vieval 0.0.7 → 0.0.9

package/dist/config.d.mts

@@ -1,2 +1,2 @@
-import { B as resolveModelByName, C as TaskDefinition, D as TaskRunContext, E as TaskReporterHooks, O as TaskRunOutput, S as TaskConcurrencyConfig, T as TaskReporterEventPayload, _ as ScopedMatrices, a as CliOpenTelemetryReportingConfig, b as TaskCaseReporterPayload, c as EvalDefinition, d as MatrixAxisValues, f as MatrixDefinition, g as MatrixValue, h as MatrixRow, i as Awaitable, l as EvalModule, m as MatrixPrimitive, n as defineEval, o as CliReportingConfig, p as MatrixLayer, r as defineTask, s as CollectedEvalEntry, t as ConfigHookPlugin, u as EvalModuleMap, v as TaskAutoRetryDelay, w as TaskExecutionPolicy, x as TaskCaseState, y as TaskCaseReporterEndPayload, z as ModelDefinition } from "./index-5R1_k2nv.mjs";
+import { C as TaskDefinition, D as TaskRunContext, E as TaskReporterHooks, O as TaskRunOutput, R as ModelDefinition, S as TaskConcurrencyConfig, T as TaskReporterEventPayload, _ as ScopedMatrices, a as CliOpenTelemetryReportingConfig, b as TaskCaseReporterPayload, c as EvalDefinition, d as MatrixAxisValues, f as MatrixDefinition, g as MatrixValue, h as MatrixRow, i as Awaitable, l as EvalModule, m as MatrixPrimitive, n as defineEval, o as CliReportingConfig, p as MatrixLayer, r as defineTask, s as CollectedEvalEntry, t as ConfigHookPlugin, u as EvalModuleMap, v as TaskAutoRetryDelay, w as TaskExecutionPolicy, x as TaskCaseState, y as TaskCaseReporterEndPayload, z as resolveModelByName } from "./index-BkjyCInx.mjs";
 export { Awaitable, CliOpenTelemetryReportingConfig, CliReportingConfig, CollectedEvalEntry, ConfigHookPlugin, EvalDefinition, EvalModule, EvalModuleMap, MatrixAxisValues, MatrixDefinition, MatrixLayer, MatrixPrimitive, MatrixRow, MatrixValue, ModelDefinition, ScopedMatrices, TaskAutoRetryDelay, TaskCaseReporterEndPayload, TaskCaseReporterPayload, TaskCaseState, TaskConcurrencyConfig, TaskDefinition, TaskExecutionPolicy, TaskReporterEventPayload, TaskReporterHooks, TaskRunContext, TaskRunOutput, defineEval, defineTask, resolveModelByName };

package/dist/config.mjs

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

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

@@ -1,4 +1,4 @@
-import { X as RunScore, Z as RunScoreKind } from "../../index-5R1_k2nv.mjs";
+import { X as RunScoreKind, Y as RunScore } from "../../index-BkjyCInx.mjs";
 
 //#region src/core/assertions/index.d.ts
 /**

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

@@ -1,4 +1,4 @@
-import { a as requiredEnvFrom, i as envFrom, n as EnvValueType, r as RequiredEnvFromOptions, t as EnvFromOptions } from "../../env-BeHv_5mo.mjs";
+import { a as requiredEnvFrom, i as envFrom, n as EnvValueType, r as RequiredEnvFromOptions, t as EnvFromOptions } from "../../env-DfWZy_n4.mjs";
 import { createOpenAI } from "@xsai-ext/providers/create";
 
 //#region src/core/inference-executors/retry-policy.d.ts

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

@@ -1,4 +1,4 @@
-import { n as requiredEnvFrom, t as envFrom } from "../../env--94B0UtW.mjs";
+import { n as requiredEnvFrom, t as envFrom } from "../../env-nV5rVErX.mjs";
 import process from "node:process";
 import { errorMessageFrom, errorNameFrom, sleep } from "@moeru/std";
 import { createOpenAI } from "@xsai-ext/providers/create";
@@ -150,15 +150,21 @@ function createOpenAIFromEnv(source = {}, defaults = {}) {
 	const apiKeyEnvKey = source.apiKey ?? "OPENAI_API_KEY";
 	const baseURLEnvKey = source.baseURL ?? "OPENAI_BASE_URL";
 	const modelEnvKey = source.model ?? "OPENAI_MODEL";
-	const apiKey = requiredEnvFrom(env[apiKeyEnvKey] ?? defaults.apiKey, {
+	const envWithDefaults = {
+		...defaults.apiKey == null ? {} : { [apiKeyEnvKey]: defaults.apiKey },
+		...defaults.baseURL == null ? {} : { [baseURLEnvKey]: defaults.baseURL },
+		...defaults.model == null ? {} : { [modelEnvKey]: defaults.model },
+		...env
+	};
+	const apiKey = requiredEnvFrom(envWithDefaults, {
 		name: apiKeyEnvKey,
 		type: "string"
 	});
-	const model = requiredEnvFrom(env[modelEnvKey] ?? defaults.model, {
+	const model = requiredEnvFrom(envWithDefaults, {
 		name: modelEnvKey,
 		type: "string"
 	});
-	const baseURL = envFrom(env[baseURLEnvKey] ?? defaults.baseURL, {
+	const baseURL = envFrom(envWithDefaults, {
 		name: baseURLEnvKey,
 		type: "string"
 	});

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

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

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

@@ -1,4 +1,4 @@
-import { q as AggregatedRunResults } from "../../../index-5R1_k2nv.mjs";
+import { K as AggregatedRunResults } from "../../../index-BkjyCInx.mjs";
 
