vieval 0.0.5 → 0.0.6

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 model resolution 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(task: ScheduledTask): 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    model(options) {\n      const requestedModelName = typeof options === 'string' ? options : options?.name\n      if (requestedModelName != null) {\n        throw new Error(`No model registry configured. Requested model: ${requestedModelName}`)\n      }\n\n      throw new Error(`No model registry configured for task inferenceExecutor id \"${task.inferenceExecutor.id}\".`)\n    },\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(task)\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\nimport { resolveModelByName } from '../../config/models'\n\n/**\n * Options for selecting a model from the execution context.\n */\nexport interface TaskModelSelectionOptions {\n  /**\n   * Model id or alias name.\n   */\n  name: string\n}\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   * Resolves model configuration for the current task.\n   *\n   * Use when:\n   * - no arguments are provided to use the model selected by run matrix/inferenceExecutor\n   * - `name` is provided to resolve a specific model id or alias\n   */\n  model: (\n    selection?: string | TaskModelSelectionOptions,\n  ) => 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 \"${key}\".`)\n        },\n      }\n    },\n  }\n}\n\nfunction resolveDefaultTaskModel(\n  models: readonly ModelDefinition[],\n  task: ScheduledTask,\n): ModelDefinition {\n  const runMatrixModelName = task.matrix.run.model\n  if (runMatrixModelName != null) {\n    const matrixSelectedModel = resolveModelByName(models, runMatrixModelName)\n    if (matrixSelectedModel != null) {\n      return matrixSelectedModel\n    }\n\n    throw new Error(`Unknown configured model \"${runMatrixModelName}\" from task.matrix.run.model.`)\n  }\n\n  const matched = resolveModelByName(models, task.inferenceExecutor.id)\n  if (matched != null) {\n    return matched\n  }\n\n  if (models.length > 1) {\n    throw new Error(\n      [\n        `Multiple configured models are available, but no default model is selected for inferenceExecutor \"${task.inferenceExecutor.id}\".`,\n        'Select one model explicitly by either:',\n        '- setting runMatrix.override.model (or task matrix run.model)',\n        '- setting project.inferenceExecutors to a matching model id',\n        '- calling context.model({ name: \"your-model-id-or-alias\" })',\n      ].join('\\n'),\n    )\n  }\n\n  if (models.length === 1) {\n    const firstModel = models[0]\n    if (firstModel != null) {\n      return firstModel\n    }\n  }\n\n  throw new Error(`No configured model found for inferenceExecutor id \"${task.inferenceExecutor.id}\".`)\n}\n\n/**\n * Creates task-scoped model resolver context for runner execution.\n *\n * Call stack:\n *\n * {@link runScheduledTasks}\n *   -> {@link createTaskExecutionContext}\n *     -> {@link resolveModelByName}\n *       -> `task.model()` / `task.model({ name })`\n */\nexport function createTaskExecutionContext(options: CreateTaskExecutionContextOptions): TaskExecutionContext {\n  return {\n    cache: options.cache ?? createNoopTaskCacheRuntime(),\n    model(selection) {\n      if (selection == null) {\n        return resolveDefaultTaskModel(options.models, options.task)\n      }\n\n      const name = typeof selection === 'string' ? selection : selection.name\n\n      const namedModel = resolveModelByName(options.models, name)\n      if (namedModel == null) {\n        throw new Error(`Unknown configured model \"${name}\".`)\n      }\n\n      return namedModel\n    },\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,8BAA8B,MAA2C;AAYhF,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,MAAM,SAAS;GACb,MAAM,qBAAqB,OAAO,YAAY,WAAW,UAAU,SAAS;AAC5E,OAAI,sBAAsB,KACxB,OAAM,IAAI,MAAM,kDAAkD,qBAAqB;AAGzF,SAAM,IAAI,MAAM,+DAA+D,KAAK,kBAAkB,GAAG,IAAI;;EAEhH;;;;;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,8BAA8B,KAAK;WAE3F,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;;;;AChO3C,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;;;;ACxTT,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;;AAGH,SAAS,wBACP,QACA,MACiB;CACjB,MAAM,qBAAqB,KAAK,OAAO,IAAI;AAC3C,KAAI,sBAAsB,MAAM;EAC9B,MAAM,sBAAsB,mBAAmB,QAAQ,mBAAmB;AAC1E,MAAI,uBAAuB,KACzB,QAAO;AAGT,QAAM,IAAI,MAAM,6BAA6B,mBAAmB,+BAA+B;;CAGjG,MAAM,UAAU,mBAAmB,QAAQ,KAAK,kBAAkB,GAAG;AACrE,KAAI,WAAW,KACb,QAAO;AAGT,KAAI,OAAO,SAAS,EAClB,OAAM,IAAI,MACR;EACE,qGAAqG,KAAK,kBAAkB,GAAG;EAC/H;EACA;EACA;EACA;EACD,CAAC,KAAK,KAAK,CACb;AAGH,KAAI,OAAO,WAAW,GAAG;EACvB,MAAM,aAAa,OAAO;AAC1B,MAAI,cAAc,KAChB,QAAO;;AAIX,OAAM,IAAI,MAAM,uDAAuD,KAAK,kBAAkB,GAAG,IAAI;;;;;;;;;;;;AAavG,SAAgB,2BAA2B,SAAkE;AAC3G,QAAO;EACL,OAAO,QAAQ,SAAS,4BAA4B;EACpD,MAAM,WAAW;AACf,OAAI,aAAa,KACf,QAAO,wBAAwB,QAAQ,QAAQ,QAAQ,KAAK;GAG9D,MAAM,OAAO,OAAO,cAAc,WAAW,YAAY,UAAU;GAEnE,MAAM,aAAa,mBAAmB,QAAQ,QAAQ,KAAK;AAC3D,OAAI,cAAc,KAChB,OAAM,IAAI,MAAM,6BAA6B,KAAK,IAAI;AAGxD,UAAO;;EAEV"}
+{"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 model resolution 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(task: ScheduledTask): 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    model(options) {\n      const requestedModelName = typeof options === 'string' ? options : options?.name\n      if (requestedModelName != null) {\n        throw new Error(`No model registry configured. Requested model: ${requestedModelName}`)\n      }\n\n      throw new Error(`No model registry configured for task inferenceExecutor id \"${task.inferenceExecutor.id}\".`)\n    },\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(task)\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\nimport { resolveModelByName } from '../../config/models'\n\n/**\n * Options for selecting a model from the execution context.\n */\nexport interface TaskModelSelectionOptions {\n  /**\n   * Model id or alias name.\n   */\n  name: string\n}\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   * Resolves model configuration for the current task.\n   *\n   * Use when:\n   * - no arguments are provided to use the model selected by run matrix/inferenceExecutor\n   * - `name` is provided to resolve a specific model id or alias\n   */\n  model: (\n    selection?: string | TaskModelSelectionOptions,\n  ) => 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 \"${key}\".`)\n        },\n      }\n    },\n  }\n}\n\nfunction resolveDefaultTaskModel(\n  models: readonly ModelDefinition[],\n  task: ScheduledTask,\n): ModelDefinition {\n  const runMatrixModelName = task.matrix.run.model\n  if (runMatrixModelName != null) {\n    const matrixSelectedModel = resolveModelByName(models, runMatrixModelName)\n    if (matrixSelectedModel != null) {\n      return matrixSelectedModel\n    }\n\n    throw new Error(`Unknown configured model \"${runMatrixModelName}\" from task.matrix.run.model.`)\n  }\n\n  const matched = resolveModelByName(models, task.inferenceExecutor.id)\n  if (matched != null) {\n    return matched\n  }\n\n  if (models.length > 1) {\n    throw new Error(\n      [\n        `Multiple configured models are available, but no default model is selected for inferenceExecutor \"${task.inferenceExecutor.id}\".`,\n        'Select one model explicitly by either:',\n        '- setting runMatrix.override.model (or task matrix run.model)',\n        '- setting project.inferenceExecutors to a matching model id',\n        '- calling context.model({ name: \"your-model-id-or-alias\" })',\n      ].join('\\n'),\n    )\n  }\n\n  if (models.length === 1) {\n    const firstModel = models[0]\n    if (firstModel != null) {\n      return firstModel\n    }\n  }\n\n  throw new Error(`No configured model found for inferenceExecutor id \"${task.inferenceExecutor.id}\".`)\n}\n\n/**\n * Creates task-scoped model resolver context for runner execution.\n *\n * Call stack:\n *\n * {@link runScheduledTasks}\n *   -> {@link createTaskExecutionContext}\n *     -> {@link resolveModelByName}\n *       -> `task.model()` / `task.model({ name })`\n */\nexport function createTaskExecutionContext(options: CreateTaskExecutionContextOptions): TaskExecutionContext {\n  return {\n    cache: options.cache ?? createNoopTaskCacheRuntime(),\n    model(selection) {\n      if (selection == null) {\n        return resolveDefaultTaskModel(options.models, options.task)\n      }\n\n      const name = typeof selection === 'string' ? selection : selection.name\n\n      const namedModel = resolveModelByName(options.models, name)\n      if (namedModel == null) {\n        throw new Error(`Unknown configured model \"${name}\".`)\n      }\n\n      return namedModel\n    },\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,8BAA8B,MAA2C;AAYhF,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,MAAM,SAAS;GACb,MAAM,qBAAqB,OAAO,YAAY,WAAW,UAAU,SAAS;AAC5E,OAAI,sBAAsB,KACxB,OAAM,IAAI,MAAM,kDAAkD,qBAAqB;AAGzF,SAAM,IAAI,MAAM,+DAA+D,KAAK,kBAAkB,GAAG,IAAI;;EAEhH;;;;;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,8BAA8B,KAAK;WAE3F,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;;;;AChO3C,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;;;;ACxTT,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;;AAGH,SAAS,wBACP,QACA,MACiB;CACjB,MAAM,qBAAqB,KAAK,OAAO,IAAI;AAC3C,KAAI,sBAAsB,MAAM;EAC9B,MAAM,sBAAsB,mBAAmB,QAAQ,mBAAmB;AAC1E,MAAI,uBAAuB,KACzB,QAAO;AAGT,QAAM,IAAI,MAAM,6BAA6B,mBAAmB,+BAA+B;;CAGjG,MAAM,UAAU,mBAAmB,QAAQ,KAAK,kBAAkB,GAAG;AACrE,KAAI,WAAW,KACb,QAAO;AAGT,KAAI,OAAO,SAAS,EAClB,OAAM,IAAI,MACR;EACE,qGAAqG,KAAK,kBAAkB,GAAG;EAC/H;EACA;EACA;EACA;EACD,CAAC,KAAK,KAAK,CACb;AAGH,KAAI,OAAO,WAAW,GAAG;EACvB,MAAM,aAAa,OAAO;AAC1B,MAAI,cAAc,KAChB,QAAO;;AAIX,OAAM,IAAI,MAAM,uDAAuD,KAAK,kBAAkB,GAAG,IAAI;;;;;;;;;;;;AAavG,SAAgB,2BAA2B,SAAkE;AAC3G,QAAO;EACL,OAAO,QAAQ,SAAS,4BAA4B;EACpD,MAAM,WAAW;AACf,OAAI,aAAa,KACf,QAAO,wBAAwB,QAAQ,QAAQ,QAAQ,KAAK;GAG9D,MAAM,OAAO,OAAO,cAAc,WAAW,YAAY,UAAU;GAEnE,MAAM,aAAa,mBAAmB,QAAQ,QAAQ,KAAK;AAC3D,OAAI,cAAc,KAChB,OAAM,IAAI,MAAM,6BAA6B,KAAK,IAAI;AAGxD,UAAO;;EAEV"}

