oh-my-knowledge 0.23.0 → 0.25.0

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 (196) hide show
  1. package/README.md +146 -84
  2. package/README.zh.md +144 -83
  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/analysis/saturation.d.ts +2 -2
  12. package/dist/src/analysis/saturation.js +2 -2
  13. package/dist/src/authoring/evolver.d.ts +12 -4
  14. package/dist/src/authoring/evolver.d.ts.map +1 -1
  15. package/dist/src/authoring/evolver.js +35 -10
  16. package/dist/src/authoring/evolver.js.map +1 -1
  17. package/dist/src/cli/i18n-dict.d.ts +1 -1
  18. package/dist/src/cli/i18n-dict.d.ts.map +1 -1
  19. package/dist/src/cli/i18n-dict.js +241 -64
  20. package/dist/src/cli/i18n-dict.js.map +1 -1
  21. package/dist/src/cli/index.js +267 -152
  22. package/dist/src/cli/index.js.map +1 -1
  23. package/dist/src/cli/parse-run-config.d.ts +32 -9
  24. package/dist/src/cli/parse-run-config.d.ts.map +1 -1
  25. package/dist/src/cli/parse-run-config.js +80 -25
  26. package/dist/src/cli/parse-run-config.js.map +1 -1
  27. package/dist/src/cli/parse-strict.d.ts +20 -0
  28. package/dist/src/cli/parse-strict.d.ts.map +1 -0
  29. package/dist/src/cli/parse-strict.js +25 -0
  30. package/dist/src/cli/parse-strict.js.map +1 -0
  31. package/dist/src/cli/progress.d.ts +1 -0
  32. package/dist/src/cli/progress.d.ts.map +1 -1
  33. package/dist/src/cli/progress.js +8 -1
  34. package/dist/src/cli/progress.js.map +1 -1
  35. package/dist/src/doctor/index.d.ts +19 -0
  36. package/dist/src/doctor/index.d.ts.map +1 -0
  37. package/dist/src/doctor/index.js +182 -0
  38. package/dist/src/doctor/index.js.map +1 -0
  39. package/dist/src/doctor/preflight.d.ts +32 -0
  40. package/dist/src/doctor/preflight.d.ts.map +1 -0
  41. package/dist/src/doctor/preflight.js +32 -0
  42. package/dist/src/doctor/preflight.js.map +1 -0
  43. package/dist/src/doctor/renderer.d.ts +13 -0
  44. package/dist/src/doctor/renderer.d.ts.map +1 -0
  45. package/dist/src/doctor/renderer.js +69 -0
  46. package/dist/src/doctor/renderer.js.map +1 -0
  47. package/dist/src/doctor/rules.d.ts +30 -0
  48. package/dist/src/doctor/rules.d.ts.map +1 -0
  49. package/dist/src/doctor/rules.js +216 -0
  50. package/dist/src/doctor/rules.js.map +1 -0
  51. package/dist/src/eval-core/bootstrap.d.ts +1 -1
  52. package/dist/src/eval-core/bootstrap.js +1 -1
  53. package/dist/src/eval-core/cache.d.ts +7 -5
  54. package/dist/src/eval-core/cache.d.ts.map +1 -1
  55. package/dist/src/eval-core/cache.js +11 -7
  56. package/dist/src/eval-core/cache.js.map +1 -1
  57. package/dist/src/eval-core/comparability.d.ts +11 -0
  58. package/dist/src/eval-core/comparability.d.ts.map +1 -0
  59. package/dist/src/eval-core/comparability.js +296 -0
  60. package/dist/src/eval-core/comparability.js.map +1 -0
  61. package/dist/src/eval-core/dependency-checker.js +1 -1
  62. package/dist/src/eval-core/dependency-checker.js.map +1 -1
  63. package/dist/src/eval-core/evaluation-execution.d.ts +23 -7
  64. package/dist/src/eval-core/evaluation-execution.d.ts.map +1 -1
  65. package/dist/src/eval-core/evaluation-execution.js +53 -21
  66. package/dist/src/eval-core/evaluation-execution.js.map +1 -1
  67. package/dist/src/eval-core/evaluation-job.d.ts +3 -5
  68. package/dist/src/eval-core/evaluation-job.d.ts.map +1 -1
  69. package/dist/src/eval-core/evaluation-job.js +2 -4
  70. package/dist/src/eval-core/evaluation-job.js.map +1 -1
  71. package/dist/src/eval-core/evaluation-reporting.d.ts +3 -1
  72. package/dist/src/eval-core/evaluation-reporting.d.ts.map +1 -1
  73. package/dist/src/eval-core/evaluation-reporting.js +67 -9