 //#region src/core/processors/results/policies/hybrid-threshold.d.ts
 /**

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

@@ -1,3 +1,3 @@
-import { $ as CreateRunnerScheduleOptions, A as RunScheduledTasksOptions, F as CreateTaskExecutionContextOptions, G as createRunnerRuntimeContext, H as collectEvalEntries, I as TaskExecutionContext, J as AggregatedRunSummary, K as AggregatedProviderSummary, L as TaskModelSelectionOptions, M as RunnerTaskState, N as ScheduledTaskExecutor, P as runScheduledTasks, Q as aggregateRunResults, R as createTaskExecutionContext, U as CreateVievalRunnerRuntimeContextOptions, V as asProjectRelativePath, W as RunnerRuntimeContext, X as RunScore, Y as RunResult, Z as RunScoreKind, at as ScheduledTaskMatrix, ct as CreateFilesystemTaskCacheRuntimeOptions, dt as CacheFileHandle, et as InferenceExecutor, ft as CacheFileOptions, it as ScheduledTask, j as RunnerExecutionError, lt as createFilesystemTaskCacheRuntime, mt as TaskCacheRuntime, nt as RunnerMatrixInput, ot as ScheduledTaskMatrixMeta, pt as CacheNamespace, q as AggregatedRunResults, rt as RunnerMatrixSelection, st as createRunnerSchedule, tt as RunnerMatrixDefinition, ut as normalizeCacheFilePathSegments } from "../../index-5R1_k2nv.mjs";
+import { $ as InferenceExecutor, A as RunScheduledTasksOptions, B as asProjectRelativePath, F as CreateTaskExecutionContextOptions, G as AggregatedProviderSummary, H as CreateVievalRunnerRuntimeContextOptions, I as TaskExecutionContext, J as RunResult, K as AggregatedRunResults, L as createTaskExecutionContext, M as RunnerTaskState, N as ScheduledTaskExecutor, P as runScheduledTasks, Q as CreateRunnerScheduleOptions, U as RunnerRuntimeContext, V as collectEvalEntries, W as createRunnerRuntimeContext, X as RunScoreKind, Y as RunScore, Z as aggregateRunResults, at as ScheduledTaskMatrixMeta, ct as createFilesystemTaskCacheRuntime, dt as CacheFileOptions, et as RunnerMatrixDefinition, ft as CacheNamespace, it as ScheduledTaskMatrix, j as RunnerExecutionError, lt as normalizeCacheFilePathSegments, nt as RunnerMatrixSelection, ot as createRunnerSchedule, pt as TaskCacheRuntime, q as AggregatedRunSummary, rt as ScheduledTask, st as CreateFilesystemTaskCacheRuntimeOptions, tt as RunnerMatrixInput, ut as CacheFileHandle } from "../../index-BkjyCInx.mjs";
 import { a as SchedulerMiddleware, c as SchedulerScopeContext, i as SchedulerConcurrencyConfig, n as getActiveScopes, o as SchedulerRuntime, r as CreateSchedulerRuntimeOptions, s as SchedulerScope, t as createSchedulerRuntime } from "../../index-fakXoZEe.mjs";
-export { AggregatedProviderSummary, AggregatedRunResults, AggregatedRunSummary, CacheFileHandle, CacheFileOptions, CacheNamespace, CreateFilesystemTaskCacheRuntimeOptions, CreateRunnerScheduleOptions, CreateSchedulerRuntimeOptions, CreateTaskExecutionContextOptions, CreateVievalRunnerRuntimeContextOptions, InferenceExecutor, RunResult, RunScheduledTasksOptions, RunScore, RunScoreKind, RunnerExecutionError, RunnerMatrixDefinition, RunnerMatrixInput, RunnerMatrixSelection, RunnerRuntimeContext, RunnerTaskState, ScheduledTask, ScheduledTaskExecutor, ScheduledTaskMatrix, ScheduledTaskMatrixMeta, SchedulerConcurrencyConfig, SchedulerMiddleware, SchedulerRuntime, SchedulerScope, SchedulerScopeContext, TaskCacheRuntime, TaskExecutionContext, TaskModelSelectionOptions, aggregateRunResults, asProjectRelativePath, collectEvalEntries, createFilesystemTaskCacheRuntime, createRunnerRuntimeContext, createRunnerSchedule, createSchedulerRuntime, createTaskExecutionContext, getActiveScopes, normalizeCacheFilePathSegments, runScheduledTasks };
+export { AggregatedProviderSummary, AggregatedRunResults, AggregatedRunSummary, CacheFileHandle, CacheFileOptions, CacheNamespace, CreateFilesystemTaskCacheRuntimeOptions, CreateRunnerScheduleOptions, CreateSchedulerRuntimeOptions, CreateTaskExecutionContextOptions, CreateVievalRunnerRuntimeContextOptions, InferenceExecutor, RunResult, RunScheduledTasksOptions, RunScore, RunScoreKind, RunnerExecutionError, RunnerMatrixDefinition, RunnerMatrixInput, RunnerMatrixSelection, RunnerRuntimeContext, RunnerTaskState, ScheduledTask, ScheduledTaskExecutor, ScheduledTaskMatrix, ScheduledTaskMatrixMeta, SchedulerConcurrencyConfig, SchedulerMiddleware, SchedulerRuntime, SchedulerScope, SchedulerScopeContext, TaskCacheRuntime, TaskExecutionContext, aggregateRunResults, asProjectRelativePath, collectEvalEntries, createFilesystemTaskCacheRuntime, createRunnerRuntimeContext, createRunnerSchedule, createSchedulerRuntime, createTaskExecutionContext, getActiveScopes, normalizeCacheFilePathSegments, runScheduledTasks };

package/dist/core/runner/index.mjs

@@ -1,5 +1,4 @@
 import { createSchedulerRuntime, getActiveScopes } from "../scheduler/index.mjs";
-import { t as resolveModelByName } from "../../models-DIGdOUpJ.mjs";
 import { createRequire } from "node:module";
 import process from "node:process";
 import { errorMessageFrom } from "@moeru/std";
@@ -304,7 +303,7 @@ function collectEvalEntries(modules, context) {
 }
 //#endregion
 //#region src/core/runner/run.ts
-function createDefaultExecutionContext(task) {
+function createDefaultExecutionContext() {
 	return {
 		cache: { namespace(name) {
 			return { file(options) {
@@ -312,11 +311,7 @@ function createDefaultExecutionContext(task) {
 				throw new Error(`Task cache runtime is not configured. Requested namespace "${name}" and key "${key}".`);
 			} };
 		} },
-		model(options) {
-			const requestedModelName = typeof options === "string" ? options : options?.name;
-			if (requestedModelName != null) throw new Error(`No model registry configured. Requested model: ${requestedModelName}`);
-			throw new Error(`No model registry configured for task inferenceExecutor id "${task.inferenceExecutor.id}".`);
-		}
+		models: []
 	};
 }
 /**
@@ -366,7 +361,7 @@ async function runScheduledTasks(tasks, executor, options = {}) {
 	async function executeScheduledTask(task) {
 		let executionContext;
 		try {
-			executionContext = options.createExecutionContext?.(task) ?? createDefaultExecutionContext(task);
+			executionContext = options.createExecutionContext?.(task) ?? createDefaultExecutionContext();
 		} catch (error) {
 			throw createRunnerExecutionError(task.id, error);
 		}
@@ -587,48 +582,19 @@ function createNoopTaskCacheRuntime() {
 		} };
 	} };
 }
-function resolveDefaultTaskModel(models, task) {
-	const runMatrixModelName = task.matrix.run.model;
-	if (runMatrixModelName != null) {
-		const matrixSelectedModel = resolveModelByName(models, runMatrixModelName);
-		if (matrixSelectedModel != null) return matrixSelectedModel;
-		throw new Error(`Unknown configured model "${runMatrixModelName}" from task.matrix.run.model.`);
-	}
-	const matched = resolveModelByName(models, task.inferenceExecutor.id);
-	if (matched != null) return matched;
-	if (models.length > 1) throw new Error([
-		`Multiple configured models are available, but no default model is selected for inferenceExecutor "${task.inferenceExecutor.id}".`,
-		"Select one model explicitly by either:",
-		"- setting runMatrix.override.model (or task matrix run.model)",
-		"- setting project.inferenceExecutors to a matching model id",
-		"- calling context.model({ name: \"your-model-id-or-alias\" })"
-	].join("\n"));
-	if (models.length === 1) {
-		const firstModel = models[0];
-		if (firstModel != null) return firstModel;
-	}
-	throw new Error(`No configured model found for inferenceExecutor id "${task.inferenceExecutor.id}".`);
-}
 /**
-* Creates task-scoped model resolver context for runner execution.
+* Creates task-scoped context data for runner execution.
 *
 * Call stack:
 *
 * {@link runScheduledTasks}
 *   -> {@link createTaskExecutionContext}
-*     -> {@link resolveModelByName}
-*       -> `task.model()` / `task.model({ name })`
+*     -> `TaskExecutionContext`
 */
 function createTaskExecutionContext(options) {
 	return {
 		cache: options.cache ?? createNoopTaskCacheRuntime(),
-		model(selection) {
-			if (selection == null) return resolveDefaultTaskModel(options.models, options.task);
-			const name = typeof selection === "string" ? selection : selection.name;
-			const namedModel = resolveModelByName(options.models, name);
-			if (namedModel == null) throw new Error(`Unknown configured model "${name}".`);
-			return namedModel;
-		}
+		models: options.models
 	};
 }
 //#endregion