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

@@ -0,0 +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";
+export { CreateSchedulerRuntimeOptions, SchedulerConcurrencyConfig, SchedulerMiddleware, SchedulerRuntime, SchedulerScope, SchedulerScopeContext, createSchedulerRuntime, getActiveScopes };

package/dist/core/scheduler/index.mjs

@@ -0,0 +1,188 @@
+import { t as createSchedulerQueue } from "../../queue-DsZQkZO_.mjs";
+//#region src/core/scheduler/runtime.ts
+const schedulerScopeOrder = [
+	"workspace",
+	"project",
+	"task",
+	"attempt",
+	"case"
+];
+/**
+* Creates the core scheduler runtime used to serialize work by scope.
+*
+* Call stack:
+*
+* {@link createSchedulerRuntime}
+*   -> `createRuntimeQueues`
+*     -> `runtime.runCase(context, execute)`
+*       -> `runWithQueues`
+*         -> `runAcquireMiddleware`
+*           -> `execute`
+*         -> `runReleaseMiddleware`
+*
+* Use when:
+* - runner code needs concurrency caps for queued case execution
+* - middleware should wrap work with acquire/release lifecycle hooks
+*
+* Expects:
+* - middleware is ordered from outermost to innermost concern
+* - concurrency caps are positive integers when provided
+*
+* Returns:
+* - a scheduler runtime with case execution support
+*/
+function createSchedulerRuntime(options = {}) {
+	const middleware = options.middleware ?? [];
+	const queues = createRuntimeQueues(options.concurrency ?? {});
+	return { runCase(context, execute) {
+		return runWithQueues(getActiveScopes(context), context, queues, () => {
+			if (middleware.length === 0) return execute();
+			return runWithMiddlewareEnvelope(middleware, context, execute);
+		});
+	} };
+}
+/**
+* Resolves the scheduler scopes that apply to a context.
+*
+* Before:
+* - `{ scope: 'case', workspaceId: 'ws', experimentId: 'exp', caseId: 'case-1' }`
+*
+* After:
+* - `['workspace', 'project', 'task', 'attempt', 'case']` up to the requested scope
+*/
+function getActiveScopes(context) {
+	const targetScopeIndex = schedulerScopeOrder.indexOf(context.scope);
+	if (targetScopeIndex < 0) return [];
+	return schedulerScopeOrder.slice(0, targetScopeIndex + 1);
+}
+function createRuntimeQueues(concurrency) {
+	const queues = /* @__PURE__ */ new Map();
+	for (const scope of schedulerScopeOrder) {
+		const scopeConcurrency = concurrency[scope];
+		if (scopeConcurrency === void 0) continue;
+		validateSchedulerConcurrency(scope, scopeConcurrency);
+		queues.set(scope, {
+			concurrency: scopeConcurrency,
+			instances: /* @__PURE__ */ new Map()
+		});
+	}
+	return queues;
+}
+async function runWithQueues(scopes, context, queues, execute, index = 0) {
+	const scope = scopes[index];
+	if (scope === void 0) return execute();
+	const queue = getScopeQueue(scope, context, queues);
+	if (queue === void 0) return runWithQueues(scopes, context, queues, execute, index + 1);
+	return queue.run(() => runWithQueues(scopes, context, queues, execute, index + 1));
+}
+function getScopeQueue(scope, context, queues) {
+	const queueRegistry = queues.get(scope);
+	if (queueRegistry === void 0) return;
+	const scopeKey = getSchedulerScopeInstanceKey(scope, context);
+	const existingQueue = queueRegistry.instances.get(scopeKey);
+	if (existingQueue !== void 0) return existingQueue;
+	const queue = createSchedulerQueue(queueRegistry.concurrency);
+	queueRegistry.instances.set(scopeKey, queue);
+	return queue;
+}
+function getSchedulerScopeInstanceKey(scope, context) {
+	const workspaceKey = `workspace:${context.workspaceId}:experiment:${context.experimentId}`;
+	const projectKey = `${workspaceKey}:project:${context.projectName ?? "(missing-project)"}`;
+	const taskKey = `${projectKey}:task:${context.taskId ?? "(missing-task)"}`;
+	const attemptKey = `${taskKey}:attempt:${context.attemptIndex ?? "(missing-attempt)"}`;
+	switch (scope) {
+		case "workspace": return workspaceKey;
+		case "project": return projectKey;
+		case "task": return taskKey;
+		case "attempt": return attemptKey;
+		case "case": return attemptKey;
+	}
+}
+async function runWithMiddlewareEnvelope(middleware, context, execute) {
+	const result = await runAcquireMiddleware(middleware, context, execute, 0);
+	try {
+		switch (result.outcome.status) {
+			case "succeeded": return result.outcome.value;
+			case "failed": throw result.outcome.error;
+			case "skipped": throw createSchedulerShortCircuitError();
+		}
+	} finally {
+		await runReleaseMiddleware(result.releaseStack, context, result.releaseStack.length - 1);
+	}
+}
+async function runAcquireMiddleware(middleware, context, execute, index) {
+	const currentMiddleware = middleware[index];
+	if (currentMiddleware === void 0) return createSchedulerExecutionResult([], execute);
+	let nextResult = createSchedulerShortCircuitResult();
+	let didCallNext = false;
+	const next = async () => {
+		didCallNext = true;
+		nextResult = await runAcquireMiddleware(middleware, context, execute, index + 1);
+	};
+	try {
+		if (currentMiddleware.onAcquire === void 0) await next();
+		else await currentMiddleware.onAcquire(context, next);
+	} catch (error) {
+		if (!didCallNext) return createSchedulerFailureResult([], error);
+		return createSchedulerFailureResult([currentMiddleware, ...nextResult.releaseStack], error);
+	}
+	return {
+		releaseStack: [currentMiddleware, ...nextResult.releaseStack],
+		outcome: nextResult.outcome
+	};
+}
+async function runReleaseMiddleware(releaseStack, context, index) {
+	const currentMiddleware = releaseStack[index];
+	if (currentMiddleware === void 0) return;
+	if (currentMiddleware.onRelease === void 0) {
+		await runReleaseMiddleware(releaseStack, context, index - 1);
+		return;
+	}
+	await currentMiddleware.onRelease(context, async () => {
+		await runReleaseMiddleware(releaseStack, context, index - 1);
+	});
+}
+async function createSchedulerExecutionResult(releaseStack, execute) {
+	try {
+		return {
+			releaseStack,
+			outcome: {
+				status: "succeeded",
+				value: await execute()
+			}
+		};
+	} catch (error) {
+		return {
+			releaseStack,
+			outcome: {
+				status: "failed",
+				error
+			}
+		};
+	}
+}
+function createSchedulerFailureResult(releaseStack, error) {
+	return {
+		releaseStack,
+		outcome: {
+			status: "failed",
+			error
+		}
+	};
+}
+function createSchedulerShortCircuitResult() {
+	return {
+		releaseStack: [],
+		outcome: { status: "skipped" }
+	};
+}
+function validateSchedulerConcurrency(scope, concurrency) {
+	if (!Number.isFinite(concurrency) || !Number.isInteger(concurrency) || concurrency <= 0) throw new Error(`Invalid scheduler concurrency for "${scope}": ${String(concurrency)}`);
+}
+function createSchedulerShortCircuitError() {
+	return /* @__PURE__ */ new Error("Scheduler middleware short-circuited execution.");
+}
+//#endregion
+export { createSchedulerRuntime, getActiveScopes };
+
+//# sourceMappingURL=index.mjs.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"index.mjs","names":[],"sources":["../../../src/core/scheduler/runtime.ts"],"sourcesContent":["import type {\n  CreateSchedulerRuntimeOptions,\n  SchedulerConcurrencyConfig,\n  SchedulerMiddleware,\n  SchedulerRuntime,\n  SchedulerScope,\n  SchedulerScopeContext,\n} from './types'\n\nimport { createSchedulerQueue } from './queue'\n\nconst schedulerScopeOrder: SchedulerScope[] = [\n  'workspace',\n  'project',\n  'task',\n  'attempt',\n  'case',\n]\n\n/**\n * Creates the core scheduler runtime used to serialize work by scope.\n *\n * Call stack:\n *\n * {@link createSchedulerRuntime}\n *   -> `createRuntimeQueues`\n *     -> `runtime.runCase(context, execute)`\n *       -> `runWithQueues`\n *         -> `runAcquireMiddleware`\n *           -> `execute`\n *         -> `runReleaseMiddleware`\n *\n * Use when:\n * - runner code needs concurrency caps for queued case execution\n * - middleware should wrap work with acquire/release lifecycle hooks\n *\n * Expects:\n * - middleware is ordered from outermost to innermost concern\n * - concurrency caps are positive integers when provided\n *\n * Returns:\n * - a scheduler runtime with case execution support\n */\nexport function createSchedulerRuntime(\n  options: CreateSchedulerRuntimeOptions = {},\n): SchedulerRuntime {\n  const middleware = options.middleware ?? []\n  const queues = createRuntimeQueues(options.concurrency ?? {})\n\n  return {\n    runCase<T>(context: SchedulerScopeContext, execute: () => Promise<T>) {\n      const activeScopes = getActiveScopes(context)\n\n      return runWithQueues(activeScopes, context, queues, () => {\n        if (middleware.length === 0) {\n          return execute()\n        }\n\n        return runWithMiddlewareEnvelope(middleware, context, execute)\n      })\n    },\n  }\n}\n\n/**\n * Resolves the scheduler scopes that apply to a context.