vieval 0.0.11 → 0.0.12

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

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

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

@@ -1,19 +1,6 @@
-import { K as AggregatedRunResults } from "../../../index-CwKBlCG9.mjs";
+import { K as AggregatedRunResults } from "../../../index-BLIlhiWT.mjs";
 
 //#region src/core/processors/results/policies/hybrid-threshold.d.ts
-/**
- * Violation emitted when result policies fail.
- */
-interface ResultPolicyViolation {
-  /**
-   * Stable policy id.
-   */
-  policyId: string;
-  /**
-   * Human-readable violation reason.
-   */
-  reason: string;
-}
 /**
  * Configures hybrid-threshold policy behavior.
  */
@@ -31,6 +18,19 @@ interface HybridThresholdPolicyOptions {
    */
   minProviderHybridScore?: number;
 }
+/**
+ * Violation emitted when result policies fail.
+ */
+interface ResultPolicyViolation {
+  /**
+   * Stable policy id.
+   */
+  policyId: string;
+  /**
+   * Human-readable violation reason.
+   */
+  reason: string;
+}
 //#endregion
 //#region src/core/processors/results/policies/max-failed-runs.d.ts
 /**
@@ -56,14 +56,14 @@ interface MaxFailedRunsPolicyOptions {
  * Configures result-processing policies for eval gating.
  */
 interface ProcessRunResultsOptions {
-  /**
-   * Threshold policy options.
-   */
-  threshold?: HybridThresholdPolicyOptions;
   /**
    * Hard-limit failed-run policy options.
    */
   maxFailedRuns?: MaxFailedRunsPolicyOptions;
+  /**
+   * Threshold policy options.
+   */
+  threshold?: HybridThresholdPolicyOptions;
 }
 /**
  * Final gate decision returned by result processors.

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

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

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

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