agent-hitch 0.2.5 → 0.2.7

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 (395) hide show
  1. package/README.md +14 -1
  2. package/dist/scripts/canary-buildkit-secrets.js +166 -0
  3. package/dist/scripts/canary-buildkit-secrets.js.map +1 -0
  4. package/dist/scripts/canary-collision-domains.js +77 -0
  5. package/dist/scripts/canary-collision-domains.js.map +1 -0
  6. package/dist/scripts/canary-docker-image-observation.js +181 -0
  7. package/dist/scripts/canary-docker-image-observation.js.map +1 -0
  8. package/dist/scripts/canary-gpu-hardware.js +95 -0
  9. package/dist/scripts/canary-gpu-hardware.js.map +1 -0
  10. package/dist/scripts/canary-harbor-load.js +365 -0
  11. package/dist/scripts/canary-harbor-load.js.map +1 -0
  12. package/dist/scripts/canary-packaged-harness.js +75 -0
  13. package/dist/scripts/canary-packaged-harness.js.map +1 -0
  14. package/dist/scripts/canary-resource-load.js +188 -0
  15. package/dist/scripts/canary-resource-load.js.map +1 -0
  16. package/dist/scripts/check-architecture.js +8 -4
  17. package/dist/scripts/check-architecture.js.map +1 -1
  18. package/dist/scripts/docker-reaper-crash-child.js +21 -0
  19. package/dist/scripts/docker-reaper-crash-child.js.map +1 -0
  20. package/dist/src/adapters/catalog.js +1 -0
  21. package/dist/src/adapters/catalog.js.map +1 -1
  22. package/dist/src/adapters/discovery.js +1 -1
  23. package/dist/src/adapters/discovery.js.map +1 -1
  24. package/dist/src/adapters/providers/claude.js +6 -0
  25. package/dist/src/adapters/providers/claude.js.map +1 -1
  26. package/dist/src/adapters/providers/codex.js +6 -0
  27. package/dist/src/adapters/providers/codex.js.map +1 -1
  28. package/dist/src/adapters/providers/deepseek.js +6 -0
  29. package/dist/src/adapters/providers/deepseek.js.map +1 -1
  30. package/dist/src/adapters/providers/opencode.js +6 -0
  31. package/dist/src/adapters/providers/opencode.js.map +1 -1
  32. package/dist/src/adapters/providers/pi.js +6 -0
  33. package/dist/src/adapters/providers/pi.js.map +1 -1
  34. package/dist/src/artifacts/index.js +1 -0
  35. package/dist/src/artifacts/index.js.map +1 -1
  36. package/dist/src/backends/harbor/backend.js +55 -104
  37. package/dist/src/backends/harbor/backend.js.map +1 -1
  38. package/dist/src/backends/harbor/environment-config.js +92 -0
  39. package/dist/src/backends/harbor/environment-config.js.map +1 -0
  40. package/dist/src/backends/harbor/index.js +5 -1
  41. package/dist/src/backends/harbor/index.js.map +1 -1
  42. package/dist/src/backends/harbor/model-proxy-config.js +23 -0
  43. package/dist/src/backends/harbor/model-proxy-config.js.map +1 -0
  44. package/dist/src/backends/harbor/process.js +177 -0
  45. package/dist/src/backends/harbor/process.js.map +1 -0
  46. package/dist/src/backends/harbor/runtime-toolchain.js +30 -0
  47. package/dist/src/backends/harbor/runtime-toolchain.js.map +1 -0
  48. package/dist/src/backends/harbor/supervisor.js +74 -0
  49. package/dist/src/backends/harbor/supervisor.js.map +1 -0
  50. package/dist/src/backends/harbor/task-resources.js +206 -0
  51. package/dist/src/backends/harbor/task-resources.js.map +1 -0
  52. package/dist/src/backends/harbor/tools.js +6 -29
  53. package/dist/src/backends/harbor/tools.js.map +1 -1
  54. package/dist/src/backends/harbor/verifier-config.js +11 -0
  55. package/dist/src/backends/harbor/verifier-config.js.map +1 -0
  56. package/dist/src/backends/index.js +4 -1
  57. package/dist/src/backends/index.js.map +1 -1
  58. package/dist/src/cli/arguments.js +14 -0
  59. package/dist/src/cli/arguments.js.map +1 -1
  60. package/dist/src/cli/commands/daemon.js +135 -8
  61. package/dist/src/cli/commands/daemon.js.map +1 -1
  62. package/dist/src/cli/commands/eval.js +214 -4
  63. package/dist/src/cli/commands/eval.js.map +1 -1
  64. package/dist/src/cli/commands/images.js +40 -0
  65. package/dist/src/cli/commands/images.js.map +1 -0
  66. package/dist/src/cli/commands/run.js +6 -3
  67. package/dist/src/cli/commands/run.js.map +1 -1
  68. package/dist/src/cli/commands/runs.js +25 -3
  69. package/dist/src/cli/commands/runs.js.map +1 -1
  70. package/dist/src/cli/commands/worker.js +79 -0
  71. package/dist/src/cli/commands/worker.js.map +1 -0
  72. package/dist/src/cli/main.js +4 -0
  73. package/dist/src/cli/main.js.map +1 -1
  74. package/dist/src/cli/output.js +72 -3
  75. package/dist/src/cli/output.js.map +1 -1
  76. package/dist/src/control-plane/build-admission.js +69 -0
  77. package/dist/src/control-plane/build-admission.js.map +1 -0
  78. package/dist/src/control-plane/build-records.js +6 -0
  79. package/dist/src/control-plane/build-records.js.map +1 -0
  80. package/dist/src/control-plane/collisions.js +40 -0
  81. package/dist/src/control-plane/collisions.js.map +1 -0
  82. package/dist/src/control-plane/eval-control-state.js +22 -0
  83. package/dist/src/control-plane/eval-control-state.js.map +1 -0
