@tangle-network/agent-eval 0.205.2 → 0.206.0

package/dist/systemone.d.ts

@@ -0,0 +1,60 @@
+import { b as CustomTokenPricing, g as CostReceiptInput } from "./cost-ledger-DbQdN3nO.js";
+import { i as AnalystFinding, n as AnalystContext, t as Analyst } from "./types-6VEehmEg.js";
+import { R as Scenario, S as JudgeConfig, w as JudgeScore } from "./types-ZUy5HGY5.js";
+import { EvaluationAnalystOptions, EvaluationContext, EvaluationJudgeOptions, EvaluationResult, Evaluator, EvaluatorOptions, asAnalyst, asJudge, createEvaluator } from "./evaluation.js";
+import { S as systemOneUsage, _ as parseSystemOneQuestions, a as SystemOneQuestion, b as parseSystemOneResult, c as SystemOneRecordedQuestions, d as SystemOneResponseError, f as SystemOneResult, g as normalizeSystemOneResult, h as SystemOneState, i as SystemOneJson, l as SystemOneRecordedRequest, m as SystemOneScoreCriteria, n as SystemOneAnswerFor, o as SystemOneQuestions, p as SystemOneResultProfile, r as SystemOneDescription, s as SystemOneRecordedQuestion, t as SystemOneAnswer, u as SystemOneRequest, v as parseSystemOneRecordedRequest, x as systemOneConfidence, y as parseSystemOneRequest } from "./systemone-protocol-DKoi1NdK.js";
+import { a as SystemOneReviewReport, c as systemOneReviewFindings, i as SystemOneReviewInput, n as SystemOneReviewAssessment, o as assessSystemOneReview, r as SystemOneReviewCheck, s as prepareSystemOneReview, t as PreparedSystemOneReview } from "./systemone-review-DGTKkLMG.js";
+//#region src/systemone.d.ts
+export type SystemOneEvaluate = EvaluatorOptions<SystemOneRequest, unknown>['execute'];
+export interface SystemOneEvaluatorOptions {
+  evaluate: SystemOneEvaluate;
+  costLedger?: EvaluatorOptions<SystemOneRequest, unknown>['costLedger'];
+  maximumCharge?: EvaluatorOptions<SystemOneRequest, unknown>['maximumCharge'];
+  pricing?: CustomTokenPricing;
+  receipt?: (response: unknown) => CostReceiptInput;
+  receiptFromError?: (error: Error) => CostReceiptInput | undefined;
+  /** Optional deployment policy; the protocol adapter does not hardcode model aliases. */
+  acceptModel?: (requested: string, served: string) => boolean;
+}
+/** Configure transport once; supply native state, questions, and model on every invocation. */
+export declare function systemOneEvaluator(options: SystemOneEvaluatorOptions): <const Q extends SystemOneQuestions>(request: SystemOneRequest<Q>, context?: EvaluationContext) => Promise<EvaluationResult<SystemOneResult<Q>>>;
+type Input<A, S extends Scenario> = {
+  artifact: A;
+  scenario: S;
+};
+type QuestionSource<I, Q extends SystemOneQuestions> = Q | ((input: I) => Q | Promise<Q>);
+export interface SystemOneOptions extends SystemOneEvaluatorOptions {
+  model: string;
+  version: string;
+  questions: SystemOneQuestions;
+}