  74. package/dist/src/eval-core/evaluation-reporting.js.map +1 -1
  75. package/dist/src/eval-core/execution-strategy.js +2 -2
  76. package/dist/src/eval-core/execution-strategy.js.map +1 -1
  77. package/dist/src/eval-core/schema.d.ts.map +1 -1
  78. package/dist/src/eval-core/schema.js +20 -2
  79. package/dist/src/eval-core/schema.js.map +1 -1
  80. package/dist/src/eval-workflows/batch-evaluation-workflow.d.ts +113 -0
  81. package/dist/src/eval-workflows/batch-evaluation-workflow.d.ts.map +1 -0
  82. package/dist/src/eval-workflows/batch-evaluation-workflow.js +217 -0
  83. package/dist/src/eval-workflows/batch-evaluation-workflow.js.map +1 -0
  84. package/dist/src/eval-workflows/evaluation-pipeline.d.ts +6 -4
  85. package/dist/src/eval-workflows/evaluation-pipeline.d.ts.map +1 -1
  86. package/dist/src/eval-workflows/evaluation-pipeline.js +44 -38
  87. package/dist/src/eval-workflows/evaluation-pipeline.js.map +1 -1
  88. package/dist/src/eval-workflows/evaluation-preparation.d.ts +4 -17
  89. package/dist/src/eval-workflows/evaluation-preparation.d.ts.map +1 -1
  90. package/dist/src/eval-workflows/evaluation-preparation.js +3 -18
  91. package/dist/src/eval-workflows/evaluation-preparation.js.map +1 -1
  92. package/dist/src/eval-workflows/run-evaluation.d.ts +27 -23
  93. package/dist/src/eval-workflows/run-evaluation.d.ts.map +1 -1
  94. package/dist/src/eval-workflows/run-evaluation.js +137 -20
  95. package/dist/src/eval-workflows/run-evaluation.js.map +1 -1
  96. package/dist/src/executors/claude-cli.d.ts.map +1 -1
  97. package/dist/src/executors/claude-cli.js +11 -6
  98. package/dist/src/executors/claude-cli.js.map +1 -1
  99. package/dist/src/executors/codex-cli-trace.d.ts +10 -0
  100. package/dist/src/executors/codex-cli-trace.d.ts.map +1 -0
  101. package/dist/src/executors/codex-cli-trace.js +123 -0
  102. package/dist/src/executors/codex-cli-trace.js.map +1 -0
  103. package/dist/src/executors/codex-cli.d.ts +18 -0
  104. package/dist/src/executors/codex-cli.d.ts.map +1 -0
  105. package/dist/src/executors/codex-cli.js +254 -0
  106. package/dist/src/executors/codex-cli.js.map +1 -0
  107. package/dist/src/executors/codex-sdk.d.ts +18 -0
  108. package/dist/src/executors/codex-sdk.d.ts.map +1 -0
  109. package/dist/src/executors/codex-sdk.js +214 -0
  110. package/dist/src/executors/codex-sdk.js.map +1 -0
  111. package/dist/src/executors/gemini.d.ts.map +1 -1
  112. package/dist/src/executors/gemini.js +28 -24
  113. package/dist/src/executors/gemini.js.map +1 -1
  114. package/dist/src/executors/index.d.ts.map +1 -1
  115. package/dist/src/executors/index.js +7 -2
  116. package/dist/src/executors/index.js.map +1 -1
  117. package/dist/src/executors/runtime-fingerprint.d.ts +7 -0
  118. package/dist/src/executors/runtime-fingerprint.d.ts.map +1 -0
  119. package/dist/src/executors/runtime-fingerprint.js +277 -0
  120. package/dist/src/executors/runtime-fingerprint.js.map +1 -0
  121. package/dist/src/executors/script.d.ts.map +1 -1
  122. package/dist/src/executors/script.js +47 -55
  123. package/dist/src/executors/script.js.map +1 -1
  124. package/dist/src/executors/shared.d.ts +78 -1
  125. package/dist/src/executors/shared.d.ts.map +1 -1
  126. package/dist/src/executors/shared.js +203 -1
  127. package/dist/src/executors/shared.js.map +1 -1
  128. package/dist/src/grading/assertions.d.ts.map +1 -1
  129. package/dist/src/grading/assertions.js +22 -7
  130. package/dist/src/grading/assertions.js.map +1 -1
  131. package/dist/src/grading/gold-cli.js +4 -4
  132. package/dist/src/grading/gold-cli.js.map +1 -1
  133. package/dist/src/grading/human-gold.d.ts +1 -1
  134. package/dist/src/grading/human-gold.js +1 -1