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 models, cache, or other task-scoped data\n   */\n  createExecutionContext?: (task: ScheduledTask) => TaskExecutionContext\n  /**\n   * Runs before the executor starts handling a task.\n   *\n   * Use when:\n   * - callers want to observe task activation before execution begins\n   *\n   * Expects:\n   * - thrown errors abort the task before executor work starts\n   */\n  onTaskStart?: (task: ScheduledTask) => void\n  /**\n   * Runs after the executor settles for a task.\n   *\n   * Use when:\n   * - callers want to observe successful and failed task completion\n   *\n   * Expects:\n   * - thrown errors abort successful runs\n   * - failed-task observers do not override the executor error for the task\n   */\n  onTaskEnd?: (task: ScheduledTask, state: RunnerTaskState) => void\n  /**\n   * Maximum number of tasks to execute concurrently.\n   *\n   * @default 1\n   */\n  maxConcurrency?: number\n}\n\nfunction createDefaultExecutionContext(): TaskExecutionContext {\n  const cache: TaskCacheRuntime = {\n    namespace(name) {\n      return {\n        file(options) {\n          const key = options.key.join('/')\n          throw new Error(`Task cache runtime is not configured. Requested namespace \"${name}\" and key \"${key}\".`)\n        },\n      }\n    },\n  }\n\n  return {\n    cache,\n    models: [],\n  }\n}\n\n/**\n * Error thrown when a scheduled run fails before producing a normalized result.\n */\nexport class RunnerExecutionError extends Error {\n  /**\n   * Stable task id that failed.\n   */\n  taskId: string\n\n  constructor(taskId: string, cause: unknown) {\n    const message = errorMessageFrom(cause) ?? 'Unknown runner execution failure.'\n    super(`Runner task \"${taskId}\" failed: ${message}`)\n    this.name = 'RunnerExecutionError'\n    this.taskId = taskId\n    this.cause = cause\n  }\n}\n\nfunction createRunnerExecutionError(taskId: string, cause: unknown): RunnerExecutionError {\n  if (cause instanceof RunnerExecutionError && cause.taskId === taskId) {\n    return cause\n  }\n\n  return new RunnerExecutionError(taskId, cause)\n}\n\n/**\n * Executes runner tasks sequentially and aggregates the normalized results.\n *\n * Call stack:\n *\n * {@link createRunnerSchedule}\n *   -> {@link runScheduledTasks}\n *     -> `executor(task)`\n *       -> {@link aggregateRunResults}\n *\n * Use when:\n * - the caller already expanded the runner matrix\n * - task execution should stay deterministic and easy to debug\n *\n * Expects:\n * - `executor` to return normalized `0..1` scores\n * - callers to handle concurrency outside this helper when needed\n * - `onTaskStart` / `onTaskEnd` hooks to be synchronous lifecycle observers\n *\n * Throws:\n * - `RunnerExecutionError` when task setup, hooks, or the executor throws\n */\nexport async function runScheduledTasks(\n  tasks: readonly ScheduledTask[],\n  executor: ScheduledTaskExecutor,\n  options: RunScheduledTasksOptions = {},\n): Promise<AggregatedRunResults> {\n  if (tasks.length === 0) {\n    return aggregateRunResults([])\n  }\n\n  async function executeScheduledTask(task: ScheduledTask): Promise<RunResult> {\n    let executionContext: TaskExecutionContext\n\n    try {\n      executionContext = options.createExecutionContext?.(task) ?? createDefaultExecutionContext()\n    }\n    catch (error) {\n      throw createRunnerExecutionError(task.id, error)\n    }\n\n    try {\n      options.onTaskStart?.(task)\n    }\n    catch (error) {\n      throw createRunnerExecutionError(task.id, error)\n    }\n\n    let runResult: RunResult\n    try {\n      runResult = await executor(task, executionContext)\n    }\n    catch (error) {\n      try {\n        options.onTaskEnd?.(task, 'failed')\n      }\n      catch {\n        // Failed-task observers must not mask the task execution failure.\n      }\n      throw createRunnerExecutionError(task.id, error)\n    }\n\n    try {\n      options.onTaskEnd?.(task, 'passed')\n    }\n    catch (error) {\n      throw createRunnerExecutionError(task.id, error)\n    }\n\n    return runResult\n  }\n\n  const maxConcurrency = options.maxConcurrency ?? 1\n  if (maxConcurrency <= 1) {\n    const results: RunResult[] = []\n    for (const task of tasks) {\n      results.push(await executeScheduledTask(task))\n    }\n    return aggregateRunResults(results)\n  }\n\n  const runWithLimit = limitConcurrency(maxConcurrency)\n  const resultPairs = await Promise.all(tasks.map(async (task, index) => {\n    const result = await runWithLimit(async () => executeScheduledTask(task))\n    return { index, result }\n  }))\n\n  const sortedResults = resultPairs\n    .sort((left, right) => left.index - right.index)\n    .map(item => item.result)\n\n  return aggregateRunResults(sortedResults)\n}\n","import { createRequire } from 'node:module'\nimport { dirname } from 'node:path'\nimport { fileURLToPath } from 'node:url'\n\nconst require = createRequire(import.meta.url)\n\n/**\n * Shared runtime context used by the vieval runner.\n *\n * Use when:\n * - runner services need stable path resolution without module-level side effects\n * - call sites want deterministic control over workspace root detection\n */\nexport interface RunnerRuntimeContext {\n  /**\n   * Absolute project root directory used for path normalization.\n   */\n  projectRootDirectory: string\n}\n\n/**\n * Options used to construct the runner runtime context.\n */\nexport interface CreateVievalRunnerRuntimeContextOptions {\n  /**\n   * Directory used to search for the nearest pnpm workspace.\n   *\n   * @default directory of this module file\n   */\n  cwd?: string\n  /**\n   * Absolute fallback directory when a pnpm workspace root is not found.