vieval 0.0.9 → 0.0.11

package/dist/config.d.mts

@@ -1,2 +1,2 @@
-import { C as TaskDefinition, D as TaskRunContext, E as TaskReporterHooks, O as TaskRunOutput, R as ModelDefinition, S as TaskConcurrencyConfig, T as TaskReporterEventPayload, _ as ScopedMatrices, a as CliOpenTelemetryReportingConfig, b as TaskCaseReporterPayload, c as EvalDefinition, d as MatrixAxisValues, f as MatrixDefinition, g as MatrixValue, h as MatrixRow, i as Awaitable, l as EvalModule, m as MatrixPrimitive, n as defineEval, o as CliReportingConfig, p as MatrixLayer, r as defineTask, s as CollectedEvalEntry, t as ConfigHookPlugin, u as EvalModuleMap, v as TaskAutoRetryDelay, w as TaskExecutionPolicy, x as TaskCaseState, y as TaskCaseReporterEndPayload, z as resolveModelByName } from "./index-BkjyCInx.mjs";
+import { C as TaskDefinition, D as TaskRunContext, E as TaskReporterHooks, O as TaskRunOutput, R as ModelDefinition, S as TaskConcurrencyConfig, T as TaskReporterEventPayload, _ as ScopedMatrices, a as CliOpenTelemetryReportingConfig, b as TaskCaseReporterPayload, c as EvalDefinition, d as MatrixAxisValues, f as MatrixDefinition, g as MatrixValue, h as MatrixRow, i as Awaitable, l as EvalModule, m as MatrixPrimitive, n as defineEval, o as CliReportingConfig, p as MatrixLayer, r as defineTask, s as CollectedEvalEntry, t as ConfigHookPlugin, u as EvalModuleMap, v as TaskAutoRetryDelay, w as TaskExecutionPolicy, x as TaskCaseState, y as TaskCaseReporterEndPayload, z as resolveModelByName } from "./index-CwKBlCG9.mjs";
 export { Awaitable, CliOpenTelemetryReportingConfig, CliReportingConfig, CollectedEvalEntry, ConfigHookPlugin, EvalDefinition, EvalModule, EvalModuleMap, MatrixAxisValues, MatrixDefinition, MatrixLayer, MatrixPrimitive, MatrixRow, MatrixValue, ModelDefinition, ScopedMatrices, TaskAutoRetryDelay, TaskCaseReporterEndPayload, TaskCaseReporterPayload, TaskCaseState, TaskConcurrencyConfig, TaskDefinition, TaskExecutionPolicy, TaskReporterEventPayload, TaskReporterHooks, TaskRunContext, TaskRunOutput, defineEval, defineTask, resolveModelByName };

package/dist/config.mjs

@@ -1,4 +1,4 @@
-import { t as resolveModelByName } from "./models-pBSRUZhY.mjs";
+import { t as resolveModelByName } from "./models-CaCOUPZw.mjs";
 //#region src/config/define.ts
 /**
 * Returns the provided vieval definition while preserving literal field types.

package/dist/config.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"config.mjs","names":[],"sources":["../src/config/define.ts"],"sourcesContent":["import type { EvalDefinition, TaskDefinition } from './types'\n\n/**\n * Returns the provided vieval definition while preserving literal field types.\n */\nexport function defineEval<const TDefinition extends EvalDefinition>(definition: TDefinition): TDefinition {\n  return definition\n}\n\n/**\n * Returns the provided task definition while preserving literal field types.\n */\nexport function defineTask<const TDefinition extends TaskDefinition>(definition: TDefinition): TDefinition {\n  return definition\n}\n"],"mappings":";;;;;AAKA,SAAgB,WAAqD,YAAsC;AACzG,QAAO;;;;;AAMT,SAAgB,WAAqD,YAAsC;AACzG,QAAO"}
+{"version":3,"file":"config.mjs","names":[],"sources":["../src/config/define.ts"],"sourcesContent":["import type { EvalDefinition, TaskDefinition } from './types'\n\n/**\n * Returns the provided vieval definition while preserving literal field types.\n */\nexport function defineEval<const TDefinition extends EvalDefinition>(definition: TDefinition): TDefinition {\n  return definition\n}\n\n/**\n * Returns the provided task definition while preserving literal field types.\n */\nexport function defineTask<const TDefinition extends TaskDefinition>(definition: TDefinition): TDefinition {\n  return definition\n}\n"],"mappings":";;;;;AAKA,SAAgB,WAAqD,YAAsC;CACzG,OAAO;AACT;;;;AAKA,SAAgB,WAAqD,YAAsC;CACzG,OAAO;AACT"}

package/dist/core/assertions/index.d.mts

@@ -1,4 +1,4 @@
-import { X as RunScoreKind, Y as RunScore } from "../../index-BkjyCInx.mjs";
+import { X as RunScoreKind, Y as RunScore } from "../../index-CwKBlCG9.mjs";
 
 //#region src/core/assertions/index.d.ts
 /**

package/dist/core/assertions/index.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"index.mjs","names":[],"sources":["../../../src/core/assertions/index.ts"],"sourcesContent":["import type { RunScore, RunScoreKind } from '../runner/aggregate'\n\n/**\n * Stores mutable evaluation state for stateful assertion flows.\n *\n * Use when:\n * - assertions need to share counters, rolling metrics, or memoized values\n * - a scenario evaluates multiple steps and expects state-aware checks\n */\nexport type AssertionState = Map<string, unknown>\n\n/**\n * Represents one tool call emitted by a model response.\n */\nexport interface ToolCall {\n  /**\n   * Tool name used by the call.\n   */\n  name: string\n  /**\n   * Tool arguments payload.\n   */\n  args: unknown\n}\n\n/**\n * Normalized assertion context for one model output.\n */\nexport interface AssertionContext {\n  /**\n   * Plain text model output used by text assertions.\n   */\n  text: string\n  /**\n   * Optional structured output parsed from the model response.\n   */\n  structuredOutput?: unknown\n  /**\n   * Optional tool calls extracted from the model response.\n   */\n  toolCalls?: readonly ToolCall[]\n  /**\n   * Shared mutable state for stateful assertion measurement.\n   */\n  state: AssertionState\n}\n\n/**\n * Result for one assertion evaluation.\n */\nexport interface AssertionOutcome {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Assertion family emitted as run score kind.\n   */\n  scoreKind: RunScoreKind\n  /**\n   * Whether the assertion passed.\n   */\n  pass: boolean\n  /**\n   * Normalized score in the `0..1` range.\n   */\n  score: number\n  /**\n   * Human-readable reason for logs and reports.\n   */\n  reason: string\n}\n\n/**\n * Async assertion function used by eval scenarios.\n */\nexport type Assertion = (context: AssertionContext) => Promise<AssertionOutcome>\n\n/**\n * Normalizes text for matching.\n *\n * Before: `\"  Hello\\nWorld  \"`\n * After: `\"hello world\"`\n */\nexport function normalizeMatchText(value: string, caseSensitive: boolean): string {\n  const compactedWhitespace = value.trim().replaceAll(/\\s+/g, ' ')\n\n  if (caseSensitive) {\n    return compactedWhitespace\n  }\n\n  return compactedWhitespace.toLowerCase()\n}\n\nfunction clampScore(score: number): number {\n  if (Number.isNaN(score)) {\n    return 0\n  }\n\n  if (score < 0) {\n    return 0\n  }\n\n  if (score > 1) {\n    return 1\n  }\n\n  return score\n}\n\nfunction createOutcome(\n  id: string,\n  scoreKind: RunScoreKind,\n  pass: boolean,\n  score: number,\n  reason: string,\n): AssertionOutcome {\n  return {\n    id,\n    pass,\n    reason,\n    score: clampScore(score),\n    scoreKind,\n  }\n}\n\n/**\n * Options for include-keyword assertions.\n */\nexport interface MustIncludeAssertionOptions {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Keywords that must be present.\n   */\n  keywords: readonly string[]\n  /**\n   * Match mode for keywords.\n   *\n   * @default 'all'\n   */\n  mode?: 'all' | 'any'\n  /**\n   * Case-sensitive matching toggle.\n   *\n   * @default false\n   */\n  caseSensitive?: boolean\n}\n\n/**\n * Creates an assertion that requires specific keywords in model text.\n *\n * Example:\n * `expectMustInclude({ id: 'tone', keywords: ['calm', 'move'] })`\n */\nexport function expectMustInclude(options: MustIncludeAssertionOptions): Assertion {\n  return async (context) => {\n    if (options.keywords.length === 0) {\n      return createOutcome(options.id, 'exact', true, 1, 'No required keywords configured.')\n    }\n\n    const caseSensitive = options.caseSensitive ?? false\n    const normalizedText = normalizeMatchText(context.text, caseSensitive)\n    const matches = options.keywords.filter((keyword) => {\n      const normalizedKeyword = normalizeMatchText(keyword, caseSensitive)\n      return normalizedText.includes(normalizedKeyword)\n    })\n\n    const mode = options.mode ?? 'all'\n    const pass = mode === 'all'\n      ? matches.length === options.keywords.length\n      : matches.length > 0\n\n    const score = options.keywords.length === 0 ? 1 : matches.length / options.keywords.length\n\n    return createOutcome(\n      options.id,\n      'exact',\n      pass,\n      score,\n      pass\n        ? `Matched ${matches.length}/${options.keywords.length} required keywords.`\n        : `Matched ${matches.length}/${options.keywords.length} required keywords.`,\n    )\n  }\n}\n\n/**\n * Options for exclude-keyword assertions.\n */\nexport interface MustExcludeAssertionOptions {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Keywords that must not appear.\n   */\n  keywords: readonly string[]\n  /**\n   * Case-sensitive matching toggle.\n   *\n   * @default false\n   */\n  caseSensitive?: boolean\n}\n\n/**\n * Creates an assertion that forbids specific keywords.\n *\n * Example:\n * `expectMustExclude({ id: 'no-engine-dump', keywords: ['bestmove', 'ponder'] })`\n */\nexport function expectMustExclude(options: MustExcludeAssertionOptions): Assertion {\n  return async (context) => {\n    if (options.keywords.length === 0) {\n      return createOutcome(options.id, 'exact', true, 1, 'No excluded keywords configured.')\n    }\n\n    const caseSensitive = options.caseSensitive ?? false\n    const normalizedText = normalizeMatchText(context.text, caseSensitive)\n    const forbiddenMatches = options.keywords.filter((keyword) => {\n      const normalizedKeyword = normalizeMatchText(keyword, caseSensitive)\n      return normalizedText.includes(normalizedKeyword)\n    })\n\n    const pass = forbiddenMatches.length === 0\n    const score = pass ? 1 : 0\n\n    return createOutcome(\n      options.id,\n      'exact',\n      pass,\n      score,\n      pass\n        ? 'No forbidden keywords found.'\n        : `Forbidden keywords found: ${forbiddenMatches.join(', ')}`,\n    )\n  }\n}\n\n/**\n * Options for regular-expression assertions.\n */\nexport interface RegexAssertionOptions {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Pattern to apply to model text.\n   */\n  pattern: RegExp\n}\n\n/**\n * Creates an assertion based on a regular expression.\n *\n * Example:\n * `expectRegex({ id: 'starts-with-act', pattern: /^<\\|ACT:/ })`\n */\nexport function expectRegex(options: RegexAssertionOptions): Assertion {\n  return async (context) => {\n    const pass = options.pattern.test(context.text)\n\n    return createOutcome(\n      options.id,\n      'exact',\n      pass,\n      pass ? 1 : 0,\n      pass ? 'Regex matched response text.' : `Regex did not match: ${options.pattern}`,\n    )\n  }\n}\n\n/**\n * Options for structured-output assertions.\n */\nexport interface StructuredOutputAssertionOptions<TValue> {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Runtime validator for structured output.\n   */\n  validate: (value: unknown) => value is TValue\n  /**\n   * Optional failure reason.\n   */\n  failureReason?: string\n}\n\n/**\n * Creates an assertion for structured model output.\n *\n * Example:\n * `expectStructuredOutput({ id: 'json-shape', validate: isMySchema })`\n */\nexport function expectStructuredOutput<TValue>(options: StructuredOutputAssertionOptions<TValue>): Assertion {\n  return async (context) => {\n    const pass = options.validate(context.structuredOutput)\n\n    return createOutcome(\n      options.id,\n      'exact',\n      pass,\n      pass ? 1 : 0,\n      pass ? 'Structured output matched validator.' : (options.failureReason ?? 'Structured output validation failed.'),\n    )\n  }\n}\n\n/**\n * Options for tool-call argument assertions.\n */\nexport interface ToolCallArgsAssertionOptions {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Tool name to inspect.\n   */\n  toolName: string\n  /**\n   * Runtime validator for tool arguments.\n   */\n  validate: (args: unknown) => boolean\n}\n\n/**\n * Creates an assertion for validating tool-call arguments.\n *\n * Example:\n * `expectToolCallArgs({ id: 'spark-command-shape', toolName: 'builtIn_sparkCommand', validate: isSparkArgs })`\n */\nexport function expectToolCallArgs(options: ToolCallArgsAssertionOptions): Assertion {\n  return async (context) => {\n    const targetCall = (context.toolCalls ?? []).find(call => call.name === options.toolName)\n\n    if (targetCall == null) {\n      return createOutcome(options.id, 'exact', false, 0, `Missing tool call: ${options.toolName}`)\n    }\n\n    const pass = options.validate(targetCall.args)\n\n    return createOutcome(\n      options.id,\n      'exact',\n      pass,\n      pass ? 1 : 0,\n      pass ? `Tool call args validated for ${options.toolName}.` : `Tool call args validation failed for ${options.toolName}.`,\n    )\n  }\n}\n\n/**\n * Rubric judge result returned by teacher-model or rubric logic.\n */\nexport interface RubricJudgeResult {\n  /**\n   * Normalized score in the `0..1` range.\n   */\n  score: number\n  /**\n   * Judge explanation text.\n   */\n  reason: string\n  /**\n   * Optional judge model id.\n   */\n  judgeModel?: string\n}\n\n/**\n * Options for rubric assertions.\n */\nexport interface RubricAssertionOptions {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Async rubric judge callback.\n   */\n  judge: (context: AssertionContext) => Promise<RubricJudgeResult>\n  /**\n   * Minimum passing score.\n   *\n   * @default 0.7\n   */\n  minScore?: number\n}\n\n/**\n * Creates a rubric assertion driven by teacher-model style scoring.\n *\n * Example:\n * `expectRubric({ id: 'human-like-tone', judge: judgeFn, minScore: 0.8 })`\n */\nexport function expectRubric(options: RubricAssertionOptions): Assertion {\n  return async (context) => {\n    const result = await options.judge(context)\n    const minScore = options.minScore ?? 0.7\n    const normalizedScore = clampScore(result.score)\n    const pass = normalizedScore >= minScore\n\n    return createOutcome(\n      options.id,\n      'judge',\n      pass,\n      normalizedScore,\n      `${result.reason}${result.judgeModel ? ` (judge: ${result.judgeModel})` : ''}`,\n    )\n  }\n}\n\n/**\n * Options for custom assertions.\n */\nexport interface CustomAssertionOptions {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Score family emitted by this custom assertion.\n   */\n  scoreKind: RunScoreKind\n  /**\n   * Custom evaluator callback.\n   */\n  evaluate: (context: AssertionContext) => Promise<{ pass: boolean, reason: string, score: number }> | { pass: boolean, reason: string, score: number }\n}\n\n/**\n * Creates a custom assertion with fully user-defined logic.\n *\n * Example:\n * `expectCustom({ id: 'stateful-window', scoreKind: 'exact', evaluate: (ctx) => ... })`\n */\nexport function expectCustom(options: CustomAssertionOptions): Assertion {\n  return async (context) => {\n    const result = await options.evaluate(context)\n\n    return createOutcome(options.id, options.scoreKind, result.pass, result.score, result.reason)\n  }\n}\n\n/**\n * Creates an inverse assertion.\n *\n * Example:\n * `expectNot(expectMustInclude({ id: 'contains-engine-word', keywords: ['bestmove'] }), { id: 'no-engine-word' })`\n */\nexport function expectNot(assertion: Assertion, options: { id: string }): Assertion {\n  return async (context) => {\n    const baseOutcome = await assertion(context)\n\n    return createOutcome(\n      options.id,\n      baseOutcome.scoreKind,\n      !baseOutcome.pass,\n      1 - baseOutcome.score,\n      `NOT(${baseOutcome.id}): ${baseOutcome.reason}`,\n    )\n  }\n}\n\n/**\n * Executes assertion list and returns all outcomes.\n *\n * Call stack:\n *\n * {@link evaluateAssertions}\n *   -> `assertion(context)`\n *     -> {@link AssertionOutcome}[]\n */\nexport async function evaluateAssertions(\n  assertions: readonly Assertion[],\n  context: Omit<AssertionContext, 'state'> & { state?: AssertionState },\n): Promise<AssertionOutcome[]> {\n  const state = context.state ?? new Map<string, unknown>()\n  const normalizedContext: AssertionContext = {\n    state,\n    structuredOutput: context.structuredOutput,\n    text: context.text,\n    toolCalls: context.toolCalls,\n  }\n\n  const outcomes: AssertionOutcome[] = []\n\n  for (const assertion of assertions) {\n    outcomes.push(await assertion(normalizedContext))\n  }\n\n  return outcomes\n}\n\n/**\n * Converts assertion outcomes to run-score tuples consumed by aggregation.\n */\nexport function toRunScores(outcomes: readonly AssertionOutcome[]): RunScore[] {\n  return outcomes.map(outcome => ({\n    kind: outcome.scoreKind,\n    score: outcome.score,\n  }))\n}\n\n/**\n * Returns failing assertion outcomes in original order.\n */\nexport function collectFailedAssertions(outcomes: readonly AssertionOutcome[]): AssertionOutcome[] {\n  return outcomes.filter(outcome => !outcome.pass)\n}\n"],"mappings":";;;;;;;AAoFA,SAAgB,mBAAmB,OAAe,eAAgC;CAChF,MAAM,sBAAsB,MAAM,MAAM,CAAC,WAAW,QAAQ,IAAI;AAEhE,KAAI,cACF,QAAO;AAGT,QAAO,oBAAoB,aAAa;;AAG1C,SAAS,WAAW,OAAuB;AACzC,KAAI,OAAO,MAAM,MAAM,CACrB,QAAO;AAGT,KAAI,QAAQ,EACV,QAAO;AAGT,KAAI,QAAQ,EACV,QAAO;AAGT,QAAO;;AAGT,SAAS,cACP,IACA,WACA,MACA,OACA,QACkB;AAClB,QAAO;EACL;EACA;EACA;EACA,OAAO,WAAW,MAAM;EACxB;EACD;;;;;;;;AAmCH,SAAgB,kBAAkB,SAAiD;AACjF,QAAO,OAAO,YAAY;AACxB,MAAI,QAAQ,SAAS,WAAW,EAC9B,QAAO,cAAc,QAAQ,IAAI,SAAS,MAAM,GAAG,mCAAmC;EAGxF,MAAM,gBAAgB,QAAQ,iBAAiB;EAC/C,MAAM,iBAAiB,mBAAmB,QAAQ,MAAM,cAAc;EACtE,MAAM,UAAU,QAAQ,SAAS,QAAQ,YAAY;GACnD,MAAM,oBAAoB,mBAAmB,SAAS,cAAc;AACpE,UAAO,eAAe,SAAS,kBAAkB;IACjD;EAGF,MAAM,QADO,QAAQ,QAAQ,WACP,QAClB,QAAQ,WAAW,QAAQ,SAAS,SACpC,QAAQ,SAAS;EAErB,MAAM,QAAQ,QAAQ,SAAS,WAAW,IAAI,IAAI,QAAQ,SAAS,QAAQ,SAAS;AAEpF,SAAO,cACL,QAAQ,IACR,SACA,MACA,OACA,OACI,WAAW,QAAQ,OAAO,GAAG,QAAQ,SAAS,OAAO,uBACrD,WAAW,QAAQ,OAAO,GAAG,QAAQ,SAAS,OAAO,qBAC1D;;;;;;;;;AA8BL,SAAgB,kBAAkB,SAAiD;AACjF,QAAO,OAAO,YAAY;AACxB,MAAI,QAAQ,SAAS,WAAW,EAC9B,QAAO,cAAc,QAAQ,IAAI,SAAS,MAAM,GAAG,mCAAmC;EAGxF,MAAM,gBAAgB,QAAQ,iBAAiB;EAC/C,MAAM,iBAAiB,mBAAmB,QAAQ,MAAM,cAAc;EACtE,MAAM,mBAAmB,QAAQ,SAAS,QAAQ,YAAY;GAC5D,MAAM,oBAAoB,mBAAmB,SAAS,cAAc;AACpE,UAAO,eAAe,SAAS,kBAAkB;IACjD;EAEF,MAAM,OAAO,iBAAiB,WAAW;EACzC,MAAM,QAAQ,OAAO,IAAI;AAEzB,SAAO,cACL,QAAQ,IACR,SACA,MACA,OACA,OACI,iCACA,6BAA6B,iBAAiB,KAAK,KAAK,GAC7D;;;;;;;;;AAwBL,SAAgB,YAAY,SAA2C;AACrE,QAAO,OAAO,YAAY;EACxB,MAAM,OAAO,QAAQ,QAAQ,KAAK,QAAQ,KAAK;AAE/C,SAAO,cACL,QAAQ,IACR,SACA,MACA,OAAO,IAAI,GACX,OAAO,iCAAiC,wBAAwB,QAAQ,UACzE;;;;;;;;;AA4BL,SAAgB,uBAA+B,SAA8D;AAC3G,QAAO,OAAO,YAAY;EACxB,MAAM,OAAO,QAAQ,SAAS,QAAQ,iBAAiB;AAEvD,SAAO,cACL,QAAQ,IACR,SACA,MACA,OAAO,IAAI,GACX,OAAO,yCAA0C,QAAQ,iBAAiB,uCAC3E;;;;;;;;;AA4BL,SAAgB,mBAAmB,SAAkD;AACnF,QAAO,OAAO,YAAY;EACxB,MAAM,cAAc,QAAQ,aAAa,EAAE,EAAE,MAAK,SAAQ,KAAK,SAAS,QAAQ,SAAS;AAEzF,MAAI,cAAc,KAChB,QAAO,cAAc,QAAQ,IAAI,SAAS,OAAO,GAAG,sBAAsB,QAAQ,WAAW;EAG/F,MAAM,OAAO,QAAQ,SAAS,WAAW,KAAK;AAE9C,SAAO,cACL,QAAQ,IACR,SACA,MACA,OAAO,IAAI,GACX,OAAO,gCAAgC,QAAQ,SAAS,KAAK,wCAAwC,QAAQ,SAAS,GACvH;;;;;;;;;AAgDL,SAAgB,aAAa,SAA4C;AACvE,QAAO,OAAO,YAAY;EACxB,MAAM,SAAS,MAAM,QAAQ,MAAM,QAAQ;EAC3C,MAAM,WAAW,QAAQ,YAAY;EACrC,MAAM,kBAAkB,WAAW,OAAO,MAAM;EAChD,MAAM,OAAO,mBAAmB;AAEhC,SAAO,cACL,QAAQ,IACR,SACA,MACA,iBACA,GAAG,OAAO,SAAS,OAAO,aAAa,YAAY,OAAO,WAAW,KAAK,KAC3E;;;;;;;;;AA4BL,SAAgB,aAAa,SAA4C;AACvE,QAAO,OAAO,YAAY;EACxB,MAAM,SAAS,MAAM,QAAQ,SAAS,QAAQ;AAE9C,SAAO,cAAc,QAAQ,IAAI,QAAQ,WAAW,OAAO,MAAM,OAAO,OAAO,OAAO,OAAO;;;;;;;;;AAUjG,SAAgB,UAAU,WAAsB,SAAoC;AAClF,QAAO,OAAO,YAAY;EACxB,MAAM,cAAc,MAAM,UAAU,QAAQ;AAE5C,SAAO,cACL,QAAQ,IACR,YAAY,WACZ,CAAC,YAAY,MACb,IAAI,YAAY,OAChB,OAAO,YAAY,GAAG,KAAK,YAAY,SACxC;;;;;;;;;;;;AAaL,eAAsB,mBACpB,YACA,SAC6B;CAE7B,MAAM,oBAAsC;EAC1C,OAFY,QAAQ,yBAAS,IAAI,KAAsB;EAGvD,kBAAkB,QAAQ;EAC1B,MAAM,QAAQ;EACd,WAAW,QAAQ;EACpB;CAED,MAAM,WAA+B,EAAE;AAEvC,MAAK,MAAM,aAAa,WACtB,UAAS,KAAK,MAAM,UAAU,kBAAkB,CAAC;AAGnD,QAAO;;;;;AAMT,SAAgB,YAAY,UAAmD;AAC7E,QAAO,SAAS,KAAI,aAAY;EAC9B,MAAM,QAAQ;EACd,OAAO,QAAQ;EAChB,EAAE;;;;;AAML,SAAgB,wBAAwB,UAA2D;AACjG,QAAO,SAAS,QAAO,YAAW,CAAC,QAAQ,KAAK"}
+{"version":3,"file":"index.mjs","names":[],"sources":["../../../src/core/assertions/index.ts"],"sourcesContent":["import type { RunScore, RunScoreKind } from '../runner/aggregate'\n\n/**\n * Stores mutable evaluation state for stateful assertion flows.\n *\n * Use when:\n * - assertions need to share counters, rolling metrics, or memoized values\n * - a scenario evaluates multiple steps and expects state-aware checks\n */\nexport type AssertionState = Map<string, unknown>\n\n/**\n * Represents one tool call emitted by a model response.\n */\nexport interface ToolCall {\n  /**\n   * Tool name used by the call.\n   */\n  name: string\n  /**\n   * Tool arguments payload.\n   */\n  args: unknown\n}\n\n/**\n * Normalized assertion context for one model output.\n */\nexport interface AssertionContext {\n  /**\n   * Plain text model output used by text assertions.\n   */\n  text: string\n  /**\n   * Optional structured output parsed from the model response.\n   */\n  structuredOutput?: unknown\n  /**\n   * Optional tool calls extracted from the model response.\n   */\n  toolCalls?: readonly ToolCall[]\n  /**\n   * Shared mutable state for stateful assertion measurement.\n   */\n  state: AssertionState\n}\n\n/**\n * Result for one assertion evaluation.\n */\nexport interface AssertionOutcome {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Assertion family emitted as run score kind.\n   */\n  scoreKind: RunScoreKind\n  /**\n   * Whether the assertion passed.\n   */\n  pass: boolean\n  /**\n   * Normalized score in the `0..1` range.\n   */\n  score: number\n  /**\n   * Human-readable reason for logs and reports.\n   */\n  reason: string\n}\n\n/**\n * Async assertion function used by eval scenarios.\n */\nexport type Assertion = (context: AssertionContext) => Promise<AssertionOutcome>\n\n/**\n * Normalizes text for matching.\n *\n * Before: `\"  Hello\\nWorld  \"`\n * After: `\"hello world\"`\n */\nexport function normalizeMatchText(value: string, caseSensitive: boolean): string {\n  const compactedWhitespace = value.trim().replaceAll(/\\s+/g, ' ')\n\n  if (caseSensitive) {\n    return compactedWhitespace\n  }\n\n  return compactedWhitespace.toLowerCase()\n}\n\nfunction clampScore(score: number): number {\n  if (Number.isNaN(score)) {\n    return 0\n  }\n\n  if (score < 0) {\n    return 0\n  }\n\n  if (score > 1) {\n    return 1\n  }\n\n  return score\n}\n\nfunction createOutcome(\n  id: string,\n  scoreKind: RunScoreKind,\n  pass: boolean,\n  score: number,\n  reason: string,\n): AssertionOutcome {\n  return {\n    id,\n    pass,\n    reason,\n    score: clampScore(score),\n    scoreKind,\n  }\n}\n\n/**\n * Options for include-keyword assertions.\n */\nexport interface MustIncludeAssertionOptions {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Keywords that must be present.\n   */\n  keywords: readonly string[]\n  /**\n   * Match mode for keywords.\n   *\n   * @default 'all'\n   */\n  mode?: 'all' | 'any'\n  /**\n   * Case-sensitive matching toggle.\n   *\n   * @default false\n   */\n  caseSensitive?: boolean\n}\n\n/**\n * Creates an assertion that requires specific keywords in model text.\n *\n * Example:\n * `expectMustInclude({ id: 'tone', keywords: ['calm', 'move'] })`\n */\nexport function expectMustInclude(options: MustIncludeAssertionOptions): Assertion {\n  return async (context) => {\n    if (options.keywords.length === 0) {\n      return createOutcome(options.id, 'exact', true, 1, 'No required keywords configured.')\n    }\n\n    const caseSensitive = options.caseSensitive ?? false\n    const normalizedText = normalizeMatchText(context.text, caseSensitive)\n    const matches = options.keywords.filter((keyword) => {\n      const normalizedKeyword = normalizeMatchText(keyword, caseSensitive)\n      return normalizedText.includes(normalizedKeyword)\n    })\n\n    const mode = options.mode ?? 'all'\n    const pass = mode === 'all'\n      ? matches.length === options.keywords.length\n      : matches.length > 0\n\n    const score = options.keywords.length === 0 ? 1 : matches.length / options.keywords.length\n\n    return createOutcome(\n      options.id,\n      'exact',\n      pass,\n      score,\n      pass\n        ? `Matched ${matches.length}/${options.keywords.length} required keywords.`\n        : `Matched ${matches.length}/${options.keywords.length} required keywords.`,\n    )\n  }\n}\n\n/**\n * Options for exclude-keyword assertions.\n */\nexport interface MustExcludeAssertionOptions {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Keywords that must not appear.\n   */\n  keywords: readonly string[]\n  /**\n   * Case-sensitive matching toggle.\n   *\n   * @default false\n   */\n  caseSensitive?: boolean\n}\n\n/**\n * Creates an assertion that forbids specific keywords.\n *\n * Example:\n * `expectMustExclude({ id: 'no-engine-dump', keywords: ['bestmove', 'ponder'] })`\n */\nexport function expectMustExclude(options: MustExcludeAssertionOptions): Assertion {\n  return async (context) => {\n    if (options.keywords.length === 0) {\n      return createOutcome(options.id, 'exact', true, 1, 'No excluded keywords configured.')\n    }\n\n    const caseSensitive = options.caseSensitive ?? false\n    const normalizedText = normalizeMatchText(context.text, caseSensitive)\n    const forbiddenMatches = options.keywords.filter((keyword) => {\n      const normalizedKeyword = normalizeMatchText(keyword, caseSensitive)\n      return normalizedText.includes(normalizedKeyword)\n    })\n\n    const pass = forbiddenMatches.length === 0\n    const score = pass ? 1 : 0\n\n    return createOutcome(\n      options.id,\n      'exact',\n      pass,\n      score,\n      pass\n        ? 'No forbidden keywords found.'\n        : `Forbidden keywords found: ${forbiddenMatches.join(', ')}`,\n    )\n  }\n}\n\n/**\n * Options for regular-expression assertions.\n */\nexport interface RegexAssertionOptions {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Pattern to apply to model text.\n   */\n  pattern: RegExp\n}\n\n/**\n * Creates an assertion based on a regular expression.\n *\n * Example:\n * `expectRegex({ id: 'starts-with-act', pattern: /^<\\|ACT:/ })`\n */\nexport function expectRegex(options: RegexAssertionOptions): Assertion {\n  return async (context) => {\n    const pass = options.pattern.test(context.text)\n\n    return createOutcome(\n      options.id,\n      'exact',\n      pass,\n      pass ? 1 : 0,\n      pass ? 'Regex matched response text.' : `Regex did not match: ${options.pattern}`,\n    )\n  }\n}\n\n/**\n * Options for structured-output assertions.\n */\nexport interface StructuredOutputAssertionOptions<TValue> {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Runtime validator for structured output.\n   */\n  validate: (value: unknown) => value is TValue\n  /**\n   * Optional failure reason.\n   */\n  failureReason?: string\n}\n\n/**\n * Creates an assertion for structured model output.\n *\n * Example:\n * `expectStructuredOutput({ id: 'json-shape', validate: isMySchema })`\n */\nexport function expectStructuredOutput<TValue>(options: StructuredOutputAssertionOptions<TValue>): Assertion {\n  return async (context) => {\n    const pass = options.validate(context.structuredOutput)\n\n    return createOutcome(\n      options.id,\n      'exact',\n      pass,\n      pass ? 1 : 0,\n      pass ? 'Structured output matched validator.' : (options.failureReason ?? 'Structured output validation failed.'),\n    )\n  }\n}\n\n/**\n * Options for tool-call argument assertions.\n */\nexport interface ToolCallArgsAssertionOptions {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Tool name to inspect.\n   */\n  toolName: string\n  /**\n   * Runtime validator for tool arguments.\n   */\n  validate: (args: unknown) => boolean\n}\n\n/**\n * Creates an assertion for validating tool-call arguments.\n *\n * Example:\n * `expectToolCallArgs({ id: 'spark-command-shape', toolName: 'builtIn_sparkCommand', validate: isSparkArgs })`\n */\nexport function expectToolCallArgs(options: ToolCallArgsAssertionOptions): Assertion {\n  return async (context) => {\n    const targetCall = (context.toolCalls ?? []).find(call => call.name === options.toolName)\n\n    if (targetCall == null) {\n      return createOutcome(options.id, 'exact', false, 0, `Missing tool call: ${options.toolName}`)\n    }\n\n    const pass = options.validate(targetCall.args)\n\n    return createOutcome(\n      options.id,\n      'exact',\n      pass,\n      pass ? 1 : 0,\n      pass ? `Tool call args validated for ${options.toolName}.` : `Tool call args validation failed for ${options.toolName}.`,\n    )\n  }\n}\n\n/**\n * Rubric judge result returned by teacher-model or rubric logic.\n */\nexport interface RubricJudgeResult {\n  /**\n   * Normalized score in the `0..1` range.\n   */\n  score: number\n  /**\n   * Judge explanation text.\n   */\n  reason: string\n  /**\n   * Optional judge model id.\n   */\n  judgeModel?: string\n}\n\n/**\n * Options for rubric assertions.\n */\nexport interface RubricAssertionOptions {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Async rubric judge callback.\n   */\n  judge: (context: AssertionContext) => Promise<RubricJudgeResult>\n  /**\n   * Minimum passing score.\n   *\n   * @default 0.7\n   */\n  minScore?: number\n}\n\n/**\n * Creates a rubric assertion driven by teacher-model style scoring.\n *\n * Example:\n * `expectRubric({ id: 'human-like-tone', judge: judgeFn, minScore: 0.8 })`\n */\nexport function expectRubric(options: RubricAssertionOptions): Assertion {\n  return async (context) => {\n    const result = await options.judge(context)\n    const minScore = options.minScore ?? 0.7\n    const normalizedScore = clampScore(result.score)\n    const pass = normalizedScore >= minScore\n\n    return createOutcome(\n      options.id,\n      'judge',\n      pass,\n      normalizedScore,\n      `${result.reason}${result.judgeModel ? ` (judge: ${result.judgeModel})` : ''}`,\n    )\n  }\n}\n\n/**\n * Options for custom assertions.\n */\nexport interface CustomAssertionOptions {\n  /**\n   * Stable assertion id.\n   */\n  id: string\n  /**\n   * Score family emitted by this custom assertion.\n   */\n  scoreKind: RunScoreKind\n  /**\n   * Custom evaluator callback.\n   */\n  evaluate: (context: AssertionContext) => Promise<{ pass: boolean, reason: string, score: number }> | { pass: boolean, reason: string, score: number }\n}\n\n/**\n * Creates a custom assertion with fully user-defined logic.\n *\n * Example:\n * `expectCustom({ id: 'stateful-window', scoreKind: 'exact', evaluate: (ctx) => ... })`\n */\nexport function expectCustom(options: CustomAssertionOptions): Assertion {\n  return async (context) => {\n    const result = await options.evaluate(context)\n\n    return createOutcome(options.id, options.scoreKind, result.pass, result.score, result.reason)\n  }\n}\n\n/**\n * Creates an inverse assertion.\n *\n * Example:\n * `expectNot(expectMustInclude({ id: 'contains-engine-word', keywords: ['bestmove'] }), { id: 'no-engine-word' })`\n */\nexport function expectNot(assertion: Assertion, options: { id: string }): Assertion {\n  return async (context) => {\n    const baseOutcome = await assertion(context)\n\n    return createOutcome(\n      options.id,\n      baseOutcome.scoreKind,\n      !baseOutcome.pass,\n      1 - baseOutcome.score,\n      `NOT(${baseOutcome.id}): ${baseOutcome.reason}`,\n    )\n  }\n}\n\n/**\n * Executes assertion list and returns all outcomes.\n *\n * Call stack:\n *\n * {@link evaluateAssertions}\n *   -> `assertion(context)`\n *     -> {@link AssertionOutcome}[]\n */\nexport async function evaluateAssertions(\n  assertions: readonly Assertion[],\n  context: Omit<AssertionContext, 'state'> & { state?: AssertionState },\n): Promise<AssertionOutcome[]> {\n  const state = context.state ?? new Map<string, unknown>()\n  const normalizedContext: AssertionContext = {\n    state,\n    structuredOutput: context.structuredOutput,\n    text: context.text,\n    toolCalls: context.toolCalls,\n  }\n\n  const outcomes: AssertionOutcome[] = []\n\n  for (const assertion of assertions) {\n    outcomes.push(await assertion(normalizedContext))\n  }\n\n  return outcomes\n}\n\n/**\n * Converts assertion outcomes to run-score tuples consumed by aggregation.\n */\nexport function toRunScores(outcomes: readonly AssertionOutcome[]): RunScore[] {\n  return outcomes.map(outcome => ({\n    kind: outcome.scoreKind,\n    score: outcome.score,\n  }))\n}\n\n/**\n * Returns failing assertion outcomes in original order.\n */\nexport function collectFailedAssertions(outcomes: readonly AssertionOutcome[]): AssertionOutcome[] {\n  return outcomes.filter(outcome => !outcome.pass)\n}\n"],"mappings":";;;;;;;AAoFA,SAAgB,mBAAmB,OAAe,eAAgC;CAChF,MAAM,sBAAsB,MAAM,KAAK,CAAC,CAAC,WAAW,QAAQ,GAAG;CAE/D,IAAI,eACF,OAAO;CAGT,OAAO,oBAAoB,YAAY;AACzC;AAEA,SAAS,WAAW,OAAuB;CACzC,IAAI,OAAO,MAAM,KAAK,GACpB,OAAO;CAGT,IAAI,QAAQ,GACV,OAAO;CAGT,IAAI,QAAQ,GACV,OAAO;CAGT,OAAO;AACT;AAEA,SAAS,cACP,IACA,WACA,MACA,OACA,QACkB;CAClB,OAAO;EACL;EACA;EACA;EACA,OAAO,WAAW,KAAK;EACvB;CACF;AACF;;;;;;;AAkCA,SAAgB,kBAAkB,SAAiD;CACjF,OAAO,OAAO,YAAY;EACxB,IAAI,QAAQ,SAAS,WAAW,GAC9B,OAAO,cAAc,QAAQ,IAAI,SAAS,MAAM,GAAG,kCAAkC;EAGvF,MAAM,gBAAgB,QAAQ,iBAAiB;EAC/C,MAAM,iBAAiB,mBAAmB,QAAQ,MAAM,aAAa;EACrE,MAAM,UAAU,QAAQ,SAAS,QAAQ,YAAY;GACnD,MAAM,oBAAoB,mBAAmB,SAAS,aAAa;GACnE,OAAO,eAAe,SAAS,iBAAiB;EAClD,CAAC;EAGD,MAAM,QADO,QAAQ,QAAQ,WACP,QAClB,QAAQ,WAAW,QAAQ,SAAS,SACpC,QAAQ,SAAS;EAErB,MAAM,QAAQ,QAAQ,SAAS,WAAW,IAAI,IAAI,QAAQ,SAAS,QAAQ,SAAS;EAEpF,OAAO,cACL,QAAQ,IACR,SACA,MACA,OACA,OACI,WAAW,QAAQ,OAAO,GAAG,QAAQ,SAAS,OAAO,uBACrD,WAAW,QAAQ,OAAO,GAAG,QAAQ,SAAS,OAAO,oBAC3D;CACF;AACF;;;;;;;AA4BA,SAAgB,kBAAkB,SAAiD;CACjF,OAAO,OAAO,YAAY;EACxB,IAAI,QAAQ,SAAS,WAAW,GAC9B,OAAO,cAAc,QAAQ,IAAI,SAAS,MAAM,GAAG,kCAAkC;EAGvF,MAAM,gBAAgB,QAAQ,iBAAiB;EAC/C,MAAM,iBAAiB,mBAAmB,QAAQ,MAAM,aAAa;EACrE,MAAM,mBAAmB,QAAQ,SAAS,QAAQ,YAAY;GAC5D,MAAM,oBAAoB,mBAAmB,SAAS,aAAa;GACnE,OAAO,eAAe,SAAS,iBAAiB;EAClD,CAAC;EAED,MAAM,OAAO,iBAAiB,WAAW;EACzC,MAAM,QAAQ,OAAO,IAAI;EAEzB,OAAO,cACL,QAAQ,IACR,SACA,MACA,OACA,OACI,iCACA,6BAA6B,iBAAiB,KAAK,IAAI,GAC7D;CACF;AACF;;;;;;;AAsBA,SAAgB,YAAY,SAA2C;CACrE,OAAO,OAAO,YAAY;EACxB,MAAM,OAAO,QAAQ,QAAQ,KAAK,QAAQ,IAAI;EAE9C,OAAO,cACL,QAAQ,IACR,SACA,MACA,OAAO,IAAI,GACX,OAAO,iCAAiC,wBAAwB,QAAQ,SAC1E;CACF;AACF;;;;;;;AA0BA,SAAgB,uBAA+B,SAA8D;CAC3G,OAAO,OAAO,YAAY;EACxB,MAAM,OAAO,QAAQ,SAAS,QAAQ,gBAAgB;EAEtD,OAAO,cACL,QAAQ,IACR,SACA,MACA,OAAO,IAAI,GACX,OAAO,yCAA0C,QAAQ,iBAAiB,sCAC5E;CACF;AACF;;;;;;;AA0BA,SAAgB,mBAAmB,SAAkD;CACnF,OAAO,OAAO,YAAY;EACxB,MAAM,cAAc,QAAQ,aAAa,CAAC,EAAA,CAAG,MAAK,SAAQ,KAAK,SAAS,QAAQ,QAAQ;EAExF,IAAI,cAAc,MAChB,OAAO,cAAc,QAAQ,IAAI,SAAS,OAAO,GAAG,sBAAsB,QAAQ,UAAU;EAG9F,MAAM,OAAO,QAAQ,SAAS,WAAW,IAAI;EAE7C,OAAO,cACL,QAAQ,IACR,SACA,MACA,OAAO,IAAI,GACX,OAAO,gCAAgC,QAAQ,SAAS,KAAK,wCAAwC,QAAQ,SAAS,EACxH;CACF;AACF;;;;;;;AA8CA,SAAgB,aAAa,SAA4C;CACvE,OAAO,OAAO,YAAY;EACxB,MAAM,SAAS,MAAM,QAAQ,MAAM,OAAO;EAC1C,MAAM,WAAW,QAAQ,YAAY;EACrC,MAAM,kBAAkB,WAAW,OAAO,KAAK;EAC/C,MAAM,OAAO,mBAAmB;EAEhC,OAAO,cACL,QAAQ,IACR,SACA,MACA,iBACA,GAAG,OAAO,SAAS,OAAO,aAAa,YAAY,OAAO,WAAW,KAAK,IAC5E;CACF;AACF;;;;;;;AA0BA,SAAgB,aAAa,SAA4C;CACvE,OAAO,OAAO,YAAY;EACxB,MAAM,SAAS,MAAM,QAAQ,SAAS,OAAO;EAE7C,OAAO,cAAc,QAAQ,IAAI,QAAQ,WAAW,OAAO,MAAM,OAAO,OAAO,OAAO,MAAM;CAC9F;AACF;;;;;;;AAQA,SAAgB,UAAU,WAAsB,SAAoC;CAClF,OAAO,OAAO,YAAY;EACxB,MAAM,cAAc,MAAM,UAAU,OAAO;EAE3C,OAAO,cACL,QAAQ,IACR,YAAY,WACZ,CAAC,YAAY,MACb,IAAI,YAAY,OAChB,OAAO,YAAY,GAAG,KAAK,YAAY,QACzC;CACF;AACF;;;;;;;;;;AAWA,eAAsB,mBACpB,YACA,SAC6B;CAE7B,MAAM,oBAAsC;EAC1C,OAFY,QAAQ,yBAAS,IAAI,IAAqB;EAGtD,kBAAkB,QAAQ;EAC1B,MAAM,QAAQ;EACd,WAAW,QAAQ;CACrB;CAEA,MAAM,WAA+B,CAAC;CAEtC,KAAK,MAAM,aAAa,YACtB,SAAS,KAAK,MAAM,UAAU,iBAAiB,CAAC;CAGlD,OAAO;AACT;;;;AAKA,SAAgB,YAAY,UAAmD;CAC7E,OAAO,SAAS,KAAI,aAAY;EAC9B,MAAM,QAAQ;EACd,OAAO,QAAQ;CACjB,EAAE;AACJ;;;;AAKA,SAAgB,wBAAwB,UAA2D;CACjG,OAAO,SAAS,QAAO,YAAW,CAAC,QAAQ,IAAI;AACjD"}

