oh-my-knowledge 0.20.0 → 0.21.0

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 (203) hide show
  1. package/README.md +8 -4
  2. package/README.zh.md +8 -4
  3. package/dist/src/analysis/coverage-analyzer.d.ts +1 -1
  4. package/dist/src/analysis/coverage-analyzer.d.ts.map +1 -1
  5. package/dist/src/analysis/failure-clusterer.d.ts +1 -1
  6. package/dist/src/analysis/failure-clusterer.d.ts.map +1 -1
  7. package/dist/src/analysis/gap-analyzer.d.ts +2 -2
  8. package/dist/src/analysis/gap-analyzer.d.ts.map +1 -1
  9. package/dist/src/analysis/gap-analyzer.js +4 -4
  10. package/dist/src/analysis/hedging-classifier.d.ts +1 -1
  11. package/dist/src/analysis/hedging-classifier.d.ts.map +1 -1
  12. package/dist/src/analysis/report-diagnostics.d.ts +1 -1
  13. package/dist/src/analysis/report-diagnostics.d.ts.map +1 -1
  14. package/dist/src/analysis/report-diagnostics.js +7 -7
  15. package/dist/src/analysis/sample-diagnostics.d.ts +1 -1
  16. package/dist/src/analysis/sample-diagnostics.d.ts.map +1 -1
  17. package/dist/src/analysis/sample-diagnostics.js +15 -15
  18. package/dist/src/authoring/evolver.d.ts +1 -1
  19. package/dist/src/authoring/evolver.d.ts.map +1 -1
  20. package/dist/src/authoring/evolver.js +5 -5
  21. package/dist/src/authoring/evolver.js.map +1 -1
  22. package/dist/src/authoring/generator.d.ts +1 -1
  23. package/dist/src/authoring/generator.d.ts.map +1 -1
  24. package/dist/src/authoring/generator.js +8 -8
  25. package/dist/src/authoring/generator.js.map +1 -1
  26. package/dist/src/cli/i18n-dict.d.ts +56 -0
  27. package/dist/src/cli/i18n-dict.d.ts.map +1 -0
  28. package/dist/src/cli/i18n-dict.js +912 -0
  29. package/dist/src/cli/i18n-dict.js.map +1 -0
  30. package/dist/src/cli/i18n.d.ts +24 -0
  31. package/dist/src/cli/i18n.d.ts.map +1 -0
  32. package/dist/src/cli/i18n.js +53 -0
  33. package/dist/src/cli/i18n.js.map +1 -0
  34. package/dist/src/cli.js +279 -408
  35. package/dist/src/cli.js.map +1 -1
  36. package/dist/src/eval-core/cache.d.ts +1 -1
  37. package/dist/src/eval-core/cache.d.ts.map +1 -1
  38. package/dist/src/eval-core/dependency-checker.d.ts +16 -1
  39. package/dist/src/eval-core/dependency-checker.d.ts.map +1 -1
  40. package/dist/src/eval-core/dependency-checker.js +79 -4
  41. package/dist/src/eval-core/dependency-checker.js.map +1 -1
  42. package/dist/src/eval-core/evaluation-execution.d.ts +3 -3
  43. package/dist/src/eval-core/evaluation-execution.d.ts.map +1 -1
  44. package/dist/src/eval-core/evaluation-execution.js +1 -1
  45. package/dist/src/eval-core/evaluation-execution.js.map +1 -1
  46. package/dist/src/eval-core/evaluation-job.d.ts +2 -2
  47. package/dist/src/eval-core/evaluation-job.d.ts.map +1 -1
  48. package/dist/src/eval-core/evaluation-reporting.d.ts +1 -1
  49. package/dist/src/eval-core/evaluation-reporting.d.ts.map +1 -1
  50. package/dist/src/eval-core/execution-strategy.d.ts +1 -1
  51. package/dist/src/eval-core/execution-strategy.d.ts.map +1 -1
  52. package/dist/src/eval-core/layer-gates.d.ts +17 -0
  53. package/dist/src/eval-core/layer-gates.d.ts.map +1 -0
  54. package/dist/src/eval-core/{ci-gates.js → layer-gates.js} +5 -5
  55. package/dist/src/eval-core/layer-gates.js.map +1 -0
  56. package/dist/src/eval-core/schema.d.ts +1 -1
  57. package/dist/src/eval-core/schema.d.ts.map +1 -1
  58. package/dist/src/eval-core/schema.js +1 -1
  59. package/dist/src/eval-core/task-planner.d.ts +1 -1
  60. package/dist/src/eval-core/task-planner.d.ts.map +1 -1
  61. package/dist/src/eval-core/task-planner.js +1 -1
  62. package/dist/src/eval-core/task-planner.js.map +1 -1
  63. package/dist/src/eval-core/verdict.d.ts +5 -2
  64. package/dist/src/eval-core/verdict.d.ts.map +1 -1
  65. package/dist/src/eval-core/verdict.js +47 -8
  66. package/dist/src/eval-core/verdict.js.map +1 -1
  67. package/dist/src/eval-workflows/each-evaluation-workflow.d.ts +7 -7
  68. package/dist/src/eval-workflows/each-evaluation-workflow.d.ts.map +1 -1
  69. package/dist/src/eval-workflows/evaluation-pipeline.d.ts +18 -3
  70. package/dist/src/eval-workflows/evaluation-pipeline.d.ts.map +1 -1
  71. package/dist/src/eval-workflows/evaluation-pipeline.js +37 -0
  72. package/dist/src/eval-workflows/evaluation-pipeline.js.map +1 -1
  73. package/dist/src/eval-workflows/evaluation-preparation.d.ts +5 -5
  74. package/dist/src/eval-workflows/evaluation-preparation.d.ts.map +1 -1
  75. package/dist/src/eval-workflows/run-evaluation.d.ts +3 -3
  76. package/dist/src/eval-workflows/run-evaluation.d.ts.map +1 -1
  77. package/dist/src/eval-workflows/run-evaluation.js +9 -3
  78. package/dist/src/eval-workflows/run-evaluation.js.map +1 -1
  79. package/dist/src/executors/anthropic-api.d.ts +1 -1
  80. package/dist/src/executors/anthropic-api.d.ts.map +1 -1
  81. package/dist/src/executors/anthropic-api.js +1 -1
  82. package/dist/src/executors/anthropic-api.js.map +1 -1
  83. package/dist/src/executors/claude-cli.d.ts +1 -1
  84. package/dist/src/executors/claude-cli.d.ts.map +1 -1
  85. package/dist/src/executors/claude-sdk-trace.d.ts +1 -1
  86. package/dist/src/executors/claude-sdk-trace.d.ts.map +1 -1
  87. package/dist/src/executors/claude-sdk.d.ts +1 -1
  88. package/dist/src/executors/claude-sdk.d.ts.map +1 -1
  89. package/dist/src/executors/gemini.d.ts +1 -1
  90. package/dist/src/executors/gemini.d.ts.map +1 -1
  91. package/dist/src/executors/index.d.ts +1 -1
  92. package/dist/src/executors/index.d.ts.map +1 -1
  93. package/dist/src/executors/openai-api.d.ts +1 -1
  94. package/dist/src/executors/openai-api.d.ts.map +1 -1
  95. package/dist/src/executors/openai-api.js +1 -1
  96. package/dist/src/executors/openai-api.js.map +1 -1
  97. package/dist/src/executors/openai-cli.d.ts +1 -1
  98. package/dist/src/executors/openai-cli.d.ts.map +1 -1
  99. package/dist/src/executors/script.d.ts +1 -1
  100. package/dist/src/executors/script.d.ts.map +1 -1
  101. package/dist/src/executors/shared.d.ts +1 -1
  102. package/dist/src/executors/shared.d.ts.map +1 -1
  103. package/dist/src/grading/assertions.d.ts +1 -1
  104. package/dist/src/grading/assertions.d.ts.map +1 -1
  105. package/dist/src/grading/debias-validate.d.ts +1 -1
  106. package/dist/src/grading/debias-validate.d.ts.map +1 -1
  107. package/dist/src/grading/debias-validate.js +3 -3
  108. package/dist/src/grading/gold-cli.d.ts +1 -1
  109. package/dist/src/grading/gold-cli.d.ts.map +1 -1
  110. package/dist/src/grading/gold-cli.js +3 -3
  111. package/dist/src/grading/gold-cli.js.map +1 -1
  112. package/dist/src/grading/index.d.ts +1 -1
  113. package/dist/src/grading/index.d.ts.map +1 -1
  114. package/dist/src/grading/judge.d.ts +1 -1
  115. package/dist/src/grading/judge.d.ts.map +1 -1
  116. package/dist/src/grading/layered-scores.d.ts +1 -1
  117. package/dist/src/grading/layered-scores.d.ts.map +1 -1
  118. package/dist/src/inputs/eval-config.d.ts +1 -1
  119. package/dist/src/inputs/eval-config.d.ts.map +1 -1
  120. package/dist/src/inputs/eval-config.js +18 -18
  121. package/dist/src/inputs/eval-config.js.map +1 -1
  122. package/dist/src/inputs/load-samples.d.ts +1 -1
  123. package/dist/src/inputs/load-samples.d.ts.map +1 -1
  124. package/dist/src/inputs/load-samples.js +4 -4
  125. package/dist/src/inputs/load-samples.js.map +1 -1
  126. package/dist/src/inputs/mcp-resolver.d.ts +1 -1
  127. package/dist/src/inputs/mcp-resolver.d.ts.map +1 -1
  128. package/dist/src/inputs/mcp-resolver.js +4 -4
  129. package/dist/src/inputs/mcp-resolver.js.map +1 -1
  130. package/dist/src/inputs/skill-loader.d.ts +1 -1
  131. package/dist/src/inputs/skill-loader.d.ts.map +1 -1
  132. package/dist/src/inputs/skill-loader.js +12 -6
  133. package/dist/src/inputs/skill-loader.js.map +1 -1
  134. package/dist/src/inputs/url-fetcher.d.ts +1 -1
  135. package/dist/src/inputs/url-fetcher.d.ts.map +1 -1
  136. package/dist/src/inputs/url-fetcher.js +2 -2
  137. package/dist/src/inputs/url-fetcher.js.map +1 -1
  138. package/dist/src/observability/skill-health-analyzer.d.ts +1 -1
  139. package/dist/src/observability/skill-health-analyzer.d.ts.map +1 -1
  140. package/dist/src/observability/trace-adapter.d.ts +1 -1
  141. package/dist/src/observability/trace-adapter.d.ts.map +1 -1
  142. package/dist/src/renderer/html-renderer.d.ts +1 -1
  143. package/dist/src/renderer/html-renderer.d.ts.map +1 -1
  144. package/dist/src/renderer/html-renderer.js +43 -2
  145. package/dist/src/renderer/html-renderer.js.map +1 -1
  146. package/dist/src/renderer/layout.d.ts +1 -1
  147. package/dist/src/renderer/layout.d.ts.map +1 -1
  148. package/dist/src/renderer/layout.js +93 -1
  149. package/dist/src/renderer/layout.js.map +1 -1
  150. package/dist/src/renderer/skill-health-renderer.d.ts +1 -1
  151. package/dist/src/renderer/skill-health-renderer.d.ts.map +1 -1
  152. package/dist/src/renderer/summary.d.ts +4 -1
  153. package/dist/src/renderer/summary.d.ts.map +1 -1
  154. package/dist/src/renderer/summary.js +115 -54
  155. package/dist/src/renderer/summary.js.map +1 -1
  156. package/dist/src/renderer/table.d.ts +1 -1
  157. package/dist/src/renderer/table.d.ts.map +1 -1
  158. package/dist/src/renderer/trends.d.ts +1 -1
  159. package/dist/src/renderer/trends.d.ts.map +1 -1
  160. package/dist/src/server/job-store.d.ts +1 -1
  161. package/dist/src/server/job-store.d.ts.map +1 -1
  162. package/dist/src/server/report-server.d.ts +1 -1
  163. package/dist/src/server/report-server.d.ts.map +1 -1
  164. package/dist/src/server/report-server.js +5 -5
  165. package/dist/src/server/report-server.js.map +1 -1
  166. package/dist/src/server/report-store.d.ts +1 -1
  167. package/dist/src/server/report-store.d.ts.map +1 -1
  168. package/dist/src/types/eval.d.ts +240 -0
  169. package/dist/src/types/eval.d.ts.map +1 -0
  170. package/dist/src/types/eval.js +2 -0
  171. package/dist/src/types/eval.js.map +1 -0
  172. package/dist/src/types/executor.d.ts +65 -0
  173. package/dist/src/types/executor.d.ts.map +1 -0
  174. package/dist/src/types/executor.js +2 -0
  175. package/dist/src/types/executor.js.map +1 -0
  176. package/dist/src/types/index.d.ts +7 -0
  177. package/dist/src/types/index.d.ts.map +1 -0
  178. package/dist/src/types/index.js +7 -0
  179. package/dist/src/types/index.js.map +1 -0
  180. package/dist/src/types/judge.d.ts +98 -0
  181. package/dist/src/types/judge.d.ts.map +1 -0
  182. package/dist/src/types/judge.js +2 -0
  183. package/dist/src/types/judge.js.map +1 -0
  184. package/dist/src/types/report.d.ts +390 -0
  185. package/dist/src/types/report.d.ts.map +1 -0
  186. package/dist/src/types/report.js +2 -0
  187. package/dist/src/types/report.js.map +1 -0
  188. package/dist/src/types/shared.d.ts +2 -0
  189. package/dist/src/types/shared.d.ts.map +1 -0