  135. package/dist/src/grading/index.d.ts +20 -15
  136. package/dist/src/grading/index.d.ts.map +1 -1
  137. package/dist/src/grading/index.js +40 -16
  138. package/dist/src/grading/index.js.map +1 -1
  139. package/dist/src/grading/judge.d.ts +1 -1
  140. package/dist/src/grading/judge.d.ts.map +1 -1
  141. package/dist/src/grading/judge.js +76 -7
  142. package/dist/src/grading/judge.js.map +1 -1
  143. package/dist/src/inputs/eval-config.js +65 -7
  144. package/dist/src/inputs/eval-config.js.map +1 -1
  145. package/dist/src/inputs/skill-loader.d.ts +1 -1
  146. package/dist/src/inputs/skill-loader.d.ts.map +1 -1
  147. package/dist/src/inputs/skill-loader.js +1 -1
  148. package/dist/src/inputs/skill-loader.js.map +1 -1
  149. package/dist/src/renderer/html-renderer.d.ts +5 -4
  150. package/dist/src/renderer/html-renderer.d.ts.map +1 -1
  151. package/dist/src/renderer/html-renderer.js +240 -96
  152. package/dist/src/renderer/html-renderer.js.map +1 -1
  153. package/dist/src/renderer/layout.d.ts +2 -1
  154. package/dist/src/renderer/layout.d.ts.map +1 -1
  155. package/dist/src/renderer/layout.js +30 -42
  156. package/dist/src/renderer/layout.js.map +1 -1
  157. package/dist/src/renderer/summary.d.ts +3 -3
  158. package/dist/src/renderer/summary.d.ts.map +1 -1
  159. package/dist/src/renderer/summary.js +233 -43
  160. package/dist/src/renderer/summary.js.map +1 -1
  161. package/dist/src/renderer/trends.d.ts.map +1 -1
  162. package/dist/src/renderer/trends.js +5 -3
  163. package/dist/src/renderer/trends.js.map +1 -1
  164. package/dist/src/server/report-server.js +4 -4
  165. package/dist/src/server/report-server.js.map +1 -1
  166. package/dist/src/server/report-store.d.ts +7 -5
  167. package/dist/src/server/report-store.d.ts.map +1 -1
  168. package/dist/src/server/report-store.js +39 -11
  169. package/dist/src/server/report-store.js.map +1 -1
  170. package/dist/src/types/doctor.d.ts +95 -0
  171. package/dist/src/types/doctor.d.ts.map +1 -0
  172. package/dist/src/types/doctor.js +2 -0
  173. package/dist/src/types/doctor.js.map +1 -0
  174. package/dist/src/types/eval.d.ts +40 -19
  175. package/dist/src/types/eval.d.ts.map +1 -1
  176. package/dist/src/types/executor.d.ts +38 -0
  177. package/dist/src/types/executor.d.ts.map +1 -1
  178. package/dist/src/types/index.d.ts +1 -0
  179. package/dist/src/types/index.d.ts.map +1 -1
  180. package/dist/src/types/index.js +1 -0
  181. package/dist/src/types/index.js.map +1 -1
  182. package/dist/src/types/judge.d.ts +21 -0
  183. package/dist/src/types/judge.d.ts.map +1 -1
  184. package/dist/src/types/report.d.ts +100 -35
  185. package/dist/src/types/report.d.ts.map +1 -1
  186. package/dist/src/types/storage.d.ts +7 -7
  187. package/dist/src/types/storage.d.ts.map +1 -1
  188. package/package.json +6 -5
  189. package/dist/src/eval-workflows/each-evaluation-workflow.d.ts +0 -153
  190. package/dist/src/eval-workflows/each-evaluation-workflow.d.ts.map +0 -1
  191. package/dist/src/eval-workflows/each-evaluation-workflow.js +0 -178
  192. package/dist/src/eval-workflows/each-evaluation-workflow.js.map +0 -1
  193. package/dist/src/executors/openai-cli.d.ts +0 -3
  194. package/dist/src/executors/openai-cli.d.ts.map +0 -1
  195. package/dist/src/executors/openai-cli.js +0 -60
  196. package/dist/src/executors/openai-cli.js.map +0 -1
@@ -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":"AASA,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;AA6J3B;;;;;;;;;;;;;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,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAC3B,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,gBAAwB,EACxB,OAAe,EACf,KAAS,EACT,eAAe,EAGf,QAAQ,EAAE,SAAS,EACnB,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,CA0IlF"}
