oh-my-knowledge 0.23.0 → 0.24.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (153) hide show
  1. package/README.md +63 -64
  2. package/README.zh.md +64 -63
  3. package/dist/src/analysis/report-diagnostics.d.ts +3 -2
  4. package/dist/src/analysis/report-diagnostics.d.ts.map +1 -1
  5. package/dist/src/analysis/report-diagnostics.js +107 -80
  6. package/dist/src/analysis/report-diagnostics.js.map +1 -1
  7. package/dist/src/analysis/sample-diagnostics.d.ts +4 -1
  8. package/dist/src/analysis/sample-diagnostics.d.ts.map +1 -1
  9. package/dist/src/analysis/sample-diagnostics.js +123 -29
  10. package/dist/src/analysis/sample-diagnostics.js.map +1 -1
  11. package/dist/src/authoring/evolver.d.ts +5 -0
  12. package/dist/src/authoring/evolver.d.ts.map +1 -1
  13. package/dist/src/authoring/evolver.js +20 -3
  14. package/dist/src/authoring/evolver.js.map +1 -1
  15. package/dist/src/cli/i18n-dict.d.ts +1 -1
  16. package/dist/src/cli/i18n-dict.d.ts.map +1 -1
  17. package/dist/src/cli/i18n-dict.js +28 -24
  18. package/dist/src/cli/i18n-dict.js.map +1 -1
  19. package/dist/src/cli/index.js +80 -66
  20. package/dist/src/cli/index.js.map +1 -1
  21. package/dist/src/cli/parse-run-config.d.ts.map +1 -1
  22. package/dist/src/cli/parse-run-config.js +4 -4
  23. package/dist/src/cli/parse-run-config.js.map +1 -1
  24. package/dist/src/cli/progress.d.ts +1 -0
  25. package/dist/src/cli/progress.d.ts.map +1 -1
  26. package/dist/src/cli/progress.js +8 -1
  27. package/dist/src/cli/progress.js.map +1 -1
  28. package/dist/src/eval-core/cache.d.ts +7 -5
  29. package/dist/src/eval-core/cache.d.ts.map +1 -1
  30. package/dist/src/eval-core/cache.js +11 -7
  31. package/dist/src/eval-core/cache.js.map +1 -1
  32. package/dist/src/eval-core/comparability.d.ts +11 -0
  33. package/dist/src/eval-core/comparability.d.ts.map +1 -0
  34. package/dist/src/eval-core/comparability.js +271 -0
  35. package/dist/src/eval-core/comparability.js.map +1 -0
  36. package/dist/src/eval-core/evaluation-execution.d.ts +4 -1
  37. package/dist/src/eval-core/evaluation-execution.d.ts.map +1 -1
  38. package/dist/src/eval-core/evaluation-execution.js +25 -17
  39. package/dist/src/eval-core/evaluation-execution.js.map +1 -1
  40. package/dist/src/eval-core/evaluation-job.d.ts +2 -2
  41. package/dist/src/eval-core/evaluation-job.d.ts.map +1 -1
  42. package/dist/src/eval-core/evaluation-job.js +2 -2
  43. package/dist/src/eval-core/evaluation-job.js.map +1 -1
  44. package/dist/src/eval-core/evaluation-reporting.d.ts +3 -1
  45. package/dist/src/eval-core/evaluation-reporting.d.ts.map +1 -1
  46. package/dist/src/eval-core/evaluation-reporting.js +64 -4
  47. package/dist/src/eval-core/evaluation-reporting.js.map +1 -1
  48. package/dist/src/eval-core/execution-strategy.js +2 -2
  49. package/dist/src/eval-core/execution-strategy.js.map +1 -1
  50. package/dist/src/eval-core/schema.d.ts.map +1 -1
  51. package/dist/src/eval-core/schema.js +13 -0
  52. package/dist/src/eval-core/schema.js.map +1 -1
  53. package/dist/src/eval-workflows/batch-evaluation-workflow.d.ts +111 -0
  54. package/dist/src/eval-workflows/batch-evaluation-workflow.d.ts.map +1 -0
  55. package/dist/src/eval-workflows/batch-evaluation-workflow.js +215 -0
  56. package/dist/src/eval-workflows/batch-evaluation-workflow.js.map +1 -0
  57. package/dist/src/eval-workflows/evaluation-pipeline.d.ts +5 -3
  58. package/dist/src/eval-workflows/evaluation-pipeline.d.ts.map +1 -1
  59. package/dist/src/eval-workflows/evaluation-pipeline.js +6 -5
  60. package/dist/src/eval-workflows/evaluation-pipeline.js.map +1 -1
  61. package/dist/src/eval-workflows/evaluation-preparation.d.ts +8 -6
  62. package/dist/src/eval-workflows/evaluation-preparation.d.ts.map +1 -1
  63. package/dist/src/eval-workflows/evaluation-preparation.js +8 -4
  64. package/dist/src/eval-workflows/evaluation-preparation.js.map +1 -1
  65. package/dist/src/eval-workflows/run-evaluation.d.ts +20 -17
  66. package/dist/src/eval-workflows/run-evaluation.d.ts.map +1 -1
  67. package/dist/src/eval-workflows/run-evaluation.js +25 -15
  68. package/dist/src/eval-workflows/run-evaluation.js.map +1 -1
  69. package/dist/src/executors/claude-cli.d.ts.map +1 -1
  70. package/dist/src/executors/claude-cli.js +11 -6
  71. package/dist/src/executors/claude-cli.js.map +1 -1
  72. package/dist/src/executors/codex-cli-trace.d.ts +10 -0
  73. package/dist/src/executors/codex-cli-trace.d.ts.map +1 -0
  74. package/dist/src/executors/codex-cli-trace.js +123 -0
  75. package/dist/src/executors/codex-cli-trace.js.map +1 -0
  76. package/dist/src/executors/codex-cli.d.ts +18 -0
  77. package/dist/src/executors/codex-cli.d.ts.map +1 -0
  78. package/dist/src/executors/codex-cli.js +254 -0
  79. package/dist/src/executors/codex-cli.js.map +1 -0
  80. package/dist/src/executors/codex-sdk.d.ts +18 -0
  81. package/dist/src/executors/codex-sdk.d.ts.map +1 -0
  82. package/dist/src/executors/codex-sdk.js +214 -0
  83. package/dist/src/executors/codex-sdk.js.map +1 -0
  84. package/dist/src/executors/gemini.d.ts.map +1 -1
  85. package/dist/src/executors/gemini.js +28 -24
  86. package/dist/src/executors/gemini.js.map +1 -1
  87. package/dist/src/executors/index.d.ts.map +1 -1
  88. package/dist/src/executors/index.js +7 -2
  89. package/dist/src/executors/index.js.map +1 -1
  90. package/dist/src/executors/runtime-fingerprint.d.ts +7 -0
  91. package/dist/src/executors/runtime-fingerprint.d.ts.map +1 -0
  92. package/dist/src/executors/runtime-fingerprint.js +277 -0
  93. package/dist/src/executors/runtime-fingerprint.js.map +1 -0
  94. package/dist/src/executors/script.d.ts.map +1 -1
  95. package/dist/src/executors/script.js +47 -55
  96. package/dist/src/executors/script.js.map +1 -1
  97. package/dist/src/executors/shared.d.ts +78 -1
  98. package/dist/src/executors/shared.d.ts.map +1 -1
  99. package/dist/src/executors/shared.js +203 -1
  100. package/dist/src/executors/shared.js.map +1 -1
  101. package/dist/src/grading/assertions.d.ts.map +1 -1
  102. package/dist/src/grading/assertions.js +18 -3
  103. package/dist/src/grading/assertions.js.map +1 -1
  104. package/dist/src/grading/index.d.ts.map +1 -1
  105. package/dist/src/grading/index.js +11 -0
  106. package/dist/src/grading/index.js.map +1 -1
  107. package/dist/src/grading/judge.d.ts.map +1 -1
  108. package/dist/src/grading/judge.js +75 -6
  109. package/dist/src/grading/judge.js.map +1 -1
  110. package/dist/src/inputs/skill-loader.d.ts +1 -1
  111. package/dist/src/inputs/skill-loader.d.ts.map +1 -1
  112. package/dist/src/inputs/skill-loader.js +1 -1
  113. package/dist/src/inputs/skill-loader.js.map +1 -1
  114. package/dist/src/renderer/html-renderer.d.ts +5 -4
  115. package/dist/src/renderer/html-renderer.d.ts.map +1 -1
  116. package/dist/src/renderer/html-renderer.js +217 -93
  117. package/dist/src/renderer/html-renderer.js.map +1 -1
  118. package/dist/src/renderer/layout.d.ts +2 -1
  119. package/dist/src/renderer/layout.d.ts.map +1 -1
  120. package/dist/src/renderer/layout.js +30 -42
  121. package/dist/src/renderer/layout.js.map +1 -1
  122. package/dist/src/renderer/summary.d.ts +3 -3
  123. package/dist/src/renderer/summary.d.ts.map +1 -1
  124. package/dist/src/renderer/summary.js +232 -42
  125. package/dist/src/renderer/summary.js.map +1 -1
  126. package/dist/src/renderer/trends.d.ts.map +1 -1
  127. package/dist/src/renderer/trends.js +5 -3
  128. package/dist/src/renderer/trends.js.map +1 -1
  129. package/dist/src/server/report-server.js +4 -4
  130. package/dist/src/server/report-server.js.map +1 -1
  131. package/dist/src/server/report-store.d.ts +7 -5
  132. package/dist/src/server/report-store.d.ts.map +1 -1
  133. package/dist/src/server/report-store.js +39 -11
  134. package/dist/src/server/report-store.js.map +1 -1
  135. package/dist/src/types/eval.d.ts +4 -2
  136. package/dist/src/types/eval.d.ts.map +1 -1
  137. package/dist/src/types/executor.d.ts +5 -0
  138. package/dist/src/types/executor.d.ts.map +1 -1
  139. package/dist/src/types/judge.d.ts +10 -0
  140. package/dist/src/types/judge.d.ts.map +1 -1
  141. package/dist/src/types/report.d.ts +120 -24
  142. package/dist/src/types/report.d.ts.map +1 -1
  143. package/dist/src/types/storage.d.ts +7 -7
  144. package/dist/src/types/storage.d.ts.map +1 -1
  145. package/package.json +2 -1
  146. package/dist/src/eval-workflows/each-evaluation-workflow.d.ts +0 -153
  147. package/dist/src/eval-workflows/each-evaluation-workflow.d.ts.map +0 -1
  148. package/dist/src/eval-workflows/each-evaluation-workflow.js +0 -178
  149. package/dist/src/eval-workflows/each-evaluation-workflow.js.map +0 -1
  150. package/dist/src/executors/openai-cli.d.ts +0 -3
  151. package/dist/src/executors/openai-cli.d.ts.map +0 -1
  152. package/dist/src/executors/openai-cli.js +0 -60
  153. package/dist/src/executors/openai-cli.js.map +0 -1
@@ -0,0 +1,215 @@
1
+ import { dirname, resolve } from 'node:path';
2
+ import { DEFAULT_OUTPUT_DIR, generateRunId, getCliVersion, getGitInfo, persistReport } from '../eval-core/evaluation-reporting.js';
3
+ import { buildEvaluationRequest, createEvaluationRun, createSucceededJob, finalizeEvaluationRun } from '../eval-core/evaluation-job.js';
4
+ import { getExecutorRuntimeFingerprint } from '../executors/runtime-fingerprint.js';
5
+ import { createFileJobStore, DEFAULT_JOBS_DIR } from '../server/job-store.js';
6
+ import { resolveArtifacts } from '../inputs/skill-loader.js';
7
+ function commonRuntime(runtimes) {
8
+ const values = Object.values(runtimes);
9
+ if (values.length === 0)
10
+ return undefined;
11
+ const first = values[0];
12
+ return values.every((runtime) => runtime.fingerprint === first.fingerprint) ? first : undefined;
13
+ }
14
+ function safeRunIdPart(value) {
15
+ return value
16
+ .replaceAll(/[\\/:]/g, '-')
17
+ .replaceAll(/[^a-zA-Z0-9._@-]/g, '_')
18
+ .replace(/^_+|_+$/g, '')
19
+ || 'skill';
20
+ }
21
+ function reportSummarySnapshot(report, treatmentVariant) {
22
+ return {
23
+ baseline: report.summary.baseline || {},
24
+ [treatmentVariant]: report.summary[treatmentVariant] || {},
25
+ };
26
+ }
27
+ function buildBatchItems(skillResults) {
28
+ return skillResults.map(({ name, skillPath, samplesPath, report, filePath }) => ({
29
+ name,
30
+ skillPath,
31
+ samplesPath,
32
+ reportId: report.id,
33
+ reportPath: filePath,
34
+ status: 'completed',
35
+ sampleCount: report.meta.sampleCount,
36
+ totalCostUSD: report.meta.totalCostUSD,
37
+ artifactHash: report.meta.artifactHashes?.[name] || null,
38