+export interface SystemOneJudgeOptions<A, S extends Scenario = Scenario, Q extends SystemOneQuestions = SystemOneQuestions> extends SystemOneEvaluatorOptions {
+  model: string;
+  version: string;
+  questions: QuestionSource<Input<A, S>, Q>;
+  renderState: (input: Input<A, S>, context: EvaluationContext) => SystemOneState | Promise<SystemOneState>;
+  dimensions?: JudgeConfig<A, S>['dimensions'];
+  weights?: Record<string, number>;
+  map?: (value: SystemOneResult<Q>, input: Input<A, S>) => JudgeScore | Promise<JudgeScore>;
+  record?: (result: EvaluationResult<SystemOneResult<Q>>, input: Input<A, S>) => void | Promise<void>;
+  appliesTo?: (scenario: S) => boolean;
+}
+/** Optional convention. Choice utilities and non-uniform scales belong in an explicit map. */
+export declare function normalizedSystemOneScore(result: SystemOneResult, weights?: Record<string, number>): JudgeScore;
+/** Convenience around the public evaluator. No domain questions, thresholds, or prose. */
+export declare function systemOneJudge<A, S extends Scenario = Scenario, Q extends SystemOneQuestions = SystemOneQuestions>(name: string, options: SystemOneJudgeOptions<A, S, Q>): JudgeConfig<A, S>;
+export interface SystemOneAnalystOptions<I, Q extends SystemOneQuestions = SystemOneQuestions> extends SystemOneEvaluatorOptions {
+  id: string;
+  description: string;
+  inputKind: Analyst<I>['inputKind'];
+  model: string;
+  version: string;
+  questions: Q | ((input: I, context: AnalystContext) => Q | Promise<Q>);
+  renderState: (input: I, context: AnalystContext) => SystemOneState | Promise<SystemOneState>;
+  findings: (value: SystemOneResult<Q>, input: I, context: AnalystContext) => AnalystFinding[] | Promise<AnalystFinding[]>;
+  record?: (result: EvaluationResult<SystemOneResult<Q>>, input: I, context: AnalystContext) => void | Promise<void>;
+}
+export declare function systemOneAnalyst<I, Q extends SystemOneQuestions = SystemOneQuestions>(options: SystemOneAnalystOptions<I, Q>): Analyst<I>;
+//#endregion
+export { EvaluationAnalystOptions, EvaluationContext, EvaluationJudgeOptions, EvaluationResult, Evaluator, EvaluatorOptions, PreparedSystemOneReview, SystemOneAnswer, SystemOneAnswerFor, SystemOneDescription, SystemOneJson, SystemOneQuestion, SystemOneQuestions, SystemOneRecordedQuestion, SystemOneRecordedQuestions, SystemOneRecordedRequest, SystemOneRequest, SystemOneResponseError, SystemOneResult, SystemOneResultProfile, SystemOneReviewAssessment, SystemOneReviewCheck, SystemOneReviewInput, SystemOneReviewReport, SystemOneScoreCriteria, SystemOneState, asAnalyst, asJudge, assessSystemOneReview, createEvaluator, normalizeSystemOneResult, parseSystemOneQuestions, parseSystemOneRecordedRequest, parseSystemOneRequest, parseSystemOneResult, prepareSystemOneReview, systemOneConfidence, systemOneReviewFindings, systemOneUsage };
+//# sourceMappingURL=systemone.d.ts.map