\n   *\n   * @default package root directory (`packages/vieval`)\n   */\n  fallbackProjectRootDirectory?: string\n}\n\n/**\n * Creates a side-effect-free runtime context for runner path normalization.\n *\n * Call stack:\n *\n * {@link createRunnerRuntimeContext}\n *   -> `findWorkspaceDir(cwd)`\n *     -> `resolve projectRootDirectory`\n *       -> `{ projectRootDirectory }`\n *\n * Use when:\n * - initializing runner infrastructure before collecting eval modules\n * - tests need deterministic root resolution behavior\n */\nexport async function createRunnerRuntimeContext(\n  options: CreateVievalRunnerRuntimeContextOptions = {},\n): Promise<RunnerRuntimeContext> {\n  const cwd = options.cwd ?? dirname(fileURLToPath(import.meta.url))\n  const fallbackProjectRootDirectory = options.fallbackProjectRootDirectory\n    ?? fileURLToPath(new URL('../../../', import.meta.url))\n\n  // NOTICE:\n  // We use dynamic `require` here because `@pnpm/find-workspace-dir` is CommonJS.\n  // Keeping this load inside the factory avoids module-level initialization side effects.\n  const { findWorkspaceDir } = require('@pnpm/find-workspace-dir') as {\n    findWorkspaceDir: (currentWorkingDirectory: string) => Promise<string | undefined>\n  }\n\n  // NOTICE:\n  // Workspace discovery is required to keep collected eval ids stable when this\n  // package is moved inside different monorepo layouts.\n  const workspaceDirectory = await findWorkspaceDir(cwd)\n\n  return {\n    projectRootDirectory: workspaceDirectory ?? fallbackProjectRootDirectory,\n  }\n}\n","import type { CollectedEvalEntry, MatrixDefinition, MatrixLayer, MatrixValue } from '../../config'\n\n/**\n * Describes the inferenceExecutor target for a scheduled eval run.\n */\nexport interface InferenceExecutor {\n  /**\n   * Stable inferenceExecutor identifier such as `openai:gpt-4.1-mini`.\n   */\n  id: string\n}\n\n/**\n * Stores the selected value for each matrix axis.\n */\nexport type RunnerMatrixSelection = Record<string, string>\n\n/**\n * Stores stable row ids for one resolved scheduled task matrix.\n */\nexport interface ScheduledTaskMatrixMeta {\n  /**\n   * Stable row id for the resolved run matrix selection.\n   */\n  runRowId: string\n  /**\n   * Stable row id for the resolved eval matrix selection.\n   */\n  evalRowId: string\n}\n\n/**\n * Stores the structured matrix payload for one scheduled task.\n */\nexport interface ScheduledTaskMatrix {\n  /**\n   * Runtime matrix selection visible to task code.\n   */\n  run: RunnerMatrixSelection\n  /**\n   * Eval-time matrix selection visible to task code.\n   */\n  eval: RunnerMatrixSelection\n  /**\n   * Stable row ids for both scopes.\n   */\n  meta: ScheduledTaskMatrixMeta\n}\n\n/**\n * Maps matrix axis names to the values that should be expanded.\n */\nexport type RunnerMatrixDefinition = MatrixDefinition\n\n/**\n * Accepts either flat axis definitions or one layered matrix object.\n */\nexport type RunnerMatrixInput = RunnerMatrixDefinition | MatrixLayer\n\nconst matrixLayerKeys = new Set(['disable', 'extend', 'override'])\nconst ambiguousMatrixDefinitionErrorMessage = 'Ambiguous matrix definition: cannot mix reserved layer keys (disable, extend, override) with matrix axis keys.'\n\n/**\n * Represents one fully expanded runner task.\n */\nexport interface ScheduledTask {\n  /**\n   * Stable task id derived from the entry, inferenceExecutor, and matrix selection.\n   */\n  id: string\n  /**\n   * The collected eval entry to execute.\n   */\n  entry: CollectedEvalEntry\n  /**\n   * The inferenceExecutor selected for this task.\n   */\n  inferenceExecutor: InferenceExecutor\n  /**\n   * The concrete scoped matrix selection for this task.\n   */\n  matrix: ScheduledTaskMatrix\n}\n\n/**\n * Configures how the runner should expand its execution matrix.\n */\nexport interface CreateRunnerScheduleOptions {\n  /**\n   * Collected eval entries that should be scheduled.\n   */\n  entries: readonly CollectedEvalEntry[]\n  /**\n   * Providers that should run each entry.\n   */\n  inferenceExecutors: readonly InferenceExecutor[]\n  /**\n   * Optional run-time matrix axes expanded as a cartesian product.\n   */\n  runMatrix?: RunnerMatrixInput\n  /**\n   * Optional eval-time matrix axes expanded as a cartesian product.\n   */\n  evalMatrix?: RunnerMatrixInput\n}\n\nfunction encodeTaskIdSegment(value: string): string {\n  return encodeURIComponent(value)\n}\n\nfunction stringifyMatrixValue(value: MatrixValue): string {\n  return String(value)\n}\n\nfunction cloneMatrixSelection(matrix: RunnerMatrixSelection): RunnerMatrixSelection {\n  return { ...matrix }\n}\n\nfunction createScheduledTaskMatrix(\n  runMatrix: RunnerMatrixSelection,\n  evalMatrix: RunnerMatrixSelection,\n): ScheduledTaskMatrix {\n  return {\n    eval: cloneMatrixSelection(evalMatrix),\n    meta: {\n      evalRowId: createStableRowId(evalMatrix),\n      runRowId: createStableRowId(runMatrix),\n    },\n    run: cloneMatrixSelection(runMatrix),\n  }\n}\n\nfunction isMatrixLayer(matrix: RunnerMatrixInput): matrix is MatrixLayer {\n  const matrixKeys = Object.keys(matrix)\n  return (\n    matrixKeys.length > 0\n    && matrixKeys.every(key => matrixLayerKeys.has(key))\n  )\n}\n\nfunction assertNonAmbiguousMatrixDefinition(matrix: RunnerMatrixInput): void {\n  const matrixKeys = Object.keys(matrix)\n  const hasReservedKeys = matrixKeys.some(key => matrixLayerKeys.has(key))\n  const hasAxisKeys = matrixKeys.some(key => !matrixLayerKeys.has(key))\n\n  if (hasReservedKeys && hasAxisKeys) {\n    throw new TypeError(ambiguousMatrixDefinitionErrorMessage)\n  }\n}\n\nfunction normalizeLayerInputToAxes(matrix: RunnerMatrixInput | undefined): MatrixLayer | undefined {\n  if (matrix == null) {\n    return undefined\n  }\n\n  