  190. package/dist/src/types/shared.js +2 -0
  191. package/dist/src/types/shared.js.map +1 -0
  192. package/dist/src/types/storage.d.ts +21 -0
  193. package/dist/src/types/storage.d.ts.map +1 -0
  194. package/dist/src/types/storage.js +2 -0
  195. package/dist/src/types/storage.js.map +1 -0
  196. package/package.json +1 -1
  197. package/dist/src/eval-core/ci-gates.d.ts +0 -17
  198. package/dist/src/eval-core/ci-gates.d.ts.map +0 -1
  199. package/dist/src/eval-core/ci-gates.js.map +0 -1
  200. package/dist/src/types.d.ts +0 -800
  201. package/dist/src/types.d.ts.map +0 -1
  202. package/dist/src/types.js +0 -2
  203. package/dist/src/types.js.map +0 -1
@@ -1 +1 @@
1
- {"version":3,"file":"evaluation-execution.js","sourceRoot":"","sources":["../../../src/eval-core/evaluation-execution.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,QAAQ,EAAE,MAAM,YAAY,CAAC;AACnD,OAAO,EAAE,kBAAkB,EAAE,MAAM,aAAa,CAAC;AACjD,OAAO,EAAE,KAAK,EAAE,MAAM,qBAAqB,CAAC;AAC5C,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C,OAAO,EAAE,wBAAwB,EAAE,MAAM,yBAAyB,CAAC;AACnE,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACnD,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AA2ClC,KAAK,UAAU,kBAAkB,CAAI,KAAU,EAAE,WAAmB,EAAE,EAA8B;IAClG,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,UAAU,MAAM;QACnB,OAAO,KAAK,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC;YAC5B,MAAM,CAAC,GAAG,KAAK,EAAE,CAAC;YAClB,MAAM,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;QACrB,CAAC;IACH,CAAC;IACD,MAAM,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,WAAW,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,EAAE,GAAG,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;AACjG,CAAC;AAED,SAAS,eAAe,CAAC,KAAc;IACrC,MAAM,OAAO,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACvE,OAAO;QACL,EAAE,EAAE,KAAK;QACT,MAAM,EAAE,IAAI;QACZ,UAAU,EAAE,CAAC;QACb,aAAa,EAAE,CAAC;QAChB,WAAW,EAAE,CAAC;QACd,YAAY,EAAE,CAAC;QACf,eAAe,EAAE,CAAC;QAClB,mBAAmB,EAAE,CAAC;QACtB,OAAO,EAAE,CAAC;QACV,UAAU,EAAE,OAAO;QACnB,QAAQ,EAAE,CAAC;QACX,KAAK,EAAE,OAAO;KACf,CAAC;AACJ,CAAC;AAED,SAAS,KAAK,CAAC,EAAU;IACvB,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC;AAC3D,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,YAAY,CAAC,EACjC,KAAK,EACL,QAAQ,EACR,aAAa,EACb,KAAK,EACL,UAAU,EACV,OAAO,EACP,WAAW,EACX,WAAW,EACX,SAAS,EACT,OAAO,EACP,OAAO,EACP,UAAU,EACV,KAAK,GAAG,CAAC,EACT,eAAe,EACf,WAAW,GAAG,CAAC,EACf,WAAW,EACX,cAAc,EACd,YAAY,GAAG,IAAI,EACnB,MAAM,GACc;IACpB,MAAM,OAAO,GAAkD,EAAE,CAAC;IAClE,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,IAAI,eAAe,GAAG,KAAK,CAAC;IAE5B,4CAA4C;IAC5C,IAAI,eAAe,EAAE,CAAC;QACpB,KAAK,MAAM,CAAC,QAAQ,EAAE,QAAQ,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,eAAe,CAAC,EAAE,CAAC;YACnE,OAAO,CAAC,QAAQ,CAAC,GAAG,EAAE,GAAG,QAAQ,EAAE,CAAC;QACtC,CAAC;IACH,CAAC;IAED,MAAM,QAAQ,GAAG,IAAI,CAAC,OAAO,EAAE,EAAE,kBAAkB,EAAE,OAAO,CAAC,CAAC;IAC9D,MAAM,KAAK,GAAyB,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,WAAW,CAAC,QAAQ,CAAC,CAAC;IAE3E,KAAK,UAAU,WAAW,CAAC,IAAU;QACnC,sDAAsD;QACtD,IAAI,eAAe,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,EAAE,EAAE,CAAC;YAC1D,OAAO,EAAE,CAAC;YACV,OAAO,EAAE,CAAC;YACV,SAAS,EAAE,CAAC;YACZ,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;YACjI,OAAO;QACT,CAAC;QAED,yEAAyE;QACzE,0EAA0E;QAC1E,yCAAyC;QACzC,IAAI,eAAe,EAAE,CAAC;YACpB,OAAO,EAAE,CAAC;YACV,OAAO,EAAE,CAAC;YACV,SAAS,EAAE,CAAC;YACZ,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;YACjI,OAAO;QACT,CAAC;QAED,OAAO,EAAE,CAAC;QACV,MAAM,GAAG,GAAG,OAAO,CAAC;QACpB,MAAM,KAAK,GAAG,KAAK,CAAC,MAAM,CAAC;QAC3B,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;QAE1G,MAAM,aAAa,GAAG,wBAAwB,CAAC,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,OAAO,CAAC,CAAC;QAEhF,IAAI,UAAsB,CAAC;QAC3B,MAAM,GAAG,GAAG,QAAQ,CAAC,KAAK,EAAE,aAAa,CAAC,WAAW,EAAE,aAAa,CAAC,KAAK,CAAC,MAAM,EAAE,aAAa,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAC5G,MAAM,MAAM,GAAG,KAAK,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;QAC/B,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,IAAI,MAAM,EAAE,CAAC;YACX,UAAU,GAAG,EAAE,GAAG,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC;QAC3C,CAAC;aAAM,CAAC;YACN,gCAAgC;YAChC,MAAM,WAAW,GAAG,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC;YAC3C,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC;gBACxD,IAAI,CAAC;oBACH,UAAU,GAAG,MAAM,QAAQ,CAAC,aAAa,CAAC,KAAK,CAAC,CAAC;gBACnD,CAAC;gBAAC,OAAO,GAAG,EAAE,CAAC;oBACb,UAAU,GAAG,eAAe,CAAC,GAAG,CAAC,CAAC;gBACpC,CAAC;gBACD,IAAI,UAAW,CAAC,EAAE,IAAI,OAAO,KAAK,WAAW;oBAAE,MAAM;gBACrD,mCAAmC;gBACnC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC,GAAG,IAAI,EAAE,KAAK,CAAC,CAAC;gBAC7D,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,WAAW,EAAE,CAAC,CAAC;gBAChI,MAAM,KAAK,CAAC,SAAS,CAAC,CAAC;YACzB,CAAC;YACD,IAAI,KAAK,IAAI,UAAW,CAAC,EAAE;gBAAE,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE,UAAW,CAAC,CAAC;QAC3D,CAAC;QACD,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC;QACtC,YAAY,IAAI,UAAW,CAAC,OAAO,CAAC;QAEpC,IAAI,OAAO,IAAI,UAAU,EAAE,CAAC;YAC1B,UAAU,CAAC;gBACT,KAAK,EAAE,WAAW;gBAClB,QAAQ,EAAE,aAAa,CAAC,QAAQ;gBAChC,SAAS,EAAE,GAAG;gBACd,KAAK;gBACL,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,OAAO,EAAE,IAAI,CAAC,OAAO;gBACrB,UAAU,EAAE,UAAW,CAAC,UAAU;gBAClC,WAAW,EAAE,UAAW,CAAC,WAAW;gBACpC,YAAY,EAAE,UAAW,CAAC,YAAY;gBACtC,OAAO,EAAE,UAAW,CAAC,OAAO;gBAC5B,aAAa,EAAE,UAAW,CAAC,MAAM,CAAC,CAAC,CAAC,UAAW,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI;aAC5E,CAAC,CAAC;QACL,CAAC;QAED,IAAI,WAAW,GAAuB,IAAI,CAAC;QAC3C,IAAI,OAAO,GAAG,CAAC,CAAC;QAChB,IAAI,UAAW,CAAC,EAAE,EAAE,CAAC;YACnB,MAAM,kBAAkB,GAAG,IAAI,CAAC,MAAM,IAAI,IAAI,CAAC,UAAU,EAAE,MAAM,IAAI,CAAC,IAAI,CAAC,UAAU,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;YAC9H,IAAI,kBAAkB,EAAE,CAAC;gBACvB,IAAI,OAAO,IAAI,UAAU,EAAE,CAAC;oBAC1B,UAAU,CAAC;wBACT,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,aAAa,CAAC,QAAQ;wBAChC,SAAS,EAAE,GAAG;wBACd,KAAK;wBACL,SAAS,EAAE,IAAI,CAAC,SAAS;wBACzB,OAAO,EAAE,IAAI,CAAC,OAAO;qBACtB,CAAC,CAAC;gBACL,CAAC;gBACD,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;gBAC9B,IAAI,CAAC;oBACH,WAAW,GAAG,MAAM,KAAK,CAAC;wBACxB,MAAM,EAAE,UAAW,CAAC,MAAO;wBAC3B,MAAM,EAAE,IAAI,CAAC,OAAO;wBACpB,QAAQ,EAAE,aAAa;wBACvB,UAAU;wBACV,aAAa,EAAE,CAAC,OAAO;wBACvB,WAAW,EAAE;4BACX,OAAO,EAAE,UAAW,CAAC,OAAO;4BAC5B,UAAU,EAAE,UAAW,CAAC,UAAU;4BAClC,QAAQ,EAAE,UAAW,CAAC,QAAQ;4BAC9B,SAAS,EAAE,UAAW,CAAC,SAAS;4BAChC,KAAK,EAAE,UAAW,CAAC,KAAK;yBACzB;wBACD,UAAU,EAAE,OAAO,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC;wBACzC,WAAW;wBACX,WAAW;wBACX,cAAc;wBACd,YAAY;qBACb,CAAC,CAAC;gBACL,CAAC;gBAAC,OAAO,GAAG,EAAE,CAAC;oBACb,WAAW,GAAG,EAAE,cAAc,EAAE,CAAC,EAAE,CAAC;oBACpC,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;oBAC7D,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,KAAK,EAAE,SAAS,GAAG,EAAE,EAAE,CAAC,CAAC;gBACnI,CAAC;gBACD,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,UAAU,CAAC;gBAClC,IAAI,WAAW,CAAC,YAAY;oBAAE,YAAY,IAAI,WAAW,CAAC,YAAY,CAAC;YACzE,CAAC;QACH,CAAC;QAED,SAAS,EAAE,CAAC;QACZ,UAAU,EAAE,CAAC;YACX,KAAK,EAAE,MAAM;YACb,QAAQ,EAAE,aAAa,CAAC,QAAQ;YAChC,SAAS;YACT,KAAK;YACL,SAAS,EAAE,IAAI,CAAC,SAAS;YACzB,OAAO,EAAE,IAAI,CAAC,OAAO;YACrB,UAAU,EAAE,UAAW,CAAC,UAAU;YAClC,WAAW,EAAE,UAAW,CAAC,WAAW;YACpC,YAAY,EAAE,UAAW,CAAC,YAAY;YACtC,OAAO,EAAE,UAAW,CAAC,OAAO;YAC5B,KAAK,EAAE,WAAW,EAAE,cAAc;SACnC,CAAC,CAAC;QAEH,IAAI,SAAsC,CAAC;QAC3C,IAAI,UAAW,CAAC,EAAE,IAAI,UAAW,CAAC,MAAM,IAAI,IAAI,CAAC,GAAG,EAAE,CAAC;YACrD,SAAS,GAAG,UAAU,CAAC,UAAW,CAAC,MAAM,EAAE,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;QAChE,CAAC;QAED,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC;YAAE,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,GAAG,EAAE,CAAC;QAC3D,MAAM,aAAa,GAAG,kBAAkB,CAAC,UAAW,EAAE,WAAW,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC,CAAC;QAEnG,qEAAqE;QACrE,sEAAsE;QACtE,qEAAqE;QACrE,IAAI,MAAM,EAAE,YAAY,IAAI,IAAI,IAAI,aAAa,CAAC,OAAO,GAAG,MAAM,CAAC,YAAY,EAAE,CAAC;YAChF,aAAa,CAAC,EAAE,GAAG,KAAK,CAAC;YACzB,aAAa,CAAC,KAAK,GAAG,oCAAoC,aAAa,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,WAAW,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;QACxI,CAAC;QACD,IAAI,MAAM,EAAE,WAAW,IAAI,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,OAAO,IAAI,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC,WAAW,EAAE,CAAC;YAClF,aAAa,CAAC,EAAE,GAAG,KAAK,CAAC;YACzB,aAAa,CAAC,KAAK,GAAG,sCAAsC,MAAM,GAAG,CAAC,OAAO,IAAI,CAAC,CAAC,YAAY,MAAM,CAAC,WAAW,IAAI,CAAC;QACxH,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,aAAa,CAAC;QAEtD,uEAAuE;QACvE,0EAA0E;QAC1E,iCAAiC;QACjC,IAAI,MAAM,EAAE,QAAQ,IAAI,IAAI,IAAI,YAAY,GAAG,MAAM,CAAC,QAAQ,IAAI,CAAC,eAAe,EAAE,CAAC;YACnF,eAAe,GAAG,IAAI,CAAC;YACvB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,0CAA0C,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,kBAAkB,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,eAAe,KAAK,CAAC,MAAM,GAAG,OAAO,2BAA2B,CAAC,CAAC;QACtM,CAAC;IACH,CAAC;IAED,IAAI,CAAC;QACH,MAAM,kBAAkB,CAAC,KAAK,EAAE,WAAW,EAAE,WAAW,CAAC,CAAC;IAC5D,CAAC;YAAS,CAAC;QACT,IAAI,KAAK;YAAE,KAAK,CAAC,IAAI,EAAE,CAAC;IAC1B,CAAC;IAED,OAAO,EAAE,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,eAAe,EAAE,CAAC;AAC7D,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,SAAS,CAAC,QAAoB,EAAE,KAAa,EAAE,YAAoB,KAAK;IAC5F,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC;QAC5B,KAAK;QACL,MAAM,EAAE,EAAE;QACV,MAAM,EAAE,IAAI;QACZ,GAAG,EAAE,OAAO,CAAC,GAAG,EAAE;QAClB,SAAS;KACV,CAAC,CAAC;IACH,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;QACf,MAAM,IAAI,KAAK,CAAC,SAAS,KAAK,MAAM,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;IACtD,CAAC;AACH,CAAC"}
1
