST_USD, LLM_INPUT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_MODEL_ATTR_KEYS, LLM_MODEL_NAME, LLM_OUTPUT_TOKENS, LLM_OUTPUT_TOKEN_ATTR_KEYS, LLM_REASONING_TOKENS, LLM_REASONING_TOKEN_ATTR_KEYS, type LlmSpanOtlpInput, OPENINFERENCE_SPAN_KIND, OUTPUT_VALUE, RUN_COST_ATTR_KEYS, SPAN_KIND_ATTR_KEYS, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, applyLlmSpanOtlpAttributes, asNumber, contextInputTokens, firstNumberAttr };

package/dist/trace-attributes.js

@@ -1,38 +1,62 @@
 import {
   INPUT_VALUE,
   LLM_CACHED_TOKENS,
+  LLM_CACHED_TOKEN_ATTR_KEYS,
   LLM_CACHE_WRITE_TOKENS,
+  LLM_CACHE_WRITE_TOKEN_ATTR_KEYS,
   LLM_CONTEXT_TOKENS,
+  LLM_COST_ATTR_KEYS,
   LLM_COST_USD,
   LLM_INPUT_TOKENS,
+  LLM_INPUT_TOKEN_ATTR_KEYS,
+  LLM_MODEL_ATTR_KEYS,
   LLM_MODEL_NAME,
   LLM_OUTPUT_TOKENS,
+  LLM_OUTPUT_TOKEN_ATTR_KEYS,
   LLM_REASONING_TOKENS,
+  LLM_REASONING_TOKEN_ATTR_KEYS,
   OPENINFERENCE_SPAN_KIND,
   OUTPUT_VALUE,
+  RUN_COST_ATTR_KEYS,
+  SPAN_KIND_ATTR_KEYS,
   TOOL_ARGS_CAPTURED,
   TOOL_LATENCY_MS,
   TOOL_NAME,
+  TOOL_NAME_ATTR_KEYS,
   applyLlmSpanOtlpAttributes,
-  contextInputTokens
-} from "./chunk-NW4AQBXD.js";
+  asNumber,
+  contextInputTokens,
+  firstNumberAttr
+} from "./chunk-K4DBDHLK.js";
 import "./chunk-PZ5AY32C.js";
 export {
   INPUT_VALUE,
   LLM_CACHED_TOKENS,
+  LLM_CACHED_TOKEN_ATTR_KEYS,
   LLM_CACHE_WRITE_TOKENS,
+  LLM_CACHE_WRITE_TOKEN_ATTR_KEYS,
   LLM_CONTEXT_TOKENS,
+  LLM_COST_ATTR_KEYS,
   LLM_COST_USD,
   LLM_INPUT_TOKENS,
+  LLM_INPUT_TOKEN_ATTR_KEYS,
+  LLM_MODEL_ATTR_KEYS,
   LLM_MODEL_NAME,
   LLM_OUTPUT_TOKENS,
+  LLM_OUTPUT_TOKEN_ATTR_KEYS,
   LLM_REASONING_TOKENS,
+  LLM_REASONING_TOKEN_ATTR_KEYS,
   OPENINFERENCE_SPAN_KIND,
   OUTPUT_VALUE,
+  RUN_COST_ATTR_KEYS,
+  SPAN_KIND_ATTR_KEYS,
   TOOL_ARGS_CAPTURED,
   TOOL_LATENCY_MS,
   TOOL_NAME,
+  TOOL_NAME_ATTR_KEYS,
   applyLlmSpanOtlpAttributes,
-  contextInputTokens
+  asNumber,
+  contextInputTokens,
+  firstNumberAttr
 };
 //# sourceMappingURL=trace-attributes.js.map