assertNonAmbiguousMatrixDefinition(matrix)\n\n  if (isMatrixLayer(matrix)) {\n    return matrix\n  }\n\n  return {\n    extend: matrix,\n  }\n}\n\nfunction dedupeAxisValues(values: readonly MatrixValue[]): string[] {\n  return Array.from(new Set(values.map(stringifyMatrixValue)))\n}\n\nfunction applyAxisValues(\n  axes: Map<string, string[]>,\n  definition: RunnerMatrixDefinition | undefined,\n  mode: 'extend' | 'override',\n): void {\n  if (definition == null) {\n    return\n  }\n\n  for (const [axis, values] of Object.entries(definition)) {\n    const nextValues = dedupeAxisValues(values)\n\n    if (mode === 'extend') {\n      const existingValues = axes.get(axis) ?? []\n      axes.set(axis, Array.from(new Set([...existingValues, ...nextValues])))\n      continue\n    }\n\n    axes.set(axis, nextValues)\n  }\n}\n\nfunction applyLayer(\n  baseAxes: ReadonlyMap<string, string[]>,\n  layer: MatrixLayer | undefined,\n): Map<string, string[]> {\n  const nextAxes = new Map<string, string[]>(\n    Array.from(baseAxes.entries()).map(([axis, values]) => [axis, [...values]]),\n  )\n\n  for (const axis of layer?.disable ?? []) {\n    nextAxes.delete(axis)\n  }\n\n  applyAxisValues(nextAxes, layer?.extend, 'extend')\n  applyAxisValues(nextAxes, layer?.override, 'override')\n\n  return nextAxes\n}\n\nfunction expandAxesToRows(axes: ReadonlyMap<string, readonly string[]>): RunnerMatrixSelection[] {\n  if (axes.size === 0) {\n    return [{}]\n  }\n\n  const dimensions = Array.from(axes.entries())\n\n  let selections: RunnerMatrixSelection[] = [{}]\n\n  for (const [axis, values] of dimensions) {\n    if (values.length === 0) {\n      return []\n    }\n\n    const nextSelections: RunnerMatrixSelection[] = []\n\n    for (const selection of selections) {\n      for (const value of values) {\n        nextSelections.push({\n          ...selection,\n          [axis]: value,\n        })\n      }\n    }\n\n    selections = nextSelections\n  }\n\n  return selections\n}\n\nfunction createStableRowId(matrix: RunnerMatrixSelection): string {\n  const segments = Object.entries(matrix)\n    .sort(([leftAxis], [rightAxis]) => leftAxis.localeCompare(rightAxis))\n    .map(([axis, value]) => `${encodeTaskIdSegment(axis)}=${encodeTaskIdSegment(value)}`)\n\n  if (segments.length === 0) {\n    return 'default'\n  }\n\n  return segments.join('&')\n}\n\nfunction createTaskId(entryId: string, inferenceExecutorId: string, runRowId: string, evalRowId: string): string {\n  const encodedEntryId = encodeTaskIdSegment(entryId)\n  const encodedProviderId = encodeTaskIdSegment(inferenceExecutorId)\n\n  return [\n    encodedEntryId,\n    encodedProviderId,\n    `run=${encodeTaskIdSegment(runRowId)}`,\n    `eval=${encodeTaskIdSegment(evalRowId)}`,\n  ].join('::')\n}\n\nfunction createResolvedRunAxes(\n  entry: CollectedEvalEntry,\n  runMatrix: RunnerMatrixInput | undefined,\n): Map<string, string[]> {\n  let resolvedAxes = new Map<string, string[]>()\n\n  for (const layerInput of [\n    runMatrix,\n    entry.matrix?.runMatrix,\n    entry.task?.matrix?.runMatrix,\n  ]) {\n    resolvedAxes = applyLayer(resolvedAxes, normalizeLayerInputToAxes(layerInput))\n  }\n\n  return resolvedAxes\n}\n\nfunction createResolvedEvalAxes(\n  entry: CollectedEvalEntry,\n  evalMatrix: RunnerMatrixInput | undefined,\n): Map<string, string[]> {\n  let resolvedAxes = new Map<string, string[]>()\n\n  for (const layerInput of [\n    evalMatrix,\n    entry.matrix?.evalMatrix,\n    entry.task?.matrix?.evalMatrix,\n  ]) {\n    resolvedAxes = applyLayer(resolvedAxes, normalizeLayerInputToAxes(layerInput))\n  }\n\n  return resolvedAxes\n}\n\n/**\n * Expands collected entries into a stable runner schedule.\n *\n * Call stack:\n *\n * {@link collectEvalEntries} (`../runner`)\n *   -> {@link createRunnerSchedule}\n *     -> {@link expandAxesToRows}\n *       -> {@link ScheduledTask}[]\n *\n * Use when:\n * - the runner already knows which eval entries are available\n * - each entry must run against multiple inferenceExecutors or matrix variants\n *\n * Expects:\n * - `entries` and `inferenceExecutors` to be provided in the desired execution order\n * - matrix axes to use insertion order when generating combinations\n */\nexport function createRunnerSchedule(options: CreateRunnerScheduleOptions): ScheduledTask[] {\n  if (options.entries.length === 0) {\n    return []\n  }\n\n  if (options.inferenceExecutors.length === 0) {\n    return []\n  }\n\n  const tasks: ScheduledTask[] = []\n\n  for (const entry of options.entries) {\n    const runSelections = expandAxesToRows(createResolvedRunAxes(entry, options.runMatrix))\n    const evalSelections = expandAxesToRows(createResolvedEvalAxes(entry, options.evalMatrix))\n\n    if (runSelections.length === 0 || evalSelections.length === 0) {\n      continue\n    }\n\n    for (const inferenceExecutor of options.inferenceExecutors) {\n      for (const runMatrix of runSelections) {\n        for (const evalMatrix of evalSelections) {\n          const isolatedMatrix = createScheduledTaskMatrix(runMatrix, evalMatrix)\n\n          tasks.push({\n            entry,\n            id: createTaskId(\n              entry.id,\n              inferenceExecutor.id,\n              isolatedMatrix.meta.runRowId,\n              isolatedMatrix.meta.evalRowId,\n            ),\n            matrix: isolatedMatrix,\n            inferenceExecutor,\n          })\n        }\n      }\n    }\n  }\n\n  return tasks\n}\n","import type { ModelDefinition } from '../../config/models'\nimport type { TaskCacheRuntime } from '../cache'\nimport type { ScheduledTask } from './schedule'\n\n/**\n * Task-scoped execution context exposed to runner executors.\n */\nexport interface TaskExecutionContext {\n  /**\n   * Deterministic cache runtime scoped to the current task project.