+ {"version":3,"file":"evaluation-execution.js","sourceRoot":"","sources":["../../../src/eval-core/evaluation-execution.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,QAAQ,EAAE,MAAM,YAAY,CAAC;AACnD,OAAO,EAAE,kBAAkB,EAAE,MAAM,aAAa,CAAC;AACjD,OAAO,EAAE,KAAK,EAAE,MAAM,qBAAqB,CAAC;AAC5C,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C,OAAO,EAAE,wBAAwB,EAAE,MAAM,yBAAyB,CAAC;AACnE,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACnD,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AA2ClC,KAAK,UAAU,kBAAkB,CAAI,KAAU,EAAE,WAAmB,EAAE,EAA8B;IAClG,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,UAAU,MAAM;QACnB,OAAO,KAAK,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC;YAC5B,MAAM,CAAC,GAAG,KAAK,EAAE,CAAC;YAClB,MAAM,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;QACrB,CAAC;IACH,CAAC;IACD,MAAM,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,WAAW,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,EAAE,GAAG,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;AACjG,CAAC;AAED,SAAS,eAAe,CAAC,KAAc;IACrC,MAAM,OAAO,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACvE,OAAO;QACL,EAAE,EAAE,KAAK;QACT,MAAM,EAAE,IAAI;QACZ,UAAU,EAAE,CAAC;QACb,aAAa,EAAE,CAAC;QAChB,WAAW,EAAE,CAAC;QACd,YAAY,EAAE,CAAC;QACf,eAAe,EAAE,CAAC;QAClB,mBAAmB,EAAE,CAAC;QACtB,OAAO,EAAE,CAAC;QACV,UAAU,EAAE,OAAO;QACnB,QAAQ,EAAE,CAAC;QACX,KAAK,EAAE,OAAO;KACf,CAAC;AACJ,CAAC;AAED,SAAS,KAAK,CAAC,EAAU;IACvB,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC;AAC3D,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,YAAY,CAAC,EACjC,KAAK,EACL,QAAQ,EACR,aAAa,EACb,KAAK,EACL,UAAU,EACV,OAAO,EACP,WAAW,EACX,WAAW,EACX,SAAS,EACT,OAAO,EACP,OAAO,EACP,UAAU,EACV,KAAK,GAAG,CAAC,EACT,eAAe,EACf,WAAW,GAAG,CAAC,EACf,WAAW,EACX,cAAc,EACd,YAAY,GAAG,IAAI,EACnB,MAAM,GACc;IACpB,MAAM,OAAO,GAAkD,EAAE,CAAC;IAClE,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,IAAI,eAAe,GAAG,KAAK,CAAC;IAE5B,4CAA4C;IAC5C,IAAI,eAAe,EAAE,CAAC;QACpB,KAAK,MAAM,CAAC,QAAQ,EAAE,QAAQ,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,eAAe,CAAC,EAAE,CAAC;YACnE,OAAO,CAAC,QAAQ,CAAC,GAAG,EAAE,GAAG,QAAQ,EAAE,CAAC;QACtC,CAAC;IACH,CAAC;IAED,MAAM,QAAQ,GAAG,IAAI,CAAC,OAAO,EAAE,EAAE,kBAAkB,EAAE,OAAO,CAAC,CAAC;IAC9D,MAAM,KAAK,GAAyB,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,WAAW,CAAC,QAAQ,CAAC,CAAC;IAE3E,KAAK,UAAU,WAAW,CAAC,IAAU;QACnC,sDAAsD;QACtD,IAAI,eAAe,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,EAAE,EAAE,CAAC;YAC1D,OAAO,EAAE,CAAC;YACV,OAAO,EAAE,CAAC;YACV,SAAS,EAAE,CAAC;YACZ,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;YACjI,OAAO;QACT,CAAC;QAED,yEAAyE;QACzE,0EAA0E;QAC1E,yCAAyC;QACzC,IAAI,eAAe,EAAE,CAAC;YACpB,OAAO,EAAE,CAAC;YACV,OAAO,EAAE,CAAC;YACV,SAAS,EAAE,CAAC;YACZ,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;YACjI,OAAO;QACT,CAAC;QAED,OAAO,EAAE,CAAC;QACV,MAAM,GAAG,GAAG,OAAO,CAAC;QACpB,MAAM,KAAK,GAAG,KAAK,CAAC,MAAM,CAAC;QAC3B,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;QAE1G,MAAM,aAAa,GAAG,wBAAwB,CAAC,IAAI,EAAE,KAAK,EAAE,SAAS,EAAE,OAAO,CAAC,CAAC;QAEhF,IAAI,UAAsB,CAAC;QAC3B,MAAM,GAAG,GAAG,QAAQ,CAAC,KAAK,EAAE,aAAa,CAAC,WAAW,EAAE,aAAa,CAAC,KAAK,CAAC,MAAM,EAAE,aAAa,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAC5G,MAAM,MAAM,GAAG,KAAK,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;QAC/B,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,IAAI,MAAM,EAAE,CAAC;YACX,UAAU,GAAG,EAAE,GAAG,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC;QAC3C,CAAC;aAAM,CAAC;YACN,gCAAgC;YAChC,MAAM,WAAW,GAAG,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC;YAC3C,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,WAAW,EAAE,OAAO,EAAE,EAAE,CAAC;gBACxD,IAAI,CAAC;oBACH,UAAU,GAAG,MAAM,QAAQ,CAAC,aAAa,CAAC,KAAK,CAAC,CAAC;gBACnD,CAAC;gBAAC,OAAO,GAAG,EAAE,CAAC;oBACb,UAAU,GAAG,eAAe,CAAC,GAAG,CAAC,CAAC;gBACpC,CAAC;gBACD,IAAI,UAAW,CAAC,EAAE,IAAI,OAAO,KAAK,WAAW;oBAAE,MAAM;gBACrD,mCAAmC;gBACnC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC,GAAG,IAAI,EAAE,KAAK,CAAC,CAAC;gBAC7D,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,WAAW,EAAE,CAAC,CAAC;gBAChI,MAAM,KAAK,CAAC,SAAS,CAAC,CAAC;YACzB,CAAC;YACD,IAAI,KAAK,IAAI,UAAW,CAAC,EAAE;gBAAE,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE,UAAW,CAAC,CAAC;QAC3D,CAAC;QACD,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC;QACtC,YAAY,IAAI,UAAW,CAAC,OAAO,CAAC;QAEpC,IAAI,OAAO,IAAI,UAAU,EAAE,CAAC;YAC1B,UAAU,CAAC;gBACT,KAAK,EAAE,WAAW;gBAClB,QAAQ,EAAE,aAAa,CAAC,QAAQ;gBAChC,SAAS,EAAE,GAAG;gBACd,KAAK;gBACL,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,OAAO,EAAE,IAAI,CAAC,OAAO;gBACrB,UAAU,EAAE,UAAW,CAAC,UAAU;gBAClC,WAAW,EAAE,UAAW,CAAC,WAAW;gBACpC,YAAY,EAAE,UAAW,CAAC,YAAY;gBACtC,OAAO,EAAE,UAAW,CAAC,OAAO;gBAC5B,aAAa,EAAE,UAAW,CAAC,MAAM,CAAC,CAAC,CAAC,UAAW,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI;aAC5E,CAAC,CAAC;QACL,CAAC;QAED,IAAI,WAAW,GAAuB,IAAI,CAAC;QAC3C,IAAI,OAAO,GAAG,CAAC,CAAC;QAChB,IAAI,UAAW,CAAC,EAAE,EAAE,CAAC;YACnB,MAAM,kBAAkB,GAAG,IAAI,CAAC,MAAM,IAAI,IAAI,CAAC,UAAU,EAAE,MAAM,IAAI,CAAC,IAAI,CAAC,UAAU,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,CAAC;YAC9H,IAAI,kBAAkB,EAAE,CAAC;gBACvB,IAAI,OAAO,IAAI,UAAU,EAAE,CAAC;oBAC1B,UAAU,CAAC;wBACT,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,aAAa,CAAC,QAAQ;wBAChC,SAAS,EAAE,GAAG;wBACd,KAAK;wBACL,SAAS,EAAE,IAAI,CAAC,SAAS;wBACzB,OAAO,EAAE,IAAI,CAAC,OAAO;qBACtB,CAAC,CAAC;gBACL,CAAC;gBACD,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;gBAC9B,IAAI,CAAC;oBACH,WAAW,GAAG,MAAM,KAAK,CAAC;wBACxB,MAAM,EAAE,UAAW,CAAC,MAAO;wBAC3B,MAAM,EAAE,IAAI,CAAC,OAAO;wBACpB,QAAQ,EAAE,aAAa;wBACvB,UAAU;wBACV,aAAa,EAAE,CAAC,OAAO;wBACvB,WAAW,EAAE;4BACX,OAAO,EAAE,UAAW,CAAC,OAAO;4BAC5B,UAAU,EAAE,UAAW,CAAC,UAAU;4BAClC,QAAQ,EAAE,UAAW,CAAC,QAAQ;4BAC9B,SAAS,EAAE,UAAW,CAAC,SAAS;4BAChC,KAAK,EAAE,UAAW,CAAC,KAAK;yBACzB;wBACD,UAAU,EAAE,OAAO,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC;wBACzC,WAAW;wBACX,WAAW;wBACX,cAAc;wBACd,YAAY;qBACb,CAAC,CAAC;gBACL,CAAC;gBAAC,OAAO,GAAG,EAAE,CAAC;oBACb,WAAW,GAAG,EAAE,cAAc,EAAE,CAAC,EAAE,CAAC;oBACpC,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;oBAC7D,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,KAAK,EAAE,SAAS,GAAG,EAAE,EAAE,CAAC,CAAC;gBACnI,CAAC;gBACD,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,UAAU,CAAC;gBAClC,IAAI,WAAW,CAAC,YAAY;oBAAE,YAAY,IAAI,WAAW,CAAC,YAAY,CAAC;YACzE,CAAC;QACH,CAAC;QAED,SAAS,EAAE,CAAC;QACZ,UAAU,EAAE,CAAC;YACX,KAAK,EAAE,MAAM;YACb,QAAQ,EAAE,aAAa,CAAC,QAAQ;YAChC,SAAS;YACT,KAAK;YACL,SAAS,EAAE,IAAI,CAAC,SAAS;YACzB,OAAO,EAAE,IAAI,CAAC,OAAO;YACrB,UAAU,EAAE,UAAW,CAAC,UAAU;YAClC,WAAW,EAAE,UAAW,CAAC,WAAW;YACpC,YAAY,EAAE,UAAW,CAAC,YAAY;YACtC,OAAO,EAAE,UAAW,CAAC,OAAO;YAC5B,KAAK,EAAE,WAAW,EAAE,cAAc;SACnC,CAAC,CAAC;QAEH,IAAI,SAAsC,CAAC;QAC3C,IAAI,UAAW,CAAC,EAAE,IAAI,UAAW,CAAC,MAAM,IAAI,IAAI,CAAC,GAAG,EAAE,CAAC;YACrD,SAAS,GAAG,UAAU,CAAC,UAAW,CAAC,MAAM,EAAE,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;QAChE,CAAC;QAED,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC;YAAE,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,GAAG,EAAE,CAAC;QAC3D,MAAM,aAAa,GAAG,kBAAkB,CAAC,UAAW,EAAE,WAAW,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC,CAAC;QAEnG,qEAAqE;QACrE,sEAAsE;QACtE,qEAAqE;QACrE,IAAI,MAAM,EAAE,YAAY,IAAI,IAAI,IAAI,aAAa,CAAC,OAAO,GAAG,MAAM,CAAC,YAAY,EAAE,CAAC;YAChF,aAAa,CAAC,EAAE,GAAG,KAAK,CAAC;YACzB,aAAa,CAAC,KAAK,GAAG,oCAAoC,aAAa,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,WAAW,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;QACxI,CAAC;QACD,IAAI,MAAM,EAAE,WAAW,IAAI,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,OAAO,IAAI,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC,WAAW,EAAE,CAAC;YAClF,aAAa,CAAC,EAAE,GAAG,KAAK,CAAC;YACzB,aAAa,CAAC,KAAK,GAAG,sCAAsC,MAAM,GAAG,CAAC,OAAO,IAAI,CAAC,CAAC,YAAY,MAAM,CAAC,WAAW,IAAI,CAAC;QACxH,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,aAAa,CAAC;QAEtD,uEAAuE;QACvE,0EAA0E;QAC1E,iCAAiC;QACjC,IAAI,MAAM,EAAE,QAAQ,IAAI,IAAI,IAAI,YAAY,GAAG,MAAM,CAAC,QAAQ,IAAI,CAAC,eAAe,EAAE,CAAC;YACnF,eAAe,GAAG,IAAI,CAAC;YACvB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,0CAA0C,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,kBAAkB,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,eAAe,KAAK,CAAC,MAAM,GAAG,OAAO,2BAA2B,CAAC,CAAC;QACtM,CAAC;IACH,CAAC;IAED,IAAI,CAAC;QACH,MAAM,kBAAkB,CAAC,KAAK,EAAE,WAAW,EAAE,WAAW,CAAC,CAAC;IAC5D,CAAC;YAAS,CAAC;QACT,IAAI,KAAK;YAAE,KAAK,CAAC,IAAI,EAAE,CAAC;IAC1B,CAAC;IAED,OAAO,EAAE,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,eAAe,EAAE,CAAC;AAC7D,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,SAAS,CAAC,QAAoB,EAAE,KAAa,EAAE,YAAoB,KAAK;IAC5F,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC;QAC5B,KAAK;QACL,MAAM,EAAE,EAAE;QACV,MAAM,EAAE,IAAI;QACZ,GAAG,EAAE,OAAO,CAAC,GAAG,EAAE;QAClB,SAAS;KACV,CAAC,CAAC;IACH,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;QACf,MAAM,IAAI,KAAK,CAAC,qBAAqB,KAAK,MAAM,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;IAClE,CAAC;AACH,CAAC"}
@@ -1,4 +1,4 @@
1
- import type { Artifact, EvaluationErrorCategory, EvaluationJob, EvaluationRequest, EvaluationRun, JudgeConfig } from '../types.js';
1
+ import type { Artifact, EvaluationErrorCategory, EvaluationJob, EvaluationRequest, EvaluationRun, JudgeConfig } from '../types/index.js';
2
2
  export declare function buildEvaluationRequest({ samplesPath, skillDir, artifacts, model, judgeModel, executor, judgeExecutor, noJudge, concurrency, timeoutMs, noCache, dryRun, blind, project, owner, tags, repeat, each, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, }: {
3
3
  samplesPath: string;
4
4
  skillDir: string;
@@ -23,7 +23,7 @@ export declare function buildEvaluationRequest({ samplesPath, skillDir, artifact
23
23
  bootstrap?: boolean;
24
24
  bootstrapSamples?: number;
25
25
  lengthDebias?: boolean;
26
- budget?: import('../types.js').EvalBudget;
26
+ budget?: import('../types/index.js').EvalBudget;
27
27
  }): EvaluationRequest;
28
28
  export declare function createEvaluationRun(runId: string, startedAt?: string): {
29
29
  run: EvaluationRun;
@@ -1 +1 @@
1
- {"version":3,"file":"evaluation-job.d.ts","sourceRoot":"","sources":["../../../src/eval-core/evaluation-job.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,uBAAuB,EAAE,aAAa,EAAE,iBAAiB,EAAE,aAAa,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAMnI,wBAAgB,sBAAsB,CAAC,EACrC,WAAW,EACX,QAAQ,EACR,SAAS,EACT,KAAK,EACL,UAAU,EACV,QAAQ,EACR,aAAa,EACb,OAAO,EACP,WAAW,EACX,SAAS,EACT,OAAO,EACP,MAAM,EACN,KAAK,EACL,OAAO,EACP,KAAK,EACL,IAAI,EACJ,MAAM,EACN,IAAI,EACJ,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,GACP,EAAE;IACD,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,QAAQ,EAAE,MAAM,CAAC;IACjB,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,OAAO,EAAE,OAAO,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,MAAM,EAAE,OAAO,CAAC;IAChB,KAAK,EAAE,OAAO,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,WAAW,EAAE,CAAC;IAC5B,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,MAAM,CAAC,EAAE,OAAO,aAAa,EAAE,UAAU,CAAC;CAC3C,GAAG,iBAAiB,CA2BpB;AAED,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,GAAE,MAAiB,GAAG;IAAE,GAAG,EAAE,aAAa,CAAC;IAAC,SAAS,EAAE,MAAM,CAAA;CAAE,CAS1H;AAED,wBAAgB,qBAAqB,CAAC,GAAG,EAAE,aAAa,EAAE,UAAU,GAAE,MAAiB,GAAG,aAAa,CAMtG;AAED,wBAAgB,iBAAiB,CAAC,GAAG,EAAE,aAAa,EAAE,UAAU,GAAE,MAAiB,GAAG,aAAa,CAMlG;AAED,wBAAgB,eAAe,CAAC,EAC9B,KAAK,EACL,OAAO,EACP,SAAoB,GACrB,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,iBAAiB,CAAC;IAC3B,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB,GAAG,aAAa,CAQhB;AAED,wBAAgB,cAAc,CAAC,GAAG,EAAE,aAAa,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,GAAE,MAAiB,GAAG,aAAa,CAQ7G;AAED,wBAAgB,kBAAkB,CAAC,EACjC,KAAK,EACL,KAAK,EACL,QAAQ,EACR,OAAO,EACP,SAAS,EACT,SAAS,EACT,UAAU,GACX,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,iBAAiB,CAAC;IAC3B,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;CACpB,GAAG,aAAa,CAYhB;AAED,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,MAAM,GAAG,uBAAuB,CA4B9E;AAED,wBAAgB,eAAe,CAAC,EAC9B,GAAG,EACH,KAAK,EACL,aAA8C,EAC9C,UAAqB,GACtB,EAAE;IACD,GAAG,EAAE,aAAa,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,CAAC,EAAE,uBAAuB,CAAC;IACxC,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB,GAAG,aAAa,CAShB"}