+ summary: reportSummarySnapshot(report, name),
39
+ ...(report.variance ? { variance: report.variance } : {}),
40
+ }));
41
+ }
42
+ function buildExecutorRuntimesBySkill({ skillDir, skillResults, executorName, model, }) {
43
+ const executorRuntimes = {};
44
+ for (const skill of skillResults) {
45
+ executorRuntimes[skill.name] =
46
+ skill.report.meta.executorRuntimes?.[skill.name]
47
+ ?? skill.report.meta.executorRuntime
48
+ ?? getExecutorRuntimeFingerprint(executorName, model, {
49
+ skillDir: skill.skillPath ? dirname(skill.skillPath) : skillDir,
50
+ });
51
+ }
52
+ return executorRuntimes;
53
+ }
54
+ export function buildBatchEvaluationReport({ batchRunId, skillDir, skillEntries, skillResults, model, judgeModel, noJudge, executorName, judgeExecutorName, project, owner, tags, concurrency, timeoutMs, totalCostUSD, repeat, judgeModels, noCache, }) {
55
+ const request = buildEvaluationRequest({
56
+ samplesPath: '',
57
+ skillDir,
58
+ artifacts: skillEntries.map((entry) => ({
59
+ name: entry.name,
60
+ kind: 'skill',
61
+ source: 'file-path',
62
+ content: null,
63
+ locator: entry.skillPath,
64
+ })),
65
+ model,
66
+ judgeModel: noJudge ? null : judgeModel,
67
+ executor: executorName,
68
+ judgeExecutor: judgeExecutorName || executorName,
69
+ noJudge,
70
+ concurrency,
71
+ timeoutMs,
72
+ noCache,
73
+ dryRun: false,
74
+ blind: false,
75
+ project,
76
+ owner,
77
+ tags,
78
+ repeat,
79
+ judgeModels,
80
+ batch: true,
81
+ });
82
+ const createdAt = new Date().toISOString();
83
+ const { run: initialRun, startedAt } = createEvaluationRun(batchRunId, createdAt);
84
+ const finishedAt = new Date().toISOString();
85
+ const run = finalizeEvaluationRun(initialRun, finishedAt);
86
+ const job = createSucceededJob({
87
+ jobId: `job-${batchRunId}`,
88
+ runId: batchRunId,
89
+ reportId: batchRunId,
90
+ request,
91
+ createdAt,
92
+ startedAt,
93
+ finishedAt,
94
+ });
95
+ const sampleCount = skillResults.reduce((sum, skill) => sum + skill.report.meta.sampleCount, 0);
96
+ const executorRuntimes = buildExecutorRuntimesBySkill({ skillDir, skillResults, executorName, model });
97
+ const executorRuntime = commonRuntime(executorRuntimes)
98
+ ?? Object.values(executorRuntimes)[0]
99
+ ?? getExecutorRuntimeFingerprint(executorName, model, { skillDir });
100
+ const isJudgeEnsemble = Boolean(judgeModels && judgeModels.length >= 2);
101
+ const totalCostReported = skillResults.every((skill) => skill.report.meta.totalCostReported !== false
102
+ && Object.values(skill.report.summary || {}).every((variant) => variant.execCostReported !== false && variant.judgeCostReported !== false));
103
+ const report = {
104
+ kind: 'batch-evaluation',
105
+ id: batchRunId,
106
+ mode: 'skill',
107
+ meta: {
108
+ mode: 'skill',
109
+ model,
110
+ judgeModel: noJudge || isJudgeEnsemble ? null : judgeModel,
111
+ executor: executorName,
112
+ skillDir,
113
+ sampleCount,
114
+ taskCount: sampleCount * 2,
115
+ totalArtifacts: skillResults.length,
116
+ totalCostUSD: Number(totalCostUSD.toFixed(6)),
117
+ ...(totalCostReported ? {} : { totalCostReported: false }),
118
+ timestamp: finishedAt,
119
+ cliVersion: getCliVersion(),
120
+ nodeVersion: process.version,
121
+ executorRuntime,
122
+ executorRuntimes,
123
+ judgeRuntime: noJudge || isJudgeEnsemble ? null : getExecutorRuntimeFingerprint(judgeExecutorName || executorName, judgeModel, { skillDir }),
124
+ ...(judgeModels && judgeModels.length >= 2 ? {
125
+ judgeModels: judgeModels.map((jc) => `${jc.executor}:${jc.model}`),
126
+ judgeRuntimes: Object.fromEntries(judgeModels.map((jc) => [`${jc.executor}:${jc.model}`, getExecutorRuntimeFingerprint(jc.executor, jc.model, { skillDir })])),
127
+ } : {}),
128
+ request,
129
+ run,
130
+ job,
131
+ gitInfo: getGitInfo(),
132
+ },
133
+ items: buildBatchItems(skillResults),
134
+ };
135
+ return { report, job };
136
+ }
137
+ export async function executeBatchEvaluationRuns({ skillDir, skillEntries, model, judgeModel, outputDir = DEFAULT_OUTPUT_DIR, project, owner, tags, noJudge = false, concurrency = 1, timeoutMs, executorName, judgeExecutorName, jobStore = null, persistJob = true, onProgress = null, onSkillProgress = null, skipPreflight = false, mcpConfig, verbose = false, repeat, judgeRepeat, judgeModels, lengthDebias, noCache = false, strictBaseline, variantAllowedSkills, runSingleEvaluation, }) {
138
+ const batchRunId = generateRunId(['batch']);
139
+ const skillResults = [];
140
+ let totalCostUSD = 0;
141
+ for (let i = 0; i < skillEntries.length; i++) {
142
+ const entry = skillEntries[i];
143
+ onSkillProgress?.({ phase: 'start', skill: entry.name, current: i + 1, total: skillEntries.length });
144
+ // Batch mode 的实验结构固定为 baseline control vs 当前 skill treatment。
145
+ const perSkillAllowedSkills = variantAllowedSkills?.[entry.name] !== undefined
146
+ ? { ...variantAllowedSkills, [entry.skillPath]: variantAllowedSkills[entry.name] }
147
+ : variantAllowedSkills;
148
+ const skillArtifacts = resolveArtifacts(resolve(skillDir), ['baseline', entry.skillPath], { strictBaseline, variantAllowedSkills: perSkillAllowedSkills }).map((artifact) => {
149
+ if (artifact.name === entry.skillPath) {
150
+ return { ...artifact, name: entry.name, experimentRole: 'treatment' };
151
+ }
152
+ return { ...artifact, experimentRole: 'control' };
153
+ });
154
+ const childRunId = `${batchRunId}-${String(i + 1).padStart(2, '0')}-${safeRunIdPart(entry.name)}`;
155
+ const { report, filePath } = await runSingleEvaluation({
156
+ samplesPath: entry.samplesPath,
157
+ skillDir,
158
+ artifacts: skillArtifacts,
159
+ model,
160
+ judgeModel,
161
+ outputDir,
162
+ noJudge,
163
+ concurrency,
164
+ timeoutMs,
165
+ noCache,
166
+ executorName,
167
+ judgeExecutorName,
168
+ jobStore: null,
169
+ persistJob: false,
170
+ onProgress,
171
+ skipPreflight: skipPreflight || i > 0,
172
+ mcpConfig,
173
+ verbose,
174
+ runId: childRunId,
175
+ judgeRepeat,
176
+ judgeModels,
177
+ lengthDebias,
178
+ });
179
+ skillResults.push({
180
+ name: entry.name,
181
+ skillPath: entry.skillPath,
182
+ samplesPath: entry.samplesPath,
183
+ report,
184
+ filePath,
185
+ });
186
+ totalCostUSD += report.meta.totalCostUSD;
187
+ onSkillProgress?.({ phase: 'done', skill: entry.name, current: i + 1, total: skillEntries.length });
188
+ }
189
+ const { report: batchReport, job } = buildBatchEvaluationReport({
190
+ batchRunId,
191
+ skillDir,
192
+ skillEntries,
193
+ skillResults,
194
+ model,
195
+ judgeModel,
196
+ noJudge,
197
+ executorName,
198
+ judgeExecutorName,
199
+ project,
200
+ owner,
201
+ tags,
202
+ concurrency,
203
+ timeoutMs,
204
+ totalCostUSD,
205
+ repeat,
206
+ judgeModels,
207
+ noCache,
208
+ });
209
+ const filePath = persistReport(batchReport, outputDir);
210
+ const resolvedJobStore = persistJob ? (jobStore ?? createFileJobStore(DEFAULT_JOBS_DIR)) : null;
211
+ if (resolvedJobStore)
212
+ await resolvedJobStore.save(job.jobId, job);
213
+ return { report: batchReport, filePath };
214
+ }
215
+ //# sourceMappingURL=batch-evaluation-workflow.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"batch-evaluation-workflow.js","sourceRoot":"","sources":["../../../src/eval-workflows/batch-evaluation-workflow.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAC7C,OAAO,EAAE,kBAAkB,EAAE,aAAa,EAAE,aAAa,EAAE,UAAU,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACnI,OAAO,EAAE,sBAAsB,EAAE,mBAAmB,EAAE,kBAAkB,EAAE,qBAAqB,EAAE,MAAM,gCAAgC,CAAC;AACxI,OAAO,EAAE,6BAA6B,EAAE,MAAM,qCAAqC,CAAC;AACpF,OAAO,EAAE,kBAAkB,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC9E,OAAO,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AAgD7D,SAAS,aAAa,CAAC,QAAoD;IACzE,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IACvC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAC1C,MAAM,KAAK,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;IACxB,OAAO,MAAM,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,WAAW,KAAK,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;AAClG,CAAC;AAED,SAAS,aAAa,CAAC,KAAa;IAClC,OAAO,KAAK;SACT,UAAU,CAAC,SAAS,EAAE,GAAG,CAAC;SAC1B,UAAU,CAAC,mBAAmB,EAAE,GAAG,CAAC;SACpC,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC;WACrB,OAAO,CAAC;AACf,CAAC;AAED,SAAS,qBAAqB,CAAC,MAAwB,EAAE,gBAAwB;IAC/E,OAAO;QACL,QAAQ,EAAE,MAAM,CAAC,OAAO,CAAC,QAAQ,IAAK,EAAqB;QAC3D,CAAC,gBAAgB,CAAC,EAAE,MAAM,CAAC,OAAO,CAAC,gBAAgB,CAAC,IAAK,EAAqB;KAC/E,CAAC;AACJ,CAAC;AAED,SAAS,eAAe,CAAC,YAAsC;IAC7D,OAAO,YAAY,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,WAAW,EAAE,MAAM,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC;QAC/E,IAAI;QACJ,SAAS;QACT,WAAW;QACX,QAAQ,EAAE,MAAM,CAAC,EAAE;QACnB,UAAU,EAAE,QAAQ;QACpB,MAAM,EAAE,WAAW;QACnB,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,WAAW;QACpC,YAAY,EAAE,MAAM,CAAC,IAAI,CAAC,YAAY;QACtC,YAAY,EAAE,MAAM,CAAC,IAAI,CAAC,cAAc,EAAE,CAAC,IAAI,CAAC,IAAI,IAAI;QACxD,OAAO,EAAE,qBAAqB,CAAC,MAAM,EAAE,IAAI,CAAC;QAC5C,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC1D,CAAC,CAAC,CAAC;AACN,CAAC;AAED,SAAS,4BAA4B,CAAC,EACpC,QAAQ,EACR,YAAY,EACZ,YAAY,EACZ,KAAK,GAMN;IACC,MAAM,gBAAgB,GAA+C,EAAE,CAAC;IACxE,KAAK,MAAM,KAAK,IAAI,YAAY,EAAE,CAAC;QACjC,gBAAgB,CAAC,KAAK,CAAC,IAAI,CAAC;YAC1B,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,gBAAgB,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC;mBAC7C,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,eAAe;mBACjC,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE;oBACpD,QAAQ,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,QAAQ;iBAChE,CAAC,CAAC;IACP,CAAC;IACD,OAAO,gBAAgB,CAAC;AAC1B,CAAC;AAED,MAAM,UAAU,0BAA0B,CAAC,EACzC,UAAU,EACV,QAAQ,EACR,YAAY,EACZ,YAAY,EACZ,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,OAAO,EACP,KAAK,EACL,IAAI,EACJ,WAAW,EACX,SAAS,EACT,YAAY,EACZ,MAAM,EACN,WAAW,EACX,OAAO,GAoBR;IACC,MAAM,OAAO,GAAG,sBAAsB,CAAC;QACrC,WAAW,EAAE,EAAE;QACf,QAAQ;QACR,SAAS,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;YACtC,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,IAAI,EAAE,OAAO;YACb,MAAM,EAAE,WAAW;YACnB,OAAO,EAAE,IAAI;YACb,OAAO,EAAE,KAAK,CAAC,SAAS;SACzB,CAAC,CAAC;QACH,KAAK;QACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;QACvC,QAAQ,EAAE,YAAY;QACtB,aAAa,EAAE,iBAAiB,IAAI,YAAY;QAChD,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO;QACP,MAAM,EAAE,KAAK;QACb,KAAK,EAAE,KAAK;QACZ,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM;QACN,WAAW;QACX,KAAK,EAAE,IAAI;KACZ,CAAC,CAAC;IACH,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC3C,MAAM,EAAE,GAAG,EAAE,UAAU,EAAE,SAAS,EAAE,GAAG,mBAAmB,CAAC,UAAU,EAAE,SAAS,CAAC,CAAC;IAClF,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,GAAG,GAAG,qBAAqB,