\n   */\n  cache: TaskCacheRuntime\n  /**\n   * Configured model registrations available to model plugins.\n   */\n  models: readonly ModelDefinition[]\n}\n\n/**\n * Inputs used to build task execution context.\n */\nexport interface CreateTaskExecutionContextOptions {\n  cache?: TaskCacheRuntime\n  models: readonly ModelDefinition[]\n  task: ScheduledTask\n}\n\nfunction createNoopTaskCacheRuntime(): TaskCacheRuntime {\n  return {\n    namespace(name) {\n      return {\n        file(options) {\n          const key = options.key.join('/')\n          throw new Error(`Task cache runtime is not configured. Requested namespace \"${name}\" and key \"${key}\".`)\n        },\n      }\n    },\n  }\n}\n\n/**\n * Creates task-scoped context data for runner execution.\n *\n * Call stack:\n *\n * {@link runScheduledTasks}\n *   -> {@link createTaskExecutionContext}\n *     -> `TaskExecutionContext`\n */\nexport function createTaskExecutionContext(options: CreateTaskExecutionContextOptions): TaskExecutionContext {\n  return {\n    cache: options.cache ?? createNoopTaskCacheRuntime(),\n    models: options.models,\n  }\n}\n"],"mappings":";;;;;;;;;;;AA2BA,SAAS,oBAAoB,OAAuB;CAClD,MAAM,aAAa,MAAM,MAAM;AAC/B,KAAI,WAAW,WAAW,EACxB,QAAO;AAGT,QAAO,WAAW,QAAQ,aAAa,IAAI;;AAG7C,SAAS,mBAAmB,WAA+B,WAAmD;AAC5G,KAAI,aAAa,QAAQ,UAAU,SAAS,EAC1C,QAAO,UAAU,WAAW,IAAI,GAAG,UAAU,MAAM,EAAE,GAAG;AAG1D,KAAI,aAAa,QAAQ,UAAU,WAAW,EAC5C;AAGF,KAAI,cAAc,mBAChB,QAAO;AAGT,KAAI,cAAc,aAChB,QAAO;AAGT,KAAI,cAAc,YAChB,QAAO;;;;;;;;;;;AAeX,SAAgB,+BAA+B,SAAqC;CAClF,MAAM,eAAe,QAAQ,IAAI,KAAI,YAAW,oBAAoB,QAAQ,CAAC;CAC7E,MAAM,YAAY,mBAAmB,QAAQ,KAAK,QAAQ,UAAU;AAEpE,KAAI,aAAa,WAAW,EAC1B,QAAO,aAAa,OAAO,CAAC,WAAW,GAAG,CAAC,YAAY,YAAY;AAGrE,KAAI,aAAa,KACf,QAAO;CAGT,MAAM,cAAc,aAAa,MAAM,GAAG,KAAK,IAAI,GAAG,aAAa,SAAS,EAAE,CAAC;CAC/E,MAAM,OAAO,aAAa,aAAa,SAAS,MAAM;AACtD,QAAO,CAAC,GAAG,aAAa,GAAG,KAAK,GAAG,YAAY;;AAGjD,eAAe,gBAAgB,MAAc,SAAyC;CACpF,MAAM,YAAY,QAAQ,KAAK;CAC/B,MAAM,gBAAgB,GAAG,KAAK,OAAO,QAAQ,IAAI,GAAG,KAAK,KAAK,CAAC,GAAG,KAAK,QAAQ,CAAC,SAAS,GAAG,CAAC,MAAM,GAAG,GAAG;AACzG,OAAM,MAAM,WAAW,EAAE,WAAW,MAAM,CAAC;AAC3C,OAAM,UAAU,eAAe,QAAQ;AACvC,OAAM,OAAO,eAAe,KAAK;;AAGnC,SAAS,sBAAsB,MAA+B;AAC5D,QAAO;EACL;EACA,MAAM,SAAS;AACb,OAAI;AACF,UAAM,OAAO,KAAK;AAClB,WAAO;WAEH;AACJ,WAAO;;;EAGX,iBAAiB;AACf,UAAO,iBAAiB,KAAK;;EAE/B,MAAM,kBAAkB;AACtB,SAAM,MAAM,QAAQ,KAAK,EAAE,EAAE,WAAW,MAAM,CAAC;AAC/C,UAAO,kBAAkB,KAAK;;EAEhC,MAAM,aAAa;AACjB,UAAO,MAAM,SAAS,KAAK;;EAE7B,MAAM,YAAY,OAAO;AACvB,SAAM,gBAAgB,MAAM,MAAM;;EAEpC,MAAM,SAAS,WAAW,SAAS;AACjC,UAAO,MAAM,SAAS,MAAM,SAAS;;EAEvC,MAAM,UAAU,OAAO,WAAW,SAAS;AACzC,SAAM,gBAAgB,MAAM,OAAO,KAAK,OAAO,SAAS,CAAC;;EAE3D,MAAM,WAAc;AAClB,UAAO,KAAK,MAAM,MAAM,SAAS,MAAM,QAAQ,CAAC;;EAElD,MAAM,UAAU,OAAO;AACrB,SAAM,gBAAgB,MAAM,GAAG,KAAK,UAAU,OAAO,MAAM,EAAE,CAAC,IAAI;;EAEpE,MAAM,mBAAsB;AAC1B,UAAO,MAAM,KAAK,UAAe;;EAEnC,MAAM,sBAAyB;AAC7B,UAAO,MAAM,KAAK,UAAa;;EAElC;;AAGH,SAAS,qBAAqB,eAAuB,WAAmC;AACtF,QAAO,EACL,KAAK,SAAS;EACZ,MAAM,uBAAuB,+BAA+B,QAAQ;AACpE,SAAO,sBAAsB,KAAK,eAAe,oBAAoB,UAAU,EAAE,GAAG,qBAAqB,CAAC;IAE7G;;;;;;;;;;;;;;;;;AAkBH,SAAgB,iCACd,SACkB;CAClB,MAAM,qBAAqB,oBAAoB,QAAQ,YAAY;CACnE,MAAM,mBAAmB,oBAAoB,QAAQ,YAAY;CACjE,MAAM,gBAAgB,KAAK,QAAQ,oBAAoB,oBAAoB,iBAAiB;AAE5F,QAAO,EACL,UAAU,MAAM;AACd,SAAO,qBAAqB,eAAe,KAAK;IAEnD;;;;ACvCH,SAAS,yBAAyB,QAAkD;AAClF,QAAO;EACL,MAAM,EACJ,GAAG,OAAO,MACX;EACD,MAAM,EACJ,GAAG,OAAO,MACX;EACD,KAAK,EACH,GAAG,OAAO,KACX;EACF;;AAGH,SAAS,qBAAqB,MAA4B;AACxD,KAAI,SAAS,WAAW,SAAS,QAC/B,QAAO;AAGT,OAAM,IAAI,UAAU,4BAA4B,KAAK,IAAI;;AAG3D,SAAS,QAAQ,QAA0C;AACzD,KAAI,OAAO,WAAW,EACpB,QAAO;AAIT,QADc,OAAO,QAAQ,KAAK,UAAU,MAAM,OAAO,EAAE,GAC5C,OAAO;;AAGxB,SAAS,oBAAoB,cAA6B,cAA4C;AACpG,KAAI,gBAAgB,QAAQ,gBAAgB,KAC1C,SAAQ,eAAe,gBAAgB;AAGzC,KAAI,gBAAgB,KAClB,QAAO;AAGT,KAAI,gBAAgB,KAClB,QAAO;AAGT,QAAO;;AAGT,SAAS,oBAAoB,QAA2C;CACtE,MAAM,UAAwB;EAC5B,OAAO,EAAE;EACT,OAAO,EAAE;EACV;AAED,MAAK,MAAM,SAAS,QAAQ;AAG1B,MAFa,qBAAqB,MAAM,KAAK,KAEhC,SAAS;AACpB,WAAQ,MAAM,KAAK,MAAM,MAAM;AAC/B;;AAGF,UAAQ,MAAM,KAAK,MAAM,MAAM;;AAGjC,QAAO;;AAGT,SAAS,iBAAiB,QAAyC;CACjE,MAAM,UAAU,oBAAoB,OAAO,OAAO;CAClD,MAAM,eAAe,QAAQ,QAAQ,MAAM;CAC3C,MAAM,eAAe,QAAQ,QAAQ,MAAM;AAE3C,QAAO;EACL,SAAS,OAAO;EAChB;EACA,eAAe,oBAAoB,cAAc,aAAa;EAC9D,IAAI,OAAO;EACX;EACA,QAAQ,yBAAyB,OAAO,OAAO;EAC/C,qBAAqB,OAAO;EAC7B;;AAGH,SAAS,sBAAsB,qBAA6B,SAA0D;CACpH,MAAM,cAAwB,EAAE;CAChC,MAAM,cAAwB,EAAE;AAEhC,MAAK,MAAM,UAAU,SAAS;EAC5B,MAAM,UAAU,oBAAoB,OAAO,OAAO;AAClD,cAAY,KAAK,GAAG,QAAQ,MAAM;AAClC,cAAY,KAAK,GAAG,QAAQ,MAAM;;CAGpC,MAAM,eAAe,QAAQ,YAAY;CACzC,MAAM,eAAe,QAAQ,YAAY;AAEzC,QAAO;EACL;EACA,eAAe,oBAAoB,cAAc,aAAa;EAC9D;EACA;EACA,UAAU,QAAQ;EACnB;;;;;;;;;;;;;;;;;;;;;AAsBH,SAAgB,oBAAoB,SAAqD;CACvF,MAAM,OAAO,QAAQ,IAAI,iBAAiB;CAG1C,MAAM,qBADuB,MAAM,KAAK,IAAI,IAAI,QAAQ,KAAI,WAAU,OAAO,oBAAoB,CAAC,CAAC,CAEhG,KAAK,wBAAwB;AAE5B,SAAO,sBAAsB,qBADL,QAAQ,QAAO,WAAU,OAAO,wBAAwB,oBAAoB,CAClC;GAClE,CACD,MAAM,MAAM,UAAU,KAAK,oBAAoB,cAAc,MAAM,oBAAoB,CAAC;CAE3F,MAAM,UAAU,sBACd,WACA,QACD;AAED,QAAO;EACL,SAAS;GACP,cAAc,QAAQ;GACtB,eAAe,QAAQ;GACvB,cAAc,QAAQ;GACtB,UAAU,QAAQ;GACnB;EACD;EACA;EACD;;;;ACvRH,MAAM,iBAAiB;AACvB,MAAM,sBAAsB;AAE5B,SAAS,cAAc,OAAuB;AAC5C,QAAO,MAAM,WAAW,MAAM,IAAI;;;;;;;;;;;AAYpC,SAAgB,sBAAsB,UAAkB,SAAuC;CAC7F,MAAM,qBAAqB,cAAc,SAAS;CAClD,MAAM,iCAAiC,cAAc,QAAQ,qBAAqB;CAClF,MAAM,uBAAuB,mBAAmB,MAAM,aAAa,GAAG;CACtE,MAAM,0BAA0B,+BAA+B,MAAM,aAAa,GAAG;AAErF,KAAI,wBAAwB,QAAQ,2BAA2B,KAC7D,QAAO;AAGT,KACE,wBAAwB,QACrB,2BAA2B,QAC3B,qBAAqB,aAAa,KAAK,wBAAwB,aAAa,CAE/E,QAAO;CAGT,MAAM,uBAAuB,QAAQ;CACrC,MAAM,mBAAmB,cAAc,SAAS,sBAAsB,SAAS,CAAC;AAEhF,KAAI,CAAC,oBAAoB,KAAK,iBAAiB,EAAE;AAC/C,MAAI,qBAAqB,KACvB,QAAO,cAAc,SAAS;AAGhC,MAAI,CAAC,iBAAiB,WAAW,MAAM,CACrC,QAAO;;AAIX,QAAO,cAAc,SAAS;;AAGhC,SAAS,sBAAsB,YAAmC;AAChE,KAAI,CAAC,WAAW,WAAW,QAAQ,CACjC,QAAO;AAGT,KAAI;AACF,SAAO,cAAc,WAAW;SAE5B;AACJ,SAAO;;;AAIX,SAAS,yBACP,YACA,kBACA,SAC2B