@@ -6,20 +6,20 @@ import { analyzeResults } from '../analysis/report-diagnostics.js';
6
6
  import { computeReportCoverage } from '../analysis/coverage-analyzer.js';
7
7
  import { computeReportGapRates } from '../analysis/gap-analyzer.js';
8
8
  import { aggregateReport, applyBlindMode, DEFAULT_OUTPUT_DIR, generateRunId, persistReport } from '../eval-core/evaluation-reporting.js';
9
- import { executeTasks, preflight } from '../eval-core/evaluation-execution.js';
10
- import { preflightDependencies, formatDependencyErrors } from '../eval-core/dependency-checker.js';
9
+ import { executeTasks, preflight, preflightAllJudges } from '../eval-core/evaluation-execution.js';
11
10
  import { createFileJobStore, DEFAULT_JOBS_DIR, } from '../server/job-store.js';
12
11
  import { stopAllServers } from '../inputs/mcp-resolver.js';
13
12
  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, }) {
13
+ 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, }) {
14
+ const effectiveJudges = judgeModels && judgeModels.length > 0
15
+ ? judgeModels
16
+ : [{ executor: judgeExecutorName, model: judgeModel }];
15
17
  const request = buildEvaluationRequest({
16
18
  samplesPath,
17
19
  skillDir,
18
20
  artifacts,
19
21
  model,
20
- judgeModel: noJudge ? null : judgeModel,
21
22
  executor: executorName,
22
- judgeExecutor: judgeExecutorName,
23
23
  noJudge,
24
24
  concurrency,
25
25
  timeoutMs,
@@ -30,9 +30,9 @@ async function initializeEvaluationRunState({ samplesPath, skillDir, artifacts,
30
30
  owner,
31
31
  tags,
32
32
  repeat,
33
- each,
33
+ batch,
34
34
  judgeRepeat,
35
- judgeModels,
35
+ judgeModels: effectiveJudges,
36
36
  bootstrap,
37
37
  bootstrapSamples,
38
38
  lengthDebias,
@@ -201,7 +201,10 @@ 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, skipConnectivity = false, verbose = false, retry = 0, existingResults,
205
+ // requires 现在由 runEvaluation 上游传给 doctor 处理; eval-pipeline 不再用
206
+ // 但保留接口字段,避免破 programmatic API (类型层面接收, 内部忽略)
207
+ requires: _requires, layeredStats = false, repeat, batch, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias = true, budget, strictBaseline, runId, }) {
205
208
  const variantNames = artifacts.map((artifact) => artifact.name);
206
209
  const runState = await initializeEvaluationRunState({
207
210
  samplesPath,
@@ -219,11 +222,11 @@ export async function executeEvaluationPipeline({ samplesPath, skillDir, samples
219
222
  project,
220
223
  owner,
221
224
  tags,
222
- runId: generateRunId(variantNames),
225
+ runId: runId ?? generateRunId(variantNames),
223
226
  jobStore,
224
227
  persistJob,
225
228
  repeat,
226
- each,
229
+ batch,
227
230
  judgeRepeat,
228
231
  judgeModels,
229
232
  bootstrap,
@@ -232,18 +235,37 @@ export async function executeEvaluationPipeline({ samplesPath, skillDir, samples
232
235
  budget,
233
236
  });
234
237
  try {
235
- if (!skipPreflight) {
238
+ // 解析 judge config + 配套 executor map 一次,后续 preflight 与 executeTasks 共用。
239
+ // production 路径 judgeModels 始终非空(RunConfig 必填);fallback 仅兼容 internal
240
+ // legacy caller 还在传 single judgeModel + judgeExecutorName 的场景。
241
+ //
242
+ // executor map 必须先于 preflight build,否则 ensemble 中第二、第三个 judge 配错
243
+ // (404 model / executor 不存在 / auth fail)只会在 grading 半途才暴露 — 那时已经
244
+ // 浪费 task execution 成本 + 报告里也不会有 judgeAgreement。
245
+ const resolvedJudgeModels = judgeModels && judgeModels.length > 0
246
+ ? judgeModels
247
+ : [{ executor: executorName, model: judgeModel }];
248
+ let resolvedJudgeExecutors;
249
