@tangle-network/agent-eval 0.205.1 → 0.206.0

package/dist/systemone.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"systemone.js","names":[],"sources":["../src/systemone.ts"],"sourcesContent":["import type { Analyst, AnalystContext, AnalystFinding } from './analyst/types'\nimport type { JudgeConfig, JudgeScore, Scenario } from './campaign/types'\nimport type { CostReceiptInput, CustomTokenPricing } from './cost-ledger'\nimport type { EvaluationContext, EvaluationResult, EvaluatorOptions } from './evaluation'\nimport { asAnalyst, asJudge, createEvaluator } from './evaluation'\nimport { jsonDocument } from './ledger-core/canonical'\nimport { weightedComposite } from './statistics'\nimport type {\n  SystemOneQuestions,\n  SystemOneRequest,\n  SystemOneResult,\n  SystemOneState,\n} from './systemone-protocol'\nimport {\n  parseSystemOneQuestions,\n  parseSystemOneRequest,\n  parseSystemOneResult,\n  systemOneUsage,\n} from './systemone-protocol'\nimport { contentHash } from './verdict-cache'\n\nexport * from './evaluation'\nexport * from './systemone-protocol'\nexport * from './systemone-review'\n\nexport type SystemOneEvaluate = EvaluatorOptions<SystemOneRequest, unknown>['execute']\nexport interface SystemOneEvaluatorOptions {\n  evaluate: SystemOneEvaluate\n  costLedger?: EvaluatorOptions<SystemOneRequest, unknown>['costLedger']\n  maximumCharge?: EvaluatorOptions<SystemOneRequest, unknown>['maximumCharge']\n  pricing?: CustomTokenPricing\n  receipt?: (response: unknown) => CostReceiptInput\n  receiptFromError?: (error: Error) => CostReceiptInput | undefined\n  /** Optional deployment policy; the protocol adapter does not hardcode model aliases. */\n  acceptModel?: (requested: string, served: string) => boolean\n}\n\n/** Configure transport once; supply native state, questions, and model on every invocation. */\nexport function systemOneEvaluator(options: SystemOneEvaluatorOptions) {\n  const config = { ...options, pricing: options.pricing && { ...options.pricing } }\n  const run = createEvaluator<SystemOneRequest, unknown>({\n    execute: (request, context) => config.evaluate(structuredClone(request), context),\n    model: (request) => request.model,\n    costLedger: config.costLedger,\n    maximumCharge: config.maximumCharge,\n    receiptFromError: config.receiptFromError,\n    receipt:\n      config.receipt ??\n      ((raw) => {\n        const usage = systemOneUsage(raw)\n        const { model } = raw as SystemOneResult\n        return {\n          model,\n          inputTokens: usage.input_tokens,\n          outputTokens: usage.output_tokens,\n          customTokenPricing: config.pricing,\n        }\n      }),\n    validate: (raw, request) => {\n      const result = parseSystemOneResult(raw, request)\n      if (config.acceptModel && !config.acceptModel(request.model, result.model)) {\n        throw new Error('Evaluation served a model rejected by the caller policy')\n      }\n    },\n  })\n  return async <const Q extends SystemOneQuestions>(\n    request: SystemOneRequest<Q>,\n    context?: EvaluationContext,\n  ): Promise<EvaluationResult<SystemOneResult<Q>>> => {\n    context?.signal?.throwIfAborted()\n    parseSystemOneRequest(request)\n    const snapshot = structuredClone(request)\n    const result = await run(snapshot, context)\n    // The validator checked these exact questions after recording the paid work.\n    return { ...result, value: result.value as SystemOneResult<Q> }\n  }\n}\n\ntype Input<A, S extends Scenario> = { artifact: A; scenario: S }\ntype QuestionSource<I, Q extends SystemOneQuestions> = Q | ((input: I) => Q | Promise<Q>)\nexport interface SystemOneOptions extends SystemOneEvaluatorOptions {\n  model: string\n  version: string\n  questions: SystemOneQuestions\n}\nexport interface SystemOneJudgeOptions<\n  A,\n  S extends Scenario = Scenario,\n  Q extends SystemOneQuestions = SystemOneQuestions,\n> extends SystemOneEvaluatorOptions {\n  model: string\n  version: string\n  questions: QuestionSource<Input<A, S>, Q>\n  renderState: (\n    input: Input<A, S>,\n    context: EvaluationContext,\n  ) => SystemOneState | Promise<SystemOneState>\n  dimensions?: JudgeConfig<A, S>['dimensions']\n  weights?: Record<string, number>\n  map?: (value: SystemOneResult<Q>, input: Input<A, S>) => JudgeScore | Promise<JudgeScore>\n  record?: (\n    result: EvaluationResult<SystemOneResult<Q>>,\n    input: Input<A, S>,\n  ) => void | Promise<void>\n  appliesTo?: (scenario: S) => boolean\n}\n\n/** Canonical identity also retains native question/alternative order; dispatch stays untouched. */\nfunction questionIdentity(questions: SystemOneQuestions) {\n  return {\n    document: jsonDocument(questions),\n    order: Object.entries(questions).map(([name, question]) => [\n      name,\n      question.type === 'choice' ? Object.keys(question.criteria) : null,\n    ]),\n  }\n}\n\nfunction