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IAC1D,MAAM,GAAG,GAAG,kBAAkB,CAAC;QAC7B,KAAK,EAAE,OAAO,UAAU,EAAE;QAC1B,KAAK,EAAE,UAAU;QACjB,QAAQ,EAAE,UAAU;QACpB,OAAO;QACP,SAAS;QACT,SAAS;QACT,UAAU;KACX,CAAC,CAAC;IACH,MAAM,WAAW,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;IAChG,MAAM,gBAAgB,GAAG,4BAA4B,CAAC,EAAE,QAAQ,EAAE,YAAY,EAAE,YAAY,EAAE,KAAK,EAAE,CAAC,CAAC;IACvG,MAAM,eAAe,GAAG,aAAa,CAAC,gBAAgB,CAAC;WAClD,MAAM,CAAC,MAAM,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC;WAClC,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE,EAAE,QAAQ,EAAE,CAAC,CAAC;IACtE,MAAM,eAAe,GAAG,OAAO,CAAC,WAAW,IAAI,WAAW,CAAC,MAAM,IAAI,CAAC,CAAC,CAAC;IACxE,MAAM,iBAAiB,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE,CACrD,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,iBAAiB,KAAK,KAAK;WAC1C,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CAC7D,OAAO,CAAC,gBAAgB,KAAK,KAAK,IAAI,OAAO,CAAC,iBAAiB,KAAK,KAAK,CAAC,CAAC,CAAC;IAEhF,MAAM,MAAM,GAA0B;QACpC,IAAI,EAAE,kBAAkB;QACxB,EAAE,EAAE,UAAU;QACd,IAAI,EAAE,OAAO;QACb,IAAI,EAAE;YACJ,IAAI,EAAE,OAAO;YACb,KAAK;YACL,UAAU,EAAE,OAAO,IAAI,eAAe,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;YAC1D,QAAQ,EAAE,YAAY;YACtB,QAAQ;YACR,WAAW;YACX,SAAS,EAAE,WAAW,GAAG,CAAC;YAC1B,cAAc,EAAE,YAAY,CAAC,MAAM;YACnC,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YAC7C,GAAG,CAAC,iBAAiB,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,KAAK,EAAE,CAAC;YAC1D,SAAS,EAAE,UAAU;YACrB,UAAU,EAAE,aAAa,EAAE;YAC3B,WAAW,EAAE,OAAO,CAAC,OAAO;YAC5B,eAAe;YACf,gBAAgB;YAChB,YAAY,EAAE,OAAO,IAAI,eAAe,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,6BAA6B,CAAC,iBAAiB,IAAI,YAAY,EAAE,UAAU,EAAE,EAAE,QAAQ,EAAE,CAAC;YAC5I,GAAG,CAAC,WAAW,IAAI,WAAW,CAAC,MAAM,IAAI,CAAC,CAAC,CAAC,CAAC;gBAC3C,WAAW,EAAE,WAAW,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,IAAI,EAAE,CAAC,KAAK,EAAE,CAAC;gBAClE,aAAa,EAAE,MAAM,CAAC,WAAW,CAC/B,WAAW,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,IAAI,EAAE,CAAC,KAAK,EAAE,EAAE,6BAA6B,CAAC,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC,KAAK,EAAE,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC,CAC5H;aACF,CAAC,CAAC,CAAC,EAAE,CAAC;YACP,OAAO;YACP,GAAG;YACH,GAAG;YACH,OAAO,EAAE,UAAU,EAAE;SACtB;QACD,KAAK,EAAE,eAAe,CAAC,YAAY,CAAC;KACrC,CAAC;IAEF,OAAO,EAAE,MAAM,EAAE,GAAG,EAAE,CAAC;AACzB,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,0BAA0B,CAAC,EAC/C,QAAQ,EACR,YAAY,EACZ,KAAK,EACL,UAAU,EACV,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAO,GAAG,KAAK,EACf,WAAW,GAAG,CAAC,EACf,SAAS,EACT,YAAY,EACZ,iBAAiB,EACjB,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,eAAe,GAAG,IAAI,EACtB,aAAa,GAAG,KAAK,EACrB,SAAS,EACT,OAAO,GAAG,KAAK,EACf,MAAM,EACN,WAAW,EACX,WAAW,EACX,YAAY,EACZ,OAAO,GAAG,KAAK,EACf,cAAc,EACd,oBAAoB,EACpB,mBAAmB,GAgCpB;IACC,MAAM,UAAU,GAAG,aAAa,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC;IAC5C,MAAM,YAAY,GAA6B,EAAE,CAAC;IAClD,IAAI,YAAY,GAAG,CAAC,CAAC;IAErB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,YAAY,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC7C,MAAM,KAAK,GAAG,YAAY,CAAC,CAAC,CAAC,CAAC;QAC9B,eAAe,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,YAAY,CAAC,MAAM,EAAE,CAAC,CAAC;QAErG,8DAA8D;QAC9D,MAAM,qBAAqB,GAAG,oBAAoB,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,SAAS;YAC5E,CAAC,CAAC,EAAE,GAAG,oBAAoB,EAAE,CAAC,KAAK,CAAC,SAAS,CAAC,EAAE,oBAAoB,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE;YAClF,CAAC,CAAC,oBAAoB,CAAC;QACzB,MAAM,cAAc,GAAG,gBAAgB,CACrC,OAAO,CAAC,QAAQ,CAAC,EACjB,CAAC,UAAU,EAAE,KAAK,CAAC,SAAS,CAAC,EAC7B,EAAE,cAAc,EAAE,oBAAoB,EAAE,qBAAqB,EAAE,CAChE,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE;YACjB,IAAI,QAAQ,CAAC,IAAI,KAAK,KAAK,CAAC,SAAS,EAAE,CAAC;gBACtC,OAAO,EAAE,GAAG,QAAQ,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,cAAc,EAAE,WAAoB,EAAE,CAAC;YACjF,CAAC;YACD,OAAO,EAAE,GAAG,QAAQ,EAAE,cAAc,EAAE,SAAkB,EAAE,CAAC;QAC7D,CAAC,CAAC,CAAC;QACH,MAAM,UAAU,GAAG,GAAG,UAAU,IAAI,MAAM,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,IAAI,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;QAClG,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,GAAG,MAAM,mBAAmB,CAAC;YACrD,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,QAAQ;YACR,SAAS,EAAE,cAAc;YACzB,KAAK;YACL,UAAU;YACV,SAAS;YACT,OAAO;YACP,WAAW;YACX,SAAS;YACT,OAAO;YACP,YAAY;YACZ,iBAAiB;YACjB,QAAQ,EAAE,IAAI;YACd,UAAU,EAAE,KAAK;YACjB,UAAU;YACV,aAAa,EAAE,aAAa,IAAI,CAAC,GAAG,CAAC;YACrC,SAAS;YACT,OAAO;YACP,KAAK,EAAE,UAAU;YACjB,WAAW;YACX,WAAW;YACX,YAAY;SACb,CAAC,CAAC;QAEH,YAAY,CAAC,IAAI,CAAC;YAChB,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,MAAM;YACN,QAAQ;SACT,CAAC,CAAC;QAEH,YAAY,IAAI,MAAM,CAAC,IAAI,CAAC,YAAY,CAAC;QACzC,eAAe,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,YAAY,CAAC,MAAM,EAAE,CAAC,CAAC;IACtG,CAAC;IAED,MAAM,EAAE,MAAM,EAAE,WAAW,EAAE,GAAG,EAAE,GAAG,0BAA0B,CAAC;QAC9D,UAAU;QACV,QAAQ;QACR,YAAY;QACZ,YAAY;QACZ,KAAK;QACL,UAAU;QACV,OAAO;QACP,YAAY;QACZ,iBAAiB;QACjB,OAAO;QACP,KAAK;QACL,IAAI;QACJ,WAAW;QACX,SAAS;QACT,YAAY;QACZ,MAAM;QACN,WAAW;QACX,OAAO;KACR,CAAC,CAAC;IACH,MAAM,QAAQ,GAAG,aAAa,CAAC,WAAW,EAAE,SAAS,CAAC,CAAC;IACvD,MAAM,gBAAgB,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,QAAQ,IAAI,kBAAkB,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAChG,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IAClE,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,QAAQ,EAAE,CAAC;AAC3C,CAAC"}
@@ -57,8 +57,8 @@ export interface EvaluationPipelineOptions {
57
57
  layeredStats?: boolean;
58
58
  /** 透传到 meta.request.repeat */
59
59
  repeat?: number;
60
- /** 透传到 meta.request.each */
61
- each?: boolean;
60
+ /** 透传到 meta.request.batch */
61
+ batch?: boolean;
62
62
  /** 透传到 meta.request.judgeRepeat 与 grade(),每条 sample × dimension judge N 次 */
63
63
  judgeRepeat?: number;
64
64
  /** Multi-judge ensemble configs (≥ 2 entries triggers ensemble mode). */
@@ -75,8 +75,10 @@ export interface EvaluationPipelineOptions {
75
75
  * isolation-disabled pre-flight warnings). True/undefined = default behavior
76
76
  * (no warning); false = user explicitly disabled, warn if ~/.claude/skills/ has content. */
77
77
  strictBaseline?: boolean;
78
+ /** Explicit persisted run id. Used by batch workflows that need stable child ids. */
79
+ runId?: string;
78
80
  }
79
- export declare function executeEvaluationPipeline({ samplesPath, skillDir, samples, tasks, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, executor, judgeExecutor, outputDir, project, owner, tags, blind, concurrency, timeoutMs, noCache, jobStore, persistJob, onProgress, skipPreflight, verbose, retry, existingResults, requires, layeredStats, repeat, each, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, strictBaseline, }: EvaluationPipelineOptions): Promise<{
81
+ export declare function executeEvaluationPipeline({ samplesPath, skillDir, samples, tasks, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, executor, judgeExecutor, outputDir, project, owner, tags, blind, concurrency, timeoutMs, noCache, jobStore, persistJob, onProgress, skipPreflight, verbose, retry, existingResults, requires, layeredStats, repeat, batch, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, strictBaseline, runId, }: EvaluationPipelineOptions): Promise<{
80
82
  report: Report;
81
83
  filePath: string | null;
82
84
  }>;
@@ -1 +1 @@
1
- {"version":3,"file":"evaluation-pipeline.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AAgBjF,OAAO,KAAK,EACV,QAAQ,EAIR,UAAU,EACV,QAAQ,EACR,gBAAgB,EAChB,MAAM,EACN,MAAM,EACN,IAAI,EACJ,aAAa,EACd,MAAM,mBAAmB,CAAC;AA4J3B;;;;;;;;;;;;;GAaG;AACH,wBAAgB,kBAAkB,CAAC,WAAW,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,GAAG,MAAM,EAAE,CAiBhF;AAQD;;;;;;;;GAQG;AACH,wBAAgB,sBAAsB,CACpC,SAAS,EAAE,QAAQ,EAAE,EACrB,cAAc,EAAE,OAAO,GAAG,SAAS,GAClC,MAAM,EAAE,CAsBV;AAgED,MAAM,WAAW,yBAAyB;IACxC,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,QAAQ,EAAE,UAAU,CAAC;IACrB,aAAa,EAAE,UAAU,CAAC;IAC1B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,QAAQ,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;IAC3B,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACrC,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,eAAe,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC,CAAC;IAChE,QAAQ,CAAC,EAAE,sBAAsB,CAAC;IAClC,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,8BAA8B;IAC9B,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,4BAA4B;IAC5B,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,6EAA6E;IAC7E,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,yEAAyE;IACzE,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,mBAAmB;IACnB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,2CAA2C;IAC3C,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,mFAAmF;IACnF,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,wBAAwB;IACxB,MAAM,CAAC,EAAE,OAAO,mBAAmB,EAAE,UAAU,CAAC;IAChD;;iGAE6F;IAC7F,cAAc,CAAC,EAAE,OAAO,CAAC;CAC1B;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,WAAW,EACX,QAAQ,EACR,OAAO,EACP,KAAK,EACL,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,QAAQ,EACR,aAAa,EACb,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAa,EACb,WAAe,EACf,SAAS,EACT,OAAe,EACf,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,aAAqB,EACrB,OAAe,EACf,KAAS,EACT,eAAe,EACf,QAAQ,EACR,YAAoB,EACpB,MAAM,EACN,IAAI,EACJ,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAmB,EACnB,MAAM,EACN,cAAc,GACf,EAAE,yBAAyB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CAwIlF"}
1
+ {"version":3,"file":"evaluation-pipeline.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AAgBjF,OAAO,KAAK,EACV,QAAQ,EAIR,UAAU,EACV,QAAQ,EACR,gBAAgB,EAChB,MAAM,EACN,MAAM,EACN,IAAI,EACJ,aAAa,EACd,MAAM,mBAAmB,CAAC;AA4J3B;;;;;;;;;;;;;GAaG;AACH,wBAAgB,kBAAkB,CAAC,WAAW,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,GAAG,MAAM,EAAE,CAiBhF;AAQD;;;;;;;;GAQG;AACH,wBAAgB,sBAAsB,CACpC,SAAS,EAAE,QAAQ,EAAE,EACrB,cAAc,EAAE,OAAO,GAAG,SAAS,GAClC,MAAM,EAAE,CAsBV;AAgED,MAAM,WAAW,yBAAyB;IACxC,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,QAAQ,EAAE,UAAU,CAAC;IACrB,aAAa,EAAE,UAAU,CAAC;IAC1B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,QAAQ,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;IAC3B,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACrC,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,eAAe,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC,CAAC;IAChE,QAAQ,CAAC,EAAE,sBAAsB,CAAC;IAClC,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,8BAA8B;IAC9B,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,6BAA6B;IAC7B,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,6EAA6E;IAC7E,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,yEAAyE;IACzE,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,mBAAmB;IACnB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,2CAA2C;IAC3C,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,mFAAmF;IACnF,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,wBAAwB;IACxB,MAAM,CAAC,EAAE,OAAO,mBAAmB,EAAE,UAAU,CAAC;IAChD;;iGAE6F;IAC7F,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,qFAAqF;IACrF,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,WAAW,EACX,QAAQ,EACR,OAAO,EACP,KAAK,EACL,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,QAAQ,EACR,aAAa,EACb,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAa,EACb,WAAe,EACf,SAAS,EACT,OAAe,EACf,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,aAAqB,EACrB,OAAe,EACf,KAAS,EACT,eAAe,EACf,QAAQ,EACR,YAAoB,EACpB,MAAM,EACN,KAAK,EACL,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAmB,EACnB,MAAM,EACN,cAAc,EACd,KAAK,GACN,EAAE,yBAAyB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CAyIlF"}