+ if (judgeModels && judgeModels.length > 0) {
250
+ const { createExecutor } = await import('../executors/index.js');
251
+ resolvedJudgeExecutors = {};
252
+ for (const jc of resolvedJudgeModels) {
253
+ if (!resolvedJudgeExecutors[jc.executor]) {
254
+ resolvedJudgeExecutors[jc.executor] = createExecutor(jc.executor);
255
+ }
256
+ }
257
+ }
258
+ else {
259
+ resolvedJudgeExecutors = { [executorName]: judgeExecutor };
260
+ }
261
+ if (!skipConnectivity) {
236
262
  if (onProgress)
237
263
  onProgress({ phase: 'preflight', jobId: runState.jobId });
264
+ // LLM 连通性: eval 唯一职责。doctor 在 runEvaluation 上游已跑,
265
+ // 包含 dep / 结构 / 元数据 / 契约检查。这里只剩 executor + 所有 judge。
238
266
  await preflight(executor, model);
239
- if (!noJudge)
240
- await preflight(judgeExecutor, judgeModel);
241
- // Dependency check: auto-extract from skill contents + merge explicit requires
242
- const skillContents = artifacts.map((a) => a.content).filter((c) => typeof c === 'string');
243
- const cwd = artifacts.find((a) => a.cwd)?.cwd || skillDir || process.cwd();
244
- const depResult = await preflightDependencies(skillContents, samples, cwd, requires, artifacts);
245
- if (!depResult.ok) {
246
- throw new Error(formatDependencyErrors(depResult.missing));
267
+ if (!noJudge) {
268
+ await preflightAllJudges(resolvedJudgeModels, resolvedJudgeExecutors);
247
269
  }
248
270
  }
249
271
  // Structural power warnings — print to stderr after preflight passes, before
@@ -253,27 +275,11 @@ export async function executeEvaluationPipeline({ samplesPath, skillDir, samples
253
275
  emitPowerWarnings(samples.length, repeat ?? 1);
254
276
  // Isolation pre-flight warning (--no-strict-baseline + ~/.claude/skills/ non-empty)
255
277
  emitIsolationWarnings(artifacts, strictBaseline);
256
- // Pre-build a per-executor map for ensemble judges. Each unique executor name
257
- // gets one ExecutorFn, shared across all judges using that executor. The default
258
- // judge's executor is also included so single-judge fallbacks have it available.
259
- let judgeExecutors;
260
- if (judgeModels && judgeModels.length >= 2) {
261
- const { createExecutor } = await import('../executors/index.js');
262
- judgeExecutors = {};
263
- const seen = new Set();
264
- for (const jc of judgeModels) {
265
- if (!seen.has(jc.executor)) {
266
- seen.add(jc.executor);
267
- judgeExecutors[jc.executor] = createExecutor(jc.executor);
268
- }
269
- }
270
- }
271
278
  const { results, totalCostUSD, skipped, budgetExhausted } = await executeTasks({
272
279
  tasks,
273
280
  executor,
274
- judgeExecutor,
281
+ executorName,
275
282
  model,
276
- judgeModel,
277
283
  noJudge,
278
284
  samplesPath,
279
285
  concurrency,
@@ -284,8 +290,8 @@ export async function executeEvaluationPipeline({ samplesPath, skillDir, samples
284
290
  retry,
285
291
  existingResults,
286
292
  judgeRepeat,
287
- judgeModels,
288
- judgeExecutors,
293
+ judgeModels: resolvedJudgeModels,
294
+ judgeExecutors: resolvedJudgeExecutors,
289
295
  lengthDebias,
290
296
  budget,
291
297
  });
@@ -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,kBAAkB,EAAE,MAAM,sCAAsC,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,eAAe,GAA8C,WAAW,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC;QACtG,CAAC,CAAC,WAAW;QACb,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,iBAAiB,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;IACzD,MAAM,OAAO,GAAG,sBAAsB,CAAC;QACrC,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,QAAQ,EAAE,YAAY;QACtB,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,EAAE,eAAe;QAC5B,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,gBAAgB,GAAG,KAAK,EACxB