  84. package/dist/src/control-plane/eval-control-work.js +39 -0
  85. package/dist/src/control-plane/eval-control-work.js.map +1 -0
  86. package/dist/src/control-plane/eval-image-resolution.js +52 -0
  87. package/dist/src/control-plane/eval-image-resolution.js.map +1 -0
  88. package/dist/src/control-plane/eval-image-services.js +30 -0
  89. package/dist/src/control-plane/eval-image-services.js.map +1 -0
  90. package/dist/src/control-plane/eval-records.js +252 -0
  91. package/dist/src/control-plane/eval-records.js.map +1 -0
  92. package/dist/src/control-plane/eval-recovery.js +181 -0
  93. package/dist/src/control-plane/eval-recovery.js.map +1 -0
  94. package/dist/src/control-plane/eval-scheduler.js +447 -0
  95. package/dist/src/control-plane/eval-scheduler.js.map +1 -0
  96. package/dist/src/control-plane/index.js +20 -0
  97. package/dist/src/control-plane/index.js.map +1 -0
  98. package/dist/src/control-plane/local-worker.js +49 -0
  99. package/dist/src/control-plane/local-worker.js.map +1 -0
  100. package/dist/src/control-plane/model-capture-planning.js +12 -0
  101. package/dist/src/control-plane/model-capture-planning.js.map +1 -0
  102. package/dist/src/control-plane/remote-result-transport.js +172 -0
  103. package/dist/src/control-plane/remote-result-transport.js.map +1 -0
  104. package/dist/src/control-plane/remote-work-coordinator.js +332 -0
  105. package/dist/src/control-plane/remote-work-coordinator.js.map +1 -0
  106. package/dist/src/control-plane/remote-work-inputs.js +266 -0
  107. package/dist/src/control-plane/remote-work-inputs.js.map +1 -0
  108. package/dist/src/control-plane/remote-work-recovery.js +181 -0
  109. package/dist/src/control-plane/remote-work-recovery.js.map +1 -0
  110. package/dist/src/control-plane/remote-worker-artifacts.js +135 -0
  111. package/dist/src/control-plane/remote-worker-artifacts.js.map +1 -0
  112. package/dist/src/control-plane/remote-worker-client.js +202 -0
  113. package/dist/src/control-plane/remote-worker-client.js.map +1 -0
  114. package/dist/src/control-plane/remote-worker-configuration.js +34 -0
  115. package/dist/src/control-plane/remote-worker-configuration.js.map +1 -0
  116. package/dist/src/control-plane/remote-worker-credentials.js +58 -0
  117. package/dist/src/control-plane/remote-worker-credentials.js.map +1 -0
  118. package/dist/src/control-plane/remote-worker-protocol.js +510 -0
  119. package/dist/src/control-plane/remote-worker-protocol.js.map +1 -0
  120. package/dist/src/control-plane/remote-worker-runner.js +214 -0
  121. package/dist/src/control-plane/remote-worker-runner.js.map +1 -0
  122. package/dist/src/control-plane/remote-workers.js +377 -0
  123. package/dist/src/control-plane/remote-workers.js.map +1 -0
  124. package/dist/src/control-plane/rerun-scheduler.js +411 -0
  125. package/dist/src/control-plane/rerun-scheduler.js.map +1 -0
  126. package/dist/src/control-plane/rerun-submission.js +56 -0
  127. package/dist/src/control-plane/rerun-submission.js.map +1 -0
  128. package/dist/src/control-plane/resources.js +182 -0
  129. package/dist/src/control-plane/resources.js.map +1 -0
  130. package/dist/src/control-plane/scheduler-eval-entry.js +32 -0
  131. package/dist/src/control-plane/scheduler-eval-entry.js.map +1 -0
  132. package/dist/src/control-plane/synthetic-result.js +21 -0
  133. package/dist/src/control-plane/synthetic-result.js.map +1 -0
  134. package/dist/src/control-plane/work-admission.js +21 -0
  135. package/dist/src/control-plane/work-admission.js.map +1 -0
  136. package/dist/src/control-plane/work-dispatcher.js +229 -0
  137. package/dist/src/control-plane/work-dispatcher.js.map +1 -0
  138. package/dist/src/controller-runtime/hash.js +12 -1
  139. package/dist/src/controller-runtime/hash.js.map +1 -1
  140. package/dist/src/controller-runtime/index.js +1 -1
  141. package/dist/src/controller-runtime/index.js.map +1 -1
  142. package/dist/src/controller-runtime/store.js +15 -6
  143. package/dist/src/controller-runtime/store.js.map +1 -1
  144. package/dist/src/daemon/client.js.map +1 -1
  145. package/dist/src/daemon/event-stream.js +46 -0
  146. package/dist/src/daemon/event-stream.js.map +1 -0
  147. package/dist/src/daemon/health.js +66 -0
  148. package/dist/src/daemon/health.js.map +1 -0
  149. package/dist/src/daemon/launcher.js +28 -1
  150. package/dist/src/daemon/launcher.js.map +1 -1
  151. package/dist/src/daemon/logging.js +23 -0
  152. package/dist/src/daemon/logging.js.map +1 -0
  153. package/dist/src/daemon/scheduler.js +26 -5
  154. package/dist/src/daemon/scheduler.js.map +1 -1
  155. package/dist/src/daemon/server.js +255 -98
  156. package/dist/src/daemon/server.js.map +1 -1
  157. package/dist/src/daemon/telemetry.js +163 -0
  158. package/dist/src/daemon/telemetry.js.map +1 -0
  159. package/dist/src/daemon/worker-routes.js +219 -0