package/dist/core/inference-executors/index.d.mts

@@ -1,4 +1,4 @@
-import { a as requiredEnvFrom, i as envFrom, n as EnvValueType, r as RequiredEnvFromOptions, t as EnvFromOptions } from "../../env-DfWZy_n4.mjs";
+import { a as requiredEnvFrom, i as envFrom, n as EnvValueType, r as RequiredEnvFromOptions, t as EnvFromOptions } from "../../env-bRH0K6fU.mjs";
 import { createOpenAI } from "@xsai-ext/providers/create";
 
 //#region src/core/inference-executors/retry-policy.d.ts

package/dist/core/inference-executors/index.mjs

@@ -1,9 +1,9 @@
-import { n as requiredEnvFrom, t as envFrom } from "../../env-nV5rVErX.mjs";
+import { n as requiredEnvFrom, t as envFrom } from "../../env-BVYeJhGA.mjs";
 import process from "node:process";
 import { errorMessageFrom, errorNameFrom, sleep } from "@moeru/std";
 import { createOpenAI } from "@xsai-ext/providers/create";
 //#region src/core/inference-executors/retry-policy.ts
-const retryableStatusCodes = new Set([
+const retryableStatusCodes = /* @__PURE__ */ new Set([
 	408,
 	425,
 	429,
@@ -12,7 +12,7 @@ const retryableStatusCodes = new Set([
 	503,
 	504
 ]);
-const retryableErrorNames = new Set(["TimeoutError", "FetchError"]);
+const retryableErrorNames = /* @__PURE__ */ new Set(["TimeoutError", "FetchError"]);
 const retryableMessagePatterns = [
 	/rate limit/i,
 	/rate-limited/i,

