@axlsdk/eval 0.13.0 → 0.13.2

package/dist/index.js

@@ -1,7 +1,7 @@
 import {
   evalCompare,
   runEval
-} from "./chunk-7B5NFD5T.js";
+} from "./chunk-K73SNPVV.js";
 
 // src/dataset.ts
 import { readFile } from "fs/promises";
@@ -46,6 +46,7 @@ function scorer(config) {
 }
 
 // src/llm-scorer.ts
+import { extractJson } from "@axlsdk/axl";
 function llmScorer(config) {
   const scorerInstance = {
     name: config.name,
@@ -77,10 +78,10 @@ ${JSON.stringify(annotations, null, 2)}` : "",
           { role: "system", content: config.system },
           { role: "user", content: prompt }
         ],
-        { model, temperature: config.temperature ?? 0.2 }
+        { model, temperature: config.temperature ?? 0.2, responseFormat: { type: "json_object" } }
       );
       scorerInstance._lastCost = response.cost;
-      const parsed = JSON.parse(response.content);
+      const parsed = JSON.parse(extractJson(response.content));
       const validated = config.schema.parse(parsed);
       return validated.score;
     }

package/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"sources":["../src/dataset.ts","../src/scorer.ts","../src/llm-scorer.ts","../src/define-eval.ts"],"sourcesContent":["import type { z } from 'zod';\nimport { readFile } from 'node:fs/promises';\nimport * as path from 'node:path';\n\nexport type DatasetItem<TInput, TAnnotations = undefined> = {\n  input: TInput;\n  annotations?: TAnnotations;\n};\n\nexport type DatasetConfig<TInput extends z.ZodType, TAnnotations extends z.ZodType = z.ZodType> = {\n  name: string;\n  schema: TInput;\n  annotations?: TAnnotations;\n  items?: DatasetItem<z.infer<TInput>, z.infer<TAnnotations>>[];\n  file?: string;\n  /** Base directory for resolving relative file paths. Defaults to cwd. */\n  basePath?: string;\n};\n\nexport type Dataset<TInput = unknown, TAnnotations = unknown> = {\n  readonly name: string;\n  readonly schema: z.ZodType;\n  readonly annotationsSchema?: z.ZodType;\n  getItems(): Promise<DatasetItem<TInput, TAnnotations>[]>;\n};\n\nexport function dataset<TInput extends z.ZodType, TAnnotations extends z.ZodType>(\n  config: DatasetConfig<TInput, TAnnotations>,\n): Dataset<z.infer<TInput>, z.infer<TAnnotations>> {\n  return {\n    name: config.name,\n    schema: config.schema,\n    annotationsSchema: config.annotations,\n    async getItems() {\n      if (config.items && config.file) {\n        throw new Error('Dataset config error: \"items\" and \"file\" are mutually exclusive');\n      }\n      if (!config.items && !config.file) {\n        throw new Error('Dataset config error: either \"items\" or \"file\" must be provided');\n      }\n      if (config.items) {\n        return config.items.map((item) => ({\n          input: config.schema.parse(item.input),\n          annotations:\n            item.annotations && config.annotations\n              ? config.annotations.parse(item.annotations)\n              : item.annotations,\n        }));\n      }\n      const resolvedPath = path.resolve(config.basePath ?? process.cwd(), config.file!);\n      const content = await readFile(resolvedPath, 'utf-8');\n      const items = JSON.parse(content) as DatasetItem<z.infer<TInput>, z.infer<TAnnotations>>[];\n      return items.map((item) => ({\n        input: config.schema.parse(item.input),\n        annotations:\n          item.annotations && config.annotations\n            ? config.annotations.parse(item.annotations)\n            : item.annotations,\n      }));\n    },\n  };\n}\n","/** Context passed to scorers by the eval runner. */\nexport type ScorerContext = {\n  /** Resolve a provider:model URI to a provider instance and model name. */\n  resolveProvider: (modelUri: string) => {\n    provider: {\n      chat(\n        messages: { role: string; content: string }[],\n        options: { model: string; temperature?: number },\n      ): Promise<{ content: string; cost?: number }>;\n    };\n    model: string;\n  };\n};\n\nexport