  160. package/dist/src/daemon/worker-routes.js.map +1 -0
  161. package/dist/src/domain/bundles.js +2 -0
  162. package/dist/src/domain/bundles.js.map +1 -0
  163. package/dist/src/domain/control-plane.js +2 -0
  164. package/dist/src/domain/control-plane.js.map +1 -0
  165. package/dist/src/domain/execution-evidence.js +2 -0
  166. package/dist/src/domain/execution-evidence.js.map +1 -0
  167. package/dist/src/domain/execution-plan.js +2 -0
  168. package/dist/src/domain/execution-plan.js.map +1 -0
  169. package/dist/src/domain/images.js +2 -0
  170. package/dist/src/domain/images.js.map +1 -0
  171. package/dist/src/domain/index.js +9 -0
  172. package/dist/src/domain/index.js.map +1 -1
  173. package/dist/src/domain/interactions.js +2 -0
  174. package/dist/src/domain/interactions.js.map +1 -0
  175. package/dist/src/domain/providers.js +2 -0
  176. package/dist/src/domain/providers.js.map +1 -0
  177. package/dist/src/domain/resources.js +2 -0
  178. package/dist/src/domain/resources.js.map +1 -0
  179. package/dist/src/domain/workers.js +2 -0
  180. package/dist/src/domain/workers.js.map +1 -0
  181. package/dist/src/evals/collect-only-rerun.js +182 -0
  182. package/dist/src/evals/collect-only-rerun.js.map +1 -0
  183. package/dist/src/evals/directory.js +30 -0
  184. package/dist/src/evals/directory.js.map +1 -0
  185. package/dist/src/evals/docker-engine-stats.js +89 -0
  186. package/dist/src/evals/docker-engine-stats.js.map +1 -0
  187. package/dist/src/evals/docker-ownership.js +51 -0
  188. package/dist/src/evals/docker-ownership.js.map +1 -0
  189. package/dist/src/evals/docker-reaper.js +144 -0
  190. package/dist/src/evals/docker-reaper.js.map +1 -0
  191. package/dist/src/evals/docker-resource-observer.js +207 -0
  192. package/dist/src/evals/docker-resource-observer.js.map +1 -0
  193. package/dist/src/evals/environment-image-planning.js +164 -0
  194. package/dist/src/evals/environment-image-planning.js.map +1 -0
  195. package/dist/src/evals/eval-artifact-planning.js +56 -0
  196. package/dist/src/evals/eval-artifact-planning.js.map +1 -0
  197. package/dist/src/evals/eval-finalization.js +10 -0
  198. package/dist/src/evals/eval-finalization.js.map +1 -0
  199. package/dist/src/evals/eval-lifecycle-events.js +16 -0
  200. package/dist/src/evals/eval-lifecycle-events.js.map +1 -0
  201. package/dist/src/evals/events.js +53 -5
  202. package/dist/src/evals/events.js.map +1 -1
  203. package/dist/src/evals/execution-evidence.js +116 -0
  204. package/dist/src/evals/execution-evidence.js.map +1 -0
  205. package/dist/src/evals/execution-leases.js +342 -0
  206. package/dist/src/evals/execution-leases.js.map +1 -0
  207. package/dist/src/evals/execution-plan-artifacts.js +72 -0
  208. package/dist/src/evals/execution-plan-artifacts.js.map +1 -0
  209. package/dist/src/evals/execution-plan-images.js +69 -0
  210. package/dist/src/evals/execution-plan-images.js.map +1 -0
  211. package/dist/src/evals/execution-plan-resources.js +174 -0
  212. package/dist/src/evals/execution-plan-resources.js.map +1 -0
  213. package/dist/src/evals/execution-plan.js +415 -0
  214. package/dist/src/evals/execution-plan.js.map +1 -0
  215. package/dist/src/evals/harbor-artifact-builder-image.js +99 -0
  216. package/dist/src/evals/harbor-artifact-builder-image.js.map +1 -0
  217. package/dist/src/evals/harbor-artifact-builder.js +354 -0
  218. package/dist/src/evals/harbor-artifact-builder.js.map +1 -0
  219. package/dist/src/evals/harbor-bridge-error.js +39 -3
  220. package/dist/src/evals/harbor-bridge-error.js.map +1 -1
  221. package/dist/src/evals/harbor-phase-timings.js +60 -0
  222. package/dist/src/evals/harbor-phase-timings.js.map +1 -0
  223. package/dist/src/evals/index.js +26 -3
  224. package/dist/src/evals/index.js.map +1 -1
  225. package/dist/src/evals/infrastructure-retry.js +218 -0
  226. package/dist/src/evals/infrastructure-retry.js.map +1 -0
  227. package/dist/src/evals/interaction-capture-import.js +49 -0
  228. package/dist/src/evals/interaction-capture-import.js.map +1 -0
  229. package/dist/src/evals/local-docker-provider.js +372 -0
  230. package/dist/src/evals/local-docker-provider.js.map +1 -0
  231. package/dist/src/evals/local-eval-planning.js +80 -0
  232. package/dist/src/evals/local-eval-planning.js.map +1 -0
  233. package/dist/src/evals/model-capture-plan.js +22 -0
  234. package/dist/src/evals/model-capture-plan.js.map +1 -0
  235. package/dist/src/evals/model-capture-runtime.js +114 -0
  236. package/dist/src/evals/model-capture-runtime.js.map +1 -0
  237. package/dist/src/evals/model-proxy-runtime-state.js +53 -0
  238. package/dist/src/evals/model-proxy-runtime-state.js.map +1 -0
  239. package/dist/src/evals/planned-execution-support.js +54 -0
  240. package/dist/src/evals/planned-execution-support.js.map +1 -0
  241. package/dist/src/evals/planned-execution.js +461 -0