validateWeights(weights: Record<string, number>, keys: string[]): void {\n  if (\n    Object.entries(weights).some(\n      ([key, weight]) => !keys.includes(key) || !Number.isFinite(weight) || weight < 0,\n    ) ||\n    Object.values(weights).reduce((sum, weight) => sum + weight, 0) <= 0\n  ) {\n    throw new TypeError('Invalid judge weights')\n  }\n}\n\n/** Optional convention. Choice utilities and non-uniform scales belong in an explicit map. */\nexport function normalizedSystemOneScore(\n  result: SystemOneResult,\n  weights?: Record<string, number>,\n): JudgeScore {\n  const dimensions: Record<string, number> = Object.create(null)\n  const distribution: NonNullable<JudgeScore['distribution']> = Object.create(null)\n  for (const [key, answer] of Object.entries(result.answers)) {\n    if (answer.type === 'choice') throw new TypeError('Choice scoring requires an explicit map')\n    if (answer.type === 'noul') {\n      dimensions[key] = answer.noul\n      distribution[key] = [\n        { score: 0, probability: 1 - answer.noul },\n        { score: 1, probability: answer.noul },\n      ]\n    } else {\n      const scale = Object.keys(answer.legend).length - 1\n      dimensions[key] = answer.score / scale\n      distribution[key] = Object.entries(answer.probabilities).map(([level, probability]) => ({\n        score: Number(level) / scale,\n        probability,\n      }))\n    }\n  }\n  const effective = weights ?? Object.fromEntries(Object.keys(dimensions).map((key) => [key, 1]))\n  validateWeights(effective, Object.keys(dimensions))\n  return {\n    dimensions,\n    composite: weightedComposite({ dims: dimensions, weights: effective }).composite,\n    notes: '',\n    scoringMethod: 'expectation',\n    distribution,\n  }\n}\n\n/** Convenience around the public evaluator. No domain questions, thresholds, or prose. */\nexport function systemOneJudge<\n  A,\n  S extends Scenario = Scenario,\n  Q extends SystemOneQuestions = SystemOneQuestions,\n>(name: string, options: SystemOneJudgeOptions<A, S, Q>): JudgeConfig<A, S> {\n  const config = { ...options }\n  const source =\n    typeof options.questions === 'function' ? options.questions : structuredClone(options.questions)\n  const weights = options.weights && { ...options.weights }\n  if (!name.trim() || !options.version.trim()) {\n    throw new TypeError('Judge name and version are required')\n  }\n  if ((typeof source === 'function' || config.map) && !config.dimensions) {\n    throw new TypeError('Dynamic questions and custom mappings require stable judge dimensions')\n  }\n  if (!config.map && typeof source !== 'function') {\n    parseSystemOneQuestions(source)\n    if (Object.values(source).some((question) => question.type === 'choice')) {\n      throw new TypeError(\n        'Choice scoring requires an explicit map; labels have no numeric ordering',\n      )\n    }\n    if (weights) validateWeights(weights, Object.keys(source))\n  }\n  const dimensions =\n    config.dimensions ??\n    Object.entries(source).map(([key, question]) => ({\n      key,\n      description: typeof question.instructions === 'string' ? question.instructions : key,\n    }))\n  const evaluate = systemOneEvaluator(config)\n  return asJudge<A, S, SystemOneResult<Q>>({\n    name,\n    version: contentHash({\n      model: config.model,\n      version: config.version,\n      questions: typeof source === 'function' ? 'dynamic' : questionIdentity(source),\n      dimensions,\n      // Hash the JSON document: optional undefined fields are absent on the wire.\n      ...(weights ? { weights } : {}),\n    }),\n    dimensions,\n    appliesTo: config.appliesTo,\n    evaluate: async (input, context = {}) =>\n      evaluate(\n        {\n          model: config.model,\n          state: await config.renderState(input, context),\n          questions: typeof source === 'function' ? await source(input) : source,\n        },\n        context,\n      ),\n    map: config.map ?? ((value) => normalizedSystemOneScore(value, weights)),\n    record: config.record,\n  })\n}\n\nexport interface SystemOneAnalystOptions<I, Q extends SystemOneQuestions = SystemOneQuestions>\n  extends SystemOneEvaluatorOptions {\n  id: string\n  description: string\n  inputKind: Analyst<I>['inputKind']\n  model: string\n  version: string\n  questions: Q | ((input: I, context: AnalystContext) => Q | Promise<Q>)\n  renderState: (input: I, context: AnalystContext) => SystemOneState | Promise<SystemOneState>\n  findings: (\n    value: SystemOneResult<Q>,\n    input: I,\n    context: AnalystContext,\n  ) => AnalystFinding[] | Promise<AnalystFinding[]>\n  record?: (\n    result: EvaluationResult<SystemOneResult<Q>>,\n    input: I,\n    context: AnalystContext,\n  ) => void | Promise<void>\n}\n\nexport function systemOneAnalyst<I, Q extends SystemOneQuestions = SystemOneQuestions>(\n  options: SystemOneAnalystOptions<I, Q>,\n): Analyst<I> {\n  const config = { ...options }\n  const source =\n    typeof options.questions === 'function' ? options.questions : structuredClone(options.questions)\n  if (!config.id.trim() || !config.version.trim() || !config.description.trim()) {\n    throw new TypeError('Analyst id, version, and description are required')\n  }\n  const evaluate = systemOneEvaluator(config)\n  return asAnalyst<I, SystemOneResult<Q>>({\n    id: config.id,\n    description: config.description,\n    inputKind: config.inputKind,\n    version: contentHash({\n      model: config.model,\n      version: config.version,\n      questions: typeof source === 'function' ? 