type ScorerFn<TOutput, TInput, TAnnotations> = (\n  output: TOutput,\n  input: TInput,\n  annotations?: TAnnotations,\n) => number;\n\nexport type ScorerConfig<TOutput = unknown, TInput = unknown, TAnnotations = unknown> = {\n  name: string;\n  description: string;\n  score: ScorerFn<TOutput, TInput, TAnnotations>;\n};\n\nexport type Scorer<TOutput = unknown, TInput = unknown, TAnnotations = unknown> = {\n  readonly name: string;\n  readonly description: string;\n  readonly isLlm: boolean;\n  score(\n    output: TOutput,\n    input: TInput,\n    annotations?: TAnnotations,\n    context?: ScorerContext,\n  ): number | Promise<number>;\n};\n\nexport function scorer<TOutput = unknown, TInput = unknown, TAnnotations = unknown>(\n  config: ScorerConfig<TOutput, TInput, TAnnotations>,\n): Scorer<TOutput, TInput, TAnnotations> {\n  return {\n    name: config.name,\n    description: config.description,\n    isLlm: false,\n    score: config.score,\n  };\n}\n","import type { z } from 'zod';\nimport type { Scorer, ScorerContext } from './scorer.js';\n\nexport type LlmScorerConfig = {\n  name: string;\n  description: string;\n  model: string;\n  system: string;\n  schema: z.ZodType<{ score: number; [key: string]: unknown }>;\n  temperature?: number;\n};\n\nexport function llmScorer(config: LlmScorerConfig): Scorer {\n  const scorerInstance: Scorer & { _lastCost?: number } = {\n    name: config.name,\n    description: config.description,\n    isLlm: true,\n    async score(\n      output: unknown,\n      input: unknown,\n      annotations?: unknown,\n      context?: ScorerContext,\n    ): Promise<number> {\n      if (!context?.resolveProvider) {\n        throw new Error(\n          `LLM scorer \"${config.name}\" has no provider. ` +\n            `Ensure you are running via runEval() with a real AxlRuntime instance.`,\n        );\n      }\n\n      const { provider, model } = context.resolveProvider(config.model);\n\n      const prompt = [\n        `Evaluate the following output.`,\n        ``,\n        `## Input`,\n        `${JSON.stringify(input, null, 2)}`,\n        annotations\n          ? `\\n## Annotations (Ground Truth)\\n${JSON.stringify(annotations, null, 2)}`\n          : '',\n        ``,\n        `## Output to Evaluate`,\n        `${JSON.stringify(output, null, 2)}`,\n        ``,\n        `Respond with valid JSON matching the required schema with a score field (0-1) and reasoning.`,\n      ].join('\\n');\n\n      const response = await provider.chat(\n        [\n          { role: 'system', content: config.system },\n          { role: 'user', content: prompt },\n        ],\n        { model, temperature: config.temperature ?? 0.2 },\n      );\n\n      scorerInstance._lastCost = response.cost;\n\n      const parsed = JSON.parse(response.content);\n      const validated = config.schema.parse(parsed);\n      return validated.score;\n    },\n  };\n\n  return scorerInstance;\n}\n","import type { EvalConfig } from './types.js';\n\nexport function defineEval(config: EvalConfig): EvalConfig {\n  return config;\n}\n"],"mappings":";;;;;;AACA,SAAS,gBAAgB;AACzB,YAAY,UAAU;AAwBf,SAAS,QACd,QACiD;AACjD,SAAO;AAAA,IACL,MAAM,OAAO;AAAA,IACb,QAAQ,OAAO;AAAA,IACf,mBAAmB,OAAO;AAAA,IAC1B,MAAM,WAAW;AACf,UAAI,OAAO,SAAS,OAAO,MAAM;AAC/B,cAAM,IAAI,MAAM,iEAAiE;AAAA,MACnF;AACA,UAAI,CAAC,OAAO,SAAS,CAAC,OAAO,MAAM;AACjC,cAAM,IAAI,MAAM,iEAAiE;AAAA,MACnF;AACA,UAAI,OAAO,OAAO;AAChB,eAAO,OAAO,MAAM,IAAI,CAAC,UAAU;AAAA,UACjC,OAAO,OAAO,OAAO,MAAM,KAAK,KAAK;AAAA,UACrC,aACE,KAAK,eAAe,OAAO,cACvB,OAAO,YAAY,MAAM,KAAK,WAAW,IACzC,KAAK;AAAA,QACb,EAAE;AAAA,MACJ;AACA,YAAM,eAAoB,aAAQ,OAAO,YAAY,QAAQ,IAAI,GAAG,OAAO,IAAK;AAChF,YAAM,UAAU,MAAM,SAAS,cAAc,OAAO;AACpD,YAAM,QAAQ,KAAK,MAAM,OAAO;AAChC,aAAO,MAAM,IAAI,CAAC,UAAU;AAAA,QAC1B,OAAO,OAAO,OAAO,MAAM,KAAK,KAAK;AAAA,QACrC,aACE,KAAK,eAAe,OAAO,cACvB,OAAO,YAAY,MAAM,KAAK,WAAW,IACzC,KAAK;AAAA,MACb,EAAE;AAAA,IACJ;AAAA,EACF;AACF;;;ACvBO,SAAS,OACd,QACuC;AACvC,SAAO;AAAA,IACL,MAAM,OAAO;AAAA,IACb,aAAa,OAAO;AAAA,IACpB,OAAO;AAAA,IACP,OAAO,OAAO;AAAA,EAChB;AACF;;;ACnCO,SAAS,UAAU,QAAiC;AACzD,QAAM,iBAAkD;AAAA,IACtD,MAAM,OAAO;AAAA,IACb,aAAa,OAAO;AAAA,IACpB,OAAO;AAAA,IACP,MAAM,MACJ,QACA,OACA,aACA,SACiB;AACjB,UAAI,CAAC,SAAS,iBAAiB;AAC7B,cAAM,IAAI;AAAA,UACR,eAAe,OAAO,IAAI;AAAA,QAE5B;AAAA,MACF;AAEA,YAAM,EAAE,UAAU,MAAM,IAAI,QAAQ,gBAAgB,OAAO,KAAK;AAEhE,YAAM,SAAS;AAAA,QACb;AAAA,QACA;AAAA,QACA;AAAA,QACA,GAAG,KAAK,UAAU,OAAO,MAAM,CAAC,CAAC;AAAA,QACjC,cACI;AAAA;AAAA,EAAoC,KAAK,UAAU,aAAa,MAAM,CAAC,CAAC,KACxE;AAAA,QACJ;AAAA,QACA;AAAA,QACA,GAAG,KAAK,UAAU,QAAQ,MAAM,CAAC,CAAC;AAAA,QAClC;AAAA,QACA;AAAA,MACF,EAAE,KAAK,IAAI;AAEX,YAAM,WAAW,MAAM,SAAS;AAAA,QAC9B;AAAA,UACE,EAAE,MAAM,UAAU,SAAS,OAAO,OAAO;AAAA,UACzC,EAAE,MAAM,QAAQ,SAAS,OAAO;AAAA,QAClC;AAAA,QACA,EAAE,OAAO,aAAa,OAAO,eAAe,IAAI;AAAA,MAClD;AAEA,qBAAe,YAAY,SAAS;AAEpC,YAAM,SAAS,KAAK,MAAM,SAAS,OAAO;AAC1C,YAAM,YAAY,OAAO,OAAO,MAAM,MAAM;AAC5C,aAAO,UAAU;AAAA,IACnB;AAAA,EACF;AAEA,SAAO;AACT;;;AC9DO,SAAS,WAAW,QAAgC;AACzD,SAAO;AACT;","names":[]}
+{"version":3,"sources":["../src/dataset.ts","../src/scorer.ts","../src/llm-scorer.ts","../src/define-eval.ts"],"sourcesContent":["import type { z } from 'zod';\nimport { readFile } from 'node:fs/promises';\nimport * as path from 'node:path';\n\nexport type DatasetItem<TInput, TAnnotations = undefined> = {\n  input: TInput;\n  annotations?: TAnnotations;\n};\n\nexport type DatasetConfig<TInput extends z.ZodType, TAnnotations extends z.ZodType = z.ZodType> = {\n  name: string;\n  schema: TInput;\n  annotations?: TAnnotations;\n  items?: DatasetItem<z.infer<TInput>, z.infer<TAnnotations>>[];\n  file?: string;\n  /** Base directory for resolving relative file paths. Defaults to cwd. */\n  basePath?: string;\n};\n\nexport type Dataset<TInput = unknown, TAnnotations = unknown> = {\n  readonly name: string;\n  readonly schema: z.ZodType;\n  readonly annotationsSchema?: z.ZodType;\n  getItems(): Promise<DatasetItem<TInput, TAnnotations>[]>;\n};\n\nexport function dataset<TInput extends z.ZodType, TAnnotations extends z.ZodType>(\n  config: DatasetConfig<TInput, TAnnotations>,\n): Dataset<z.infer<TInput>, z.infer<TAnnotations>> {\n  return {\n    name: config.name,\n    schema: config.schema,\n    annotationsSchema: config.annotations,\n    async getItems() {\n      if (config.items && config.file) {\n        throw new Error('Dataset config error: \"items\" and \"file\" are mutually exclusive');\n      }\n      if (!config.items && !config.file) {\n        throw new Error('Dataset config error: either \"items\" or \"file\" must be provided');\n      }\n      if (config.items) {\n        return config.items.map((item) => ({\n          input: config.schema.parse(item.input),\n          annotations:\n            item.annotations && config.annotations\n              ? config.annotations.parse(item.annotations)\n              : item.annotations,\n        }));\n      }\n      const resolvedPath = path.resolve(config.basePath ?? process.cwd(), config.file!);\n      const content = await readFile(resolvedPath, 'utf-8');\n      const items = JSON.parse(content) as DatasetItem<z.infer<TInput>, z.infer<TAnnotations>>[];\n      return items.map((item) => ({\n        input: config.schema.parse(item.input),\n        annotations:\n          item.annotations && config.annotations\n            ? config.annotations.parse(item.annotations)\n            : item.annotations,\n      }));\n    },\n  };\n}\n","/** Context passed to scorers by the eval runner. */\nexport type ScorerContext = {\n  /** Resolve a provider:model URI to a provider instance and model name. */\n  resolveProvider: (modelUri: string) => {\n    provider: {\n      chat(\n        messages: { role: string; content: string }[],\n        options: {\n          model: string;\n          temperature?: number;\n          responseFormat?: { type: string; json_schema?: unknown };\n        },\n      ): Promise<{ content: string; cost?: number }>;\n    };\n    model: string;\n  };\n};\n\nexport