  242. package/dist/src/evals/planned-execution.js.map +1 -0
  243. package/dist/src/evals/planned-remote-execution.js +40 -0
  244. package/dist/src/evals/planned-remote-execution.js.map +1 -0
  245. package/dist/src/evals/planned-retry-lifecycle.js +186 -0
  246. package/dist/src/evals/planned-retry-lifecycle.js.map +1 -0
  247. package/dist/src/evals/prepared-harness.js +76 -0
  248. package/dist/src/evals/prepared-harness.js.map +1 -0
  249. package/dist/src/evals/progress.js +4 -4
  250. package/dist/src/evals/progress.js.map +1 -1
  251. package/dist/src/evals/records.js +20 -3
  252. package/dist/src/evals/records.js.map +1 -1
  253. package/dist/src/evals/recovery.js +192 -0
  254. package/dist/src/evals/recovery.js.map +1 -0
  255. package/dist/src/evals/remote-infrastructure-retry.js +99 -0
  256. package/dist/src/evals/remote-infrastructure-retry.js.map +1 -0
  257. package/dist/src/evals/request.js +14 -1
  258. package/dist/src/evals/request.js.map +1 -1
  259. package/dist/src/evals/rerun-inputs.js +89 -0
  260. package/dist/src/evals/rerun-inputs.js.map +1 -0
  261. package/dist/src/evals/rerun-slots.js +12 -3
  262. package/dist/src/evals/rerun-slots.js.map +1 -1
  263. package/dist/src/evals/rerun-types.js +49 -0
  264. package/dist/src/evals/rerun-types.js.map +1 -0
  265. package/dist/src/evals/rerun.js +113 -75
  266. package/dist/src/evals/rerun.js.map +1 -1
  267. package/dist/src/evals/result-helpers.js +65 -0
  268. package/dist/src/evals/result-helpers.js.map +1 -0
  269. package/dist/src/evals/resume-state.js +36 -0
  270. package/dist/src/evals/resume-state.js.map +1 -0
  271. package/dist/src/evals/service-types.js +2 -0
  272. package/dist/src/evals/service-types.js.map +1 -0
  273. package/dist/src/evals/service.js +327 -237
  274. package/dist/src/evals/service.js.map +1 -1
  275. package/dist/src/evals/task-resources.js +209 -0
  276. package/dist/src/evals/task-resources.js.map +1 -0
  277. package/dist/src/evals/trial-environment-evidence.js +87 -0
  278. package/dist/src/evals/trial-environment-evidence.js.map +1 -0
  279. package/dist/src/evals/trial-execution-evidence.js +13 -0
  280. package/dist/src/evals/trial-execution-evidence.js.map +1 -0
  281. package/dist/src/evals/trial-import-identity.js +31 -0
  282. package/dist/src/evals/trial-import-identity.js.map +1 -0
  283. package/dist/src/evals/trial-import.js +166 -168
  284. package/dist/src/evals/trial-import.js.map +1 -1
  285. package/dist/src/evals/trial-publication-recovery.js +104 -0
  286. package/dist/src/evals/trial-publication-recovery.js.map +1 -0
  287. package/dist/src/evals/trial-publication.js +25 -0
  288. package/dist/src/evals/trial-publication.js.map +1 -0
  289. package/dist/src/evals/verifier-diagnostics.js +235 -0
  290. package/dist/src/evals/verifier-diagnostics.js.map +1 -0
  291. package/dist/src/evals/work-item-artifacts.js +23 -0
  292. package/dist/src/evals/work-item-artifacts.js.map +1 -0
  293. package/dist/src/foundation/config.js +17 -0
  294. package/dist/src/foundation/config.js.map +1 -1
  295. package/dist/src/foundation/credential-redaction.js +137 -0
  296. package/dist/src/foundation/credential-redaction.js.map +1 -0
  297. package/dist/src/foundation/environment-image-references.js +59 -0
  298. package/dist/src/foundation/environment-image-references.js.map +1 -0
  299. package/dist/src/foundation/executable.js +45 -5
  300. package/dist/src/foundation/executable.js.map +1 -1
  301. package/dist/src/foundation/index.js +5 -2
  302. package/dist/src/foundation/index.js.map +1 -1
  303. package/dist/src/foundation/process-identity.js +69 -0
  304. package/dist/src/foundation/process-identity.js.map +1 -0
  305. package/dist/src/images/context.js +47 -0
  306. package/dist/src/images/context.js.map +1 -0
  307. package/dist/src/images/docker-buildkit.js +137 -0
  308. package/dist/src/images/docker-buildkit.js.map +1 -0
  309. package/dist/src/images/dockerfile.js +67 -0
  310. package/dist/src/images/dockerfile.js.map +1 -0
  311. package/dist/src/images/gc.js +261 -0
  312. package/dist/src/images/gc.js.map +1 -0
  313. package/dist/src/images/index.js +10 -0
  314. package/dist/src/images/index.js.map +1 -0
  315. package/dist/src/images/manifest.js +109 -0
  316. package/dist/src/images/manifest.js.map +1 -0
  317. package/dist/src/images/ownership.js +5 -0
  318. package/dist/src/images/ownership.js.map +1 -0
  319. package/dist/src/images/records.js +32 -0
  320. package/dist/src/images/records.js.map +1 -0
  321. package/dist/src/images/registry.js +154 -0
  322. package/dist/src/images/registry.js.map +1 -0
  323. package/dist/src/images/service.js +224 -0
  324. package/dist/src/images/service.js.map +1 -0
  325. package/dist/src/model-access/capture.js +255 -0
  326. package/dist/src/model-access/capture.js.map +1 -0
  327. package/dist/src/model-access/index.js +5 -0