\n *\n * Before:\n * - `{ scope: 'case', workspaceId: 'ws', experimentId: 'exp', caseId: 'case-1' }`\n *\n * After:\n * - `['workspace', 'project', 'task', 'attempt', 'case']` up to the requested scope\n */\nexport function getActiveScopes(context: SchedulerScopeContext): SchedulerScope[] {\n  const targetScopeIndex = schedulerScopeOrder.indexOf(context.scope)\n\n  if (targetScopeIndex < 0) {\n    return []\n  }\n\n  return schedulerScopeOrder.slice(0, targetScopeIndex + 1)\n}\n\nfunction createRuntimeQueues(concurrency: SchedulerConcurrencyConfig) {\n  const queues = new Map<SchedulerScope, SchedulerScopeQueueRegistry>()\n\n  for (const scope of schedulerScopeOrder) {\n    const scopeConcurrency = concurrency[scope]\n\n    if (scopeConcurrency === undefined) {\n      continue\n    }\n\n    validateSchedulerConcurrency(scope, scopeConcurrency)\n\n    queues.set(scope, {\n      concurrency: scopeConcurrency,\n      instances: new Map<string, ReturnType<typeof createSchedulerQueue>>(),\n    })\n  }\n\n  return queues\n}\n\nasync function runWithQueues<T>(\n  scopes: SchedulerScope[],\n  context: SchedulerScopeContext,\n  queues: Map<SchedulerScope, SchedulerScopeQueueRegistry>,\n  execute: () => Promise<T>,\n  index = 0,\n): Promise<T> {\n  const scope = scopes[index]\n\n  if (scope === undefined) {\n    return execute()\n  }\n\n  const queue = getScopeQueue(scope, context, queues)\n\n  if (queue === undefined) {\n    return runWithQueues(scopes, context, queues, execute, index + 1)\n  }\n\n  return queue.run(() => runWithQueues(scopes, context, queues, execute, index + 1))\n}\n\ninterface SchedulerScopeQueueRegistry {\n  concurrency: number\n  instances: Map<string, ReturnType<typeof createSchedulerQueue>>\n}\n\ninterface SchedulerEnvelopeResult<T> {\n  releaseStack: SchedulerMiddleware[]\n  outcome: SchedulerExecutionOutcome<T>\n}\n\ninterface SchedulerExecutionFailure {\n  error: unknown\n  status: 'failed'\n}\n\ninterface SchedulerExecutionSkipped {\n  status: 'skipped'\n}\n\ninterface SchedulerExecutionSuccess<T> {\n  status: 'succeeded'\n  value: T\n}\n\ntype SchedulerExecutionOutcome<T>\n  = | SchedulerExecutionFailure\n    | SchedulerExecutionSkipped\n    | SchedulerExecutionSuccess<T>\n\nfunction getScopeQueue(\n  scope: SchedulerScope,\n  context: SchedulerScopeContext,\n  queues: Map<SchedulerScope, SchedulerScopeQueueRegistry>,\n) {\n  const queueRegistry = queues.get(scope)\n\n  if (queueRegistry === undefined) {\n    return undefined\n  }\n\n  const scopeKey = getSchedulerScopeInstanceKey(scope, context)\n  const existingQueue = queueRegistry.instances.get(scopeKey)\n\n  if (existingQueue !== undefined) {\n    return existingQueue\n  }\n\n  const queue = createSchedulerQueue(queueRegistry.concurrency)\n  queueRegistry.instances.set(scopeKey, queue)\n  return queue\n}\n\nfunction getSchedulerScopeInstanceKey(\n  scope: SchedulerScope,\n  context: SchedulerScopeContext,\n): string {\n  const workspaceKey = `workspace:${context.workspaceId}:experiment:${context.experimentId}`\n  const projectKey = `${workspaceKey}:project:${context.projectName ?? '(missing-project)'}`\n  const taskKey = `${projectKey}:task:${context.taskId ?? '(missing-task)'}`\n  const attemptKey = `${taskKey}:attempt:${context.attemptIndex ?? '(missing-attempt)'}`\n\n  switch (scope) {\n    case 'workspace':\n      return workspaceKey\n    case 'project':\n      return projectKey\n    case 'task':\n      return taskKey\n    case 'attempt':\n      return attemptKey\n    case 'case':\n      return attemptKey\n  }\n}\n\nasync function runWithMiddlewareEnvelope<T>(\n  middleware: SchedulerMiddleware[],\n  context: SchedulerScopeContext,\n  execute: () => Promise<T>,\n): Promise<T> {\n  const result = await runAcquireMiddleware(middleware, context, execute, 0)\n\n  try {\n    switch (result.outcome.status) {\n      case 'succeeded':\n        return result.outcome.value\n      case 'failed':\n        throw result.outcome.error\n      case 'skipped':\n        throw createSchedulerShortCircuitError()\n    }\n  }\n  finally {\n    await runReleaseMiddleware(result.releaseStack, context, result.releaseStack.length - 1)\n  }\n}\n\nasync function runAcquireMiddleware<T>(\n  middleware: SchedulerMiddleware[],\n  context: SchedulerScopeContext,\n  execute: () => Promise<T>,\n  index: number,\n): Promise<SchedulerEnvelopeResult<T>> {\n  const currentMiddleware = middleware[index]\n\n  if (currentMiddleware === undefined) {\n    return createSchedulerExecutionResult([], execute)\n  }\n\n  let nextResult = createSchedulerShortCircuitResult<T>()\n  let didCallNext = false\n\n  const next = async () => {\n    didCallNext = true\n    nextResult = await runAcquireMiddleware(middleware, context, execute, index + 1)\n  }\n\n  try {\n    if (currentMiddleware.onAcquire === undefined) {\n      await next()\n    }\n    else {\n      await currentMiddleware.onAcquire(context, next)\n    }\n  }\n  catch (error) {\n    if (!didCallNext) {\n      return createSchedulerFailureResult([], error)\n    }\n\n    return createSchedulerFailureResult(\n      [currentMiddleware, ...nextResult.releaseStack],\n      error,\n    )\n  }\n\n  return {\n    releaseStack: [currentMiddleware, ...nextResult.releaseStack],\n    outcome: nextResult.outcome,\n  }\n}\n\nasync function runReleaseMiddleware(\n  releaseStack: SchedulerMiddleware[],\n  context: SchedulerScopeContext,\n  index: number,\n): Promise<void> {\n  const currentMiddleware = releaseStack[index]\n\n  if (currentMiddleware === undefined) {\n    return\n  }\n\n  if (currentMiddleware.onRelease === undefined) {\n    await runReleaseMiddleware(releaseStack, context, index - 1)\n    return\n  }\n\n  await currentMiddleware.onRelease(context, async () => {\n    await runReleaseMiddleware(releaseStack, context, index - 1)\n  })\n}\n\nasync function createSchedulerExecutionResult<T>(\n  releaseStack: SchedulerMiddleware[],\n  execute: () => Promise<T>,\n): Promise<SchedulerEnvelopeResult<T>> {\n  try {\n    return {\n      releaseStack,\n      outcome: {\n        status: 'succeeded',\n        value: await execute(),\n      },\n    }\n  }\n  catch (error) {\n    return {\n      releaseStack,\n      outcome: {\n        status: 'failed',\n        error,\n      },\n    }\n  }\n}\n\nfunction createSchedulerFailureResult<T>(\n  releaseStack: SchedulerMiddleware[],\n  error: unknown,\n): SchedulerEnvelopeResult<T> {\n  return {\n    releaseStack,\n    outcome: {\n      status: 'failed',\n      error,\n    },\n  }\n}\n\nfunction createSchedulerShortCircuitResult<T>(): SchedulerEnvelopeResult<T> {\n  return {\n    releaseStack: [],\n    outcome: {\n      status: 'skipped',\n    },\n  }\n}\n\nfunction validateSchedulerConcurrency(scope: SchedulerScope, concurrency: number): void {\n  if (!Number.isFinite(concurrency) || !Number.isInteger(concurrency) || concurrency <= 0) {\n    throw new Error(`Invalid scheduler concurrency for \"${scope}\": ${String(concurrency)}`)\n  }\n}\n\nfunction createSchedulerShortCircuitError(): Error {\n  return new Error('Scheduler middleware short-circuited execution.')\n}\n"],"mappings":";;AAWA,MAAM,sBAAwC;CAC5C;CACA;CACA;CACA;CACA;CACD;;;;;;;;;;;;;;;;;;;;;;;;;AA0BD,SAAgB,uBACd,UAAyC,EAAE,EACzB;CAClB,MAAM,aAAa,QAAQ,cAAc,EAAE;CAC3C,MAAM,SAAS,oBAAoB,QAAQ,eAAe,EAAE,CAAC;AAE7D,QAAO,EACL,QAAW,SAAgC,SAA2B;AAGpE,SAAO,cAFc,gBAAgB,QAAQ,EAEV,SAAS,cAAc;AACxD,OAAI,WAAW,WAAW,EACxB,QAAO,SAAS;AAGlB,UAAO,0BAA0B,YAAY,SAAS,QAAQ;IAC9D;IAEL;;;;;;;;;;;AAYH,SAAgB,gBAAgB,SAAkD;CAChF,MAAM,mBAAmB,oBAAoB,QAAQ,QAAQ,MAAM;AAEnE,KAAI,mBAAmB,EACrB,QAAO,EAAE;AAGX,QAAO,oBAAoB,MAAM,GAAG,mBAAmB,EAAE;;AAG3D,SAAS,oBAAoB,aAAyC;CACpE,MAAM,yBAAS,IAAI,KAAkD;AAErE,MAAK,MAAM,SAAS,qBAAqB;EACvC,MAAM,mBAAmB,YAAY;AAErC,MAAI,qBAAqB,KAAA,EACvB;AAGF,+BAA6B,OAAO,iBAAiB;AAErD,SAAO,IAAI,OAAO;GAChB,aAAa;GACb,2BAAW,IAAI,KAAsD;GACtE,CAAC;;AAGJ,QAAO;;AAGT,eAAe,cACb,QACA,SACA,QACA,SACA,QAAQ,GACI;CACZ,MAAM,QAAQ,OAAO;AAErB,KAAI,UAAU,KAAA,EACZ,QAAO,SAAS;CAGlB,MAAM,QAAQ,cAAc,OAAO,SAAS,OAAO;AAEnD,KAAI,UAAU,KAAA,EACZ,QAAO,cAAc,QAAQ,SAAS,QAAQ,SAAS,QAAQ,EAAE;AAGnE,QAAO,MAAM,UAAU,cAAc,QAAQ,SAAS,QAAQ,SAAS,QAAQ,EAAE,CAAC;;AAgCpF,SAAS,cACP,OACA,SACA,QACA;CACA,MAAM,gBAAgB,OAAO,IAAI,MAAM;AAEvC,KAAI,kBAAkB,KAAA,EACpB;CAGF,MAAM,WAAW,6BAA6B,OAAO,QAAQ;CAC7D,MAAM,gBAAgB,cAAc,UAAU,IAAI,SAAS;AAE3D,KAAI,kBAAkB,KAAA,EACpB,QAAO;CAGT,MAAM,QAAQ,qBAAqB,cAAc,YAAY;AAC7D,eAAc,UAAU,IAAI,UAAU,MAAM;AAC5C,QAAO;;AAGT,SAAS,6BACP,OACA,SACQ;CACR,MAAM,eAAe,aAAa,QAAQ,YAAY,cAAc,QAAQ;CAC5E,MAAM,aAAa,GAAG,aAAa,WAAW,QAAQ,eAAe;CACrE,MAAM,UAAU,GAAG,WAAW,QAAQ,QAAQ,UAAU;CACxD,MAAM,aAAa,GAAG,QAAQ,WAAW,QAAQ,gBAAgB;AAEjE,SAAQ,OAAR;EACE,KAAK,YACH,QAAO;EACT,KAAK,UACH,QAAO;EACT,KAAK,OACH,QAAO;EACT,KAAK,UACH,QAAO;EACT,KAAK,OACH,QAAO;;;AAIb,eAAe,0BACb,YACA,SACA,SACY;CACZ,MAAM,SAAS,MAAM,qBAAqB,YAAY,SAAS,SAAS,EAAE;AAE1E,KAAI;AACF,UAAQ,OAAO,QAAQ,QAAvB;GACE,KAAK,YACH,QAAO,OAAO,QAAQ;GACxB,KAAK,SACH,OAAM,OAAO,QAAQ;GACvB,KAAK,UACH,OAAM,kCAAkC;;WAGtC;AACN,QAAM,qBAAqB,OAAO,cAAc,SAAS,OAAO,aAAa,SAAS,EAAE;;;AAI5F,eAAe,qBACb,YACA,SACA,SACA,OACqC;CACrC,MAAM,oBAAoB,WAAW;AAErC,KAAI,sBAAsB,KAAA,EACxB,QAAO,+BAA+B,EAAE,EAAE,QAAQ;CAGpD,IAAI,aAAa,mCAAsC;CACvD,IAAI,cAAc;CAElB,MAAM,OAAO,YAAY;AACvB,gBAAc;AACd,eAAa,MAAM,qBAAqB,YAAY,SAAS,SAAS,QAAQ,EAAE;;AAGlF,KAAI;AACF,MAAI,kBAAkB,cAAc,KAAA,EAClC,OAAM,MAAM;MAGZ,OAAM,kBAAkB,UAAU,SAAS,KAAK;UAG7C,OAAO;AACZ,MAAI,CAAC,YACH,QAAO,6BAA6B,EAAE,EAAE,MAAM;AAGhD,SAAO,6BACL,CAAC,mBAAmB,GAAG,WAAW,aAAa,EAC/C,MACD;;AAGH,QAAO;EACL,cAAc,CAAC,mBAAmB,GAAG,WAAW,aAAa;EAC7D,SAAS,WAAW;EACrB;;AAGH,eAAe,qBACb,cACA,SACA,OACe;CACf,MAAM,oBAAoB,aAAa;AAEvC,KAAI,sBAAsB,KAAA,EACxB;AAGF,KAAI,kBAAkB,cAAc,KAAA,GAAW;AAC7C,QAAM,qBAAqB,cAAc,SAAS,QAAQ,EAAE;AAC5D;;AAGF,OAAM,kBAAkB,UAAU,SAAS,YAAY;AACrD,QAAM,qBAAqB,cAAc,SAAS,QAAQ,EAAE;GAC5D;;AAGJ,eAAe,+BACb,cACA,SACqC;AACrC,KAAI;AACF,SAAO;GACL;GACA,SAAS;IACP,QAAQ;IACR,OAAO,MAAM,SAAS;IACvB;GACF;UAEI,OAAO;AACZ,SAAO;GACL;GACA,SAAS;IACP,QAAQ;IACR;IACD;GACF;;;AAIL,SAAS,6BACP,cACA,OAC4B;AAC5B,QAAO;EACL;EACA,SAAS;GACP,QAAQ;GACR;GACD;EACF;;AAGH,SAAS,oCAAmE;AAC1E,QAAO;EACL,cAAc,EAAE;EAChB,SAAS,EACP,QAAQ,WACT;EACF;;AAGH,SAAS,6BAA6B,OAAuB,aAA2B;AACtF,KAAI,CAAC,OAAO,SAAS,YAAY,IAAI,CAAC,OAAO,UAAU,YAAY,IAAI,eAAe,EACpF,OAAM,IAAI,MAAM,sCAAsC,MAAM,KAAK,OAAO,YAAY,GAAG;;AAI3F,SAAS,mCAA0C;AACjD,wBAAO,IAAI,MAAM,kDAAkD"}