package/dist/traces.d.ts

@@ -1200,6 +1200,22 @@ declare const TOOL_ARGS_CAPTURED = "tool.args_captured";
 declare const TOOL_LATENCY_MS = "tool.latency_ms";
 declare const INPUT_VALUE = "input.value";
 declare const OUTPUT_VALUE = "output.value";
+declare const SPAN_KIND_ATTR_KEYS: readonly ["openinference.span.kind", "inference.observation_kind"];
+declare const LLM_MODEL_ATTR_KEYS: readonly ["llm.model_name", "inference.llm.model_name", "llm.model", ...string[], "tangle.model"];
+declare const LLM_INPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt", "inference.llm.input_tokens", "llm.input_tokens", ...string[], "tangle.tokens.in", "tokens.in"];
+declare const LLM_OUTPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.completion", "inference.llm.output_tokens", "llm.output_tokens", ...string[], "tangle.tokens.out", "tokens.out"];
+/** Reasoning-token subset of output, when a producer exposes it separately. */
+declare const LLM_REASONING_TOKEN_ATTR_KEYS: readonly ["llm.token_count.reasoning", "inference.llm.reasoning_tokens", "llm.reasoning_tokens", "gen_ai.usage.reasoning_tokens", "gen_ai.usage.reasoning_output_tokens", "reasoning_tokens", "reasoning_output_tokens", "tangle.tokens.reasoning", "gen_ai.usage.output_tokens_details.reasoning_tokens", "gen_ai.usage.completion_tokens_details.reasoning_tokens"];
+declare const LLM_CACHED_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt_cache_hit", "inference.llm.cached_tokens", "llm.cached_tokens", "gen_ai.usage.cached_tokens", "gen_ai.usage.prompt_tokens_details.cached_tokens", "gen_ai.usage.input_tokens_details.cached_tokens", "gen_ai.usage.cache_read_tokens", "gen_ai.usage.cache_read_input_tokens", "cache_read_tokens", "cache_read_input_tokens", "input_cache_read", "tangle.tokens.cached"];
+declare const LLM_CACHE_WRITE_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt_cache_write", "inference.llm.cache_write_tokens", "llm.cache_write_tokens", "gen_ai.usage.cache_creation_tokens", "gen_ai.usage.cache_creation_input_tokens", "cache_creation_tokens", "cache_creation_input_tokens", "input_cache_creation", "tangle.tokens.cache_write"];
+declare const LLM_COST_ATTR_KEYS: readonly ["llm.cost_usd", "inference.llm.cost.total", "llm.cost.total", "gen_ai.usage.cost", "gen_ai.usage.cost_usd", "tangle.cost.usd", "cost.usd", "cost"];
+/** Explicit run-total cost keys safe to preserve on an untyped span. */
+declare const RUN_COST_ATTR_KEYS: readonly ["tangle.cost.usd", "cost.usd"];
+declare const TOOL_NAME_ATTR_KEYS: readonly ["tool.name", "inference.tool.name"];
+/** Read a numeric attribute, tolerating numeric strings; `null` if absent or invalid. */
+declare function asNumber(value: unknown): number | null;
+/** First finite numeric value across a list of candidate attribute keys. */
+declare function firstNumberAttr(attributes: Record<string, unknown>, keys: readonly string[]): number | null;
 /** Sum producer-supplied, mutually exclusive prompt-token categories. */
 declare function contextInputTokens(usage: {
     inputTokens?: number | null;
@@ -1221,18 +1237,6 @@ declare function applyLlmSpanOtlpAttributes(attributes: Record<string, unknown>,
 
 /** Canonical OpenInference-over-OTLP attribute vocabulary used at the trace boundary. */
 