  328. package/dist/src/model-access/index.js.map +1 -0
  329. package/dist/src/model-access/policy.js +49 -0
  330. package/dist/src/model-access/policy.js.map +1 -0
  331. package/dist/src/model-access/proxy.js +336 -0
  332. package/dist/src/model-access/proxy.js.map +1 -0
  333. package/dist/src/model-access/records.js +167 -0
  334. package/dist/src/model-access/records.js.map +1 -0
  335. package/dist/src/runs/bundle.js +450 -0
  336. package/dist/src/runs/bundle.js.map +1 -0
  337. package/dist/src/runs/events.js +12 -1
  338. package/dist/src/runs/events.js.map +1 -1
  339. package/dist/src/runs/executor.js +24 -17
  340. package/dist/src/runs/executor.js.map +1 -1
  341. package/dist/src/runs/index.js +3 -1
  342. package/dist/src/runs/index.js.map +1 -1
  343. package/dist/src/runs/manifest.js +5 -5
  344. package/dist/src/runs/manifest.js.map +1 -1
  345. package/dist/src/runs/request.js +6 -1
  346. package/dist/src/runs/request.js.map +1 -1
  347. package/dist/src/runs/training-candidate.js +161 -0
  348. package/dist/src/runs/training-candidate.js.map +1 -0
  349. package/dist/src/trajectories/provider-capture.js +15 -24
  350. package/dist/src/trajectories/provider-capture.js.map +1 -1
  351. package/dist/src/trajectories/providers/deepseek.js +3 -3
  352. package/dist/src/trajectories/providers/deepseek.js.map +1 -1
  353. package/dist/src/workers/index.js +3 -0
  354. package/dist/src/workers/index.js.map +1 -0
  355. package/dist/src/workers/remote-harbor-work-spec.js +97 -0
  356. package/dist/src/workers/remote-harbor-work-spec.js.map +1 -0
  357. package/dist/src/workers/remote-harbor-worker.js +220 -0
  358. package/dist/src/workers/remote-harbor-worker.js.map +1 -0
  359. package/docs/schemas/environment-build-index.schema.json +13 -0
  360. package/docs/schemas/environment-build-record.schema.json +25 -0
  361. package/docs/schemas/environment-image-manifest.schema.json +62 -0
  362. package/docs/schemas/eval-control.schema.json +36 -0
  363. package/docs/schemas/eval-execution-plan.schema.json +216 -0
  364. package/docs/schemas/eval-request.schema.json +2 -0
  365. package/docs/schemas/eval-rerun-result.schema.json +53 -0
  366. package/docs/schemas/eval-rerun-state.schema.json +61 -0
  367. package/docs/schemas/eval-rerun-submission.schema.json +58 -0
  368. package/docs/schemas/eval-result.schema.json +34 -1
  369. package/docs/schemas/eval-submission-input.schema.json +14 -0
  370. package/docs/schemas/eval-submission.schema.json +68 -0
  371. package/docs/schemas/eval-trial-publication.schema.json +39 -0
  372. package/docs/schemas/execution-evidence.schema.json +96 -0
  373. package/docs/schemas/execution-lease.schema.json +50 -0
  374. package/docs/schemas/execution-provider-status.schema.json +65 -0
  375. package/docs/schemas/execution-worker.schema.json +53 -0
  376. package/docs/schemas/harbor-process-exit.schema.json +14 -0
  377. package/docs/schemas/interaction-capture-ref.schema.json +45 -0
  378. package/docs/schemas/local-provider-execution.schema.json +33 -0
  379. package/docs/schemas/model-interaction.schema.json +41 -0
  380. package/docs/schemas/remote-result-envelope.schema.json +32 -0
  381. package/docs/schemas/remote-tree-envelope.schema.json +25 -0
  382. package/docs/schemas/remote-work-offer.schema.json +58 -0
  383. package/docs/schemas/remote-work-receipt.schema.json +50 -0
  384. package/docs/schemas/remote-worker-artifact.schema.json +17 -0
  385. package/docs/schemas/remote-worker-event.schema.json +18 -0
  386. package/docs/schemas/remote-worker-heartbeat.schema.json +37 -0
  387. package/docs/schemas/remote-worker-registration.schema.json +52 -0
  388. package/docs/schemas/result-bundle-index.schema.json +133 -0
  389. package/docs/schemas/run-environment-images.schema.json +39 -0
  390. package/docs/schemas/training-data-candidate.schema.json +46 -0
  391. package/integrations/harbor/hitch_harbor_agent.py +300 -547
  392. package/integrations/harbor/hitch_harbor_environment.py +331 -0
  393. package/integrations/harbor/hitch_harbor_task_resources.py +275 -0
  394. package/integrations/harbor/hitch_harbor_verifier.py +337 -0
  395. package/package.json +16 -2
@@ -0,0 +1,192 @@
1
+ import path from "node:path";
2
+ import { readHarborRawResult } from "../backends/index.js";
3
+ import { HitchError, readJSON } from "../foundation/index.js";
4
+ import { DEFAULT_EXECUTION_LEASE_HEARTBEAT_MS, heartbeatExecutionLease, markExecutionLeaseLost, readExecutionLeases, reissueExecutionLease, releaseExecutionLease } from "./execution-leases.js";
5
+ import { LocalDockerExecutionProvider, readLocalDockerProcessRecordByLease, waitForLocalDockerProcessTerminal } from "./local-docker-provider.js";
6
+ import { loadEvalResumeState } from "./resume-state.js";
7
+ import { mergeEvalProgressTrial, writeEvalProgress } from "./progress.js";
8
+ import { assertBackendTrialSet } from "./result-helpers.js";