type ScorerFn<TOutput, TInput, TAnnotations> = (\n  output: TOutput,\n  input: TInput,\n  annotations?: TAnnotations,\n) => number;\n\nexport type ScorerConfig<TOutput = unknown, TInput = unknown, TAnnotations = unknown> = {\n  name: string;\n  description: string;\n  score: ScorerFn<TOutput, TInput, TAnnotations>;\n};\n\nexport type Scorer<TOutput = unknown, TInput = unknown, TAnnotations = unknown> = {\n  readonly name: string;\n  readonly description: string;\n  readonly isLlm: boolean;\n  score(\n    output: TOutput,\n    input: TInput,\n    annotations?: TAnnotations,\n    context?: ScorerContext,\n  ): number | Promise<number>;\n};\n\nexport function scorer<TOutput = unknown, TInput = unknown, TAnnotations = unknown>(\n  config: ScorerConfig<TOutput, TInput, TAnnotations>,\n): Scorer<TOutput, TInput, TAnnotations> {\n  return {\n    name: config.name,\n    description: config.description,\n    isLlm: false,\n    score: config.score,\n  };\n}\n","import type { z } from 'zod';\nimport type { Scorer, ScorerContext } from './scorer.js';\nimport { extractJson } from '@axlsdk/axl';\n\nexport type LlmScorerConfig = {\n  name: string;\n  description: string;\n  model: string;\n  system: string;\n  schema: z.ZodType<{ score: number; [key: string]: unknown }>;\n  temperature?: number;\n};\n\nexport function llmScorer(config: LlmScorerConfig): Scorer {\n  const scorerInstance: Scorer & { _lastCost?: number } = {\n    name: config.name,\n    description: config.description,\n    isLlm: true,\n    async score(\n      output: unknown,\n      input: unknown,\n      annotations?: unknown,\n      context?: ScorerContext,\n    ): Promise<number> {\n      if (!context?.resolveProvider) {\n        throw new Error(\n          `LLM scorer \"${config.name}\" has no provider. ` +\n            `Ensure you are running via runEval() with a real AxlRuntime instance.`,\n        );\n      }\n\n      const { provider, model } = context.resolveProvider(config.model);\n\n      const prompt = [\n        `Evaluate the following output.`,\n        ``,\n        `## Input`,\n        `${JSON.stringify(input, null, 2)}`,\n        annotations\n          ? `\\n## Annotations (Ground Truth)\\n${JSON.stringify(annotations, null, 2)}`\n          : '',\n        ``,\n        `## Output to Evaluate`,\n        `${JSON.stringify(output, null, 2)}`,\n        ``,\n        `Respond with valid JSON matching the required schema with a score field (0-1) and reasoning.`,\n      ].join('\\n');\n\n      const response = await provider.chat(\n        [\n          { role: 'system', content: config.system },\n          { role: 'user', content: prompt },\n        ],\n        { model, temperature: config.temperature ?? 0.2, responseFormat: { type: 'json_object' } },\n      );\n\n      scorerInstance._lastCost = response.cost;\n\n      const parsed = JSON.parse(extractJson(response.content));\n      const validated = config.schema.parse(parsed);\n      return validated.score;\n    },\n  };\n\n  return scorerInstance;\n}\n","import type { EvalConfig } from './types.js';\n\nexport function defineEval(config: EvalConfig): EvalConfig {\n  return config;\n}\n"],"mappings":";;;;;;AACA,SAAS,gBAAgB;AACzB,YAAY,UAAU;AAwBf,SAAS,QACd,QACiD;AACjD,SAAO;AAAA,IACL,MAAM,OAAO;AAAA,IACb,QAAQ,OAAO;AAAA,IACf,mBAAmB,OAAO;AAAA,IAC1B,MAAM,WAAW;AACf,UAAI,OAAO,SAAS,OAAO,MAAM;AAC/B,cAAM,IAAI,MAAM,iEAAiE;AAAA,MACnF;AACA,UAAI,CAAC,OAAO,SAAS,CAAC,OAAO,MAAM;AACjC,cAAM,IAAI,MAAM,iEAAiE;AAAA,MACnF;AACA,UAAI,OAAO,OAAO;AAChB,eAAO,OAAO,MAAM,IAAI,CAAC,UAAU;AAAA,UACjC,OAAO,OAAO,OAAO,MAAM,KAAK,KAAK;AAAA,UACrC,aACE,KAAK,eAAe,OAAO,cACvB,OAAO,YAAY,MAAM,KAAK,WAAW,IACzC,KAAK;AAAA,QACb,EAAE;AAAA,MACJ;AACA,YAAM,eAAoB,aAAQ,OAAO,YAAY,QAAQ,IAAI,GAAG,OAAO,IAAK;AAChF,YAAM,UAAU,MAAM,SAAS,cAAc,OAAO;AACpD,YAAM,QAAQ,KAAK,MAAM,OAAO;AAChC,aAAO,MAAM,IAAI,CAAC,UAAU;AAAA,QAC1B,OAAO,OAAO,OAAO,MAAM,KAAK,KAAK;AAAA,QACrC,aACE,KAAK,eAAe,OAAO,cACvB,OAAO,YAAY,MAAM,KAAK,WAAW,IACzC,KAAK;AAAA,MACb,EAAE;AAAA,IACJ;AAAA,EACF;AACF;;;ACnBO,SAAS,OACd,QACuC;AACvC,SAAO;AAAA,IACL,MAAM,OAAO;AAAA,IACb,aAAa,OAAO;AAAA,IACpB,OAAO;AAAA,IACP,OAAO,OAAO;AAAA,EAChB;AACF;;;ACjDA,SAAS,mBAAmB;AAWrB,SAAS,UAAU,QAAiC;AACzD,QAAM,iBAAkD;AAAA,IACtD,MAAM,OAAO;AAAA,IACb,aAAa,OAAO;AAAA,IACpB,OAAO;AAAA,IACP,MAAM,MACJ,QACA,OACA,aACA,SACiB;AACjB,UAAI,CAAC,SAAS,iBAAiB;AAC7B,cAAM,IAAI;AAAA,UACR,eAAe,OAAO,IAAI;AAAA,QAE5B;AAAA,MACF;AAEA,YAAM,EAAE,UAAU,MAAM,IAAI,QAAQ,gBAAgB,OAAO,KAAK;AAEhE,YAAM,SAAS;AAAA,QACb;AAAA,QACA;AAAA,QACA;AAAA,QACA,GAAG,KAAK,UAAU,OAAO,MAAM,CAAC,CAAC;AAAA,QACjC,cACI;AAAA;AAAA,EAAoC,KAAK,UAAU,aAAa,MAAM,CAAC,CAAC,KACxE;AAAA,QACJ;AAAA,QACA;AAAA,QACA,GAAG,KAAK,UAAU,QAAQ,MAAM,CAAC,CAAC;AAAA,QAClC;AAAA,QACA;AAAA,MACF,EAAE,KAAK,IAAI;AAEX,YAAM,WAAW,MAAM,SAAS;AAAA,QAC9B;AAAA,UACE,EAAE,MAAM,UAAU,SAAS,OAAO,OAAO;AAAA,UACzC,EAAE,MAAM,QAAQ,SAAS,OAAO;AAAA,QAClC;AAAA,QACA,EAAE,OAAO,aAAa,OAAO,eAAe,KAAK,gBAAgB,EAAE,MAAM,cAAc,EAAE;AAAA,MAC3F;AAEA,qBAAe,YAAY,SAAS;AAEpC,YAAM,SAAS,KAAK,MAAM,YAAY,SAAS,OAAO,CAAC;AACvD,YAAM,YAAY,OAAO,OAAO,MAAM,MAAM;AAC5C,aAAO,UAAU;AAAA,IACnB;AAAA,EACF;AAEA,SAAO;AACT;;;AC/DO,SAAS,WAAW,QAAgC;AACzD,SAAO;AACT;","names":[]}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@axlsdk/eval",
-  "version": "0.13.0",
+  "version": "0.13.2",
   "description": "Evaluation framework for Axl agentic workflows",
   "type": "module",
   "main": "./dist/index.cjs",
@@ -25,7 +25,7 @@
     "dist"
   ],
   "dependencies": {
-    "@axlsdk/axl": "0.13.0"
+    "@axlsdk/axl": "0.13.2"
   },
   "devDependencies": {
     "@types/node": "^25.2.3",

package/dist/chunk-7B5NFD5T.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/compare.ts","../src/runner.ts"],"sourcesContent":["import type { EvalResult, EvalComparison, EvalRegression, EvalImprovement } from './types.js';\n\nfunction round(n: number): number {\n  return Math.round(n * 100) / 100;\n}\n\nexport function evalCompare(baseline: EvalResult, candidate: EvalResult): EvalComparison {\n  if (baseline.dataset !== candidate.dataset) {\n    throw new Error(\n      `Cannot compare evals from different datasets: \"${baseline.dataset}\" vs \"${candidate.dataset}\"`,\n    );\n  }\n\n  const baselineScorerNames = Object.keys(baseline.summary.scorers).sort();\n  const candidateScorerNames = Object.keys(candidate.summary.scorers).sort();\n  if (JSON.stringify(baselineScorerNames) !== JSON.stringify(candidateScorerNames)) {\n    throw new Error(`Cannot compare evals with different scorers`);\n  }\n\n  const scorers: EvalComparison['scorers'] = {};\n  for (const name of baselineScorerNames) {\n    const bMean = baseline.summary.scorers[name].mean;\n    const cMean = candidate.summary.scorers[name].mean;\n    const delta = round(cMean - bMean);\n    const deltaPercent = bMean > 0 ? round((delta / bMean) * 100) : 0;\n    scorers[name] = { baselineMean: bMean, candidateMean: cMean, delta, deltaPercent };\n  }\n\n  const regressions: EvalRegression[] = [];\n  const improvements: EvalImprovement[] = [];\n\n  const minLength = Math.min(baseline.items.length, candidate.items.length);\n  for (let i = 0; i < minLength; i++) {\n    const bItem = baseline.items[i];\n    const cItem = candidate.items[i];\n    if (bItem.error || cItem.error) continue;\n    for (const name of baselineScorerNames) {\n      const bScore = bItem.scores[name];\n      const cScore = cItem.scores[name];\n      if (bScore == null || cScore == null) continue;\n      const delta = round(cScore - bScore);\n      if (delta < -0.1)\n        regressions.push({\n          input: bItem.input,\n          scorer: name,\n          baselineScore: bScore,\n          candidateScore: cScore,\n          delta,\n        });\n      else if (delta > 0.1)\n        improvements.push({\n          input: bItem.input,\n          scorer: name,\n          baselineScore: bScore,\n          candidateScore: cScore,\n          delta,\n        });\n    }\n  }\n\n  const parts: string[] = [];\n  for (const name of baselineScorerNames) {\n    const s = scorers[name];\n    if (Math.abs(s.delta) > 0.001) {\n      const direction = s.delta > 0 ? 