-declare const SPAN_KIND_ATTR_KEYS: readonly ["openinference.span.kind", "inference.observation_kind"];
-declare const LLM_MODEL_ATTR_KEYS: readonly ["llm.model_name", "inference.llm.model_name", "llm.model", ...string[], "tangle.model"];
-declare const LLM_INPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt", "inference.llm.input_tokens", "llm.input_tokens", ...string[], "tangle.tokens.in", "tokens.in"];
-declare const LLM_OUTPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.completion", "inference.llm.output_tokens", "llm.output_tokens", ...string[], "tangle.tokens.out", "tokens.out"];
-/** Reasoning-token subset of output, when a producer exposes it separately. */
-declare const LLM_REASONING_TOKEN_ATTR_KEYS: readonly ["llm.token_count.reasoning", "inference.llm.reasoning_tokens", "llm.reasoning_tokens", "gen_ai.usage.reasoning_tokens", "gen_ai.usage.reasoning_output_tokens", "reasoning_tokens", "reasoning_output_tokens", "tangle.tokens.reasoning", "gen_ai.usage.output_tokens_details.reasoning_tokens", "gen_ai.usage.completion_tokens_details.reasoning_tokens"];
-declare const LLM_CACHED_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt_cache_hit", "inference.llm.cached_tokens", "llm.cached_tokens", "gen_ai.usage.cached_tokens", "gen_ai.usage.prompt_tokens_details.cached_tokens", "gen_ai.usage.input_tokens_details.cached_tokens", "gen_ai.usage.cache_read_tokens", "gen_ai.usage.cache_read_input_tokens", "cache_read_tokens", "cache_read_input_tokens", "input_cache_read", "tangle.tokens.cached"];
-declare const LLM_CACHE_WRITE_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt_cache_write", "inference.llm.cache_write_tokens", "llm.cache_write_tokens", "gen_ai.usage.cache_creation_tokens", "gen_ai.usage.cache_creation_input_tokens", "cache_creation_tokens", "cache_creation_input_tokens", "input_cache_creation", "tangle.tokens.cache_write"];
-declare const LLM_COST_ATTR_KEYS: readonly ["llm.cost_usd", "inference.llm.cost.total", "llm.cost.total", "gen_ai.usage.cost", "gen_ai.usage.cost_usd", "tangle.cost.usd", "cost.usd", "cost"];
-/** Explicit run-total cost keys safe to preserve on an untyped span. */
-declare const RUN_COST_ATTR_KEYS: readonly ["tangle.cost.usd", "cost.usd"];
-declare const TOOL_NAME_ATTR_KEYS: readonly ["tool.name", "inference.tool.name"];
 type ToolSpanOtlpInput = Pick<ToolSpan, 'toolName' | 'args' | 'argsCaptured' | 'result' | 'latencyMs'>;
 declare function applyToolSpanOtlpAttributes(attributes: Record<string, unknown>, span: ToolSpanOtlpInput): void;
 declare function traceSpanKindToOpenInferenceKind(kind: string): string;
@@ -2032,10 +2036,6 @@ declare function inferOtlpKind(attrs: Record<string, unknown>): TraceAnalystSpan
 declare function extractOtlpAttributes(raw: Record<string, unknown>): Record<string, unknown>;
 declare function stringField(raw: Record<string, unknown>, key: string): string | undefined;
 declare function asString(v: unknown): string | null;
-/** Read a numeric attribute, tolerating numeric strings; `null` if absent/NaN. */
-declare function asNumber(v: unknown): number | null;
-/** First finite numeric value across a list of candidate attribute keys. */
-declare function firstNumberAttr(attrs: Record<string, unknown>, keys: readonly string[]): number | null;
 /** First non-empty string value across a list of candidate attribute keys. */
 declare function firstStringAttr(attrs: Record<string, unknown>, keys: readonly string[]): string | null;
 