1
+ {"version":3,"file":"evaluation-job.d.ts","sourceRoot":"","sources":["../../../src/eval-core/evaluation-job.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,uBAAuB,EAAE,aAAa,EAAE,iBAAiB,EAAE,aAAa,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAMzI,wBAAgB,sBAAsB,CAAC,EACrC,WAAW,EACX,QAAQ,EACR,SAAS,EACT,KAAK,EACL,UAAU,EACV,QAAQ,EACR,aAAa,EACb,OAAO,EACP,WAAW,EACX,SAAS,EACT,OAAO,EACP,MAAM,EACN,KAAK,EACL,OAAO,EACP,KAAK,EACL,IAAI,EACJ,MAAM,EACN,IAAI,EACJ,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,GACP,EAAE;IACD,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,QAAQ,EAAE,MAAM,CAAC;IACjB,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,OAAO,EAAE,OAAO,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,MAAM,EAAE,OAAO,CAAC;IAChB,KAAK,EAAE,OAAO,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,WAAW,EAAE,CAAC;IAC5B,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,MAAM,CAAC,EAAE,OAAO,mBAAmB,EAAE,UAAU,CAAC;CACjD,GAAG,iBAAiB,CA2BpB;AAED,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,GAAE,MAAiB,GAAG;IAAE,GAAG,EAAE,aAAa,CAAC;IAAC,SAAS,EAAE,MAAM,CAAA;CAAE,CAS1H;AAED,wBAAgB,qBAAqB,CAAC,GAAG,EAAE,aAAa,EAAE,UAAU,GAAE,MAAiB,GAAG,aAAa,CAMtG;AAED,wBAAgB,iBAAiB,CAAC,GAAG,EAAE,aAAa,EAAE,UAAU,GAAE,MAAiB,GAAG,aAAa,CAMlG;AAED,wBAAgB,eAAe,CAAC,EAC9B,KAAK,EACL,OAAO,EACP,SAAoB,GACrB,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,iBAAiB,CAAC;IAC3B,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB,GAAG,aAAa,CAQhB;AAED,wBAAgB,cAAc,CAAC,GAAG,EAAE,aAAa,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,GAAE,MAAiB,GAAG,aAAa,CAQ7G;AAED,wBAAgB,kBAAkB,CAAC,EACjC,KAAK,EACL,KAAK,EACL,QAAQ,EACR,OAAO,EACP,SAAS,EACT,SAAS,EACT,UAAU,GACX,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,iBAAiB,CAAC;IAC3B,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;CACpB,GAAG,aAAa,CAYhB;AAED,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,MAAM,GAAG,uBAAuB,CA4B9E;AAED,wBAAgB,eAAe,CAAC,EAC9B,GAAG,EACH,KAAK,EACL,aAA8C,EAC9C,UAAqB,GACtB,EAAE;IACD,GAAG,EAAE,aAAa,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,CAAC,EAAE,uBAAuB,CAAC;IACxC,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB,GAAG,aAAa,CAShB"}
@@ -1,4 +1,4 @@
1
- import type { Artifact, Report, Sample, Task, VariantResult, EvaluationJob, EvaluationRequest, EvaluationRun } from '../types.js';
1
+ import type { Artifact, Report, Sample, Task, VariantResult, EvaluationJob, EvaluationRequest, EvaluationRun } from '../types/index.js';
2
2
  export declare const DEFAULT_OUTPUT_DIR: string;
3
3
  interface AggregateReportOptions {
4
4
  runId: string;
@@ -1 +1 @@
1
- {"version":3,"file":"evaluation-reporting.d.ts","sourceRoot":"","sources":["../../../src/eval-core/evaluation-reporting.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EACV,QAAQ,EACR,MAAM,EACN,MAAM,EACN,IAAI,EACJ,aAAa,EAIb,aAAa,EACb,iBAAiB,EACjB,aAAa,EACd,MAAM,aAAa,CAAC;AAgBrB,eAAO,MAAM,kBAAkB,EAAE,MAAuD,CAAC;AA8CzF,UAAU,sBAAsB;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC,CAAC;IACvD,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,OAAO,CAAC,EAAE,iBAAiB,CAAC;IAC5B,GAAG,CAAC,EAAE,aAAa,CAAC;IACpB,GAAG,CAAC,EAAE,aAAa,CAAC;IACpB,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AAED,wBAAgB,eAAe,CAAC,EAC9B,KAAK,EACL,QAAQ,EACR,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,OAAO,EACP,KAAK,EACL,OAAO,EACP,YAAY,EACZ,SAAS,EACT,OAAO,EACP,GAAG,EACH,GAAG,EACH,YAAY,GACb,EAAE,sBAAsB,GAAG,MAAM,CAmGjC;AAED,wBAAgB,cAAc,CAAC,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,EAAE,SAAS,EAAE,MAAM,GAAG,IAAI,CAmC1F;AAED,MAAM,WAAW,iBAAiB;IAChC,EAAE,EAAE,MAAM,CAAC;CACZ;AAED,wBAAgB,aAAa,CAAC,MAAM,EAAE,iBAAiB,EAAE,SAAS,EAAE,MAAM,GAAG,IAAI,GAAG,MAAM,GAAG,IAAI,CAMhG;AAED,wBAAgB,aAAa,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,MAAM,CASxD"}
1
+ {"version":3,"file":"evaluation-reporting.d.ts","sourceRoot":"","sources":["../../../src/eval-core/evaluation-reporting.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EACV,QAAQ,EACR,MAAM,EACN,MAAM,EACN,IAAI,EACJ,aAAa,EAIb,aAAa,EACb,iBAAiB,EACjB,aAAa,EACd,MAAM,mBAAmB,CAAC;AAgB3B,eAAO,MAAM,kBAAkB,EAAE,MAAuD,CAAC;AA8CzF,UAAU,sBAAsB;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC,CAAC;IACvD,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,OAAO,CAAC,EAAE,iBAAiB,CAAC;IAC5B,GAAG,CAAC,EAAE,aAAa,CAAC;IACpB,GAAG,CAAC,EAAE,aAAa,CAAC;IACpB,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AAED,wBAAgB,eAAe,CAAC,EAC9B,KAAK,EACL,QAAQ,EACR,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,OAAO,EACP,KAAK,EACL,OAAO,EACP,YAAY,EACZ,SAAS,EACT,OAAO,EACP,GAAG,EACH,GAAG,EACH,YAAY,GACb,EAAE,sBAAsB,GAAG,MAAM,CAmGjC;AAED,wBAAgB,cAAc,CAAC,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,EAAE,SAAS,EAAE,MAAM,GAAG,IAAI,CAmC1F;AAED,MAAM,WAAW,iBAAiB;IAChC,EAAE,EAAE,MAAM,CAAC;CACZ;AAED,wBAAgB,aAAa,CAAC,MAAM,EAAE,iBAAiB,EAAE,SAAS,EAAE,MAAM,GAAG,IAAI,GAAG,MAAM,GAAG,IAAI,CAMhG;AAED,wBAAgB,aAAa,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,MAAM,CASxD"}
@@ -1,4 +1,4 @@
1
- import type { Artifact, ExecutionStrategyKind, ExecutorInput, ExperimentType, Task, VariantConfig } from '../types.js';
1
+ import type { Artifact, ExecutionStrategyKind, ExecutorInput, ExperimentType, Task, VariantConfig } from '../types/index.js';
2
2
  export interface ExecutionPlan {
3
3
  strategy: ExecutionStrategyKind;
4
4
  cacheSystem: string;
@@ -1 +1 @@
1
- {"version":3,"file":"execution-strategy.d.ts","sourceRoot":"","sources":["../../../src/eval-core/execution-strategy.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,QAAQ,EAAE,qBAAqB,EAAE,aAAa,EAAE,cAAc,EAAE,IAAI,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAEvH,MAAM,WAAW,aAAa;IAC5B,QAAQ,EAAE,qBAAqB,CAAC;IAChC,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,aAAa,CAAC;CACtB;AAOD,wBAAgB,gCAAgC,CAAC,QAAQ,EAAE,QAAQ,GAAG,qBAAqB,CAc1F;AAED,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,QAAQ,GAAG,cAAc,CAIxE;AAED,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,QAAQ,GAAG,aAAa,CAyBpE;AAOD,wBAAgB,wBAAwB,CAAC,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,OAAO,GAAG,aAAa,CA+DxH"}
1
+ {"version":3,"file":"execution-strategy.d.ts","sourceRoot":"","sources":["../../../src/eval-core/execution-strategy.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,QAAQ,EAAE,qBAAqB,EAAE,aAAa,EAAE,cAAc,EAAE,IAAI,EAAE,aAAa,EAAE,MAAM,mBAAmB,CAAC;AAE7H,MAAM,WAAW,aAAa;IAC5B,QAAQ,EAAE,qBAAqB,CAAC;IAChC,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,aAAa,CAAC;CACtB;AAOD,wBAAgB,gCAAgC,CAAC,QAAQ,EAAE,QAAQ,GAAG,qBAAqB,CAc1F;AAED,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,QAAQ,GAAG,cAAc,CAIxE;AAED,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,QAAQ,GAAG,aAAa,CAyBpE;AAOD,wBAAgB,wBAAwB,CAAC,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,OAAO,GAAG,aAAa,CA+DxH"}
@@ -0,0 +1,17 @@
1
+ import type { VariantSummary } from '../types/index.js';
2
+ export interface LayerGateResult {
3
+ allPass: boolean;
4
+ lines: string[];
5
+ }
6
+ /**
7
+ * Three-layer gate:事实 / 行为 / LLM 评价 三层各自与 threshold 独立比较,
8
+ * 任一层低于 threshold 即 FAIL。这样某一层崩盘会被暴露,不会被 composite 合成分
9
+ * 均化掩盖。
10
+ *
11
+ * 三层都缺(eval-samples 既没定义断言也没定义 rubric)时直接 FAIL + 引导,
12
+ * 不走 composite fallback——符合 0-1 窗口期不做兼容原则。
13
+ *
14
+ * 抽为纯函数便于单测;handleGate 只负责 IO(读 config、打印、退出码)。
15
+ */
16
+ export declare function evaluateLayerGates(summary: Record<string, VariantSummary>, threshold: number): LayerGateResult;
17
+ //# sourceMappingURL=layer-gates.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"layer-gates.d.ts","sourceRoot":"","sources":["../../../src/eval-core/layer-gates.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AAExD,MAAM,WAAW,eAAe;IAC9B,OAAO,EAAE,OAAO,CAAC;IACjB,KAAK,EAAE,MAAM,EAAE,CAAC;CACjB;AAED;;;;;;;;;GASG;AACH,wBAAgB,kBAAkB,CAChC,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,EACvC,SAAS,EAAE,MAAM,GAChB,eAAe,CAgCjB"}
@@ -1,14 +1,14 @@
1
1
  /**
2
- * PR-3 three-gate CI 门禁:事实 / 行为 / LLM 评价 三层各自与 threshold 独立比较,
2
+ * Three-layer gate:事实 / 行为 / LLM 评价 三层各自与 threshold 独立比较,
3
3
  * 任一层低于 threshold 即 FAIL。这样某一层崩盘会被暴露,不会被 composite 合成分
4
4
  * 均化掩盖。
5
5
  *
6
6
  * 三层都缺(eval-samples 既没定义断言也没定义 rubric)时直接 FAIL + 引导,
7
- * 不走 composite fallback——符合 0-1 窗口期不做兼容原则与 PR-3 精神。
7
+ * 不走 composite fallback——符合 0-1 窗口期不做兼容原则。
8
8
  *
9
- * 抽为纯函数便于单测;handleCi 只负责 IO(读 config、打印、退出码)。
9
+ * 抽为纯函数便于单测;handleGate 只负责 IO(读 config、打印、退出码)。
10
10
  */
11
- export function evaluateCiGates(summary, threshold) {
11
+ export function evaluateLayerGates(summary, threshold) {
12
12
  let allPass = true;
13
13
  const lines = [];
14
14
  for (const [variant, stats] of Object.entries(summary || {})) {
@@ -39,4 +39,4 @@ export function evaluateCiGates(summary, threshold) {
39
39
  }
40
40
  return { allPass, lines };
41
41
  }
42
- //# sourceMappingURL=ci-gates.js.map
42
+ //# sourceMappingURL=layer-gates.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"layer-gates.js","sourceRoot":"","sources":["../../../src/eval-core/layer-gates.ts"],"names":[],"mappings":"AAOA;;;;;;;;;GASG;AACH,MAAM,UAAU,kBAAkB,CAChC,OAAuC,EACvC,SAAiB;IAEjB,IAAI,OAAO,GAAG,IAAI,CAAC;IACnB,MAAM,KAAK,GAAa,EAAE,CAAC;IAE3B,KAAK,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,IAAI,EAAE,CAAC,EAAE,CAAC;QAC7D,MAAM,MAAM,GAAwD;YAClE,EAAE,KAAK,EAAE,WAAW,EAAS,KAAK,EAAE,KAAK,CAAC,YAAY,EAAE;YACxD,EAAE,KAAK,EAAE,eAAe,EAAK,KAAK,EAAE,KAAK,CAAC,gBAAgB,EAAE;YAC5D,EAAE,KAAK,EAAE,gBAAgB,EAAI,KAAK,EAAE,KAAK,CAAC,aAAa,EAAE;SAC1D,CAAC;QACF,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC;QAElE,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACzB,KAAK,CAAC,IAAI,CAAC,SAAS,OAAO,+FAA+F,CAAC,CAAC;YAC5H,OAAO,GAAG,KAAK,CAAC;YAChB,SAAS;QACX,CAAC;QAED,IAAI,WAAW,GAAG,IAAI,CAAC;QACvB,MAAM,KAAK,GAAa,EAAE,CAAC;QAC3B,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;YACxB,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC;YACvB,MAAM,IAAI,GAAG,CAAC,IAAI,SAAS,CAAC;YAC5B,IAAI,CAAC,IAAI;gBAAE,WAAW,GAAG,KAAK,CAAC;YAC/B,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;QAC9D,CAAC;QACD,MAAM,MAAM,GAAG,WAAW,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;QAC7C,KAAK,CAAC,IAAI,CAAC,GAAG,MAAM,KAAK,OAAO,MAAM,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,eAAe,SAAS,GAAG,CAAC,CAAC;QACpF,IAAI,CAAC,WAAW;YAAE,OAAO,GAAG,KAAK,CAAC;IACpC,CAAC;IAED,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;AAC5B,CAAC"}
@@ -2,7 +2,7 @@
2
2
  * Domain model definitions for oh-my-knowledge.
3
3
  * Single source of truth for object structures used across runner, renderer, and server.
4
4
  */
5
- import type { ExecResult, GradeResult, VariantResult, VariantSummary } from '../types.js';
5
+ import type { ExecResult, GradeResult, VariantResult, VariantSummary } from '../types/index.js';
6
6
  /**
7
7
  * Build a VariantResult from execution and grading results.