'dynamic' : questionIdentity(source),\n    }),\n    cost: { kind: 'llm', models: [config.model] },\n    evaluate: async (input, context, analystContext) =>\n      evaluate(\n        {\n          model: config.model,\n          state: await config.renderState(input, analystContext),\n          questions: typeof source === 'function' ? await source(input, analystContext) : source,\n        },\n        context,\n      ),\n    map: config.findings,\n    record: config.record,\n  })\n}\n"],"mappings":";;;;;;;;AAsCA,SAAgB,mBAAmB,SAAoC;CACrE,MAAM,SAAS;EAAE,GAAG;EAAS,SAAS,QAAQ,WAAW,EAAE,GAAG,QAAQ,QAAQ;CAAE;CAChF,MAAM,MAAM,gBAA2C;EACrD,UAAU,SAAS,YAAY,OAAO,SAAS,gBAAgB,OAAO,GAAG,OAAO;EAChF,QAAQ,YAAY,QAAQ;EAC5B,YAAY,OAAO;EACnB,eAAe,OAAO;EACtB,kBAAkB,OAAO;EACzB,SACE,OAAO,aACL,QAAQ;GACR,MAAM,QAAQ,eAAe,GAAG;GAChC,MAAM,EAAE,UAAU;GAClB,OAAO;IACL;IACA,aAAa,MAAM;IACnB,cAAc,MAAM;IACpB,oBAAoB,OAAO;GAC7B;EACF;EACF,WAAW,KAAK,YAAY;GAC1B,MAAM,SAAS,qBAAqB,KAAK,OAAO;GAChD,IAAI,OAAO,eAAe,CAAC,OAAO,YAAY,QAAQ,OAAO,OAAO,KAAK,GACvE,MAAM,IAAI,MAAM,yDAAyD;EAE7E;CACF,CAAC;CACD,OAAO,OACL,SACA,YACkD;EAClD,SAAS,QAAQ,eAAe;EAChC,sBAAsB,OAAO;EAC7B,MAAM,WAAW,gBAAgB,OAAO;EACxC,MAAM,SAAS,MAAM,IAAI,UAAU,OAAO;EAE1C,OAAO;GAAE,GAAG;GAAQ,OAAO,OAAO;EAA4B;CAChE;AACF;;AAgCA,SAAS,iBAAiB,WAA+B;CACvD,OAAO;EACL,UAAU,aAAa,SAAS;EAChC,OAAO,OAAO,QAAQ,SAAS,CAAC,CAAC,KAAK,CAAC,MAAM,cAAc,CACzD,MACA,SAAS,SAAS,WAAW,OAAO,KAAK,SAAS,QAAQ,IAAI,IAChE,CAAC;CACH;AACF;AAEA,SAAS,gBAAgB,SAAiC,MAAsB;CAC9E,IACE,OAAO,QAAQ,OAAO,CAAC,CAAC,MACrB,CAAC,KAAK,YAAY,CAAC,KAAK,SAAS,GAAG,KAAK,CAAC,OAAO,SAAS,MAAM,KAAK,SAAS,CACjF,KACA,OAAO,OAAO,OAAO,CAAC,CAAC,QAAQ,KAAK,WAAW,MAAM,QAAQ,CAAC,KAAK,GAEnE,MAAM,IAAI,UAAU,uBAAuB;AAE/C;;AAGA,SAAgB,yBACd,QACA,SACY;CACZ,MAAM,aAAqC,OAAO,OAAO,IAAI;CAC7D,MAAM,eAAwD,OAAO,OAAO,IAAI;CAChF,KAAK,MAAM,CAAC,KAAK,WAAW,OAAO,QAAQ,OAAO,OAAO,GAAG;EAC1D,IAAI,OAAO,SAAS,UAAU,MAAM,IAAI,UAAU,yCAAyC;EAC3F,IAAI,OAAO,SAAS,QAAQ;GAC1B,WAAW,OAAO,OAAO;GACzB,aAAa,OAAO,CAClB;IAAE,OAAO;IAAG,aAAa,IAAI,OAAO;GAAK,GACzC;IAAE,OAAO;IAAG,aAAa,OAAO;GAAK,CACvC;EACF,OAAO;GACL,MAAM,QAAQ,OAAO,KAAK,OAAO,MAAM,CAAC,CAAC,SAAS;GAClD,WAAW,OAAO,OAAO,QAAQ;GACjC,aAAa,OAAO,OAAO,QAAQ,OAAO,aAAa,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB;IACtF,OAAO,OAAO,KAAK,IAAI;IACvB;GACF,EAAE;EACJ;CACF;CACA,MAAM,YAAY,WAAW,OAAO,YAAY,OAAO,KAAK,UAAU,CAAC,CAAC,KAAK,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC;CAC9F,gBAAgB,WAAW,OAAO,KAAK,UAAU,CAAC;CAClD,OAAO;EACL;EACA,WAAW,kBAAkB;GAAE,MAAM;GAAY,SAAS;EAAU,CAAC,CAAC,CAAC;EACvE,OAAO;EACP,eAAe;EACf;CACF;AACF;;AAGA,SAAgB,eAId,MAAc,SAA4D;CAC1E,MAAM,SAAS,EAAE,GAAG,QAAQ;CAC5B,MAAM,SACJ,OAAO,QAAQ,cAAc,aAAa,QAAQ,YAAY,gBAAgB,QAAQ,SAAS;CACjG,MAAM,UAAU,QAAQ,WAAW,EAAE,GAAG,QAAQ,QAAQ;CACxD,IAAI,CAAC,KAAK,KAAK,KAAK,CAAC,QAAQ,QAAQ,KAAK,GACxC,MAAM,IAAI,UAAU,qCAAqC;CAE3D,KAAK,OAAO,WAAW,cAAc,OAAO,QAAQ,CAAC,OAAO,YAC1D,MAAM,IAAI,UAAU,uEAAuE;CAE7F,IAAI,CAAC,OAAO,OAAO,OAAO,WAAW,YAAY;EAC/C,wBAAwB,MAAM;EAC9B,IAAI,OAAO,OAAO,MAAM,CAAC,CAAC,MAAM,aAAa,SAAS,SAAS,QAAQ,GACrE,MAAM,IAAI,UACR,0EACF;EAEF,IAAI,SAAS,gBAAgB,SAAS,OAAO,KAAK,MAAM,CAAC;CAC3D;CACA,MAAM,aACJ,OAAO,cACP,OAAO,QAAQ,MAAM,CAAC,CAAC,KAAK,CAAC,KAAK,eAAe;EAC/C;EACA,aAAa,OAAO,SAAS,iBAAiB,WAAW,SAAS,eAAe;CACnF,EAAE;CACJ,MAAM,WAAW,mBAAmB,MAAM;CAC1C,OAAO,QAAkC;EACvC;EACA,SAAS,YAAY;GACnB,OAAO,OAAO;GACd,SAAS,OAAO;GAChB,WAAW,OAAO,WAAW,aAAa,YAAY,iBAAiB,MAAM;GAC7E;GAEA,GAAI,UAAU,EAAE,QAAQ,IAAI,CAAC;EAC/B,CAAC;EACD;EACA,WAAW,OAAO;EAClB,UAAU,OAAO,OAAO,UAAU,CAAC,MACjC,SACE;GACE,OAAO,OAAO;GACd,OAAO,MAAM,OAAO,YAAY,OAAO,OAAO;GAC9C,WAAW,OAAO,WAAW,aAAa,MAAM,OAAO,KAAK,IAAI;EAClE,GACA,OACF;EACF,KAAK,OAAO,SAAS,UAAU,yBAAyB,OAAO,OAAO;EACtE,QAAQ,OAAO;CACjB,CAAC;AACH;AAuBA,SAAgB,iBACd,SACY;CACZ,MAAM,SAAS,EAAE,GAAG,QAAQ;CAC5B,MAAM,SACJ,OAAO,QAAQ,cAAc,aAAa,QAAQ,YAAY,gBAAgB,QAAQ,SAAS;CACjG,IAAI,CAAC,OAAO,GAAG,KAAK,KAAK,CAAC,OAAO,QAAQ,KAAK,KAAK,CAAC,OAAO,YAAY,KAAK,GAC1E,MAAM,IAAI,UAAU,mDAAmD;CAEzE,MAAM,WAAW,mBAAmB,MAAM;CAC1C,OAAO,UAAiC;EACtC,IAAI,OAAO;EACX,aAAa,OAAO;EACpB,WAAW,OAAO;EAClB,SAAS,YAAY;GACnB,OAAO,OAAO;GACd,SAAS,OAAO;GAChB,WAAW,OAAO,WAAW,aAAa,YAAY,iBAAiB,MAAM;EAC/E,CAAC;EACD,MAAM;GAAE,MAAM;GAAO,QAAQ,CAAC,OAAO,KAAK;EAAE;EAC5C,UAAU,OAAO,OAAO,SAAS,mBAC/B,SACE;GACE,OAAO,OAAO;GACd,OAAO,MAAM,OAAO,YAAY,OAAO,cAAc;GACrD,WAAW,OAAO,WAAW,aAAa,MAAM,OAAO,OAAO,cAAc,IAAI;EAClF,GACA,OACF;EACF,KAAK,OAAO;EACZ,QAAQ,OAAO;CACjB,CAAC;AACH"}