,OAAO,GAAG,KAAK,EACf,KAAK,GAAG,CAAC,EACT,eAAe;AACf,+DAA+D;AAC/D,8CAA8C;AAC9C,QAAQ,EAAE,SAAS,EACnB,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,uEAAuE;QACvE,qEAAqE;QACrE,+DAA+D;QAC/D,EAAE;QACF,iEAAiE;QACjE,iEAAiE;QACjE,iDAAiD;QACjD,MAAM,mBAAmB,GAAG,WAAW,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC;YAC/D,CAAC,CAAC,WAAW;YACb,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,YAAY,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;QACpD,IAAI,sBAAkD,CAAC;QACvD,IAAI,WAAW,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC1C,MAAM,EAAE,cAAc,EAAE,GAAG,MAAM,MAAM,CAAC,uBAAuB,CAAC,CAAC;YACjE,sBAAsB,GAAG,EAAE,CAAC;YAC5B,KAAK,MAAM,EAAE,IAAI,mBAAmB,EAAE,CAAC;gBACrC,IAAI,CAAC,sBAAsB,CAAC,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC;oBACzC,sBAAsB,CAAC,EAAE,CAAC,QAAQ,CAAC,GAAG,cAAc,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC;gBACpE,CAAC;YACH,CAAC;QACH,CAAC;aAAM,CAAC;YACN,sBAAsB,GAAG,EAAE,CAAC,YAAY,CAAC,EAAE,aAAa,EAAE,CAAC;QAC7D,CAAC;QAED,IAAI,CAAC,gBAAgB,EAAE,CAAC;YACtB,IAAI,UAAU;gBAAE,UAAU,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,KAAK,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC;YAC1E,kDAAkD;YAClD,qDAAqD;YACrD,MAAM,SAAS,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAC;YACjC,IAAI,CAAC,OAAO,EAAE,CAAC;gBACb,MAAM,kBAAkB,CAAC,mBAAmB,EAAE,sBAAsB,CAAC,CAAC;YACxE,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,MAAM,EAAE,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,eAAe,EAAE,GAAG,MAAM,YAAY,CAAC;YAC7E,KAAK;YACL,QAAQ;YACR,YAAY;YACZ,KAAK;YACL,OAAO;YACP,WAAW;YACX,WAAW;YACX,SAAS;YACT,OAAO;YACP,OAAO;YACP,UAAU;YACV,KAAK;YACL,eAAe;YACf,WAAW;YACX,WAAW,EAAE,mBAAmB;YAChC,cAAc,EAAE,sBAAsB;YACtC,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[];
@@ -19,9 +19,9 @@ 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, judgeModels, executorName, samplesPath, skillDir, tasks, variantNames, }: {
23
23
  model: string;
24
- judgeModel: string;
24
+ judgeModels: JudgeConfig[];
25
25
  executorName: string;
26
26
  samplesPath: string;
27
27
  skillDir: string;
@@ -30,7 +30,7 @@ export declare function buildDryRunTaskReport({ model, judgeModel, executorName,
30
30
  }): {
31
31
  dryRun: true;
32
32
  model: string;
33
- judgeModel: string;
33
+ judgeModels: JudgeConfig[];
34
34
  variants: string[];
35
35
  executor: string;
36
36
  samplesPath: string;
@@ -52,17 +52,4 @@ export declare function buildDryRunTaskReport({ model, judgeModel, executorName,
52
52
  hasSystem: boolean;
53
53
  }[];
54
54
  };
55
- export declare function buildDryRunEachArtifacts(skillEntries: Array<{
56
- name: string;
57
- skillPath: string;
58
- samplesPath: string;
59
- }>): {
60
- artifacts: {
61
- name: string;
62
- samplesPath: string;
63
- sampleCount: number;
64
- taskCount: number;
65
- }[];
66
- totalTasks: number;
67
- };
68
55
  //# sourceMappingURL=evaluation-preparation.d.ts.map
@@ -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,WAAW,EACX,YAAY,EACZ,WAAW,EACX,QAAQ,EACR,KAAK,EACL,YAAY,GACb,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,WAAW,EAAE,CAAC;IAC3B,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"}
@@ -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 +42,11 @@ 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, judgeModels, executorName, samplesPath, skillDir, tasks, variantNames, }) {
46
46
  return {
47
47
  dryRun: true,
48
48
  model,
49
- judgeModel,
49
+ judgeModels,
50
50
  variants: variantNames,
51
51
  executor: executorName,
52
52
  samplesPath,
@@ -72,19 +72,4 @@ export function buildDryRunTaskReport({ model, judgeModel, executorName, samples
72
72
  }),
73
73
  };
74
74
  }
75
- export function buildDryRunEachArtifacts(skillEntries) {