8
8
  */
@@ -1 +1 @@
1
- {"version":3,"file":"schema.d.ts","sourceRoot":"","sources":["../../../src/eval-core/schema.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,WAAW,EAAE,aAAa,EAAE,cAAc,EAA0B,MAAM,aAAa,CAAC;AA2BlH;;GAEG;AACH,UAAU,mBAAmB;IAC3B,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE;QAAE,aAAa,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAC;QAAC,YAAY,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,KAAK,CAAC;YAAE,IAAI,EAAE,MAAM,CAAC;YAAC,KAAK,EAAE,MAAM,CAAC;YAAC,QAAQ,EAAE,OAAO,CAAC;YAAC,QAAQ,CAAC,EAAE,MAAM,CAAA;SAAE,CAAC,CAAA;KAAE,CAAC;CACvK;AAED,wBAAgB,kBAAkB,CAAC,UAAU,EAAE,UAAU,EAAE,WAAW,EAAE,WAAW,GAAG,IAAI,EAAE,OAAO,CAAC,EAAE,mBAAmB,GAAG,aAAa,CAgFxI;AAED;;GAEG;AACH,wBAAgB,mBAAmB,CAAC,OAAO,EAAE,aAAa,EAAE,GAAG,cAAc,CA8F5E"}
1
+ {"version":3,"file":"schema.d.ts","sourceRoot":"","sources":["../../../src/eval-core/schema.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,WAAW,EAAE,aAAa,EAAE,cAAc,EAA0B,MAAM,mBAAmB,CAAC;AA2BxH;;GAEG;AACH,UAAU,mBAAmB;IAC3B,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE;QAAE,aAAa,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAC;QAAC,YAAY,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,KAAK,CAAC;YAAE,IAAI,EAAE,MAAM,CAAC;YAAC,KAAK,EAAE,MAAM,CAAC;YAAC,QAAQ,EAAE,OAAO,CAAC;YAAC,QAAQ,CAAC,EAAE,MAAM,CAAA;SAAE,CAAC,CAAA;KAAE,CAAC;CACvK;AAED,wBAAgB,kBAAkB,CAAC,UAAU,EAAE,UAAU,EAAE,WAAW,EAAE,WAAW,GAAG,IAAI,EAAE,OAAO,CAAC,EAAE,mBAAmB,GAAG,aAAa,CAgFxI;AAED;;GAEG;AACH,wBAAgB,mBAAmB,CAAC,OAAO,EAAE,aAAa,EAAE,GAAG,cAAc,CA8F5E"}
@@ -164,7 +164,7 @@ export function buildVariantSummary(entries) {
164
164
  };
165
165
  })(),
166
166
  ...(() => {
167
- // 保留 0 分样本(评委打"完全不合格"是合法低分,不是缺失)。
167
+ // 保留 0 分用例(评委打"完全不合格"是合法低分,不是缺失)。
168
168
  // 仅 filter null / undefined(真正缺数据,如该 sample 无对应断言或未配 judge)。
169
169
  const factScores = ok.map((e) => e.layeredScores?.factScore).filter((s) => s != null);
170
170
  const behaviorScores = ok.map((e) => e.layeredScores?.behaviorScore).filter((s) => s != null);
@@ -1,4 +1,4 @@
1
- import type { Artifact, Sample, Task } from '../types.js';
1
+ import type { Artifact, Sample, Task } from '../types/index.js';
2
2
  export declare function buildTasks(samples: Sample[], variants: string[], skills: Record<string, string | null>): Task[];
3
3
  export declare function buildTasksFromArtifacts(samples: Sample[], artifacts: Artifact[]): Task[];
4
4
  //# sourceMappingURL=task-planner.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"task-planner.d.ts","sourceRoot":"","sources":["../../../src/eval-core/task-planner.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,aAAa,CAAC;AAE1D,wBAAgB,UAAU,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE,QAAQ,EAAE,MAAM,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC,GAAG,IAAI,EAAE,CAQ/G;AAED,wBAAgB,uBAAuB,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE,SAAS,EAAE,QAAQ,EAAE,GAAG,IAAI,EAAE,CAyBxF"}
1
+ {"version":3,"file":"task-planner.d.ts","sourceRoot":"","sources":["../../../src/eval-core/task-planner.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,mBAAmB,CAAC;AAEhE,wBAAgB,UAAU,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE,QAAQ,EAAE,MAAM,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC,GAAG,IAAI,EAAE,CAQ/G;AAED,wBAAgB,uBAAuB,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE,SAAS,EAAE,QAAQ,EAAE,GAAG,IAAI,EAAE,CAyBxF"}
@@ -23,7 +23,7 @@ export function buildTasksFromArtifacts(samples, artifacts) {
23
23
  assertions: sample.assertions || null,
24
24
  dimensions: sample.dimensions || null,
25
25
  artifactContent: artifact.content,
26
- cwd: artifact.cwd || sample.cwd || null,
26
+ cwd: artifact.cwd || artifact.skillRoot || sample.cwd || null,
27
27
  _sample: sample,
28
28
  });
29
29
  }
@@ -1 +1 @@
1
- {"version":3,"file":"task-planner.js","sourceRoot":"","sources":["../../../src/eval-core/task-planner.ts"],"names":[],"mappings":"AAEA,MAAM,UAAU,UAAU,CAAC,OAAiB,EAAE,QAAkB,EAAE,MAAqC;IACrG,MAAM,SAAS,GAAe,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QACvD,IAAI,EAAE,OAAO;QACb,IAAI,EAAE,OAAO,KAAK,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,OAAO;QACnD,MAAM,EAAE,OAAO,KAAK,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,QAAQ;QACtD,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,IAAI,IAAI;KACjC,CAAC,CAAC,CAAC;IACJ,OAAO,uBAAuB,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;AACrD,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,OAAiB,EAAE,SAAqB;IAC9E,MAAM,KAAK,GAAW,EAAE,CAAC;IAEzB,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;YACjC,MAAM,UAAU,GAAG,MAAM,CAAC,OAAO;gBAC/B,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,eAAe,MAAM,CAAC,OAAO,UAAU;gBACzD,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC;YAElB,KAAK,CAAC,IAAI,CAAC;gBACT,SAAS,EAAE,MAAM,CAAC,SAAS;gBAC3B,OAAO,EAAE,QAAQ,CAAC,IAAI;gBACtB,QAAQ;gBACR,MAAM,EAAE,UAAU;gBAClB,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,IAAI;gBAC7B,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;gBACrC,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;gBACrC,eAAe,EAAE,QAAQ,CAAC,OAAO;gBACjC,GAAG,EAAE,QAAQ,CAAC,GAAG,IAAI,MAAM,CAAC,GAAG,IAAI,IAAI;gBACvC,OAAO,EAAE,MAAM;aAChB,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,OAAO,KAAK,CAAC;AACf,CAAC"}
1
+ {"version":3,"file":"task-planner.js","sourceRoot":"","sources":["../../../src/eval-core/task-planner.ts"],"names":[],"mappings":"AAEA,MAAM,UAAU,UAAU,CAAC,OAAiB,EAAE,QAAkB,EAAE,MAAqC;IACrG,MAAM,SAAS,GAAe,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QACvD,IAAI,EAAE,OAAO;QACb,IAAI,EAAE,OAAO,KAAK,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,OAAO;QACnD,MAAM,EAAE,OAAO,KAAK,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,QAAQ;QACtD,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,IAAI,IAAI;KACjC,CAAC,CAAC,CAAC;IACJ,OAAO,uBAAuB,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;AACrD,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,OAAiB,EAAE,SAAqB;IAC9E,MAAM,KAAK,GAAW,EAAE,CAAC;IAEzB,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;YACjC,MAAM,UAAU,GAAG,MAAM,CAAC,OAAO;gBAC/B,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,eAAe,MAAM,CAAC,OAAO,UAAU;gBACzD,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC;YAElB,KAAK,CAAC,IAAI,CAAC;gBACT,SAAS,EAAE,MAAM,CAAC,SAAS;gBAC3B,OAAO,EAAE,QAAQ,CAAC,IAAI;gBACtB,QAAQ;gBACR,MAAM,EAAE,UAAU;gBAClB,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,IAAI;gBAC7B,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;gBACrC,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;gBACrC,eAAe,EAAE,QAAQ,CAAC,OAAO;gBACjC,GAAG,EAAE,QAAQ,CAAC,GAAG,IAAI,QAAQ,CAAC,SAAS,IAAI,MAAM,CAAC,GAAG,IAAI,IAAI;gBAC7D,OAAO,EAAE,MAAM;aAChB,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,OAAO,KAAK,CAAC;AACf,CAAC"}
@@ -26,7 +26,7 @@
26
26
  * proven optimal. Each rule's source (NIST AI 800-3 / Krippendorff thresholds /
27
27
  * empirical) is documented inline so users can audit and override.
28
28
  */
29
- import type { Report } from '../types.js';
29
+ import type { Report } from '../types/index.js';
30
30
  export type VerdictLevel = 'PROGRESS' | 'CAUTIOUS' | 'REGRESS' | 'NOISE' | 'UNDERPOWERED' | 'SOLO';
31
31
  export interface VerdictResult {
32
32
  level: VerdictLevel;
@@ -44,6 +44,9 @@ export interface VerdictResult {
44
44
  significance?: string;
45
45
  layerWinners?: string;
46
46
  sampleSize?: string;
47
+ /** Stability claim:CV / variance summary if --repeat ≥ 2, 否则显式说"未测量"
48
+ * 让用户感受到 single-run 的盲区,而不是默默不提。 */
49
+ stability?: string;
47
50
  judgeAgreement?: string;
48
51
  shipRecommendation?: string;
49
52
  };
@@ -52,7 +55,7 @@ export interface VerdictResult {
52
55
  }
53
56
  export interface VerdictOptions {
54
57
  /** Three-layer ci-gate threshold; defaults to 3.5 (matches `omk bench ci`). */
55
- ciThreshold?: number;
58
+ gateThreshold?: number;
56
59
  /**
57
60
  * Magnitude (in raw score points) below which a "significant" diff is treated
58
61
  * as practically negligible (statistically real but too small to matter).
@@ -1 +1 @@
1
- {"version":3,"file":"verdict.d.ts","sourceRoot":"","sources":["../../../src/eval-core/verdict.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAEH,OAAO,KAAK,EAAE,MAAM,EAAyC,MAAM,aAAa,CAAC;AAGjF,MAAM,MAAM,YAAY,GACpB,UAAU,GACV,UAAU,GACV,SAAS,GACT,OAAO,GACP,cAAc,GACd,MAAM,CAAC;AAEX,MAAM,WAAW,aAAa;IAC5B,KAAK,EAAE,YAAY,CAAC;IACpB,6CAA6C;IAC7C,QAAQ,EAAE,MAAM,CAAC;IACjB,yEAAyE;IACzE,OAAO,CAAC,EAAE,KAAK,CAAC;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,YAAY,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC/F,6DAA6D;IAC7D,SAAS,EAAE;QACT,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,UAAU,CAAC,EAAE,MAAM,CAAC;QACpB,cAAc,CAAC,EAAE,MAAM,CAAC;QACxB,kBAAkB,CAAC,EAAE,MAAM,CAAC;KAC7B,CAAC;IACF,gEAAgE;IAChE,QAAQ,EAAE,MAAM,EAAE,CAAC;CACpB;AAED,MAAM,WAAW,cAAc;IAC7B,+EAA+E;IAC/E,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;;;;OAIG;IACH,kBAAkB,CAAC,EAAE,MAAM,CAAC;CAC7B;AAED;;GAEG;AACH,wBAAgB,cAAc,CAAC,MAAM,EAAE,MAAM,EAAE,OAAO,GAAE,cAAmB,GAAG,aAAa,CAqE1F;AAoLD;;;GAGG;AACH,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,aAAa,EAAE,OAAO,GAAE;IAAE,OAAO,CAAC,EAAE,OAAO,CAAA;CAAO,GAAG,MAAM,CAgBpG"}
1
+ {"version":3,"file":"verdict.d.ts","sourceRoot":"","sources":["../../../src/eval-core/verdict.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAEH,OAAO,KAAK,EAAE,MAAM,EAAyC,MAAM,mBAAmB,CAAC;AAGvF,MAAM,MAAM,YAAY,GACpB,UAAU,GACV,UAAU,GACV,SAAS,GACT,OAAO,GACP,cAAc,GACd,MAAM,CAAC;AAEX,MAAM,WAAW,aAAa;IAC5B,KAAK,EAAE,YAAY,CAAC;IACpB,6CAA6C;IAC7C,QAAQ,EAAE,MAAM,CAAC;IACjB,yEAAyE;IACzE,OAAO,CAAC,EAAE,KAAK,CAAC;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,YAAY,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC/F,6DAA6D;IAC7D,SAAS,EAAE;QACT,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,UAAU,CAAC,EAAE,MAAM,CAAC;QACpB;6CACqC;QACrC,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,cAAc,CAAC,EAAE,MAAM,CAAC;QACxB,kBAAkB,CAAC,EAAE,MAAM,CAAC;KAC7B,CAAC;IACF,gEAAgE;IAChE,QAAQ,EAAE,MAAM,EAAE,CAAC;CACpB;AAED,MAAM,WAAW,cAAc;IAC7B,+EAA+E;IAC/E,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB;;;;OAIG;IACH,kBAAkB,CAAC,EAAE,MAAM,CAAC;CAC7B;AAED;;GAEG;AACH,wBAAgB,cAAc,CAAC,MAAM,EAAE,MAAM,EAAE,OAAO,GAAE,cAAmB,GAAG,aAAa,CAwE1F;AAsND;;;GAGG;AACH,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,aAAa,EAAE,OAAO,GAAE;IAAE,OAAO,CAAC,EAAE,OAAO,CAAA;CAAO,GAAG,MAAM,CAiBpG"}
@@ -26,27 +26,28 @@
26
26
  * proven optimal. Each rule's source (NIST AI 800-3 / Krippendorff thresholds /
27
27
  * empirical) is documented inline so users can audit and override.
28
28
  */
29
- import { evaluateCiGates } from './ci-gates.js';
29
+ import { evaluateLayerGates } from './layer-gates.js';
30
30
  /**
31
31
  * Compute a verdict for a finished report. Pure function — no I/O.
32
32
  */
33
33
  export function computeVerdict(report, options = {}) {
34
- const { ciThreshold = 3.5, triviallySmallDiff = 0.1 } = options;
34
+ const { gateThreshold = 3.5, triviallySmallDiff = 0.1 } = options;
35
35
  const variants = report.meta?.variants ?? [];
36
36
  const summary = report.summary ?? {};
37
37
  const sampleCount = report.meta?.sampleCount ?? 0;
38
38
  if (variants.length < 2) {
39
39
  // Single-variant — no comparison possible. Just report whether the variant
40
40
  // passes its own three-layer gate.
41
- const gate = evaluateCiGates(summary, ciThreshold);
41
+ const gate = evaluateLayerGates(summary, gateThreshold);
42
42
  return {
43
43
  level: 'SOLO',
44
44
  headline: gate.allPass
45
- ? `SOLO · single variant, three-layer gate PASS @ threshold ${ciThreshold}`
45
+ ? `SOLO · single variant, three-layer gate PASS @ threshold ${gateThreshold}`
46
46
  : `SOLO · single variant, three-layer gate FAIL — see ci output`,
47
47
  rationale: {
48
48
  layerWinners: gate.lines.join('; '),
49
49
  sampleSize: `N=${sampleCount}`,
50
+ stability: formatStability(report),
50
51
  },
51
52
  variants,
52
53
  };
@@ -58,7 +59,7 @@ export function computeVerdict(report, options = {}) {
58
59
  const pairs = explicitPairs.length > 0
59
60
  ? explicitPairs
60
61
  : variants.slice(1).map((treatment) => ({ control: variants[0], treatment }));
61
- const perPair = pairs.map((p) => verdictForPair(p, summary, sampleCount, report, ciThreshold, triviallySmallDiff));
62
+ const perPair = pairs.map((p) => verdictForPair(p, summary, sampleCount, report, gateThreshold, triviallySmallDiff));
62
63
  // Worst-case roll-up: REGRESS dominates, then CAUTIOUS, then NOISE/UNDERPOWERED, then PROGRESS.
63
64
  const order = ['REGRESS', 'CAUTIOUS', 'UNDERPOWERED', 'NOISE', 'PROGRESS'];
64
65
  let topLevel = 'PROGRESS';
@@ -75,6 +76,7 @@ export function computeVerdict(report, options = {}) {
75
76
  : 'no pairwise comparison available — was --bootstrap used?';
76
77
  const layerWinners = formatLayerWinners(summary, variants);
77
78
  const sampleSize = formatSampleSize(report);
79
+ const stability = formatStability(report);
78
80
  const judgeAgreement = formatJudgeAgreement(report);
79
81
  const shipRecommendation = recommendation(topLevel, perPair);
80
82
  return {
@@ -87,6 +89,7 @@ export function computeVerdict(report, options = {}) {
87
89
  significance,
88
90
  layerWinners,
89
91
  sampleSize,
92
+ stability,
90
93
  judgeAgreement,
91
94
  shipRecommendation,
92
95
  },
@@ -104,11 +107,11 @@ export function computeVerdict(report, options = {}) {
104
107
  * 5. If diff CI is positive AND treatment passes all gates → PROGRESS.
105
108
  * 6. If diff CI is positive but treatment broke a gate → CAUTIOUS.
106
109
  */
107
- function verdictForPair(pair, summary, sampleCount, report, ciThreshold, triviallySmallDiff) {
110
+ function verdictForPair(pair, summary, sampleCount, report, gateThreshold, triviallySmallDiff) {
108
111
  const { control, treatment, diffBootstrapCI: diff } = pair;
109
112
  // Layer-gate check: did any layer fall below threshold for either variant?
110
- const cGate = evaluateCiGates({ [control]: summary[control] }, ciThreshold);
111
- const tGate = evaluateCiGates({ [treatment]: summary[treatment] }, ciThreshold);
113
+ const cGate = evaluateLayerGates({ [control]: summary[control] }, gateThreshold);
114
+ const tGate = evaluateLayerGates({ [treatment]: summary[treatment] }, gateThreshold);
112
115
  // No bootstrap CI available → fall back to point-estimate diff comparison.