;CAC3B,MAAM,WAAW,sBAAsB,WAAW;AAElD,KAAI,CAAC,SACH,QAAO;CAGT,MAAM,mBAAmB,sBAAsB,UAAU,QAAQ;AAEjE,KAAI,CAAC,iBAAiB,SAAS,eAAe,CAC5C,QAAO;CAGT,MAAM,YAAY,SAAS,kBAAkB,eAAe;AAE5D,KAAI,UAAU,WAAW,EACvB,QAAO;CAGT,MAAM,oBAAoB,QAAQ,iBAAiB;CACnD,MAAM,YAAY,sBAAsB,MAAM,KAAK;AAEnD,QAAO;EACL,GAAG,iBAAiB;EACpB;EACA;EACA,IAAI,UAAU,WAAW,IAAI,YAAY,GAAG,UAAU,GAAG;EACzD,MAAM;EACP;;;;;;;;;;;;;;;;AAiBH,SAAgB,mBACd,SACA,SACsB;AACtB,QAAO,OAAO,QAAQ,QAAQ,CAC3B,SAAS,CAAC,YAAY,sBAAsB;EAC3C,MAAM,QAAQ,yBAAyB,YAAY,kBAAkB,QAAQ;AAE7E,MAAI,CAAC,MACH,QAAO,EAAE;AAGX,SAAO,CAAC,MAAM;GACd,CACD,MAAM,MAAM,UAAU,KAAK,GAAG,cAAc,MAAM,GAAG,CAAC;;;;AC9C3D,SAAS,gCAAsD;AAY7D,QAAO;EACL,OAZ8B,EAC9B,UAAU,MAAM;AACd,UAAO,EACL,KAAK,SAAS;IACZ,MAAM,MAAM,QAAQ,IAAI,KAAK,IAAI;AACjC,UAAM,IAAI,MAAM,8DAA8D,KAAK,aAAa,IAAI,IAAI;MAE3G;KAEJ;EAIC,QAAQ,EAAE;EACX;;;;;AAMH,IAAa,uBAAb,cAA0C,MAAM;;;;CAI9C;CAEA,YAAY,QAAgB,OAAgB;EAC1C,MAAM,UAAU,iBAAiB,MAAM,IAAI;AAC3C,QAAM,gBAAgB,OAAO,YAAY,UAAU;AACnD,OAAK,OAAO;AACZ,OAAK,SAAS;AACd,OAAK,QAAQ;;;AAIjB,SAAS,2BAA2B,QAAgB,OAAsC;AACxF,KAAI,iBAAiB,wBAAwB,MAAM,WAAW,OAC5D,QAAO;AAGT,QAAO,IAAI,qBAAqB,QAAQ,MAAM;;;;;;;;;;;;;;;;;;;;;;;;AAyBhD,eAAsB,kBACpB,OACA,UACA,UAAoC,EAAE,EACP;AAC/B,KAAI,MAAM,WAAW,EACnB,QAAO,oBAAoB,EAAE,CAAC;CAGhC,eAAe,qBAAqB,MAAyC;EAC3E,IAAI;AAEJ,MAAI;AACF,sBAAmB,QAAQ,yBAAyB,KAAK,IAAI,+BAA+B;WAEvF,OAAO;AACZ,SAAM,2BAA2B,KAAK,IAAI,MAAM;;AAGlD,MAAI;AACF,WAAQ,cAAc,KAAK;WAEtB,OAAO;AACZ,SAAM,2BAA2B,KAAK,IAAI,MAAM;;EAGlD,IAAI;AACJ,MAAI;AACF,eAAY,MAAM,SAAS,MAAM,iBAAiB;WAE7C,OAAO;AACZ,OAAI;AACF,YAAQ,YAAY,MAAM,SAAS;WAE/B;AAGN,SAAM,2BAA2B,KAAK,IAAI,MAAM;;AAGlD,MAAI;AACF,WAAQ,YAAY,MAAM,SAAS;WAE9B,OAAO;AACZ,SAAM,2BAA2B,KAAK,IAAI,MAAM;;AAGlD,SAAO;;CAGT,MAAM,iBAAiB,QAAQ,kBAAkB;AACjD,KAAI,kBAAkB,GAAG;EACvB,MAAM,UAAuB,EAAE;AAC/B,OAAK,MAAM,QAAQ,MACjB,SAAQ,KAAK,MAAM,qBAAqB,KAAK,CAAC;AAEhD,SAAO,oBAAoB,QAAQ;;CAGrC,MAAM,eAAe,iBAAiB,eAAe;AAUrD,QAAO,qBATa,MAAM,QAAQ,IAAI,MAAM,IAAI,OAAO,MAAM,UAAU;AAErE,SAAO;GAAE;GAAO,QADD,MAAM,aAAa,YAAY,qBAAqB,KAAK,CAAC;GACjD;GACxB,CAAC,EAGA,MAAM,MAAM,UAAU,KAAK,QAAQ,MAAM,MAAM,CAC/C,KAAI,SAAQ,KAAK,OAAO,CAEc;;;;ACzN3C,MAAM,UAAU,cAAc,OAAO,KAAK,IAAI;;;;;;;;;;;;;;;AAgD9C,eAAsB,2BACpB,UAAmD,EAAE,EACtB;CAC/B,MAAM,MAAM,QAAQ,OAAO,QAAQ,cAAc,OAAO,KAAK,IAAI,CAAC;CAClE,MAAM,+BAA+B,QAAQ,gCACxC,cAAc,IAAI,IAAI,aAAa,OAAO,KAAK,IAAI,CAAC;CAKzD,MAAM,EAAE,qBAAqB,QAAQ,2BAA2B;AAShE,QAAO,EACL,sBAHyB,MAAM,iBAAiB,IAAI,IAGR,8BAC7C;;;;ACdH,MAAM,kBAAkB,IAAI,IAAI;CAAC;CAAW;CAAU;CAAW,CAAC;AAClE,MAAM,wCAAwC;AA8C9C,SAAS,oBAAoB,OAAuB;AAClD,QAAO,mBAAmB,MAAM;;AAGlC,SAAS,qBAAqB,OAA4B;AACxD,QAAO,OAAO,MAAM;;AAGtB,SAAS,qBAAqB,QAAsD;AAClF,QAAO,EAAE,GAAG,QAAQ;;AAGtB,SAAS,0BACP,WACA,YACqB;AACrB,QAAO;EACL,MAAM,qBAAqB,WAAW;EACtC,MAAM;GACJ,WAAW,kBAAkB,WAAW;GACxC,UAAU,kBAAkB,UAAU;GACvC;EACD,KAAK,qBAAqB,UAAU;EACrC;;AAGH,SAAS,cAAc,QAAkD;CACvE,MAAM,aAAa,OAAO,KAAK,OAAO;AACtC,QACE,WAAW,SAAS,KACjB,WAAW,OAAM,QAAO,gBAAgB,IAAI,IAAI,CAAC;;AAIxD,SAAS,mCAAmC,QAAiC;CAC3E,MAAM,aAAa,OAAO,KAAK,OAAO;CACtC,MAAM,kBAAkB,WAAW,MAAK,QAAO,gBAAgB,IAAI,IAAI,CAAC;CACxE,MAAM,cAAc,WAAW,MAAK,QAAO,CAAC,gBAAgB,IAAI,IAAI,CAAC;AAErE,KAAI,mBAAmB,YACrB,OAAM,IAAI,UAAU,sCAAsC;;AAI9D,SAAS,0BAA0B,QAAgE;AACjG,KAAI,UAAU,KACZ;AAGF,oCAAmC,OAAO;AAE1C,KAAI,cAAc,OAAO,CACvB,QAAO;AAGT,QAAO,EACL,QAAQ,QACT;;AAGH,SAAS,iBAAiB,QAA0C;AAClE,QAAO,MAAM,KAAK,IAAI,IAAI,OAAO,IAAI,qBAAqB,CAAC,CAAC;;AAG9D,SAAS,gBACP,MACA,YACA,MACM;AACN,KAAI,cAAc,KAChB;AAGF,MAAK,MAAM,CAAC,MAAM,WAAW,OAAO,QAAQ,WAAW,EAAE;EACvD,MAAM,aAAa,iBAAiB,OAAO;AAE3C,MAAI,SAAS,UAAU;GACrB,MAAM,iBAAiB,KAAK,IAAI,KAAK,IAAI,EAAE;AAC3C,QAAK,IAAI,MAAM,MAAM,KAAK,IAAI,IAAI,CAAC,GAAG,gBAAgB,GAAG,WAAW,CAAC,CAAC,CAAC;AACvE;;AAGF,OAAK,IAAI,MAAM,WAAW;;;AAI9B,SAAS,WACP,UACA,OACuB;CACvB,MAAM,WAAW,IAAI,IACnB,MAAM,KAAK,SAAS,SAAS,CAAC,CAAC,KAAK,CAAC,MAAM,YAAY,CAAC,MAAM,CAAC,GAAG,OAAO,CAAC,CAAC,CAC5E;AAED,MAAK,MAAM,QAAQ,OAAO,WAAW,EAAE,CACrC,UAAS,OAAO,KAAK;AAGvB,iBAAgB,UAAU,OAAO,QAAQ,SAAS;AAClD,iBAAgB,UAAU,OAAO,UAAU,WAAW;AAEtD,QAAO;;AAGT,SAAS,iBAAiB,MAAuE;AAC/F,KAAI,KAAK,SAAS,EAChB,QAAO,CAAC,EAAE,CAAC;CAGb,MAAM,aAAa,MAAM,KAAK,KAAK,SAAS,CAAC;CAE7C,IAAI,aAAsC,CAAC,EAAE,CAAC;AAE9C,MAAK,MAAM,CAAC,MAAM,WAAW,YAAY;AACvC,MAAI,OAAO,WAAW,EACpB,QAAO,EAAE;EAGX,MAAM,iBAA0C,EAAE;AAElD,OAAK,MAAM,aAAa,WACtB,MAAK,MAAM,SAAS,OAClB,gBAAe,KAAK;GAClB,GAAG;IACF,OAAO;GACT,CAAC;AAIN,eAAa;;AAGf,QAAO;;AAGT,SAAS,kBAAkB,QAAuC;CAChE,MAAM,WAAW,OAAO,QAAQ,OAAO,CACpC,MAAM,CAAC,WAAW,CAAC,eAAe,SAAS,cAAc,UAAU,CAAC,CACpE,KAAK,CAAC,MAAM,WAAW,GAAG,oBAAoB,KAAK,CAAC,GAAG,oBAAoB,MAAM,GAAG;AAEvF,KAAI,SAAS,WAAW,EACtB,QAAO;AAGT,QAAO,SAAS,KAAK,IAAI;;AAG3B,SAAS,aAAa,SAAiB,qBAA6B,UAAkB,WAA2B;AAI/G,QAAO;EAHgB,oBAAoB,QAAQ;EACzB,oBAAoB,oBAAoB;EAKhE,OAAO,oBAAoB,SAAS;EACpC,QAAQ,oBAAoB,UAAU;EACvC,CAAC,KAAK,KAAK;;AAGd,SAAS,sBACP,OACA,WACuB;CACvB,IAAI,+BAAe,IAAI,KAAuB;AAE9C,MAAK,MAAM,cAAc;EACvB;EACA,MAAM,QAAQ;EACd,MAAM,MAAM,QAAQ;EACrB,CACC,gBAAe,WAAW,cAAc,0BAA0B,WAAW,CAAC;AAGhF,QAAO;;AAGT,SAAS,uBACP,OACA,YACuB;CACvB,IAAI,+BAAe,IAAI,KAAuB;AAE9C,MAAK,MAAM,cAAc;EACvB;EACA,MAAM,QAAQ;EACd,MAAM,MAAM,QAAQ;EACrB,CACC,gBAAe,WAAW,cAAc,0BAA0B,WAAW,CAAC;AAGhF,QAAO;;;;;;;;;;;;;;;;;;;;AAqBT,SAAgB,qBAAqB,SAAuD;AAC1F,KAAI,QAAQ,QAAQ,WAAW,EAC7B,QAAO,EAAE;AAGX,KAAI,QAAQ,mBAAmB,WAAW,EACxC,QAAO,EAAE;CAGX,MAAM,QAAyB,EAAE;AAEjC,MAAK,MAAM,SAAS,QAAQ,SAAS;EACnC,MAAM,gBAAgB,iBAAiB,sBAAsB,OAAO,QAAQ,UAAU,CAAC;EACvF,MAAM,iBAAiB,iBAAiB,uBAAuB,OAAO,QAAQ,WAAW,CAAC;AAE1F,MAAI,cAAc,WAAW,KAAK,eAAe,WAAW,EAC1D;AAGF,OAAK,MAAM,qBAAqB,QAAQ,mBACtC,MAAK,MAAM,aAAa,cACtB,MAAK,MAAM,cAAc,gBAAgB;GACvC,MAAM,iBAAiB,0BAA0B,WAAW,WAAW;AAEvE,SAAM,KAAK;IACT;IACA,IAAI,aACF,MAAM,IACN,kBAAkB,IAClB,eAAe,KAAK,UACpB,eAAe,KAAK,UACrB;IACD,QAAQ;IACR;IACD,CAAC;;;AAMV,QAAO;;;;AC1UT,SAAS,6BAA+C;AACtD,QAAO,EACL,UAAU,MAAM;AACd,SAAO,EACL,KAAK,SAAS;GACZ,MAAM,MAAM,QAAQ,IAAI,KAAK,IAAI;AACjC,SAAM,IAAI,MAAM,8DAA8D,KAAK,aAAa,IAAI,IAAI;KAE3G;IAEJ;;;;;;;;;;;AAYH,SAAgB,2BAA2B,SAAkE;AAC3G,QAAO;EACL,OAAO,QAAQ,SAAS,4BAA4B;EACpD,QAAQ,QAAQ;EACjB"}