package/dist/traces.js

@@ -27,8 +27,8 @@ import {
   scoreTraceInsightReadiness,
   tokenizeDomainWords,
   traceAnalystOnRunComplete
-} from "./chunk-6JQE3YVE.js";
-import "./chunk-TVXPJJ7H.js";
+} from "./chunk-ZYHJNKI3.js";
+import "./chunk-H5UD2323.js";
 import {
   extractUsage,
   extractUsageFromResponse,
@@ -50,7 +50,7 @@ import {
   TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
   TRACE_ANALYST_SUBAGENT_DESCRIPTION,
   analyzeTraces
-} from "./chunk-63KG3ASG.js";
+} from "./chunk-FIUFRXP3.js";
 import {
   DEFAULT_REDACTION_RULES,
   REDACTION_VERSION,
@@ -59,27 +59,15 @@ import {
 } from "./chunk-GGE4NNQT.js";
 import {
   DEFAULT_TRACE_ANALYST_BUDGETS,
-  LLM_CACHED_TOKEN_ATTR_KEYS,
-  LLM_CACHE_WRITE_TOKEN_ATTR_KEYS,
-  LLM_COST_ATTR_KEYS,
-  LLM_INPUT_TOKEN_ATTR_KEYS,
-  LLM_MODEL_ATTR_KEYS,
-  LLM_OUTPUT_TOKEN_ATTR_KEYS,
-  LLM_REASONING_TOKEN_ATTR_KEYS,
   OtlpFileTraceStore,
-  RUN_COST_ATTR_KEYS,
-  SPAN_KIND_ATTR_KEYS,
   SpanNotFoundError,
-  TOOL_NAME_ATTR_KEYS,
   TRACE_ANALYST_TRUNCATION_MARKER_PREFIX,
   TraceFileMissingError,
   TraceNotFoundError,
   applyToolSpanOtlpAttributes,
-  asNumber,
   asString,
   buildTraceAnalystTools,
   extractOtlpAttributes,
-  firstNumberAttr,
   firstStringAttr,
   inferOtlpKind,
   projectOtlpFlatLine,
@@ -87,7 +75,7 @@ import {
   stringField,
   traceAnalystFunctionGroup,
   traceSpanKindToOpenInferenceKind
-} from "./chunk-BUBU3YHT.js";
+} from "./chunk-6QIM2EAP.js";
 import {
   RunIntegrityError,
   assertRunCaptured,
@@ -120,21 +108,33 @@ import "./chunk-ONWEPEDO.js";
 import {
   INPUT_VALUE,
   LLM_CACHED_TOKENS,
+  LLM_CACHED_TOKEN_ATTR_KEYS,
   LLM_CACHE_WRITE_TOKENS,
+  LLM_CACHE_WRITE_TOKEN_ATTR_KEYS,
   LLM_CONTEXT_TOKENS,
+  LLM_COST_ATTR_KEYS,
   LLM_COST_USD,
   LLM_INPUT_TOKENS,
+  LLM_INPUT_TOKEN_ATTR_KEYS,
+  LLM_MODEL_ATTR_KEYS,
   LLM_MODEL_NAME,
   LLM_OUTPUT_TOKENS,
+  LLM_OUTPUT_TOKEN_ATTR_KEYS,
   LLM_REASONING_TOKENS,
+  LLM_REASONING_TOKEN_ATTR_KEYS,
   OPENINFERENCE_SPAN_KIND,
   OUTPUT_VALUE,
+  RUN_COST_ATTR_KEYS,
+  SPAN_KIND_ATTR_KEYS,
   TOOL_ARGS_CAPTURED,
   TOOL_LATENCY_MS,
   TOOL_NAME,
+  TOOL_NAME_ATTR_KEYS,
   applyLlmSpanOtlpAttributes,
-  contextInputTokens
-} from "./chunk-NW4AQBXD.js";
+  asNumber,
+  contextInputTokens,
+  firstNumberAttr
+} from "./chunk-K4DBDHLK.js";
 import "./chunk-PZ5AY32C.js";
 export {
   DEFAULT_REDACTION_RULES,