113
116
  if (!diff) {
114
117
  const cMean = avgComposite(summary[control]);
@@ -227,6 +230,40 @@ function formatSampleSize(report) {
227
230
  return `N=${n}, saturated @ N=${v.atN ?? '?'} (${v.confidence} confidence)`;
228
231
  return `N=${n}, not yet saturated (${v.confidence} confidence)`;
229
232
  }
233
+ /**
234
+ * Stability rationale. 三种状态:
235
+ * - --repeat ≥ 2 + 有 variance 数据: 报告 CV (variation coefficient) 主指标
236
+ * - --repeat ≥ 2 但 variance 缺失: 异常,标 "—" 提示数据丢失
237
+ * - --repeat < 2: 显式说"未测量,需 --repeat ≥ 2",而不是默默不提
238
+ *
239
+ * 单轮场景关键:不是"稳定 = 100%"(常见误读),而是"测不到稳定性"。
240
+ * Verdict 必须诚实交代这个盲区,不能让用户以为没说就是 OK。
241
+ */
242
+ function formatStability(report) {
243
+ const runs = report.variance?.runs ?? report.meta?.request?.repeat ?? 1;
244
+ if (runs < 2) {
245
+ return '稳定性未测量(单轮评测,需 --repeat ≥ 2 才能测 CV)';
246
+ }
247
+ const variance = report.variance?.perVariant;
248
+ if (!variance || Object.keys(variance).length === 0) {
249
+ return `runs=${runs} 但 variance 数据缺失`;
250
+ }
251
+ // CV = stddev / mean,取所有 variant 的中位数(单 variant 易有 NaN/0,聚合更稳)
252
+ const cvs = [];
253
+ for (const v of Object.values(variance)) {
254
+ if (typeof v.stddev === 'number' && typeof v.mean === 'number' && v.mean > 0) {
255
+ cvs.push(v.stddev / v.mean);
256
+ }
257
+ }
258
+ if (cvs.length === 0)
259
+ return `runs=${runs}, CV 计算失败(stddev/mean 数据缺失)`;
260
+ cvs.sort((a, b) => a - b);
261
+ const median = cvs[Math.floor(cvs.length / 2)];
262
+ const cvPct = (median * 100).toFixed(1);
263
+ // 阈值参考 terminology-spec §5:<5% 稳 / 5-15% 中 / >15% 不稳
264
+ const verdict = median < 0.05 ? '稳定' : median < 0.15 ? '中等' : '不稳';
265
+ return `CV=${cvPct}% (${verdict}, runs=${runs}; 阈值 <5%=稳/5-15%=中/>15%=不稳)`;
266
+ }
230
267
  function formatJudgeAgreement(report) {
231
268
  const a = report.meta?.humanAgreement;
232
269
  if (!a)
@@ -267,6 +304,8 @@ export function formatVerdictText(result, options = {}) {
267
304
  lines.push(` Layer winners: ${result.rationale.layerWinners}`);
268
305
  if (result.rationale.sampleSize)
269
306
  lines.push(` Sample size: ${result.rationale.sampleSize}`);
307
+ if (result.rationale.stability)
308
+ lines.push(` Stability: ${result.rationale.stability}`);
270
309
  if (result.rationale.judgeAgreement)
271
310
  lines.push(` Judge α: ${result.rationale.judgeAgreement}`);
272
311
  if (result.rationale.shipRecommendation)
@@ -1 +1 @@
1
- {"version":3,"file":"verdict.js","sourceRoot":"","sources":["../../../src/eval-core/verdict.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAGH,OAAO,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAuChD;;GAEG;AACH,MAAM,UAAU,cAAc,CAAC,MAAc,EAAE,UAA0B,EAAE;IACzE,MAAM,EAAE,WAAW,GAAG,GAAG,EAAE,kBAAkB,GAAG,GAAG,EAAE,GAAG,OAAO,CAAC;IAChE,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,EAAE,QAAQ,IAAI,EAAE,CAAC;IAC7C,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC;IACrC,MAAM,WAAW,GAAG,MAAM,CAAC,IAAI,EAAE,WAAW,IAAI,CAAC,CAAC;IAElD,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,2EAA2E;QAC3E,mCAAmC;QACnC,MAAM,IAAI,GAAG,eAAe,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;QACnD,OAAO;YACL,KAAK,EAAE,MAAM;YACb,QAAQ,EAAE,IAAI,CAAC,OAAO;gBACpB,CAAC,CAAC,4DAA4D,WAAW,EAAE;gBAC3E,CAAC,CAAC,8DAA8D;YAClE,SAAS,EAAE;gBACT,YAAY,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC;gBACnC,UAAU,EAAE,KAAK,WAAW,EAAE;aAC/B;YACD,QAAQ;SACT,CAAC;IACJ,CAAC;IAED,kFAAkF;IAClF,iFAAiF;IACjF,+EAA+E;IAC/E,MAAM,aAAa,GAAG,MAAM,CAAC,IAAI,EAAE,eAAe,IAAI,EAAE,CAAC;IACzD,MAAM,KAAK,GAA4B,aAAa,CAAC,MAAM,GAAG,CAAC;QAC7D,CAAC,CAAC,aAAa;QACf,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC;IAChF,MAAM,OAAO,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,cAAc,CAAC,CAAC,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,EAAE,WAAW,EAAE,kBAAkB,CAAC,CAAC,CAAC;IAEnH,gGAAgG;IAChG,MAAM,KAAK,GAAmB,CAAC,SAAS,EAAE,UAAU,EAAE,cAAc,EAAE,OAAO,EAAE,UAAU,CAAC,CAAC;IAC3F,IAAI,QAAQ,GAAiB,UAAU,CAAC;IACxC,KAAK,MAAM,KAAK,IAAI,KAAK,EAAE,CAAC;QAC1B,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,KAAK,CAAC,EAAE,CAAC;YAC3C,QAAQ,GAAG,KAAK,CAAC;YACjB,MAAM;QACR,CAAC;IACH,CAAC;IAED,0EAA0E;IAC1E,MAAM,cAAc,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,QAAQ,CAAC,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IAE/E,MAAM,YAAY,GAAG,cAAc;QACjC,CAAC,CAAC,kBAAkB,CAAC,cAAc,CAAC;QACpC,CAAC,CAAC,0DAA0D,CAAC;IAE/D,MAAM,YAAY,GAAG,kBAAkB,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC;IAC3D,MAAM,UAAU,GAAG,gBAAgB,CAAC,MAAM,CAAC,CAAC;IAC5C,MAAM,cAAc,GAAG,oBAAoB,CAAC,MAAM,CAAC,CAAC;IACpD,MAAM,kBAAkB,GAAG,cAAc,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IAE7D,OAAO;QACL,KAAK,EAAE,QAAQ;QACf,QAAQ,EAAE,cAAc;YACtB,CAAC,CAAC,GAAG,QAAQ,MAAM,cAAc,CAAC,SAAS,OAAO,cAAc,CAAC,OAAO,KAAK,cAAc,CAAC,QAAQ,EAAE;YACtG,CAAC,CAAC,GAAG,QAAQ,MAAM,QAAQ,CAAC,MAAM,WAAW;QAC/C,OAAO;QACP,SAAS,EAAE;YACT,YAAY;YACZ,YAAY;YACZ,UAAU;YACV,cAAc;YACd,kBAAkB;SACnB;QACD,QAAQ;KACT,CAAC;AACJ,CAAC;AAED;;;;;;;;;;GAUG;AACH,SAAS,cAAc,CACrB,IAA2B,EAC3B,OAAuC,EACvC,WAAmB,EACnB,MAAc,EACd,WAAmB,EACnB,kBAA0B;IAE1B,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,eAAe,EAAE,IAAI,EAAE,GAAG,IAAI,CAAC;IAE3D,2EAA2E;IAC3E,MAAM,KAAK,GAAG,eAAe,CAAC,EAAE,CAAC,OAAO,CAAC,EAAE,OAAO,CAAC,OAAO,CAAC,EAAE,EAAE,WAAW,CAAC,CAAC;IAC5E,MAAM,KAAK,GAAG,eAAe,CAAC,EAAE,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC,SAAS,CAAC,EAAE,EAAE,WAAW,CAAC,CAAC;IAEhF,2EAA2E;IAC3E,IAAI,CAAC,IAAI,EAAE,CAAC;QACV,MAAM,KAAK,GAAG,YAAY,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC;QAC7C,MAAM,KAAK,GAAG,YAAY,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC;QAC/C,MAAM,KAAK,GAAG,KAAK,GAAG,KAAK,CAAC;QAC5B,MAAM,QAAQ,GAAG,KAAK,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,kCAAkC,CAAC;QACjG,IAAI,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,kBAAkB,EAAE,CAAC;YACzC,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC;QAC1D,CAAC;QACD,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS;YACzC,QAAQ;SACT,CAAC;IACJ,CAAC;IAED,MAAM,YAAY,GAAG,KAAK,IAAI,CAAC,QAAQ,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,QAAQ,QAAQ,IAAI,CAAC,GAAG,KAAK,IAAI,CAAC,IAAI,GAAG,CAAC;IAEzG,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC;QACtB,wFAAwF;QACxF,gFAAgF;QAChF,MAAM,UAAU,GAAG,MAAM,CAAC,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC,SAAS,CAAC,CAAC;QACtE,MAAM,YAAY,GAChB,WAAW,GAAG,EAAE;YAChB,CAAC,UAAU,IAAI,CAAC,UAAU,CAAC,SAAS,IAAI,UAAU,CAAC,UAAU,KAAK,MAAM,CAAC,CAAC;QAC5E,IAAI,YAAY,EAAE,CAAC;YACjB,OAAO;gBACL,OAAO;gBACP,SAAS;gBACT,KAAK,EAAE,cAAc;gBACrB,QAAQ,EAAE,GAAG,YAAY,QAAQ,WAAW,mBAAmB;aAChE,CAAC;QACJ,CAAC;QACD,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,OAAO,EAAE,QAAQ,EAAE,GAAG,YAAY,eAAe,EAAE,CAAC;IAC1F,CAAC;IAED,uBAAuB;IACvB,IAAI,IAAI,CAAC,QAAQ,GAAG,CAAC,EAAE,CAAC;QACtB,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,SAAS,EAAE,QAAQ,EAAE,GAAG,YAAY,oBAAoB,EAAE,CAAC;IACjG,CAAC;IAED,wDAAwD;IACxD,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;QACnB,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,UAAU;YACjB,QAAQ,EAAE,GAAG,YAAY,8CAA8C;SACxE,CAAC;IACJ,CAAC;IAED,wBAAwB;IACxB,IAAI,IAAI,CAAC,QAAQ,GAAG,kBAAkB,EAAE,CAAC;QACvC,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,UAAU;YACjB,QAAQ,EAAE,GAAG,YAAY,qCAAqC;SAC/D,CAAC;IACJ,CAAC;IAED,4EAA4E;IAC5E,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;QACnB,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,UAAU;YACjB,QAAQ,EAAE,GAAG,YAAY,2CAA2C;SACrE,CAAC;IACJ,CAAC;IAED,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,UAAU,EAAE,QAAQ,EAAE,GAAG,YAAY,cAAc,EAAE,CAAC;AAC5F,CAAC;AAED,SAAS,YAAY,CAAC,CAA6B;IACjD,IAAI,CAAC,CAAC;QAAE,OAAO,CAAC,CAAC;IACjB,IAAI,OAAO,CAAC,CAAC,iBAAiB,KAAK,QAAQ;QAAE,OAAO,CAAC,CAAC,iBAAiB,CAAC;IACxE,6EAA6E;IAC7E,MAAM,MAAM,GAAG,CAAC,CAAC,CAAC,YAAY,EAAE,CAAC,CAAC,gBAAgB,EAAE,CAAC,CAAC,aAAa,CAAC,CAAC,MAAM,CACzE,CAAC,CAAC,EAAe,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAC1C,CAAC;IACF,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAClC,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;AAC3D,CAAC;AAED,SAAS,kBAAkB,CAAC,CAA4C;IACtE,OAAO,GAAG,CAAC,CAAC,KAAK,MAAM,CAAC,CAAC,QAAQ,EAAE,CAAC;AACtC,CAAC;AAED,SAAS,kBAAkB,CAAC,OAAuC,EAAE,QAAkB;IACrF,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,GAAG,CAAC;IACpC,MAAM,OAAO,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;IAC5B,MAAM,SAAS,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;IAC9B,MAAM,CAAC,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;IAC3B,MAAM,CAAC,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IAC7B,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC;QAAE,OAAO,GAAG,CAAC;IACzB,MAAM,MAAM,GAA4D;QACtE,CAAC,MAAM,EAAE,CAAC,CAAC,YAAY,EAAE,CAAC,CAAC,YAAY,CAAC;QACxC,CAAC,UAAU,EAAE,CAAC,CAAC,gBAAgB,EAAE,CAAC,CAAC,gBAAgB,CAAC;QACpD,CAAC,OAAO,EAAE,CAAC,CAAC,aAAa,EAAE,CAAC,CAAC,aAAa,CAAC;KAC5C,CAAC;IACF,OAAO,MAAM;SACV,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,EAAE,EAAE,EAAE,CAAC,EAAE,EAAE;QACtB,IAAI,OAAO,EAAE,KAAK,QAAQ,IAAI,OAAO,EAAE,KAAK,QAAQ;YAAE,OAAO,GAAG,IAAI,IAAI,CAAC;QACzE,MAAM,KAAK,GAAG,EAAE,GAAG,EAAE,CAAC;QACtB,MAAM,IAAI,GAAG,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;QACnC,OAAO,GAAG,IAAI,KAAK,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;IACnF,CAAC,CAAC;SACD,IAAI,CAAC,IAAI,CAAC,CAAC;AAChB,CAAC;AAED,SAAS,gBAAgB,CAAC,MAAc;IACtC,MAAM,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,WAAW,IAAI,CAAC,CAAC;IACxC,MAAM,GAAG,GAAG,MAAM,CAAC,QAAQ,EAAE,UAAU,EAAE,QAAQ,CAAC;IAClD,IAAI,CAAC,GAAG;QAAE,OAAO,KAAK,CAAC,EAAE,CAAC;IAC1B,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,EAAE,QAAQ,IAAI,EAAE,CAAC;IAC7C,MAAM,SAAS,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;IAC9B,MAAM,CAAC,GAAG,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACjD,IAAI,CAAC,CAAC;QAAE,OAAO,KAAK,CAAC,EAAE,CAAC;IACxB,IAAI,CAAC,CAAC,SAAS;QAAE,OAAO,KAAK,CAAC,mBAAmB,CAAC,CAAC,GAAG,IAAI,GAAG,KAAK,CAAC,CAAC,UAAU,cAAc,CAAC;IAC7F,OAAO,KAAK,CAAC,wBAAwB,CAAC,CAAC,UAAU,cAAc,CAAC;AAClE,CAAC;AAED,SAAS,oBAAoB,CAAC,MAAc;IAC1C,MAAM,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,cAAc,CAAC;IACtC,IAAI,CAAC,CAAC;QAAE,OAAO,SAAS,CAAC;IACzB,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC;QACnC,CAAC,CAAC,WAAW;QACb,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,GAAG,CAAC,CAAC,CAAC,QAAQ;YAC3B,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,KAAK,CAAC,CAAC,CAAC,YAAY;gBACjC,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,GAAG,CAAC,CAAC,CAAC,MAAM;oBACzB,CAAC,CAAC,MAAM,CAAC;IACX,OAAO,KAAK,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,OAAO,aAAa,CAAC,CAAC,aAAa,EAAE,CAAC;AAC3G,CAAC;AAED,SAAS,cAAc,CAAC,KAAmB,EAAE,QAAwC;IACnF,QAAQ,KAAK,EAAE,CAAC;QACd,KAAK,UAAU;YACb,OAAO,sEAAsE,CAAC;QAChF,KAAK,UAAU;YACb,OAAO,uIAAuI,CAAC;QACjJ,KAAK,SAAS;YACZ,OAAO,mFAAmF,CAAC;QAC7F,KAAK,OAAO;YACV,OAAO,0EAA0E,CAAC;QACpF,KAAK,cAAc;YACjB,OAAO,6DAA6D,CAAC;QACvE,KAAK,MAAM;YACT,OAAO,2FAA2F,CAAC;IACvG,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,iBAAiB,CAAC,MAAqB,EAAE,UAAiC,EAAE;IAC1F,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,KAAK,CAAC,IAAI,CAAC,YAAY,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;IACvC,KAAK,CAAC,IAAI,CAAC,KAAK,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC;IACnC,IAAI,MAAM,CAAC,SAAS,CAAC,YAAY;QAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,MAAM,CAAC,SAAS,CAAC,YAAY,EAAE,CAAC,CAAC;IACnG,IAAI,MAAM,CAAC,SAAS,CAAC,UAAU;QAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,MAAM,CAAC,SAAS,CAAC,UAAU,EAAE,CAAC,CAAC;IAC/F,IAAI,MAAM,CAAC,SAAS,CAAC,cAAc;QAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,MAAM,CAAC,SAAS,CAAC,cAAc,EAAE,CAAC,CAAC;IACvG,IAAI,MAAM,CAAC,SAAS,CAAC,kBAAkB;QAAE,KAAK,CAAC,IAAI,CAAC,KAAK,MAAM,CAAC,SAAS,CAAC,kBAAkB,EAAE,CAAC,CAAC;IAChG,IAAI,OAAO,CAAC,OAAO,IAAI,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACnE,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;QACjC,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;YAC/B,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,KAAK,KAAK,CAAC,CAAC,SAAS,OAAO,CAAC,CAAC,OAAO,MAAM,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC;QAC/E,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC"}
1