@@ -11,7 +11,7 @@ import { preflightDependencies, formatDependencyErrors } from '../eval-core/depe
11
11
  import { createFileJobStore, DEFAULT_JOBS_DIR, } from '../server/job-store.js';
12
12
  import { stopAllServers } from '../inputs/mcp-resolver.js';
13
13
  import { buildEvaluationRequest, createFailedJob, createEvaluationRun, createQueuedJob, createSucceededJob, finalizeEvaluationRun, markJobRunning, failEvaluationRun, } from '../eval-core/evaluation-job.js';
14
- async function initializeEvaluationRunState({ samplesPath, skillDir, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, concurrency, timeoutMs, noCache, blind, project, owner, tags, runId, jobStore, persistJob, repeat, each, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, }) {
14
+ async function initializeEvaluationRunState({ samplesPath, skillDir, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, concurrency, timeoutMs, noCache, blind, project, owner, tags, runId, jobStore, persistJob, repeat, batch, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, }) {
15
15
  const request = buildEvaluationRequest({
16
16
  samplesPath,
17
17
  skillDir,
@@ -30,7 +30,7 @@ async function initializeEvaluationRunState({ samplesPath, skillDir, artifacts,
30
30
  owner,
31
31
  tags,
32
32
  repeat,
33
- each,
33
+ batch,
34
34
  judgeRepeat,
35
35
  judgeModels,
36
36
  bootstrap,
@@ -201,7 +201,7 @@ function finalizeEvaluationReport({ report, results, artifacts, variantNames, bl
201
201
  }
202
202
  return report;
203
203
  }
204
- export async function executeEvaluationPipeline({ samplesPath, skillDir, samples, tasks, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, executor, judgeExecutor, outputDir = DEFAULT_OUTPUT_DIR, project, owner, tags, blind = false, concurrency = 1, timeoutMs, noCache = false, jobStore = null, persistJob = true, onProgress = null, skipPreflight = false, verbose = false, retry = 0, existingResults, requires, layeredStats = false, repeat, each, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias = true, budget, strictBaseline, }) {
204
+ export async function executeEvaluationPipeline({ samplesPath, skillDir, samples, tasks, artifacts, model, judgeModel, noJudge, executorName, judgeExecutorName, executor, judgeExecutor, outputDir = DEFAULT_OUTPUT_DIR, project, owner, tags, blind = false, concurrency = 1, timeoutMs, noCache = false, jobStore = null, persistJob = true, onProgress = null, skipPreflight = false, verbose = false, retry = 0, existingResults, requires, layeredStats = false, repeat, batch, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias = true, budget, strictBaseline, runId, }) {
205
205
  const variantNames = artifacts.map((artifact) => artifact.name);
206
206
  const runState = await initializeEvaluationRunState({
207
207
  samplesPath,
@@ -219,11 +219,11 @@ export async function executeEvaluationPipeline({ samplesPath, skillDir, samples
219
219
  project,
220
220
  owner,
221
221
  tags,
222
- runId: generateRunId(variantNames),
222
+ runId: runId ?? generateRunId(variantNames),
223
223
  jobStore,
224
224
  persistJob,
225
225
  repeat,
226
- each,
226
+ batch,
227
227
  judgeRepeat,
228
228
  judgeModels,
229
229
  bootstrap,
@@ -271,6 +271,7 @@ export async function executeEvaluationPipeline({ samplesPath, skillDir, samples
271
271
  const { results, totalCostUSD, skipped, budgetExhausted } = await executeTasks({
272
272
  tasks,
273
273
  executor,
274
+ executorName,
274
275
  judgeExecutor,
275
276
  model,
276
277
  judgeModel,
@@ -1 +1 @@
1
- {"version":3,"file":"evaluation-pipeline.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAChE,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,cAAc,EAAE,MAAM,mCAAmC,CAAC;AACnE,OAAO,EAAE,qBAAqB,EAAE,MAAM,kCAAkC,CAAC;AACzE,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,OAAO,EAAE,eAAe,EAAE,cAAc,EAAE,kBAAkB,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACzI,OAAO,EAAE,YAAY,EAAE,SAAS,EAAE,MAAM,sCAAsC,CAAC;AAC/E,OAAO,EAAE,qBAAqB,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AAEnG,OAAO,EACL,kBAAkB,EAClB,gBAAgB,GACjB,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAC3D,OAAO,EACL,sBAAsB,EACtB,eAAe,EACf,mBAAmB,EACnB,eAAe,EACf,kBAAkB,EAClB,qBAAqB,EACrB,cAAc,EACd,iBAAiB,GAClB,MAAM,gCAAgC,CAAC;AA4BxC,KAAK,UAAU,4BAA4B,CAAC,EAC1C,WAAW,EACX,QAAQ,EACR,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,WAAW,EACX,SAAS,EACT,OAAO,EACP,KAAK,EACL,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAK,EACL,QAAQ,EACR,UAAU,EACV,MAAM,EACN,IAAI,EACJ,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,GA4BP;IACC,MAAM,OAAO,GAAG,sBAAsB,CAAC;QACrC,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;QACvC,QAAQ,EAAE,YAAY;QACtB,aAAa,EAAE,iBAAiB;QAChC,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO;QACP,MAAM,EAAE,KAAK;QACb,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM;QACN,IAAI;QACJ,WAAW;QACX,WAAW;QACX,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;KACP,CAAC,CAAC;IACH,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC3C,MAAM,EAAE,GAAG,EAAE,UAAU,EAAE,SAAS,EAAE,GAAG,mBAAmB,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC7E,MAAM,KAAK,GAAG,OAAO,KAAK,EAAE,CAAC;IAC7B,MAAM,gBAAgB,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,QAAQ,IAAI,kBAAkB,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAChG,MAAM,SAAS,GAAG,eAAe,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC,CAAC;IACjE,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IACpE,MAAM,UAAU,GAAG,cAAc,CAAC,SAAS,EAAE,KAAK,EAAE,SAAS,CAAC,CAAC;IAC/D,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,KAAK,EAAE,UAAU,CAAC,CAAC;IACrE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,UAAU,EAAE,UAAU,EAAE,gBAAgB,EAAE,CAAC;AACnG,CAAC;AAED,SAAS,qBAAqB,CAAC,KAAyB;IACtD,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,GAAG,GAAG,qBAAqB,CAAC,KAAK,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IAChE,MAAM,GAAG,GAAG,kBAAkB,CAAC;QAC7B,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,QAAQ,EAAE,KAAK,CAAC,KAAK;QACrB,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,UAAU;KACX,CAAC,CAAC;IACH,OAAO,EAAE,GAAG,EAAE,GAAG,EAAE,CAAC;AACtB,CAAC;AAED,KAAK,UAAU,oBAAoB,CAAC,KAAyB,EAAE,GAAkB;IAC/E,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC;QAC3B,MAAM,KAAK,CAAC,gBAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IACtD,CAAC;AACH,CAAC;AAED,KAAK,UAAU,gBAAgB,CAAC,KAAyB,EAAE,GAAY;IACrE,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,SAAS,GAAG,eAAe,CAAC;QAChC,GAAG,EAAE,EAAE,GAAG,KAAK,CAAC,UAAU,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,UAAU,EAAE,SAAS,EAAE;QACnG,KAAK,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;QACvD,UAAU;KACX,CAAC,CAAC;IACH,KAAK,iBAAiB,CAAC,KAAK,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IACrD,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC;QAC3B,MAAM,KAAK,CAAC,gBAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC5D,CAAC;AACH,CAAC;AAED;;;;GAIG;AACH,SAAS,kBAAkB,CAAC,WAAmB;IAC7C,IAAI,CAAC,WAAW,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC;QAAE,OAAO,IAAI,CAAC;IAC1D,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;QACnD,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IACzE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED;;;;;;;;;;;;;GAaG;AACH,MAAM,UAAU,kBAAkB,CAAC,WAAmB,EAAE,MAAc;IACpE,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,IAAI,WAAW,GAAG,CAAC,EAAE,CAAC;QACpB,QAAQ,CAAC,IAAI,CACX,OAAO,WAAW,6GAA6G,CAChI,CAAC;IACJ,CAAC;SAAM,IAAI,WAAW,GAAG,EAAE,EAAE,CAAC;QAC5B,QAAQ,CAAC,IAAI,CACX,OAAO,WAAW,kIAAkI,CACrJ,CAAC;IACJ,CAAC;IACD,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;QACf,QAAQ,CAAC,IAAI,CACX,0IAA0I,CAC3I,CAAC;IACJ,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,SAAS,iBAAiB,CAAC,WAAmB,EAAE,MAAc;IAC5D,KAAK,MAAM,CAAC,IAAI,kBAAkB,CAAC,WAAW,EAAE,MAAM,CAAC,EAAE,CAAC;QACxD,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACjC,CAAC;AACH,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,sBAAsB,CACpC,SAAqB,EACrB,cAAmC;IAEnC,qDAAqD;IACrD,IAAI,cAAc,KAAK,KAAK;QAAE,OAAO,EAAE,CAAC;IAExC,MAAM,eAAe,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,UAAU,CAAC,CAAC;IACrE,IAAI,CAAC,eAAe;QAAE,OAAO,EAAE,CAAC;IAEhC,kFAAkF;IAClF,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC;IACvD,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC;QAAE,OAAO,EAAE,CAAC;IAEtC,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,CAAC;QACH,UAAU,GAAG,WAAW,CAAC,SAAS,CAAC,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC;IACvF,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;IACD,IAAI,UAAU,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAEhC,OAAO;QACL,mEAAmE,UAAU,+EAA+E;KAC7J,CAAC;AACJ,CAAC;AAED,SAAS,qBAAqB,CAAC,SAAqB,EAAE,cAAmC;IACvF,KAAK,MAAM,CAAC,IAAI,sBAAsB,CAAC,SAAS,EAAE,cAAc,CAAC,EAAE,CAAC;QAClE,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACjC,CAAC;AACH,CAAC;AAED,SAAS,wBAAwB,CAAC,EAChC,MAAM,EACN,OAAO,EACP,SAAS,EACT,YAAY,EACZ,KAAK,EACL,WAAW,EACX,OAAO,GASR;IACC,qEAAqE;IACrE,2EAA2E;IAC3E,0DAA0D;IAC1D,MAAM,CAAC,QAAQ,GAAG,cAAc,CAAC,MAAM,EAAE,EAAE,OAAO,EAAE,CAAC,CAAC;IAEtD,MAAM,WAAW,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,aAAa,EAAE,EAAE,CAAC,CACjE,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,aAAa,EAAE,EAAE,CAAC,aAAa,CAAC,SAAS,IAAI,aAAa,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CACpH,CAAC,CAAC;IACH,IAAI,WAAW,EAAE,CAAC;QAChB,MAAM,gBAAgB,GAAG,MAAM,CAAC,WAAW,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC5G,MAAM,YAAY,GAAG,MAAM,CAAC,WAAW,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,GAAG,IAAI,IAAI,CAAC,CAAC,CAAC,CAAC;QAC5G,MAAM,QAAQ,GAAG,qBAAqB,CAAC,MAAM,EAAE,gBAAgB,EAAE,YAAY,CAAC,CAAC;QAC/E,IAAI,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACrC,MAAM,CAAC,QAAS,CAAC,QAAQ,GAAG,QAAQ,CAAC;QACvC,CAAC;IACH,CAAC;IAED,6EAA6E;IAC7E,2EAA2E;IAC3E,gFAAgF;IAChF,MAAM,UAAU,GAAG,qBAAqB,CAAC,MAAM,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;IACvE,IAAI,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACvC,MAAM,WAAW,GAAG,kBAAkB,CAAC,WAAW,CAAC,CAAC;QACpD,KAAK,MAAM,OAAO,IAAI,YAAY,EAAE,CAAC;YACnC,MAAM,EAAE,GAAG,UAAU,CAAC,OAAO,CAAC,CAAC;YAC/B,IAAI,CAAC,EAAE;gBAAE,SAAS;YAClB,EAAE,CAAC,WAAW,GAAG,WAAW,CAAC;YAC7B,EAAE,CAAC,WAAW,GAAG,WAAW,CAAC;QAC/B,CAAC;QACD,MAAM,CAAC,QAAS,CAAC,UAAU,GAAG,UAAU,CAAC;IAC3C,CAAC;IAED,IAAI,KAAK,EAAE,CAAC;QACV,cAAc,CAAC,MAAM,EAAE,YAAY,EAAE,GAAG,YAAY,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,WAAW,EAAE,CAAC,CAAC;IACnF,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAsDD,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,EAC9C,WAAW,EACX,QAAQ,EACR,OAAO,EACP,KAAK,EACL,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,QAAQ,EACR,aAAa,EACb,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAK,GAAG,KAAK,EACb,WAAW,GAAG,CAAC,EACf,SAAS,EACT,OAAO,GAAG,KAAK,EACf,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,aAAa,GAAG,KAAK,EACrB,OAAO,GAAG,KAAK,EACf,KAAK,GAAG,CAAC,EACT,eAAe,EACf,QAAQ,EACR,YAAY,GAAG,KAAK,EACpB,MAAM,EACN,IAAI,EACJ,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,GAAG,IAAI,EACnB,MAAM,EACN,cAAc,GACY;IAC1B,MAAM,YAAY,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IAChE,MAAM,QAAQ,GAAG,MAAM,4BAA4B,CAAC;QAClD,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,UAAU;QACV,OAAO;QACP,YAAY;QACZ,iBAAiB;QACjB,WAAW;QACX,SAAS;QACT,OAAO;QACP,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,KAAK,EAAE,aAAa,CAAC,YAAY,CAAC;QAClC,QAAQ;QACR,UAAU;QACV,MAAM;QACN,IAAI;QACJ,WAAW;QACX,WAAW;QACX,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;KACP,CAAC,CAAC;IAEH,IAAI,CAAC;QACH,IAAI,CAAC,aAAa,EAAE,CAAC;YACnB,IAAI,UAAU;gBAAE,UAAU,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,KAAK,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC;YAC1E,MAAM,SAAS,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAC;YACjC,IAAI,CAAC,OAAO;gBAAE,MAAM,SAAS,CAAC,aAAa,EAAE,UAAU,CAAC,CAAC;YAEzD,+EAA+E;YAC/E,MAAM,aAAa,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAAC,CAAC;YACxG,MAAM,GAAG,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,GAAG,IAAI,QAAQ,IAAI,OAAO,CAAC,GAAG,EAAE,CAAC;YAC3E,MAAM,SAAS,GAAG,MAAM,qBAAqB,CAAC,aAAa,EAAE,OAAO,EAAE,GAAG,EAAE,QAAQ,EAAE,SAAS,CAAC,CAAC;YAChG,IAAI,CAAC,SAAS,CAAC,EAAE,EAAE,CAAC;gBAClB,MAAM,IAAI,KAAK,CAAC,sBAAsB,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC;YAC7D,CAAC;QACH,CAAC;QAED,6EAA6E;QAC7E,+EAA+E;QAC/E,+EAA+E;QAC/E,4DAA4D;QAC5D,iBAAiB,CAAC,OAAO,CAAC,MAAM,EAAE,MAAM,IAAI,CAAC,CAAC,CAAC;QAC/C,oFAAoF;QACpF,qBAAqB,CAAC,SAAS,EAAE,cAAc,CAAC,CAAC;QAEjD,8EAA8E;QAC9E,iFAAiF;QACjF,iFAAiF;QACjF,IAAI,cAAsD,CAAC;QAC3D,IAAI,WAAW,IAAI,WAAW,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;YAC3C,MAAM,EAAE,cAAc,EAAE,GAAG,MAAM,MAAM,CAAC,uBAAuB,CAAC,CAAC;YACjE,cAAc,GAAG,EAAE,CAAC;YACpB,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;YAC/B,KAAK,MAAM,EAAE,IAAI,WAAW,EAAE,CAAC;gBAC7B,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC;oBAC3B,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC;oBACtB,cAAc,CAAC,EAAE,CAAC,QAAQ,CAAC,GAAG,cAAc,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC;gBAC5D,CAAC;YACH,CAAC;QACH,CAAC;QAED,MAAM,EAAE,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,eAAe,EAAE,GAAG,MAAM,YAAY,CAAC;YAC7E,KAAK;YACL,QAAQ;YACR,aAAa;YACb,KAAK;YACL,UAAU;YACV,OAAO;YACP,WAAW;YACX,WAAW;YACX,SAAS;YACT,OAAO;YACP,OAAO;YACP,UAAU;YACV,KAAK;YACL,eAAe;YACf,WAAW;YACX,WAAW;YACX,cAAc;YACd,YAAY;YACZ,MAAM;SACP,CAAC,CAAC;QACH,IAAI,OAAO,GAAG,CAAC,IAAI,UAAU,EAAE,CAAC;YAC9B,UAAU,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,CAAC,MAAM,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,SAAS,EAAE,EAAE,EAAE,OAAO,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;QACzH,CAAC;QAED,MAAM,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,qBAAqB,CAAC,QAAQ,CAAC,CAAC;QACrD,MAAM,MAAM,GAAG,wBAAwB,CAAC;YACtC,MAAM,EAAE,eAAe,CAAC;gBACtB,KAAK,EAAE,QAAQ,CAAC,KAAK;gBACrB,QAAQ,EAAE,YAAY;gBACtB,KAAK;gBACL,UAAU;gBACV,OAAO;gBACP,YAAY;gBACZ,OAAO;gBACP,KAAK;gBACL,OAAO;gBACP,YAAY;gBACZ,SAAS;gBACT,OAAO,EAAE,QAAQ,CAAC,OAAO;gBACzB,GAAG;gBACH,GAAG;gBACH,YAAY;aACb,CAAC;YACF,OAAO;YACP,SAAS;YACT,YAAY;YACZ,KAAK;YACL,WAAW;YACX,OAAO;SACR,CAAC,CAAC;QACH,IAAI,eAAe,EAAE,CAAC;YACpB,MAAM,CAAC,IAAI,CAAC,eAAe,GAAG,IAAI,CAAC;QACrC,CAAC;QACD,IAAI,MAAM,EAAE,CAAC;YACX,MAAM,CAAC,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QAC9B,CAAC;QACD,MAAM,QAAQ,GAAG,aAAa,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;QAClD,MAAM,oBAAoB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QAC1C,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC;IAC9B,CAAC;IAAC,OAAO,GAAY,EAAE,CAAC;QACtB,MAAM,gBAAgB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QACtC,MAAM,GAAG,CAAC;IACZ,CAAC;YAAS,CAAC;QACT,MAAM,cAAc,EAAE,CAAC;IACzB,CAAC;AACH,CAAC"}
1
+ {"version":3,"file":"evaluation-pipeline.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAChE,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,cAAc,EAAE,MAAM,mCAAmC,CAAC;AACnE,OAAO,EAAE,qBAAqB,EAAE,MAAM,kCAAkC,CAAC;AACzE,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,OAAO,EAAE,eAAe,EAAE,cAAc,EAAE,kBAAkB,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AACzI,OAAO,EAAE,YAAY,EAAE,SAAS,EAAE,MAAM,sCAAsC,CAAC;AAC/E,OAAO,EAAE,qBAAqB,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AAEnG,OAAO,EACL,kBAAkB,EAClB,gBAAgB,GACjB,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAC3D,OAAO,EACL,sBAAsB,EACtB,eAAe,EACf,mBAAmB,EACnB,eAAe,EACf,kBAAkB,EAClB,qBAAqB,EACrB,cAAc,EACd,iBAAiB,GAClB,MAAM,gCAAgC,CAAC;AA4BxC,KAAK,UAAU,4BAA4B,CAAC,EAC1C,WAAW,EACX,QAAQ,EACR,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,WAAW,EACX,SAAS,EACT,OAAO,EACP,KAAK,EACL,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAK,EACL,QAAQ,EACR,UAAU,EACV,MAAM,EACN,KAAK,EACL,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,GA4BP;IACC,MAAM,OAAO,GAAG,sBAAsB,CAAC;QACrC,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;QACvC,QAAQ,EAAE,YAAY;QACtB,aAAa,EAAE,iBAAiB;QAChC,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO;QACP,MAAM,EAAE,KAAK;QACb,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM;QACN,KAAK;QACL,WAAW;QACX,WAAW;QACX,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;KACP,CAAC,CAAC;IACH,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC3C,MAAM,EAAE,GAAG,EAAE,UAAU,EAAE,SAAS,EAAE,GAAG,mBAAmB,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC7E,MAAM,KAAK,GAAG,OAAO,KAAK,EAAE,CAAC;IAC7B,MAAM,gBAAgB,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,QAAQ,IAAI,kBAAkB,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAChG,MAAM,SAAS,GAAG,eAAe,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC,CAAC;IACjE,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IACpE,MAAM,UAAU,GAAG,cAAc,CAAC,SAAS,EAAE,KAAK,EAAE,SAAS,CAAC,CAAC;IAC/D,IAAI,gBAAgB;QAAE,MAAM,gBAAgB,CAAC,IAAI,CAAC,KAAK,EAAE,UAAU,CAAC,CAAC;IACrE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,UAAU,EAAE,UAAU,EAAE,gBAAgB,EAAE,CAAC;AACnG,CAAC;AAED,SAAS,qBAAqB,CAAC,KAAyB;IACtD,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,GAAG,GAAG,qBAAqB,CAAC,KAAK,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IAChE,MAAM,GAAG,GAAG,kBAAkB,CAAC;QAC7B,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,QAAQ,EAAE,KAAK,CAAC,KAAK;QACrB,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,UAAU;KACX,CAAC,CAAC;IACH,OAAO,EAAE,GAAG,EAAE,GAAG,EAAE,CAAC;AACtB,CAAC;AAED,KAAK,UAAU,oBAAoB,CAAC,KAAyB,EAAE,GAAkB;IAC/E,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC;QAC3B,MAAM,KAAK,CAAC,gBAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IACtD,CAAC;AACH,CAAC;AAED,KAAK,UAAU,gBAAgB,CAAC,KAAyB,EAAE,GAAY;IACrE,MAAM,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC5C,MAAM,SAAS,GAAG,eAAe,CAAC;QAChC,GAAG,EAAE,EAAE,GAAG,KAAK,CAAC,UAAU,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,UAAU,EAAE,SAAS,EAAE;QACnG,KAAK,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;QACvD,UAAU;KACX,CAAC,CAAC;IACH,KAAK,iBAAiB,CAAC,KAAK,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IACrD,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC;QAC3B,MAAM,KAAK,CAAC,gBAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;IAC5D,CAAC;AACH,CAAC;AAED;;;;GAIG;AACH,SAAS,kBAAkB,CAAC,WAAmB;IAC7C,IAAI,CAAC,WAAW,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC;QAAE,OAAO,IAAI,CAAC;IAC1D,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;QACnD,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IACzE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED;;;;;;;;;;;;;GAaG;AACH,MAAM,UAAU,kBAAkB,CAAC,WAAmB,EAAE,MAAc;IACpE,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,IAAI,WAAW,GAAG,CAAC,EAAE,CAAC;QACpB,QAAQ,CAAC,IAAI,CACX,OAAO,WAAW,6GAA6G,CAChI,CAAC;IACJ,CAAC;SAAM,IAAI,WAAW,GAAG,EAAE,EAAE,CAAC;QAC5B,QAAQ,CAAC,IAAI,CACX,OAAO,WAAW,kIAAkI,CACrJ,CAAC;IACJ,CAAC;IACD,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;QACf,QAAQ,CAAC,IAAI,CACX,0IAA0I,CAC3I,CAAC;IACJ,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,SAAS,iBAAiB,CAAC,WAAmB,EAAE,MAAc;IAC5D,KAAK,MAAM,CAAC,IAAI,kBAAkB,CAAC,WAAW,EAAE,MAAM,CAAC,EAAE,CAAC;QACxD,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACjC,CAAC;AACH,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,sBAAsB,CACpC,SAAqB,EACrB,cAAmC;IAEnC,qDAAqD;IACrD,IAAI,cAAc,KAAK,KAAK;QAAE,OAAO,EAAE,CAAC;IAExC,MAAM,eAAe,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,UAAU,CAAC,CAAC;IACrE,IAAI,CAAC,eAAe;QAAE,OAAO,EAAE,CAAC;IAEhC,kFAAkF;IAClF,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC;IACvD,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC;QAAE,OAAO,EAAE,CAAC;IAEtC,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,CAAC;QACH,UAAU,GAAG,WAAW,CAAC,SAAS,CAAC,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC;IACvF,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;IACD,IAAI,UAAU,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAEhC,OAAO;QACL,mEAAmE,UAAU,+EAA+E;KAC7J,CAAC;AACJ,CAAC;AAED,SAAS,qBAAqB,CAAC,SAAqB,EAAE,cAAmC;IACvF,KAAK,MAAM,CAAC,IAAI,sBAAsB,CAAC,SAAS,EAAE,cAAc,CAAC,EAAE,CAAC;QAClE,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACjC,CAAC;AACH,CAAC;AAED,SAAS,wBAAwB,CAAC,EAChC,MAAM,EACN,OAAO,EACP,SAAS,EACT,YAAY,EACZ,KAAK,EACL,WAAW,EACX,OAAO,GASR;IACC,qEAAqE;IACrE,2EAA2E;IAC3E,0DAA0D;IAC1D,MAAM,CAAC,QAAQ,GAAG,cAAc,CAAC,MAAM,EAAE,EAAE,OAAO,EAAE,CAAC,CAAC;IAEtD,MAAM,WAAW,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,aAAa,EAAE,EAAE,CAAC,CACjE,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,aAAa,EAAE,EAAE,CAAC,aAAa,CAAC,SAAS,IAAI,aAAa,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CACpH,CAAC,CAAC;IACH,IAAI,WAAW,EAAE,CAAC;QAChB,MAAM,gBAAgB,GAAG,MAAM,CAAC,WAAW,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC5G,MAAM,YAAY,GAAG,MAAM,CAAC,WAAW,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,GAAG,IAAI,IAAI,CAAC,CAAC,CAAC,CAAC;QAC5G,MAAM,QAAQ,GAAG,qBAAqB,CAAC,MAAM,EAAE,gBAAgB,EAAE,YAAY,CAAC,CAAC;QAC/E,IAAI,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACrC,MAAM,CAAC,QAAS,CAAC,QAAQ,GAAG,QAAQ,CAAC;QACvC,CAAC;IACH,CAAC;IAED,6EAA6E;IAC7E,2EAA2E;IAC3E,gFAAgF;IAChF,MAAM,UAAU,GAAG,qBAAqB,CAAC,MAAM,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;IACvE,IAAI,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACvC,MAAM,WAAW,GAAG,kBAAkB,CAAC,WAAW,CAAC,CAAC;QACpD,KAAK,MAAM,OAAO,IAAI,YAAY,EAAE,CAAC;YACnC,MAAM,EAAE,GAAG,UAAU,CAAC,OAAO,CAAC,CAAC;YAC/B,IAAI,CAAC,EAAE;gBAAE,SAAS;YAClB,EAAE,CAAC,WAAW,GAAG,WAAW,CAAC;YAC7B,EAAE,CAAC,WAAW,GAAG,WAAW,CAAC;QAC/B,CAAC;QACD,MAAM,CAAC,QAAS,CAAC,UAAU,GAAG,UAAU,CAAC;IAC3C,CAAC;IAED,IAAI,KAAK,EAAE,CAAC;QACV,cAAc,CAAC,MAAM,EAAE,YAAY,EAAE,GAAG,YAAY,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,WAAW,EAAE,CAAC,CAAC;IACnF,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAwDD,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,EAC9C,WAAW,EACX,QAAQ,EACR,OAAO,EACP,KAAK,EACL,SAAS,EACT,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,QAAQ,EACR,aAAa,EACb,SAAS,GAAG,kBAAkB,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,KAAK,GAAG,KAAK,EACb,WAAW,GAAG,CAAC,EACf,SAAS,EACT,OAAO,GAAG,KAAK,EACf,QAAQ,GAAG,IAAI,EACf,UAAU,GAAG,IAAI,EACjB,UAAU,GAAG,IAAI,EACjB,aAAa,GAAG,KAAK,EACrB,OAAO,GAAG,KAAK,EACf,KAAK,GAAG,CAAC,EACT,eAAe,EACf,QAAQ,EACR,YAAY,GAAG,KAAK,EACpB,MAAM,EACN,KAAK,EACL,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,GAAG,IAAI,EACnB,MAAM,EACN,cAAc,EACd,KAAK,GACqB;IAC1B,MAAM,YAAY,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IAChE,MAAM,QAAQ,GAAG,MAAM,4BAA4B,CAAC;QAClD,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,UAAU;QACV,OAAO;QACP,YAAY;QACZ,iBAAiB;QACjB,WAAW;QACX,SAAS;QACT,OAAO;QACP,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,KAAK,EAAE,KAAK,IAAI,aAAa,CAAC,YAAY,CAAC;QAC3C,QAAQ;QACR,UAAU;QACV,MAAM;QACN,KAAK;QACL,WAAW;QACX,WAAW;QACX,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;KACP,CAAC,CAAC;IAEH,IAAI,CAAC;QACH,IAAI,CAAC,aAAa,EAAE,CAAC;YACnB,IAAI,UAAU;gBAAE,UAAU,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,KAAK,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC;YAC1E,MAAM,SAAS,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAC;YACjC,IAAI,CAAC,OAAO;gBAAE,MAAM,SAAS,CAAC,aAAa,EAAE,UAAU,CAAC,CAAC;YAEzD,+EAA+E;YAC/E,MAAM,aAAa,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAAC,CAAC;YACxG,MAAM,GAAG,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,GAAG,IAAI,QAAQ,IAAI,OAAO,CAAC,GAAG,EAAE,CAAC;YAC3E,MAAM,SAAS,GAAG,MAAM,qBAAqB,CAAC,aAAa,EAAE,OAAO,EAAE,GAAG,EAAE,QAAQ,EAAE,SAAS,CAAC,CAAC;YAChG,IAAI,CAAC,SAAS,CAAC,EAAE,EAAE,CAAC;gBAClB,MAAM,IAAI,KAAK,CAAC,sBAAsB,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC;YAC7D,CAAC;QACH,CAAC;QAED,6EAA6E;QAC7E,+EAA+E;QAC/E,+EAA+E;QAC/E,4DAA4D;QAC5D,iBAAiB,CAAC,OAAO,CAAC,MAAM,EAAE,MAAM,IAAI,CAAC,CAAC,CAAC;QAC/C,oFAAoF;QACpF,qBAAqB,CAAC,SAAS,EAAE,cAAc,CAAC,CAAC;QAEjD,8EAA8E;QAC9E,iFAAiF;QACjF,iFAAiF;QACjF,IAAI,cAAsD,CAAC;QAC3D,IAAI,WAAW,IAAI,WAAW,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;YAC3C,MAAM,EAAE,cAAc,EAAE,GAAG,MAAM,MAAM,CAAC,uBAAuB,CAAC,CAAC;YACjE,cAAc,GAAG,EAAE,CAAC;YACpB,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;YAC/B,KAAK,MAAM,EAAE,IAAI,WAAW,EAAE,CAAC;gBAC7B,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC;oBAC3B,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC;oBACtB,cAAc,CAAC,EAAE,CAAC,QAAQ,CAAC,GAAG,cAAc,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC;gBAC5D,CAAC;YACH,CAAC;QACH,CAAC;QAED,MAAM,EAAE,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,eAAe,EAAE,GAAG,MAAM,YAAY,CAAC;YAC7E,KAAK;YACL,QAAQ;YACR,YAAY;YACZ,aAAa;YACb,KAAK;YACL,UAAU;YACV,OAAO;YACP,WAAW;YACX,WAAW;YACX,SAAS;YACT,OAAO;YACP,OAAO;YACP,UAAU;YACV,KAAK;YACL,eAAe;YACf,WAAW;YACX,WAAW;YACX,cAAc;YACd,YAAY;YACZ,MAAM;SACP,CAAC,CAAC;QACH,IAAI,OAAO,GAAG,CAAC,IAAI,UAAU,EAAE,CAAC;YAC9B,UAAU,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,CAAC,MAAM,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,SAAS,EAAE,EAAE,EAAE,OAAO,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;QACzH,CAAC;QAED,MAAM,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,qBAAqB,CAAC,QAAQ,CAAC,CAAC;QACrD,MAAM,MAAM,GAAG,wBAAwB,CAAC;YACtC,MAAM,EAAE,eAAe,CAAC;gBACtB,KAAK,EAAE,QAAQ,CAAC,KAAK;gBACrB,QAAQ,EAAE,YAAY;gBACtB,KAAK;gBACL,UAAU;gBACV,OAAO;gBACP,YAAY;gBACZ,OAAO;gBACP,KAAK;gBACL,OAAO;gBACP,YAAY;gBACZ,SAAS;gBACT,OAAO,EAAE,QAAQ,CAAC,OAAO;gBACzB,GAAG;gBACH,GAAG;gBACH,YAAY;aACb,CAAC;YACF,OAAO;YACP,SAAS;YACT,YAAY;YACZ,KAAK;YACL,WAAW;YACX,OAAO;SACR,CAAC,CAAC;QACH,IAAI,eAAe,EAAE,CAAC;YACpB,MAAM,CAAC,IAAI,CAAC,eAAe,GAAG,IAAI,CAAC;QACrC,CAAC;QACD,IAAI,MAAM,EAAE,CAAC;YACX,MAAM,CAAC,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QAC9B,CAAC;QACD,MAAM,QAAQ,GAAG,aAAa,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;QAClD,MAAM,oBAAoB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QAC1C,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC;IAC9B,CAAC;IAAC,OAAO,GAAY,EAAE,CAAC;QACtB,MAAM,gBAAgB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QACtC,MAAM,GAAG,CAAC;IACZ,CAAC;YAAS,CAAC;QACT,MAAM,cAAc,EAAE,CAAC;IACzB,CAAC;AACH,CAAC"}
@@ -1,5 +1,5 @@
1
1
  import type { DependencyRequirements } from '../eval-core/dependency-checker.js';
2
- import type { Artifact, Sample, Task, VariantSpec } from '../types/index.js';
2
+ import type { Artifact, JudgeConfig, Sample, Task, VariantSpec } from '../types/index.js';
3
3
  export interface PreparedEvaluationRun {
4
4
  samples: Sample[];
5
5
  artifacts: Artifact[];
@@ -19,18 +19,16 @@ export declare function prepareEvaluationRun({ samplesPath, skillDir, variantSpe
19
19
  /** — eval.yaml 显式 allowedSkills per variant (overrides strictBaseline default). */
20
20
  variantAllowedSkills?: Record<string, string[]>;
21
21
  }): Promise<PreparedEvaluationRun>;
22
- export declare function buildDryRunTaskReport({ model, judgeModel, executorName, samplesPath, skillDir, tasks, variantNames, }: {
22
+ export declare function buildDryRunTaskReport({ model, judgeModel, judgeModels, executorName, samplesPath, skillDir, tasks, variantNames, }: {
23
23
  model: string;
24
24
  judgeModel: string;
25
+ judgeModels?: JudgeConfig[];
25
26
  executorName: string;
26
27
  samplesPath: string;
27
28
  skillDir: string;
28
29
  tasks: Task[];
29
30
  variantNames: string[];
30
31
  }): {
31
- dryRun: true;
32
- model: string;
33
- judgeModel: string;
34
32
  variants: string[];
35
33
  executor: string;
36
34
  samplesPath: string;
@@ -51,8 +49,12 @@ export declare function buildDryRunTaskReport({ model, judgeModel, executorName,
51
49
  hasDimensions: boolean;
52
50
  hasSystem: boolean;
53
51
  }[];
52
+ judgeModels?: string[] | undefined;
53
+ dryRun: true;
54
+ model: string;
55
+ judgeModel: string | null;
54
56
  };
55
- export declare function buildDryRunEachArtifacts(skillEntries: Array<{
57
+ export declare function buildDryRunBatchArtifacts(skillEntries: Array<{
56
58
  name: string;
57
59
  skillPath: string;
58
60
  samplesPath: string;
@@ -1 +1 @@
1
- {"version":3,"file":"evaluation-preparation.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-preparation.ts"],"names":[],"mappings":"AAOA,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AACjF,OAAO,KAAK,EAAE,QAAQ,EAAc,MAAM,EAAE,IAAI,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAEzF,MAAM,WAAW,qBAAqB;IACpC,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,QAAQ,CAAC,EAAE,sBAAsB,CAAC;CACnC;AAED,wBAAsB,oBAAoB,CAAC,EACzC,WAAW,EACX,QAAQ,EACR,YAAY,EACZ,SAAS,EACT,MAAM,EACN,SAAS,EACT,cAAc,EACd,oBAAoB,GACrB,EAAE;IACD,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,WAAW,EAAE,CAAC;IAC5B,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,MAAM,EAAE,OAAO,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,0DAA0D;IAC1D,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,oFAAoF;IACpF,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;CACjD,GAAG,OAAO,CAAC,qBAAqB,CAAC,CA6CjC;AAED,wBAAgB,qBAAqB,CAAC,EACpC,KAAK,EACL,UAAU,EACV,YAAY,EACZ,WAAW,EACX,QAAQ,EACR,KAAK,EACL,YAAY,GACb,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB;;;;;;;;;;;;;;;;;;;;;;;;EA6BA;AAED,wBAAgB,wBAAwB,CAAC,YAAY,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,SAAS,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,CAAC;;;;;;;;EAerH"}
1
+ {"version":3,"file":"evaluation-preparation.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-preparation.ts"],"names":[],"mappings":"AAOA,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,oCAAoC,CAAC;AACjF,OAAO,KAAK,EAAE,QAAQ,EAAE,WAAW,EAAc,MAAM,EAAE,IAAI,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAEtG,MAAM,WAAW,qBAAqB;IACpC,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,QAAQ,CAAC,EAAE,sBAAsB,CAAC;CACnC;AAED,wBAAsB,oBAAoB,CAAC,EACzC,WAAW,EACX,QAAQ,EACR,YAAY,EACZ,SAAS,EACT,MAAM,EACN,SAAS,EACT,cAAc,EACd,oBAAoB,GACrB,EAAE;IACD,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,WAAW,EAAE,CAAC;IAC5B,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,MAAM,EAAE,OAAO,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,0DAA0D;IAC1D,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,oFAAoF;IACpF,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;CACjD,GAAG,OAAO,CAAC,qBAAqB,CAAC,CA6CjC;AAED,wBAAgB,qBAAqB,CAAC,EACpC,KAAK,EACL,UAAU,EACV,WAAW,EACX,YAAY,EACZ,WAAW,EACX,QAAQ,EACR,KAAK,EACL,YAAY,GACb,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,WAAW,EAAE,CAAC;IAC5B,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB;;;;;;;;;;;;;;;;;;;;;;;;;EAiCA;AAED,wBAAgB,yBAAyB,CAAC,YAAY,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,SAAS,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,CAAC;;;;;;;;EAetH"}
@@ -16,7 +16,7 @@ export async function prepareEvaluationRun({ samplesPath, skillDir, variantSpecs
16
16
  roleByName[spec.name] = spec.role;
17
17
  for (const artifact of resolvedArtifacts) {
18
18
  if (artifact.experimentRole)
19
- continue; // preserve if already set (e.g. each-workflow)
19
+ continue; // preserve if already set (e.g. batch workflow)
20
20
  const role = roleByName[artifact.name];
21
21
  if (role)
22
22
  artifact.experimentRole = role;
@@ -42,11 +42,15 @@ export async function prepareEvaluationRun({ samplesPath, skillDir, variantSpecs
42
42
  requires,
43
43
  };
44
44
  }