package/dist/wire/index.d.ts

@@ -7,8 +7,14 @@ import { Hono } from 'hono';
 type CostChannel = 'agent' | 'judge' | 'verifier' | 'analyst' | 'driver' | (string & {});
 interface CostUsage {
     inputTokens: number;
+    /** Includes reasoning tokens when the provider bills them as output. */
     outputTokens: number;
+    /** Reasoning-token subset of outputTokens, when reported. */
+    reasoningTokens?: number;
+    /** Prompt tokens served from a provider cache. */
     cachedTokens?: number;
+    /** Prompt tokens written into a provider cache. */
+    cacheWriteTokens?: number;
 }
 interface CostCallBase {
     callId: string;
@@ -75,7 +81,9 @@ interface ChannelRollup {
     calls: number;
     inputTokens: number;
     outputTokens: number;
+    reasoningTokens?: number;
     cachedTokens: number;
+    cacheWriteTokens?: number;
     costUsd: number;
     unpricedCalls: number;
     unknownUsageCalls: number;
@@ -87,7 +95,9 @@ interface CostLedgerSummary {
     reservedCostUsd: number;
     inputTokens: number;
     outputTokens: number;
+    reasoningTokens?: number;
     cachedTokens: number;
+    cacheWriteTokens?: number;
     totalCostUsd: number;
     byChannel: ChannelRollup[];
     unpricedModels: string[];
@@ -101,6 +111,10 @@ interface CostLedgerFilter {
     phase?: string;
     tags?: Record<string, string>;
 }
+interface CostLedgerWaitOptions {
+    /** Maximum time to wait for active provider calls. Default 5 seconds. */
+    timeoutMs?: number;
+}
 /** Append-only storage. `append` must atomically reject stale revisions. */
 interface CostLedgerPersistence {
     read(): {
@@ -120,6 +134,7 @@ declare class CostLedger {
     private readonly records;
     private readonly activeCallIds;
     private readonly lateCallIds;
+    private readonly idleWaiters;
     private completedTasks;
     private revision;
     private costLimitPersisted;
@@ -127,6 +142,8 @@ declare class CostLedger {
     private readonly persistence?;
     constructor(input?: number | CostLedgerOptions);
     runPaidCall<T>(input: RunPaidCallInput<T>): Promise<PaidCallResult<T>>;
+    /** Wait until every call started by this ledger has produced a durable outcome. */
+    waitForIdle(options?: CostLedgerWaitOptions): Promise<boolean>;
     /** Settle a call left pending by a crashed process after reconciling with the provider. */
     reconcile(callId: string, observed: CostReceiptInput, options?: {
         error?: string;
@@ -137,6 +154,7 @@ declare class CostLedger {
     costPerCompletedTask(): number | null;
     private execute;
     private captureLateOutcome;
+    private releaseActiveCall;
     private commitOutcome;
     private captureFailure;
     private commitReceipt;
@@ -152,7 +170,7 @@ declare class CostLedger {
  * Keeping callback contracts structural lets those subpaths compose while the
  * concrete {@link CostLedger} retains its private durable state.
  */
-type CostLedgerHandle = Pick<CostLedger, keyof CostLedger>;
+type CostLedgerHandle = Pick<CostLedger, Exclude<keyof CostLedger, 'waitForIdle'>> & Partial<Pick<CostLedger, 'waitForIdle'>>;
 
 type RunStatus = 'running' | 'completed' | 'failed' | 'aborted';
 interface BudgetSpec {

package/dist/wire/index.js

@@ -34,9 +34,9 @@ import {
   runRpcOnce,
   startServer,
   startServerAsync
-} from "../chunk-JTMFT5QZ.js";
+} from "../chunk-4I2E3LLO.js";
 import "../chunk-NJC7U437.js";
-import "../chunk-MLHRFWQK.js";
+import "../chunk-JHOJHHU7.js";
 import "../chunk-VI2UW6B6.js";
 import "../chunk-PC4UYEBM.js";
 import "../chunk-ONWEPEDO.js";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-eval",
-  "version": "0.118.2",
+  "version": "0.119.0",
   "description": "Evaluate and improve AI agents from runs, traces, judges, and feedback. Compare candidates, cluster failures, measure lift, and gate releases.",
   "homepage": "https://github.com/tangle-network/agent-eval#readme",
   "repository": {
@@ -119,6 +119,11 @@
       "import": "./dist/hosted/index.js",
       "default": "./dist/hosted/index.js"
     },
+    "./primeintellect": {
+      "types": "./dist/primeintellect/index.d.ts",
+      "import": "./dist/primeintellect/index.js",
+      "default": "./dist/primeintellect/index.js"
+    },
     "./openapi.json": {
       "default": "./dist/openapi.json"
     }