package/dist/{env-BeHv_5mo.d.mts → env-DfWZy_n4.d.mts}

@@ -7,6 +7,10 @@ type EnvValueType = 'string';
  * Common options for env readers.
  */
 interface EnvFromOptions {
+  /**
+   * Env key to read and use in error messages.
+   */
+  name: string;
   /**
    * Expected env value type.
    */
@@ -17,10 +21,6 @@ interface EnvFromOptions {
    * @default false
    */
   required?: boolean;
-  /**
-   * Optional key name used for clearer error messages.
-   */
-  name?: string;
 }
 /**
  * Env options used by the required helper.
@@ -28,20 +28,25 @@ interface EnvFromOptions {
  * `required` is intentionally omitted because this helper is always required.
  */
 type RequiredEnvFromOptions = Omit<EnvFromOptions, 'required'>;
+type EnvSource = Record<string, string | undefined>;
 /**
  * Parses one env value with optional required behavior.
  *
  * Example:
- * `const apiKey = envFrom(process.env.OPENAI_API_KEY, { type: 'string', required: true, name: 'OPENAI_API_KEY' })`
+ * `const apiKey = envFrom(process.env, { type: 'string', required: true, name: 'OPENAI_API_KEY' })`
  */
-declare function envFrom(value: string | undefined, options: EnvFromOptions): string | undefined;
+declare function envFrom<TEnv extends EnvSource>(env: TEnv, options: EnvFromOptions & {
+  name: keyof TEnv & string;
+}): string | undefined;
 /**
  * Parses one required env value.
  *
  * Example:
- * `const apiKey = requiredEnvFrom(process.env.OPENAI_API_KEY, { type: 'string', name: 'OPENAI_API_KEY' })`
+ * `const apiKey = requiredEnvFrom(process.env, { type: 'string', name: 'OPENAI_API_KEY' })`
  */
-declare function requiredEnvFrom(value: string | undefined, options: RequiredEnvFromOptions): string;
+declare function requiredEnvFrom<TEnv extends EnvSource>(env: TEnv, options: RequiredEnvFromOptions & {
+  name: keyof TEnv & string;
+}): string;
 //#endregion
 export { requiredEnvFrom as a, envFrom as i, EnvValueType as n, RequiredEnvFromOptions as r, EnvFromOptions as t };
-//# sourceMappingURL=env-BeHv_5mo.d.mts.map
+//# sourceMappingURL=env-DfWZy_n4.d.mts.map

package/dist/env-nV5rVErX.mjs

@@ -0,0 +1,35 @@
+//#region src/core/inference-executors/env.ts
+function assertNonEmptyString(value, options) {
+	if (value == null || value.trim().length === 0) {
+		if (options.required === true) throw new Error(`Missing required ${options.name}.`);
+		return;
+	}
+	return value;
+}
+/**
+* Parses one env value with optional required behavior.
+*
+* Example:
+* `const apiKey = envFrom(process.env, { type: 'string', required: true, name: 'OPENAI_API_KEY' })`
+*/
+function envFrom(env, options) {
+	if (options.type === "string") return assertNonEmptyString(env[options.name], options);
+}
+/**
+* Parses one required env value.
+*
+* Example:
+* `const apiKey = requiredEnvFrom(process.env, { type: 'string', name: 'OPENAI_API_KEY' })`
+*/
+function requiredEnvFrom(env, options) {
+	const parsed = envFrom(env, {
+		...options,
+		required: true
+	});
+	if (parsed == null) throw new Error(`Missing required ${options.name}.`);
+	return parsed;
+}
+//#endregion
+export { requiredEnvFrom as n, envFrom as t };
+
+//# sourceMappingURL=env-nV5rVErX.mjs.map

package/dist/env-nV5rVErX.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"env-nV5rVErX.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   * Env key to read and use in error messages.\n   */\n  name: string\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\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\ntype EnvSource = Record<string, string | undefined>\n\nfunction assertNonEmptyString(value: string | undefined, options: EnvFromOptions): string | undefined {\n  if (value == null || value.trim().length === 0) {\n    if (options.required === true) {\n      throw new Error(`Missing required ${options.name}.`)\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, { type: 'string', required: true, name: 'OPENAI_API_KEY' })`\n */\nexport function envFrom<TEnv extends EnvSource>(\n  env: TEnv,\n  options: EnvFromOptions & { name: keyof TEnv & string },\n): string | undefined {\n  if (options.type === 'string') {\n    return assertNonEmptyString(env[options.name], options)\n  }\n\n  return undefined\n}\n\n/**\n * Parses one required env value.\n *\n * Example:\n * `const apiKey = requiredEnvFrom(process.env, { type: 'string', name: 'OPENAI_API_KEY' })`\n */\nexport function requiredEnvFrom<TEnv extends EnvSource>(\n  env: TEnv,\n  options: RequiredEnvFromOptions & { name: keyof TEnv & string },\n): string {\n  const parsed = envFrom(env, {\n    ...options,\n    required: true,\n  })\n\n  if (parsed == null) {\n    throw new Error(`Missing required ${options.name}.`)\n  }\n\n  return parsed\n}\n"],"mappings":";AAkCA,SAAS,qBAAqB,OAA2B,SAA6C;AACpG,KAAI,SAAS,QAAQ,MAAM,MAAM,CAAC,WAAW,GAAG;AAC9C,MAAI,QAAQ,aAAa,KACvB,OAAM,IAAI,MAAM,oBAAoB,QAAQ,KAAK,GAAG;AAGtD;;AAGF,QAAO;;;;;;;;AAST,SAAgB,QACd,KACA,SACoB;AACpB,KAAI,QAAQ,SAAS,SACnB,QAAO,qBAAqB,IAAI,QAAQ,OAAO,QAAQ;;;;;;;;AAY3D,SAAgB,gBACd,KACA,SACQ;CACR,MAAM,SAAS,QAAQ,KAAK;EAC1B,GAAG;EACH,UAAU;EACX,CAAC;AAEF,KAAI,UAAU,KACZ,OAAM,IAAI,MAAM,oBAAoB,QAAQ,KAAK,GAAG;AAGtD,QAAO"}