package/dist/core/inference-executors/index.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"index.mjs","names":["defaultSleep"],"sources":["../../../src/core/inference-executors/retry-policy.ts","../../../src/core/inference-executors/adapters.ts","../../../src/core/inference-executors/remote-providers/openai/index.ts"],"sourcesContent":["import { sleep as defaultSleep, errorMessageFrom, errorNameFrom } from '@moeru/std'\n\n/**\n * Describes how provider retries should behave.\n *\n * ASCII flow:\n * attempt -> run request -> success return\n * attempt -> run request -> retriable failure -> sleep -> next attempt\n * attempt -> run request -> non-retriable failure -> throw\n */\nexport interface RetryPolicy {\n  /**\n   * Maximum number of total attempts, including the first try.\n   */\n  maxAttempts: number\n  /**\n   * Returns the wait time for a retry attempt.\n   */\n  delayMs: (attempt: number) => number\n  /**\n   * Determines whether an error can be retried safely.\n   */\n  shouldRetry: (error: unknown) => boolean\n  /**\n   * Suspends execution between retries.\n   */\n  sleep: (milliseconds: number) => Promise<void>\n}\n\n/**\n * Configures a retry policy before a provider call is executed.\n *\n * Use when:\n * - you want the default retry classifier but need to tune attempts or delay\n * - you need to replace the sleeper in tests\n *\n * Expects:\n * - `maxAttempts` to be a finite integer greater than or equal to `1`\n * - `delayMs` to return a non-negative wait time in milliseconds\n */\nexport interface RetryPolicyOptions {\n  /**\n   * Maximum total attempts, including the first request.\n   *\n   * @default 3\n   */\n  maxAttempts?: number\n  /**\n   * Computes the delay for a retry attempt.\n   *\n   * The attempt number starts at `1` for the first retry.\n   */\n  delayMs?: (attempt: number) => number\n  /**\n   * Overrides the retry classifier.\n   */\n  shouldRetry?: (error: unknown) => boolean\n  /**\n   * Overrides the sleeper used between attempts.\n   */\n  sleep?: (milliseconds: number) => Promise<void>\n}\n\nconst retryableStatusCodes = new Set([408, 425, 429, 500, 502, 503, 504])\nconst retryableErrorNames = new Set(['TimeoutError', 'FetchError'])\nconst retryableMessagePatterns = [\n  /rate limit/i,\n  /rate-limited/i,\n  /temporarily unavailable/i,\n  /service unavailable/i,\n  /server error/i,\n  /fetch failed/i,\n  /network error/i,\n  /socket hang up/i,\n  /econnreset/i,\n  /econnrefused/i,\n  /eai_again/i,\n  /etimedout/i,\n  /timed out/i,\n  /timeout/i,\n]\n\nfunction getStatusCode(error: unknown): number | undefined {\n  if (error == null || typeof error !== 'object') {\n    return undefined\n  }\n\n  const maybeStatusCode = (error as { statusCode?: unknown }).statusCode\n  if (typeof maybeStatusCode === 'number') {\n    return maybeStatusCode\n  }\n\n  const maybeStatus = (error as { status?: unknown }).status\n  if (typeof maybeStatus === 'number') {\n    return maybeStatus\n  }\n\n  const response = (error as { response?: unknown }).response\n  if (response == null || typeof response !== 'object') {\n    return undefined\n  }\n\n  const responseStatus = (response as { status?: unknown }).status\n  return typeof responseStatus === 'number' ? responseStatus : undefined\n}\n\n/**\n * Returns true when a provider failure is temporary and a retry is reasonable.\n *\n * Use when:\n * - the upstream failure is a transport problem or a 5xx/429 response\n *\n * Expects:\n * - provider errors to expose a status code, name, or message when possible\n */\nexport function isRetriableProviderError(error: unknown): boolean {\n  const statusCode = getStatusCode(error)\n\n  if (statusCode != null) {\n    return retryableStatusCodes.has(statusCode)\n  }\n\n  const errorName = errorNameFrom(error)\n  if (errorName != null && retryableErrorNames.has(errorName)) {\n    return true\n  }\n\n  const errorMessage = errorMessageFrom(error)\n  if (errorMessage == null) {\n    return false\n  }\n\n  return retryableMessagePatterns.some(pattern => pattern.test(errorMessage))\n}\n\nfunction defaultDelayMs(attempt: number): number {\n  return 500 * 2 ** (attempt - 1)\n}\n\n/**\n * Creates a retry policy for provider work.\n *\n * Use when:\n * - you need a reusable retry runner for eval-time provider calls\n * - you want to keep retry behavior deterministic in tests\n *\n * Expects:\n * - callers to treat `maxAttempts` as total attempts, not retries\n *\n * Throws:\n * - `RangeError` when `maxAttempts` is not a finite integer greater than or equal to `1`\n */\nfunction assertValidMaxAttempts(value: number): number {\n  if (!Number.isFinite(value) || !Number.isInteger(value) || value < 1) {\n    throw new RangeError('maxAttempts must be a finite integer greater than or equal to 1.')\n  }\n\n  return value\n}\n\nexport function createRetryPolicy(options: RetryPolicyOptions = {}): RetryPolicy {\n  const maxAttempts = assertValidMaxAttempts(options.maxAttempts ?? 3)\n\n  return {\n    maxAttempts,\n    delayMs: options.delayMs ?? defaultDelayMs,\n    shouldRetry: options.shouldRetry ?? isRetriableProviderError,\n    sleep: options.sleep ?? defaultSleep,\n  }\n}\n\n/**\n * Runs an operation with bounded retries.\n *\n * Use when:\n * - you are calling an LLM provider or other temporary upstream dependency\n * - non-retriable failures should bubble immediately\n *\n * Expects:\n * - the operation to be idempotent across attempts\n */\nexport async function runWithRetry<T>(operation: () => Promise<T>, policy: RetryPolicy = createRetryPolicy()): Promise<T> {\n  for (let attempt = 1; attempt <= policy.maxAttempts; attempt += 1) {\n    try {\n      return await operation()\n    }\n    catch (error) {\n      if (attempt >= policy.maxAttempts || !policy.shouldRetry(error)) {\n        throw error\n      }\n\n      const delayMilliseconds = policy.delayMs(attempt)\n      if (delayMilliseconds > 0) {\n        await policy.sleep(delayMilliseconds)\n      }\n    }\n  }\n\n  throw new Error('Retry loop exited without returning a value.')\n}\n","import type { RetryPolicy, RetryPolicyOptions } from './retry-policy'\n\nimport { createRetryPolicy, runWithRetry } from './retry-policy'\n\n/**\n * Bundles a provider with the retry policy used to call it.\n *\n * Use when:\n * - a provider instance should travel with the retry runner that governs it\n * - you want call sites to share one retry configuration object\n */\nexport interface ProviderAdapter<TProvider> {\n  /**\n   * The underlying provider instance.\n   */\n  provider: TProvider\n  /**\n   * The retry policy used for provider calls.\n   */\n  retryPolicy: RetryPolicy\n  /**\n   * Runs a provider-dependent operation with the adapter retry policy.\n   */\n  runWithRetry: <TResult>(operation: () => Promise<TResult>) => Promise<TResult>\n}\n\n/**\n * Creates a provider adapter with the default retry policy.\n *\n * Use when:\n * - you have a provider instance and want a consistent retry wrapper\n *\n * Expects:\n * - the provider to be safe to reuse across attempts\n */\nexport function createProviderAdapter<TProvider>(provider: TProvider, options: RetryPolicyOptions = {}): ProviderAdapter<TProvider> {\n  const retryPolicy = createRetryPolicy(options)\n\n  return {\n    provider,\n    retryPolicy,\n    runWithRetry: operation => runWithRetry(operation, retryPolicy),\n  }\n}\n","import type { ProviderAdapter } from '../../adapters'\nimport type { RetryPolicyOptions } from '../../retry-policy'\n\nimport process from 'node:process'\n\nimport { createOpenAI } from '@xsai-ext/providers/create'\n\nimport { createProviderAdapter } from '../../adapters'\nimport { envFrom, requiredEnvFrom } from '../../env'\n\n/**\n * Represents the OpenAI provider instance returned by xsai.\n */\nexport type OpenAIProvider = ReturnType<typeof createOpenAI>\n\n/**\n * Represents the OpenAI adapter used by vieval.\n */\nexport type OpenAIProviderAdapter = ProviderAdapter<OpenAIProvider>\n\n/**\n * Configures env key names and source for OpenAI provider setup.\n */\nexport interface OpenAIEnvSourceOptions {\n  /**\n   * Environment object used for variable lookup.\n   *\n   * @default process.env\n   */\n  env?: NodeJS.ProcessEnv\n  /**\n   * Env key name for API key.\n   *\n   * @default 'OPENAI_API_KEY'\n   */\n  apiKey?: string\n  /**\n   * Env key name for base URL.\n   *\n   * @default 'OPENAI_BASE_URL'\n   */\n  baseURL?: string\n  /**\n   * Env key name for model.\n   *\n   * @default 'OPENAI_MODEL'\n   */\n  model?: string\n}\n\n/**\n * Configures fallback defaults when env values are missing.\n */\nexport interface OpenAIFromEnvDefaultOptions {\n  /**\n   * API key fallback value.\n   */\n  apiKey?: string\n  /**\n   * Base URL fallback value.\n   */\n  baseURL?: string\n  /**\n   * Model fallback value.\n   */\n  model?: string\n  /**\n   * Retry policy override passed to provider adapter.\n   */\n  retryOptions?: RetryPolicyOptions\n}\n\n/**\n * Result produced by `createOpenAIFromEnv`.\n */\nexport interface OpenAIFromEnvResult {\n  adapter: OpenAIProviderAdapter\n  apiKey: string\n  baseURL?: string\n  model: string\n}\n\n/**\n * Minimal response shape returned by text-generation calls.\n */\nexport interface OpenAITextGenerationResult {\n  /**\n   * Text output from the provider.\n   *\n   * Some OpenAI-compatible implementations may return `null`.\n   */\n  text?: string | null\n}\n\n/**\n * Normalizes provider text output to a safe string.\n *\n * Before: `{ text: null }`\n * After: `''`\n *\n * Before: `{ text: 'hello' }`\n * After: `'hello'`\n */\nexport function normalizeOpenAITextOutput(result: OpenAITextGenerationResult): string {\n  return typeof result.text === 'string' ? result.text : ''\n}\n\n/**\n * Creates an OpenAI provider adapter using environment variables with defaults.\n *\n * Example:\n * `const runtime = createOpenAIFromEnv({}, { model: 'gpt-4.1-mini' })`\n */\nexport function createOpenAIFromEnv(\n  source: OpenAIEnvSourceOptions = {},\n  defaults: OpenAIFromEnvDefaultOptions = {},\n): OpenAIFromEnvResult {\n  const env = source.env ?? process.env\n  const apiKeyEnvKey = source.apiKey ?? 'OPENAI_API_KEY'\n  const baseURLEnvKey = source.baseURL ?? 'OPENAI_BASE_URL'\n  const modelEnvKey = source.model ?? 'OPENAI_MODEL'\n\n  const envWithDefaults = {\n    ...(defaults.apiKey == null ? {} : { [apiKeyEnvKey]: defaults.apiKey }),\n    ...(defaults.baseURL == null ? {} : { [baseURLEnvKey]: defaults.baseURL }),\n    ...(defaults.model == null ? {} : { [modelEnvKey]: defaults.model }),\n    ...env,\n  }\n\n  const apiKey = requiredEnvFrom(envWithDefaults, {\n    name: apiKeyEnvKey,\n    type: 'string',\n  })\n  const model = requiredEnvFrom(envWithDefaults, {\n    name: modelEnvKey,\n    type: 'string',\n  })\n  const baseURL = envFrom(envWithDefaults, {\n    name: baseURLEnvKey,\n    type: 'string',\n  })\n  const adapter = createOpenAIProviderAdapter(apiKey, baseURL, defaults.retryOptions)\n\n  return {\n    adapter,\n    apiKey,\n    baseURL,\n    model,\n  }\n}\n\n/**\n * Creates an OpenAI provider adapter for eval-time requests.\n *\n * Use when:\n * - an eval needs the OpenAI SDK surface plus the shared retry runner\n *\n * Expects:\n * - `apiKey` and `baseURL` to point at an OpenAI-compatible endpoint\n * - `retryOptions` to follow the same invariants as `createRetryPolicy`\n */\nexport function createOpenAIProviderAdapter(apiKey: string, baseURL?: string, retryOptions: RetryPolicyOptions = {}): OpenAIProviderAdapter {\n  return createProviderAdapter(createOpenAI(apiKey, baseURL), retryOptions)\n}\n"],"mappings":";;;;;AA+DA,MAAM,uBAAuB,IAAI,IAAI;CAAC;CAAK;CAAK;CAAK;CAAK;CAAK;CAAK;CAAI,CAAC;AACzE,MAAM,sBAAsB,IAAI,IAAI,CAAC,gBAAgB,aAAa,CAAC;AACnE,MAAM,2BAA2B;CAC/B;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACD;AAED,SAAS,cAAc,OAAoC;AACzD,KAAI,SAAS,QAAQ,OAAO,UAAU,SACpC;CAGF,MAAM,kBAAmB,MAAmC;AAC5D,KAAI,OAAO,oBAAoB,SAC7B,QAAO;CAGT,MAAM,cAAe,MAA+B;AACpD,KAAI,OAAO,gBAAgB,SACzB,QAAO;CAGT,MAAM,WAAY,MAAiC;AACnD,KAAI,YAAY,QAAQ,OAAO,aAAa,SAC1C;CAGF,MAAM,iBAAkB,SAAkC;AAC1D,QAAO,OAAO,mBAAmB,WAAW,iBAAiB,KAAA;;;;;;;;;;;AAY/D,SAAgB,yBAAyB,OAAyB;CAChE,MAAM,aAAa,cAAc,MAAM;AAEvC,KAAI,cAAc,KAChB,QAAO,qBAAqB,IAAI,WAAW;CAG7C,MAAM,YAAY,cAAc,MAAM;AACtC,KAAI,aAAa,QAAQ,oBAAoB,IAAI,UAAU,CACzD,QAAO;CAGT,MAAM,eAAe,iBAAiB,MAAM;AAC5C,KAAI,gBAAgB,KAClB,QAAO;AAGT,QAAO,yBAAyB,MAAK,YAAW,QAAQ,KAAK,aAAa,CAAC;;AAG7E,SAAS,eAAe,SAAyB;AAC/C,QAAO,MAAM,MAAM,UAAU;;;;;;;;;;;;;;;AAgB/B,SAAS,uBAAuB,OAAuB;AACrD,KAAI,CAAC,OAAO,SAAS,MAAM,IAAI,CAAC,OAAO,UAAU,MAAM,IAAI,QAAQ,EACjE,OAAM,IAAI,WAAW,mEAAmE;AAG1F,QAAO;;AAGT,SAAgB,kBAAkB,UAA8B,EAAE,EAAe;AAG/E,QAAO;EACL,aAHkB,uBAAuB,QAAQ,eAAe,EAAE;EAIlE,SAAS,QAAQ,WAAW;EAC5B,aAAa,QAAQ,eAAe;EACpC,OAAO,QAAQ,SAASA;EACzB;;;;;;;;;;;;AAaH,eAAsB,aAAgB,WAA6B,SAAsB,mBAAmB,EAAc;AACxH,MAAK,IAAI,UAAU,GAAG,WAAW,OAAO,aAAa,WAAW,EAC9D,KAAI;AACF,SAAO,MAAM,WAAW;UAEnB,OAAO;AACZ,MAAI,WAAW,OAAO,eAAe,CAAC,OAAO,YAAY,MAAM,CAC7D,OAAM;EAGR,MAAM,oBAAoB,OAAO,QAAQ,QAAQ;AACjD,MAAI,oBAAoB,EACtB,OAAM,OAAO,MAAM,kBAAkB;;AAK3C,OAAM,IAAI,MAAM,+CAA+C;;;;;;;;;;;;;ACnKjE,SAAgB,sBAAiC,UAAqB,UAA8B,EAAE,EAA8B;CAClI,MAAM,cAAc,kBAAkB,QAAQ;AAE9C,QAAO;EACL;EACA;EACA,eAAc,cAAa,aAAa,WAAW,YAAY;EAChE;;;;;;;;;;;;;AC6DH,SAAgB,0BAA0B,QAA4C;AACpF,QAAO,OAAO,OAAO,SAAS,WAAW,OAAO,OAAO;;;;;;;;AASzD,SAAgB,oBACd,SAAiC,EAAE,EACnC,WAAwC,EAAE,EACrB;CACrB,MAAM,MAAM,OAAO,OAAO,QAAQ;CAClC,MAAM,eAAe,OAAO,UAAU;CACtC,MAAM,gBAAgB,OAAO,WAAW;CACxC,MAAM,cAAc,OAAO,SAAS;CAEpC,MAAM,kBAAkB;EACtB,GAAI,SAAS,UAAU,OAAO,EAAE,GAAG,GAAG,eAAe,SAAS,QAAQ;EACtE,GAAI,SAAS,WAAW,OAAO,EAAE,GAAG,GAAG,gBAAgB,SAAS,SAAS;EACzE,GAAI,SAAS,SAAS,OAAO,EAAE,GAAG,GAAG,cAAc,SAAS,OAAO;EACnE,GAAG;EACJ;CAED,MAAM,SAAS,gBAAgB,iBAAiB;EAC9C,MAAM;EACN,MAAM;EACP,CAAC;CACF,MAAM,QAAQ,gBAAgB,iBAAiB;EAC7C,MAAM;EACN,MAAM;EACP,CAAC;CACF,MAAM,UAAU,QAAQ,iBAAiB;EACvC,MAAM;EACN,MAAM;EACP,CAAC;AAGF,QAAO;EACL,SAHc,4BAA4B,QAAQ,SAAS,SAAS,aAAa;EAIjF;EACA;EACA;EACD;;;;;;;;;;;;AAaH,SAAgB,4BAA4B,QAAgB,SAAkB,eAAmC,EAAE,EAAyB;AAC1I,QAAO,sBAAsB,aAAa,QAAQ,QAAQ,EAAE,aAAa"}
+{"version":3,"file":"index.mjs","names":["defaultSleep"],"sources":["../../../src/core/inference-executors/retry-policy.ts","../../../src/core/inference-executors/adapters.ts","../../../src/core/inference-executors/remote-providers/openai/index.ts"],"sourcesContent":["import { sleep as defaultSleep, errorMessageFrom, errorNameFrom } from '@moeru/std'\n\n/**\n * Describes how provider retries should behave.\n *\n * ASCII flow:\n * attempt -> run request -> success return\n * attempt -> run request -> retriable failure -> sleep -> next attempt\n * attempt -> run request -> non-retriable failure -> throw\n */\nexport interface RetryPolicy {\n  /**\n   * Maximum number of total attempts, including the first try.\n   */\n  maxAttempts: number\n  /**\n   * Returns the wait time for a retry attempt.\n   */\n  delayMs: (attempt: number) => number\n  /**\n   * Determines whether an error can be retried safely.\n   */\n  shouldRetry: (error: unknown) => boolean\n  /**\n   * Suspends execution between retries.\n   */\n  sleep: (milliseconds: number) => Promise<void>\n}\n\n/**\n * Configures a retry policy before a provider call is executed.\n *\n * Use when:\n * - you want the default retry classifier but need to tune attempts or delay\n * - you need to replace the sleeper in tests\n *\n * Expects:\n * - `maxAttempts` to be a finite integer greater than or equal to `1`\n * - `delayMs` to return a non-negative wait time in milliseconds\n */\nexport interface RetryPolicyOptions {\n  /**\n   * Maximum total attempts, including the first request.\n   *\n   * @default 3\n   */\n  maxAttempts?: number\n  /**\n   * Computes the delay for a retry attempt.\n   *\n   * The attempt number starts at `1` for the first retry.\n   */\n  delayMs?: (attempt: number) => number\n  /**\n   * Overrides the retry classifier.\n   */\n  shouldRetry?: (error: unknown) => boolean\n  /**\n   * Overrides the sleeper used between attempts.\n   */\n  sleep?: (milliseconds: number) => Promise<void>\n}\n\nconst retryableStatusCodes = new Set([408, 425, 429, 500, 502, 503, 504])\nconst retryableErrorNames = new Set(['TimeoutError', 'FetchError'])\nconst retryableMessagePatterns = [\n  /rate limit/i,\n  /rate-limited/i,\n  /temporarily unavailable/i,\n  /service unavailable/i,\n  /server error/i,\n  /fetch failed/i,\n  /network error/i,\n  /socket hang up/i,\n  /econnreset/i,\n  /econnrefused/i,\n  /eai_again/i,\n  /etimedout/i,\n  /timed out/i,\n  /timeout/i,\n]\n\nfunction getStatusCode(error: unknown): number | undefined {\n  if (error == null || typeof error !== 'object') {\n    return undefined\n  }\n\n  const maybeStatusCode = (error as { statusCode?: unknown }).statusCode\n  if (typeof maybeStatusCode === 'number') {\n    return maybeStatusCode\n  }\n\n  const maybeStatus = (error as { status?: unknown }).status\n  if (typeof maybeStatus === 'number') {\n    return maybeStatus\n  }\n\n  const response = (error as { response?: unknown }).response\n  if (response == null || typeof response !== 'object') {\n    return undefined\n  }\n\n  const responseStatus = (response as { status?: unknown }).status\n  return typeof responseStatus === 'number' ? responseStatus : undefined\n}\n\n/**\n * Returns true when a provider failure is temporary and a retry is reasonable.\n *\n * Use when:\n * - the upstream failure is a transport problem or a 5xx/429 response\n *\n * Expects:\n * - provider errors to expose a status code, name, or message when possible\n */\nexport function isRetriableProviderError(error: unknown): boolean {\n  const statusCode = getStatusCode(error)\n\n  if (statusCode != null) {\n    return retryableStatusCodes.has(statusCode)\n  }\n\n  const errorName = errorNameFrom(error)\n  if (errorName != null && retryableErrorNames.has(errorName)) {\n    return true\n  }\n\n  const errorMessage = errorMessageFrom(error)\n  if (errorMessage == null) {\n    return false\n  }\n\n  return retryableMessagePatterns.some(pattern => pattern.test(errorMessage))\n}\n\nfunction defaultDelayMs(attempt: number): number {\n  return 500 * 2 ** (attempt - 1)\n}\n\n/**\n * Creates a retry policy for provider work.\n *\n * Use when:\n * - you need a reusable retry runner for eval-time provider calls\n * - you want to keep retry behavior deterministic in tests\n *\n * Expects:\n * - callers to treat `maxAttempts` as total attempts, not retries\n *\n * Throws:\n * - `RangeError` when `maxAttempts` is not a finite integer greater than or equal to `1`\n */\nfunction assertValidMaxAttempts(value: number): number {\n  if (!Number.isFinite(value) || !Number.isInteger(value) || value < 1) {\n    throw new RangeError('maxAttempts must be a finite integer greater than or equal to 1.')\n  }\n\n  return value\n}\n\nexport function createRetryPolicy(options: RetryPolicyOptions = {}): RetryPolicy {\n  const maxAttempts = assertValidMaxAttempts(options.maxAttempts ?? 3)\n\n  return {\n    maxAttempts,\n    delayMs: options.delayMs ?? defaultDelayMs,\n    shouldRetry: options.shouldRetry ?? isRetriableProviderError,\n    sleep: options.sleep ?? defaultSleep,\n  }\n}\n\n/**\n * Runs an operation with bounded retries.\n *\n * Use when:\n * - you are calling an LLM provider or other temporary upstream dependency\n * - non-retriable failures should bubble immediately\n *\n * Expects:\n * - the operation to be idempotent across attempts\n */\nexport async function runWithRetry<T>(operation: () => Promise<T>, policy: RetryPolicy = createRetryPolicy()): Promise<T> {\n  for (let attempt = 1; attempt <= policy.maxAttempts; attempt += 1) {\n    try {\n      return await operation()\n    }\n    catch (error) {\n      if (attempt >= policy.maxAttempts || !policy.shouldRetry(error)) {\n        throw error\n      }\n\n      const delayMilliseconds = policy.delayMs(attempt)\n      if (delayMilliseconds > 0) {\n        await policy.sleep(delayMilliseconds)\n      }\n    }\n  }\n\n  throw new Error('Retry loop exited without returning a value.')\n}\n","import type { RetryPolicy, RetryPolicyOptions } from './retry-policy'\n\nimport { createRetryPolicy, runWithRetry } from './retry-policy'\n\n/**\n * Bundles a provider with the retry policy used to call it.\n *\n * Use when:\n * - a provider instance should travel with the retry runner that governs it\n * - you want call sites to share one retry configuration object\n */\nexport interface ProviderAdapter<TProvider> {\n  /**\n   * The underlying provider instance.\n   */\n  provider: TProvider\n  /**\n   * The retry policy used for provider calls.\n   */\n  retryPolicy: RetryPolicy\n  /**\n   * Runs a provider-dependent operation with the adapter retry policy.\n   */\n  runWithRetry: <TResult>(operation: () => Promise<TResult>) => Promise<TResult>\n}\n\n/**\n * Creates a provider adapter with the default retry policy.\n *\n * Use when:\n * - you have a provider instance and want a consistent retry wrapper\n *\n * Expects:\n * - the provider to be safe to reuse across attempts\n */\nexport function createProviderAdapter<TProvider>(provider: TProvider, options: RetryPolicyOptions = {}): ProviderAdapter<TProvider> {\n  const retryPolicy = createRetryPolicy(options)\n\n  return {\n    provider,\n    retryPolicy,\n    runWithRetry: operation => runWithRetry(operation, retryPolicy),\n  }\n}\n","import type { ProviderAdapter } from '../../adapters'\nimport type { RetryPolicyOptions } from '../../retry-policy'\n\nimport process from 'node:process'\n\nimport { createOpenAI } from '@xsai-ext/providers/create'\n\nimport { createProviderAdapter } from '../../adapters'\nimport { envFrom, requiredEnvFrom } from '../../env'\n\n/**\n * Represents the OpenAI provider instance returned by xsai.\n */\nexport type OpenAIProvider = ReturnType<typeof createOpenAI>\n\n/**\n * Represents the OpenAI adapter used by vieval.\n */\nexport type OpenAIProviderAdapter = ProviderAdapter<OpenAIProvider>\n\n/**\n * Configures env key names and source for OpenAI provider setup.\n */\nexport interface OpenAIEnvSourceOptions {\n  /**\n   * Environment object used for variable lookup.\n   *\n   * @default process.env\n   */\n  env?: NodeJS.ProcessEnv\n  /**\n   * Env key name for API key.\n   *\n   * @default 'OPENAI_API_KEY'\n   */\n  apiKey?: string\n  /**\n   * Env key name for base URL.\n   *\n   * @default 'OPENAI_BASE_URL'\n   */\n  baseURL?: string\n  /**\n   * Env key name for model.\n   *\n   * @default 'OPENAI_MODEL'\n   */\n  model?: string\n}\n\n/**\n * Configures fallback defaults when env values are missing.\n */\nexport interface OpenAIFromEnvDefaultOptions {\n  /**\n   * API key fallback value.\n   */\n  apiKey?: string\n  /**\n   * Base URL fallback value.\n   */\n  baseURL?: string\n  /**\n   * Model fallback value.\n   */\n  model?: string\n  /**\n   * Retry policy override passed to provider adapter.\n   */\n  retryOptions?: RetryPolicyOptions\n}\n\n/**\n * Result produced by `createOpenAIFromEnv`.\n */\nexport interface OpenAIFromEnvResult {\n  adapter: OpenAIProviderAdapter\n  apiKey: string\n  baseURL?: string\n  model: string\n}\n\n/**\n * Minimal response shape returned by text-generation calls.\n */\nexport interface OpenAITextGenerationResult {\n  /**\n   * Text output from the provider.\n   *\n   * Some OpenAI-compatible implementations may return `null`.\n   */\n  text?: string | null\n}\n\n/**\n * Normalizes provider text output to a safe string.\n *\n * Before: `{ text: null }`\n * After: `''`\n *\n * Before: `{ text: 'hello' }`\n * After: `'hello'`\n */\nexport function normalizeOpenAITextOutput(result: OpenAITextGenerationResult): string {\n  return typeof result.text === 'string' ? result.text : ''\n}\n\n/**\n * Creates an OpenAI provider adapter using environment variables with defaults.\n *\n * Example:\n * `const runtime = createOpenAIFromEnv({}, { model: 'gpt-4.1-mini' })`\n */\nexport function createOpenAIFromEnv(\n  source: OpenAIEnvSourceOptions = {},\n  defaults: OpenAIFromEnvDefaultOptions = {},\n): OpenAIFromEnvResult {\n  const env = source.env ?? process.env\n  const apiKeyEnvKey = source.apiKey ?? 'OPENAI_API_KEY'\n  const baseURLEnvKey = source.baseURL ?? 'OPENAI_BASE_URL'\n  const modelEnvKey = source.model ?? 'OPENAI_MODEL'\n\n  const envWithDefaults = {\n    ...(defaults.apiKey == null ? {} : { [apiKeyEnvKey]: defaults.apiKey }),\n    ...(defaults.baseURL == null ? {} : { [baseURLEnvKey]: defaults.baseURL }),\n    ...(defaults.model == null ? {} : { [modelEnvKey]: defaults.model }),\n    ...env,\n  }\n\n  const apiKey = requiredEnvFrom(envWithDefaults, {\n    name: apiKeyEnvKey,\n    type: 'string',\n  })\n  const model = requiredEnvFrom(envWithDefaults, {\n    name: modelEnvKey,\n    type: 'string',\n  })\n  const baseURL = envFrom(envWithDefaults, {\n    name: baseURLEnvKey,\n    type: 'string',\n  })\n  const adapter = createOpenAIProviderAdapter(apiKey, baseURL, defaults.retryOptions)\n\n  return {\n    adapter,\n    apiKey,\n    baseURL,\n    model,\n  }\n}\n\n/**\n * Creates an OpenAI provider adapter for eval-time requests.\n *\n * Use when:\n * - an eval needs the OpenAI SDK surface plus the shared retry runner\n *\n * Expects:\n * - `apiKey` and `baseURL` to point at an OpenAI-compatible endpoint\n * - `retryOptions` to follow the same invariants as `createRetryPolicy`\n */\nexport function createOpenAIProviderAdapter(apiKey: string, baseURL?: string, retryOptions: RetryPolicyOptions = {}): OpenAIProviderAdapter {\n  return createProviderAdapter(createOpenAI(apiKey, baseURL), retryOptions)\n}\n"],"mappings":";;;;;AA+DA,MAAM,uCAAuB,IAAI,IAAI;CAAC;CAAK;CAAK;CAAK;CAAK;CAAK;CAAK;AAAG,CAAC;AACxE,MAAM,sCAAsB,IAAI,IAAI,CAAC,gBAAgB,YAAY,CAAC;AAClE,MAAM,2BAA2B;CAC/B;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF;AAEA,SAAS,cAAc,OAAoC;CACzD,IAAI,SAAS,QAAQ,OAAO,UAAU,UACpC;CAGF,MAAM,kBAAmB,MAAmC;CAC5D,IAAI,OAAO,oBAAoB,UAC7B,OAAO;CAGT,MAAM,cAAe,MAA+B;CACpD,IAAI,OAAO,gBAAgB,UACzB,OAAO;CAGT,MAAM,WAAY,MAAiC;CACnD,IAAI,YAAY,QAAQ,OAAO,aAAa,UAC1C;CAGF,MAAM,iBAAkB,SAAkC;CAC1D,OAAO,OAAO,mBAAmB,WAAW,iBAAiB,KAAA;AAC/D;;;;;;;;;;AAWA,SAAgB,yBAAyB,OAAyB;CAChE,MAAM,aAAa,cAAc,KAAK;CAEtC,IAAI,cAAc,MAChB,OAAO,qBAAqB,IAAI,UAAU;CAG5C,MAAM,YAAY,cAAc,KAAK;CACrC,IAAI,aAAa,QAAQ,oBAAoB,IAAI,SAAS,GACxD,OAAO;CAGT,MAAM,eAAe,iBAAiB,KAAK;CAC3C,IAAI,gBAAgB,MAClB,OAAO;CAGT,OAAO,yBAAyB,MAAK,YAAW,QAAQ,KAAK,YAAY,CAAC;AAC5E;AAEA,SAAS,eAAe,SAAyB;CAC/C,OAAO,MAAM,MAAM,UAAU;AAC/B;;;;;;;;;;;;;;AAeA,SAAS,uBAAuB,OAAuB;CACrD,IAAI,CAAC,OAAO,SAAS,KAAK,KAAK,CAAC,OAAO,UAAU,KAAK,KAAK,QAAQ,GACjE,MAAM,IAAI,WAAW,kEAAkE;CAGzF,OAAO;AACT;AAEA,SAAgB,kBAAkB,UAA8B,CAAC,GAAgB;CAG/E,OAAO;EACL,aAHkB,uBAAuB,QAAQ,eAAe,CAGtD;EACV,SAAS,QAAQ,WAAW;EAC5B,aAAa,QAAQ,eAAe;EACpC,OAAO,QAAQ,SAASA;CAC1B;AACF;;;;;;;;;;;AAYA,eAAsB,aAAgB,WAA6B,SAAsB,kBAAkB,GAAe;CACxH,KAAK,IAAI,UAAU,GAAG,WAAW,OAAO,aAAa,WAAW,GAC9D,IAAI;EACF,OAAO,MAAM,UAAU;CACzB,SACO,OAAO;EACZ,IAAI,WAAW,OAAO,eAAe,CAAC,OAAO,YAAY,KAAK,GAC5D,MAAM;EAGR,MAAM,oBAAoB,OAAO,QAAQ,OAAO;EAChD,IAAI,oBAAoB,GACtB,MAAM,OAAO,MAAM,iBAAiB;CAExC;CAGF,MAAM,IAAI,MAAM,8CAA8C;AAChE;;;;;;;;;;;;ACpKA,SAAgB,sBAAiC,UAAqB,UAA8B,CAAC,GAA+B;CAClI,MAAM,cAAc,kBAAkB,OAAO;CAE7C,OAAO;EACL;EACA;EACA,eAAc,cAAa,aAAa,WAAW,WAAW;CAChE;AACF;;;;;;;;;;;;AC4DA,SAAgB,0BAA0B,QAA4C;CACpF,OAAO,OAAO,OAAO,SAAS,WAAW,OAAO,OAAO;AACzD;;;;;;;AAQA,SAAgB,oBACd,SAAiC,CAAC,GAClC,WAAwC,CAAC,GACpB;CACrB,MAAM,MAAM,OAAO,OAAO,QAAQ;CAClC,MAAM,eAAe,OAAO,UAAU;CACtC,MAAM,gBAAgB,OAAO,WAAW;CACxC,MAAM,cAAc,OAAO,SAAS;CAEpC,MAAM,kBAAkB;EACtB,GAAI,SAAS,UAAU,OAAO,CAAC,IAAI,GAAG,eAAe,SAAS,OAAO;EACrE,GAAI,SAAS,WAAW,OAAO,CAAC,IAAI,GAAG,gBAAgB,SAAS,QAAQ;EACxE,GAAI,SAAS,SAAS,OAAO,CAAC,IAAI,GAAG,cAAc,SAAS,MAAM;EAClE,GAAG;CACL;CAEA,MAAM,SAAS,gBAAgB,iBAAiB;EAC9C,MAAM;EACN,MAAM;CACR,CAAC;CACD,MAAM,QAAQ,gBAAgB,iBAAiB;EAC7C,MAAM;EACN,MAAM;CACR,CAAC;CACD,MAAM,UAAU,QAAQ,iBAAiB;EACvC,MAAM;EACN,MAAM;CACR,CAAC;CAGD,OAAO;EACL,SAHc,4BAA4B,QAAQ,SAAS,SAAS,YAG9D;EACN;EACA;EACA;CACF;AACF;;;;;;;;;;;AAYA,SAAgB,4BAA4B,QAAgB,SAAkB,eAAmC,CAAC,GAA0B;CAC1I,OAAO,sBAAsB,aAAa,QAAQ,OAAO,GAAG,YAAY;AAC1E"}

package/dist/core/processors/results/index.d.mts

@@ -1,4 +1,4 @@
-import { K as AggregatedRunResults } from "../../../index-BkjyCInx.mjs";
+import { K as AggregatedRunResults } from "../../../index-CwKBlCG9.mjs";
 