package/dist/{env-BFSjny07.mjs → env--94B0UtW.mjs}

@@ -38,4 +38,4 @@ function requiredEnvFrom(value, options) {
 //#endregion
 export { requiredEnvFrom as n, envFrom as t };
 
-//# sourceMappingURL=env-BFSjny07.mjs.map
+//# sourceMappingURL=env--94B0UtW.mjs.map

package/dist/{env-BFSjny07.mjs.map → env--94B0UtW.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"env-BFSjny07.mjs","names":[],"sources":["../src/core/inference-executors/env.ts"],"sourcesContent":["/**\n * Supported env value coercion types.\n */\nexport type EnvValueType = 'string'\n\n/**\n * Common options for env readers.\n */\nexport interface EnvFromOptions {\n  /**\n   * Expected env value type.\n   */\n  type: EnvValueType\n  /**\n   * Whether an empty or missing value should throw.\n   *\n   * @default false\n   */\n  required?: boolean\n  /**\n   * Optional key name used for clearer error messages.\n   */\n  name?: string\n}\n\n/**\n * Env options used by the required helper.\n *\n * `required` is intentionally omitted because this helper is always required.\n */\nexport type RequiredEnvFromOptions = Omit<EnvFromOptions, 'required'>\n\nfunction assertNonEmptyString(value: string | undefined, options: EnvFromOptions): string | undefined {\n  if (value == null || value.trim().length === 0) {\n    if (options.required === true) {\n      const label = options.name ?? 'environment variable'\n      throw new Error(`Missing required ${label}.`)\n    }\n\n    return undefined\n  }\n\n  return value\n}\n\n/**\n * Parses one env value with optional required behavior.\n *\n * Example:\n * `const apiKey = envFrom(process.env.OPENAI_API_KEY, { type: 'string', required: true, name: 'OPENAI_API_KEY' })`\n */\nexport function envFrom(\n  value: string | undefined,\n  options: EnvFromOptions,\n): string | undefined {\n  if (options.type === 'string') {\n    return assertNonEmptyString(value, options)\n  }\n\n  return undefined\n}\n\n/**\n * Parses one required env value.\n *\n * Example:\n * `const apiKey = requiredEnvFrom(process.env.OPENAI_API_KEY, { type: 'string', name: 'OPENAI_API_KEY' })`\n */\nexport function requiredEnvFrom(\n  value: string | undefined,\n  options: RequiredEnvFromOptions,\n): string {\n  const parsed = envFrom(value, {\n    ...options,\n    required: true,\n  })\n\n  if (parsed == null) {\n    const label = options.name ?? 'environment variable'\n    throw new Error(`Missing required ${label}.`)\n  }\n\n  return parsed\n}\n"],"mappings":";AAgCA,SAAS,qBAAqB,OAA2B,SAA6C;AACpG,KAAI,SAAS,QAAQ,MAAM,MAAM,CAAC,WAAW,GAAG;AAC9C,MAAI,QAAQ,aAAa,MAAM;GAC7B,MAAM,QAAQ,QAAQ,QAAQ;AAC9B,SAAM,IAAI,MAAM,oBAAoB,MAAM,GAAG;;AAG/C;;AAGF,QAAO;;;;;;;;AAST,SAAgB,QACd,OACA,SACoB;AACpB,KAAI,QAAQ,SAAS,SACnB,QAAO,qBAAqB,OAAO,QAAQ;;;;;;;;AAY/C,SAAgB,gBACd,OACA,SACQ;CACR,MAAM,SAAS,QAAQ,OAAO;EAC5B,GAAG;EACH,UAAU;EACX,CAAC;AAEF,KAAI,UAAU,MAAM;EAClB,MAAM,QAAQ,QAAQ,QAAQ;AAC9B,QAAM,IAAI,MAAM,oBAAoB,MAAM,GAAG;;AAG/C,QAAO"}
+{"version":3,"file":"env--94B0UtW.mjs","names":[],"sources":["../src/core/inference-executors/env.ts"],"sourcesContent":["/**\n * Supported env value coercion types.\n */\nexport type EnvValueType = 'string'\n\n/**\n * Common options for env readers.\n */\nexport interface EnvFromOptions {\n  /**\n   * Expected env value type.\n   */\n  type: EnvValueType\n  /**\n   * Whether an empty or missing value should throw.\n   *\n   * @default false\n   */\n  required?: boolean\n  /**\n   * Optional key name used for clearer error messages.\n   */\n  name?: string\n}\n\n/**\n * Env options used by the required helper.\n *\n * `required` is intentionally omitted because this helper is always required.\n */\nexport type RequiredEnvFromOptions = Omit<EnvFromOptions, 'required'>\n\nfunction assertNonEmptyString(value: string | undefined, options: EnvFromOptions): string | undefined {\n  if (value == null || value.trim().length === 0) {\n    if (options.required === true) {\n      const label = options.name ?? 'environment variable'\n      throw new Error(`Missing required ${label}.`)\n    }\n\n    return undefined\n  }\n\n  return value\n}\n\n/**\n * Parses one env value with optional required behavior.\n *\n * Example:\n * `const apiKey = envFrom(process.env.OPENAI_API_KEY, { type: 'string', required: true, name: 'OPENAI_API_KEY' })`\n */\nexport function envFrom(\n  value: string | undefined,\n  options: EnvFromOptions,\n): string | undefined {\n  if (options.type === 'string') {\n    return assertNonEmptyString(value, options)\n  }\n\n  return undefined\n}\n\n/**\n * Parses one required env value.\n *\n * Example:\n * `const apiKey = requiredEnvFrom(process.env.OPENAI_API_KEY, { type: 'string', name: 'OPENAI_API_KEY' })`\n */\nexport function requiredEnvFrom(\n  value: string | undefined,\n  options: RequiredEnvFromOptions,\n): string {\n  const parsed = envFrom(value, {\n    ...options,\n    required: true,\n  })\n\n  if (parsed == null) {\n    const label = options.name ?? 'environment variable'\n    throw new Error(`Missing required ${label}.`)\n  }\n\n  return parsed\n}\n"],"mappings":";AAgCA,SAAS,qBAAqB,OAA2B,SAA6C;AACpG,KAAI,SAAS,QAAQ,MAAM,MAAM,CAAC,WAAW,GAAG;AAC9C,MAAI,QAAQ,aAAa,MAAM;GAC7B,MAAM,QAAQ,QAAQ,QAAQ;AAC9B,SAAM,IAAI,MAAM,oBAAoB,MAAM,GAAG;;AAG/C;;AAGF,QAAO;;;;;;;;AAST,SAAgB,QACd,OACA,SACoB;AACpB,KAAI,QAAQ,SAAS,SACnB,QAAO,qBAAqB,OAAO,QAAQ;;;;;;;;AAY/C,SAAgB,gBACd,OACA,SACQ;CACR,MAAM,SAAS,QAAQ,OAAO;EAC5B,GAAG;EACH,UAAU;EACX,CAAC;AAEF,KAAI,UAAU,MAAM;EAClB,MAAM,QAAQ,QAAQ,QAAQ;AAC9B,QAAM,IAAI,MAAM,oBAAoB,MAAM,GAAG;;AAG/C,QAAO"}