package/docs/changes/jev-evidence-review.md

@@ -9,7 +9,7 @@ The optional `examples/jev-behavior-review.ts` recipe covers reward gaming, eval
 policy/authorization/data-boundary breaches, injection-following, contradicted completion,
 oversight bypass, matched-baseline anomalies and cross-run behavior. It composes the existing
 `jevAnalyst`; the taxonomy is not installed in the generic evaluator and is not a calibrated
-safety detector. See [behavioral reviews](../jev-behavior-review.md) for evidence and limits.
+safety detector. See [behavioral reviews](../systemone-behavior-review.md) for evidence and limits.
 
 No dependency, transport, scheduler, package entrypoint, default model, production action or
 spending policy changes. Existing evaluator, judge, analyst and protocol APIs remain unchanged.

package/docs/changes/systemone-provider-neutral.md

@@ -0,0 +1,36 @@
+# System One is provider-neutral
+
+The Jev surface is renamed to System One, because the same typed decision protocol is now served by
+TypeSafe (Jev), Cloudflare Workers AI (Clef, Clef-flash) and Perplexity (pplx-decider). The wire
+format does not change, so recorded requests and answers stay readable. There are no aliases: each
+consumer renames in the pull request that bumps its pin.
+
+| Before | After |
+| --- | --- |
+| `@tangle-network/agent-eval/jev`, `/jev/protocol` | `/systemone`, `/systemone/protocol` |
+| `Jev*` types (`JevRequest`, `JevResult`, `JevQuestions`, `JevRecordedRequest`, ...) | `SystemOne*` |
+| `parseJev{Request,RecordedRequest,Questions,Result}`, `jevUsage` | `parseSystemOne*`, `systemOneUsage` |
+| `JevResponseError` (message prefix `Jev:`) | `SystemOneResponseError` (prefix `System One:`) |
+| `jevEvaluator`, `jevJudge`, `jevAnalyst`, `normalizedJevScore` | `systemOneEvaluator`, `systemOneJudge`, `systemOneAnalyst`, `normalizedSystemOneScore` |
+| `prepareJevReview`, `assessJevReview`, `jevReviewFindings` | `prepareSystemOneReview`, `assessSystemOneReview`, `systemOneReviewFindings` |
+| `claimIntegrityJev{Questions,Review}`, `jevSignal`, signal id `jev`, version prefix `claim-integrity-jev-` | `claimIntegritySystemOne*`, `systemOneSignal`, id `systemone`, prefix `claim-integrity-systemone-` |
+
+Historical records keep their old signal ids and versions; a reader that filters on `jev` must also
+accept `systemone`.
+
+New in `/systemone/protocol`:
+
+- `normalizeSystemOneResult(raw, request, profile)` removes a provider's declared rounding before
+  the strict parser and gives every provider the same `confidence`. The route owner supplies the
+  profile (`{ decimals?, confidence: canonical | provider-defined }`). Answers that already pass
+  the strict checks are unchanged.
+- `systemOneConfidence(answer)` is that one definition, TypeSafe's published formulas.
+
+New in `/meta-eval`: `brierScore(pairs)`. Equal-mass ECE and the reliability table already come from
+`calibrationFromPairs(pairs, ..., { binning: equal-frequency, range: { lo: 0, hi: 1 } })`.
+
+Without normalization the strict parser refuses a recorded TypeSafe score answer whose rounded score
+is not the expectation of its rounded probabilities (8.79 against 8.86), so a transport that passes
+TypeSafe bytes straight to `parseSystemOneResult` should normalize with `{ decimals: 2, confidence:
+canonical }`. See [System One evaluations](../systemone.md#providers-rounding-and-confidence) for
+the recorded provider profiles.

package/docs/public-api.md

@@ -7,11 +7,11 @@ Generated by `pnpm api:census` on demand — this is a dated reading, not a gate
 | measure | count |
 | --- | --- |
 | export subpaths | 33 |
-| published value exports (subpath x symbol) | 1589 |
-| distinct symbols | 1401 |
-| production | 1048 |
-| planned | 278 |
-| none | 263 |
+| published value exports (subpath x symbol) | 1614 |
+| distinct symbols | 1422 |
+| production | 1052 |
+| planned | 296 |
+| none | 266 |
 
 Type-only exports are not listed: a type binds no runtime surface, and removing one cannot break a caller at run time.
 
@@ -84,7 +84,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 
 ### `.`
 
-360 value exports — 279 production, 40 planned, 41 none.
+371 value exports — 281 production, 44 planned, 46 none.
 
 | symbol | consumer | evidence |
 | --- | --- | --- |
@@ -150,6 +150,8 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `checkCanaries` | production | blueprint-agent:scripts/experiments/lib/contamination-preflight.ts |
 | `checkServedModel` | production | this package: src/integrity/preflight.ts:33 |
 | `checkTraceContracts` | production | creative-agent:eval/lib/trace-contracts.ts |
+| `chiSquared` | none | only this package's tests: src/statistics/inference.test.ts:2 |
+| `chiSquaredPValue` | none | only this package's tests: src/statistics/inference.test.ts:2 |
 | `clamp01` | production | agent-dev-container:products/intelligence/api/src/lib/consultant/executor.ts |
 | `classifyFailure` | production | phony:products/builder/api/src/eval/autoresearch/trace-analyst.ts |
 | `classifyFailureReason` | production | this package: src/diagnosis/first-failure.ts:39 |
@@ -246,6 +248,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `FileSystemTraceStore` | production | agent-builder:scripts/eval.ts |
 | `fileVerdictCache` | production | blueprint-agent:scripts/experiments/calibrate-instrument.ts |
 | `FindingsStore` | production | agent-builder:scripts/run-canonical-analyst-loop.ts |
+| `fisherExactTwoSided` | none | only this package's tests: src/statistics/inference.test.ts:2 |
 | `formatScorecardDiff` | production | ai-trading-blueprint:evals/src/trading/scorecard-integration.ts |
 | `gainHistogram` | production | blueprint-agent:scripts/experiments/vb-runrecord-analysis.ts |
 | `gateTreatmentApplied` | production | blueprint-agent:scripts/experiments/lib/validity-gates.ts |
@@ -286,6 +289,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `judgeSpans` | production | this package: src/builder-eval/three-layer-eval.ts:25 |
 | `knowledgeReadinessTracePayload` | production | agent-dev-container:products/intelligence/api/src/routes/project-engines.ts |
 | `leaderboard` | production | blueprint-agent:scripts/experiments/lib/cross-harness-leaderboard.ts |
+| `linearRegressionSlope` | none | only this package's tests: src/statistics/inference.test.ts:2 |
 | `lintTraceContractSpec` | planned | doc: docs/trace-contracts.md |
 | `LlmCallError` | production | agent-dev-container:products/intelligence/api/src/lib/router-model-owner.ts |
 | `llmJudge` | production | agent-runtime:examples/agentic-data-creation/offline-fixtures.ts |
@@ -314,6 +318,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `mulberry32` | production | agent-knowledge:src/memory/holdout.ts |
 | `MultiLayerVerifier` | production | agent-builder:src/lib/.server/eval/loops/forge-refinement.ts |
 | `NoopRawProviderSink` | production | agent-dev-container:products/intelligence/api/src/lib/eval-engine.ts |
+| `normalTwoSidedPValue` | none | only this package's tests: src/statistics/inference.test.ts:2 |
 | `notBlocked` | none | only this package's tests: src/oracle.test.ts:2 |
 | `NotFoundError` | production | agent-runtime:src/errors.ts |
 | `objectiveEval` | production | agent-knowledge:src/research-loop.ts |
@@ -398,6 +403,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `runSemanticConceptJudge` | production | agent-dev-container:products/intelligence/api/src/lib/optimization-engine.ts |
 | `runsForScenario` | production | phony:products/builder/api/src/eval/gate.ts |
 | `runTaskScore` | production | discovery-lab:tools/strict-screen.mjs |
+| `sampleMedian` | planned | doc: docs/concepts.md |
 | `scoreKnowledgeReadiness` | production | agent-builder:src/lib/.server/runtime/agent-task-readiness.ts |
 | `scorePolicyEditReadiness` | none | — |
 | `scoreRedTeamOutput` | production | agent-dev-container:products/intelligence/api/src/lib/optimization-gate.ts |
@@ -409,12 +415,15 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `ServedCrossFamilyError` | none | only this package's tests: src/integrity/served-model.test.ts:2 |
 | `servedModelAcceptable` | production | this package: src/integrity/preflight.ts:33 |
 | `shareAllowed` | planned | doc: docs/redaction.md |
+| `spanRecordsToTraceAnalysisStore` | planned | doc: docs/trace-analysis.md |
 | `spearmanR` | production | blueprint-agent:scripts/experiments/lib/benchmark-validity-report.mjs |
 | `stripFencedJson` | production | blueprint-agent:scripts/experiments/compare-analysis-reports.ts |
+| `studentTTwoSidedPValue` | production | this package: src/baseline.ts:19 |
 | `subjectiveEval` | production | creative-agent:eval/control/creative-onboarding.ts |
 | `summarizeBackendIntegrity` | production | agent-dev-container:products/intelligence/api/src/lib/run-provenance.ts |
 | `summarizeNumberSeries` | production | this package: src/campaign/cell-aggregates.ts:12 |
 | `summarizePreferenceMemory` | production | agent-builder:src/lib/.server/eval/loops/auto-research-runner.ts |
+| `summarizeSample` | planned | doc: docs/concepts.md |
 | `summaryTable` | production | blueprint-agent:scripts/experiments/vb-runrecord-analysis.ts |
 | `textInSnapshot` | none | only this package's tests: src/oracle.test.ts:2 |
 | `toAgentProfileJson` | production | blueprint-agent:scripts/experiments/lib/agent-profile-cell.ts |
@@ -425,6 +434,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `traceContract` | production | creative-agent:eval/lib/trace-contracts.ts |
 | `TraceEmitter` | production | agent-builder:src/lib/.server/runtime/trace-runtime.ts |
 | `transientDispatchFailure` | planned | doc: docs/eval-surface-map.md |
+| `twoByTwoPValue` | planned | doc: docs/concepts.md |
 | `UNKNOWN_MODEL` | production | this package: src/campaign/presets/run-profile-matrix.ts:54 |
 | `urlContains` | none | only this package's tests: src/oracle.test.ts:2 |
 | `userQuestionsForKnowledgeGaps` | production | agent-dev-container:products/intelligence/api/src/routes/project-engines.ts |
@@ -441,6 +451,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `viteDeployRunner` | production | blueprint-agent:scripts/experiments/lib/deploy-runner.ts |
 | `weightedComposite` | production | agent-app:src/eval/index.ts |
 | `weightedMean` | production | blueprint-agent:scripts/experiments/lib/synthesis-quality-judge.ts |
+| `welchsTTest` | production | this package: src/contract/analyze-runs.ts:25 |
 | `WILCOXON_EXACT_MAX_N` | none | — |
 | `wilcoxonSignedRank` | production | agent-builder:src/lib/.server/eval/loops/differential-eval.ts |
 | `wilson` | production | agent-dev-container:products/intelligence/api/src/lib/project-analysis-coverage-recommendations.ts |
@@ -489,7 +500,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `AnalystRegistry` | production | blueprint-agent:scripts/experiments/vb-multi-analyst.ts |
 | `analystUsageReceiptFromPrimeUsage` | production | this package: src/analyst/benchmark-runner-prime.ts:19 |
 | `appendVerificationArtifactsToOtlp` | production | this package: src/analyst/benchmark-public-data.ts:32 |
-| `askTraceQuestions` | planned | doc: docs/jev-behavior-review.md |
+| `askTraceQuestions` | planned | doc: docs/systemone-behavior-review.md |
 | `assertProposalFindings` | production | agent-runtime:src/improvement/code-execution.ts |
 | `behavioralAnalyst` | production | agent-dev-container:products/intelligence/api/src/lib/deep-trace-analyst.ts |
 | `bindAnalyst` | planned | doc: docs/trace-analysis.md |
@@ -644,7 +655,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 
 ### `./campaign`
 