9
+ import { importEvalTrialRuns, validateEvalTrialReferences } from "./trial-import.js";
10
+ import { startEvalModelCaptureRuntime } from "./model-capture-runtime.js";
11
+ import { readModelProxyRuntimeState } from "./model-proxy-runtime-state.js";
12
+ export async function recoverLocalDockerEvalLeases(input) {
13
+ const emit = input.emit ?? (() => { });
14
+ const recovered = [];
15
+ let failure;
16
+ let captureRuntime;
17
+ try {
18
+ captureRuntime = await restoreModelCaptureForRecovery(input);
19
+ for (const lease of input.leases) {
20
+ if (!activeLease(lease))
21
+ continue;
22
+ try {
23
+ await recoverLease(input, lease, emit, captureRuntime);
24
+ recovered.push(lease.lease_id);
25
+ }
26
+ catch (error) {
27
+ const typed = error instanceof HitchError;
28
+ failure ??= {
29
+ code: typed ? error.code : "execution_state_ambiguous",
30
+ message: error?.message || String(error),
31
+ };
32
+ const current = (await readExecutionLeases(input.evalDirectory)).find((entry) => entry.lease_id === lease.lease_id);
33
+ if (current && activeLease(current))
34
+ await markExecutionLeaseLost({
35
+ evalDirectory: input.evalDirectory,
36
+ leaseId: current.lease_id,
37
+ expectedEpoch: current.epoch,
38
+ });
39
+ emit({ type: "eval.lease.recovery-failed", lease_id: lease.lease_id, code: failure.code });
40
+ emit({ type: "eval.work.lost", work_id: lease.work_id, lease_id: lease.lease_id, code: failure.code });
41
+ }
42
+ }
43
+ }
44
+ catch (error) {
45
+ failure = { code: "execution_state_ambiguous", message: error?.message || String(error) };
46
+ for (const lease of input.leases.filter(activeLease)) {
47
+ const current = (await readExecutionLeases(input.evalDirectory)).find((entry) => entry.lease_id === lease.lease_id);
48
+ if (current && activeLease(current))
49
+ await markExecutionLeaseLost({ evalDirectory: input.evalDirectory, leaseId: current.lease_id, expectedEpoch: current.epoch });
50
+ emit({ type: "eval.lease.recovery-failed", lease_id: lease.lease_id, code: failure.code });
51
+ emit({ type: "eval.work.lost", work_id: lease.work_id, lease_id: lease.lease_id, code: failure.code });
52
+ }
53
+ }
54
+ finally {
55
+ await captureRuntime?.close().catch(() => undefined);
56
+ }
57
+ return failure
58
+ ? { status: "ambiguous", recovered_lease_ids: recovered, ...failure }
59
+ : { status: "resumable", recovered_lease_ids: recovered };
60
+ }
61
+ async function recoverLease(input, lease, emit, captureRuntime) {
62
+ if (Date.parse(lease.expires_at) <= Date.now())
63
+ emit({ type: "lease.expired", work_id: lease.work_id, lease_id: lease.lease_id, lease_epoch: lease.epoch, worker_id: lease.worker_id });
64
+ if (lease.provider !== "local-docker")
65
+ throw ambiguous(`unsupported recovery provider: ${lease.provider}`);
66
+ const provider = new LocalDockerExecutionProvider({
67
+ root: input.root,
68
+ workerId: lease.worker_id,
69
+ status: () => { throw new Error("recovery does not inspect provider capacity"); },
70
+ });
71
+ const record = await readLocalDockerProcessRecordByLease({ root: input.root, evalId: input.evalId, leaseId: lease.lease_id }).catch(() => {
72
+ throw ambiguous(`local provider has no durable process identity for ${lease.lease_id}`);
73
+ });
74
+ if (record.lease_epoch === lease.epoch - 1)
75
+ await provider.adoptLeaseEpoch(lease.lease_id, record.lease_epoch, lease.epoch);
76
+ else if (record.lease_epoch !== lease.epoch)
77
+ throw ambiguous(`local provider epoch does not match ${lease.lease_id}`);
78
+ const probe = await provider.recover(lease);
79
+ emit({ type: "eval.lease.recovery-probed", lease_id: lease.lease_id, lease_epoch: lease.epoch, state: probe.state });
80
+ if (probe.state === "released") {
81
+ await releaseExecutionLease({ evalDirectory: input.evalDirectory, leaseId: lease.lease_id, expectedEpoch: lease.epoch });
82
+ emit({ type: "lease.recovered", work_id: lease.work_id, lease_id: lease.lease_id, lease_epoch: lease.epoch, state: "already-released" });
83
+ return;
84
+ }
85
+ if (probe.state !== "running" && probe.state !== "terminal-uncollected") {
86
+ throw ambiguous(`local provider classified ${lease.lease_id} as ${probe.state}`);
87
+ }
88
+ const reissued = await reissueExecutionLease({ evalDirectory: input.evalDirectory, leaseId: lease.lease_id, expectedEpoch: lease.epoch });
89
+ await provider.adoptLeaseEpoch(lease.lease_id, lease.epoch, reissued.epoch);
90
+ emit({ type: "eval.lease.reissued", lease_id: lease.lease_id, previous_epoch: lease.epoch, lease_epoch: reissued.epoch });
91
+ let heartbeatFailure;
92
+ let heartbeatTail = Promise.resolve();
93
+ const timer = setInterval(() => {
94
+ heartbeatTail = heartbeatTail.then(async () => {
95
+ await heartbeatExecutionLease({ evalDirectory: input.evalDirectory, leaseId: lease.lease_id, expectedEpoch: reissued.epoch });
96
+ }).catch((error) => { heartbeatFailure ??= error; });