'improves' : 'regresses';\n      const sign = s.delta > 0 ? '+' : '';\n      parts.push(\n        `'${name}' ${direction} by ${sign}${s.deltaPercent.toFixed(1)}% (${s.baselineMean.toFixed(2)} -> ${s.candidateMean.toFixed(2)})`,\n      );\n    }\n  }\n  const summaryStr = `candidate ${parts.length > 0 ? parts.join(', ') : 'no significant changes'} with ${regressions.length} regressions and ${improvements.length} improvements`;\n\n  return {\n    baseline: { id: baseline.id, metadata: baseline.metadata },\n    candidate: { id: candidate.id, metadata: candidate.metadata },\n    scorers,\n    regressions,\n    improvements,\n    summary: summaryStr,\n  };\n}\n","import type { AxlRuntime } from '@axlsdk/axl';\nimport type { EvalConfig, EvalResult, EvalItem, EvalSummary } from './types.js';\nimport type { ScorerContext } from './scorer.js';\nimport { randomUUID } from 'node:crypto';\n\nfunction computeStats(scores: number[]): {\n  mean: number;\n  min: number;\n  max: number;\n  p50: number;\n  p95: number;\n} {\n  if (scores.length === 0) return { mean: 0, min: 0, max: 0, p50: 0, p95: 0 };\n  const sorted = [...scores].sort((a, b) => a - b);\n  const mean = sorted.reduce((a, b) => a + b, 0) / sorted.length;\n  const min = sorted[0];\n  const max = sorted[sorted.length - 1];\n  const p50 = sorted[Math.floor(sorted.length * 0.5)];\n  const p95 = sorted[Math.floor(sorted.length * 0.95)];\n  return { mean: round(mean), min: round(min), max: round(max), p50: round(p50), p95: round(p95) };\n}\n\nfunction round(n: number): number {\n  return Math.round(n * 1000) / 1000;\n}\n\nfunction parseCost(cost: string): number {\n  const match = cost.match(/^\\$?([\\d.]+)$/);\n  if (!match) throw new Error(`Invalid cost format: \"${cost}\"`);\n  return parseFloat(match[1]);\n}\n\nexport async function runEval(\n  config: EvalConfig,\n  executeWorkflow: (\n    input: unknown,\n    runtime: AxlRuntime,\n  ) => Promise<{ output: unknown; cost?: number }>,\n  runtime: AxlRuntime,\n): Promise<EvalResult> {\n  const startTime = Date.now();\n  const id = randomUUID();\n  const items = await config.dataset.getItems();\n  const concurrency = config.concurrency ?? 5;\n  const budgetLimit = config.budget ? parseCost(config.budget) : undefined;\n\n  // Create a scorer context that LLM scorers use to resolve providers.\n  const scorerContext: ScorerContext = {\n    resolveProvider: (uri: string) => {\n      if (typeof runtime.resolveProvider !== 'function') {\n        throw new Error(\n          `LLM scorers require a runtime with resolveProvider(). ` +\n            `Ensure you are using a real AxlRuntime instance, not a mock.`,\n        );\n      }\n      return runtime.resolveProvider(uri);\n    },\n  };\n\n  const evalItems: EvalItem[] = [];\n  let totalCost = 0;\n  let budgetExceeded = false;\n\n  async function processItem(item: (typeof items)[0]): Promise<void> {\n    if (budgetExceeded) {\n      evalItems.push({\n        input: item.input,\n        annotations: item.annotations,\n        output: null,\n        error: 'Budget exceeded',\n        scores: {},\n      });\n      return;\n    }\n\n    const evalItem: EvalItem = {\n      input: item.input,\n      annotations: item.annotations,\n      output: null,\n      scores: {},\n    };\n    try {\n      const result = await executeWorkflow(item.input, runtime);\n      evalItem.output = result.output;\n      if (result.cost != null) {\n        totalCost += result.cost;\n      }\n    } catch (err) {\n      evalItem.error = err instanceof Error ? err.message : String(err);\n      evalItems.push(evalItem);\n      return;\n    }\n\n    if (budgetLimit != null && totalCost > budgetLimit) {\n      budgetExceeded = true;\n    }\n\n    for (const scorer of config.scorers) {\n      try {\n        const score = await scorer.score(\n          evalItem.output,\n          