-166 value exports — 117 production, 22 planned, 27 none.
+168 value exports — 119 production, 22 planned, 27 none.
 
 | symbol | consumer | evidence |
 | --- | --- | --- |
@@ -665,6 +676,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `buildTraceAnalystSurfaceDispatch` | none | only this package's tests: src/campaign/analyst-surface.test.ts:4 |
 | `campaignBreakdown` | production | this package: src/campaign/external-text-evaluation.ts:11 |
 | `campaignCellSearchResult` | production | this package: src/campaign/presets/run-optimization.ts:51 |
+| `campaignCoverage` | production | this package: src/campaign/presets/run-optimization.ts:32 |
 | `campaignMeanComposite` | production | agent-dev-container:products/intelligence/api/src/lib/optimization-engine.ts |
 | `campaignMeasurementDigest` | production | this package: src/contract/self-improve-method.ts:14 |
 | `campaignScenarioIdentity` | production | agent-runtime:src/improvement/method-identity.ts |
@@ -681,6 +693,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `costFromLedgerSummary` | production | supervisor-lab:bench/drain/seat.ts |
 | `createProfileMatrixPlan` | production | discovery-lab:tools/run-profile-confirmation.mjs |
 | `createReferenceEquivalenceJudge` | none | only this package's tests: src/reference-equivalence-judge.test.ts:13 |
+| `createRetainedDispatch` | production | this package: scripts/prove-retained-dispatch-integrity.ts:9 |
 | `createRunCostLedger` | production | agent-dev-container:products/intelligence/api/src/lib/eval-engine.ts |
 | `createSearchHistoryReceipt` | production | this package: src/campaign/search-ledger-recording.ts:27 |
 | `crowdedFrontierParent` | production | this package: scripts/search-sim.ts:146 |
@@ -713,13 +726,13 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `inMemoryCampaignStorage` | production | agent-dev-container:products/sandbox/evals/src/auto-optimization.ts |
 | `isProposedCandidate` | production | supervisor-lab:bench/sequential-arena.ts |
 | `isTransientTransportFailure` | production | gtm-agent:eval/suites/persona.ts |
-| `LabeledScenarioStoreError` | none | only this package's tests: tests/campaign/run-campaign.test.ts:6 |
+| `LabeledScenarioStoreError` | none | only this package's tests: tests/campaign/run-campaign.test.ts:7 |
 | `labelTrustRank` | production | this package: src/campaign/labeled-store/fs-adapter.ts:40 |
 | `llmJudge` | production | agent-runtime:examples/agentic-data-creation/offline-fixtures.ts |
 | `loadEvalFixture` | planned | doc: docs/eval-fixtures.md |
 | `loadEvalFixtureScenarios` | planned | example: examples/eval-fixtures-quickstart/index.ts:11 |
 | `loopProvenanceArgsFromResult` | production | this package: src/contract/self-improve.ts:36 |
-| `loopProvenanceSpans` | none | only this package's tests: tests/campaign/presets.test.ts:6 |
+| `loopProvenanceSpans` | none | only this package's tests: tests/campaign/presets.test.ts:7 |
 | `makePlaybackDispatch` | none | only this package's tests: src/campaign/presets/playback.test.ts:6 |
 | `makeProposalFinding` | production | agent-runtime:examples/improve/improve.ts |
 | `neutralizationGate` | planned | consumer tests: tax-agent:tests/eval/benchmarks/taxcalc/evolve.ts |
@@ -907,9 +920,9 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 
 | symbol | consumer | evidence |
 | --- | --- | --- |
-| `asAnalyst` | production | this package: src/jev.ts:5 |
-| `asJudge` | production | this package: src/jev.ts:5 |
-| `createEvaluator` | production | this package: src/jev.ts:5 |
+| `asAnalyst` | production | this package: src/systemone.ts:5 |
+| `asJudge` | production | this package: src/systemone.ts:5 |
+| `createEvaluator` | production | this package: src/systemone.ts:5 |
 
 ### `./experiment`
 
@@ -1061,40 +1074,6 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `TraceSpanEventSchema` | planned | example: examples/hosted-ingest-server/server.ts:33 |
 | `UnixNanoTimestampSchema` | none | — |
 