 //#region src/core/processors/results/policies/hybrid-threshold.d.ts
 /**

package/dist/core/processors/results/index.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"index.mjs","names":[],"sources":["../../../../src/core/processors/results/policies/hybrid-threshold.ts","../../../../src/core/processors/results/policies/max-failed-runs.ts","../../../../src/core/processors/results/index.ts"],"sourcesContent":["import type { AggregatedRunResults } from '../../../runner/aggregate'\n\n/**\n * Violation emitted when result policies fail.\n */\nexport interface ResultPolicyViolation {\n  /**\n   * Stable policy id.\n   */\n  policyId: string\n  /**\n   * Human-readable violation reason.\n   */\n  reason: string\n}\n\n/**\n * Configures hybrid-threshold policy behavior.\n */\nexport interface HybridThresholdPolicyOptions {\n  /**\n   * Minimum required overall hybrid score.\n   *\n   * @default 0.7\n   */\n  minOverallHybridScore?: number\n  /**\n   * Minimum required inferenceExecutor hybrid score.\n   *\n   * @default 0.6\n   */\n  minProviderHybridScore?: number\n}\n\n/**\n * Evaluates threshold policy against aggregated results.\n */\nexport function evaluateHybridThresholdPolicy(\n  results: AggregatedRunResults,\n  options: HybridThresholdPolicyOptions = {},\n): ResultPolicyViolation[] {\n  const minOverallHybridScore = options.minOverallHybridScore ?? 0.7\n  const minProviderHybridScore = options.minProviderHybridScore ?? 0.6\n\n  const violations: ResultPolicyViolation[] = []\n\n  const overallHybridAverage = results.overall.hybridAverage\n  if (overallHybridAverage == null || overallHybridAverage < minOverallHybridScore) {\n    violations.push({\n      policyId: 'threshold:overall-hybrid',\n      reason: `Overall hybrid average ${overallHybridAverage ?? 'null'} is below ${minOverallHybridScore}.`,\n    })\n  }\n\n  for (const inferenceExecutor of results.inferenceExecutors) {\n    if (inferenceExecutor.hybridAverage == null || inferenceExecutor.hybridAverage < minProviderHybridScore) {\n      violations.push({\n        policyId: 'threshold:inferenceExecutor-hybrid',\n        reason: `Provider ${inferenceExecutor.inferenceExecutorId} hybrid average ${inferenceExecutor.hybridAverage ?? 'null'} is below ${minProviderHybridScore}.`,\n      })\n    }\n  }\n\n  return violations\n}\n","import type { AggregatedRunResults } from '../../../runner/aggregate'\nimport type { ResultPolicyViolation } from './hybrid-threshold'\n\n/**\n * Configures hard-limit policy for failed runs.\n */\nexport interface MaxFailedRunsPolicyOptions {\n  /**\n   * Maximum allowed failed run count.\n   *\n   * @default 0\n   */\n  maxFailedRuns?: number\n  /**\n   * Hybrid score threshold below which a run counts as failed.\n   *\n   * @default 0.6\n   */\n  minRunHybridScore?: number\n}\n\n/**\n * Evaluates hard-limit policy for failed runs.\n */\nexport function evaluateMaxFailedRunsPolicy(\n  results: AggregatedRunResults,\n  options: MaxFailedRunsPolicyOptions = {},\n): ResultPolicyViolation[] {\n  const maxFailedRuns = options.maxFailedRuns ?? 0\n  const minRunHybridScore = options.minRunHybridScore ?? 0.6\n\n  const failedRuns = results.runs.filter((run) => {\n    if (run.hybridAverage == null) {\n      return true\n    }\n\n    return run.hybridAverage < minRunHybridScore\n  })\n\n  if (failedRuns.length <= maxFailedRuns) {\n    return []\n  }\n\n  return [{\n    policyId: 'hard-limit:max-failed-runs',\n    reason: `Failed runs ${failedRuns.length} exceed maxFailedRuns ${maxFailedRuns} with minRunHybridScore ${minRunHybridScore}.`,\n  }]\n}\n","import type { AggregatedRunResults } from '../../runner/aggregate'\nimport type { HybridThresholdPolicyOptions, ResultPolicyViolation } from './policies/hybrid-threshold'\nimport type { MaxFailedRunsPolicyOptions } from './policies/max-failed-runs'\n\nimport { evaluateHybridThresholdPolicy } from './policies/hybrid-threshold'\nimport { evaluateMaxFailedRunsPolicy } from './policies/max-failed-runs'\n\n/**\n * Configures result-processing policies for eval gating.\n */\nexport interface ProcessRunResultsOptions {\n  /**\n   * Threshold policy options.\n   */\n  threshold?: HybridThresholdPolicyOptions\n  /**\n   * Hard-limit failed-run policy options.\n   */\n  maxFailedRuns?: MaxFailedRunsPolicyOptions\n}\n\n/**\n * Final gate decision returned by result processors.\n */\nexport interface ResultGateDecision {\n  /**\n   * Whether the result batch passes all policies.\n   */\n  pass: boolean\n  /**\n   * Collected policy violations.\n   */\n  violations: ResultPolicyViolation[]\n}\n\n/**\n * Processes aggregated run results through built-in gating policies.\n *\n * Call stack:\n *\n * {@link runScheduledTasks}\n *   -> {@link aggregateRunResults}\n *     -> {@link processRunResults}\n *       -> {@link evaluateHybridThresholdPolicy}\n *       -> {@link evaluateMaxFailedRunsPolicy}\n *         -> {@link ResultGateDecision}\n */\nexport function processRunResults(\n  results: AggregatedRunResults,\n  options: ProcessRunResultsOptions = {},\n): ResultGateDecision {\n  const thresholdViolations = evaluateHybridThresholdPolicy(results, options.threshold)\n  const maxFailedRunsViolations = evaluateMaxFailedRunsPolicy(results, options.maxFailedRuns)\n\n  const violations = [\n    ...thresholdViolations,\n    ...maxFailedRunsViolations,\n  ]\n\n  return {\n    pass: violations.length === 0,\n    violations,\n  }\n}\n\nexport type {\n  HybridThresholdPolicyOptions,\n  MaxFailedRunsPolicyOptions,\n  ResultPolicyViolation,\n}\n"],"mappings":";;;;AAqCA,SAAgB,8BACd,SACA,UAAwC,EAAE,EACjB;CACzB,MAAM,wBAAwB,QAAQ,yBAAyB;CAC/D,MAAM,yBAAyB,QAAQ,0BAA0B;CAEjE,MAAM,aAAsC,EAAE;CAE9C,MAAM,uBAAuB,QAAQ,QAAQ;AAC7C,KAAI,wBAAwB,QAAQ,uBAAuB,sBACzD,YAAW,KAAK;EACd,UAAU;EACV,QAAQ,0BAA0B,wBAAwB,OAAO,YAAY,sBAAsB;EACpG,CAAC;AAGJ,MAAK,MAAM,qBAAqB,QAAQ,mBACtC,KAAI,kBAAkB,iBAAiB,QAAQ,kBAAkB,gBAAgB,uBAC/E,YAAW,KAAK;EACd,UAAU;EACV,QAAQ,YAAY,kBAAkB,oBAAoB,kBAAkB,kBAAkB,iBAAiB,OAAO,YAAY,uBAAuB;EAC1J,CAAC;AAIN,QAAO;;;;;;;ACvCT,SAAgB,4BACd,SACA,UAAsC,EAAE,EACf;CACzB,MAAM,gBAAgB,QAAQ,iBAAiB;CAC/C,MAAM,oBAAoB,QAAQ,qBAAqB;CAEvD,MAAM,aAAa,QAAQ,KAAK,QAAQ,QAAQ;AAC9C,MAAI,IAAI,iBAAiB,KACvB,QAAO;AAGT,SAAO,IAAI,gBAAgB;GAC3B;AAEF,KAAI,WAAW,UAAU,cACvB,QAAO,EAAE;AAGX,QAAO,CAAC;EACN,UAAU;EACV,QAAQ,eAAe,WAAW,OAAO,wBAAwB,cAAc,0BAA0B,kBAAkB;EAC5H,CAAC;;;;;;;;;;;;;;;;ACCJ,SAAgB,kBACd,SACA,UAAoC,EAAE,EAClB;CACpB,MAAM,sBAAsB,8BAA8B,SAAS,QAAQ,UAAU;CACrF,MAAM,0BAA0B,4BAA4B,SAAS,QAAQ,cAAc;CAE3F,MAAM,aAAa,CACjB,GAAG,qBACH,GAAG,wBACJ;AAED,QAAO;EACL,MAAM,WAAW,WAAW;EAC5B;EACD"}
+{"version":3,"file":"index.mjs","names":[],"sources":["../../../../src/core/processors/results/policies/hybrid-threshold.ts","../../../../src/core/processors/results/policies/max-failed-runs.ts","../../../../src/core/processors/results/index.ts"],"sourcesContent":["import type { AggregatedRunResults } from '../../../runner/aggregate'\n\n/**\n * Violation emitted when result policies fail.\n */\nexport interface ResultPolicyViolation {\n  /**\n   * Stable policy id.\n   */\n  policyId: string\n  /**\n   * Human-readable violation reason.\n   */\n  reason: string\n}\n\n/**\n * Configures hybrid-threshold policy behavior.\n */\nexport interface HybridThresholdPolicyOptions {\n  /**\n   * Minimum required overall hybrid score.\n   *\n   * @default 0.7\n   */\n  minOverallHybridScore?: number\n  /**\n   * Minimum required inferenceExecutor hybrid score.\n   *\n   * @default 0.6\n   */\n  minProviderHybridScore?: number\n}\n\n/**\n * Evaluates threshold policy against aggregated results.\n */\nexport function evaluateHybridThresholdPolicy(\n  results: AggregatedRunResults,\n  options: HybridThresholdPolicyOptions = {},\n): ResultPolicyViolation[] {\n  const minOverallHybridScore = options.minOverallHybridScore ?? 0.7\n  const minProviderHybridScore = options.minProviderHybridScore ?? 0.6\n\n  const violations: ResultPolicyViolation[] = []\n\n  const overallHybridAverage = results.overall.hybridAverage\n  if (overallHybridAverage == null || overallHybridAverage < minOverallHybridScore) {\n    violations.push({\n      policyId: 'threshold:overall-hybrid',\n      reason: `Overall hybrid average ${overallHybridAverage ?? 'null'} is below ${minOverallHybridScore}.`,\n    })\n  }\n\n  for (const inferenceExecutor of results.inferenceExecutors) {\n    if (inferenceExecutor.hybridAverage == null || inferenceExecutor.hybridAverage < minProviderHybridScore) {\n      violations.push({\n        policyId: 'threshold:inferenceExecutor-hybrid',\n        reason: `Provider ${inferenceExecutor.inferenceExecutorId} hybrid average ${inferenceExecutor.hybridAverage ?? 'null'} is below ${minProviderHybridScore}.`,\n      })\n    }\n  }\n\n  return violations\n}\n","import type { AggregatedRunResults } from '../../../runner/aggregate'\nimport type { ResultPolicyViolation } from './hybrid-threshold'\n\n/**\n * Configures hard-limit policy for failed runs.\n */\nexport interface MaxFailedRunsPolicyOptions {\n  /**\n   * Maximum allowed failed run count.\n   *\n   * @default 0\n   */\n  maxFailedRuns?: number\n  /**\n   * Hybrid score threshold below which a run counts as failed.\n   *\n   * @default 0.6\n   */\n  minRunHybridScore?: number\n}\n\n/**\n * Evaluates hard-limit policy for failed runs.\n */\nexport function evaluateMaxFailedRunsPolicy(\n  results: AggregatedRunResults,\n  options: MaxFailedRunsPolicyOptions = {},\n): ResultPolicyViolation[] {\n  const maxFailedRuns = options.maxFailedRuns ?? 0\n  const minRunHybridScore = options.minRunHybridScore ?? 0.6\n\n  const failedRuns = results.runs.filter((run) => {\n    if (run.hybridAverage == null) {\n      return true\n    }\n\n    return run.hybridAverage < minRunHybridScore\n  })\n\n  if (failedRuns.length <= maxFailedRuns) {\n    return []\n  }\n\n  return [{\n    policyId: 'hard-limit:max-failed-runs',\n    reason: `Failed runs ${failedRuns.length} exceed maxFailedRuns ${maxFailedRuns} with minRunHybridScore ${minRunHybridScore}.`,\n  }]\n}\n","import type { AggregatedRunResults } from '../../runner/aggregate'\nimport type { HybridThresholdPolicyOptions, ResultPolicyViolation } from './policies/hybrid-threshold'\nimport type { MaxFailedRunsPolicyOptions } from './policies/max-failed-runs'\n\nimport { evaluateHybridThresholdPolicy } from './policies/hybrid-threshold'\nimport { evaluateMaxFailedRunsPolicy } from './policies/max-failed-runs'\n\n/**\n * Configures result-processing policies for eval gating.\n */\nexport interface ProcessRunResultsOptions {\n  /**\n   * Threshold policy options.\n   */\n  threshold?: HybridThresholdPolicyOptions\n  /**\n   * Hard-limit failed-run policy options.\n   */\n  maxFailedRuns?: MaxFailedRunsPolicyOptions\n}\n\n/**\n * Final gate decision returned by result processors.\n */\nexport interface ResultGateDecision {\n  /**\n   * Whether the result batch passes all policies.\n   */\n  pass: boolean\n  /**\n   * Collected policy violations.\n   */\n  violations: ResultPolicyViolation[]\n}\n\n/**\n * Processes aggregated run results through built-in gating policies.\n *\n * Call stack:\n *\n * {@link runScheduledTasks}\n *   -> {@link aggregateRunResults}\n *     -> {@link processRunResults}\n *       -> {@link evaluateHybridThresholdPolicy}\n *       -> {@link evaluateMaxFailedRunsPolicy}\n *         -> {@link ResultGateDecision}\n */\nexport function processRunResults(\n  results: AggregatedRunResults,\n  options: ProcessRunResultsOptions = {},\n): ResultGateDecision {\n  const thresholdViolations = evaluateHybridThresholdPolicy(results, options.threshold)\n  const maxFailedRunsViolations = evaluateMaxFailedRunsPolicy(results, options.maxFailedRuns)\n\n  const violations = [\n    ...thresholdViolations,\n    ...maxFailedRunsViolations,\n  ]\n\n  return {\n    pass: violations.length === 0,\n    violations,\n  }\n}\n\nexport type {\n  HybridThresholdPolicyOptions,\n  MaxFailedRunsPolicyOptions,\n  ResultPolicyViolation,\n}\n"],"mappings":";;;;AAqCA,SAAgB,8BACd,SACA,UAAwC,CAAC,GAChB;CACzB,MAAM,wBAAwB,QAAQ,yBAAyB;CAC/D,MAAM,yBAAyB,QAAQ,0BAA0B;CAEjE,MAAM,aAAsC,CAAC;CAE7C,MAAM,uBAAuB,QAAQ,QAAQ;CAC7C,IAAI,wBAAwB,QAAQ,uBAAuB,uBACzD,WAAW,KAAK;EACd,UAAU;EACV,QAAQ,0BAA0B,wBAAwB,OAAO,YAAY,sBAAsB;CACrG,CAAC;CAGH,KAAK,MAAM,qBAAqB,QAAQ,oBACtC,IAAI,kBAAkB,iBAAiB,QAAQ,kBAAkB,gBAAgB,wBAC/E,WAAW,KAAK;EACd,UAAU;EACV,QAAQ,YAAY,kBAAkB,oBAAoB,kBAAkB,kBAAkB,iBAAiB,OAAO,YAAY,uBAAuB;CAC3J,CAAC;CAIL,OAAO;AACT;;;;;;ACxCA,SAAgB,4BACd,SACA,UAAsC,CAAC,GACd;CACzB,MAAM,gBAAgB,QAAQ,iBAAiB;CAC/C,MAAM,oBAAoB,QAAQ,qBAAqB;CAEvD,MAAM,aAAa,QAAQ,KAAK,QAAQ,QAAQ;EAC9C,IAAI,IAAI,iBAAiB,MACvB,OAAO;EAGT,OAAO,IAAI,gBAAgB;CAC7B,CAAC;CAED,IAAI,WAAW,UAAU,eACvB,OAAO,CAAC;CAGV,OAAO,CAAC;EACN,UAAU;EACV,QAAQ,eAAe,WAAW,OAAO,wBAAwB,cAAc,0BAA0B,kBAAkB;CAC7H,CAAC;AACH;;;;;;;;;;;;;;;ACAA,SAAgB,kBACd,SACA,UAAoC,CAAC,GACjB;CACpB,MAAM,sBAAsB,8BAA8B,SAAS,QAAQ,SAAS;CACpF,MAAM,0BAA0B,4BAA4B,SAAS,QAAQ,aAAa;CAE1F,MAAM,aAAa,CACjB,GAAG,qBACH,GAAG,uBACL;CAEA,OAAO;EACL,MAAM,WAAW,WAAW;EAC5B;CACF;AACF"}

package/dist/core/runner/index.d.mts

@@ -1,3 +1,3 @@
-import { $ as InferenceExecutor, A as RunScheduledTasksOptions, B as asProjectRelativePath, F as CreateTaskExecutionContextOptions, G as AggregatedProviderSummary, H as CreateVievalRunnerRuntimeContextOptions, I as TaskExecutionContext, J as RunResult, K as AggregatedRunResults, L as createTaskExecutionContext, M as RunnerTaskState, N as ScheduledTaskExecutor, P as runScheduledTasks, Q as CreateRunnerScheduleOptions, U as RunnerRuntimeContext, V as collectEvalEntries, W as createRunnerRuntimeContext, X as RunScoreKind, Y as RunScore, Z as aggregateRunResults, at as ScheduledTaskMatrixMeta, ct as createFilesystemTaskCacheRuntime, dt as CacheFileOptions, et as RunnerMatrixDefinition, ft as CacheNamespace, it as ScheduledTaskMatrix, j as RunnerExecutionError, lt as normalizeCacheFilePathSegments, nt as RunnerMatrixSelection, ot as createRunnerSchedule, pt as TaskCacheRuntime, q as AggregatedRunSummary, rt as ScheduledTask, st as CreateFilesystemTaskCacheRuntimeOptions, tt as RunnerMatrixInput, ut as CacheFileHandle } from "../../index-BkjyCInx.mjs";
-import { a as SchedulerMiddleware, c as SchedulerScopeContext, i as SchedulerConcurrencyConfig, n as getActiveScopes, o as SchedulerRuntime, r as CreateSchedulerRuntimeOptions, s as SchedulerScope, t as createSchedulerRuntime } from "../../index-fakXoZEe.mjs";
+import { $ as InferenceExecutor, A as RunScheduledTasksOptions, B as asProjectRelativePath, F as CreateTaskExecutionContextOptions, G as AggregatedProviderSummary, H as CreateVievalRunnerRuntimeContextOptions, I as TaskExecutionContext, J as RunResult, K as AggregatedRunResults, L as createTaskExecutionContext, M as RunnerTaskState, N as ScheduledTaskExecutor, P as runScheduledTasks, Q as CreateRunnerScheduleOptions, U as RunnerRuntimeContext, V as collectEvalEntries, W as createRunnerRuntimeContext, X as RunScoreKind, Y as RunScore, Z as aggregateRunResults, at as ScheduledTaskMatrixMeta, ct as createFilesystemTaskCacheRuntime, dt as CacheFileOptions, et as RunnerMatrixDefinition, ft as CacheNamespace, it as ScheduledTaskMatrix, j as RunnerExecutionError, lt as normalizeCacheFilePathSegments, nt as RunnerMatrixSelection, ot as createRunnerSchedule, pt as TaskCacheRuntime, q as AggregatedRunSummary, rt as ScheduledTask, st as CreateFilesystemTaskCacheRuntimeOptions, tt as RunnerMatrixInput, ut as CacheFileHandle } from "../../index-CwKBlCG9.mjs";
+import { a as SchedulerMiddleware, c as SchedulerScopeContext, i as SchedulerConcurrencyConfig, n as getActiveScopes, o as SchedulerRuntime, r as CreateSchedulerRuntimeOptions, s as SchedulerScope, t as createSchedulerRuntime } from "../../index-Be5I1ZJL.mjs";
 export { AggregatedProviderSummary, AggregatedRunResults, AggregatedRunSummary, CacheFileHandle, CacheFileOptions, CacheNamespace, CreateFilesystemTaskCacheRuntimeOptions, CreateRunnerScheduleOptions, CreateSchedulerRuntimeOptions, CreateTaskExecutionContextOptions, CreateVievalRunnerRuntimeContextOptions, InferenceExecutor, RunResult, RunScheduledTasksOptions, RunScore, RunScoreKind, RunnerExecutionError, RunnerMatrixDefinition, RunnerMatrixInput, RunnerMatrixSelection, RunnerRuntimeContext, RunnerTaskState, ScheduledTask, ScheduledTaskExecutor, ScheduledTaskMatrix, ScheduledTaskMatrixMeta, SchedulerConcurrencyConfig, SchedulerMiddleware, SchedulerRuntime, SchedulerScope, SchedulerScopeContext, TaskCacheRuntime, TaskExecutionContext, aggregateRunResults, asProjectRelativePath, collectEvalEntries, createFilesystemTaskCacheRuntime, createRunnerRuntimeContext, createRunnerSchedule, createSchedulerRuntime, createTaskExecutionContext, getActiveScopes, normalizeCacheFilePathSegments, runScheduledTasks };

package/dist/core/runner/index.mjs

@@ -2,11 +2,11 @@ import { createSchedulerRuntime, getActiveScopes } from "../scheduler/index.mjs"
 import { createRequire } from "node:module";
 import process from "node:process";
 import { errorMessageFrom } from "@moeru/std";
-import { basename, dirname, join, relative } from "node:path";
 import { access, mkdir, readFile, rename, writeFile } from "node:fs/promises";
+import { basename, dirname, join, relative } from "node:path";
 import { fileURLToPath } from "node:url";
-import { Buffer } from "node:buffer";
 import { createReadStream, createWriteStream } from "node:fs";
+import { Buffer } from "node:buffer";
 import { limitConcurrency } from "@vitest/runner/utils";
 //#region src/core/cache/filesystem.ts
 function sanitizePathSegment(value) {
@@ -425,7 +425,7 @@ async function createRunnerRuntimeContext(options = {}) {
 }
 //#endregion
 //#region src/core/runner/schedule.ts
-const matrixLayerKeys = new Set([
+const matrixLayerKeys = /* @__PURE__ */ new Set([
 	"disable",
 	"extend",
 	"override"
@@ -475,7 +475,7 @@ function applyAxisValues(axes, definition, mode) {
 		const nextValues = dedupeAxisValues(values);
 		if (mode === "extend") {
 			const existingValues = axes.get(axis) ?? [];
-			axes.set(axis, Array.from(new Set([...existingValues, ...nextValues])));
+			axes.set(axis, Array.from(/* @__PURE__ */ new Set([...existingValues, ...nextValues])));
 			continue;
 		}
 		axes.set(axis, nextValues);