97
+ }, DEFAULT_EXECUTION_LEASE_HEARTBEAT_MS);
98
+ timer.unref();
99
+ try {
100
+ if (input.cancelRequested && probe.state === "running")
101
+ await provider.cancel(lease.lease_id, reissued.epoch);
102
+ const terminal = await waitForLocalDockerProcessTerminal({ root: input.root, leaseId: lease.lease_id, epoch: reissued.epoch });
103
+ await heartbeatTail;
104
+ if (heartbeatFailure !== undefined)
105
+ throw heartbeatFailure;
106
+ try {
107
+ await collectRecoveredWork(input, reissued, terminal.backend_directory, emit, captureRuntime);
108
+ }
109
+ catch (error) {
110
+ if (!input.cancelRequested || error.code !== "recovery_collection_missing")
111
+ throw error;
112
+ }
113
+ if (!input.cancelRequested && terminal.process_exit_code !== null && terminal.process_exit_code !== 0) {
114
+ throw new HitchError(`recovered Harbor process exited with code ${terminal.process_exit_code}`, { code: "harbor_failed", exitCode: 13 });
115
+ }
116
+ await provider.release(lease.lease_id, reissued.epoch);
117
+ await releaseExecutionLease({ evalDirectory: input.evalDirectory, leaseId: lease.lease_id, expectedEpoch: reissued.epoch });
118
+ emit({ type: "lease.released", work_id: lease.work_id, lease_id: lease.lease_id, lease_epoch: reissued.epoch, worker_id: lease.worker_id });
119
+ emit({ type: "eval.lease.recovered", lease_id: lease.lease_id, lease_epoch: reissued.epoch, state: "released" });
120
+ emit({ type: "lease.recovered", lease_id: lease.lease_id, lease_epoch: reissued.epoch, state: "released" });
121
+ }
122
+ finally {
123
+ clearInterval(timer);
124
+ await heartbeatTail;
125
+ }
126
+ }
127
+ async function collectRecoveredWork(input, lease, backendDirectory, emit, captureRuntime) {
128
+ const state = await loadEvalResumeState(input.evalDirectory);
129
+ const work = state.executionPlan.work_items.find((item) => item.work_id === lease.work_id);
130
+ if (!work || work.logical_attempt === null || work.task_ids.length !== 1)
131
+ throw ambiguous("recovered lease does not match one planned task slot");
132
+ const jobDirectory = path.join(input.evalDirectory, backendDirectory, "job");
133
+ const rawResult = await readHarborRawResult(jobDirectory);
134
+ if (!rawResult)
135
+ throw new HitchError("recovered Harbor work item has no aggregate result", { code: "recovery_collection_missing", exitCode: 12 });
136
+ const resolvedRevision = await readJSON(path.join(input.evalDirectory, "resolution.json"));
137
+ const request = await readJSON(path.join(input.evalDirectory, "request.json"));
138
+ const runtime = state.plan.controller_runtime;
139
+ const runtimeId = typeof runtime?.runtime_id === "string" ? runtime.runtime_id : undefined;
140
+ const refs = await importEvalTrialRuns({
141
+ root: input.root,
142
+ evalId: input.evalId,
143
+ evalDirectory: input.evalDirectory,
144
+ harborJobDirectory: jobDirectory,
145
+ expectedAttempt: work.logical_attempt,
146
+ request,
147
+ resolvedRevision,
148
+ benchmarkId: state.progress.benchmark_id,
149
+ benchmarkRevision: state.progress.benchmark_revision,
150
+ ...(runtimeId ? { runtimeId } : {}),
151
+ env: input.env ?? process.env,
152
+ ...(state.executionPlan.model_capture ? { modelCapturePlan: captureRuntime?.plan ?? state.executionPlan.model_capture } : {}),
153
+ ...(captureRuntime?.exporter ? { interactionCaptureExporter: captureRuntime.exporter } : {}),
154
+ rawResult,
155
+ }, state.progress.trials);
156
+ const workRefs = refs.filter((ref) => ref.task_id === work.task_ids[0] && ref.attempt === work.logical_attempt);
157
+ assertBackendTrialSet(rawResult, workRefs);
158
+ await validateEvalTrialReferences(input.root, input.evalId, workRefs, {
159
+ benchmarkId: state.progress.benchmark_id,
160
+ benchmarkRevision: state.progress.benchmark_revision,
161
+ });
162
+ let progress = state.progress;
163
+ for (const ref of workRefs)
164
+ progress = mergeEvalProgressTrial(progress, ref);
165
+ if (progress.generation !== state.progress.generation)
166
+ await writeEvalProgress(input.evalDirectory, progress);
167
+ emit({ type: "eval.work-item.recovered", work_id: work.work_id, lease_id: lease.lease_id, trials: workRefs.length });
168
+ }
169
+ async function restoreModelCaptureForRecovery(input) {
170
+ if (!input.leases.some(activeLease))
171
+ return undefined;
172
+ const state = await loadEvalResumeState(input.evalDirectory);
173
+ const plan = state.executionPlan.model_capture;
174
+ if (!plan || plan.effective_mode !== "proxy" && plan.effective_mode !== "hybrid")
175
+ return undefined;
176
+ if (!await readModelProxyRuntimeState(input.evalDirectory, input.evalId, plan)) {
177
+ throw ambiguous("recoverable model proxy has no persisted endpoint identity");
178
+ }
179
+ const runtime = await startEvalModelCaptureRuntime({ plan, evalId: input.evalId, evalDirectory: input.evalDirectory, env: input.env ?? process.env });
180