package/dist/systemone.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"systemone.d.ts","names":[],"sources":["../src/systemone.ts"],"mappings":";;;;;;;YAyBY,oBAAoB,iBAAiB;iBAChC;EACf,UAAU;EACV,aAAa,iBAAiB;EAC9B,gBAAgB,iBAAiB;EACjC,UAAU;EACV,WAAW,sBAAsB;EACjC,oBAAoB,OAAO,UAAU;;EAErC,eAAe,mBAAmB;;;wBAIpB,mBAAmB,SAAS,mCA2BtB,UAAU,oBAAkB,SACrC,iBAAiB,IAAE,UAClB,sBACT,QAAQ,iBAAiB,gBAAgB;KAUzC,MAAM,GAAG,UAAU;EAAc,UAAU;EAAG,UAAU;;KACxD,eAAe,GAAG,UAAU,sBAAsB,MAAM,OAAO,MAAM,IAAI,QAAQ;iBACrE,yBAAyB;EACxC;EACA;EACA,WAAW;;iBAEI,sBACf,GACA,UAAU,WAAW,UACrB,UAAU,qBAAqB,4BACvB;EACR;EACA;EACA,WAAW,eAAe,MAAM,GAAG,IAAI;EACvC,cACE,OAAO,MAAM,GAAG,IAChB,SAAS,sBACN,iBAAiB,QAAQ;EAC9B,aAAa,YAAY,GAAG;EAC5B,UAAU;EACV,OAAO,OAAO,gBAAgB,IAAI,OAAO,MAAM,GAAG,OAAO,aAAa,QAAQ;EAC9E,UACE,QAAQ,iBAAiB,gBAAgB,KACzC,OAAO,MAAM,GAAG,cACN;EACZ,aAAa,UAAU;;;wBA0BT,yBACd,QAAQ,iBACR,UAAU,yBACT;;wBAgCa,eACd,GACA,UAAU,WAAW,UACrB,UAAU,qBAAqB,oBAC/B,cAAc,SAAS,sBAAsB,GAAG,GAAG,KAAK,YAAY,GAAG;iBAqDxD,wBAAwB,GAAG,UAAU,qBAAqB,4BACjE;EACR;EACA;EACA,WAAW,QAAQ;EACnB;EACA;EACA,WAAW,MAAM,OAAO,GAAG,SAAS,mBAAmB,IAAI,QAAQ;EACnE,cAAc,OAAO,GAAG,SAAS,mBAAmB,iBAAiB,QAAQ;EAC7E,WACE,OAAO,gBAAgB,IACvB,OAAO,GACP,SAAS,mBACN,mBAAmB,QAAQ;EAChC,UACE,QAAQ,iBAAiB,gBAAgB,KACzC,OAAO,GACP,SAAS,0BACC;;wBAGE,iBAAiB,GAAG,UAAU,qBAAqB,oBACjE,SAAS,wBAAwB,GAAG,KACnC,QAAQ"}