package/dist/core/runner/index.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"index.mjs","names":[],"sources":["../../../src/core/cache/filesystem.ts","../../../src/core/runner/aggregate.ts","../../../src/core/runner/collect.ts","../../../src/core/runner/run.ts","../../../src/core/runner/runtime-context.ts","../../../src/core/runner/schedule.ts","../../../src/core/runner/task-context.ts"],"sourcesContent":["import type { CacheFileHandle, CacheFileOptions, CacheNamespace, TaskCacheRuntime } from './types'\n\nimport process from 'node:process'\n\nimport { Buffer } from 'node:buffer'\nimport { createReadStream, createWriteStream } from 'node:fs'\nimport { access, mkdir, readFile, rename, writeFile } from 'node:fs/promises'\nimport { dirname, join } from 'node:path'\n\n/**\n * Options for creating the filesystem-backed task cache runtime.\n */\nexport interface CreateFilesystemTaskCacheRuntimeOptions {\n  /**\n   * Absolute cache root directory.\n   */\n  cacheRootDirectory: string\n  /**\n   * Project identifier under one workspace cache scope.\n   */\n  projectName: string\n  /**\n   * Workspace identifier used to share cache roots across projects.\n   */\n  workspaceId: string\n}\n\nfunction sanitizePathSegment(value: string): string {\n  const normalized = value.trim()\n  if (normalized.length === 0) {\n    return 'default'\n  }\n\n  return normalized.replace(/[^\\w.-]+/g, '-')\n}\n\nfunction normalizeExtension(extension: string | undefined, mediaType: string | undefined): string | undefined {\n  if (extension != null && extension.length > 0) {\n    return extension.startsWith('.') ? extension.slice(1) : extension\n  }\n\n  if (mediaType == null || mediaType.length === 0) {\n    return undefined\n  }\n\n  if (mediaType === 'application/json') {\n    return 'json'\n  }\n\n  if (mediaType === 'text/plain') {\n    return 'txt'\n  }\n\n  if (mediaType === 'audio/wav') {\n    return 'wav'\n  }\n\n  return undefined\n}\n\n/**\n * Normalizes cache file options into deterministic relative path segments.\n *\n * Before:\n * - `{ key: ['cases', 'dataset hash', 'v1'], ext: 'json' }`\n *\n * After:\n * - `['cases', 'dataset-hash', 'v1.json']`\n */\nexport function normalizeCacheFilePathSegments(options: CacheFileOptions): string[] {\n  const sanitizedKey = options.key.map(segment => sanitizePathSegment(segment))\n  const extension = normalizeExtension(options.ext, options.mediaType)\n\n  if (sanitizedKey.length === 0) {\n    return extension == null ? ['artifact'] : [`artifact.${extension}`]\n  }\n\n  if (extension == null) {\n    return sanitizedKey\n  }\n\n  const withoutTail = sanitizedKey.slice(0, Math.max(0, sanitizedKey.length - 1))\n  const tail = sanitizedKey[sanitizedKey.length - 1] ?? 'artifact'\n  return [...withoutTail, `${tail}.${extension}`]\n}\n\nasync function writeAtomically(path: string, content: Buffer | string): Promise<void> {\n  const directory = dirname(path)\n  const temporaryPath = `${path}.tmp-${process.pid}-${Date.now()}-${Math.random().toString(36).slice(2, 10)}`\n  await mkdir(directory, { recursive: true })\n  await writeFile(temporaryPath, content)\n  await rename(temporaryPath, path)\n}\n\nfunction createCacheFileHandle(path: string): CacheFileHandle {\n  return {\n    path,\n    async exists() {\n      try {\n        await access(path)\n        return true\n      }\n      catch {\n        return false\n      }\n    },\n    openReadStream() {\n      return createReadStream(path)\n    },\n    async openWriteStream() {\n      await mkdir(dirname(path), { recursive: true })\n      return createWriteStream(path)\n    },\n    async readBuffer() {\n      return await readFile(path)\n    },\n    async writeBuffer(value) {\n      await writeAtomically(path, value)\n    },\n    async readText(encoding = 'utf-8') {\n      return await readFile(path, encoding)\n    },\n    async writeText(value, encoding = 'utf-8') {\n      await writeAtomically(path, Buffer.from(value, encoding))\n    },\n    async readJson<T>() {\n      return JSON.parse(await readFile(path, 'utf-8')) as T\n    },\n    async writeJson(value) {\n      await writeAtomically(path, `${JSON.stringify(value, null, 2)}\\n`)\n    },\n    async loadAsCasesInput<T>() {\n      return await this.readJson<T[]>()\n    },\n    async loadAsExpectFixture<T>() {\n      return await this.readJson<T>()\n    },\n  }\n}\n\nfunction createCacheNamespace(baseDirectory: string, namespace: string): CacheNamespace {\n  return {\n    file(options) {\n      const relativePathSegments = normalizeCacheFilePathSegments(options)\n      return createCacheFileHandle(join(baseDirectory, sanitizePathSegment(namespace), ...relativePathSegments))\n    },\n  }\n}\n\n/**\n * Creates a deterministic filesystem-backed task cache runtime.\n *\n * Use when:\n * - eval tasks need reproducible cache paths for expensive pre-processing outputs\n * - benchmark adapters need one artifact-oriented API for text/json/binary reads and writes\n *\n * Expects:\n * - `cacheRootDirectory` to be writable by the running process\n * - `workspaceId` + `projectName` to stay stable for reproducible paths\n *\n * Returns:\n * - task cache runtime that resolves namespaced file handles under:\n *   `<cacheRootDirectory>/<workspaceId>/<projectName>/<namespace>/...`\n */\nexport function createFilesystemTaskCacheRuntime(\n  options: CreateFilesystemTaskCacheRuntimeOptions,\n): TaskCacheRuntime {\n  const workspaceDirectory = sanitizePathSegment(options.workspaceId)\n  const projectDirectory = sanitizePathSegment(options.projectName)\n  const baseDirectory = join(options.cacheRootDirectory, workspaceDirectory, projectDirectory)\n\n  return {\n    namespace(name) {\n      return createCacheNamespace(baseDirectory, name)\n    },\n  }\n}\n","import type { ScheduledTaskMatrix } from './schedule'\n\n/**\n * Identifies the scoring family for a single eval score.\n */\nexport type RunScoreKind = 'exact' | 'judge'\n\n/**\n * Represents one normalized score emitted by a completed eval run.\n */\nexport interface RunScore {\n  /**\n   * Score family used for aggregation.\n   */\n  kind: RunScoreKind\n  /**\n   * Normalized score in the `0..1` range.\n   */\n  score: number\n}\n\n/**\n * Captures the output of one scheduled runner task.\n */\nexport interface RunResult {\n  /**\n   * Stable run id, usually copied from the scheduled task id.\n   */\n  id: string\n  /**\n   * Collected eval entry id.\n   */\n  entryId: string\n  /**\n   * Stable inferenceExecutor id.\n   */\n  inferenceExecutorId: string\n  /**\n   * Concrete matrix selection used by the run.\n   */\n  matrix: ScheduledTaskMatrix\n  /**\n   * Raw scores emitted by the eval.\n   */\n  scores: readonly RunScore[]\n}\n\n/**\n * Stores the per-run score averages after normalization.\n */\nexport interface AggregatedRunSummary {\n  /**\n   * Stable run id.\n   */\n  id: string\n  /**\n   * Collected eval entry id.\n   */\n  entryId: string\n  /**\n   * Stable inferenceExecutor id.\n   */\n  inferenceExecutorId: string\n  /**\n   * Concrete matrix selection used by the run.\n   */\n  matrix: ScheduledTaskMatrix\n  /**\n   * Mean of exact-match scores or `null` when absent.\n   */\n  exactAverage: number | null\n  /**\n   * Mean of judge-based scores or `null` when absent.\n   */\n  judgeAverage: number | null\n  /**\n   * Hybrid average. Uses both families when present, otherwise falls back to the\n   * single available family.\n   */\n  hybridAverage: number | null\n}\n\n/**\n * Stores inferenceExecutor-level score aggregates across multiple runs.\n */\nexport interface AggregatedProviderSummary {\n  /**\n   * Stable inferenceExecutor id.\n   */\n  inferenceExecutorId: string\n  /**\n   * Number of runs included in this inferenceExecutor bucket.\n   */\n  runCount: number\n  /**\n   * Mean of all exact-match scores or `null` when absent.\n   */\n  exactAverage: number | null\n  /**\n   * Mean of all judge-based scores or `null` when absent.\n   */\n  judgeAverage: number | null\n  /**\n   * Hybrid average derived from the inferenceExecutor exact and judge means.\n   */\n  hybridAverage: number | null\n}\n\n/**\n * Stores the final aggregation output for a batch of runner results.\n */\nexport interface AggregatedRunResults {\n  /**\n   * Per-run normalized score summaries.\n   */\n  runs: AggregatedRunSummary[]\n  /**\n   * Provider-level summaries sorted by inferenceExecutor id.\n   */\n  inferenceExecutors: AggregatedProviderSummary[]\n  /**\n   * Overall summary across every run.\n   */\n  overall: {\n    exactAverage: number | null\n    judgeAverage: number | null\n    hybridAverage: number | null\n    runCount: number\n  }\n}\n\ninterface ScoreBuckets {\n  exact: number[]\n  judge: number[]\n}\n\nfunction cloneScheduledTaskMatrix(matrix: ScheduledTaskMatrix): ScheduledTaskMatrix {\n  return {\n    eval: {\n      ...matrix.eval,\n    },\n    meta: {\n      ...matrix.meta,\n    },\n    run: {\n      ...matrix.run,\n    },\n  }\n}\n\nfunction assertKnownScoreKind(kind: string): RunScoreKind {\n  if (kind === 'exact' || kind === 'judge') {\n    return kind\n  }\n\n  throw new TypeError(`Unknown eval score kind \"${kind}\".`)\n}\n\nfunction average(scores: readonly number[]): number | null {\n  if (scores.length === 0) {\n    return null\n  }\n\n  const total = scores.reduce((sum, score) => sum + score, 0)\n  return total / scores.length\n}\n\nfunction createHybridAverage(exactAverage: number | null, judgeAverage: number | null): number | null {\n  if (exactAverage != null && judgeAverage != null) {\n    return (exactAverage + judgeAverage) / 2\n  }\n\n  if (exactAverage != null) {\n    return exactAverage\n  }\n\n  if (judgeAverage != null) {\n    return judgeAverage\n  }\n\n  return null\n}\n\nfunction collectScoreBuckets(scores: readonly RunScore[]): ScoreBuckets {\n  const buckets: ScoreBuckets = {\n    exact: [],\n    judge: [],\n  }\n\n  for (const score of scores) {\n    const kind = assertKnownScoreKind(score.kind)\n\n    if (kind === 'exact') {\n      buckets.exact.push(score.score)\n      continue\n    }\n\n    buckets.judge.push(score.score)\n  }\n\n  return buckets\n}\n\nfunction createRunSummary(result: RunResult): AggregatedRunSummary {\n  const buckets = collectScoreBuckets(result.scores)\n  const exactAverage = average(buckets.exact)\n  const judgeAverage = average(buckets.judge)\n\n  return {\n    entryId: result.entryId,\n    exactAverage,\n    hybridAverage: createHybridAverage(exactAverage, judgeAverage),\n    id: result.id,\n    judgeAverage,\n    matrix: cloneScheduledTaskMatrix(result.matrix),\n    inferenceExecutorId: result.inferenceExecutorId,\n  }\n}\n\nfunction createProviderSummary(inferenceExecutorId: string, results: readonly RunResult[]): AggregatedProviderSummary {\n  const exactScores: number[] = []\n  const judgeScores: number[] = []\n\n  for (const result of results) {\n    const buckets = collectScoreBuckets(result.scores)\n    exactScores.push(...buckets.exact)\n    judgeScores.push(...buckets.judge)\n  }\n\n  const exactAverage = average(exactScores)\n  const judgeAverage = average(judgeScores)\n\n  return {\n    exactAverage,\n    hybridAverage: createHybridAverage(exactAverage, judgeAverage),\n    judgeAverage,\n    inferenceExecutorId,\n    runCount: results.length,\n  }\n}\n\n/**\n * Aggregates exact-match and judge-based scores into hybrid runner summaries.\n *\n * Call stack:\n *\n * {@link runScheduledTasks}\n *   -> {@link aggregateRunResults}\n *     -> {@link createRunSummary}\n *     -> {@link createProviderSummary}\n *       -> `report output`\n *\n * Use when:\n * - a runner batch mixes deterministic exact checks with judge-based grading\n * - inferenceExecutor comparison should preserve both score families and one hybrid view\n *\n * Expects:\n * - each score to be normalized to the `0..1` range before aggregation\n * - `scores.kind` to use only `'exact'` or `'judge'`\n */\nexport function aggregateRunResults(results: readonly RunResult[]): AggregatedRunResults {\n  const runs = results.map(createRunSummary)\n\n  const inferenceExecutorIds = Array.from(new Set(results.map(result => result.inferenceExecutorId)))\n  const inferenceExecutors = inferenceExecutorIds\n    .map((inferenceExecutorId) => {\n      const providerResults = results.filter(result => result.inferenceExecutorId === inferenceExecutorId)\n      return createProviderSummary(inferenceExecutorId, providerResults)\n    })\n    .sort((left, right) => left.inferenceExecutorId.localeCompare(right.inferenceExecutorId))\n\n  const overall = createProviderSummary(\n    'overall',\n    results,\n  )\n\n  return {\n    overall: {\n      exactAverage: overall.exactAverage,\n      hybridAverage: overall.hybridAverage,\n      judgeAverage: overall.judgeAverage,\n      runCount: overall.runCount,\n    },\n    inferenceExecutors,\n    runs,\n  }\n}\n","import type { CollectedEvalEntry, EvalModule, EvalModuleMap } from '../../config'\nimport type { RunnerRuntimeContext } from './runtime-context'\n\nimport { basename, dirname, relative } from 'node:path'\nimport { fileURLToPath } from 'node:url'\n\nconst evalFileSuffix = '.eval.ts'\nconst absolutePathPattern = /^(?:[A-Z]:\\/|\\/|\\\\\\\\)/i\n\nfunction normalizePath(value: string): string {\n  return value.replaceAll('\\\\', '/')\n}\n\n/**\n * Converts a file path into a project-relative path when possible.\n *\n * Before: `/repo/plugins/airi-plugin-game-chess/src/agent/evals/chess-commentary.eval.ts`\n * After: `plugins/airi-plugin-game-chess/src/agent/evals/chess-commentary.eval.ts`\n *\n * Before: `D:/repo/plugins/airi-plugin-game-chess/src/agent/evals/chess-commentary.eval.ts`\n * After: `D:/repo/plugins/airi-plugin-game-chess/src/agent/evals/chess-commentary.eval.ts`\n */\nexport function asProjectRelativePath(filePath: string, context: RunnerRuntimeContext): string {\n  const normalizedFilePath = normalizePath(filePath)\n  const normalizedProjectRootDirectory = normalizePath(context.projectRootDirectory)\n  const filePathWindowsDrive = normalizedFilePath.match(/^[A-Z]:\\//i)?.[0]\n  const projectRootWindowsDrive = normalizedProjectRootDirectory.match(/^[A-Z]:\\//i)?.[0]\n\n  if (filePathWindowsDrive != null && projectRootWindowsDrive == null) {\n    return normalizedFilePath\n  }\n\n  if (\n    filePathWindowsDrive != null\n    && projectRootWindowsDrive != null\n    && filePathWindowsDrive.toLowerCase() !== projectRootWindowsDrive.toLowerCase()\n  ) {\n    return normalizedFilePath\n  }\n\n  const projectRootDirectory = context.projectRootDirectory\n  const relativeFilePath = normalizePath(relative(projectRootDirectory, filePath))\n\n  if (!absolutePathPattern.test(relativeFilePath)) {\n    if (relativeFilePath === '..') {\n      return normalizePath(filePath)\n    }\n\n    if (!relativeFilePath.startsWith('../')) {\n      return relativeFilePath\n    }\n  }\n\n  return normalizePath(filePath)\n}\n\nfunction resolveModuleFilePath(moduleHref: string): string | null {\n  if (!moduleHref.startsWith('file:')) {\n    return null\n  }\n\n  try {\n    return fileURLToPath(moduleHref)\n  }\n  catch {\n    return null\n  }\n}\n\nfunction createCollectedEvalEntry(\n  moduleHref: string,\n  moduleDefinition: EvalModule,\n  context: RunnerRuntimeContext,\n): CollectedEvalEntry | null {\n  const filePath = resolveModuleFilePath(moduleHref)\n\n  if (!filePath) {\n    return null\n  }\n\n  const relativeFilePath = asProjectRelativePath(filePath, context)\n\n  if (!relativeFilePath.endsWith(evalFileSuffix)) {\n    return null\n  }\n\n  const entryName = basename(relativeFilePath, evalFileSuffix)\n\n  if (entryName.length === 0) {\n    return null\n  }\n\n  const relativeDirectory = dirname(relativeFilePath)\n  const directory = relativeDirectory === '.' ? '' : relativeDirectory\n\n  return {\n    ...moduleDefinition.default,\n    directory,\n    filePath,\n    id: directory.length === 0 ? entryName : `${directory}/${entryName}`,\n    name: entryName,\n  }\n}\n\n/**\n * Collects loaded vieval modules into sorted runner entries with stable ids.\n *\n * Call stack:\n *\n * `import.meta.glob(...)`\n *   -> {@link collectEvalEntries}\n *     -> {@link createCollectedEvalEntry}\n *       -> {@link CollectedEvalEntry}[]\n *\n * Use when:\n * - the runner has already loaded candidate eval modules\n * - downstream scheduling needs stable entry ids and directory metadata\n */\nexport function collectEvalEntries(\n  modules: EvalModuleMap,\n  context: RunnerRuntimeContext,\n): CollectedEvalEntry[] {\n  return Object.entries(modules)\n    .flatMap(([moduleHref, moduleDefinition]) => {\n      const entry = createCollectedEvalEntry(moduleHref, moduleDefinition, context)\n\n      if (!entry) {\n        return []\n      }\n\n      return [entry]\n    })\n    .sort((left, right) => left.id.localeCompare(right.id))\n}\n","import type { TaskCacheRuntime } from '../cache'\nimport type { AggregatedRunResults, RunResult } from './aggregate'\nimport type { ScheduledTask } from './schedule'\nimport type { TaskExecutionContext } from './task-context'\n\nimport { errorMessageFrom } from '@moeru/std'\nimport { limitConcurrency } from '@vitest/runner/utils'\n\nimport { aggregateRunResults } from './aggregate'\n\n/**\n * Executes one scheduled runner task and returns a normalized run result.\n *\n * Use when:\n * - a scheduler already selected the task and execution context\n * - the caller wants a typed executor contract for runner workers\n *\n * Expects:\n * - the task context to be ready for model resolution and task-scoped work\n *\n * Returns:\n * - a normalized run result with score entries ready for aggregation\n */\nexport type ScheduledTaskExecutor = (\n  task: ScheduledTask,\n  context: TaskExecutionContext,\n) => Promise<RunResult>\n\n/**\n * Terminal task state reported by runner lifecycle hooks.\n *\n * Use when:\n * - reporting the outcome of one scheduled task to lifecycle observers\n *\n * Expects:\n * - hooks treat the value as final for the completed task\n */\nexport type RunnerTaskState = 'passed' | 'failed'\n\n/**\n * Optional runner execution hooks used while processing scheduled tasks.\n *\n * Use when:\n * - callers want lifecycle visibility around sequential task execution\n * - task execution should remain deterministic while still observable\n *\n * Expects:\n * - hook functions are synchronous lifecycle observers\n */\nexport interface RunScheduledTasksOptions {\n  /**\n   * Creates per-task execution context.\n   *\n   * Use when:\n   * - executor code needs per-task models, cache, or other task-scoped data\n   */\n  createExecutionContext?: (task: ScheduledTask) => TaskExecutionContext\n  /**\n   * Runs before the executor starts handling a task.\n   *\n   * Use when:\n   * - callers want to observe task activation before execution begins\n   *\n   * Expects:\n   * - thrown errors abort the task before executor work starts\n   */\n  onTaskStart?: (task: ScheduledTask) => void\n  /**\n   * Runs after the executor settles for a task.\n   *\n   * Use when:\n   * - callers want to observe successful and failed task completion\n   *\n   * Expects:\n   * - thrown errors abort successful runs\n   * - failed-task observers do not override the executor error for the task\n   */\n  onTaskEnd?: (task: ScheduledTask, state: RunnerTaskState) => void\n  /**\n   * Maximum number of tasks to execute concurrently.\n   *\n   * @default 1\n   */\n  maxConcurrency?: number\n}\n\nfunction createDefaultExecutionContext(): TaskExecutionContext {\n  const cache: TaskCacheRuntime = {\n    namespace(name) {\n      return {\n        file(options) {\n          const key = options.key.join('/')\n          throw new Error(`Task cache runtime is not configured. Requested namespace \"${name}\" and key \"${key}\".`)\n        },\n      }\n    },\n  }\n\n  return {\n    cache,\n    models: [],\n  }\n}\n\n/**\n * Error thrown when a scheduled run fails before producing a normalized result.\n */\nexport class RunnerExecutionError extends Error {\n  /**\n   * Stable task id that failed.\n   */\n  taskId: string\n\n  constructor(taskId: string, cause: unknown) {\n    const message = errorMessageFrom(cause) ?? 'Unknown runner execution failure.'\n    super(`Runner task \"${taskId}\" failed: ${message}`)\n    this.name = 'RunnerExecutionError'\n    this.taskId = taskId\n    this.cause = cause\n  }\n}\n\nfunction createRunnerExecutionError(taskId: string, cause: unknown): RunnerExecutionError {\n  if (cause instanceof RunnerExecutionError && cause.taskId === taskId) {\n    return cause\n  }\n\n  return new RunnerExecutionError(taskId, cause)\n}\n\n/**\n * Executes runner tasks sequentially and aggregates the normalized results.\n *\n * Call stack:\n *\n * {@link createRunnerSchedule}\n *   -> {@link runScheduledTasks}\n *     -> `executor(task)`\n *       -> {@link aggregateRunResults}\n *\n * Use when:\n * - the caller already expanded the runner matrix\n * - task execution should stay deterministic and easy to debug\n *\n * Expects:\n * - `executor` to return normalized `0..1` scores\n * - callers to handle concurrency outside this helper when needed\n * - `onTaskStart` / `onTaskEnd` hooks to be synchronous lifecycle observers\n *\n * Throws:\n * - `RunnerExecutionError` when task setup, hooks, or the executor throws\n */\nexport async function runScheduledTasks(\n  tasks: readonly ScheduledTask[],\n  executor: ScheduledTaskExecutor,\n  options: RunScheduledTasksOptions = {},\n): Promise<AggregatedRunResults> {\n  if (tasks.length === 0) {\n    return aggregateRunResults([])\n  }\n\n  async function executeScheduledTask(task: ScheduledTask): Promise<RunResult> {\n    let executionContext: TaskExecutionContext\n\n    try {\n      executionContext = options.createExecutionContext?.(task) ?? createDefaultExecutionContext()\n    }\n    catch (error) {\n      throw createRunnerExecutionError(task.id, error)\n    }\n\n    try {\n      options.onTaskStart?.(task)\n    }\n    catch (error) {\n      throw createRunnerExecutionError(task.id, error)\n    }\n\n    let runResult: RunResult\n    try {\n      runResult = await executor(task, executionContext)\n    }\n    catch (error) {\n      try {\n        options.onTaskEnd?.(task, 'failed')\n      }\n      catch {\n        // Failed-task observers must not mask the task execution failure.\n      }\n      throw createRunnerExecutionError(task.id, error)\n    }\n\n    try {\n      options.onTaskEnd?.(task, 'passed')\n    }\n    catch (error) {\n      throw createRunnerExecutionError(task.id, error)\n    }\n\n    return runResult\n  }\n\n  const maxConcurrency = options.maxConcurrency ?? 1\n  if (maxConcurrency <= 1) {\n    const results: RunResult[] = []\n    for (const task of tasks) {\n      results.push(await executeScheduledTask(task))\n    }\n    return aggregateRunResults(results)\n  }\n\n  const runWithLimit = limitConcurrency(maxConcurrency)\n  const resultPairs = await Promise.all(tasks.map(async (task, index) => {\n    const result = await runWithLimit(async () => executeScheduledTask(task))\n    return { index, result }\n  }))\n\n  const sortedResults = resultPairs\n    .sort((left, right) => left.index - right.index)\n    .map(item => item.result)\n\n  return aggregateRunResults(sortedResults)\n}\n","import { createRequire } from 'node:module'\nimport { dirname } from 'node:path'\nimport { fileURLToPath } from 'node:url'\n\nconst require = createRequire(import.meta.url)\n\n/**\n * Shared runtime context used by the vieval runner.\n *\n * Use when:\n * - runner services need stable path resolution without module-level side effects\n * - call sites want deterministic control over workspace root detection\n */\nexport interface RunnerRuntimeContext {\n  /**\n   * Absolute project root directory used for path normalization.\n   */\n  projectRootDirectory: string\n}\n\n/**\n * Options used to construct the runner runtime context.\n */\nexport interface CreateVievalRunnerRuntimeContextOptions {\n  /**\n   * Directory used to search for the nearest pnpm workspace.\n   *\n   * @default directory of this module file\n   */\n  cwd?: string\n  /**\n   * Absolute fallback directory when a pnpm workspace root is not found.\n   *\n   * @default package root directory (`packages/vieval`)\n   */\n  fallbackProjectRootDirectory?: string\n}\n\n/**\n * Creates a side-effect-free runtime context for runner path normalization.\n *\n * Call stack:\n *\n * {@link createRunnerRuntimeContext}\n *   -> `findWorkspaceDir(cwd)`\n *     -> `resolve projectRootDirectory`\n *       -> `{ projectRootDirectory }`\n *\n * Use when:\n * - initializing runner infrastructure before collecting eval modules\n * - tests need deterministic root resolution behavior\n */\nexport async function createRunnerRuntimeContext(\n  options: CreateVievalRunnerRuntimeContextOptions = {},\n): Promise<RunnerRuntimeContext> {\n  const cwd = options.cwd ?? dirname(fileURLToPath(import.meta.url))\n  const fallbackProjectRootDirectory = options.fallbackProjectRootDirectory\n    ?? fileURLToPath(new URL('../../../', import.meta.url))\n\n  // NOTICE:\n  // We use dynamic `require` here because `@pnpm/find-workspace-dir` is CommonJS.\n  // Keeping this load inside the factory avoids module-level initialization side effects.\n  const { findWorkspaceDir } = require('@pnpm/find-workspace-dir') as {\n    findWorkspaceDir: (currentWorkingDirectory: string) => Promise<string | undefined>\n  }\n\n  // NOTICE:\n  // Workspace discovery is required to keep collected eval ids stable when this\n  // package is moved inside different monorepo layouts.\n  const workspaceDirectory = await findWorkspaceDir(cwd)\n\n  return {\n    projectRootDirectory: workspaceDirectory ?? fallbackProjectRootDirectory,\n  }\n}\n","import type { CollectedEvalEntry, MatrixDefinition, MatrixLayer, MatrixValue } from '../../config'\n\n/**\n * Describes the inferenceExecutor target for a scheduled eval run.\n */\nexport interface InferenceExecutor {\n  /**\n   * Stable inferenceExecutor identifier such as `openai:gpt-4.1-mini`.\n   */\n  id: string\n}\n\n/**\n * Stores the selected value for each matrix axis.\n */\nexport type RunnerMatrixSelection = Record<string, string>\n\n/**\n * Stores stable row ids for one resolved scheduled task matrix.\n */\nexport interface ScheduledTaskMatrixMeta {\n  /**\n   * Stable row id for the resolved run matrix selection.\n   */\n  runRowId: string\n  /**\n   * Stable row id for the resolved eval matrix selection.\n   */\n  evalRowId: string\n}\n\n/**\n * Stores the structured matrix payload for one scheduled task.\n */\nexport interface ScheduledTaskMatrix {\n  /**\n   * Runtime matrix selection visible to task code.\n   */\n  run: RunnerMatrixSelection\n  /**\n   * Eval-time matrix selection visible to task code.\n   */\n  eval: RunnerMatrixSelection\n  /**\n   * Stable row ids for both scopes.\n   */\n  meta: ScheduledTaskMatrixMeta\n}\n\n/**\n * Maps matrix axis names to the values that should be expanded.\n */\nexport type RunnerMatrixDefinition = MatrixDefinition\n\n/**\n * Accepts either flat axis definitions or one layered matrix object.\n */\nexport type RunnerMatrixInput = RunnerMatrixDefinition | MatrixLayer\n\nconst matrixLayerKeys = new Set(['disable', 'extend', 'override'])\nconst ambiguousMatrixDefinitionErrorMessage = 'Ambiguous matrix definition: cannot mix reserved layer keys (disable, extend, override) with matrix axis keys.'\n\n/**\n * Represents one fully expanded runner task.\n */\nexport interface ScheduledTask {\n  /**\n   * Stable task id derived from the entry, inferenceExecutor, and matrix selection.\n   */\n  id: string\n  /**\n   * The collected eval entry to execute.\n   */\n  entry: CollectedEvalEntry\n  /**\n   * The inferenceExecutor selected for this task.\n   */\n  inferenceExecutor: InferenceExecutor\n  /**\n   * The concrete scoped matrix selection for this task.\n   */\n  matrix: ScheduledTaskMatrix\n}\n\n/**\n * Configures how the runner should expand its execution matrix.\n */\nexport interface CreateRunnerScheduleOptions {\n  /**\n   * Collected eval entries that should be scheduled.\n   */\n  entries: readonly CollectedEvalEntry[]\n  /**\n   * Providers that should run each entry.\n   */\n  inferenceExecutors: readonly InferenceExecutor[]\n  /**\n   * Optional run-time matrix axes expanded as a cartesian product.\n   */\n  runMatrix?: RunnerMatrixInput\n  /**\n   * Optional eval-time matrix axes expanded as a cartesian product.\n   */\n  evalMatrix?: RunnerMatrixInput\n}\n\nfunction encodeTaskIdSegment(value: string): string {\n  return encodeURIComponent(value)\n}\n\nfunction stringifyMatrixValue(value: MatrixValue): string {\n  return String(value)\n}\n\nfunction cloneMatrixSelection(matrix: RunnerMatrixSelection): RunnerMatrixSelection {\n  return { ...matrix }\n}\n\nfunction createScheduledTaskMatrix(\n  runMatrix: RunnerMatrixSelection,\n  evalMatrix: RunnerMatrixSelection,\n): ScheduledTaskMatrix {\n  return {\n    eval: cloneMatrixSelection(evalMatrix),\n    meta: {\n      evalRowId: createStableRowId(evalMatrix),\n      runRowId: createStableRowId(runMatrix),\n    },\n    run: cloneMatrixSelection(runMatrix),\n  }\n}\n\nfunction isMatrixLayer(matrix: RunnerMatrixInput): matrix is MatrixLayer {\n  const matrixKeys = Object.keys(matrix)\n  return (\n    matrixKeys.length > 0\n    && matrixKeys.every(key => matrixLayerKeys.has(key))\n  )\n}\n\nfunction assertNonAmbiguousMatrixDefinition(matrix: RunnerMatrixInput): void {\n  const matrixKeys = Object.keys(matrix)\n  const hasReservedKeys = matrixKeys.some(key => matrixLayerKeys.has(key))\n  const hasAxisKeys = matrixKeys.some(key => !matrixLayerKeys.has(key))\n\n  if (hasReservedKeys && hasAxisKeys) {\n    throw new TypeError(ambiguousMatrixDefinitionErrorMessage)\n  }\n}\n\nfunction normalizeLayerInputToAxes(matrix: RunnerMatrixInput | undefined): MatrixLayer | undefined {\n  if (matrix == null) {\n    return undefined\n  }\n\n  assertNonAmbiguousMatrixDefinition(matrix)\n\n  if (isMatrixLayer(matrix)) {\n    return matrix\n  }\n\n  return {\n    extend: matrix,\n  }\n}\n\nfunction dedupeAxisValues(values: readonly MatrixValue[]): string[] {\n  return Array.from(new Set(values.map(stringifyMatrixValue)))\n}\n\nfunction applyAxisValues(\n  axes: Map<string, string[]>,\n  definition: RunnerMatrixDefinition | undefined,\n  mode: 'extend' | 'override',\n): void {\n  if (definition == null) {\n    return\n  }\n\n  for (const [axis, values] of Object.entries(definition)) {\n    const nextValues = dedupeAxisValues(values)\n\n    if (mode === 'extend') {\n      const existingValues = axes.get(axis) ?? []\n      axes.set(axis, Array.from(new Set([...existingValues, ...nextValues])))\n      continue\n    }\n\n    axes.set(axis, nextValues)\n  }\n}\n\nfunction applyLayer(\n  baseAxes: ReadonlyMap<string, string[]>,\n  layer: MatrixLayer | undefined,\n): Map<string, string[]> {\n  const nextAxes = new Map<string, string[]>(\n    Array.from(baseAxes.entries()).map(([axis, values]) => [axis, [...values]]),\n  )\n\n  for (const axis of layer?.disable ?? []) {\n    nextAxes.delete(axis)\n  }\n\n  applyAxisValues(nextAxes, layer?.extend, 'extend')\n  applyAxisValues(nextAxes, layer?.override, 'override')\n\n  return nextAxes\n}\n\nfunction expandAxesToRows(axes: ReadonlyMap<string, readonly string[]>): RunnerMatrixSelection[] {\n  if (axes.size === 0) {\n    return [{}]\n  }\n\n  const dimensions = Array.from(axes.entries())\n\n  let selections: RunnerMatrixSelection[] = [{}]\n\n  for (const [axis, values] of dimensions) {\n    if (values.length === 0) {\n      return []\n    }\n\n    const nextSelections: RunnerMatrixSelection[] = []\n\n    for (const selection of selections) {\n      for (const value of values) {\n        nextSelections.push({\n          ...selection,\n          [axis]: value,\n        })\n      }\n    }\n\n    selections = nextSelections\n  }\n\n  return selections\n}\n\nfunction createStableRowId(matrix: RunnerMatrixSelection): string {\n  const segments = Object.entries(matrix)\n    .sort(([leftAxis], [rightAxis]) => leftAxis.localeCompare(rightAxis))\n    .map(([axis, value]) => `${encodeTaskIdSegment(axis)}=${encodeTaskIdSegment(value)}`)\n\n  if (segments.length === 0) {\n    return 'default'\n  }\n\n  return segments.join('&')\n}\n\nfunction createTaskId(entryId: string, inferenceExecutorId: string, runRowId: string, evalRowId: string): string {\n  const encodedEntryId = encodeTaskIdSegment(entryId)\n  const encodedProviderId = encodeTaskIdSegment(inferenceExecutorId)\n\n  return [\n    encodedEntryId,\n    encodedProviderId,\n    `run=${encodeTaskIdSegment(runRowId)}`,\n    `eval=${encodeTaskIdSegment(evalRowId)}`,\n  ].join('::')\n}\n\nfunction createResolvedRunAxes(\n  entry: CollectedEvalEntry,\n  runMatrix: RunnerMatrixInput | undefined,\n): Map<string, string[]> {\n  let resolvedAxes = new Map<string, string[]>()\n\n  for (const layerInput of [\n    runMatrix,\n    entry.matrix?.runMatrix,\n    entry.task?.matrix?.runMatrix,\n  ]) {\n    resolvedAxes = applyLayer(resolvedAxes, normalizeLayerInputToAxes(layerInput))\n  }\n\n  return resolvedAxes\n}\n\nfunction createResolvedEvalAxes(\n  entry: CollectedEvalEntry,\n  evalMatrix: RunnerMatrixInput | undefined,\n): Map<string, string[]> {\n  let resolvedAxes = new Map<string, string[]>()\n\n  for (const layerInput of [\n    evalMatrix,\n    entry.matrix?.evalMatrix,\n    entry.task?.matrix?.evalMatrix,\n  ]) {\n    resolvedAxes = applyLayer(resolvedAxes, normalizeLayerInputToAxes(layerInput))\n  }\n\n  return resolvedAxes\n}\n\n/**\n * Expands collected entries into a stable runner schedule.