package/dist/{env-BTq3dV7C.d.mts → env-BeHv_5mo.d.mts}

@@ -44,4 +44,4 @@ declare function envFrom(value: string | undefined, options: EnvFromOptions): st
 declare function requiredEnvFrom(value: string | undefined, options: RequiredEnvFromOptions): string;
 //#endregion
 export { requiredEnvFrom as a, envFrom as i, EnvValueType as n, RequiredEnvFromOptions as r, EnvFromOptions as t };
-//# sourceMappingURL=env-BTq3dV7C.d.mts.map
+//# sourceMappingURL=env-BeHv_5mo.d.mts.map

package/dist/{expect-extensions-QLXESWjn.mjs → expect-extensions-DCSqlneN.mjs}

@@ -194,4 +194,4 @@ function installVievalExpectMatchers() {
 //#endregion
 export { getRuntimeExpect as n, installVievalExpectMatchers as t };
 
-//# sourceMappingURL=expect-extensions-QLXESWjn.mjs.map
+//# sourceMappingURL=expect-extensions-DCSqlneN.mjs.map

package/dist/{expect-extensions-QLXESWjn.mjs.map → expect-extensions-DCSqlneN.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"expect-extensions-QLXESWjn.mjs","names":[],"sources":["../src/testing/runtime-expect.ts","../src/testing/expect-extensions.ts"],"sourcesContent":["import type { ExpectStatic, MatchersObject, MatcherState, Tester } from '@vitest/expect'\n\nimport {\n  addCustomEqualityTesters,\n  ASYMMETRIC_MATCHERS_OBJECT,\n  chai,\n  ChaiStyleAssertions,\n  customMatchers,\n  getState,\n  GLOBAL_EXPECT,\n  JestAsymmetricMatchers,\n  JestChaiExpect,\n  JestExtend,\n  setState,\n} from '@vitest/expect'\n\nlet isPluginInstalled = false\nlet runtimeExpectInstance: ExpectStatic | undefined\n\n/**\n * Installs Vitest expect plugins once for process-local runtime assertions.\n *\n * Use when:\n * - running eval tasks outside Vitest worker runtime\n * - building an `expect` instance that does not rely on Vitest internal state\n *\n * Expects:\n * - `@vitest/expect` is available in runtime dependencies\n *\n * Returns:\n * - nothing; side-effects are applied to `chai`\n */\nfunction ensureRuntimeExpectPluginsInstalled(): void {\n  if (isPluginInstalled) {\n    return\n  }\n\n  chai.use(JestExtend)\n  chai.use(JestChaiExpect)\n  chai.use(ChaiStyleAssertions)\n  chai.use(JestAsymmetricMatchers)\n  isPluginInstalled = true\n}\n\n/**\n * Creates a Vitest-compatible `expect` instance without worker-state coupling.\n *\n * Use when:\n * - CLI runtime needs assertion helpers from `vieval/expect`\n * - code is executed outside `vitest run`\n *\n * Expects:\n * - plugins from {@link ensureRuntimeExpectPluginsInstalled} are installed\n * - callers do not depend on Vitest worker-only features (snapshot/poll internals)\n *\n * Returns:\n * - standalone expect instance with core matcher APIs and `extend`\n */\nfunction createRuntimeExpect(): ExpectStatic {\n  ensureRuntimeExpectPluginsInstalled()\n\n  const runtimeExpect = ((value: unknown, message?: string) => {\n    const currentState = getState(runtimeExpect)\n    setState({ assertionCalls: currentState.assertionCalls + 1 }, runtimeExpect)\n    return chai.expect(value, message)\n  }) as unknown as ExpectStatic\n\n  Object.assign(runtimeExpect, chai.expect)\n  Object.assign(runtimeExpect, (globalThis as Record<PropertyKey, unknown>)[ASYMMETRIC_MATCHERS_OBJECT] as object)\n\n  runtimeExpect.getState = () => getState(runtimeExpect)\n  runtimeExpect.setState = (state: Partial<MatcherState>) => setState(state, runtimeExpect)\n  runtimeExpect.assert = chai.assert\n  // NOTICE:\n  // Chai's public `ExpectStatic` type does not expose Vitest's plugin-added `extend`.\n  // Runtime `chai.expect.extend` exists after `JestExtend` plugin installation.\n  // Source/context: `@vitest/expect` plugin pipeline in `dist/index.js`.\n  // Removal condition: remove this cast if upstream exposes `extend` on Chai expect types.\n  const chaiExpectWithExtend = chai.expect as unknown as {\n    extend: (expect: ExpectStatic, matchers: MatchersObject) => void\n  }\n  runtimeExpect.extend = (matchers: MatchersObject) => chaiExpectWithExtend.extend(runtimeExpect, matchers)\n  runtimeExpect.addEqualityTesters = (customTesters: Tester[]) => addCustomEqualityTesters(customTesters)\n  runtimeExpect.unreachable = (message?: string) => {\n    chai.assert.fail(`expected${message ? ` \"${message}\" ` : ' '}not to be reached`)\n  }\n\n  runtimeExpect.setState({\n    assertionCalls: 0,\n    currentTestName: '',\n    expectedAssertionsNumber: null,\n    expectedAssertionsNumberErrorGen: null,\n    isExpectingAssertions: false,\n    isExpectingAssertionsError: null,\n  })\n\n  runtimeExpect.extend(customMatchers)\n\n  return runtimeExpect\n}\n\n/**\n * Returns process-local runtime `expect` instance used by Vieval.\n *\n * Use when:\n * - you need matcher assertions in eval files and CLI runtime\n * - importing from `vitest` would crash outside Vitest worker contexts\n *\n * Expects:\n * - single-process usage (instance is memoized per process)\n *\n * Returns:\n * - memoized runtime `expect` instance\n */\nexport function getRuntimeExpect(): ExpectStatic {\n  if (runtimeExpectInstance != null) {\n    return runtimeExpectInstance\n  }\n\n  runtimeExpectInstance = createRuntimeExpect()\n  Object.defineProperty(globalThis, GLOBAL_EXPECT, {\n    configurable: true,\n    value: runtimeExpectInstance,\n    writable: true,\n  })\n\n  return runtimeExpectInstance\n}\n","import type { RubricJudgeResult, ToolCall } from '../core/assertions'\n\nimport { normalizeMatchText } from '../core/assertions'\nimport { getRuntimeExpect } from './runtime-expect'\n\n/**\n * Options for keyword-based matcher behavior.\n */\nexport interface KeywordMatcherOptions {\n  /**\n   * Case-sensitive matching toggle.\n   *\n   * @default false\n   */\n  caseSensitive?: boolean\n  /**\n   * Match mode.\n   *\n   * @default 'all'\n   */\n  mode?: 'all' | 'any'\n}\n\n/**\n * Shape used by tool-call matchers.\n */\nexport interface ToolCallContainer {\n  /**\n   * Tool calls to inspect.\n   */\n  toolCalls?: readonly ToolCall[]\n}\n\nfunction toKeywordArray(keywords: string | readonly string[]): readonly string[] {\n  if (typeof keywords === 'string') {\n    return [keywords]\n  }\n\n  return keywords\n}\n\n/**\n * Registers vieval custom matchers on Vitest `expect`.\n *\n * Call stack:\n *\n * {@link installVievalExpectMatchers}\n *   -> `expect.extend(...)`\n *     -> `expect(received).toMustInclude(...)`\n *     -> `expect(received).toScoreRubricGreaterThan(...)`\n *\n * Use when:\n * - eval suites need domain assertions while preserving native Vitest ergonomics\n * - callers want native `.not` chaining with the same matchers\n */\nexport function installVievalExpectMatchers(): void {\n  const expect = getRuntimeExpect()\n\n  expect.extend({\n    toMustExclude(received: unknown, keywords: string | readonly string[], options: KeywordMatcherOptions = {}) {\n      const keywordList = toKeywordArray(keywords)\n\n      if (typeof received !== 'string') {\n        return {\n          message: () => 'Expected received value to be a string.',\n          pass: false,\n        }\n      }\n\n      const normalizedText = normalizeMatchText(received, options.caseSensitive ?? false)\n      const forbiddenMatches = keywordList.filter((keyword) => {\n        return normalizedText.includes(normalizeMatchText(keyword, options.caseSensitive ?? false))\n      })\n\n      const pass = forbiddenMatches.length === 0\n\n      return {\n        message: () => {\n          if (pass) {\n            return `Expected text to include forbidden keywords: ${keywordList.join(', ')}`\n          }\n\n          return `Expected text not to include forbidden keywords, but matched: ${forbiddenMatches.join(', ')}`\n        },\n        pass,\n      }\n    },\n\n    toMustInclude(received: unknown, keywords: string | readonly string[], options: KeywordMatcherOptions = {}) {\n      const keywordList = toKeywordArray(keywords)\n\n      if (typeof received !