+ {"version":3,"file":"verdict.js","sourceRoot":"","sources":["../../../src/eval-core/verdict.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAGH,OAAO,EAAE,kBAAkB,EAAE,MAAM,kBAAkB,CAAC;AA0CtD;;GAEG;AACH,MAAM,UAAU,cAAc,CAAC,MAAc,EAAE,UAA0B,EAAE;IACzE,MAAM,EAAE,aAAa,GAAG,GAAG,EAAE,kBAAkB,GAAG,GAAG,EAAE,GAAG,OAAO,CAAC;IAClE,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,EAAE,QAAQ,IAAI,EAAE,CAAC;IAC7C,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC;IACrC,MAAM,WAAW,GAAG,MAAM,CAAC,IAAI,EAAE,WAAW,IAAI,CAAC,CAAC;IAElD,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,2EAA2E;QAC3E,mCAAmC;QACnC,MAAM,IAAI,GAAG,kBAAkB,CAAC,OAAO,EAAE,aAAa,CAAC,CAAC;QACxD,OAAO;YACL,KAAK,EAAE,MAAM;YACb,QAAQ,EAAE,IAAI,CAAC,OAAO;gBACpB,CAAC,CAAC,4DAA4D,aAAa,EAAE;gBAC7E,CAAC,CAAC,8DAA8D;YAClE,SAAS,EAAE;gBACT,YAAY,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC;gBACnC,UAAU,EAAE,KAAK,WAAW,EAAE;gBAC9B,SAAS,EAAE,eAAe,CAAC,MAAM,CAAC;aACnC;YACD,QAAQ;SACT,CAAC;IACJ,CAAC;IAED,kFAAkF;IAClF,iFAAiF;IACjF,+EAA+E;IAC/E,MAAM,aAAa,GAAG,MAAM,CAAC,IAAI,EAAE,eAAe,IAAI,EAAE,CAAC;IACzD,MAAM,KAAK,GAA4B,aAAa,CAAC,MAAM,GAAG,CAAC;QAC7D,CAAC,CAAC,aAAa;QACf,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC;IAChF,MAAM,OAAO,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,cAAc,CAAC,CAAC,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,EAAE,aAAa,EAAE,kBAAkB,CAAC,CAAC,CAAC;IAErH,gGAAgG;IAChG,MAAM,KAAK,GAAmB,CAAC,SAAS,EAAE,UAAU,EAAE,cAAc,EAAE,OAAO,EAAE,UAAU,CAAC,CAAC;IAC3F,IAAI,QAAQ,GAAiB,UAAU,CAAC;IACxC,KAAK,MAAM,KAAK,IAAI,KAAK,EAAE,CAAC;QAC1B,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,KAAK,CAAC,EAAE,CAAC;YAC3C,QAAQ,GAAG,KAAK,CAAC;YACjB,MAAM;QACR,CAAC;IACH,CAAC;IAED,0EAA0E;IAC1E,MAAM,cAAc,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,QAAQ,CAAC,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IAE/E,MAAM,YAAY,GAAG,cAAc;QACjC,CAAC,CAAC,kBAAkB,CAAC,cAAc,CAAC;QACpC,CAAC,CAAC,0DAA0D,CAAC;IAE/D,MAAM,YAAY,GAAG,kBAAkB,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC;IAC3D,MAAM,UAAU,GAAG,gBAAgB,CAAC,MAAM,CAAC,CAAC;IAC5C,MAAM,SAAS,GAAG,eAAe,CAAC,MAAM,CAAC,CAAC;IAC1C,MAAM,cAAc,GAAG,oBAAoB,CAAC,MAAM,CAAC,CAAC;IACpD,MAAM,kBAAkB,GAAG,cAAc,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IAE7D,OAAO;QACL,KAAK,EAAE,QAAQ;QACf,QAAQ,EAAE,cAAc;YACtB,CAAC,CAAC,GAAG,QAAQ,MAAM,cAAc,CAAC,SAAS,OAAO,cAAc,CAAC,OAAO,KAAK,cAAc,CAAC,QAAQ,EAAE;YACtG,CAAC,CAAC,GAAG,QAAQ,MAAM,QAAQ,CAAC,MAAM,WAAW;QAC/C,OAAO;QACP,SAAS,EAAE;YACT,YAAY;YACZ,YAAY;YACZ,UAAU;YACV,SAAS;YACT,cAAc;YACd,kBAAkB;SACnB;QACD,QAAQ;KACT,CAAC;AACJ,CAAC;AAED;;;;;;;;;;GAUG;AACH,SAAS,cAAc,CACrB,IAA2B,EAC3B,OAAuC,EACvC,WAAmB,EACnB,MAAc,EACd,aAAqB,EACrB,kBAA0B;IAE1B,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,eAAe,EAAE,IAAI,EAAE,GAAG,IAAI,CAAC;IAE3D,2EAA2E;IAC3E,MAAM,KAAK,GAAG,kBAAkB,CAAC,EAAE,CAAC,OAAO,CAAC,EAAE,OAAO,CAAC,OAAO,CAAC,EAAE,EAAE,aAAa,CAAC,CAAC;IACjF,MAAM,KAAK,GAAG,kBAAkB,CAAC,EAAE,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC,SAAS,CAAC,EAAE,EAAE,aAAa,CAAC,CAAC;IAErF,2EAA2E;IAC3E,IAAI,CAAC,IAAI,EAAE,CAAC;QACV,MAAM,KAAK,GAAG,YAAY,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC;QAC7C,MAAM,KAAK,GAAG,YAAY,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC;QAC/C,MAAM,KAAK,GAAG,KAAK,GAAG,KAAK,CAAC;QAC5B,MAAM,QAAQ,GAAG,KAAK,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,kCAAkC,CAAC;QACjG,IAAI,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,kBAAkB,EAAE,CAAC;YACzC,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC;QAC1D,CAAC;QACD,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS;YACzC,QAAQ;SACT,CAAC;IACJ,CAAC;IAED,MAAM,YAAY,GAAG,KAAK,IAAI,CAAC,QAAQ,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,QAAQ,QAAQ,IAAI,CAAC,GAAG,KAAK,IAAI,CAAC,IAAI,GAAG,CAAC;IAEzG,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC;QACtB,wFAAwF;QACxF,gFAAgF;QAChF,MAAM,UAAU,GAAG,MAAM,CAAC,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC,SAAS,CAAC,CAAC;QACtE,MAAM,YAAY,GAChB,WAAW,GAAG,EAAE;YAChB,CAAC,UAAU,IAAI,CAAC,UAAU,CAAC,SAAS,IAAI,UAAU,CAAC,UAAU,KAAK,MAAM,CAAC,CAAC;QAC5E,IAAI,YAAY,EAAE,CAAC;YACjB,OAAO;gBACL,OAAO;gBACP,SAAS;gBACT,KAAK,EAAE,cAAc;gBACrB,QAAQ,EAAE,GAAG,YAAY,QAAQ,WAAW,mBAAmB;aAChE,CAAC;QACJ,CAAC;QACD,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,OAAO,EAAE,QAAQ,EAAE,GAAG,YAAY,eAAe,EAAE,CAAC;IAC1F,CAAC;IAED,uBAAuB;IACvB,IAAI,IAAI,CAAC,QAAQ,GAAG,CAAC,EAAE,CAAC;QACtB,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,SAAS,EAAE,QAAQ,EAAE,GAAG,YAAY,oBAAoB,EAAE,CAAC;IACjG,CAAC;IAED,wDAAwD;IACxD,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;QACnB,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,UAAU;YACjB,QAAQ,EAAE,GAAG,YAAY,8CAA8C;SACxE,CAAC;IACJ,CAAC;IAED,wBAAwB;IACxB,IAAI,IAAI,CAAC,QAAQ,GAAG,kBAAkB,EAAE,CAAC;QACvC,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,UAAU;YACjB,QAAQ,EAAE,GAAG,YAAY,qCAAqC;SAC/D,CAAC;IACJ,CAAC;IAED,4EAA4E;IAC5E,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;QACnB,OAAO;YACL,OAAO;YACP,SAAS;YACT,KAAK,EAAE,UAAU;YACjB,QAAQ,EAAE,GAAG,YAAY,2CAA2C;SACrE,CAAC;IACJ,CAAC;IAED,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,UAAU,EAAE,QAAQ,EAAE,GAAG,YAAY,cAAc,EAAE,CAAC;AAC5F,CAAC;AAED,SAAS,YAAY,CAAC,CAA6B;IACjD,IAAI,CAAC,CAAC;QAAE,OAAO,CAAC,CAAC;IACjB,IAAI,OAAO,CAAC,CAAC,iBAAiB,KAAK,QAAQ;QAAE,OAAO,CAAC,CAAC,iBAAiB,CAAC;IACxE,6EAA6E;IAC7E,MAAM,MAAM,GAAG,CAAC,CAAC,CAAC,YAAY,EAAE,CAAC,CAAC,gBAAgB,EAAE,CAAC,CAAC,aAAa,CAAC,CAAC,MAAM,CACzE,CAAC,CAAC,EAAe,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAC1C,CAAC;IACF,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAClC,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;AAC3D,CAAC;AAED,SAAS,kBAAkB,CAAC,CAA4C;IACtE,OAAO,GAAG,CAAC,CAAC,KAAK,MAAM,CAAC,CAAC,QAAQ,EAAE,CAAC;AACtC,CAAC;AAED,SAAS,kBAAkB,CAAC,OAAuC,EAAE,QAAkB;IACrF,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,GAAG,CAAC;IACpC,MAAM,OAAO,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;IAC5B,MAAM,SAAS,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;IAC9B,MAAM,CAAC,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;IAC3B,MAAM,CAAC,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;IAC7B,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC;QAAE,OAAO,GAAG,CAAC;IACzB,MAAM,MAAM,GAA4D;QACtE,CAAC,MAAM,EAAE,CAAC,CAAC,YAAY,EAAE,CAAC,CAAC,YAAY,CAAC;QACxC,CAAC,UAAU,EAAE,CAAC,CAAC,gBAAgB,EAAE,CAAC,CAAC,gBAAgB,CAAC;QACpD,CAAC,OAAO,EAAE,CAAC,CAAC,aAAa,EAAE,CAAC,CAAC,aAAa,CAAC;KAC5C,CAAC;IACF,OAAO,MAAM;SACV,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,EAAE,EAAE,EAAE,CAAC,EAAE,EAAE;QACtB,IAAI,OAAO,EAAE,KAAK,QAAQ,IAAI,OAAO,EAAE,KAAK,QAAQ;YAAE,OAAO,GAAG,IAAI,IAAI,CAAC;QACzE,MAAM,KAAK,GAAG,EAAE,GAAG,EAAE,CAAC;QACtB,MAAM,IAAI,GAAG,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;QACnC,OAAO,GAAG,IAAI,KAAK,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;IACnF,CAAC,CAAC;SACD,IAAI,CAAC,IAAI,CAAC,CAAC;AAChB,CAAC;AAED,SAAS,gBAAgB,CAAC,MAAc;IACtC,MAAM,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,WAAW,IAAI,CAAC,CAAC;IACxC,MAAM,GAAG,GAAG,MAAM,CAAC,QAAQ,EAAE,UAAU,EAAE,QAAQ,CAAC;IAClD,IAAI,CAAC,GAAG;QAAE,OAAO,KAAK,CAAC,EAAE,CAAC;IAC1B,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,EAAE,QAAQ,IAAI,EAAE,CAAC;IAC7C,MAAM,SAAS,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;IAC9B,MAAM,CAAC,GAAG,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACjD,IAAI,CAAC,CAAC;QAAE,OAAO,KAAK,CAAC,EAAE,CAAC;IACxB,IAAI,CAAC,CAAC,SAAS;QAAE,OAAO,KAAK,CAAC,mBAAmB,CAAC,CAAC,GAAG,IAAI,GAAG,KAAK,CAAC,CAAC,UAAU,cAAc,CAAC;IAC7F,OAAO,KAAK,CAAC,wBAAwB,CAAC,CAAC,UAAU,cAAc,CAAC;AAClE,CAAC;AAED;;;;;;;;GAQG;AACH,SAAS,eAAe,CAAC,MAAc;IACrC,MAAM,IAAI,GAAG,MAAM,CAAC,QAAQ,EAAE,IAAI,IAAI,MAAM,CAAC,IAAI,EAAE,OAAO,EAAE,MAAM,IAAI,CAAC,CAAC;IACxE,IAAI,IAAI,GAAG,CAAC,EAAE,CAAC;QACb,OAAO,oCAAoC,CAAC;IAC9C,CAAC;IACD,MAAM,QAAQ,GAAG,MAAM,CAAC,QAAQ,EAAE,UAAU,CAAC;IAC7C,IAAI,CAAC,QAAQ,IAAI,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpD,OAAO,QAAQ,IAAI,kBAAkB,CAAC;IACxC,CAAC;IACD,+DAA+D;IAC/D,MAAM,GAAG,GAAa,EAAE,CAAC;IACzB,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;QACxC,IAAI,OAAO,CAAC,CAAC,MAAM,KAAK,QAAQ,IAAI,OAAO,CAAC,CAAC,IAAI,KAAK,QAAQ,IAAI,CAAC,CAAC,IAAI,GAAG,CAAC,EAAE,CAAC;YAC7E,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC;QAC9B,CAAC;IACH,CAAC;IACD,IAAI,GAAG,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,QAAQ,IAAI,6BAA6B,CAAC;IACvE,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC1B,MAAM,MAAM,GAAG,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC;IAC/C,MAAM,KAAK,GAAG,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;IACxC,qDAAqD;IACrD,MAAM,OAAO,GAAG,MAAM,GAAG,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC;IACnE,OAAO,MAAM,KAAK,MAAM,OAAO,UAAU,IAAI,6BAA6B,CAAC;AAC7E,CAAC;AAED,SAAS,oBAAoB,CAAC,MAAc;IAC1C,MAAM,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,cAAc,CAAC;IACtC,IAAI,CAAC,CAAC;QAAE,OAAO,SAAS,CAAC;IACzB,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC;QACnC,CAAC,CAAC,WAAW;QACb,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,GAAG,CAAC,CAAC,CAAC,QAAQ;YAC3B,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,KAAK,CAAC,CAAC,CAAC,YAAY;gBACjC,CAAC,CAAC,CAAC,CAAC,KAAK,IAAI,GAAG,CAAC,CAAC,CAAC,MAAM;oBACzB,CAAC,CAAC,MAAM,CAAC;IACX,OAAO,KAAK,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,OAAO,aAAa,CAAC,CAAC,aAAa,EAAE,CAAC;AAC3G,CAAC;AAED,SAAS,cAAc,CAAC,KAAmB,EAAE,QAAwC;IACnF,QAAQ,KAAK,EAAE,CAAC;QACd,KAAK,UAAU;YACb,OAAO,sEAAsE,CAAC;QAChF,KAAK,UAAU;YACb,OAAO,uIAAuI,CAAC;QACjJ,KAAK,SAAS;YACZ,OAAO,mFAAmF,CAAC;QAC7F,KAAK,OAAO;YACV,OAAO,0EAA0E,CAAC;QACpF,KAAK,cAAc;YACjB,OAAO,6DAA6D,CAAC;QACvE,KAAK,MAAM;YACT,OAAO,2FAA2F,CAAC;IACvG,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,iBAAiB,CAAC,MAAqB,EAAE,UAAiC,EAAE;IAC1F,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,KAAK,CAAC,IAAI,CAAC,YAAY,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;IACvC,KAAK,CAAC,IAAI,CAAC,KAAK,MAAM,CAAC,QAAQ,EAAE,CAAC,CAAC;IACnC,IAAI,MAAM,CAAC,SAAS,CAAC,YAAY;QAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,MAAM,CAAC,SAAS,CAAC,YAAY,EAAE,CAAC,CAAC;IACnG,IAAI,MAAM,CAAC,SAAS,CAAC,UAAU;QAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,MAAM,CAAC,SAAS,CAAC,UAAU,EAAE,CAAC,CAAC;IAC/F,IAAI,MAAM,CAAC,SAAS,CAAC,SAAS;QAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,MAAM,CAAC,SAAS,CAAC,SAAS,EAAE,CAAC,CAAC;IAC7F,IAAI,MAAM,CAAC,SAAS,CAAC,cAAc;QAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,MAAM,CAAC,SAAS,CAAC,cAAc,EAAE,CAAC,CAAC;IACvG,IAAI,MAAM,CAAC,SAAS,CAAC,kBAAkB;QAAE,KAAK,CAAC,IAAI,CAAC,KAAK,MAAM,CAAC,SAAS,CAAC,kBAAkB,EAAE,CAAC,CAAC;IAChG,IAAI,OAAO,CAAC,OAAO,IAAI,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACnE,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;QACjC,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;YAC/B,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,KAAK,KAAK,CAAC,CAAC,SAAS,OAAO,CAAC,CAAC,OAAO,MAAM,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC;QAC/E,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC"}
@@ -1,4 +1,4 @@
1
- import type { Artifact, JobStore, ProgressCallback, Report, VarianceData, VariantResult, VariantSummary } from '../types.js';
1
+ import type { Artifact, JobStore, ProgressCallback, Report, VarianceData, VariantResult, VariantSummary } from '../types/index.js';
2
2
  interface RunSingleEvaluationOptions {
3
3
  samplesPath: string;
4
4
  skillDir: string;
@@ -20,7 +20,7 @@ interface RunSingleEvaluationOptions {
20
20
  /** Forwarded to grade(); each sample × dimension is judged N times. Default 1. */
21
21
  judgeRepeat?: number;
22
22
  /** Forwarded to pipeline; ≥ 2 entries triggers multi-judge ensemble mode. */
23
- judgeModels?: import('../types.js').JudgeConfig[];
23
+ judgeModels?: import('../types/index.js').JudgeConfig[];
24
24
  /** v0.21 Phase 3a length-debias toggle. Default true. */
25
25
  lengthDebias?: boolean;
26
26
  }
@@ -79,9 +79,9 @@ export declare function buildEachReport({ skillDir, skillEntries, skillResults,
79
79
  artifactHashes: {
80
80
  [k: string]: string;
81
81
  };
82
- request: import("../types.js").EvaluationRequest;
83
- run: import("../types.js").EvaluationRun;
84
- job: import("../types.js").EvaluationJob;
82
+ request: import("../types/eval.js").EvaluationRequest;
83
+ run: import("../types/eval.js").EvaluationRun;
84
+ job: import("../types/eval.js").EvaluationJob;
85
85
  };
86
86
  summary: {};
87
87
  results: never[];
@@ -98,7 +98,7 @@ export declare function buildEachReport({ skillDir, skillEntries, skillResults,
98
98
  };
99
99
  artifacts: EachSkillResult[];
100
100
  };