45
- export function buildDryRunTaskReport({ model, judgeModel, executorName, samplesPath, skillDir, tasks, variantNames, }) {
45
+ export function buildDryRunTaskReport({ model, judgeModel, judgeModels, executorName, samplesPath, skillDir, tasks, variantNames, }) {
46
+ const judgeModelList = judgeModels && judgeModels.length >= 2
47
+ ? judgeModels.map((jc) => `${jc.executor}:${jc.model}`)
48
+ : undefined;
46
49
  return {
47
50
  dryRun: true,
48
51
  model,
49
- judgeModel,
52
+ judgeModel: judgeModelList ? null : judgeModel,
53
+ ...(judgeModelList ? { judgeModels: judgeModelList } : {}),
50
54
  variants: variantNames,
51
55
  executor: executorName,
52
56
  samplesPath,
@@ -72,7 +76,7 @@ export function buildDryRunTaskReport({ model, judgeModel, executorName, samples
72
76
  }),
73
77
  };
74
78
  }
75
- export function buildDryRunEachArtifacts(skillEntries) {
79
+ export function buildDryRunBatchArtifacts(skillEntries) {
76
80
  const artifacts = skillEntries.map((entry) => {
77
81
  const { samples } = loadSamples(entry.samplesPath);
78
82
  return {
@@ -1 +1 @@
1
- {"version":3,"file":"evaluation-preparation.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-preparation.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,uBAAuB,EAAE,MAAM,8BAA8B,CAAC;AACvE,OAAO,EAAE,WAAW,EAAE,MAAM,2BAA2B,CAAC;AACxD,OAAO,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AAC7D,OAAO,EAAE,kBAAkB,EAAE,MAAM,oCAAoC,CAAC;AACxE,OAAO,EAAE,aAAa,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAC1E,OAAO,EAAE,WAAW,EAAE,MAAM,0BAA0B,CAAC;AAYvD,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,EACzC,WAAW,EACX,QAAQ,EACR,YAAY,EACZ,SAAS,EACT,MAAM,EACN,SAAS,EACT,cAAc,EACd,oBAAoB,GAYrB;IACC,MAAM,EAAE,OAAO,EAAE,QAAQ,EAAE,GAAG,WAAW,CAAC,WAAW,CAAC,CAAC;IAEvD,4EAA4E;IAC5E,MAAM,kBAAkB,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACjE,MAAM,iBAAiB,GAAG,SAAS,IAAI,gBAAgB,CACrD,OAAO,CAAC,QAAQ,CAAC,EACjB,kBAAkB,EAClB,EAAE,cAAc,EAAE,oBAAoB,EAAE,CACzC,CAAC;IAEF,iFAAiF;IACjF,MAAM,UAAU,GAAwC,EAAE,CAAC;IAC3D,KAAK,MAAM,IAAI,IAAI,YAAY;QAAE,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC;IACnE,KAAK,MAAM,QAAQ,IAAI,iBAAiB,EAAE,CAAC;QACzC,IAAI,QAAQ,CAAC,cAAc;YAAE,SAAS,CAAE,+CAA+C;QACvF,MAAM,IAAI,GAAG,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;QACvC,IAAI,IAAI;YAAE,QAAQ,CAAC,cAAc,GAAG,IAAI,CAAC;IAC3C,CAAC;IAED,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,MAAM,UAAU,GAAsB,aAAa,CAAC,SAAS,CAAC,CAAC;QAC/D,MAAM,WAAW,GAAG,UAAU,CAAC,CAAC,CAAC,MAAM,cAAc,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC,CAAC,CAAC,IAAI,GAAG,EAAU,CAAC;QAC/F,MAAM,WAAW,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;IAC1C,CAAC;IAED,IAAI,iBAAiB,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACnC,MAAM,IAAI,KAAK,CACb,sBAAsB;cACpB,qBAAqB,OAAO,CAAC,QAAQ,CAAC,IAAI;cAC1C,sCAAsC;cACtC,wEAAwE,CAC3E,CAAC;IACJ,CAAC;IAED,MAAM,KAAK,GAAG,uBAAuB,CAAC,OAAO,EAAE,iBAAiB,CAAC,CAAC;IAClE,MAAM,YAAY,GAAG,iBAAiB,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IAExE,OAAO;QACL,OAAO;QACP,SAAS,EAAE,iBAAiB;QAC5B,KAAK;QACL,YAAY;QACZ,QAAQ;KACT,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,EACpC,KAAK,EACL,UAAU,EACV,YAAY,EACZ,WAAW,EACX,QAAQ,EACR,KAAK,EACL,YAAY,GASb;IACC,OAAO;QACL,MAAM,EAAE,IAAa;QACrB,KAAK;QACL,UAAU;QACV,QAAQ,EAAE,YAAY;QACtB,QAAQ,EAAE,YAAY;QACtB,WAAW;QACX,QAAQ;QACR,UAAU,EAAE,KAAK,CAAC,MAAM;QACxB,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;YACxB,MAAM,MAAM,GAAG,kBAAkB,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YACjD,OAAO;gBACL,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,OAAO,EAAE,IAAI,CAAC,OAAO;gBACrB,YAAY,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI;gBAChC,cAAc,EAAE,IAAI,CAAC,QAAQ,CAAC,MAAM;gBACpC,iBAAiB,EAAE,MAAM,CAAC,iBAAiB;gBAC3C,cAAc,EAAE,MAAM,CAAC,cAAc;gBACrC,cAAc,EAAE,MAAM,CAAC,cAAc;gBACrC,GAAG,EAAE,IAAI,CAAC,GAAG;gBACb,aAAa,EAAE,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC;gBACxC,SAAS,EAAE,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC;gBAC/B,aAAa,EAAE,OAAO,CAAC,IAAI,CAAC,UAAU,EAAE,MAAM,CAAC;gBAC/C,aAAa,EAAE,OAAO,CAAC,IAAI,CAAC,UAAU,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC;gBAC9E,SAAS,EAAE,OAAO,CAAC,IAAI,CAAC,eAAe,CAAC;aACzC,CAAC;QACJ,CAAC,CAAC;KACH,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,wBAAwB,CAAC,YAA6E;IACpH,MAAM,SAAS,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;QAC3C,MAAM,EAAE,OAAO,EAAE,GAAG,WAAW,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;QACnD,OAAO;YACL,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,WAAW,EAAE,OAAO,CAAC,MAAM;YAC3B,SAAS,EAAE,OAAO,CAAC,MAAM,GAAG,CAAC;SAC9B,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,OAAO;QACL,SAAS;QACT,UAAU,EAAE,SAAS,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,QAAQ,EAAE,EAAE,CAAC,GAAG,GAAG,QAAQ,CAAC,SAAS,EAAE,CAAC,CAAC;KAC7E,CAAC;AACJ,CAAC"}
1
+ {"version":3,"file":"evaluation-preparation.js","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-preparation.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,uBAAuB,EAAE,MAAM,8BAA8B,CAAC;AACvE,OAAO,EAAE,WAAW,EAAE,MAAM,2BAA2B,CAAC;AACxD,OAAO,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AAC7D,OAAO,EAAE,kBAAkB,EAAE,MAAM,oCAAoC,CAAC;AACxE,OAAO,EAAE,aAAa,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAC1E,OAAO,EAAE,WAAW,EAAE,MAAM,0BAA0B,CAAC;AAYvD,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,EACzC,WAAW,EACX,QAAQ,EACR,YAAY,EACZ,SAAS,EACT,MAAM,EACN,SAAS,EACT,cAAc,EACd,oBAAoB,GAYrB;IACC,MAAM,EAAE,OAAO,EAAE,QAAQ,EAAE,GAAG,WAAW,CAAC,WAAW,CAAC,CAAC;IAEvD,4EAA4E;IAC5E,MAAM,kBAAkB,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACjE,MAAM,iBAAiB,GAAG,SAAS,IAAI,gBAAgB,CACrD,OAAO,CAAC,QAAQ,CAAC,EACjB,kBAAkB,EAClB,EAAE,cAAc,EAAE,oBAAoB,EAAE,CACzC,CAAC;IAEF,iFAAiF;IACjF,MAAM,UAAU,GAAwC,EAAE,CAAC;IAC3D,KAAK,MAAM,IAAI,IAAI,YAAY;QAAE,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC;IACnE,KAAK,MAAM,QAAQ,IAAI,iBAAiB,EAAE,CAAC;QACzC,IAAI,QAAQ,CAAC,cAAc;YAAE,SAAS,CAAE,gDAAgD;QACxF,MAAM,IAAI,GAAG,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;QACvC,IAAI,IAAI;YAAE,QAAQ,CAAC,cAAc,GAAG,IAAI,CAAC;IAC3C,CAAC;IAED,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,MAAM,UAAU,GAAsB,aAAa,CAAC,SAAS,CAAC,CAAC;QAC/D,MAAM,WAAW,GAAG,UAAU,CAAC,CAAC,CAAC,MAAM,cAAc,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC,CAAC,CAAC,IAAI,GAAG,EAAU,CAAC;QAC/F,MAAM,WAAW,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;IAC1C,CAAC;IAED,IAAI,iBAAiB,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACnC,MAAM,IAAI,KAAK,CACb,sBAAsB;cACpB,qBAAqB,OAAO,CAAC,QAAQ,CAAC,IAAI;cAC1C,sCAAsC;cACtC,wEAAwE,CAC3E,CAAC;IACJ,CAAC;IAED,MAAM,KAAK,GAAG,uBAAuB,CAAC,OAAO,EAAE,iBAAiB,CAAC,CAAC;IAClE,MAAM,YAAY,GAAG,iBAAiB,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IAExE,OAAO;QACL,OAAO;QACP,SAAS,EAAE,iBAAiB;QAC5B,KAAK;QACL,YAAY;QACZ,QAAQ;KACT,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,EACpC,KAAK,EACL,UAAU,EACV,WAAW,EACX,YAAY,EACZ,WAAW,EACX,QAAQ,EACR,KAAK,EACL,YAAY,GAUb;IACC,MAAM,cAAc,GAAG,WAAW,IAAI,WAAW,CAAC,MAAM,IAAI,CAAC;QAC3D,CAAC,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,IAAI,EAAE,CAAC,KAAK,EAAE,CAAC;QACvD,CAAC,CAAC,SAAS,CAAC;IACd,OAAO;QACL,MAAM,EAAE,IAAa;QACrB,KAAK;QACL,UAAU,EAAE,cAAc,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU;QAC9C,GAAG,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC1D,QAAQ,EAAE,YAAY;QACtB,QAAQ,EAAE,YAAY;QACtB,WAAW;QACX,QAAQ;QACR,UAAU,EAAE,KAAK,CAAC,MAAM;QACxB,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;YACxB,MAAM,MAAM,GAAG,kBAAkB,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YACjD,OAAO;gBACL,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,OAAO,EAAE,IAAI,CAAC,OAAO;gBACrB,YAAY,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI;gBAChC,cAAc,EAAE,IAAI,CAAC,QAAQ,CAAC,MAAM;gBACpC,iBAAiB,EAAE,MAAM,CAAC,iBAAiB;gBAC3C,cAAc,EAAE,MAAM,CAAC,cAAc;gBACrC,cAAc,EAAE,MAAM,CAAC,cAAc;gBACrC,GAAG,EAAE,IAAI,CAAC,GAAG;gBACb,aAAa,EAAE,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC;gBACxC,SAAS,EAAE,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC;gBAC/B,aAAa,EAAE,OAAO,CAAC,IAAI,CAAC,UAAU,EAAE,MAAM,CAAC;gBAC/C,aAAa,EAAE,OAAO,CAAC,IAAI,CAAC,UAAU,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC;gBAC9E,SAAS,EAAE,OAAO,CAAC,IAAI,CAAC,eAAe,CAAC;aACzC,CAAC;QACJ,CAAC,CAAC;KACH,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,yBAAyB,CAAC,YAA6E;IACrH,MAAM,SAAS,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;QAC3C,MAAM,EAAE,OAAO,EAAE,GAAG,WAAW,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;QACnD,OAAO;YACL,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,WAAW,EAAE,OAAO,CAAC,MAAM;YAC3B,SAAS,EAAE,OAAO,CAAC,MAAM,GAAG,CAAC;SAC9B,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,OAAO;QACL,SAAS;QACT,UAAU,EAAE,SAAS,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,QAAQ,EAAE,EAAE,CAAC,GAAG,GAAG,QAAQ,CAAC,SAAS,EAAE,CAAC,CAAC;KAC7E,CAAC;AACJ,CAAC"}
@@ -1,4 +1,4 @@
1
- import type { Artifact, JobStore, ProgressCallback, Report, VarianceData, VariantSpec } from '../types/index.js';
1
+ import type { Artifact, BatchEvaluationReport, EvaluationReport, JobStore, ProgressCallback, Report, VarianceData, VariantSpec } from '../types/index.js';
2
2
  export interface SkillProgressInfo {
3
3
  phase: string;
4
4
  skill: string;
@@ -27,9 +27,9 @@ interface CommonEvaluationOptions {
27
27
  /** --repeat N. 1 表示单次(默认); > 1 时在 runMultiple 层聚合 variance。
28
28
  * 记入 report.meta.request.repeat 让 meta 如实反映用户输入。 */
29
29
  repeat?: number;
30
- /** --each 模式标记, true 表示当前评测是 each 批量流程(每个 skill 独立对比 baseline)。
31
- * 记入 report.meta.request.each。 */
32
- each?: boolean;
30
+ /** --batch 模式标记, true 表示当前评测是 skill batch 流程。
31
+ * 记入 report.meta.request.batch。 */
32
+ batch?: boolean;
33
33
  /** --judge-repeat N. 每条 sample × dimension 调 LLM judge N 次, 输出 stddev (judge 自一致性).
34
34
  * 默认 1 (单次). 用于量化 LLM judge 在该 rubric 上的稳定性 — stddev 高 = 评分噪声大. */
35
35
  judgeRepeat?: number;
@@ -45,6 +45,7 @@ interface CommonEvaluationOptions {
45
45
  lengthDebias?: boolean;
46
46
  /** hard budget caps. */
47
47
  budget?: import('../types/index.js').EvalBudget;
48
+ noCache?: boolean;
48
49
  }
49
50
  export interface RunEvaluationOptions extends CommonEvaluationOptions {
50
51
  samplesPath: string;
@@ -53,9 +54,10 @@ export interface RunEvaluationOptions extends CommonEvaluationOptions {
53
54
  artifacts?: Artifact[];
54
55
  dryRun?: boolean;
55
56
  blind?: boolean;
56
- noCache?: boolean;
57
57
  retry?: number;
58
58
  resume?: string;
59
+ /** Explicit persisted run id. Used by batch workflows that need stable child ids. */
60
+ runId?: string;
59
61
  /** strict-baseline default (CLI `--strict-baseline` default true).