== 'string') {\n        return {\n          message: () => 'Expected received value to be a string.',\n          pass: false,\n        }\n      }\n\n      const normalizedText = normalizeMatchText(received, options.caseSensitive ?? false)\n      const matches = keywordList.filter((keyword) => {\n        return normalizedText.includes(normalizeMatchText(keyword, options.caseSensitive ?? false))\n      })\n\n      const mode = options.mode ?? 'all'\n      const pass = mode === 'all' ? matches.length === keywordList.length : matches.length > 0\n\n      return {\n        message: () => {\n          if (pass) {\n            return `Expected text not to match required keywords, but matched: ${matches.join(', ')}`\n          }\n\n          return `Expected text to match required keywords (${mode}), but matched ${matches.length}/${keywordList.length}.`\n        },\n        pass,\n      }\n    },\n\n    toScoreRubricGreaterThan(received: unknown, threshold: number) {\n      const score = typeof received === 'number'\n        ? received\n        : (received as RubricJudgeResult | null)?.score\n\n      if (typeof score !== 'number') {\n        return {\n          message: () => 'Expected received value to be a number or RubricJudgeResult.',\n          pass: false,\n        }\n      }\n\n      const pass = score > threshold\n\n      return {\n        message: () => {\n          if (pass) {\n            return `Expected rubric score ${score} to be less than or equal to ${threshold}.`\n          }\n\n          return `Expected rubric score ${score} to be greater than ${threshold}.`\n        },\n        pass,\n      }\n    },\n\n    toSatisfyStructuredOutput<T>(received: unknown, validator: (value: unknown) => value is T) {\n      const pass = validator(received)\n\n      return {\n        message: () => pass\n          ? 'Expected structured output validator to fail.'\n          : 'Expected structured output validator to pass.',\n        pass,\n      }\n    },\n\n    toSatisfyToolCallArgs(\n      received: unknown,\n      toolName: string,\n      validator: (args: unknown) => boolean,\n    ) {\n      const toolCalls = (received as ToolCallContainer | null)?.toolCalls\n\n      if (toolCalls == null) {\n        return {\n          message: () => 'Expected received value to provide toolCalls array.',\n          pass: false,\n        }\n      }\n\n      const targetCall = toolCalls.find(call => call.name === toolName)\n      if (targetCall == null) {\n        return {\n          message: () => `Expected tool call ${toolName} to exist.`,\n          pass: false,\n        }\n      }\n\n      const pass = validator(targetCall.args)\n\n      return {\n        message: () => pass\n          ? `Expected tool call args for ${toolName} to fail validation.`\n          : `Expected tool call args for ${toolName} to pass validation.`,\n        pass,\n      }\n    },\n  })\n}\n\ninterface VievalCustomMatchers {\n  /**\n   * Asserts that text includes required keywords.\n   *\n   * Example:\n   * `expect('calm answer').toMustInclude(['calm'])`\n   */\n  toMustInclude: (keywords: string | readonly string[], options?: KeywordMatcherOptions) => void\n  /**\n   * Asserts that text excludes forbidden keywords.\n   *\n   * Example:\n   * `expect('calm answer').toMustExclude(['bestmove'])`\n   */\n  toMustExclude: (keywords: string | readonly string[], options?: KeywordMatcherOptions) => void\n  /**\n   * Asserts rubric score is greater than a threshold.\n   *\n   * Example:\n   * `expect({ score: 0.91 }).toScoreRubricGreaterThan(0.8)`\n   */\n  toScoreRubricGreaterThan: (threshold: number) => void\n  /**\n   * Asserts structured output satisfies a validator.\n   *\n   * Example:\n   * `expect(value).toSatisfyStructuredOutput(isMyShape)`\n   */\n  toSatisfyStructuredOutput: <TValue>(validator: (value: unknown) => value is TValue) => void\n  /**\n   * Asserts selected tool-call args satisfy validator.\n   *\n   * Example:\n   * `expect({ toolCalls }).toSatisfyToolCallArgs('builtIn_sparkCommand', isSparkArgs)`\n   */\n  toSatisfyToolCallArgs: (toolName: string, validator: (args: unknown) => boolean) => void\n}\n\n/* eslint-disable unused-imports/no-unused-vars */\ndeclare module '@vitest/expect' {\n  interface Matchers<T = any> extends VievalCustomMatchers {}\n  interface Assertion<T = any> extends VievalCustomMatchers {}\n}\n\ndeclare module 'vitest' {\n  interface Assertion extends VievalCustomMatchers {}\n  interface Matchers<T = any> extends VievalCustomMatchers {}\n}\n/* eslint-enable unused-imports/no-unused-vars */\n"],"mappings":";;;AAgBA,IAAI,oBAAoB;AACxB,IAAI;;;;;;;;;;;;;;AAeJ,SAAS,sCAA4C;AACnD,KAAI,kBACF;AAGF,MAAK,IAAI,WAAW;AACpB,MAAK,IAAI,eAAe;AACxB,MAAK,IAAI,oBAAoB;AAC7B,MAAK,IAAI,uBAAuB;AAChC,qBAAoB;;;;;;;;;;;;;;;;AAiBtB,SAAS,sBAAoC;AAC3C,sCAAqC;CAErC,MAAM,kBAAkB,OAAgB,YAAqB;AAE3D,WAAS,EAAE,gBADU,SAAS,cAAc,CACJ,iBAAiB,GAAG,EAAE,cAAc;AAC5E,SAAO,KAAK,OAAO,OAAO,QAAQ;;AAGpC,QAAO,OAAO,eAAe,KAAK,OAAO;AACzC,QAAO,OAAO,eAAgB,WAA4C,4BAAsC;AAEhH,eAAc,iBAAiB,SAAS,cAAc;AACtD,eAAc,YAAY,UAAiC,SAAS,OAAO,cAAc;AACzF,eAAc,SAAS,KAAK;CAM5B,MAAM,uBAAuB,KAAK;AAGlC,eAAc,UAAU,aAA6B,qBAAqB,OAAO,eAAe,SAAS;AACzG,eAAc,sBAAsB,kBAA4B,yBAAyB,cAAc;AACvG,eAAc,eAAe,YAAqB;AAChD,OAAK,OAAO,KAAK,WAAW,UAAU,KAAK,QAAQ,MAAM,IAAI,mBAAmB;;AAGlF,eAAc,SAAS;EACrB,gBAAgB;EAChB,iBAAiB;EACjB,0BAA0B;EAC1B,kCAAkC;EAClC,uBAAuB;EACvB,4BAA4B;EAC7B,CAAC;AAEF,eAAc,OAAO,eAAe;AAEpC,QAAO;;;;;;;;;;;;;;;AAgBT,SAAgB,mBAAiC;AAC/C,KAAI,yBAAyB,KAC3B,QAAO;AAGT,yBAAwB,qBAAqB;AAC7C,QAAO,eAAe,YAAY,eAAe;EAC/C,cAAc;EACd,OAAO;EACP,UAAU;EACX,CAAC;AAEF,QAAO;;;;AC7FT,SAAS,eAAe,UAAyD;AAC/E,KAAI,OAAO,aAAa,SACtB,QAAO,CAAC,SAAS;AAGnB,QAAO;;;;;;;;;;;;;;;;AAiBT,SAAgB,8BAAoC;AACnC,mBAAkB,CAE1B,OAAO;EACZ,cAAc,UAAmB,UAAsC,UAAiC,EAAE,EAAE;GAC1G,MAAM,cAAc,eAAe,SAAS;AAE5C,OAAI,OAAO,aAAa,SACtB,QAAO;IACL,eAAe;IACf,MAAM;IACP;GAGH,MAAM,iBAAiB,mBAAmB,UAAU,QAAQ,iBAAiB,MAAM;GACnF,MAAM,mBAAmB,YAAY,QAAQ,YAAY;AACvD,WAAO,eAAe,SAAS,mBAAmB,SAAS,QAAQ,iBAAiB,MAAM,CAAC;KAC3F;GAEF,MAAM,OAAO,iBAAiB,WAAW;AAEzC,UAAO;IACL,eAAe;AACb,SAAI,KACF,QAAO,gDAAgD,YAAY,KAAK,KAAK;AAG/E,YAAO,iEAAiE,iBAAiB,KAAK,KAAK;;IAErG;IACD;;EAGH,cAAc,UAAmB,UAAsC,UAAiC,EAAE,EAAE;GAC1G,MAAM,cAAc,eAAe,SAAS;AAE5C,OAAI,OAAO,aAAa,SACtB,QAAO;IACL,eAAe;IACf,MAAM;IACP;GAGH,MAAM,iBAAiB,mBAAmB,UAAU,QAAQ,iBAAiB,MAAM;GACnF,MAAM,UAAU,YAAY,QAAQ,YAAY;AAC9C,WAAO,eAAe,SAAS,mBAAmB,SAAS,QAAQ,iBAAiB,MAAM,CAAC;KAC3F;GAEF,MAAM,OAAO,QAAQ,QAAQ;GAC7B,MAAM,OAAO,SAAS,QAAQ,QAAQ,WAAW,YAAY,SAAS,QAAQ,SAAS;AAEvF,UAAO;IACL,eAAe;AACb,SAAI,KACF,QAAO,8DAA8D,QAAQ,KAAK,KAAK;AAGzF,YAAO,6CAA6C,KAAK,iBAAiB,QAAQ,OAAO,GAAG,YAAY,OAAO;;IAEjH;IACD;;EAGH,yBAAyB,UAAmB,WAAmB;GAC7D,MAAM,QAAQ,OAAO,aAAa,WAC9B,WACC,UAAuC;AAE5C,OAAI,OAAO,UAAU,SACnB,QAAO;IACL,eAAe;IACf,MAAM;IACP;GAGH,MAAM,OAAO,QAAQ;AAErB,UAAO;IACL,eAAe;AACb,SAAI,KACF,QAAO,yBAAyB,MAAM,+BAA+B,UAAU;AAGjF,YAAO,yBAAyB,MAAM,sBAAsB,UAAU;;IAExE;IACD;;EAGH,0BAA6B,UAAmB,WAA2C;GACzF,MAAM,OAAO,UAAU,SAAS;AAEhC,UAAO;IACL,eAAe,OACX,kDACA;IACJ;IACD;;EAGH,sBACE,UACA,UACA,WACA;GACA,MAAM,YAAa,UAAuC;AAE1D,OAAI,aAAa,KACf,QAAO;IACL,eAAe;IACf,MAAM;IACP;GAGH,MAAM,aAAa,UAAU,MAAK,SAAQ,KAAK,SAAS,SAAS;AACjE,OAAI,cAAc,KAChB,QAAO;IACL,eAAe,sBAAsB,SAAS;IAC9C,MAAM;IACP;GAGH,MAAM,OAAO,UAAU,WAAW,KAAK;AAEvC,UAAO;IACL,eAAe,OACX,+BAA+B,SAAS,wBACxC,+BAA+B,SAAS;IAC5C;IACD;;EAEJ,CAAC"}