\n *\n * Call stack:\n *\n * {@link collectEvalEntries} (`../runner`)\n *   -> {@link createRunnerSchedule}\n *     -> {@link expandAxesToRows}\n *       -> {@link ScheduledTask}[]\n *\n * Use when:\n * - the runner already knows which eval entries are available\n * - each entry must run against multiple inferenceExecutors or matrix variants\n *\n * Expects:\n * - `entries` and `inferenceExecutors` to be provided in the desired execution order\n * - matrix axes to use insertion order when generating combinations\n */\nexport function createRunnerSchedule(options: CreateRunnerScheduleOptions): ScheduledTask[] {\n  if (options.entries.length === 0) {\n    return []\n  }\n\n  if (options.inferenceExecutors.length === 0) {\n    return []\n  }\n\n  const tasks: ScheduledTask[] = []\n\n  for (const entry of options.entries) {\n    const runSelections = expandAxesToRows(createResolvedRunAxes(entry, options.runMatrix))\n    const evalSelections = expandAxesToRows(createResolvedEvalAxes(entry, options.evalMatrix))\n\n    if (runSelections.length === 0 || evalSelections.length === 0) {\n      continue\n    }\n\n    for (const inferenceExecutor of options.inferenceExecutors) {\n      for (const runMatrix of runSelections) {\n        for (const evalMatrix of evalSelections) {\n          const isolatedMatrix = createScheduledTaskMatrix(runMatrix, evalMatrix)\n\n          tasks.push({\n            entry,\n            id: createTaskId(\n              entry.id,\n              inferenceExecutor.id,\n              isolatedMatrix.meta.runRowId,\n              isolatedMatrix.meta.evalRowId,\n            ),\n            matrix: isolatedMatrix,\n            inferenceExecutor,\n          })\n        }\n      }\n    }\n  }\n\n  return tasks\n}\n","import type { ModelDefinition } from '../../config/models'\nimport type { TaskCacheRuntime } from '../cache'\nimport type { ScheduledTask } from './schedule'\n\n/**\n * Task-scoped execution context exposed to runner executors.\n */\nexport interface TaskExecutionContext {\n  /**\n   * Deterministic cache runtime scoped to the current task project.\n   */\n  cache: TaskCacheRuntime\n  /**\n   * Configured model registrations available to model plugins.\n   */\n  models: readonly ModelDefinition[]\n}\n\n/**\n * Inputs used to build task execution context.\n */\nexport interface CreateTaskExecutionContextOptions {\n  cache?: TaskCacheRuntime\n  models: readonly ModelDefinition[]\n  task: ScheduledTask\n}\n\nfunction createNoopTaskCacheRuntime(): TaskCacheRuntime {\n  return {\n    namespace(name) {\n      return {\n        file(options) {\n          const key = options.key.join('/')\n          throw new Error(`Task cache runtime is not configured. Requested namespace \"${name}\" and key \"${key}\".`)\n        },\n      }\n    },\n  }\n}\n\n/**\n * Creates task-scoped context data for runner execution.\n *\n * Call stack:\n *\n * {@link runScheduledTasks}\n *   -> {@link createTaskExecutionContext}\n *     -> `TaskExecutionContext`\n */\nexport function createTaskExecutionContext(options: CreateTaskExecutionContextOptions): TaskExecutionContext {\n  return {\n    cache: options.cache ?? createNoopTaskCacheRuntime(),\n    models: options.models,\n  }\n}\n"],"mappings":";;;;;;;;;;;AA2BA,SAAS,oBAAoB,OAAuB;CAClD,MAAM,aAAa,MAAM,MAAM;AAC/B,KAAI,WAAW,WAAW,EACxB,QAAO;AAGT,QAAO,WAAW,QAAQ,aAAa,IAAI;;AAG7C,SAAS,mBAAmB,WAA+B,WAAmD;AAC5G,KAAI,aAAa,QAAQ,UAAU,SAAS,EAC1C,QAAO,UAAU,WAAW,IAAI,GAAG,UAAU,MAAM,EAAE,GAAG;AAG1D,KAAI,aAAa,QAAQ,UAAU,WAAW,EAC5C;AAGF,KAAI,cAAc,mBAChB,QAAO;AAGT,KAAI,cAAc,aAChB,QAAO;AAGT,KAAI,cAAc,YAChB,QAAO;;;;;;;;;;;AAeX,SAAgB,+BAA+B,SAAqC;CAClF,MAAM,eAAe,QAAQ,IAAI,KAAI,YAAW,oBAAoB,QAAQ,CAAC;CAC7E,MAAM,YAAY,mBAAmB,QAAQ,KAAK,QAAQ,UAAU;AAEpE,KAAI,aAAa,WAAW,EAC1B,QAAO,aAAa,OAAO,CAAC,WAAW,GAAG,CAAC,YAAY,YAAY;AAGrE,KAAI,aAAa,KACf,QAAO;CAGT,MAAM,cAAc,aAAa,MAAM,GAAG,KAAK,IAAI,GAAG,aAAa,SAAS,EAAE,CAAC;CAC/E,MAAM,OAAO,aAAa,aAAa,SAAS,MAAM;AACtD,QAAO,CAAC,GAAG,aAAa,GAAG,KAAK,GAAG,YAAY;;AAGjD,eAAe,gBAAgB,MAAc,SAAyC;CACpF,MAAM,YAAY,QAAQ,KAAK;CAC/B,MAAM,gBAAgB,GAAG,KAAK,OAAO,QAAQ,IAAI,GAAG,KAAK,KAAK,CAAC,GAAG,KAAK,QAAQ,CAAC,SAAS,GAAG,CAAC,MAAM,GAAG,GAAG;AACzG,OAAM,MAAM,WAAW,EAAE,WAAW,MAAM,CAAC;AAC3C,OAAM,UAAU,eAAe,QAAQ;AACvC,OAAM,OAAO,eAAe,KAAK;;AAGnC,SAAS,sBAAsB,MAA+B;AAC5D,QAAO;EACL;EACA,MAAM,SAAS;AACb,OAAI;AACF,UAAM,OAAO,KAAK;AAClB,WAAO;WAEH;AACJ,WAAO;;;EAGX,iBAAiB;AACf,UAAO,iBAAiB,KAAK;;EAE/B,MAAM,kBAAkB;AACtB,SAAM,MAAM,QAAQ,KAAK,EAAE,EAAE,WAAW,MAAM,CAAC;AAC/C,UAAO,kBAAkB,KAAK;;EAEhC,MAAM,aAAa;AACjB,UAAO,MAAM,SAAS,KAAK;;EAE7B,MAAM,YAAY,OAAO;AACvB,SAAM,gBAAgB,MAAM,MAAM;;EAEpC,MAAM,SAAS,WAAW,SAAS;AACjC,UAAO,MAAM,SAAS,MAAM,SAAS;;EAEvC,MAAM,UAAU,OAAO,WAAW,SAAS;AACzC,SAAM,gBAAgB,MAAM,OAAO,KAAK,OAAO,SAAS,CAAC;;EAE3D,MAAM,WAAc;AAClB,UAAO,KAAK,MAAM,MAAM,SAAS,MAAM,QAAQ,CAAC;;EAElD,MAAM,UAAU,OAAO;AACrB,SAAM,gBAAgB,MAAM,GAAG,KAAK,UAAU,OAAO,MAAM,EAAE,CAAC,IAAI;;EAEpE,MAAM,mBAAsB;AAC1B,UAAO,MAAM,KAAK,UAAe;;EAEnC,MAAM,sBAAyB;AAC7B,UAAO,MAAM,KAAK,UAAa;;EAElC;;AAGH,SAAS,qBAAqB,eAAuB,WAAmC;AACtF,QAAO,EACL,KAAK,SAAS;EACZ,MAAM,uBAAuB,+BAA+B,QAAQ;AACpE,SAAO,sBAAsB,KAAK,eAAe,oBAAoB,UAAU,EAAE,GAAG,qBAAqB,CAAC;IAE7G;;;;;;;;;;;;;;;;;AAkBH,SAAgB,iCACd,SACkB;CAClB,MAAM,qBAAqB,oBAAoB,QAAQ,YAAY;CACnE,MAAM,mBAAmB,oBAAoB,QAAQ,YAAY;CACjE,MAAM,gBAAgB,KAAK,QAAQ,oBAAoB,oBAAoB,iBAAiB;AAE5F,QAAO,EACL,UAAU,MAAM;AACd,SAAO,qBAAqB,eAAe,KAAK;IAEnD;;;;ACvCH,SAAS,yBAAyB,QAAkD;AAClF,QAAO;EACL,MAAM,EACJ,GAAG,OAAO,MACX;EACD,MAAM,EACJ,GAAG,OAAO,MACX;EACD,KAAK,EACH,GAAG,OAAO,KACX;EACF;;AAGH,SAAS,qBAAqB,MAA4B;AACxD,KAAI,SAAS,WAAW,SAAS,QAC/B,QAAO;AAGT,OAAM,IAAI,UAAU,4BAA4B,KAAK,IAAI;;AAG3D,SAAS,QAAQ,QAA0C;AACzD,KAAI,OAAO,WAAW,EACpB,QAAO;AAIT,QADc,OAAO,QAAQ,KAAK,UAAU,MAAM,OAAO,EAAE,GAC5C,OAAO;;AAGxB,SAAS,oBAAoB,cAA6B,cAA4C;AACpG,KAAI,gBAAgB,QAAQ,gBAAgB,KAC1C,SAAQ,eAAe,gBAAgB;AAGzC,KAAI,gBAAgB,KAClB,QAAO;AAGT,KAAI,gBAAgB,KAClB,QAAO;AAGT,QAAO;;AAGT,SAAS,oBAAoB,QAA2C;CACtE,MAAM,UAAwB;EAC5B,OAAO,EAAE;EACT,OAAO,EAAE;EACV;AAED,MAAK,MAAM,SAAS,QAAQ;AAG1B,MAFa,qBAAqB,MAAM,KAAK,KAEhC,SAAS;AACpB,WAAQ,MAAM,KAAK,MAAM,MAAM;AAC/B;;AAGF,UAAQ,MAAM,KAAK,MAAM,MAAM;;AAGjC,QAAO;;AAGT,SAAS,iBAAiB,QAAyC;CACjE,MAAM,UAAU,oBAAoB,OAAO,OAAO;CAClD,MAAM,eAAe,QAAQ,QAAQ,MAAM;CAC3C,MAAM,eAAe,QAAQ,QAAQ,MAAM;AAE3C,QAAO;EACL,SAAS,OAAO;EAChB;EACA,eAAe,oBAAoB,cAAc,aAAa;EAC9D,IAAI,OAAO;EACX;EACA,QAAQ,yBAAyB,OAAO,OAAO;EAC/C,qBAAqB,OAAO;EAC7B;;AAGH,SAAS,sBAAsB,qBAA6B,SAA0D;CACpH,MAAM,cAAwB,EAAE;CAChC,MAAM,cAAwB,EAAE;AAEhC,MAAK,MAAM,UAAU,SAAS;EAC5B,MAAM,UAAU,oBAAoB,OAAO,OAAO;AAClD,cAAY,KAAK,GAAG,QAAQ,MAAM;AAClC,cAAY,KAAK,GAAG,QAAQ,MAAM;;CAGpC,MAAM,eAAe,QAAQ,YAAY;CACzC,MAAM,eAAe,QAAQ,YAAY;AAEzC,QAAO;EACL;EACA,eAAe,oBAAoB,cAAc,aAAa;EAC9D;EACA;EACA,UAAU,QAAQ;EACnB;;;;;;;;;;;;;;;;;;;;;AAsBH,SAAgB,oBAAoB,SAAqD;CACvF,MAAM,OAAO,QAAQ,IAAI,iBAAiB;CAG1C,MAAM,qBADuB,MAAM,KAAK,IAAI,IAAI,QAAQ,KAAI,WAAU,OAAO,oBAAoB,CAAC,CAAC,CAEhG,KAAK,wBAAwB;AAE5B,SAAO,sBAAsB,qBADL,QAAQ,QAAO,WAAU,OAAO,wBAAwB,oBAAoB,CAClC;GAClE,CACD,MAAM,MAAM,UAAU,KAAK,oBAAoB,cAAc,MAAM,oBAAoB,CAAC;CAE3F,MAAM,UAAU,sBACd,WACA,QACD;AAED,QAAO;EACL,SAAS;GACP,cAAc,QAAQ;GACtB,eAAe,QAAQ;GACvB,cAAc,QAAQ;GACtB,UAAU,QAAQ;GACnB;EACD;EACA;EACD;;;;ACvRH,MAAM,iBAAiB;AACvB,MAAM,sBAAsB;AAE5B,SAAS,cAAc,OAAuB;AAC5C,QAAO,MAAM,WAAW,MAAM,IAAI;;;;;;;;;;;AAYpC,SAAgB,sBAAsB,UAAkB,SAAuC;CAC7F,MAAM,qBAAqB,cAAc,SAAS;CAClD,MAAM,iCAAiC,cAAc,QAAQ,qBAAqB;CAClF,MAAM,uBAAuB,mBAAmB,MAAM,aAAa,GAAG;CACtE,MAAM,0BAA0B,+BAA+B,MAAM,aAAa,GAAG;AAErF,KAAI,wBAAwB,QAAQ,2BAA2B,KAC7D,QAAO;AAGT,KACE,wBAAwB,QACrB,2BAA2B,QAC3B,qBAAqB,aAAa,KAAK,wBAAwB,aAAa,CAE/E,QAAO;CAGT,MAAM,uBAAuB,QAAQ;CACrC,MAAM,mBAAmB,cAAc,SAAS,sBAAsB,SAAS,CAAC;AAEhF,KAAI,CAAC,oBAAoB,KAAK,iBAAiB,EAAE;AAC/C,MAAI,qBAAqB,KACvB,QAAO,cAAc,SAAS;AAGhC,MAAI,CAAC,iBAAiB,WAAW,MAAM,CACrC,QAAO;;AAIX,QAAO,cAAc,SAAS;;AAGhC,SAAS,sBAAsB,YAAmC;AAChE,KAAI,CAAC,WAAW,WAAW,QAAQ,CACjC,QAAO;AAGT,KAAI;AACF,SAAO,cAAc,WAAW;SAE5B;AACJ,SAAO;;;AAIX,SAAS,yBACP,YACA,kBACA,SAC2B;CAC3B,MAAM,WAAW,sBAAsB,WAAW;AAElD,KAAI,CAAC,SACH,QAAO;CAGT,MAAM,mBAAmB,sBAAsB,UAAU,QAAQ;AAEjE,KAAI,CAAC,iBAAiB,SAAS,eAAe,CAC5C,QAAO;CAGT,MAAM,YAAY,SAAS,kBAAkB,eAAe;AAE5D,KAAI,UAAU,WAAW,EACvB,QAAO;CAGT,MAAM,oBAAoB,QAAQ,iBAAiB;CACnD,MAAM,YAAY,sBAAsB,MAAM,KAAK;AAEnD,QAAO;EACL,GAAG,iBAAiB;EACpB;EACA;EACA,IAAI,UAAU,WAAW,IAAI,YAAY,GAAG,UAAU,GAAG;EACzD,MAAM;EACP;;;;;;;;;;;;;;;;AAiBH,SAAgB,mBACd,SACA,SACsB;AACtB,QAAO,OAAO,QAAQ,QAAQ,CAC3B,SAAS,CAAC,YAAY,sBAAsB;EAC3C,MAAM,QAAQ,yBAAyB,YAAY,kBAAkB,QAAQ;AAE7E,MAAI,CAAC,MACH,QAAO,EAAE;AAGX,SAAO,CAAC,MAAM;GACd,CACD,MAAM,MAAM,UAAU,KAAK,GAAG,cAAc,MAAM,GAAG,CAAC;;;;AC9C3D,SAAS,gCAAsD;AAY7D,QAAO;EACL,OAZ8B,EAC9B,UAAU,MAAM;AACd,UAAO,EACL,KAAK,SAAS;IACZ,MAAM,MAAM,QAAQ,IAAI,KAAK,IAAI;AACjC,UAAM,IAAI,MAAM,8DAA8D,KAAK,aAAa,IAAI,IAAI;MAE3G;KAEJ;EAIC,QAAQ,EAAE;EACX;;;;;AAMH,IAAa,uBAAb,cAA0C,MAAM;;;;CAI9C;CAEA,YAAY,QAAgB,OAAgB;EAC1C,MAAM,UAAU,iBAAiB,MAAM,IAAI;AAC3C,QAAM,gBAAgB,OAAO,YAAY,UAAU;AACnD,OAAK,OAAO;AACZ,OAAK,SAAS;AACd,OAAK,QAAQ;;;AAIjB,SAAS,2BAA2B,QAAgB,OAAsC;AACxF,KAAI,iBAAiB,wBAAwB,MAAM,WAAW,OAC5D,QAAO;AAGT,QAAO,IAAI,qBAAqB,QAAQ,MAAM;;;;;;;;;;;;;;;;;;;;;;;;AAyBhD,eAAsB,kBACpB,OACA,UACA,UAAoC,EAAE,EACP;AAC/B,KAAI,MAAM,WAAW,EACnB,QAAO,oBAAoB,EAAE,CAAC;CAGhC,eAAe,qBAAqB,MAAyC;EAC3E,IAAI;AAEJ,MAAI;AACF,sBAAmB,QAAQ,yBAAyB,KAAK,IAAI,+BAA+B;WAEvF,OAAO;AACZ,SAAM,2BAA2B,KAAK,IAAI,MAAM;;AAGlD,MAAI;AACF,WAAQ,cAAc,KAAK;WAEtB,OAAO;AACZ,SAAM,2BAA2B,KAAK,IAAI,MAAM;;EAGlD,IAAI;AACJ,MAAI;AACF,eAAY,MAAM,SAAS,MAAM,iBAAiB;WAE7C,OAAO;AACZ,OAAI;AACF,YAAQ,YAAY,MAAM,SAAS;WAE/B;AAGN,SAAM,2BAA2B,KAAK,IAAI,MAAM;;AAGlD,MAAI;AACF,WAAQ,YAAY,MAAM,SAAS;WAE9B,OAAO;AACZ,SAAM,2BAA2B,KAAK,IAAI,MAAM;;AAGlD,SAAO;;CAGT,MAAM,iBAAiB,QAAQ,kBAAkB;AACjD,KAAI,kBAAkB,GAAG;EACvB,MAAM,UAAuB,EAAE;AAC/B,OAAK,MAAM,QAAQ,MACjB,SAAQ,KAAK,MAAM,qBAAqB,KAAK,CAAC;AAEhD,SAAO,oBAAoB,QAAQ;;CAGrC,MAAM,eAAe,iBAAiB,eAAe;AAUrD,QAAO,qBATa,MAAM,QAAQ,IAAI,MAAM,IAAI,OAAO,MAAM,UAAU;AAErE,SAAO;GAAE;GAAO,QADD,MAAM,aAAa,YAAY,qBAAqB,KAAK,CAAC;GACjD;GACxB,CAAC,EAGA,MAAM,MAAM,UAAU,KAAK,QAAQ,MAAM,MAAM,CAC/C,KAAI,SAAQ,KAAK,OAAO,CAEc;;;;ACzN3C,MAAM,UAAU,cAAc,OAAO,KAAK,IAAI;;;;;;;;;;;;;;;AAgD9C,eAAsB,2BACpB,UAAmD,EAAE,EACtB;CAC/B,MAAM,MAAM,QAAQ,OAAO,QAAQ,cAAc,OAAO,KAAK,IAAI,CAAC;CAClE,MAAM,+BAA+B,QAAQ,gCACxC,cAAc,IAAI,IAAI,aAAa,OAAO,KAAK,IAAI,CAAC;CAKzD,MAAM,EAAE,qBAAqB,QAAQ,2BAA2B;AAShE,QAAO,EACL,sBAHyB,MAAM,iBAAiB,IAAI,IAGR,8BAC7C;;;;ACdH,MAAM,kBAAkB,IAAI,IAAI;CAAC;CAAW;CAAU;CAAW,CAAC;AAClE,MAAM,wCAAwC;AA8C9C,SAAS,oBAAoB,OAAuB;AAClD,QAAO,mBAAmB,MAAM;;AAGlC,SAAS,qBAAqB,OAA4B;AACxD,QAAO,OAAO,MAAM;;AAGtB,SAAS,qBAAqB,QAAsD;AAClF,QAAO,EAAE,GAAG,QAAQ;;AAGtB,SAAS,0BACP,WACA,YACqB;AACrB,QAAO;EACL,MAAM,qBAAqB,WAAW;EACtC,MAAM;GACJ,WAAW,kBAAkB,WAAW;GACxC,UAAU,kBAAkB,UAAU;GACvC;EACD,KAAK,qBAAqB,UAAU;EACrC;;AAGH,SAAS,cAAc,QAAkD;CACvE,MAAM,aAAa,OAAO,KAAK,OAAO;AACtC,QACE,WAAW,SAAS,KACjB,WAAW,OAAM,QAAO,gBAAgB,IAAI,IAAI,CAAC;;AAIxD,SAAS,mCAAmC,QAAiC;CAC3E,MAAM,aAAa,OAAO,KAAK,OAAO;CACtC,MAAM,kBAAkB,WAAW,MAAK,QAAO,gBAAgB,IAAI,IAAI,CAAC;CACxE,MAAM,cAAc,WAAW,MAAK,QAAO,CAAC,gBAAgB,IAAI,IAAI,CAAC;AAErE,KAAI,mBAAmB,YACrB,OAAM,IAAI,UAAU,sCAAsC;;AAI9D,SAAS,0BAA0B,QAAgE;AACjG,KAAI,UAAU,KACZ;AAGF,oCAAmC,OAAO;AAE1C,KAAI,cAAc,OAAO,CACvB,QAAO;AAGT,QAAO,EACL,QAAQ,QACT;;AAGH,SAAS,iBAAiB,QAA0C;AAClE,QAAO,MAAM,KAAK,IAAI,IAAI,OAAO,IAAI,qBAAqB,CAAC,CAAC;;AAG9D,SAAS,gBACP,MACA,YACA,MACM;AACN,KAAI,cAAc,KAChB;AAGF,MAAK,MAAM,CAAC,MAAM,WAAW,OAAO,QAAQ,WAAW,EAAE;EACvD,MAAM,aAAa,iBAAiB,OAAO;AAE3C,MAAI,SAAS,UAAU;GACrB,MAAM,iBAAiB,KAAK,IAAI,KAAK,IAAI,EAAE;AAC3C,QAAK,IAAI,MAAM,MAAM,KAAK,IAAI,IAAI,CAAC,GAAG,gBAAgB,GAAG,WAAW,CAAC,CAAC,CAAC;AACvE;;AAGF,OAAK,IAAI,MAAM,WAAW;;;AAI9B,SAAS,WACP,UACA,OACuB;CACvB,MAAM,WAAW,IAAI,IACnB,MAAM,KAAK,SAAS,SAAS,CAAC,CAAC,KAAK,CAAC,MAAM,YAAY,CAAC,MAAM,CAAC,GAAG,OAAO,CAAC,CAAC,CAC5E;AAED,MAAK,MAAM,QAAQ,OAAO,WAAW,EAAE,CACrC,UAAS,OAAO,KAAK;AAGvB,iBAAgB,UAAU,OAAO,QAAQ,SAAS;AAClD,iBAAgB,UAAU,OAAO,UAAU,WAAW;AAEtD,QAAO;;AAGT,SAAS,iBAAiB,MAAuE;AAC/F,KAAI,KAAK,SAAS,EAChB,QAAO,CAAC,EAAE,CAAC;CAGb,MAAM,aAAa,MAAM,KAAK,KAAK,SAAS,CAAC;CAE7C,IAAI,aAAsC,CAAC,EAAE,CAAC;AAE9C,MAAK,MAAM,CAAC,MAAM,WAAW,YAAY;AACvC,MAAI,OAAO,WAAW,EACpB,QAAO,EAAE;EAGX,MAAM,iBAA0C,EAAE;AAElD,OAAK,MAAM,aAAa,WACtB,MAAK,MAAM,SAAS,OAClB,gBAAe,KAAK;GAClB,GAAG;IACF,OAAO;GACT,CAAC;AAIN,eAAa;;AAGf,QAAO;;AAGT,SAAS,kBAAkB,QAAuC;CAChE,MAAM,WAAW,OAAO,QAAQ,OAAO,CACpC,MAAM,CAAC,WAAW,CAAC,eAAe,SAAS,cAAc,UAAU,CAAC,CACpE,KAAK,CAAC,MAAM,WAAW,GAAG,oBAAoB,KAAK,CAAC,GAAG,oBAAoB,MAAM,GAAG;AAEvF,KAAI,SAAS,WAAW,EACtB,QAAO;AAGT,QAAO,SAAS,KAAK,IAAI;;AAG3B,SAAS,aAAa,SAAiB,qBAA6B,UAAkB,WAA2B;AAI/G,QAAO;EAHgB,oBAAoB,QAAQ;EACzB,oBAAoB,oBAAoB;EAKhE,OAAO,oBAAoB,SAAS;EACpC,QAAQ,oBAAoB,UAAU;EACvC,CAAC,KAAK,KAAK;;AAGd,SAAS,sBACP,OACA,WACuB;CACvB,IAAI,+BAAe,IAAI,KAAuB;AAE9C,MAAK,MAAM,cAAc;EACvB;EACA,MAAM,QAAQ;EACd,MAAM,MAAM,QAAQ;EACrB,CACC,gBAAe,WAAW,cAAc,0BAA0B,WAAW,CAAC;AAGhF,QAAO;;AAGT,SAAS,uBACP,OACA,YACuB;CACvB,IAAI,+BAAe,IAAI,KAAuB;AAE9C,MAAK,MAAM,cAAc;EACvB;EACA,MAAM,QAAQ;EACd,MAAM,MAAM,QAAQ;EACrB,CACC,gBAAe,WAAW,cAAc,0BAA0B,WAAW,CAAC;AAGhF,QAAO;;;;;;;;;;;;;;;;;;;;AAqBT,SAAgB,qBAAqB,SAAuD;AAC1F,KAAI,QAAQ,QAAQ,WAAW,EAC7B,QAAO,EAAE;AAGX,KAAI,QAAQ,mBAAmB,WAAW,EACxC,QAAO,EAAE;CAGX,MAAM,QAAyB,EAAE;AAEjC,MAAK,MAAM,SAAS,QAAQ,SAAS;EACnC,MAAM,gBAAgB,iBAAiB,sBAAsB,OAAO,QAAQ,UAAU,CAAC;EACvF,MAAM,iBAAiB,iBAAiB,uBAAuB,OAAO,QAAQ,WAAW,CAAC;AAE1F,MAAI,cAAc,WAAW,KAAK,eAAe,WAAW,EAC1D;AAGF,OAAK,MAAM,qBAAqB,QAAQ,mBACtC,MAAK,MAAM,aAAa,cACtB,MAAK,MAAM,cAAc,gBAAgB;GACvC,MAAM,iBAAiB,0BAA0B,WAAW,WAAW;AAEvE,SAAM,KAAK;IACT;IACA,IAAI,aACF,MAAM,IACN,kBAAkB,IAClB,eAAe,KAAK,UACpB,eAAe,KAAK,UACrB;IACD,QAAQ;IACR;IACD,CAAC;;;AAMV,QAAO;;;;AC1UT,SAAS,6BAA+C;AACtD,QAAO,EACL,UAAU,MAAM;AACd,SAAO,EACL,KAAK,SAAS;GACZ,MAAM,MAAM,QAAQ,IAAI,KAAK,IAAI;AACjC,SAAM,IAAI,MAAM,8DAA8D,KAAK,aAAa,IAAI,IAAI;KAE3G;IAEJ;;;;;;;;;;;AAYH,SAAgB,2BAA2B,SAAkE;AAC3G,QAAO;EACL,OAAO,QAAQ,SAAS,4BAA4B;EACpD,QAAQ,QAAQ;EACjB"}
+{"version":3,"file":"index.mjs","names":[],"sources":["../../../src/core/cache/filesystem.ts","../../../src/core/runner/aggregate.ts","../../../src/core/runner/collect.ts","../../../src/core/runner/run.ts","../../../src/core/runner/runtime-context.ts","../../../src/core/runner/schedule.ts","../../../src/core/runner/task-context.ts"],"sourcesContent":["import type { CacheFileHandle, CacheFileOptions, CacheNamespace, TaskCacheRuntime } from './types'\n\nimport process from 'node:process'\n\nimport { Buffer } from 'node:buffer'\nimport { createReadStream, createWriteStream } from 'node:fs'\nimport { access, mkdir, readFile, rename, writeFile } from 'node:fs/promises'\nimport { dirname, join } from 'node:path'\n\n/**\n * Options for creating the filesystem-backed task cache runtime.\n */\nexport interface CreateFilesystemTaskCacheRuntimeOptions {\n  /**\n   * Absolute cache root directory.\n   */\n  cacheRootDirectory: string\n  /**\n   * Project identifier under one workspace cache scope.\n   */\n  projectName: string\n  /**\n   * Workspace identifier used to share cache roots across projects.\n   */\n  workspaceId: string\n}\n\nfunction sanitizePathSegment(value: string): string {\n  const normalized = value.trim()\n  if (normalized.length === 0) {\n    return 'default'\n  }\n\n  return normalized.replace(/[^\\w.-]+/g, '-')\n}\n\nfunction normalizeExtension(extension: string | undefined, mediaType: string | undefined): string | undefined {\n  if (extension != null && extension.length > 0) {\n    return extension.startsWith('.') ? extension.slice(1) : extension\n  }\n\n  if (mediaType == null || mediaType.length === 0) {\n    return undefined\n  }\n\n  if (mediaType === 'application/json') {\n    return 'json'\n  }\n\n  if (mediaType === 'text/plain') {\n    return 'txt'\n  }\n\n  if (mediaType === 'audio/wav') {\n    return 'wav'\n  }\n\n  return undefined\n}\n\n/**\n * Normalizes cache file options into deterministic relative path segments.\n *\n * Before:\n * - `{ key: ['cases', 'dataset hash', 'v1'], ext: 'json' }`\n *\n * After:\n * - `['cases', 'dataset-hash', 'v1.json']`\n */\nexport function normalizeCacheFilePathSegments(options: CacheFileOptions): string[] {\n  const sanitizedKey = options.key.map(segment => sanitizePathSegment(segment))\n  const extension = normalizeExtension(options.ext, options.mediaType)\n\n  if (sanitizedKey.length === 0) {\n    return extension == null ? ['artifact'] : [`artifact.${extension}`]\n  }\n\n  if (extension == null) {\n    return sanitizedKey\n  }\n\n  const withoutTail = sanitizedKey.slice(0, Math.max(0, sanitizedKey.length - 1))\n  const tail = sanitizedKey[sanitizedKey.length - 1] ?? 'artifact'\n  return [...withoutTail, `${tail}.${extension}`]\n}\n\nasync function writeAtomically(path: string, content: Buffer | string): Promise<void> {\n  const directory = dirname(path)\n  const temporaryPath = `${path}.tmp-${process.pid}-${Date.now()}-${Math.random().toString(36).slice(2, 10)}`\n  await mkdir(directory, { recursive: true })\n  await writeFile(temporaryPath, content)\n  await rename(temporaryPath, path)\n}\n\nfunction createCacheFileHandle(path: string): CacheFileHandle {\n  return {\n    path,\n    async exists() {\n      try {\n        await access(path)\n        return true\n      }\n      catch {\n        return false\n      }\n    },\n    openReadStream() {\n      return createReadStream(path)\n    },\n    async openWriteStream() {\n      await mkdir(dirname(path), { recursive: true })\n      return createWriteStream(path)\n    },\n    async readBuffer() {\n      return await readFile(path)\n    },\n    async writeBuffer(value) {\n      await writeAtomically(path, value)\n    },\n    async readText(encoding = 'utf-8') {\n      return await readFile(path, encoding)\n    },\n    async writeText(value, encoding = 'utf-8') {\n      await writeAtomically(path, Buffer.from(value, encoding))\n    },\n    async readJson<T>() {\n      return JSON.parse(await readFile(path, 'utf-8')) as T\n    },\n    async writeJson(value) {\n      await writeAtomically(path, `${JSON.stringify(value, null, 2)}\\n`)\n    },\n    async loadAsCasesInput<T>() {\n      return await this.readJson<T[]>()\n    },\n    async loadAsExpectFixture<T>() {\n      return await this.readJson<T>()\n    },\n  }\n}\n\nfunction createCacheNamespace(baseDirectory: string, namespace: string): CacheNamespace {\n  return {\n    file(options) {\n      const relativePathSegments = normalizeCacheFilePathSegments(options)\n      return createCacheFileHandle(join(baseDirectory, sanitizePathSegment(namespace), ...relativePathSegments))\n    },\n  }\n}\n\n/**\n * Creates a deterministic filesystem-backed task cache runtime.\n *\n * Use when:\n * - eval tasks need reproducible cache paths for expensive pre-processing outputs\n * - benchmark adapters need one artifact-oriented API for text/json/binary reads and writes\n *\n * Expects:\n * - `cacheRootDirectory` to be writable by the running process\n * - `workspaceId` + `projectName` to stay stable for reproducible paths\n *\n * Returns:\n * - task cache runtime that resolves namespaced file handles under:\n *   `<cacheRootDirectory>/<workspaceId>/<projectName>/<namespace>/...`\n */\nexport function createFilesystemTaskCacheRuntime(\n  options: CreateFilesystemTaskCacheRuntimeOptions,\n): TaskCacheRuntime {\n  const workspaceDirectory = sanitizePathSegment(options.workspaceId)\n  const projectDirectory = sanitizePathSegment(options.projectName)\n  const baseDirectory = join(options.cacheRootDirectory, workspaceDirectory, projectDirectory)\n\n  return {\n    namespace(name) {\n      return createCacheNamespace(baseDirectory, name)\n    },\n  }\n}\n","import type { ScheduledTaskMatrix } from './schedule'\n\n/**\n * Identifies the scoring family for a single eval score.\n */\nexport type RunScoreKind = 'exact' | 'judge'\n\n/**\n * Represents one normalized score emitted by a completed eval run.\n */\nexport interface RunScore {\n  /**\n   * Score family used for aggregation.\n   */\n  kind: RunScoreKind\n  /**\n   * Normalized score in the `0..1` range.\n   */\n  score: number\n}\n\n/**\n * Captures the output of one scheduled runner task.\n */\nexport interface RunResult {\n  /**\n   * Stable run id, usually copied from the scheduled task id.\n   */\n  id: string\n  /**\n   * Collected eval entry id.\n   */\n  entryId: string\n  /**\n   * Stable inferenceExecutor id.\n   */\n  inferenceExecutorId: string\n  /**\n   * Concrete matrix selection used by the run.\n   */\n  matrix: ScheduledTaskMatrix\n  /**\n   * Raw scores emitted by the eval.\n   */\n  scores: readonly RunScore[]\n}\n\n/**\n * Stores the per-run score averages after normalization.\n */\nexport interface AggregatedRunSummary {\n  /**\n   * Stable run id.\n   */\n  id: string\n  /**\n   * Collected eval entry id.\n   */\n  entryId: string\n  /**\n   * Stable inferenceExecutor id.\n   */\n  inferenceExecutorId: string\n  /**\n   * Concrete matrix selection used by the run.\n   */\n  matrix: ScheduledTaskMatrix\n  /**\n   * Mean of exact-match scores or `null` when absent.\n   */\n  exactAverage: number | null\n  /**\n   * Mean of judge-based scores or `null` when absent.\n   */\n  judgeAverage: number | null\n  /**\n   * Hybrid average. Uses both families when present, otherwise falls back to the\n   * single available family.\n   */\n  hybridAverage: number | null\n}\n\n/**\n * Stores inferenceExecutor-level score aggregates across multiple runs.\n */\nexport interface AggregatedProviderSummary {\n  /**\n   * Stable inferenceExecutor id.\n   */\n  inferenceExecutorId: string\n  /**\n   * Number of runs included in this inferenceExecutor bucket.\n   */\n  runCount: number\n  /**\n   * Mean of all exact-match scores or `null` when absent.\n   */\n  exactAverage: number | null\n  /**\n   * Mean of all judge-based scores or `null` when absent.\n   */\n  judgeAverage: number | null\n  /**\n   * Hybrid average derived from the inferenceExecutor exact and judge means.\n   */\n  hybridAverage: number | null\n}\n\n/**\n * Stores the final aggregation output for a batch of runner results.\n */\nexport interface AggregatedRunResults {\n  /**\n   * Per-run normalized score summaries.\n   */\n  runs: AggregatedRunSummary[]\n  /**\n   * Provider-level summaries sorted by inferenceExecutor id.\n   */\n  inferenceExecutors: AggregatedProviderSummary[]\n  /**\n   * Overall summary across every run.\n   */\n  overall: {\n    exactAverage: number | null\n    judgeAverage: number | null\n    hybridAverage: number | null\n    runCount: number\n  }\n}\n\ninterface ScoreBuckets {\n  exact: number[]\n  judge: number[]\n}\n\nfunction cloneScheduledTaskMatrix(matrix: ScheduledTaskMatrix): ScheduledTaskMatrix {\n  return {\n    eval: {\n      ...matrix.eval,\n    },\n    meta: {\n      ...matrix.meta,\n    },\n    run: {\n      ...matrix.run,\n    },\n  }\n}\n\nfunction assertKnownScoreKind(kind: string): RunScoreKind {\n  if (kind === 'exact' || kind === 'judge') {\n    return kind\n  }\n\n  throw new TypeError(`Unknown eval score kind \"${kind}\".`)\n}\n\nfunction average(scores: readonly number[]): number | null {\n  if (scores.length === 0) {\n    return null\n  }\n\n  const total = scores.reduce((sum, score) => sum + score, 0)\n  return total / scores.length\n}\n\nfunction createHybridAverage(exactAverage: number | null, judgeAverage: number | null): number | null {\n  if (exactAverage != null && judgeAverage != null) {\n    return (exactAverage + judgeAverage) / 2\n  }\n\n  if (exactAverage != null) {\n    return exactAverage\n  }\n\n  if (judgeAverage != null) {\n    return judgeAverage\n  }\n\n  return null\n}\n\nfunction collectScoreBuckets(scores: readonly RunScore[]): ScoreBuckets {\n  const buckets: ScoreBuckets = {\n    exact: [],\n    judge: [],\n  }\n\n  for (const score of scores) {\n    const kind = assertKnownScoreKind(score.kind)\n\n    if (kind === 'exact') {\n      buckets.exact.push(score.score)\n      continue\n    }\n\n    buckets.judge.push(score.score)\n  }\n\n  return buckets\n}\n\nfunction createRunSummary(result: RunResult): AggregatedRunSummary {\n  const buckets = collectScoreBuckets(result.scores)\n  const exactAverage = average(buckets.exact)\n  const judgeAverage = average(buckets.judge)\n\n  return {\n    entryId: result.entryId,\n    exactAverage,\n    hybridAverage: createHybridAverage(exactAverage, judgeAverage),\n    id: result.id,\n    judgeAverage,\n    matrix: cloneScheduledTaskMatrix(result.matrix),\n    inferenceExecutorId: result.inferenceExecutorId,\n  }\n}\n\nfunction createProviderSummary(inferenceExecutorId: string, results: readonly RunResult[]): AggregatedProviderSummary {\n  const exactScores: number[] = []\n  const judgeScores: number[] = []\n\n  for (const result of results) {\n    const buckets = collectScoreBuckets(result.scores)\n    exactScores.push(...buckets.exact)\n    judgeScores.push(...buckets.judge)\n  }\n\n  const exactAverage = average(exactScores)\n  const judgeAverage = average(judgeScores)\n\n  return {\n    exactAverage,\n    hybridAverage: createHybridAverage(exactAverage, judgeAverage),\n    judgeAverage,\n    inferenceExecutorId,\n    runCount: results.length,\n  }\n}\n\n/**\n * Aggregates exact-match and judge-based scores into hybrid runner summaries.\n *\n * Call stack:\n *\n * {@link runScheduledTasks}\n *   -> {@link aggregateRunResults}\n *     -> {@link createRunSummary}\n *     -> {@link createProviderSummary}\n *       -> `report output`\n *\n * Use when:\n * - a runner batch mixes deterministic exact checks with judge-based grading\n * - inferenceExecutor comparison should preserve both score families and one hybrid view\n *\n * Expects:\n * - each score to be normalized to the `0..1` range before aggregation\n * - `scores.kind` to use only `'exact'` or `'judge'`\n */\nexport function aggregateRunResults(results: readonly RunResult[]): AggregatedRunResults {\n  const runs = results.map(createRunSummary)\n\n  const inferenceExecutorIds = Array.from(new Set(results.map(result => result.inferenceExecutorId)))\n  const inferenceExecutors = inferenceExecutorIds\n    .map((inferenceExecutorId) => {\n      const providerResults = results.filter(result => result.inferenceExecutorId === inferenceExecutorId)\n      return createProviderSummary(inferenceExecutorId, providerResults)\n    })\n    .sort((left, right) => left.inferenceExecutorId.localeCompare(right.inferenceExecutorId))\n\n  const overall = createProviderSummary(\n    'overall',\n    results,\n  )\n\n  return {\n    overall: {\n      exactAverage: overall.exactAverage,\n      hybridAverage: overall.hybridAverage,\n      judgeAverage: overall.judgeAverage,\n      runCount: overall.runCount,\n    },\n    inferenceExecutors,\n    runs,\n  }\n}\n","import type { CollectedEvalEntry, EvalModule, EvalModuleMap } from '../../config'\nimport type { RunnerRuntimeContext } from './runtime-context'\n\nimport { basename, dirname, relative } from 'node:path'\nimport { fileURLToPath } from 'node:url'\n\nconst evalFileSuffix = '.eval.ts'\nconst absolutePathPattern = /^(?:[A-Z]:\\/|\\/|\\\\\\\\)/i\n\nfunction normalizePath(value: string): string {\n  return value.replaceAll('\\\\', '/')\n}\n\n/**\n * Converts a file path into a project-relative path when possible.\n *\n * Before: `/repo/plugins/airi-plugin-game-chess/src/agent/evals/chess-commentary.eval.ts`\n * After: `plugins/airi-plugin-game-chess/src/agent/evals/chess-commentary.eval.ts`\n *\n * Before: `D:/repo/plugins/airi-plugin-game-chess/src/agent/evals/chess-commentary.eval.ts`\n * After: `D:/repo/plugins/airi-plugin-game-chess/src/agent/evals/chess-commentary.eval.ts`\n */\nexport function asProjectRelativePath(filePath: string, context: RunnerRuntimeContext): string {\n  const normalizedFilePath = normalizePath(filePath)\n  const normalizedProjectRootDirectory = normalizePath(context.projectRootDirectory)\n  const filePathWindowsDrive = normalizedFilePath.match(/^[A-Z]:\\//i)?.[0]\n  const projectRootWindowsDrive = normalizedProjectRootDirectory.match(/^[A-Z]:\\//i)?.[0]\n\n  if (filePathWindowsDrive != null && projectRootWindowsDrive == null) {\n    return normalizedFilePath\n  }\n\n  if (\n    filePathWindowsDrive != null\n    && projectRootWindowsDrive != null\n    && filePathWindowsDrive.toLowerCase() !== projectRootWindowsDrive.toLowerCase()\n  ) {\n    return normalizedFilePath\n  }\n\n  const projectRootDirectory = context.projectRootDirectory\n  const relativeFilePath = normalizePath(relative(projectRootDirectory, filePath))\n\n  if (!absolutePathPattern.test(relativeFilePath)) {\n    if (relativeFilePath === '..') {\n      return normalizePath(filePath)\n    }\n\n    if (!relativeFilePath.startsWith('../')) {\n      return relativeFilePath\n    }\n  }\n\n  return normalizePath(filePath)\n}\n\nfunction resolveModuleFilePath(moduleHref: string): string | null {\n  if (!moduleHref.startsWith('file:')) {\n    return null\n  }\n\n  try {\n    return fileURLToPath(moduleHref)\n  }\n  catch {\n    return null\n  }\n}\n\nfunction createCollectedEvalEntry(\n  moduleHref: string,\n  moduleDefinition: EvalModule,\n  context: RunnerRuntimeContext,\n): CollectedEvalEntry | null {\n  const filePath = resolveModuleFilePath(moduleHref)\n\n  if (!filePath) {\n    return null\n  }\n\n  const relativeFilePath = asProjectRelativePath(filePath, context)\n\n  if (!relativeFilePath.endsWith(evalFileSuffix)) {\n    return null\n  }\n\n  const entryName = basename(relativeFilePath, evalFileSuffix)\n\n  if (entryName.length === 0) {\n    return null\n  }\n\n  const relativeDirectory = dirname(relativeFilePath)\n  const directory = relativeDirectory === '.' ? '' : relativeDirectory\n\n  return {\n    ...moduleDefinition.default,\n    directory,\n    filePath,\n    id: directory.length === 0 ? entryName : `${directory}/${entryName}`,\n    name: entryName,\n  }\n}\n\n/**\n * Collects loaded vieval modules into sorted runner entries with stable ids.\n *\n * Call stack:\n *\n * `import.meta.glob(...)