60
62
  * When undefined, treated as true. baseline-kind variants get allowedSkills=[]
61
63
  * unless eval.yaml explicitly overrides per-variant. */
@@ -64,13 +66,13 @@ export interface RunEvaluationOptions extends CommonEvaluationOptions {
64
66
  * Keyed by variant name. Always wins over strictBaseline default. */
65
67
  variantAllowedSkills?: Record<string, string[]>;
66
68
  }
67
- export interface RunEachEvaluationOptions extends CommonEvaluationOptions {
69
+ export interface RunBatchEvaluationOptions extends CommonEvaluationOptions {
68
70
  skillDir: string;
69
71
  dryRun?: boolean;
70
72
  onSkillProgress?: ((info: SkillProgressInfo) => void) | null;
71
- /** strict-baseline default. each mode 仍尊重此 flag。 */
73
+ /** strict-baseline default. batch mode 仍尊重此 flag。 */
72
74
  strictBaseline?: boolean;
73
- /** per-variant allowedSkills override (rare in each mode but supported). */
75
+ /** per-variant allowedSkills override (rare in batch mode but supported). */
74
76
  variantAllowedSkills?: Record<string, string[]>;
75
77
  }
76
78
  export interface RunMultipleOptions extends RunEvaluationOptions {
@@ -98,7 +100,8 @@ interface DryRunTask {
98
100
  interface DryRunBase {
99
101
  dryRun: true;
100
102
  model: string;
101
- judgeModel: string;
103
+ judgeModel: string | null;
104
+ judgeModels?: string[];
102
105
  executor: string;
103
106
  skillDir: string;
104
107
  totalTasks: number;
@@ -108,28 +111,28 @@ export interface DryRunReport extends DryRunBase {
108
111
  samplesPath: string;
109
112
  tasks: DryRunTask[];
110
113
  }
111
- export declare function runEvaluation({ samplesPath, skillDir, variantSpecs, artifacts, model, judgeModel, outputDir, project, owner, tags, noJudge, dryRun, blind, concurrency, timeoutMs, noCache, executorName, judgeExecutorName, jobStore, persistJob, onProgress, skipPreflight, mcpConfig, verbose, retry, resume, layeredStats, repeat, each, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, strictBaseline, variantAllowedSkills, }: RunEvaluationOptions): Promise<{
114
+ export declare function runEvaluation({ samplesPath, skillDir, variantSpecs, artifacts, model, judgeModel, outputDir, project, owner, tags, noJudge, dryRun, blind, concurrency, timeoutMs, noCache, executorName, judgeExecutorName, jobStore, persistJob, onProgress, skipPreflight, mcpConfig, verbose, retry, resume, runId, layeredStats, repeat, batch, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, strictBaseline, variantAllowedSkills, }: RunEvaluationOptions): Promise<{
112
115
  report: Report | DryRunReport;
113
116
  filePath: string | null;
114
117
  }>;
115
- interface DryRunEachSkill {
118
+ interface DryRunBatchSkill {
116
119
  name: string;
117
120
  samplesPath: string;
118
121
  sampleCount: number;
119
122
  taskCount: number;
120
123
  }
121
- export interface DryRunEachReport extends DryRunBase {
122
- each: true;
124
+ export interface DryRunBatchReport extends DryRunBase {
125
+ batch: true;
123
126
  totalArtifacts: number;
124
- artifacts: DryRunEachSkill[];
127
+ artifacts: DryRunBatchSkill[];
125
128
  }
126
129
  export declare function buildVarianceData(runs: Report[]): VarianceData | null;
127
- export declare function runEachEvaluation({ skillDir, model, judgeModel, outputDir, project, owner, tags, noJudge, dryRun, concurrency, timeoutMs, executorName, judgeExecutorName, jobStore, persistJob, onProgress, onSkillProgress, skipPreflight, mcpConfig, verbose, repeat, judgeRepeat, judgeModels, lengthDebias, strictBaseline, variantAllowedSkills, }: RunEachEvaluationOptions): Promise<{
128
- report: Report | DryRunEachReport;
130
+ export declare function runBatchEvaluation({ skillDir, model, judgeModel, outputDir, project, owner, tags, noJudge, dryRun, concurrency, timeoutMs, executorName, judgeExecutorName, jobStore, persistJob, onProgress, onSkillProgress, skipPreflight, mcpConfig, verbose, repeat, judgeRepeat, judgeModels, lengthDebias, noCache, strictBaseline, variantAllowedSkills, }: RunBatchEvaluationOptions): Promise<{
131
+ report: BatchEvaluationReport | DryRunBatchReport;
129
132
  filePath: string | null;
130
133
  }>;
131
134
  export declare function runMultiple({ repeat, onRepeatProgress, ...config }: RunMultipleOptions): Promise<{
132
- report: Report;
135
+ report: EvaluationReport;
133
136
  aggregated: VarianceData | null;
134
137
  filePath: string | null;
135
138
  }>;
@@ -1 +1 @@
1
- {"version":3,"file":"run-evaluation.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/run-evaluation.ts"],"names":[],"mappings":"AAaA,OAAO,KAAK,EACV,QAAQ,EAER,QAAQ,EACR,gBAAgB,EAChB,MAAM,EAIN,YAAY,EAGZ,WAAW,EAGZ,MAAM,mBAAmB,CAAC;AAI3B,MAAM,WAAW,iBAAiB;IAChC,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,UAAU,uBAAuB;IAC/B,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;IAC3B,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACrC,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAIlB,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB;yDACqD;IACrD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB;uCACmC;IACnC,IAAI,CAAC,EAAE,OAAO,CAAC;IACf;wEACoE;IACpE,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;iFAC6E;IAC7E,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,8EAA8E;IAC9E,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,2CAA2C;IAC3C,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B;2DACuD;IACvD,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,wBAAwB;IACxB,MAAM,CAAC,EAAE,OAAO,mBAAmB,EAAE,UAAU,CAAC;CACjD;AAED,MAAM,WAAW,oBAAqB,SAAQ,uBAAuB;IACnE,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,CAAC,EAAE,WAAW,EAAE,CAAC;IAC7B,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB;;6DAEyD;IACzD,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB;0EACsE;IACtE,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;CACjD;AAED,MAAM,WAAW,wBAAyB,SAAQ,uBAAuB;IACvE,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,eAAe,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE,iBAAiB,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;IAC7D,oDAAoD;IACpD,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,4EAA4E;IAC5E,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;CACjD;AAED,MAAM,WAAW,kBAAmB,SAAQ,oBAAoB;IAC9D,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,gBAAgB,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;CAC5E;AAED,UAAU,UAAU;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;IAC/B,cAAc,EAAE,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACnC,iBAAiB,EAAE,MAAM,CAAC;IAC1B,cAAc,EAAE,MAAM,CAAC;IACvB,cAAc,EAAE,MAAM,CAAC;IACvB,GAAG,EAAE,MAAM,GAAG,IAAI,CAAC;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,OAAO,CAAC;IACnB,aAAa,EAAE,OAAO,CAAC;IACvB,aAAa,EAAE,OAAO,CAAC;IACvB,SAAS,EAAE,OAAO,CAAC;CACpB;AAED,UAAU,UAAU;IAClB,MAAM,EAAE,IAAI,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,YAAa,SAAQ,UAAU;IAC9C,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,UAAU,EAAE,CAAC;CACrB;AAED,wBAAsB,aAAa,CAAC,EAClC,WAAW,EACX,QAAQ,EACR,YAAiB,EACjB,SAAS,EACT,KAAqB,EACrB,UAAwB,EACxB,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,MAAc,EACd,KAAa,EACb,WAAe,EACf,SAAS,EACT,OAAe,EACf,YAAuB,EACvB,iBAAiB,EACjB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,aAAqB,EACrB,SAAS,EACT,OAAe,EACf,KAAS,EACT,MAAM,EACN,YAAoB,EACpB,MAAM,EACN,IAAI,EACJ,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,EACN,cAAc,EACd,oBAAoB,GACrB,EAAE,oBAAoB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,GAAG,YAAY,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CAmH5F;AAED,UAAU,eAAe;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,gBAAiB,SAAQ,UAAU;IAClD,IAAI,EAAE,IAAI,CAAC;IACX,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,eAAe,EAAE,CAAC;CAC9B;AAyHD,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,YAAY,GAAG,IAAI,CA6ErE;AAED,wBAAsB,iBAAiB,CAAC,EACtC,QAAQ,EACR,KAAqB,EACrB,UAAwB,EACxB,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,MAAc,EACd,WAAe,EACf,SAAS,EACT,YAAuB,EACvB,iBAAiB,EACjB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,eAAsB,EACtB,aAAqB,EACrB,SAAS,EACT,OAAe,EACf,MAAM,EACN,WAAW,EACX,WAAW,EACX,YAAY,EACZ,cAAc,EACd,oBAAoB,GACrB,EAAE,wBAAwB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,GAAG,gBAAgB,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CA4DpG;AAED,wBAAsB,WAAW,CAAC,EAAE,MAAU,EAAE,gBAAgB,EAAE,GAAG,MAAM,EAAE,EAAE,kBAAkB;;;;GAyBhG"}
1
+ {"version":3,"file":"run-evaluation.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/run-evaluation.ts"],"names":[],"mappings":"AAaA,OAAO,KAAK,EACV,QAAQ,EACR,qBAAqB,EACrB,gBAAgB,EAEhB,QAAQ,EACR,gBAAgB,EAChB,MAAM,EAIN,YAAY,EAGZ,WAAW,EAGZ,MAAM,mBAAmB,CAAC;AAI3B,MAAM,WAAW,iBAAiB;IAChC,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,UAAU,uBAAuB;IAC/B,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;IAC3B,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACrC,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAIlB,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB;yDACqD;IACrD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB;wCACoC;IACpC,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB;wEACoE;IACpE,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;iFAC6E;IAC7E,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,8EAA8E;IAC9E,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,2CAA2C;IAC3C,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B;2DACuD;IACvD,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,wBAAwB;IACxB,MAAM,CAAC,EAAE,OAAO,mBAAmB,EAAE,UAAU,CAAC;IAChD,OAAO,CAAC,EAAE,OAAO,CAAC;CACnB;AAED,MAAM,WAAW,oBAAqB,SAAQ,uBAAuB;IACnE,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,CAAC,EAAE,WAAW,EAAE,CAAC;IAC7B,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,qFAAqF;IACrF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;6DAEyD;IACzD,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB;0EACsE;IACtE,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;CACjD;AAED,MAAM,WAAW,yBAA0B,SAAQ,uBAAuB;IACxE,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,eAAe,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE,iBAAiB,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;IAC7D,qDAAqD;IACrD,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,6EAA6E;IAC7E,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;CACjD;AAED,MAAM,WAAW,kBAAmB,SAAQ,oBAAoB;IAC9D,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,gBAAgB,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;CAC5E;AAED,UAAU,UAAU;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;IAC/B,cAAc,EAAE,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACnC,iBAAiB,EAAE,MAAM,CAAC;IAC1B,cAAc,EAAE,MAAM,CAAC;IACvB,cAAc,EAAE,MAAM,CAAC;IACvB,GAAG,EAAE,MAAM,GAAG,IAAI,CAAC;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,OAAO,CAAC;IACnB,aAAa,EAAE,OAAO,CAAC;IACvB,aAAa,EAAE,OAAO,CAAC;IACvB,SAAS,EAAE,OAAO,CAAC;CACpB;AAED,UAAU,UAAU;IAClB,MAAM,EAAE,IAAI,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,WAAW,CAAC,EAAE,MAAM,EAAE,CAAC;IACvB,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,YAAa,SAAQ,UAAU;IAC9C,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,UAAU,EAAE,CAAC;CACrB;AAED,wBAAsB,aAAa,CAAC,EAClC,WAAW,EACX,QAAQ,EACR,YAAiB,EACjB,SAAS,EACT,KAAqB,EACrB,UAAwB,EACxB,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,MAAc,EACd,KAAa,EACb,WAAe,EACf,SAAS,EACT,OAAe,EACf,YAAuB,EACvB,iBAAiB,EACjB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,aAAqB,EACrB,SAAS,EACT,OAAe,EACf,KAAS,EACT,MAAM,EACN,KAAK,EACL,YAAoB,EACpB,MAAM,EACN,KAAK,EACL,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,EACN,cAAc,EACd,oBAAoB,GACrB,EAAE,oBAAoB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,GAAG,YAAY,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CAuH5F;AAED,UAAU,gBAAgB;IACxB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,iBAAkB,SAAQ,UAAU;IACnD,KAAK,EAAE,IAAI,CAAC;IACZ,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,gBAAgB,EAAE,CAAC;CAC/B;AAyHD,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,YAAY,GAAG,IAAI,CA6ErE;AAED,wBAAsB,kBAAkB,CAAC,EACvC,QAAQ,EACR,KAAqB,EACrB,UAAwB,EACxB,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,MAAc,EACd,WAAe,EACf,SAAS,EACT,YAAuB,EACvB,iBAAiB,EACjB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,eAAsB,EACtB,aAAqB,EACrB,SAAS,EACT,OAAe,EACf,MAAM,EACN,WAAW,EACX,WAAW,EACX,YAAY,EACZ,OAAe,EACf,cAAc,EACd,oBAAoB,GACrB,EAAE,yBAAyB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,qBAAqB,GAAG,iBAAiB,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CAiErH;AAED,wBAAsB,WAAW,CAAC,EAAE,MAAU,EAAE,gBAAgB,EAAE,GAAG,MAAM,EAAE,EAAE,kBAAkB;;;;GAyBhG"}