+{"version":3,"file":"expect-extensions-DCSqlneN.mjs","names":[],"sources":["../src/testing/runtime-expect.ts","../src/testing/expect-extensions.ts"],"sourcesContent":["import type { ExpectStatic, MatchersObject, MatcherState, Tester } from '@vitest/expect'\n\nimport {\n  addCustomEqualityTesters,\n  ASYMMETRIC_MATCHERS_OBJECT,\n  chai,\n  ChaiStyleAssertions,\n  customMatchers,\n  getState,\n  GLOBAL_EXPECT,\n  JestAsymmetricMatchers,\n  JestChaiExpect,\n  JestExtend,\n  setState,\n} from '@vitest/expect'\n\nlet isPluginInstalled = false\nlet runtimeExpectInstance: ExpectStatic | undefined\n\n/**\n * Installs Vitest expect plugins once for process-local runtime assertions.\n *\n * Use when:\n * - running eval tasks outside Vitest worker runtime\n * - building an `expect` instance that does not rely on Vitest internal state\n *\n * Expects:\n * - `@vitest/expect` is available in runtime dependencies\n *\n * Returns:\n * - nothing; side-effects are applied to `chai`\n */\nfunction ensureRuntimeExpectPluginsInstalled(): void {\n  if (isPluginInstalled) {\n    return\n  }\n\n  chai.use(JestExtend)\n  chai.use(JestChaiExpect)\n  chai.use(ChaiStyleAssertions)\n  chai.use(JestAsymmetricMatchers)\n  isPluginInstalled = true\n}\n\n/**\n * Creates a Vitest-compatible `expect` instance without worker-state coupling.\n *\n * Use when:\n * - CLI runtime needs assertion helpers from `vieval/expect`\n * - code is executed outside `vitest run`\n *\n * Expects:\n * - plugins from {@link ensureRuntimeExpectPluginsInstalled} are installed\n * - callers do not depend on Vitest worker-only features (snapshot/poll internals)\n *\n * Returns:\n * - standalone expect instance with core matcher APIs and `extend`\n */\nfunction createRuntimeExpect(): ExpectStatic {\n  ensureRuntimeExpectPluginsInstalled()\n\n  const runtimeExpect = ((value: unknown, message?: string) => {\n    const currentState = getState(runtimeExpect)\n    setState({ assertionCalls: currentState.assertionCalls + 1 }, runtimeExpect)\n    return chai.expect(value, message)\n  }) as unknown as ExpectStatic\n\n  Object.assign(runtimeExpect, chai.expect)\n  Object.assign(runtimeExpect, (globalThis as Record<PropertyKey, unknown>)[ASYMMETRIC_MATCHERS_OBJECT] as object)\n\n  runtimeExpect.getState = () => getState(runtimeExpect)\n  runtimeExpect.setState = (state: Partial<MatcherState>) => setState(state, runtimeExpect)\n  runtimeExpect.assert = chai.assert\n  // NOTICE:\n  // Chai's public `ExpectStatic` type does not expose Vitest's plugin-added `extend`.\n  // Runtime `chai.expect.extend` exists after `JestExtend` plugin installation.\n  // Source/context: `@vitest/expect` plugin pipeline in `dist/index.js`.\n  // Removal condition: remove this cast if upstream exposes `extend` on Chai expect types.\n  const chaiExpectWithExtend = chai.expect as unknown as {\n    extend: (expect: ExpectStatic, matchers: MatchersObject) => void\n  }\n  runtimeExpect.extend = (matchers: MatchersObject) => chaiExpectWithExtend.extend(runtimeExpect, matchers)\n  runtimeExpect.addEqualityTesters = (customTesters: Tester[]) => addCustomEqualityTesters(customTesters)\n  runtimeExpect.unreachable = (message?: string) => {\n    chai.assert.fail(`expected${message ? ` \"${message}\" ` : ' '}not to be reached`)\n  }\n\n  runtimeExpect.setState({\n    assertionCalls: 0,\n    currentTestName: '',\n    expectedAssertionsNumber: null,\n    expectedAssertionsNumberErrorGen: null,\n    isExpectingAssertions: false,\n    isExpectingAssertionsError: null,\n  })\n\n  runtimeExpect.extend(customMatchers)\n\n  return runtimeExpect\n}\n\n/**\n * Returns process-local runtime `expect` instance used by Vieval.\n *\n * Use when:\n * - you need matcher assertions in eval files and CLI runtime\n * - importing from `vitest` would crash outside Vitest worker contexts\n *\n * Expects:\n * - single-process usage (instance is memoized per process)\n *\n * Returns:\n * - memoized runtime `expect` instance\n */\nexport function getRuntimeExpect(): ExpectStatic {\n  if (runtimeExpectInstance != null) {\n    return runtimeExpectInstance\n  }\n\n  runtimeExpectInstance = createRuntimeExpect()\n  Object.defineProperty(globalThis, GLOBAL_EXPECT, {\n    configurable: true,\n    value: runtimeExpectInstance,\n    writable: true,\n  })\n\n  return runtimeExpectInstance\n}\n","import type { RubricJudgeResult, ToolCall } from '../core/assertions'\n\nimport { normalizeMatchText } from '../core/assertions'\nimport { getRuntimeExpect } from './runtime-expect'\n\n/**\n * Options for keyword-based matcher behavior.\n */\nexport interface KeywordMatcherOptions {\n  /**\n   * Case-sensitive matching toggle.\n   *\n   * @default false\n   */\n  caseSensitive?: boolean\n  /**\n   * Match mode.\n   *\n   * @default 'all'\n   */\n  mode?: 'all' | 'any'\n}\n\n/**\n * Shape used by tool-call matchers.\n */\nexport interface ToolCallContainer {\n  /**\n   * Tool calls to inspect.\n   */\n  toolCalls?: readonly ToolCall[]\n}\n\nfunction toKeywordArray(keywords: string | readonly string[]): readonly string[] {\n  if (typeof keywords === 'string') {\n    return [keywords]\n  }\n\n  return keywords\n}\n\n/**\n * Registers vieval custom matchers on Vitest `expect`.\n *\n * Call stack:\n *\n * {@link installVievalExpectMatchers}\n *   -> `expect.extend(...)`\n *     -> `expect(received).toMustInclude(...)`\n *     -> `expect(received).toScoreRubricGreaterThan(...)`\n *\n * Use when:\n * - eval suites need domain assertions while preserving native Vitest ergonomics\n * - callers want native `.not` chaining with the same matchers\n */\nexport function installVievalExpectMatchers(): void {\n  const expect = getRuntimeExpect()\n\n  expect.extend({\n    toMustExclude(received: unknown, keywords: string | readonly string[], options: KeywordMatcherOptions = {}) {\n      const keywordList = toKeywordArray(keywords)\n\n      if (typeof received !== 'string') {\n        return {\n          message: () => 'Expected received value to be a string.',\n          pass: false,\n        }\n      }\n\n      const normalizedText = normalizeMatchText(received, options.caseSensitive ?? false)\n      const forbiddenMatches = keywordList.filter((keyword) => {\n        return normalizedText.includes(normalizeMatchText(keyword, options.caseSensitive ?? false))\n      })\n\n      const pass = forbiddenMatches.length === 0\n\n      return {\n        message: () => {\n          if (pass) {\n            return `Expected text to include forbidden keywords: ${keywordList.join(', ')}`\n          }\n\n          return `Expected text not to include forbidden keywords, but matched: ${forbiddenMatches.join(', ')}`\n        },\n        pass,\n      }\n    },\n\n    toMustInclude(received: unknown, keywords: string | readonly string[], options: KeywordMatcherOptions = {}) {\n      const keywordList = toKeywordArray(keywords)\n\n      if (typeof received !