`\n *   -> {@link collectEvalEntries}\n *     -> {@link createCollectedEvalEntry}\n *       -> {@link CollectedEvalEntry}[]\n *\n * Use when:\n * - the runner has already loaded candidate eval modules\n * - downstream scheduling needs stable entry ids and directory metadata\n */\nexport function collectEvalEntries(\n  modules: EvalModuleMap,\n  context: RunnerRuntimeContext,\n): CollectedEvalEntry[] {\n  return Object.entries(modules)\n    .flatMap(([moduleHref, moduleDefinition]) => {\n      const entry = createCollectedEvalEntry(moduleHref, moduleDefinition, context)\n\n      if (!entry) {\n        return []\n      }\n\n      return [entry]\n    })\n    .sort((left, right) => left.id.localeCompare(right.id))\n}\n","import type { TaskCacheRuntime } from '../cache'\nimport type { AggregatedRunResults, RunResult } from './aggregate'\nimport type { ScheduledTask } from './schedule'\nimport type { TaskExecutionContext } from './task-context'\n\nimport { errorMessageFrom } from '@moeru/std'\nimport { limitConcurrency } from '@vitest/runner/utils'\n\nimport { aggregateRunResults } from './aggregate'\n\n/**\n * Executes one scheduled runner task and returns a normalized run result.\n *\n * Use when:\n * - a scheduler already selected the task and execution context\n * - the caller wants a typed executor contract for runner workers\n *\n * Expects:\n * - the task context to be ready for model resolution and task-scoped work\n *\n * Returns:\n * - a normalized run result with score entries ready for aggregation\n */\nexport type ScheduledTaskExecutor = (\n  task: ScheduledTask,\n  context: TaskExecutionContext,\n) => Promise<RunResult>\n\n/**\n * Terminal task state reported by runner lifecycle hooks.\n *\n * Use when:\n * - reporting the outcome of one scheduled task to lifecycle observers\n *\n * Expects:\n * - hooks treat the value as final for the completed task\n */\nexport type RunnerTaskState = 'passed' | 'failed'\n\n/**\n * Optional runner execution hooks used while processing scheduled tasks.\n *\n * Use when:\n * - callers want lifecycle visibility around sequential task execution\n * - task execution should remain deterministic while still observable\n *\n * Expects:\n * - hook functions are synchronous lifecycle observers\n */\nexport interface RunScheduledTasksOptions {\n  /**\n   * Creates per-task execution context.\n   *\n   * Use when:\n   * - executor code needs per-task models, cache, or other task-scoped data\n   */\n  createExecutionContext?: (task: ScheduledTask) => TaskExecutionContext\n  /**\n   * Runs before the executor starts handling a task.\n   *\n   * Use when:\n   * - callers want to observe task activation before execution begins\n   *\n   * Expects:\n   * - thrown errors abort the task before executor work starts\n   */\n  onTaskStart?: (task: ScheduledTask) => void\n  /**\n   * Runs after the executor settles for a task.\n   *\n   * Use when:\n   * - callers want to observe successful and failed task completion\n   *\n   * Expects:\n   * - thrown errors abort successful runs\n   * - failed-task observers do not override the executor error for the task\n   */\n  onTaskEnd?: (task: ScheduledTask, state: RunnerTaskState) => void\n  /**\n   * Maximum number of tasks to execute concurrently.\n   *\n   * @default 1\n   */\n  maxConcurrency?: number\n}\n\nfunction createDefaultExecutionContext(): TaskExecutionContext {\n  const cache: TaskCacheRuntime = {\n    namespace(name) {\n      return {\n        file(options) {\n          const key = options.key.join('/')\n          throw new Error(`Task cache runtime is not configured. Requested namespace \"${name}\" and key \"${key}\".`)\n        },\n      }\n    },\n  }\n\n  return {\n    cache,\n    models: [],\n  }\n}\n\n/**\n * Error thrown when a scheduled run fails before producing a normalized result.\n */\nexport class RunnerExecutionError extends Error {\n  /**\n   * Stable task id that failed.\n   */\n  taskId: string\n\n  constructor(taskId: string, cause: unknown) {\n    const message = errorMessageFrom(cause) ?? 'Unknown runner execution failure.'\n    super(`Runner task \"${taskId}\" failed: ${message}`)\n    this.name = 'RunnerExecutionError'\n    this.taskId = taskId\n    this.cause = cause\n  }\n}\n\nfunction createRunnerExecutionError(taskId: string, cause: unknown): RunnerExecutionError {\n  if (cause instanceof RunnerExecutionError && cause.taskId === taskId) {\n    return cause\n  }\n\n  return new RunnerExecutionError(taskId, cause)\n}\n\n/**\n * Executes runner tasks sequentially and aggregates the normalized results.\n *\n * Call stack:\n *\n * {@link createRunnerSchedule}\n *   -> {@link runScheduledTasks}\n *     -> `executor(task)`\n *       -> {@link aggregateRunResults}\n *\n * Use when:\n * - the caller already expanded the runner matrix\n * - task execution should stay deterministic and easy to debug\n *\n * Expects:\n * - `executor` to return normalized `0..1` scores\n * - callers to handle concurrency outside this helper when needed\n * - `onTaskStart` / `onTaskEnd` hooks to be synchronous lifecycle observers\n *\n * Throws:\n * - `RunnerExecutionError` when task setup, hooks, or the executor throws\n */\nexport async function runScheduledTasks(\n  tasks: readonly ScheduledTask[],\n  executor: ScheduledTaskExecutor,\n  options: RunScheduledTasksOptions = {},\n): Promise<AggregatedRunResults> {\n  if (tasks.length === 0) {\n    return aggregateRunResults([])\n  }\n\n  async function executeScheduledTask(task: ScheduledTask): Promise<RunResult> {\n    let executionContext: TaskExecutionContext\n\n    try {\n      executionContext = options.createExecutionContext?.(task) ?? createDefaultExecutionContext()\n    }\n    catch (error) {\n      throw createRunnerExecutionError(task.id, error)\n    }\n\n    try {\n      options.onTaskStart?.(task)\n    }\n    catch (error) {\n      throw createRunnerExecutionError(task.id, error)\n    }\n\n    let runResult: RunResult\n    try {\n      runResult = await executor(task, executionContext)\n    }\n    catch (error) {\n      try {\n        options.onTaskEnd?.(task, 'failed')\n      }\n      catch {\n        // Failed-task observers must not mask the task execution failure.\n      }\n      throw createRunnerExecutionError(task.id, error)\n    }\n\n    try {\n      options.onTaskEnd?.(task, 'passed')\n    }\n    catch (error) {\n      throw createRunnerExecutionError(task.id, error)\n    }\n\n    return runResult\n  }\n\n  const maxConcurrency = options.maxConcurrency ?? 1\n  if (maxConcurrency <= 1) {\n    const results: RunResult[] = []\n    for (const task of tasks) {\n      results.push(await executeScheduledTask(task))\n    }\n    return aggregateRunResults(results)\n  }\n\n  const runWithLimit = limitConcurrency(maxConcurrency)\n  const resultPairs = await Promise.all(tasks.map(async (task, index) => {\n    const result = await runWithLimit(async () => executeScheduledTask(task))\n    return { index, result }\n  }))\n\n  const sortedResults = resultPairs\n    .sort((left, right) => left.index - right.index)\n    .map(item => item.result)\n\n  return aggregateRunResults(sortedResults)\n}\n","import { createRequire } from 'node:module'\nimport { dirname } from 'node:path'\nimport { fileURLToPath } from 'node:url'\n\nconst require = createRequire(import.meta.url)\n\n/**\n * Shared runtime context used by the vieval runner.\n *\n * Use when:\n * - runner services need stable path resolution without module-level side effects\n * - call sites want deterministic control over workspace root detection\n */\nexport interface RunnerRuntimeContext {\n  /**\n   * Absolute project root directory used for path normalization.\n   */\n  projectRootDirectory: string\n}\n\n/**\n * Options used to construct the runner runtime context.\n */\nexport interface CreateVievalRunnerRuntimeContextOptions {\n  /**\n   * Directory used to search for the nearest pnpm workspace.\n   *\n   * @default directory of this module file\n   */\n  cwd?: string\n  /**\n   * Absolute fallback directory when a pnpm workspace root is not found.\n   *\n   * @default package root directory (`packages/vieval`)\n   */\n  fallbackProjectRootDirectory?: string\n}\n\n/**\n * Creates a side-effect-free runtime context for runner path normalization.\n *\n * Call stack:\n *\n * {@link createRunnerRuntimeContext}\n *   -> `findWorkspaceDir(cwd)`\n *     -> `resolve projectRootDirectory`\n *       -> `{ projectRootDirectory }`\n *\n * Use when:\n * - initializing runner infrastructure before collecting eval modules\n * - tests need deterministic root resolution behavior\n */\nexport async function createRunnerRuntimeContext(\n  options: CreateVievalRunnerRuntimeContextOptions = {},\n): Promise<RunnerRuntimeContext> {\n  const cwd = options.cwd ?? dirname(fileURLToPath(import.meta.url))\n  const fallbackProjectRootDirectory = options.fallbackProjectRootDirectory\n    ?? fileURLToPath(new URL('../../../', import.meta.url))\n\n  // NOTICE:\n  // We use dynamic `require` here because `@pnpm/find-workspace-dir` is CommonJS.\n  // Keeping this load inside the factory avoids module-level initialization side effects.\n  const { findWorkspaceDir } = require('@pnpm/find-workspace-dir') as {\n    findWorkspaceDir: (currentWorkingDirectory: string) => Promise<string | undefined>\n  }\n\n  // NOTICE:\n  // Workspace discovery is required to keep collected eval ids stable when this\n  // package is moved inside different monorepo layouts.\n  const workspaceDirectory = await findWorkspaceDir(cwd)\n\n  return {\n    projectRootDirectory: workspaceDirectory ?? fallbackProjectRootDirectory,\n  }\n}\n","import type { CollectedEvalEntry, MatrixDefinition, MatrixLayer, MatrixValue } from '../../config'\n\n/**\n * Describes the inferenceExecutor target for a scheduled eval run.\n */\nexport interface InferenceExecutor {\n  /**\n   * Stable inferenceExecutor identifier such as `openai:gpt-4.1-mini`.\n   */\n  id: string\n}\n\n/**\n * Stores the selected value for each matrix axis.\n */\nexport type RunnerMatrixSelection = Record<string, string>\n\n/**\n * Stores stable row ids for one resolved scheduled task matrix.\n */\nexport interface ScheduledTaskMatrixMeta {\n  /**\n   * Stable row id for the resolved run matrix selection.\n   */\n  runRowId: string\n  /**\n   * Stable row id for the resolved eval matrix selection.\n   */\n  evalRowId: string\n}\n\n/**\n * Stores the structured matrix payload for one scheduled task.\n */\nexport interface ScheduledTaskMatrix {\n  /**\n   * Runtime matrix selection visible to task code.\n   */\n  run: RunnerMatrixSelection\n  /**\n   * Eval-time matrix selection visible to task code.\n   */\n  eval: RunnerMatrixSelection\n  /**\n   * Stable row ids for both scopes.\n   */\n  meta: ScheduledTaskMatrixMeta\n}\n\n/**\n * Maps matrix axis names to the values that should be expanded.\n */\nexport type RunnerMatrixDefinition = MatrixDefinition\n\n/**\n * Accepts either flat axis definitions or one layered matrix object.\n */\nexport type RunnerMatrixInput = RunnerMatrixDefinition | MatrixLayer\n\nconst matrixLayerKeys = new Set(['disable', 'extend', 'override'])\nconst ambiguousMatrixDefinitionErrorMessage = 'Ambiguous matrix definition: cannot mix reserved layer keys (disable, extend, override) with matrix axis keys.'\n\n/**\n * Represents one fully expanded runner task.\n */\nexport interface ScheduledTask {\n  /**\n   * Stable task id derived from the entry, inferenceExecutor, and matrix selection.\n   */\n  id: string\n  /**\n   * The collected eval entry to execute.\n   */\n  entry: CollectedEvalEntry\n  /**\n   * The inferenceExecutor selected for this task.\n   */\n  inferenceExecutor: InferenceExecutor\n  /**\n   * The concrete scoped matrix selection for this task.\n   */\n  matrix: ScheduledTaskMatrix\n}\n\n/**\n * Configures how the runner should expand its execution matrix.\n */\nexport interface CreateRunnerScheduleOptions {\n  /**\n   * Collected eval entries that should be scheduled.\n   */\n  entries: readonly CollectedEvalEntry[]\n  /**\n   * Providers that should run each entry.\n   */\n  inferenceExecutors: readonly InferenceExecutor[]\n  /**\n   * Optional run-time matrix axes expanded as a cartesian product.\n   */\n  runMatrix?: RunnerMatrixInput\n  /**\n   * Optional eval-time matrix axes expanded as a cartesian product.\n   */\n  evalMatrix?: RunnerMatrixInput\n}\n\nfunction encodeTaskIdSegment(value: string): string {\n  return encodeURIComponent(value)\n}\n\nfunction stringifyMatrixValue(value: MatrixValue): string {\n  return String(value)\n}\n\nfunction cloneMatrixSelection(matrix: RunnerMatrixSelection): RunnerMatrixSelection {\n  return { ...matrix }\n}\n\nfunction createScheduledTaskMatrix(\n  runMatrix: RunnerMatrixSelection,\n  evalMatrix: RunnerMatrixSelection,\n): ScheduledTaskMatrix {\n  return {\n    eval: cloneMatrixSelection(evalMatrix),\n    meta: {\n      evalRowId: createStableRowId(evalMatrix),\n      runRowId: createStableRowId(runMatrix),\n    },\n    run: cloneMatrixSelection(runMatrix),\n  }\n}\n\nfunction isMatrixLayer(matrix: RunnerMatrixInput): matrix is MatrixLayer {\n  const matrixKeys = Object.keys(matrix)\n  return (\n    matrixKeys.length > 0\n    && matrixKeys.every(key => matrixLayerKeys.has(key))\n  )\n}\n\nfunction assertNonAmbiguousMatrixDefinition(matrix: RunnerMatrixInput): void {\n  const matrixKeys = Object.keys(matrix)\n  const hasReservedKeys = matrixKeys.some(key => matrixLayerKeys.has(key))\n  const hasAxisKeys = matrixKeys.some(key => !matrixLayerKeys.has(key))\n\n  if (hasReservedKeys && hasAxisKeys) {\n    throw new TypeError(ambiguousMatrixDefinitionErrorMessage)\n  }\n}\n\nfunction normalizeLayerInputToAxes(matrix: RunnerMatrixInput | undefined): MatrixLayer | undefined {\n  if (matrix == null) {\n    return undefined\n  }\n\n  assertNonAmbiguousMatrixDefinition(matrix)\n\n  if (isMatrixLayer(matrix)) {\n    return matrix\n  }\n\n  return {\n    extend: matrix,\n  }\n}\n\nfunction dedupeAxisValues(values: readonly MatrixValue[]): string[] {\n  return Array.from(new Set(values.map(stringifyMatrixValue)))\n}\n\nfunction applyAxisValues(\n  axes: Map<string, string[]>,\n  definition: RunnerMatrixDefinition | undefined,\n  mode: 'extend' | 'override',\n): void {\n  if (definition == null) {\n    return\n  }\n\n  for (const [axis, values] of Object.entries(definition)) {\n    const nextValues = dedupeAxisValues(values)\n\n    if (mode === 'extend') {\n      const existingValues = axes.get(axis) ?? []\n      axes.set(axis, Array.from(new Set([...existingValues, ...nextValues])))\n      continue\n    }\n\n    axes.set(axis, nextValues)\n  }\n}\n\nfunction applyLayer(\n  baseAxes: ReadonlyMap<string, string[]>,\n  layer: MatrixLayer | undefined,\n): Map<string, string[]> {\n  const nextAxes = new Map<string, string[]>(\n    Array.from(baseAxes.entries()).map(([axis, values]) => [axis, [...values]]),\n  )\n\n  for (const axis of layer?.disable ?? []) {\n    nextAxes.delete(axis)\n  }\n\n  applyAxisValues(nextAxes, layer?.extend, 'extend')\n  applyAxisValues(nextAxes, layer?.override, 'override')\n\n  return nextAxes\n}\n\nfunction expandAxesToRows(axes: ReadonlyMap<string, readonly string[]>): RunnerMatrixSelection[] {\n  if (axes.size === 0) {\n    return [{}]\n  }\n\n  const dimensions = Array.from(axes.entries())\n\n  let selections: RunnerMatrixSelection[] = [{}]\n\n  for (const [axis, values] of dimensions) {\n    if (values.length === 0) {\n      return []\n    }\n\n    const nextSelections: RunnerMatrixSelection[] = []\n\n    for (const selection of selections) {\n      for (const value of values) {\n        nextSelections.push({\n          ...selection,\n          [axis]: value,\n        })\n      }\n    }\n\n    selections = nextSelections\n  }\n\n  return selections\n}\n\nfunction createStableRowId(matrix: RunnerMatrixSelection): string {\n  const segments = Object.entries(matrix)\n    .sort(([leftAxis], [rightAxis]) => leftAxis.localeCompare(rightAxis))\n    .map(([axis, value]) => `${encodeTaskIdSegment(axis)}=${encodeTaskIdSegment(value)}`)\n\n  if (segments.length === 0) {\n    return 'default'\n  }\n\n  return segments.join('&')\n}\n\nfunction createTaskId(entryId: string, inferenceExecutorId: string, runRowId: string, evalRowId: string): string {\n  const encodedEntryId = encodeTaskIdSegment(entryId)\n  const encodedProviderId = encodeTaskIdSegment(inferenceExecutorId)\n\n  return [\n    encodedEntryId,\n    encodedProviderId,\n    `run=${encodeTaskIdSegment(runRowId)}`,\n    `eval=${encodeTaskIdSegment(evalRowId)}`,\n  ].join('::')\n}\n\nfunction createResolvedRunAxes(\n  entry: CollectedEvalEntry,\n  runMatrix: RunnerMatrixInput | undefined,\n): Map<string, string[]> {\n  let resolvedAxes = new Map<string, string[]>()\n\n  for (const layerInput of [\n    runMatrix,\n    entry.matrix?.runMatrix,\n    entry.task?.matrix?.runMatrix,\n  ]) {\n    resolvedAxes = applyLayer(resolvedAxes, normalizeLayerInputToAxes(layerInput))\n  }\n\n  return resolvedAxes\n}\n\nfunction createResolvedEvalAxes(\n  entry: CollectedEvalEntry,\n  evalMatrix: RunnerMatrixInput | undefined,\n): Map<string, string[]> {\n  let resolvedAxes = new Map<string, string[]>()\n\n  for (const layerInput of [\n    evalMatrix,\n    entry.matrix?.evalMatrix,\n    entry.task?.matrix?.evalMatrix,\n  ]) {\n    resolvedAxes = applyLayer(resolvedAxes, normalizeLayerInputToAxes(layerInput))\n  }\n\n  return resolvedAxes\n}\n\n/**\n * Expands collected entries into a stable runner schedule.\n *\n * Call stack:\n *\n * {@link collectEvalEntries} (`../runner`)\n *   -> {@link createRunnerSchedule}\n *     -> {@link expandAxesToRows}\n *       -> {@link ScheduledTask}[]\n *\n * Use when:\n * - the runner already knows which eval entries are available\n * - each entry must run against multiple inferenceExecutors or matrix variants\n *\n * Expects:\n * - `entries` and `inferenceExecutors` to be provided in the desired execution order\n * - matrix axes to use insertion order when generating combinations\n */\nexport function createRunnerSchedule(options: CreateRunnerScheduleOptions): ScheduledTask[] {\n  if (options.entries.length === 0) {\n    return []\n  }\n\n  if (options.inferenceExecutors.length === 0) {\n    return []\n  }\n\n  const tasks: ScheduledTask[] = []\n\n  for (const entry of options.entries) {\n    const runSelections = expandAxesToRows(createResolvedRunAxes(entry, options.runMatrix))\n    const evalSelections = expandAxesToRows(createResolvedEvalAxes(entry, options.evalMatrix))\n\n    if (runSelections.length === 0 || evalSelections.length === 0) {\n      continue\n    }\n\n    for (const inferenceExecutor of options.inferenceExecutors) {\n      for (const runMatrix of runSelections) {\n        for (const evalMatrix of evalSelections) {\n          const isolatedMatrix = createScheduledTaskMatrix(runMatrix, evalMatrix)\n\n          tasks.push({\n            entry,\n            id: createTaskId(\n              entry.id,\n              inferenceExecutor.id,\n              isolatedMatrix.meta.runRowId,\n              isolatedMatrix.meta.evalRowId,\n            ),\n            matrix: isolatedMatrix,\n            inferenceExecutor,\n          })\n        }\n      }\n    }\n  }\n\n  return tasks\n}\n","import type { ModelDefinition } from '../../config/models'\nimport type { TaskCacheRuntime } from '../cache'\nimport type { ScheduledTask } from './schedule'\n\n/**\n * Task-scoped execution context exposed to runner executors.\n */\nexport interface TaskExecutionContext {\n  /**\n   * Deterministic cache runtime scoped to the current task project.\n   */\n  cache: TaskCacheRuntime\n  /**\n   * Configured model registrations available to model plugins.\n   */\n  models: readonly ModelDefinition[]\n}\n\n/**\n * Inputs used to build task execution context.\n */\nexport interface CreateTaskExecutionContextOptions {\n  cache?: TaskCacheRuntime\n  models: readonly ModelDefinition[]\n  task: ScheduledTask\n}\n\nfunction createNoopTaskCacheRuntime(): TaskCacheRuntime {\n  return {\n    namespace(name) {\n      return {\n        file(options) {\n          const key = options.key.join('/')\n          throw new Error(`Task cache runtime is not configured. Requested namespace \"${name}\" and key \"${key}\".`)\n        },\n      }\n    },\n  }\n}\n\n/**\n * Creates task-scoped context data for runner execution.\n *\n * Call stack:\n *\n * {@link runScheduledTasks}\n *   -> {@link createTaskExecutionContext}\n *     -> `TaskExecutionContext`\n */\nexport function createTaskExecutionContext(options: CreateTaskExecutionContextOptions): TaskExecutionContext {\n  return {\n    cache: options.cache ?? createNoopTaskCacheRuntime(),\n    models: options.models,\n  }\n}\n"],"mappings":";;;;;;;;;;;AA2BA,SAAS,oBAAoB,OAAuB;CAClD,MAAM,aAAa,MAAM,KAAK;CAC9B,IAAI,WAAW,WAAW,GACxB,OAAO;CAGT,OAAO,WAAW,QAAQ,aAAa,GAAG;AAC5C;AAEA,SAAS,mBAAmB,WAA+B,WAAmD;CAC5G,IAAI,aAAa,QAAQ,UAAU,SAAS,GAC1C,OAAO,UAAU,WAAW,GAAG,IAAI,UAAU,MAAM,CAAC,IAAI;CAG1D,IAAI,aAAa,QAAQ,UAAU,WAAW,GAC5C;CAGF,IAAI,cAAc,oBAChB,OAAO;CAGT,IAAI,cAAc,cAChB,OAAO;CAGT,IAAI,cAAc,aAChB,OAAO;AAIX;;;;;;;;;;AAWA,SAAgB,+BAA+B,SAAqC;CAClF,MAAM,eAAe,QAAQ,IAAI,KAAI,YAAW,oBAAoB,OAAO,CAAC;CAC5E,MAAM,YAAY,mBAAmB,QAAQ,KAAK,QAAQ,SAAS;CAEnE,IAAI,aAAa,WAAW,GAC1B,OAAO,aAAa,OAAO,CAAC,UAAU,IAAI,CAAC,YAAY,WAAW;CAGpE,IAAI,aAAa,MACf,OAAO;CAGT,MAAM,cAAc,aAAa,MAAM,GAAG,KAAK,IAAI,GAAG,aAAa,SAAS,CAAC,CAAC;CAC9E,MAAM,OAAO,aAAa,aAAa,SAAS,MAAM;CACtD,OAAO,CAAC,GAAG,aAAa,GAAG,KAAK,GAAG,WAAW;AAChD;AAEA,eAAe,gBAAgB,MAAc,SAAyC;CACpF,MAAM,YAAY,QAAQ,IAAI;CAC9B,MAAM,gBAAgB,GAAG,KAAK,OAAO,QAAQ,IAAI,GAAG,KAAK,IAAI,EAAE,GAAG,KAAK,OAAO,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,MAAM,GAAG,EAAE;CACxG,MAAM,MAAM,WAAW,EAAE,WAAW,KAAK,CAAC;CAC1C,MAAM,UAAU,eAAe,OAAO;CACtC,MAAM,OAAO,eAAe,IAAI;AAClC;AAEA,SAAS,sBAAsB,MAA+B;CAC5D,OAAO;EACL;EACA,MAAM,SAAS;GACb,IAAI;IACF,MAAM,OAAO,IAAI;IACjB,OAAO;GACT,QACM;IACJ,OAAO;GACT;EACF;EACA,iBAAiB;GACf,OAAO,iBAAiB,IAAI;EAC9B;EACA,MAAM,kBAAkB;GACtB,MAAM,MAAM,QAAQ,IAAI,GAAG,EAAE,WAAW,KAAK,CAAC;GAC9C,OAAO,kBAAkB,IAAI;EAC/B;EACA,MAAM,aAAa;GACjB,OAAO,MAAM,SAAS,IAAI;EAC5B;EACA,MAAM,YAAY,OAAO;GACvB,MAAM,gBAAgB,MAAM,KAAK;EACnC;EACA,MAAM,SAAS,WAAW,SAAS;GACjC,OAAO,MAAM,SAAS,MAAM,QAAQ;EACtC;EACA,MAAM,UAAU,OAAO,WAAW,SAAS;GACzC,MAAM,gBAAgB,MAAM,OAAO,KAAK,OAAO,QAAQ,CAAC;EAC1D;EACA,MAAM,WAAc;GAClB,OAAO,KAAK,MAAM,MAAM,SAAS,MAAM,OAAO,CAAC;EACjD;EACA,MAAM,UAAU,OAAO;GACrB,MAAM,gBAAgB,MAAM,GAAG,KAAK,UAAU,OAAO,MAAM,CAAC,EAAE,GAAG;EACnE;EACA,MAAM,mBAAsB;GAC1B,OAAO,MAAM,KAAK,SAAc;EAClC;EACA,MAAM,sBAAyB;GAC7B,OAAO,MAAM,KAAK,SAAY;EAChC;CACF;AACF;AAEA,SAAS,qBAAqB,eAAuB,WAAmC;CACtF,OAAO,EACL,KAAK,SAAS;EACZ,MAAM,uBAAuB,+BAA+B,OAAO;EACnE,OAAO,sBAAsB,KAAK,eAAe,oBAAoB,SAAS,GAAG,GAAG,oBAAoB,CAAC;CAC3G,EACF;AACF;;;;;;;;;;;;;;;;AAiBA,SAAgB,iCACd,SACkB;CAClB,MAAM,qBAAqB,oBAAoB,QAAQ,WAAW;CAClE,MAAM,mBAAmB,oBAAoB,QAAQ,WAAW;CAChE,MAAM,gBAAgB,KAAK,QAAQ,oBAAoB,oBAAoB,gBAAgB;CAE3F,OAAO,EACL,UAAU,MAAM;EACd,OAAO,qBAAqB,eAAe,IAAI;CACjD,EACF;AACF;;;ACxCA,SAAS,yBAAyB,QAAkD;CAClF,OAAO;EACL,MAAM,EACJ,GAAG,OAAO,KACZ;EACA,MAAM,EACJ,GAAG,OAAO,KACZ;EACA,KAAK,EACH,GAAG,OAAO,IACZ;CACF;AACF;AAEA,SAAS,qBAAqB,MAA4B;CACxD,IAAI,SAAS,WAAW,SAAS,SAC/B,OAAO;CAGT,MAAM,IAAI,UAAU,4BAA4B,KAAK,GAAG;AAC1D;AAEA,SAAS,QAAQ,QAA0C;CACzD,IAAI,OAAO,WAAW,GACpB,OAAO;CAIT,OADc,OAAO,QAAQ,KAAK,UAAU,MAAM,OAAO,CAC9C,IAAI,OAAO;AACxB;AAEA,SAAS,oBAAoB,cAA6B,cAA4C;CACpG,IAAI,gBAAgB,QAAQ,gBAAgB,MAC1C,QAAQ,eAAe,gBAAgB;CAGzC,IAAI,gBAAgB,MAClB,OAAO;CAGT,IAAI,gBAAgB,MAClB,OAAO;CAGT,OAAO;AACT;AAEA,SAAS,oBAAoB,QAA2C;CACtE,MAAM,UAAwB;EAC5B,OAAO,CAAC;EACR,OAAO,CAAC;CACV;CAEA,KAAK,MAAM,SAAS,QAAQ;EAG1B,IAFa,qBAAqB,MAAM,IAEjC,MAAM,SAAS;GACpB,QAAQ,MAAM,KAAK,MAAM,KAAK;GAC9B;EACF;EAEA,QAAQ,MAAM,KAAK,MAAM,KAAK;CAChC;CAEA,OAAO;AACT;AAEA,SAAS,iBAAiB,QAAyC;CACjE,MAAM,UAAU,oBAAoB,OAAO,MAAM;CACjD,MAAM,eAAe,QAAQ,QAAQ,KAAK;CAC1C,MAAM,eAAe,QAAQ,QAAQ,KAAK;CAE1C,OAAO;EACL,SAAS,OAAO;EAChB;EACA,eAAe,oBAAoB,cAAc,YAAY;EAC7D,IAAI,OAAO;EACX;EACA,QAAQ,yBAAyB,OAAO,MAAM;EAC9C,qBAAqB,OAAO;CAC9B;AACF;AAEA,SAAS,sBAAsB,qBAA6B,SAA0D;CACpH,MAAM,cAAwB,CAAC;CAC/B,MAAM,cAAwB,CAAC;CAE/B,KAAK,MAAM,UAAU,SAAS;EAC5B,MAAM,UAAU,oBAAoB,OAAO,MAAM;EACjD,YAAY,KAAK,GAAG,QAAQ,KAAK;EACjC,YAAY,KAAK,GAAG,QAAQ,KAAK;CACnC;CAEA,MAAM,eAAe,QAAQ,WAAW;CACxC,MAAM,eAAe,QAAQ,WAAW;CAExC,OAAO;EACL;EACA,eAAe,oBAAoB,cAAc,YAAY;EAC7D;EACA;EACA,UAAU,QAAQ;CACpB;AACF;;;;;;;;;;;;;;;;;;;;AAqBA,SAAgB,oBAAoB,SAAqD;CACvF,MAAM,OAAO,QAAQ,IAAI,gBAAgB;CAGzC,MAAM,qBADuB,MAAM,KAAK,IAAI,IAAI,QAAQ,KAAI,WAAU,OAAO,mBAAmB,CAAC,CACnD,CAAC,CAC5C,KAAK,wBAAwB;EAE5B,OAAO,sBAAsB,qBADL,QAAQ,QAAO,WAAU,OAAO,wBAAwB,mBAChB,CAAC;CACnE,CAAC,CAAC,CACD,MAAM,MAAM,UAAU,KAAK,oBAAoB,cAAc,MAAM,mBAAmB,CAAC;CAE1F,MAAM,UAAU,sBACd,WACA,OACF;CAEA,OAAO;EACL,SAAS;GACP,cAAc,QAAQ;GACtB,eAAe,QAAQ;GACvB,cAAc,QAAQ;GACtB,UAAU,QAAQ;EACpB;EACA;EACA;CACF;AACF;;;ACxRA,MAAM,iBAAiB;AACvB,MAAM,sBAAsB;AAE5B,SAAS,cAAc,OAAuB;CAC5C,OAAO,MAAM,WAAW,MAAM,GAAG;AACnC;;;;;;;;;;AAWA,SAAgB,sBAAsB,UAAkB,SAAuC;CAC7F,MAAM,qBAAqB,cAAc,QAAQ;CACjD,MAAM,iCAAiC,cAAc,QAAQ,oBAAoB;CACjF,MAAM,uBAAuB,mBAAmB,MAAM,YAAY,CAAC,GAAG;CACtE,MAAM,0BAA0B,+BAA+B,MAAM,YAAY,CAAC,GAAG;CAErF,IAAI,wBAAwB,QAAQ,2BAA2B,MAC7D,OAAO;CAGT,IACE,wBAAwB,QACrB,2BAA2B,QAC3B,qBAAqB,YAAY,MAAM,wBAAwB,YAAY,GAE9E,OAAO;CAGT,MAAM,uBAAuB,QAAQ;CACrC,MAAM,mBAAmB,cAAc,SAAS,sBAAsB,QAAQ,CAAC;CAE/E,IAAI,CAAC,oBAAoB,KAAK,gBAAgB,GAAG;EAC/C,IAAI,qBAAqB,MACvB,OAAO,cAAc,QAAQ;EAG/B,IAAI,CAAC,iBAAiB,WAAW,KAAK,GACpC,OAAO;CAEX;CAEA,OAAO,cAAc,QAAQ;AAC/B;AAEA,SAAS,sBAAsB,YAAmC;CAChE,IAAI,CAAC,WAAW,WAAW,OAAO,GAChC,OAAO;CAGT,IAAI;EACF,OAAO,cAAc,UAAU;CACjC,QACM;EACJ,OAAO;CACT;AACF;AAEA,SAAS,yBACP,YACA,kBACA,SAC2B;CAC3B,MAAM,WAAW,sBAAsB,UAAU;CAEjD,IAAI,CAAC,UACH,OAAO;CAGT,MAAM,mBAAmB,sBAAsB,UAAU,OAAO;CAEhE,IAAI,CAAC,iBAAiB,SAAS,cAAc,GAC3C,OAAO;CAGT,MAAM,YAAY,SAAS,kBAAkB,cAAc;CAE3D,IAAI,UAAU,WAAW,GACvB,OAAO;CAGT,MAAM,oBAAoB,QAAQ,gBAAgB;CAClD,MAAM,YAAY,sBAAsB,MAAM,KAAK;CAEnD,OAAO;EACL,GAAG,iBAAiB;EACpB;EACA;EACA,IAAI,UAAU,WAAW,IAAI,YAAY,GAAG,UAAU,GAAG;EACzD,MAAM;CACR;AACF;;;;;;;;;;;;;;;AAgBA,SAAgB,mBACd,SACA,SACsB;CACtB,OAAO,OAAO,QAAQ,OAAO,CAAC,CAC3B,SAAS,CAAC,YAAY,sBAAsB;EAC3C,MAAM,QAAQ,yBAAyB,YAAY,kBAAkB,OAAO;EAE5E,IAAI,CAAC,OACH,OAAO,CAAC;EAGV,OAAO,CAAC,KAAK;CACf,CAAC,CAAC,CACD,MAAM,MAAM,UAAU,KAAK,GAAG,cAAc,MAAM,EAAE,CAAC;AAC1D;;;AC/CA,SAAS,gCAAsD;CAY7D,OAAO;EACL,OAAA,EAXA,UAAU,MAAM;GACd,OAAO,EACL,KAAK,SAAS;IACZ,MAAM,MAAM,QAAQ,IAAI,KAAK,GAAG;IAChC,MAAM,IAAI,MAAM,8DAA8D,KAAK,aAAa,IAAI,GAAG;GACzG,EACF;EACF,EAII;EACJ,QAAQ,CAAC;CACX;AACF;;;;AAKA,IAAa,uBAAb,cAA0C,MAAM;;;;CAI9C;CAEA,YAAY,QAAgB,OAAgB;EAC1C,MAAM,UAAU,iBAAiB,KAAK,KAAK;EAC3C,MAAM,gBAAgB,OAAO,YAAY,SAAS;EAClD,KAAK,OAAO;EACZ,KAAK,SAAS;EACd,KAAK,QAAQ;CACf;AACF;AAEA,SAAS,2BAA2B,QAAgB,OAAsC;CACxF,IAAI,iBAAiB,wBAAwB,MAAM,WAAW,QAC5D,OAAO;CAGT,OAAO,IAAI,qBAAqB,QAAQ,KAAK;AAC/C;;;;;;;;;;;;;;;;;;;;;;;AAwBA,eAAsB,kBACpB,OACA,UACA,UAAoC,CAAC,GACN;CAC/B,IAAI,MAAM,WAAW,GACnB,OAAO,oBAAoB,CAAC,CAAC;CAG/B,eAAe,qBAAqB,MAAyC;EAC3E,IAAI;EAEJ,IAAI;GACF,mBAAmB,QAAQ,yBAAyB,IAAI,KAAK,8BAA8B;EAC7F,SACO,OAAO;GACZ,MAAM,2BAA2B,KAAK,IAAI,KAAK;EACjD;EAEA,IAAI;GACF,QAAQ,cAAc,IAAI;EAC5B,SACO,OAAO;GACZ,MAAM,2BAA2B,KAAK,IAAI,KAAK;EACjD;EAEA,IAAI;EACJ,IAAI;GACF,YAAY,MAAM,SAAS,MAAM,gBAAgB;EACnD,SACO,OAAO;GACZ,IAAI;IACF,QAAQ,YAAY,MAAM,QAAQ;GACpC,QACM,CAEN;GACA,MAAM,2BAA2B,KAAK,IAAI,KAAK;EACjD;EAEA,IAAI;GACF,QAAQ,YAAY,MAAM,QAAQ;EACpC,SACO,OAAO;GACZ,MAAM,2BAA2B,KAAK,IAAI,KAAK;EACjD;EAEA,OAAO;CACT;CAEA,MAAM,iBAAiB,QAAQ,kBAAkB;CACjD,IAAI,kBAAkB,GAAG;EACvB,MAAM,UAAuB,CAAC;EAC9B,KAAK,MAAM,QAAQ,OACjB,QAAQ,KAAK,MAAM,qBAAqB,IAAI,CAAC;EAE/C,OAAO,oBAAoB,OAAO;CACpC;CAEA,MAAM,eAAe,iBAAiB,cAAc;CAUpD,OAAO,qBAJe,MALI,QAAQ,IAAI,MAAM,IAAI,OAAO,MAAM,UAAU;EAErE,OAAO;GAAE;GAAO,QAAA,MADK,aAAa,YAAY,qBAAqB,IAAI,CAAC;EACjD;CACzB,CAAC,CAAC,EAAA,CAGC,MAAM,MAAM,UAAU,KAAK,QAAQ,MAAM,KAAK,CAAC,CAC/C,KAAI,SAAQ,KAAK,MAEmB,CAAC;AAC1C;;;AC1NA,MAAM,UAAU,cAAc,OAAO,KAAK,GAAG;;;;;;;;;;;;;;;AAgD7C,eAAsB,2BACpB,UAAmD,CAAC,GACrB;CAC/B,MAAM,MAAM,QAAQ,OAAO,QAAQ,cAAc,OAAO,KAAK,GAAG,CAAC;CACjE,MAAM,+BAA+B,QAAQ,gCACxC,cAAc,IAAI,IAAI,aAAa,OAAO,KAAK,GAAG,CAAC;CAKxD,MAAM,EAAE,qBAAqB,QAAQ,0BAA0B;CAS/D,OAAO,EACL,sBAAsB,MAHS,iBAAiB,GAAG,KAGP,6BAC9C;AACF;;;ACfA,MAAM,kCAAkB,IAAI,IAAI;CAAC;CAAW;CAAU;AAAU,CAAC;AACjE,MAAM,wCAAwC;AA8C9C,SAAS,oBAAoB,OAAuB;CAClD,OAAO,mBAAmB,KAAK;AACjC;AAEA,SAAS,qBAAqB,OAA4B;CACxD,OAAO,OAAO,KAAK;AACrB;AAEA,SAAS,qBAAqB,QAAsD;CAClF,OAAO,EAAE,GAAG,OAAO;AACrB;AAEA,SAAS,0BACP,WACA,YACqB;CACrB,OAAO;EACL,MAAM,qBAAqB,UAAU;EACrC,MAAM;GACJ,WAAW,kBAAkB,UAAU;GACvC,UAAU,kBAAkB,SAAS;EACvC;EACA,KAAK,qBAAqB,SAAS;CACrC;AACF;AAEA,SAAS,cAAc,QAAkD;CACvE,MAAM,aAAa,OAAO,KAAK,MAAM;CACrC,OACE,WAAW,SAAS,KACjB,WAAW,OAAM,QAAO,gBAAgB,IAAI,GAAG,CAAC;AAEvD;AAEA,SAAS,mCAAmC,QAAiC;CAC3E,MAAM,aAAa,OAAO,KAAK,MAAM;CACrC,MAAM,kBAAkB,WAAW,MAAK,QAAO,gBAAgB,IAAI,GAAG,CAAC;CACvE,MAAM,cAAc,WAAW,MAAK,QAAO,CAAC,gBAAgB,IAAI,GAAG,CAAC;CAEpE,IAAI,mBAAmB,aACrB,MAAM,IAAI,UAAU,qCAAqC;AAE7D;AAEA,SAAS,0BAA0B,QAAgE;CACjG,IAAI,UAAU,MACZ;CAGF,mCAAmC,MAAM;CAEzC,IAAI,cAAc,MAAM,GACtB,OAAO;CAGT,OAAO,EACL,QAAQ,OACV;AACF;AAEA,SAAS,iBAAiB,QAA0C;CAClE,OAAO,MAAM,KAAK,IAAI,IAAI,OAAO,IAAI,oBAAoB,CAAC,CAAC;AAC7D;AAEA,SAAS,gBACP,MACA,YACA,MACM;CACN,IAAI,cAAc,MAChB;CAGF,KAAK,MAAM,CAAC,MAAM,WAAW,OAAO,QAAQ,UAAU,GAAG;EACvD,MAAM,aAAa,iBAAiB,MAAM;EAE1C,IAAI,SAAS,UAAU;GACrB,MAAM,iBAAiB,KAAK,IAAI,IAAI,KAAK,CAAC;GAC1C,KAAK,IAAI,MAAM,MAAM,qBAAK,IAAI,IAAI,CAAC,GAAG,gBAAgB,GAAG,UAAU,CAAC,CAAC,CAAC;GACtE;EACF;EAEA,KAAK,IAAI,MAAM,UAAU;CAC3B;AACF;AAEA,SAAS,WACP,UACA,OACuB;CACvB,MAAM,WAAW,IAAI,IACnB,MAAM,KAAK,SAAS,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,YAAY,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,CAAC,CAC5E;CAEA,KAAK,MAAM,QAAQ,OAAO,WAAW,CAAC,GACpC,SAAS,OAAO,IAAI;CAGtB,gBAAgB,UAAU,OAAO,QAAQ,QAAQ;CACjD,gBAAgB,UAAU,OAAO,UAAU,UAAU;CAErD,OAAO;AACT;AAEA,SAAS,iBAAiB,MAAuE;CAC/F,IAAI,KAAK,SAAS,GAChB,OAAO,CAAC,CAAC,CAAC;CAGZ,MAAM,aAAa,MAAM,KAAK,KAAK,QAAQ,CAAC;CAE5C,IAAI,aAAsC,CAAC,CAAC,CAAC;CAE7C,KAAK,MAAM,CAAC,MAAM,WAAW,YAAY;EACvC,IAAI,OAAO,WAAW,GACpB,OAAO,CAAC;EAGV,MAAM,iBAA0C,CAAC;EAEjD,KAAK,MAAM,aAAa,YACtB,KAAK,MAAM,SAAS,QAClB,eAAe,KAAK;GAClB,GAAG;IACF,OAAO;EACV,CAAC;EAIL,aAAa;CACf;CAEA,OAAO;AACT;AAEA,SAAS,kBAAkB,QAAuC;CAChE,MAAM,WAAW,OAAO,QAAQ,MAAM,CAAC,CACpC,MAAM,CAAC,WAAW,CAAC,eAAe,SAAS,cAAc,SAAS,CAAC,CAAC,CACpE,KAAK,CAAC,MAAM,WAAW,GAAG,oBAAoB,IAAI,EAAE,GAAG,oBAAoB,KAAK,GAAG;CAEtF,IAAI,SAAS,WAAW,GACtB,OAAO;CAGT,OAAO,SAAS,KAAK,GAAG;AAC1B;AAEA,SAAS,aAAa,SAAiB,qBAA6B,UAAkB,WAA2B;CAI/G,OAAO;EAHgB,oBAAoB,OAI5B;EAHW,oBAAoB,mBAI5B;EAChB,OAAO,oBAAoB,QAAQ;EACnC,QAAQ,oBAAoB,SAAS;CACvC,CAAC,CAAC,KAAK,IAAI;AACb;AAEA,SAAS,sBACP,OACA,WACuB;CACvB,IAAI,+BAAe,IAAI,IAAsB;CAE7C,KAAK,MAAM,cAAc;EACvB;EACA,MAAM,QAAQ;EACd,MAAM,MAAM,QAAQ;CACtB,GACE,eAAe,WAAW,cAAc,0BAA0B,UAAU,CAAC;CAG/E,OAAO;AACT;AAEA,SAAS,uBACP,OACA,YACuB;CACvB,IAAI,+BAAe,IAAI,IAAsB;CAE7C,KAAK,MAAM,cAAc;EACvB;EACA,MAAM,QAAQ;EACd,MAAM,MAAM,QAAQ;CACtB,GACE,eAAe,WAAW,cAAc,0BAA0B,UAAU,CAAC;CAG/E,OAAO;AACT;;;;;;;;;;;;;;;;;;;AAoBA,SAAgB,qBAAqB,SAAuD;CAC1F,IAAI,QAAQ,QAAQ,WAAW,GAC7B,OAAO,CAAC;CAGV,IAAI,QAAQ,mBAAmB,WAAW,GACxC,OAAO,CAAC;CAGV,MAAM,QAAyB,CAAC;CAEhC,KAAK,MAAM,SAAS,QAAQ,SAAS;EACnC,MAAM,gBAAgB,iBAAiB,sBAAsB,OAAO,QAAQ,SAAS,CAAC;EACtF,MAAM,iBAAiB,iBAAiB,uBAAuB,OAAO,QAAQ,UAAU,CAAC;EAEzF,IAAI,cAAc,WAAW,KAAK,eAAe,WAAW,GAC1D;EAGF,KAAK,MAAM,qBAAqB,QAAQ,oBACtC,KAAK,MAAM,aAAa,eACtB,KAAK,MAAM,cAAc,gBAAgB;GACvC,MAAM,iBAAiB,0BAA0B,WAAW,UAAU;GAEtE,MAAM,KAAK;IACT;IACA,IAAI,aACF,MAAM,IACN,kBAAkB,IAClB,eAAe,KAAK,UACpB,eAAe,KAAK,SACtB;IACA,QAAQ;IACR;GACF,CAAC;EACH;CAGN;CAEA,OAAO;AACT;;;AC3UA,SAAS,6BAA+C;CACtD,OAAO,EACL,UAAU,MAAM;EACd,OAAO,EACL,KAAK,SAAS;GACZ,MAAM,MAAM,QAAQ,IAAI,KAAK,GAAG;GAChC,MAAM,IAAI,MAAM,8DAA8D,KAAK,aAAa,IAAI,GAAG;EACzG,EACF;CACF,EACF;AACF;;;;;;;;;;AAWA,SAAgB,2BAA2B,SAAkE;CAC3G,OAAO;EACL,OAAO,QAAQ,SAAS,2BAA2B;EACnD,QAAQ,QAAQ;CAClB;AACF"}