76
- const artifacts = skillEntries.map((entry) => {
77
- const { samples } = loadSamples(entry.samplesPath);
78
- return {
79
- name: entry.name,
80
- samplesPath: entry.samplesPath,
81
- sampleCount: samples.length,
82
- taskCount: samples.length * 2,
83
- };
84
- });
85
- return {
86
- artifacts,
87
- totalTasks: artifacts.reduce((sum, artifact) => sum + artifact.taskCount, 0),
88
- };
89
- }
90
75
  //# sourceMappingURL=evaluation-preparation.js.map
@@ -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,WAAW,EACX,YAAY,EACZ,WAAW,EACX,QAAQ,EACR,KAAK,EACL,YAAY,GASb;IACC,OAAO;QACL,MAAM,EAAE,IAAa;QACrB,KAAK;QACL,WAAW;QACX,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"}
@@ -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;
@@ -7,7 +7,6 @@ export interface SkillProgressInfo {
7
7
  }
8
8
  interface CommonEvaluationOptions {
9
9
  model?: string;
10
- judgeModel?: string;
11
10
  outputDir?: string | null;
12
11
  project?: string;
13
12
  owner?: string;
@@ -16,35 +15,39 @@ interface CommonEvaluationOptions {
16
15
  concurrency?: number;
17
16
  timeoutMs?: number;
18
17
  executorName?: string;
19
- judgeExecutorName?: string;
20
18
  jobStore?: JobStore | null;
21
19
  persistJob?: boolean;
22
20
  onProgress?: ProgressCallback | null;
23
- skipPreflight?: boolean;
21
+ /** 跳过 LLM 模型连通性检测。--resume 时自动 true(已经验过)。 */
22
+ skipConnectivity?: boolean;
23
+ /** 用户语言, 透传给 doctor 报告渲染。 */
24
+ lang?: 'zh' | 'en';
24
25
  mcpConfig?: string;
25
26
  verbose?: boolean;
26
27
  layeredStats?: boolean;
27
28
  /** --repeat N. 1 表示单次(默认); > 1 时在 runMultiple 层聚合 variance。
28
29
  * 记入 report.meta.request.repeat 让 meta 如实反映用户输入。 */
29
30
  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;
31
+ /** --batch 模式标记, true 表示当前评测是 skill batch 流程。
32
+ * 记入 report.meta.request.batch。 */
33
+ batch?: boolean;
33
34
  /** --judge-repeat N. 每条 sample × dimension 调 LLM judge N 次, 输出 stddev (judge 自一致性).
34
35
  * 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
36
  judgeRepeat?: number;
36
- /** --judge-models executor:model,executor:model,... — multi-judge ensemble. ≥ 2 个
37
- * judge 时每条 sample × dimension 由所有 judge 各自打分, 输出 inter-judge agreement. */
37
+ /** Unified judge config. 1 entry = single judge, ≥ 2 = ensemble + inter-judge agreement.
38
+ * Optional at the API surface; runEvaluation defaults to `[{executor, model: 'haiku'}]`
39
+ * when omitted. RunConfig (parseRunConfig 出口) 保证非空。 */
38
40
  judgeModels?: import('../types/index.js').JudgeConfig[];
39
41
  /** --bootstrap. Distribution-free CI on each variant mean + pairwise diff. */
40
42
  bootstrap?: boolean;
41
43
  /** --bootstrap-samples N. Default 1000. */
42
44
  bootstrapSamples?: number;
43
- /** v0.21 Phase 3a length-debias toggle. Default true (judge prompt v3-cot-length).
45
+ /** length-debias toggle. Default true (judge prompt v3-cot-length).
44
46
  * CLI passes false when --no-debias-length is set. */
45
47
  lengthDebias?: boolean;
46
48
  /** hard budget caps. */
47
49
  budget?: import('../types/index.js').EvalBudget;
50
+ noCache?: boolean;
48
51
  }
49
52
  export interface RunEvaluationOptions extends CommonEvaluationOptions {
50
53
  samplesPath: string;
@@ -53,9 +56,10 @@ export interface RunEvaluationOptions extends CommonEvaluationOptions {
53
56
  artifacts?: Artifact[];
54
57
  dryRun?: boolean;
55
58
  blind?: boolean;
56
- noCache?: boolean;
57
59
  retry?: number;
58
60
  resume?: string;
61
+ /** Explicit persisted run id. Used by batch workflows that need stable child ids. */
62
+ runId?: string;
59
63
  /** strict-baseline default (CLI `--strict-baseline` default true).