== 'string') {\n        return {\n          message: () => 'Expected received value to be a string.',\n          pass: false,\n        }\n      }\n\n      const normalizedText = normalizeMatchText(received, options.caseSensitive ?? false)\n      const matches = keywordList.filter((keyword) => {\n        return normalizedText.includes(normalizeMatchText(keyword, options.caseSensitive ?? false))\n      })\n\n      const mode = options.mode ?? 'all'\n      const pass = mode === 'all' ? matches.length === keywordList.length : matches.length > 0\n\n      return {\n        message: () => {\n          if (pass) {\n            return `Expected text not to match required keywords, but matched: ${matches.join(', ')}`\n          }\n\n          return `Expected text to match required keywords (${mode}), but matched ${matches.length}/${keywordList.length}.`\n        },\n        pass,\n      }\n    },\n\n    toScoreRubricGreaterThan(received: unknown, threshold: number) {\n      const score = typeof received === 'number'\n        ? received\n        : (received as RubricJudgeResult | null)?.score\n\n      if (typeof score !== 'number') {\n        return {\n          message: () => 'Expected received value to be a number or RubricJudgeResult.',\n          pass: false,\n        }\n      }\n\n      const pass = score > threshold\n\n      return {\n        message: () => {\n          if (pass) {\n            return `Expected rubric score ${score} to be less than or equal to ${threshold}.`\n          }\n\n          return `Expected rubric score ${score} to be greater than ${threshold}.`\n        },\n        pass,\n      }\n    },\n\n    toSatisfyStructuredOutput<T>(received: unknown, validator: (value: unknown) => value is T) {\n      const pass = validator(received)\n\n      return {\n        message: () => pass\n          ? 'Expected structured output validator to fail.'\n          : 'Expected structured output validator to pass.',\n        pass,\n      }\n    },\n\n    toSatisfyToolCallArgs(\n      received: unknown,\n      toolName: string,\n      validator: (args: unknown) => boolean,\n    ) {\n      const toolCalls = (received as ToolCallContainer | null)?.toolCalls\n\n      if (toolCalls == null) {\n        return {\n          message: () => 'Expected received value to provide toolCalls array.',\n          pass: false,\n        }\n      }\n\n      const targetCall = toolCalls.find(call => call.name === toolName)\n      if (targetCall == null) {\n        return {\n          message: () => `Expected tool call ${toolName} to exist.`,\n          pass: false,\n        }\n      }\n\n      const pass = validator(targetCall.args)\n\n      return {\n        message: () => pass\n          ? `Expected tool call args for ${toolName} to fail validation.`\n          : `Expected tool call args for ${toolName} to pass validation.`,\n        pass,\n      }\n    },\n  })\n}\n\ninterface VievalCustomMatchers {\n  /**\n   * Asserts that text includes required keywords.\n   *\n   * Example:\n   * `expect('calm answer').toMustInclude(['calm'])`\n   */\n  toMustInclude: (keywords: string | readonly string[], options?: KeywordMatcherOptions) => void\n  /**\n   * Asserts that text excludes forbidden keywords.\n   *\n   * Example:\n   * `expect('calm answer').toMustExclude(['bestmove'])`\n   */\n  toMustExclude: (keywords: string | readonly string[], options?: KeywordMatcherOptions) => void\n  /**\n   * Asserts rubric score is greater than a threshold.\n   *\n   * Example:\n   * `expect({ score: 0.91 }).toScoreRubricGreaterThan(0.8)`\n   */\n  toScoreRubricGreaterThan: (threshold: number) => void\n  /**\n   * Asserts structured output satisfies a validator.\n   *\n   * Example:\n   * `expect(value).toSatisfyStructuredOutput(isMyShape)`\n   */\n  toSatisfyStructuredOutput: <TValue>(validator: (value: unknown) => value is TValue) => void\n  /**\n   * Asserts selected tool-call args satisfy validator.\n   *\n   * Example:\n   * `expect({ toolCalls }).toSatisfyToolCallArgs('builtIn_sparkCommand', isSparkArgs)`\n   */\n  toSatisfyToolCallArgs: (toolName: string, validator: (args: unknown) => boolean) => void\n}\n\n/* eslint-disable unused-imports/no-unused-vars */\ndeclare module '@vitest/expect' {\n  interface Matchers<T = any> extends VievalCustomMatchers {}\n  interface Assertion<T = any> extends VievalCustomMatchers {}\n}\n\ndeclare module 'vitest' {\n  interface Assertion extends VievalCustomMatchers {}\n  interface Matchers<T = any> extends VievalCustomMatchers {}\n}\n/* eslint-enable unused-imports/no-unused-vars */\n"],"mappings":";;;AAgBA,IAAI,oBAAoB;AACxB,IAAI;;;;;;;;;;;;;;AAeJ,SAAS,sCAA4C;AACnD,KAAI,kBACF;AAGF,MAAK,IAAI,WAAW;AACpB,MAAK,IAAI,eAAe;AACxB,MAAK,IAAI,oBAAoB;AAC7B,MAAK,IAAI,uBAAuB;AAChC,qBAAoB;;;;;;;;;;;;;;;;AAiBtB,SAAS,sBAAoC;AAC3C,sCAAqC;CAErC,MAAM,kBAAkB,OAAgB,YAAqB;AAE3D,WAAS,EAAE,gBADU,SAAS,cAAc,CACJ,iBAAiB,GAAG,EAAE,cAAc;AAC5E,SAAO,KAAK,OAAO,OAAO,QAAQ;;AAGpC,QAAO,OAAO,eAAe,KAAK,OAAO;AACzC,QAAO,OAAO,eAAgB,WAA4C,4BAAsC;AAEhH,eAAc,iBAAiB,SAAS,cAAc;AACtD,eAAc,YAAY,UAAiC,SAAS,OAAO,cAAc;AACzF,eAAc,SAAS,KAAK;CAM5B,MAAM,uBAAuB,KAAK;AAGlC,eAAc,UAAU,aAA6B,qBAAqB,OAAO,eAAe,SAAS;AACzG,eAAc,sBAAsB,kBAA4B,yBAAyB,cAAc;AACvG,eAAc,eAAe,YAAqB;AAChD,OAAK,OAAO,KAAK,WAAW,UAAU,KAAK,QAAQ,MAAM,IAAI,mBAAmB;;AAGlF,eAAc,SAAS;EACrB,gBAAgB;EAChB,iBAAiB;EACjB,0BAA0B;EAC1B,kCAAkC;EAClC,uBAAuB;EACvB,4BAA4B;EAC7B,CAAC;AAEF,eAAc,OAAO,eAAe;AAEpC,QAAO;;;;;;;;;;;;;;;AAgBT,SAAgB,mBAAiC;AAC/C,KAAI,yBAAyB,KAC3B,QAAO;AAGT,yBAAwB,qBAAqB;AAC7C,QAAO,eAAe,YAAY,eAAe;EAC/C,cAAc;EACd,OAAO;EACP,UAAU;EACX,CAAC;AAEF,QAAO;;;;AC7FT,SAAS,eAAe,UAAyD;AAC/E,KAAI,OAAO,aAAa,SACtB,QAAO,CAAC,SAAS;AAGnB,QAAO;;;;;;;;;;;;;;;;AAiBT,SAAgB,8BAAoC;AACnC,mBAAkB,CAE1B,OAAO;EACZ,cAAc,UAAmB,UAAsC,UAAiC,EAAE,EAAE;GAC1G,MAAM,cAAc,eAAe,SAAS;AAE5C,OAAI,OAAO,aAAa,SACtB,QAAO;IACL,eAAe;IACf,MAAM;IACP;GAGH,MAAM,iBAAiB,mBAAmB,UAAU,QAAQ,iBAAiB,MAAM;GACnF,MAAM,mBAAmB,YAAY,QAAQ,YAAY;AACvD,WAAO,eAAe,SAAS,mBAAmB,SAAS,QAAQ,iBAAiB,MAAM,CAAC;KAC3F;GAEF,MAAM,OAAO,iBAAiB,WAAW;AAEzC,UAAO;IACL,eAAe;AACb,SAAI,KACF,QAAO,gDAAgD,YAAY,KAAK,KAAK;AAG/E,YAAO,iEAAiE,iBAAiB,KAAK,KAAK;;IAErG;IACD;;EAGH,cAAc,UAAmB,UAAsC,UAAiC,EAAE,EAAE;GAC1G,MAAM,cAAc,eAAe,SAAS;AAE5C,OAAI,OAAO,aAAa,SACtB,QAAO;IACL,eAAe;IACf,MAAM;IACP;GAGH,MAAM,iBAAiB,mBAAmB,UAAU,QAAQ,iBAAiB,MAAM;GACnF,MAAM,UAAU,YAAY,QAAQ,YAAY;AAC9C,WAAO,eAAe,SAAS,mBAAmB,SAAS,QAAQ,iBAAiB,MAAM,CAAC;KAC3F;GAEF,MAAM,OAAO,QAAQ,QAAQ;GAC7B,MAAM,OAAO,SAAS,QAAQ,QAAQ,WAAW,YAAY,SAAS,QAAQ,SAAS;AAEvF,UAAO;IACL,eAAe;AACb,SAAI,KACF,QAAO,8DAA8D,QAAQ,KAAK,KAAK;AAGzF,YAAO,6CAA6C,KAAK,iBAAiB,QAAQ,OAAO,GAAG,YAAY,OAAO;;IAEjH;IACD;;EAGH,yBAAyB,UAAmB,WAAmB;GAC7D,MAAM,QAAQ,OAAO,aAAa,WAC9B,WACC,UAAuC;AAE5C,OAAI,OAAO,UAAU,SACnB,QAAO;IACL,eAAe;IACf,MAAM;IACP;GAGH,MAAM,OAAO,QAAQ;AAErB,UAAO;IACL,eAAe;AACb,SAAI,KACF,QAAO,yBAAyB,MAAM,+BAA+B,UAAU;AAGjF,YAAO,yBAAyB,MAAM,sBAAsB,UAAU;;IAExE;IACD;;EAGH,0BAA6B,UAAmB,WAA2C;GACzF,MAAM,OAAO,UAAU,SAAS;AAEhC,UAAO;IACL,eAAe,OACX,kDACA;IACJ;IACD;;EAGH,sBACE,UACA,UACA,WACA;GACA,MAAM,YAAa,UAAuC;AAE1D,OAAI,aAAa,KACf,QAAO;IACL,eAAe;IACf,MAAM;IACP;GAGH,MAAM,aAAa,UAAU,MAAK,SAAQ,KAAK,SAAS,SAAS;AACjE,OAAI,cAAc,KAChB,QAAO;IACL,eAAe,sBAAsB,SAAS;IAC9C,MAAM;IACP;GAGH,MAAM,OAAO,UAAU,WAAW,KAAK;AAEvC,UAAO;IACL,eAAe,OACX,+BAA+B,SAAS,wBACxC,+BAA+B,SAAS;IAC5C;IACD;;EAEJ,CAAC"}

package/dist/expect.mjs

@@ -1,4 +1,4 @@
-import { n as getRuntimeExpect, t as installVievalExpectMatchers } from "./expect-extensions-QLXESWjn.mjs";
+import { n as getRuntimeExpect, t as installVievalExpectMatchers } from "./expect-extensions-DCSqlneN.mjs";
 //#region src/expect.ts
 let isInstalled = false;
 function ensureExpectMatchersInstalled() {