package/dist/core/scheduler/index.d.mts

@@ -1,2 +1,2 @@
-import { a as SchedulerMiddleware, c as SchedulerScopeContext, i as SchedulerConcurrencyConfig, n as getActiveScopes, o as SchedulerRuntime, r as CreateSchedulerRuntimeOptions, s as SchedulerScope, t as createSchedulerRuntime } from "../../index-fakXoZEe.mjs";
+import { a as SchedulerMiddleware, c as SchedulerScopeContext, i as SchedulerConcurrencyConfig, n as getActiveScopes, o as SchedulerRuntime, r as CreateSchedulerRuntimeOptions, s as SchedulerScope, t as createSchedulerRuntime } from "../../index-Be5I1ZJL.mjs";
 export { CreateSchedulerRuntimeOptions, SchedulerConcurrencyConfig, SchedulerMiddleware, SchedulerRuntime, SchedulerScope, SchedulerScopeContext, createSchedulerRuntime, getActiveScopes };

package/dist/core/scheduler/index.mjs

@@ -1,4 +1,4 @@
-import { t as createSchedulerQueue } from "../../queue-DsZQkZO_.mjs";
+import { t as createSchedulerQueue } from "../../queue-BL86z2W_.mjs";
 //#region src/core/scheduler/runtime.ts
 const schedulerScopeOrder = [
 	"workspace",
@@ -45,7 +45,7 @@ function createSchedulerRuntime(options = {}) {
 * Resolves the scheduler scopes that apply to a context.
 *
 * Before:
-* - `{ scope: 'case', workspaceId: 'ws', experimentId: 'exp', caseId: 'case-1' }`
+* - `{ scope: 'case', workspaceId: 'ws', projectName: 'project', caseId: 'case-1' }`
 *
 * After:
 * - `['workspace', 'project', 'task', 'attempt', 'case']` up to the requested scope
@@ -86,7 +86,7 @@ function getScopeQueue(scope, context, queues) {
 	return queue;
 }
 function getSchedulerScopeInstanceKey(scope, context) {
-	const workspaceKey = `workspace:${context.workspaceId}:experiment:${context.experimentId}`;
+	const workspaceKey = `workspace:${context.workspaceId}`;
 	const projectKey = `${workspaceKey}:project:${context.projectName ?? "(missing-project)"}`;
 	const taskKey = `${projectKey}:task:${context.taskId ?? "(missing-task)"}`;
 	const attemptKey = `${taskKey}:attempt:${context.attemptIndex ?? "(missing-attempt)"}`;