60
64
  * When undefined, treated as true. baseline-kind variants get allowedSkills=[]
61
65
  * unless eval.yaml explicitly overrides per-variant. */
@@ -64,13 +68,13 @@ export interface RunEvaluationOptions extends CommonEvaluationOptions {
64
68
  * Keyed by variant name. Always wins over strictBaseline default. */
65
69
  variantAllowedSkills?: Record<string, string[]>;
66
70
  }
67
- export interface RunEachEvaluationOptions extends CommonEvaluationOptions {
71
+ export interface RunBatchEvaluationOptions extends CommonEvaluationOptions {
68
72
  skillDir: string;
69
73
  dryRun?: boolean;
70
74
  onSkillProgress?: ((info: SkillProgressInfo) => void) | null;
71
- /** strict-baseline default. each mode 仍尊重此 flag。 */
75
+ /** strict-baseline default. batch mode 仍尊重此 flag。 */
72
76
  strictBaseline?: boolean;
73
- /** per-variant allowedSkills override (rare in each mode but supported). */
77
+ /** per-variant allowedSkills override (rare in batch mode but supported). */
74
78
  variantAllowedSkills?: Record<string, string[]>;
75
79
  }
76
80
  export interface RunMultipleOptions extends RunEvaluationOptions {
@@ -98,7 +102,7 @@ interface DryRunTask {
98
102
  interface DryRunBase {
99
103
  dryRun: true;
100
104
  model: string;
101
- judgeModel: string;
105
+ judgeModels: import('../types/index.js').JudgeConfig[];
102
106
  executor: string;
103
107
  skillDir: string;
104
108
  totalTasks: number;
@@ -108,28 +112,28 @@ export interface DryRunReport extends DryRunBase {
108
112
  samplesPath: string;
109
113
  tasks: DryRunTask[];
110
114
  }
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<{
115
+ export declare function runEvaluation({ samplesPath, skillDir, variantSpecs, artifacts, model, outputDir, project, owner, tags, noJudge, dryRun, blind, concurrency, timeoutMs, noCache, executorName, jobStore, persistJob, onProgress, skipConnectivity, lang, mcpConfig, verbose, retry, resume, runId, layeredStats, repeat, batch, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, strictBaseline, variantAllowedSkills, }: RunEvaluationOptions): Promise<{
112
116
  report: Report | DryRunReport;
113
117
  filePath: string | null;
114
118
  }>;
115
- interface DryRunEachSkill {
119
+ interface DryRunBatchSkill {
116
120
  name: string;
117
121
  samplesPath: string;
118
122
  sampleCount: number;
119
123
  taskCount: number;
120
124
  }
121
- export interface DryRunEachReport extends DryRunBase {
122
- each: true;
125
+ export interface DryRunBatchReport extends DryRunBase {
126
+ batch: true;
123
127
  totalArtifacts: number;
124
- artifacts: DryRunEachSkill[];
128
+ artifacts: DryRunBatchSkill[];
125
129
  }
126
130
  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;
131
+ export declare function runBatchEvaluation({ skillDir, model, outputDir, project, owner, tags, noJudge, dryRun, concurrency, timeoutMs, executorName, jobStore, persistJob, onProgress, onSkillProgress, skipConnectivity, lang, mcpConfig, verbose, repeat, judgeRepeat, judgeModels, lengthDebias, noCache, strictBaseline, variantAllowedSkills, }: RunBatchEvaluationOptions): Promise<{
132
+ report: BatchEvaluationReport | DryRunBatchReport;
129
133
  filePath: string | null;
130
134
  }>;
131
135
  export declare function runMultiple({ repeat, onRepeatProgress, ...config }: RunMultipleOptions): Promise<{
132
- report: Report;
136
+ report: EvaluationReport;
133
137
  aggregated: VarianceData | null;
134
138
  filePath: string | null;
135
139
  }>;
@@ -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":"AAYA,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,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,QAAQ,CAAC,EAAE,QAAQ,GAAG,IAAI,CAAC;IAC3B,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACrC,8CAA8C;IAC9C,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAC3B,6BAA6B;IAC7B,IAAI,CAAC,EAAE,IAAI,GAAG,IAAI,CAAC;IACnB,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;;4DAEwD;IACxD,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,WAAW,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACvD,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,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,MAAc,EACd,KAAa,EACb,WAAe,EACf,SAAS,EACT,OAAe,EACf,YAAuB,EACvB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,gBAAwB,EACxB,IAAW,EACX,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,CAqK5F;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,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,MAAc,EACd,WAAe,EACf,SAAS,EACT,YAAuB,EACvB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,eAAsB,EACtB,gBAAwB,EACxB,IAAW,EACX,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,CAoIrH;AAED,wBAAsB,WAAW,CAAC,EAAE,MAAU,EAAE,gBAAgB,EAAE,GAAG,MAAM,EAAE,EAAE,kBAAkB;;;;GAyBhG"}