item.input,\n          item.annotations,\n          scorerContext,\n        );\n        // Accumulate LLM scorer cost if available\n        if (scorer.isLlm) {\n          const lastCost = (scorer as unknown as { _lastCost?: number })._lastCost;\n          if (lastCost != null) {\n            totalCost += lastCost;\n          }\n        }\n        if (score < 0 || score > 1) {\n          if (!evalItem.errors) evalItem.errors = [];\n          evalItem.errors.push(\n            `Scorer \"${scorer.name}\" returned out-of-range score ${score} for input ${JSON.stringify(item.input)}`,\n          );\n          evalItem.scores[scorer.name] = null;\n        } else {\n          evalItem.scores[scorer.name] = round(score);\n        }\n      } catch (err) {\n        // Accumulate cost even on failure — the LLM call may have succeeded\n        // before a downstream error (e.g., invalid JSON, schema validation)\n        if (scorer.isLlm) {\n          const lastCost = (scorer as unknown as { _lastCost?: number })._lastCost;\n          if (lastCost != null) {\n            totalCost += lastCost;\n          }\n        }\n        if (!evalItem.errors) evalItem.errors = [];\n        evalItem.errors.push(\n          `Scorer \"${scorer.name}\" threw: ${err instanceof Error ? err.message : String(err)}`,\n        );\n        evalItem.scores[scorer.name] = null;\n      }\n    }\n    evalItems.push(evalItem);\n  }\n\n  let index = 0;\n  async function runNext(): Promise<void> {\n    while (index < items.length) {\n      const currentIndex = index++;\n      await processItem(items[currentIndex]);\n    }\n  }\n\n  const workers = Array.from({ length: Math.min(concurrency, items.length) }, () => runNext());\n  await Promise.all(workers);\n\n  const failures = evalItems.filter((i) => i.error).length;\n  const scorerNames = config.scorers.map((s) => s.name);\n  const scorerStats: EvalSummary['scorers'] = {};\n  for (const name of scorerNames) {\n    const scores = evalItems\n      .filter((i) => !i.error && i.scores[name] != null)\n      .map((i) => i.scores[name] as number);\n    scorerStats[name] = computeStats(scores);\n  }\n\n  return {\n    id,\n    workflow: config.workflow,\n    dataset: config.dataset.name,\n    metadata: config.metadata ?? {},\n    timestamp: new Date().toISOString(),\n    totalCost,\n    duration: Date.now() - startTime,\n    items: evalItems,\n    summary: { count: items.length, failures, scorers: scorerStats },\n  };\n}\n"],"mappings":";AAEA,SAAS,MAAM,GAAmB;AAChC,SAAO,KAAK,MAAM,IAAI,GAAG,IAAI;AAC/B;AAEO,SAAS,YAAY,UAAsB,WAAuC;AACvF,MAAI,SAAS,YAAY,UAAU,SAAS;AAC1C,UAAM,IAAI;AAAA,MACR,kDAAkD,SAAS,OAAO,SAAS,UAAU,OAAO;AAAA,IAC9F;AAAA,EACF;AAEA,QAAM,sBAAsB,OAAO,KAAK,SAAS,QAAQ,OAAO,EAAE,KAAK;AACvE,QAAM,uBAAuB,OAAO,KAAK,UAAU,QAAQ,OAAO,EAAE,KAAK;AACzE,MAAI,KAAK,UAAU,mBAAmB,MAAM,KAAK,UAAU,oBAAoB,GAAG;AAChF,UAAM,IAAI,MAAM,6CAA6C;AAAA,EAC/D;AAEA,QAAM,UAAqC,CAAC;AAC5C,aAAW,QAAQ,qBAAqB;AACtC,UAAM,QAAQ,SAAS,QAAQ,QAAQ,IAAI,EAAE;AAC7C,UAAM,QAAQ,UAAU,QAAQ,QAAQ,IAAI,EAAE;AAC9C,UAAM,QAAQ,MAAM,QAAQ,KAAK;AACjC,UAAM,eAAe,QAAQ,IAAI,MAAO,QAAQ,QAAS,GAAG,IAAI;AAChE,YAAQ,IAAI,IAAI,EAAE,cAAc,OAAO,eAAe,OAAO,OAAO,aAAa;AAAA,EACnF;AAEA,QAAM,cAAgC,CAAC;AACvC,QAAM,eAAkC,CAAC;AAEzC,QAAM,YAAY,KAAK,IAAI,SAAS,MAAM,QAAQ,UAAU,MAAM,MAAM;AACxE,WAAS,IAAI,GAAG,IAAI,WAAW,KAAK;AAClC,UAAM,QAAQ,SAAS,MAAM,CAAC;AAC9B,UAAM,QAAQ,UAAU,MAAM,CAAC;AAC/B,QAAI,MAAM,SAAS,MAAM,MAAO;AAChC,eAAW,QAAQ,qBAAqB;AACtC,YAAM,SAAS,MAAM,OAAO,IAAI;AAChC,YAAM,SAAS,MAAM,OAAO,IAAI;AAChC,UAAI,UAAU,QAAQ,UAAU,KAAM;AACtC,YAAM,QAAQ,MAAM,SAAS,MAAM;AACnC,UAAI,QAAQ;AACV,oBAAY,KAAK;AAAA,UACf,OAAO,MAAM;AAAA,UACb,QAAQ;AAAA,UACR,eAAe;AAAA,UACf,gBAAgB;AAAA,UAChB;AAAA,QACF,CAAC;AAAA,eACM,QAAQ;AACf,qBAAa,KAAK;AAAA,UAChB,OAAO,MAAM;AAAA,UACb,QAAQ;AAAA,UACR,eAAe;AAAA,UACf,gBAAgB;AAAA,UAChB;AAAA,QACF,CAAC;AAAA,IACL;AAAA,EACF;AAEA,QAAM,QAAkB,CAAC;AACzB,aAAW,QAAQ,qBAAqB;AACtC,UAAM,IAAI,QAAQ,IAAI;AACtB,QAAI,KAAK,IAAI,EAAE,KAAK,IAAI,MAAO;AAC7B,YAAM,YAAY,EAAE,QAAQ,IAAI,aAAa;AAC7C,YAAM,OAAO,EAAE,QAAQ,IAAI,MAAM;AACjC,YAAM;AAAA,QACJ,IAAI,IAAI,KAAK,SAAS,OAAO,IAAI,GAAG,EAAE,aAAa,QAAQ,CAAC,CAAC,MAAM,EAAE,aAAa,QAAQ,CAAC,CAAC,OAAO,EAAE,cAAc,QAAQ,CAAC,CAAC;AAAA,MAC/H;AAAA,IACF;AAAA,EACF;AACA,QAAM,aAAa,aAAa,MAAM,SAAS,IAAI,MAAM,KAAK,IAAI,IAAI,wBAAwB,SAAS,YAAY,MAAM,oBAAoB,aAAa,MAAM;AAEhK,SAAO;AAAA,IACL,UAAU,EAAE,IAAI,SAAS,IAAI,UAAU,SAAS,SAAS;AAAA,IACzD,WAAW,EAAE,IAAI,UAAU,IAAI,UAAU,UAAU,SAAS;AAAA,IAC5D;AAAA,IACA;AAAA,IACA;AAAA,IACA,SAAS;AAAA,EACX;AACF;;;AC9EA,SAAS,kBAAkB;AAE3B,SAAS,aAAa,QAMpB;AACA,MAAI,OAAO,WAAW,EAAG,QAAO,EAAE,MAAM,GAAG,KAAK,GAAG,KAAK,GAAG,KAAK,GAAG,KAAK,EAAE;AAC1E,QAAM,SAAS,CAAC,GAAG,MAAM,EAAE,KAAK,CAAC,GAAG,MAAM,IAAI,CAAC;AAC/C,QAAM,OAAO,OAAO,OAAO,CAAC,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,OAAO;AACxD,QAAM,MAAM,OAAO,CAAC;AACpB,QAAM,MAAM,OAAO,OAAO,SAAS,CAAC;AACpC,QAAM,MAAM,OAAO,KAAK,MAAM,OAAO,SAAS,GAAG,CAAC;AAClD,QAAM,MAAM,OAAO,KAAK,MAAM,OAAO,SAAS,IAAI,CAAC;AACnD,SAAO,EAAE,MAAMA,OAAM,IAAI,GAAG,KAAKA,OAAM,GAAG,GAAG,KAAKA,OAAM,GAAG,GAAG,KAAKA,OAAM,GAAG,GAAG,KAAKA,OAAM,GAAG,EAAE;AACjG;AAEA,SAASA,OAAM,GAAmB;AAChC,SAAO,KAAK,MAAM,IAAI,GAAI,IAAI;AAChC;AAEA,SAAS,UAAU,MAAsB;AACvC,QAAM,QAAQ,KAAK,MAAM,eAAe;AACxC,MAAI,CAAC,MAAO,OAAM,IAAI,MAAM,yBAAyB,IAAI,GAAG;AAC5D,SAAO,WAAW,MAAM,CAAC,CAAC;AAC5B;AAEA,eAAsB,QACpB,QACA,iBAIA,SACqB;AACrB,QAAM,YAAY,KAAK,IAAI;AAC3B,QAAM,KAAK,WAAW;AACtB,QAAM,QAAQ,MAAM,OAAO,QAAQ,SAAS;AAC5C,QAAM,cAAc,OAAO,eAAe;AAC1C,QAAM,cAAc,OAAO,SAAS,UAAU,OAAO,MAAM,IAAI;AAG/D,QAAM,gBAA+B;AAAA,IACnC,iBAAiB,CAAC,QAAgB;AAChC,UAAI,OAAO,QAAQ,oBAAoB,YAAY;AACjD,cAAM,IAAI;AAAA,UACR;AAAA,QAEF;AAAA,MACF;AACA,aAAO,QAAQ,gBAAgB,GAAG;AAAA,IACpC;AAAA,EACF;AAEA,QAAM,YAAwB,CAAC;AAC/B,MAAI,YAAY;AAChB,MAAI,iBAAiB;AAErB,iBAAe,YAAY,MAAwC;AACjE,QAAI,gBAAgB;AAClB,gBAAU,KAAK;AAAA,QACb,OAAO,KAAK;AAAA,QACZ,aAAa,KAAK;AAAA,QAClB,QAAQ;AAAA,QACR,OAAO;AAAA,QACP,QAAQ,CAAC;AAAA,MACX,CAAC;AACD;AAAA,IACF;AAEA,UAAM,WAAqB;AAAA,MACzB,OAAO,KAAK;AAAA,MACZ,aAAa,KAAK;AAAA,MAClB,QAAQ;AAAA,MACR,QAAQ,CAAC;AAAA,IACX;AACA,QAAI;AACF,YAAM,SAAS,MAAM,gBAAgB,KAAK,OAAO,OAAO;AACxD,eAAS,SAAS,OAAO;AACzB,UAAI,OAAO,QAAQ,MAAM;AACvB,qBAAa,OAAO;AAAA,MACtB;AAAA,IACF,SAAS,KAAK;AACZ,eAAS,QAAQ,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAChE,gBAAU,KAAK,QAAQ;AACvB;AAAA,IACF;AAEA,QAAI,eAAe,QAAQ,YAAY,aAAa;AAClD,uBAAiB;AAAA,IACnB;AAEA,eAAW,UAAU,OAAO,SAAS;AACnC,UAAI;AACF,cAAM,QAAQ,MAAM,OAAO;AAAA,UACzB,SAAS;AAAA,UACT,KAAK;AAAA,UACL,KAAK;AAAA,UACL;AAAA,QACF;AAEA,YAAI,OAAO,OAAO;AAChB,gBAAM,WAAY,OAA6C;AAC/D,cAAI,YAAY,MAAM;AACpB,yBAAa;AAAA,UACf;AAAA,QACF;AACA,YAAI,QAAQ,KAAK,QAAQ,GAAG;AAC1B,cAAI,CAAC,SAAS,OAAQ,UAAS,SAAS,CAAC;AACzC,mBAAS,OAAO;AAAA,YACd,WAAW,OAAO,IAAI,iCAAiC,KAAK,cAAc,KAAK,UAAU,KAAK,KAAK,CAAC;AAAA,UACtG;AACA,mBAAS,OAAO,OAAO,IAAI,IAAI;AAAA,QACjC,OAAO;AACL,mBAAS,OAAO,OAAO,IAAI,IAAIA,OAAM,KAAK;AAAA,QAC5C;AAAA,MACF,SAAS,KAAK;AAGZ,YAAI,OAAO,OAAO;AAChB,gBAAM,WAAY,OAA6C;AAC/D,cAAI,YAAY,MAAM;AACpB,yBAAa;AAAA,UACf;AAAA,QACF;AACA,YAAI,CAAC,SAAS,OAAQ,UAAS,SAAS,CAAC;AACzC,iBAAS,OAAO;AAAA,UACd,WAAW,OAAO,IAAI,YAAY,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC;AAAA,QACpF;AACA,iBAAS,OAAO,OAAO,IAAI,IAAI;AAAA,MACjC;AAAA,IACF;AACA,cAAU,KAAK,QAAQ;AAAA,EACzB;AAEA,MAAI,QAAQ;AACZ,iBAAe,UAAyB;AACtC,WAAO,QAAQ,MAAM,QAAQ;AAC3B,YAAM,eAAe;AACrB,YAAM,YAAY,MAAM,YAAY,CAAC;AAAA,IACvC;AAAA,EACF;AAEA,QAAM,UAAU,MAAM,KAAK,EAAE,QAAQ,KAAK,IAAI,aAAa,MAAM,MAAM,EAAE,GAAG,MAAM,QAAQ,CAAC;AAC3F,QAAM,QAAQ,IAAI,OAAO;AAEzB,QAAM,WAAW,UAAU,OAAO,CAAC,MAAM,EAAE,KAAK,EAAE;AAClD,QAAM,cAAc,OAAO,QAAQ,IAAI,CAAC,MAAM,EAAE,IAAI;AACpD,QAAM,cAAsC,CAAC;AAC7C,aAAW,QAAQ,aAAa;AAC9B,UAAM,SAAS,UACZ,OAAO,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,OAAO,IAAI,KAAK,IAAI,EAChD,IAAI,CAAC,MAAM,EAAE,OAAO,IAAI,CAAW;AACtC,gBAAY,IAAI,IAAI,aAAa,MAAM;AAAA,EACzC;AAEA,SAAO;AAAA,IACL;AAAA,IACA,UAAU,OAAO;AAAA,IACjB,SAAS,OAAO,QAAQ;AAAA,IACxB,UAAU,OAAO,YAAY,CAAC;AAAA,IAC9B,YAAW,oBAAI,KAAK,GAAE,YAAY;AAAA,IAClC;AAAA,IACA,UAAU,KAAK,IAAI,IAAI;AAAA,IACvB,OAAO;AAAA,IACP,SAAS,EAAE,OAAO,MAAM,QAAQ,UAAU,SAAS,YAAY;AAAA,EACjE;AACF;","names":["round"]}