-### `./jev`
-
-15 value exports — 8 production, 6 planned, 1 none.
-
-| symbol | consumer | evidence |
-| --- | --- | --- |
-| `asAnalyst` | production | this package: src/jev.ts:5 |
-| `asJudge` | production | this package: src/jev.ts:5 |
-| `assessJevReview` | planned | doc: docs/changes/jev-evidence-review.md |
-| `createEvaluator` | production | this package: src/jev.ts:5 |
-| `jevAnalyst` | planned | example: examples/jev-behavior-review.ts:3 |
-| `jevEvaluator` | planned | doc: docs/jev-behavior-review.md |
-| `jevJudge` | planned | example: examples/jev-product-integrity.ts:4 |
-| `JevResponseError` | none | — |
-| `jevReviewFindings` | planned | example: examples/jev-behavior-review.ts:3 |
-| `jevUsage` | production | this package: src/jev.ts:7 |
-| `normalizedJevScore` | planned | doc: docs/jev.md |
-| `parseJevQuestions` | production | this package: src/jev.ts:7 |
-| `parseJevRequest` | production | this package: src/jev-review.ts:9 |
-| `parseJevResult` | production | this package: src/jev-review.ts:9 |
-| `prepareJevReview` | production | this package: src/rl/claim-integrity.ts:39 |
-
-### `./jev/protocol`
-
-5 value exports — 4 production, 0 planned, 1 none.
-
-| symbol | consumer | evidence |
-| --- | --- | --- |
-| `JevResponseError` | none | — |
-| `jevUsage` | production | this package: src/jev.ts:7 |
-| `parseJevQuestions` | production | this package: src/jev.ts:7 |
-| `parseJevRequest` | production | this package: src/jev-review.ts:9 |
-| `parseJevResult` | production | this package: src/jev-review.ts:9 |
-
 ### `./ledger-core`
 
 20 value exports — 19 production, 0 planned, 1 none.
@@ -1116,7 +1095,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `readTrustedHeadFile` | production | this package: src/ledger-core/journal.ts:55 |
 | `replayLedgerText` | production | this package: src/campaign/search-ledger.ts:23 |
 | `trustedHeadPathFor` | production | this package: src/ledger-core/journal.ts:55 |
-| `tryAcquireAtomicFileLock` | production | this package: src/campaign/single-run-lock.ts:17 |
+| `tryAcquireAtomicFileLock` | production | this package: scripts/prove-retained-dispatch.ts:10 |
 | `tryWithLedgerFileLock` | production | this package: src/campaign/search-ledger-file.ts:3 |
 | `verifyEntriesAgainstTrustedHead` | production | this package: src/ledger-core/journal.ts:55 |
 | `withLedgerFileLock` | production | this package: src/analyst/benchmark-response-cache.ts:6 |
@@ -1134,12 +1113,13 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 
 ### `./meta-eval`
 
-28 value exports — 13 production, 14 planned, 1 none.
+32 value exports — 13 production, 18 planned, 1 none.
 
 | symbol | consumer | evidence |
 | --- | --- | --- |
 | `auditEvaluator` | planned | example: examples/evaluation-integrity/index.ts:15 |
 | `auditProbabilityPolicy` | planned | doc: docs/changes/probability-policy-audit.md |
+| `brierScore` | planned | doc: docs/systemone.md |
 | `calibrateJudge` | production | creative-agent:eval/calibrate-judges.ts |
 | `calibrateJudgeContinuous` | planned | consumer tests: tax-agent:tests/eval/lib/judge-sentinel.ts |
 | `calibrationCurve` | planned | doc: docs/eval-surface-map.md |
@@ -1149,6 +1129,8 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `correlationStudy` | production | phony:products/builder/api/src/eval/outcomes.ts |
 | `definePlant` | planned | doc: docs/eval-surface-map.md |
 | `evalHealthStamp` | production | creative-agent:eval/lib/judge-sentinel.ts |
+| `evaluatorAdmissionPolicySchema` | planned | doc: docs/evaluator-construction.md |
+| `evaluatorAuditObservationSchema` | planned | doc: docs/evaluator-construction.md |
 | `fileSentinelStore` | production | creative-agent:eval/lib/judge-sentinel.ts |
 | `FileSystemOutcomeStore` | production | phony:products/builder/api/src/eval/outcomes.ts |
 | `InMemoryOutcomeStore` | planned | consumer tests: phony:products/builder/api/src/eval/outcomes-correlate.test.ts |
@@ -1156,6 +1138,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `judgeSentinelReport` | production | creative-agent:eval/lib/judge-sentinel.ts |
 | `OutcomeStoreError` | planned | doc: docs/outcome-validity.md |
 | `perturbEvidence` | planned | doc: docs/plants.md |
+| `planEvaluatorAudit` | planned | doc: docs/evaluator-construction.md |
 | `plantByPerturbation` | planned | doc: docs/plants.md |
 | `positionalBias` | planned | doc: docs/concepts.md |
 | `rubricPredictiveValidity` | production | physim:apps/server/src/scripts/agent-eval/validate-rubrics.ts |
@@ -1294,14 +1277,14 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `buildRlDataset` | production | this package: src/rl/corpus.ts:24 |
 | `buildVerifiedFindingRow` | none | only this package's tests: src/rl/verified-findings-dataset.test.ts:5 |
 | `campaignToRunRecords` | production | agent-builder:src/lib/.server/eval/loops/auto-research-runner.ts |
-| `checkerReadSignal` | planned | doc: docs/jev-behavior-review.md |
+| `checkerReadSignal` | planned | doc: docs/systemone-behavior-review.md |
 | `CLAIM_INTEGRITY_VERSION` | none | — |
 | `claimBrief` | none | — |
 | `claimIntegrityDigest` | none | only this package's tests: src/rl/claim-integrity.test.ts:4 |
-| `claimIntegrityJevQuestions` | none | — |
-| `claimIntegrityJevReview` | planned | doc: docs/jev-behavior-review.md |
-| `claimIntegrityTraceQuestions` | planned | doc: docs/jev-behavior-review.md |
-| `claimIntegrityVerdict` | planned | doc: docs/jev-behavior-review.md |
+| `claimIntegritySystemOneQuestions` | none | — |
+| `claimIntegritySystemOneReview` | planned | doc: docs/systemone-behavior-review.md |
+| `claimIntegrityTraceQuestions` | planned | doc: docs/systemone-behavior-review.md |
+| `claimIntegrityVerdict` | planned | doc: docs/systemone-behavior-review.md |
 | `compareAdaptationCurves` | planned | doc: docs/design/statistics-decisions.md |
 | `datasheetToMarkdown` | none | only this package's tests: src/rl/dataset.test.ts:7 |
 | `defaultBehaviorFeatures` | none | only this package's tests: src/rl/sim-fidelity.test.ts:3 |
@@ -1312,29 +1295,28 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `extractStepRewards` | none | only this package's tests: tests/rl-process-reward.test.ts:3 |
 | `extractVerifiableReward` | planned | named in gtm-agent (bind not in the import graph) |
 | `extractVerifiableRewardsFromRecords` | production | creative-agent:eval/canonical-export.ts |
-| `farmingSignal` | planned | doc: docs/jev-behavior-review.md |
+| `farmingSignal` | planned | doc: docs/systemone-behavior-review.md |
 | `FileSystemOutcomeStore` | production | phony:products/builder/api/src/eval/outcomes.ts |
 | `filterDeterministicallyRewarded` | production | this package: src/rl/reward-hacking.ts:43 |
 | `firstPassK` | planned | doc: docs/statistical-evidence.md |
 | `fitBradleyTerry` | none | only this package's tests: tests/rl-tournament.test.ts:2 |
 | `GRADER_REFERENCE_PHRASES` | none | — |
-| `graderReferenceSignal` | planned | doc: docs/jev-behavior-review.md |
+| `graderReferenceSignal` | planned | doc: docs/systemone-behavior-review.md |
 | `injectIrrelevantClause` | none | only this package's tests: tests/rl-contamination.test.ts:2 |
 | `InMemoryOutcomeStore` | planned | consumer tests: phony:products/builder/api/src/eval/outcomes-correlate.test.ts |
 | `inverseProbabilityWeighting` | planned | consumer tests: agent-knowledge:tests/memory-holdout.test.ts |
-| `jevSignal` | planned | doc: docs/jev-behavior-review.md |
 | `jsDivergence` | none | only this package's tests: src/rl/sim-fidelity.test.ts:3 |
 | `loadVerifiedFindingsDataset` | planned | doc: docs/verified-labels-flywheel.md |
-| `monitorSignal` | planned | doc: docs/jev-behavior-review.md |
+| `monitorSignal` | planned | doc: docs/systemone-behavior-review.md |
 | `offPolicyEstimateAll` | none | only this package's tests: tests/rl-off-policy.test.ts:3 |
 | `paretoFrontier` | production | agent-dev-container:products/sandbox/evals/run.ts |
-| `parseSourceScopeReferee` | planned | doc: docs/jev-behavior-review.md |
+| `parseSourceScopeReferee` | planned | doc: docs/systemone-behavior-review.md |
 | `PredictiveValidityResearcher` | planned | doc: docs/design/mlbenchmarks-book-review.md |
 | `prmTrainingPairs` | planned | doc: examples/fine-tune-with-prime-rl/README.md |
 | `quantileEdges` | none | only this package's tests: src/rl/sim-fidelity.test.ts:3 |
 | `readCorpus` | planned | named in agent-runtime (bind not in the import graph) |
 | `REFEREE_VERDICTS` | none | — |
-| `refereeSignal` | planned | doc: docs/jev-behavior-review.md |
+| `refereeSignal` | planned | doc: docs/systemone-behavior-review.md |
 | `renameVariables` | none | only this package's tests: tests/rl-contamination.test.ts:2 |
 | `ruleHolds` | none | only this package's tests: src/rl/claim-integrity.test.ts:4 |
 | `runAdaptationCurve` | planned | doc: docs/statistical-evidence.md |
@@ -1343,14 +1325,15 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `runEvalCampaign` | production | agent-builder:scripts/eval.ts |
 | `runRLCampaign` | planned | doc: docs/outcome-validity.md |
 | `runwiseStepRewardSummary` | none | only this package's tests: tests/rl-process-reward.test.ts:3 |
-| `scopeSignals` | planned | doc: docs/jev-behavior-review.md |
+| `scopeSignals` | planned | doc: docs/systemone-behavior-review.md |
 | `selfConsistency` | none | only this package's tests: tests/rl-adversarial-and-compute.test.ts:2 |
 | `selfNormalizedImportanceWeighting` | planned | consumer tests: agent-knowledge:tests/memory-holdout.test.ts |
 | `shuffleOrder` | none | — |
 | `simFidelityReport` | none | only this package's tests: src/rl/sim-fidelity.test.ts:3 |
-| `sourceScopeRefereeMessages` | planned | doc: docs/jev-behavior-review.md |
+| `sourceScopeRefereeMessages` | planned | doc: docs/systemone-behavior-review.md |
 | `stepRewardsToJsonl` | none | only this package's tests: src/rollout/gate-properties.test.ts:38 |
 | `summarizeVerifiedFindings` | none | only this package's tests: src/rl/verified-findings-dataset.test.ts:5 |
+| `systemOneSignal` | planned | doc: docs/systemone-behavior-review.md |
 | `toAnthropicFormat` | production | blueprint-agent:scripts/experiments/lib/steering-dataset.ts |
 | `toDpoJsonl` | production | this package: src/rl/dataset.ts:27 |
 | `toDpoRows` | production | this package: src/rl/dataset.ts:27 |
@@ -1369,7 +1352,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 
 ### `./rollout`
 