package/dist/{jev.js → systemone.js}

@@ -2,11 +2,11 @@ import { o as jsonDocument } from "./canonical-BsZVgBhW.js";
 import { s as weightedComposite } from "./descriptive-CnU8Yq29.js";
 import { n as contentHash } from "./verdict-cache-BQEsVy9n.js";
 import { asAnalyst, asJudge, createEvaluator } from "./evaluation.js";
-import { JevResponseError, jevUsage, parseJevQuestions, parseJevRecordedRequest, parseJevRequest, parseJevResult } from "./jev-protocol.js";
-import { n as jevReviewFindings, r as prepareJevReview, t as assessJevReview } from "./jev-review-DZEjDzlF.js";
-//#region src/jev.ts
+import { SystemOneResponseError, normalizeSystemOneResult, parseSystemOneQuestions, parseSystemOneRecordedRequest, parseSystemOneRequest, parseSystemOneResult, systemOneConfidence, systemOneUsage } from "./systemone-protocol.js";
+import { n as prepareSystemOneReview, r as systemOneReviewFindings, t as assessSystemOneReview } from "./systemone-review-DTClwgmE.js";
+//#region src/systemone.ts
 /** Configure transport once; supply native state, questions, and model on every invocation. */
-function jevEvaluator(options) {
+function systemOneEvaluator(options) {
 	const config = {
 		...options,
 		pricing: options.pricing && { ...options.pricing }
@@ -18,7 +18,7 @@ function jevEvaluator(options) {
 		maximumCharge: config.maximumCharge,
 		receiptFromError: config.receiptFromError,
 		receipt: config.receipt ?? ((raw) => {
-			const usage = jevUsage(raw);
+			const usage = systemOneUsage(raw);
 			const { model } = raw;
 			return {
 				model,
@@ -28,13 +28,13 @@ function jevEvaluator(options) {
 			};
 		}),
 		validate: (raw, request) => {
-			const result = parseJevResult(raw, request);
+			const result = parseSystemOneResult(raw, request);
 			if (config.acceptModel && !config.acceptModel(request.model, result.model)) throw new Error("Evaluation served a model rejected by the caller policy");
 		}
 	});
 	return async (request, context) => {
 		context?.signal?.throwIfAborted();
-		parseJevRequest(request);
+		parseSystemOneRequest(request);
 		const snapshot = structuredClone(request);
 		const result = await run(snapshot, context);
 		return {
@@ -54,7 +54,7 @@ function validateWeights(weights, keys) {
 	if (Object.entries(weights).some(([key, weight]) => !keys.includes(key) || !Number.isFinite(weight) || weight < 0) || Object.values(weights).reduce((sum, weight) => sum + weight, 0) <= 0) throw new TypeError("Invalid judge weights");
 }
 /** Optional convention. Choice utilities and non-uniform scales belong in an explicit map. */
-function normalizedJevScore(result, weights) {
+function normalizedSystemOneScore(result, weights) {
 	const dimensions = Object.create(null);
 	const distribution = Object.create(null);
 	for (const [key, answer] of Object.entries(result.answers)) {
@@ -91,14 +91,14 @@ function normalizedJevScore(result, weights) {
 	};
 }
 /** Convenience around the public evaluator. No domain questions, thresholds, or prose. */
-function jevJudge(name, options) {
+function systemOneJudge(name, options) {
 	const config = { ...options };
 	const source = typeof options.questions === "function" ? options.questions : structuredClone(options.questions);
 	const weights = options.weights && { ...options.weights };
 	if (!name.trim() || !options.version.trim()) throw new TypeError("Judge name and version are required");
 	if ((typeof source === "function" || config.map) && !config.dimensions) throw new TypeError("Dynamic questions and custom mappings require stable judge dimensions");
 	if (!config.map && typeof source !== "function") {
-		parseJevQuestions(source);
+		parseSystemOneQuestions(source);
 		if (Object.values(source).some((question) => question.type === "choice")) throw new TypeError("Choice scoring requires an explicit map; labels have no numeric ordering");
 		if (weights) validateWeights(weights, Object.keys(source));
 	}
@@ -106,7 +106,7 @@ function jevJudge(name, options) {
 		key,
 		description: typeof question.instructions === "string" ? question.instructions : key
 	}));
-	const evaluate = jevEvaluator(config);
+	const evaluate = systemOneEvaluator(config);
 	return asJudge({
 		name,
 		version: contentHash({
@@ -123,15 +123,15 @@ function jevJudge(name, options) {
 			state: await config.renderState(input, context),
 			questions: typeof source === "function" ? await source(input) : source
 		}, context),
-		map: config.map ?? ((value) => normalizedJevScore(value, weights)),
+		map: config.map ?? ((value) => normalizedSystemOneScore(value, weights)),
 		record: config.record
 	});
 }
-function jevAnalyst(options) {
+function systemOneAnalyst(options) {
 	const config = { ...options };
 	const source = typeof options.questions === "function" ? options.questions : structuredClone(options.questions);
 	if (!config.id.trim() || !config.version.trim() || !config.description.trim()) throw new TypeError("Analyst id, version, and description are required");
-	const evaluate = jevEvaluator(config);
+	const evaluate = systemOneEvaluator(config);
 	return asAnalyst({
 		id: config.id,
 		description: config.description,
@@ -155,6 +155,6 @@ function jevAnalyst(options) {
 	});
 }
 //#endregion
-export { JevResponseError, asAnalyst, asJudge, assessJevReview, createEvaluator, jevAnalyst, jevEvaluator, jevJudge, jevReviewFindings, jevUsage, normalizedJevScore, parseJevQuestions, parseJevRecordedRequest, parseJevRequest, parseJevResult, prepareJevReview };
+export { SystemOneResponseError, asAnalyst, asJudge, assessSystemOneReview, createEvaluator, normalizeSystemOneResult, normalizedSystemOneScore, parseSystemOneQuestions, parseSystemOneRecordedRequest, parseSystemOneRequest, parseSystemOneResult, prepareSystemOneReview, systemOneAnalyst, systemOneConfidence, systemOneEvaluator, systemOneJudge, systemOneReviewFindings, systemOneUsage };
 
-//# sourceMappingURL=jev.js.map
+//# sourceMappingURL=systemone.js.map

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.