101
- job: import("../types.js").EvaluationJob;
101
+ job: import("../types/eval.js").EvaluationJob;
102
102
  };
103
103
  export declare function executeEachEvaluationRuns({ skillDir, skillEntries, model, judgeModel, outputDir, project, owner, tags, noJudge, concurrency, timeoutMs, executorName, judgeExecutorName, jobStore, persistJob, onProgress, onSkillProgress, skipPreflight, mcpConfig, verbose, repeat, judgeRepeat, judgeModels, lengthDebias, runSingleEvaluation, }: {
104
104
  skillDir: string;
@@ -132,7 +132,7 @@ export declare function executeEachEvaluationRuns({ skillDir, skillEntries, mode
132
132
  verbose?: boolean;
133
133
  repeat?: number;
134
134
  judgeRepeat?: number;
135
- judgeModels?: import('../types.js').JudgeConfig[];
135
+ judgeModels?: import('../types/index.js').JudgeConfig[];
136
136
  lengthDebias?: boolean;
137
137
  runSingleEvaluation: (options: RunSingleEvaluationOptions) => Promise<{
138
138
  report: Report;
@@ -1 +1 @@
1
- {"version":3,"file":"each-evaluation-workflow.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/each-evaluation-workflow.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,QAAQ,EAAE,QAAQ,EAAE,gBAAgB,EAAE,MAAM,EAAE,YAAY,EAAE,aAAa,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAE7H,UAAU,0BAA0B;IAClC,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,IAAI,CAAC;IAChB,OAAO,EAAE,OAAO,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,EAAE,IAAI,CAAC;IACf,UAAU,EAAE,KAAK,CAAC;IAClB,UAAU,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACpC,aAAa,EAAE,OAAO,CAAC;IACvB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,kFAAkF;IAClF,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,6EAA6E;IAC7E,WAAW,CAAC,EAAE,OAAO,aAAa,EAAE,WAAW,EAAE,CAAC;IAClD,yDAAyD;IACzD,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AAED,MAAM,WAAW,eAAe;IAC9B,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IACxC,0DAA0D;IAC1D,QAAQ,CAAC,EAAE,YAAY,CAAC;IACxB,OAAO,EAAE,KAAK,CAAC;QACb,SAAS,EAAE,MAAM,CAAC;QAClB,QAAQ,EAAE;YACR,QAAQ,EAAE,aAAa,CAAC;YACxB,KAAK,EAAE,aAAa,CAAC;SACtB,CAAC;KACH,CAAC,CAAC;CACJ;AAsBD,wBAAgB,eAAe,CAAC,EAC9B,QAAQ,EACR,YAAY,EACZ,YAAY,EACZ,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,OAAO,EACP,KAAK,EACL,IAAI,EACJ,MAAM,EACN,WAAW,EACX,SAAS,EACT,YAAY,EACZ,MAAM,GACP,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC9E,YAAY,EAAE,eAAe,EAAE,CAAC;IAChC,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,MAAM,EAAE,OAAO,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAwEA;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,QAAQ,EACR,YAAY,EACZ,KAAK,EACL,UAAU,EACV,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,WAAe,EACf,SAAS,EACT,YAAY,EACZ,iBAAiB,EACjB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,eAAsB,EACtB,aAAqB,EACrB,SAAS,EACT,OAAe,EACf,MAAM,EACN,WAAW,EACX,WAAW,EACX,YAAY,EACZ,mBAAmB,GACpB,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC9E,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,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,EAAE,MAAM,CAAC;IACrB,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,eAAe,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;IAC5G,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,OAAO,aAAa,EAAE,WAAW,EAAE,CAAC;IAClD,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,mBAAmB,EAAE,CAAC,OAAO,EAAE,0BAA0B,KAAK,OAAO,CAAC;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;KAAE,CAAC,CAAC;CACpH,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CAqFvD"}
1
+ {"version":3,"file":"each-evaluation-workflow.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/each-evaluation-workflow.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,QAAQ,EAAE,QAAQ,EAAE,gBAAgB,EAAE,MAAM,EAAE,YAAY,EAAE,aAAa,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AAEnI,UAAU,0BAA0B;IAClC,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,IAAI,CAAC;IAChB,OAAO,EAAE,OAAO,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,EAAE,IAAI,CAAC;IACf,UAAU,EAAE,KAAK,CAAC;IAClB,UAAU,EAAE,gBAAgB,GAAG,IAAI,CAAC;IACpC,aAAa,EAAE,OAAO,CAAC;IACvB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,kFAAkF;IAClF,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,6EAA6E;IAC7E,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,yDAAyD;IACzD,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AAED,MAAM,WAAW,eAAe;IAC9B,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IACxC,0DAA0D;IAC1D,QAAQ,CAAC,EAAE,YAAY,CAAC;IACxB,OAAO,EAAE,KAAK,CAAC;QACb,SAAS,EAAE,MAAM,CAAC;QAClB,QAAQ,EAAE;YACR,QAAQ,EAAE,aAAa,CAAC;YACxB,KAAK,EAAE,aAAa,CAAC;SACtB,CAAC;KACH,CAAC,CAAC;CACJ;AAsBD,wBAAgB,eAAe,CAAC,EAC9B,QAAQ,EACR,YAAY,EACZ,YAAY,EACZ,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,iBAAiB,EACjB,OAAO,EACP,KAAK,EACL,IAAI,EACJ,MAAM,EACN,WAAW,EACX,SAAS,EACT,YAAY,EACZ,MAAM,GACP,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC9E,YAAY,EAAE,eAAe,EAAE,CAAC;IAChC,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,MAAM,EAAE,OAAO,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAwEA;AAED,wBAAsB,yBAAyB,CAAC,EAC9C,QAAQ,EACR,YAAY,EACZ,KAAK,EACL,UAAU,EACV,SAA8B,EAC9B,OAAO,EACP,KAAK,EACL,IAAI,EACJ,OAAe,EACf,WAAe,EACf,SAAS,EACT,YAAY,EACZ,iBAAiB,EACjB,QAAe,EACf,UAAiB,EACjB,UAAiB,EACjB,eAAsB,EACtB,aAAqB,EACrB,SAAS,EACT,OAAe,EACf,MAAM,EACN,WAAW,EACX,WAAW,EACX,YAAY,EACZ,mBAAmB,GACpB,EAAE;IACD,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC9E,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,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,EAAE,MAAM,CAAC;IACrB,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,eAAe,CAAC,EAAE,CAAC,CAAC,IAAI,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,CAAC,GAAG,IAAI,CAAC;IAC5G,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,OAAO,mBAAmB,EAAE,WAAW,EAAE,CAAC;IACxD,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,mBAAmB,EAAE,CAAC,OAAO,EAAE,0BAA0B,KAAK,OAAO,CAAC;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;KAAE,CAAC,CAAC;CACpH,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CAqFvD"}
@@ -1,5 +1,20 @@
1
1
  import type { DependencyRequirements } from '../eval-core/dependency-checker.js';
2
- import type { Artifact, ExecutorFn, JobStore, ProgressCallback, Report, Sample, Task, VariantResult } from '../types.js';
2
+ import type { Artifact, ExecutorFn, JobStore, ProgressCallback, Report, Sample, Task, VariantResult } from '../types/index.js';
3
+ /**
4
+ * Compute structural power warnings (pure function, no I/O — for testing).
5
+ *
6
+ * Not MDE / power-analysis predictions (we don't have σ pre-run; predicting
7
+ * "CI half-width ~ ±0.4" before any data exists is hand-wave). These are
8
+ * **hard-floor + experience-based** thresholds:
9
+ * - n < 5: any conclusion unreliable, CI uselessly wide
10
+ * - 5 ≤ n < 20: only large effects (Cohen's d > 0.8) detectable
11
+ * - repeat=1: stability cannot be measured at all
12
+ *
13
+ * Real power claims happen post-hoc via `bench verdict` UNDERPOWERED state +
14
+ * saturation curves. This is the upfront "you might be wasting the run"
15
+ * heads-up, not a gate.
16
+ */
17
+ export declare function buildPowerWarnings(sampleCount: number, repeat: number): string[];
3
18
  export interface EvaluationPipelineOptions {
4
19
  samplesPath: string;
5
20
  skillDir: string;
@@ -37,7 +52,7 @@ export interface EvaluationPipelineOptions {
37
52
  /** 透传到 meta.request.judgeRepeat 与 grade(),每条 sample × dimension judge N 次 */
38
53
  judgeRepeat?: number;
39
54
  /** Multi-judge ensemble configs (≥ 2 entries triggers ensemble mode). */
40
- judgeModels?: import('../types.js').JudgeConfig[];
55
+ judgeModels?: import('../types/index.js').JudgeConfig[];
41
56
  /** --bootstrap. */
42
57
  bootstrap?: boolean;
43
58
  /** --bootstrap-samples N. Default 1000. */
@@ -45,7 +60,7 @@ export interface EvaluationPipelineOptions {
45
60
  /** v0.21 length-debias toggle. Default true; --no-debias-length flips to false. */
46
61
  lengthDebias?: boolean;
47
62
  /** v0.22 — hard budget caps. */
48
- budget?: import('../types.js').EvalBudget;
63
+ budget?: import('../types/index.js').EvalBudget;
49
64
  }
50
65
  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, }: EvaluationPipelineOptions): Promise<{
51
66
  report: Report;
@@ -1 +1 @@
1
- {"version":3,"file":"evaluation-pipeline.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"AAQA,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,aAAa,CAAC;AA+MrB,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,aAAa,EAAE,WAAW,EAAE,CAAC;IAClD,mBAAmB;IACnB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,2CAA2C;IAC3C,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,mFAAmF;IACnF,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,gCAAgC;IAChC,MAAM,CAAC,EAAE,OAAO,aAAa,EAAE,UAAU,CAAC;CAC3C;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,GACP,EAAE,yBAAyB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CA+HlF"}
1
+ {"version":3,"file":"evaluation-pipeline.d.ts","sourceRoot":"","sources":["../../../src/eval-workflows/evaluation-pipeline.ts"],"names":[],"mappings":"AAQA,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;AA2DD,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,gCAAgC;IAChC,MAAM,CAAC,EAAE,OAAO,mBAAmB,EAAE,UAAU,CAAC;CACjD;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,GACP,EAAE,yBAAyB,GAAG,OAAO,CAAC;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,CAAC,CAqIlF"}
@@ -95,6 +95,38 @@ function computeTestSetHash(samplesPath) {
95
95
  return null;
96
96
  }
97
97
  }
98
+ /**
99
+ * Compute structural power warnings (pure function, no I/O — for testing).
100
+ *
101
+ * Not MDE / power-analysis predictions (we don't have σ pre-run; predicting
102
+ * "CI half-width ~ ±0.4" before any data exists is hand-wave). These are
103
+ * **hard-floor + experience-based** thresholds:
104
+ * - n < 5: any conclusion unreliable, CI uselessly wide
105
+ * - 5 ≤ n < 20: only large effects (Cohen's d > 0.8) detectable
106
+ * - repeat=1: stability cannot be measured at all
107
+ *
108
+ * Real power claims happen post-hoc via `bench verdict` UNDERPOWERED state +
109
+ * saturation curves. This is the upfront "you might be wasting the run"
110
+ * heads-up, not a gate.
111
+ */
112
+ export function buildPowerWarnings(sampleCount, repeat) {
113
+ const warnings = [];
114
+ if (sampleCount < 5) {
115
+ warnings.push(`⚠ N=${sampleCount} < 5 (exploration-only): any conclusion is unreliable, CI will be uselessly wide. Decisions need ≥20 cases.`);
116
+ }
117
+ else if (sampleCount < 20) {
118
+ warnings.push(`⚠ N=${sampleCount} < 20 (large-effect-only, Cohen's d > 0.8): medium effects (d ≈ 0.5) hard to detect. For confident decisions consider ≥20 cases.`);
119
+ }
120
+ if (repeat < 2) {
121
+ warnings.push(`⚠ --repeat=1: single-run cannot measure stability (CV will be marked "not measured"). Use --repeat 3+ to detect within-variant variance.`);
122
+ }
123
+ return warnings;
124
+ }
125
+ function emitPowerWarnings(sampleCount, repeat) {
126
+ for (const w of buildPowerWarnings(sampleCount, repeat)) {
127
+ process.stderr.write(`${w}\n`);
128
+ }
129
+ }
98
130
  function finalizeEvaluationReport({ report, results, artifacts, variantNames, blind, samplesPath, }) {
99
131
  report.analysis = analyzeResults(report);
100
132
  const hasToolData = Object.values(results).some((sampleResults) => (Object.values(sampleResults).some((variantResult) => variantResult.toolCalls && variantResult.toolCalls.length > 0)));
@@ -171,6 +203,11 @@ export async function executeEvaluationPipeline({ samplesPath, skillDir, samples
171
203
  throw new Error(formatDependencyErrors(depResult.missing));
172
204
  }
173
205
  }
206
+ // Structural power warnings — print to stderr after preflight passes, before
207
+ // tasks start. These are *not* MDE / power-analysis predictions (we don't have
208
+ // σ before the run); they're hard-floor + experience-based thresholds. Verdict
209
+ // gate (computeVerdict) handles real power claims post-hoc.
210
+ emitPowerWarnings(samples.length, repeat ?? 1);
174
211
  // Pre-build a per-executor map for ensemble judges. Each unique executor name
175
212
  // gets one ExecutorFn, shared across all judges using that executor. The default
176
213
  // judge's executor is also included so single-judge fallbacks have it available.