-67 value exports — 42 production, 15 planned, 10 none.
+68 value exports — 42 production, 16 planned, 10 none.
 
 | symbol | consumer | evidence |
 | --- | --- | --- |
@@ -1401,13 +1384,14 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `gateGamedOutcome` | none | only this package's tests: src/rollout/gate-properties.test.ts:56 |
 | `HARBOR_IMPORT_GAP` | none | only this package's tests: src/rollout/interchange/harbor.test.ts:6 |
 | `isRealnessGated` | production | starter-foundry:src/lib/held-out-gate.ts |
-| `isRolloutLine` | production | this package: src/supervisor-run/rollout-nodes.ts:26 |
+| `isRolloutLine` | production | this package: src/supervisor-run/rollout-nodes.ts:27 |
 | `isTrainableSplit` | production | this package: src/rollout/release/hf-dataset.ts:24 |
 | `measureFormatGate` | production | this package: src/rollout/release/hf-dataset.ts:26 |
 | `mintRolloutRows` | production | agent-builder:src/lib/.server/eval/loops/auto-research-runner.ts |
 | `observedScore` | production | this package: src/rl/reward-hacking.ts:40 |
 | `observedSplitScore` | production | starter-foundry:src/lib/held-out-gate.ts |
 | `openOpencodeDb` | production | discovery-lab:dashboard/lib/transcript.mjs |
+| `parseClaudeTranscript` | planned | doc: docs/rollout.md |
 | `parseRolloutReleaseArgs` | none | only this package's tests: src/rollout/release/hf-dataset.test.ts:9 |
 | `planPushCommand` | none | only this package's tests: src/rollout/release/hf-dataset.test.ts:9 |
 | `readClaudeTranscript` | planned | doc: docs/rollout.md |
@@ -1430,7 +1414,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `toHarborTrajectory` | planned | doc: docs/rollout.md |
 | `toJsonl` | production | supervisor-lab:bench/vertical-rollout-package.ts |
 | `toRewardRows` | planned | doc: docs/rollout.md |
-| `toRftItem` | none | only this package's tests: src/rollout/reward-invariant.test.ts:33 |
+| `toRftItem` | none | only this package's tests: src/rollout/mint.test.ts:9 |
 | `toRftItems` | production | this package: src/rollout/release/hf-dataset.ts:22 |
 | `toSftRows` | production | this package: src/rl/dataset.ts:27 |
 | `toVerifiersRolloutOutput` | none | only this package's tests: src/rollout/exporters.test.ts:2 |
@@ -1539,6 +1523,46 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `writeSupervisorRunReport` | planned | named in agent-runtime (bind not in the import graph) |
 | `writeSupervisorRunReportSafe` | planned | named in agent-runtime (bind not in the import graph) |
 
+### `./systemone`
+
+18 value exports — 8 production, 10 planned, 0 none.
+
+| symbol | consumer | evidence |
+| --- | --- | --- |
+| `asAnalyst` | production | this package: src/systemone.ts:5 |
+| `asJudge` | production | this package: src/systemone.ts:5 |
+| `assessSystemOneReview` | planned | doc: docs/systemone-behavior-review.md |
+| `createEvaluator` | production | this package: src/systemone.ts:5 |
+| `normalizedSystemOneScore` | planned | doc: docs/systemone.md |
+| `normalizeSystemOneResult` | planned | doc: docs/systemone.md |
+| `parseSystemOneQuestions` | production | this package: src/systemone.ts:14 |
+| `parseSystemOneRecordedRequest` | planned | example: examples/systemone-replay.ts:7 |
+| `parseSystemOneRequest` | production | this package: src/systemone-review.ts:11 |
+| `parseSystemOneResult` | production | this package: src/systemone-review.ts:11 |
+| `prepareSystemOneReview` | production | this package: src/rl/claim-integrity.ts:47 |
+| `systemOneAnalyst` | planned | example: examples/systemone-behavior-review.ts:3 |
+| `systemOneConfidence` | planned | doc: docs/systemone.md |
+| `systemOneEvaluator` | planned | doc: docs/systemone-behavior-review.md |
+| `systemOneJudge` | planned | example: examples/systemone-product-integrity.ts:4 |
+| `SystemOneResponseError` | planned | doc: docs/systemone.md |
+| `systemOneReviewFindings` | planned | example: examples/systemone-behavior-review.ts:3 |
+| `systemOneUsage` | production | this package: src/systemone.ts:14 |
+
+### `./systemone/protocol`
+
+8 value exports — 4 production, 4 planned, 0 none.
+
+| symbol | consumer | evidence |
+| --- | --- | --- |
+| `normalizeSystemOneResult` | planned | doc: docs/systemone.md |
+| `parseSystemOneQuestions` | production | this package: src/systemone.ts:14 |
+| `parseSystemOneRecordedRequest` | planned | example: examples/systemone-replay.ts:7 |
+| `parseSystemOneRequest` | production | this package: src/systemone-review.ts:11 |
+| `parseSystemOneResult` | production | this package: src/systemone-review.ts:11 |
+| `systemOneConfidence` | planned | doc: docs/systemone.md |
+| `SystemOneResponseError` | planned | doc: docs/systemone.md |
+| `systemOneUsage` | production | this package: src/systemone.ts:14 |
+
 ### `./trace-attributes`
 
 28 value exports — 26 production, 1 planned, 1 none.
@@ -1679,7 +1703,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 
 ### `./traces`
 
-124 value exports — 96 production, 8 planned, 20 none.
+125 value exports — 96 production, 9 planned, 20 none.
 
 | symbol | consumer | evidence |
 | --- | --- | --- |
@@ -1702,7 +1726,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `convertTraceStoresToOtlp` | planned | consumer tests: legal-agent:tests/eval/lib/traces-to-otlp.ts |
 | `createBoundedTraceAnalysisStore` | production | braid:src/adapters/analysis/trace-store.ts |
 | `DEFAULT_TRACE_ANALYST_BUDGETS` | production | agent-builder:src/lib/.server/eval/stores/d1-trace-analysis-store-adapter.ts |
-| `defaultProviderRedactor` | production | this package: src/llm-client.ts:53 |
+| `defaultProviderRedactor` | production | this package: src/llm-client.ts:54 |
 | `defaultTraceInsightPanel` | none | only this package's tests: src/trace-analyst/insights.test.ts:3 |
 | `describeTraceInsightScope` | production | blueprint-agent:scripts/experiments/lib/analyze-vb-run/report-html.ts |
 | `detectCredential` | production | this package: scripts/redaction-corpus.ts:9 |
@@ -1763,7 +1787,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `OUTPUT_VALUE` | production | agent-runtime:src/runtime/supervise-surface.ts |
 | `planTraceInsightQuestions` | none | only this package's tests: src/trace-analyst/insights.test.ts:3 |
 | `projectOtlpFlatLine` | production | this package: src/trace-analyst/otlp-to-run-records.ts:60 |
-| `providerFromBaseUrl` | production | this package: src/llm-client.ts:53 |
+| `providerFromBaseUrl` | production | this package: src/llm-client.ts:54 |
 | `redact` | production | this package: scripts/redaction-corpus.ts:9 |
 | `redactForShare` | production | this package: scripts/redaction-corpus.ts:9 |
 | `REDACTION_PROFILES` | production | this package: scripts/redaction-corpus.ts:9 |
@@ -1779,6 +1803,7 @@ The `none` set was reviewed symbol by symbol on 2026-08-21. Four rules decided m
 | `shareAllowed` | planned | doc: docs/redaction.md |
 | `SPAN_KIND_ATTR_KEYS` | production | traces:src/run-span-tree.ts |
 | `SpanNotFoundError` | production | this package: src/trace-analyst/store-boundary.ts:1 |
+| `spanRecordsToTraceAnalysisStore` | planned | doc: docs/trace-analysis.md |
 | `stringField` | planned | named in agent-dev-container (bind not in the import graph) |
 | `throwIfRunIncomplete` | none | only this package's tests: tests/run-integrity.test.ts:3 |
 | `tokenizeDomainWords` | planned | consumer tests: blueprint-agent:scripts/experiments/__tests__/analyze-vb-run.test.ts |

package/docs/rollout.md

@@ -92,7 +92,7 @@ Validation preserves source order and rejects duplicate pairs or missing ordinal
 | the gate checks themselves | `src/rollout/gate-checks.ts` | `GATE_CHECK_IDS`, `GATE_CHECKS`, `GATE_POLICIES`, `gateErrors(subject, policy)` — the one list every entry point draws from. The subject is the LINE (`{outcome, steps}`), not the outcome alone: a per-step reward is training signal and lives outside `outcome`. `readReward` / `payloadIsPopulated` are the total readers every check narrows through, so a value the gate cannot classify is refused rather than read as clean |
 | ledger file API | `src/rollout/ledger.ts` | `writeRolloutLedger`, `appendRolloutLines`, `readRolloutLedger` |
 | minting from records | `src/rollout/mint.ts` | `mintRolloutRows(records, traceStore)`: RunRecord joined to trace via shared `runId`. `unmintableReasons(record)`: the same door's refusals, reported without throwing |
-| harness-store intake | `src/rollout/readers/` | `openOpencodeDb` + `readOpencodeSessionMessages` (opencode sqlite), `findClaudeTranscripts` + `readClaudeTranscript` (Claude Code project jsonl) |
+| harness-store intake | `src/rollout/readers/` | `openOpencodeDb` + `readOpencodeSessionMessages` (opencode sqlite), `findClaudeTranscripts` + `readClaudeTranscript` (Claude Code project JSONL), `parseClaudeTranscript` (retained JSONL text) |
 | interchange | `src/rollout/interchange/harbor.ts` | `toHarborTrajectory` / `toHarborTrajectories` / `fromHarborTrajectory` / `relabelImportedSplit` (Harbor ATIF-v1.7); all root-exported as well as on the `/rollout` subpath |
 | exporters | `src/rollout/exporters.ts` | `toSftRows`, `toRewardRows`, `toVerifiersRolloutOutputs` (Prime Intellect), `toRftItems` (OpenAI RFT), `toJsonl` |
 | RL exporters | `src/rl/exporters.ts` | `toDpoRows`, `toGrpoRows`, `toSftRows`, `toPrmRows`; all training inputs are `MintedRolloutLine[]` or line-referenced artifacts with explicit line context |
@@ -200,3 +200,11 @@ A step carrying none of the four produces no span — absent stays absent.
 Letta's trajectory-v1 is a strict subset of what we need from ATIF for this purpose: no per-step or aggregate cost, no multi-agent/subagent structure, and no token-id or logprob channel.
 A Letta sink would therefore carry strictly less than the ATIF one while adding a second format to keep correct against schema drift.
 Do not add one without a concrete consumer that reads Letta and cannot read ATIF.
+
+### Read retained Claude transcripts
+
+Use `parseClaudeTranscript(raw, options)` from `@tangle-network/agent-eval/rollout` for JSONL already resolved from a retained artifact.
+It returns the same messages, tool calls/results, usage and source gaps as `readClaudeTranscript(path, options)`.
+Set `includeSidechain: true` when reading a native subagent transcript.
+The caller retains the original bytes and verifies their artifact identity before parsing.
+This parser does not read CAS, store another copy, redact content or establish capture completeness.