+ if (!runtime.route || !runtime.exporter) {
181
+ await runtime.close().catch(() => undefined);
182
+ throw ambiguous("recoverable model proxy endpoint could not be restored");
183
+ }
184
+ return runtime;
185
+ }
186
+ function activeLease(lease) {
187
+ return lease.state === "offered" || lease.state === "accepted" || lease.state === "running" || lease.state === "releasing";
188
+ }
189
+ function ambiguous(message) {
190
+ return new HitchError(message, { code: "execution_state_ambiguous", exitCode: 12 });
191
+ }
192
+ //# sourceMappingURL=recovery.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"recovery.js","sourceRoot":"","sources":["../../../src/evals/recovery.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,mBAAmB,EAAE,MAAM,sBAAsB,CAAC;AAE3D,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,wBAAwB,CAAC;AAC9D,OAAO,EAAE,oCAAoC,EAAE,uBAAuB,EAAE,sBAAsB,EAAE,mBAAmB,EAAE,qBAAqB,EAAE,qBAAqB,EAAE,MAAM,uBAAuB,CAAC;AACjM,OAAO,EAAE,4BAA4B,EAAE,mCAAmC,EAAE,iCAAiC,EAAE,MAAM,4BAA4B,CAAC;AAClJ,OAAO,EAAE,mBAAmB,EAAE,MAAM,mBAAmB,CAAC;AACxD,OAAO,EAAE,sBAAsB,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAC1E,OAAO,EAAE,qBAAqB,EAAE,MAAM,qBAAqB,CAAC;AAC5D,OAAO,EAAE,mBAAmB,EAAE,2BAA2B,EAAE,MAAM,mBAAmB,CAAC;AACrF,OAAO,EAAE,4BAA4B,EAAE,MAAM,4BAA4B,CAAC;AAE1E,OAAO,EAAE,0BAA0B,EAAE,MAAM,gCAAgC,CAAC;AAS5E,MAAM,CAAC,KAAK,UAAU,4BAA4B,CAAC,KAQlD;IACC,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,IAAI,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;IACtC,MAAM,SAAS,GAAa,EAAE,CAAC;IAC/B,IAAI,OAAsD,CAAC;IAC3D,IAAI,cAAmD,CAAC;IACxD,IAAI,CAAC;QACH,cAAc,GAAG,MAAM,8BAA8B,CAAC,KAAK,CAAC,CAAC;QAC7D,KAAK,MAAM,KAAK,IAAI,KAAK,CAAC,MAAM,EAAE,CAAC;YACjC,IAAI,CAAC,WAAW,CAAC,KAAK,CAAC;gBAAE,SAAS;YAClC,IAAI,CAAC;gBACH,MAAM,YAAY,CAAC,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,cAAc,CAAC,CAAC;gBACvD,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;YACjC,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,MAAM,KAAK,GAAG,KAAK,YAAY,UAAU,CAAC;gBAC1C,OAAO,KAAK;oBACV,IAAI,EAAE,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,2BAA2B;oBACtD,OAAO,EAAG,KAAe,EAAE,OAAO,IAAI,MAAM,CAAC,KAAK,CAAC;iBACpD,CAAC;gBACF,MAAM,OAAO,GAAG,CAAC,MAAM,mBAAmB,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,KAAK,CAAC,QAAQ,CAAC,CAAC;gBACpH,IAAI,OAAO,IAAI,WAAW,CAAC,OAAO,CAAC;oBAAE,MAAM,sBAAsB,CAAC;wBAChE,aAAa,EAAE,KAAK,CAAC,aAAa;wBAClC,OAAO,EAAE,OAAO,CAAC,QAAQ;wBACzB,aAAa,EAAE,OAAO,CAAC,KAAK;qBAC7B,CAAC,CAAC;gBACH,IAAI,CAAC,EAAE,IAAI,EAAE,4BAA4B,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;gBAC3F,IAAI,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;YACzG,CAAC;QACH,CAAC;IACH,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,OAAO,GAAG,EAAE,IAAI,EAAE,2BAA2B,EAAE,OAAO,EAAG,KAAe,EAAE,OAAO,IAAI,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC;QACrG,KAAK,MAAM,KAAK,IAAI,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,WAAW,CAAC,EAAE,CAAC;YACrD,MAAM,OAAO,GAAG,CAAC,MAAM,mBAAmB,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,KAAK,CAAC,QAAQ,CAAC,CAAC;YACpH,IAAI,OAAO,IAAI,WAAW,CAAC,OAAO,CAAC;gBAAE,MAAM,sBAAsB,CAAC,EAAE,aAAa,EAAE,KAAK,CAAC,aAAa,EAAE,OAAO,EAAE,OAAO,CAAC,QAAQ,EAAE,aAAa,EAAE,OAAO,CAAC,KAAK,EAAE,CAAC,CAAC;YACnK,IAAI,CAAC,EAAE,IAAI,EAAE,4BAA4B,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;YAC3F,IAAI,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;QACzG,CAAC;IACH,CAAC;YAAS,CAAC;QACT,MAAM,cAAc,EAAE,KAAK,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAC;IACvD,CAAC;IACD,OAAO,OAAO;QACZ,CAAC,CAAC,EAAE,MAAM,EAAE,WAAW,EAAE,mBAAmB,EAAE,SAAS,EAAE,GAAG,OAAO,EAAE;QACrE,CAAC,CAAC,EAAE,MAAM,EAAE,WAAW,EAAE,mBAAmB,EAAE,SAAS,EAAE,CAAC;AAC9D,CAAC;AAED,KAAK,UAAU,YAAY,CACzB,KAAkH,EAClH,KAAuB,EACvB,IAA8C,EAC9C,cAAwC;IAExC,IAAI,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,UAAU,CAAC,IAAI,IAAI,CAAC,GAAG,EAAE;QAAE,IAAI,CAAC,EAAE,IAAI,EAAE,eAAe,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,WAAW,EAAE,KAAK,CAAC,KAAK,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,CAAC,CAAC;IACxL,IAAI,KAAK,CAAC,QAAQ,KAAK,cAAc;QAAE,MAAM,SAAS,CAAC,kCAAkC,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC3G,MAAM,QAAQ,GAAG,IAAI,4BAA4B,CAAC;QAChD,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,QAAQ,EAAE,KAAK,CAAC,SAAS;QACzB,MAAM,EAAE,GAAG,EAAE,GAAG,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC,CAAC,CAAC;KAClF,CAAC,CAAC;IACH,MAAM,MAAM,GAAG,MAAM,mCAAmC,CAAC,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE;QACvI,MAAM,SAAS,CAAC,sDAAsD,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC1F,CAAC,CAAC,CAAC;IACH,IAAI,MAAM,CAAC,WAAW,KAAK,KAAK,CAAC,KAAK,GAAG,CAAC;QAAE,MAAM,QAAQ,CAAC,eAAe,CAAC,KAAK,CAAC,QAAQ,EAAE,MAAM,CAAC,WAAW,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC;SACvH,IAAI,MAAM,CAAC,WAAW,KAAK,KAAK,CAAC,KAAK;QAAE,MAAM,SAAS,CAAC,uCAAuC,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;IACtH,MAAM,KAAK,GAAG,MAAM,QAAQ,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;IAC5C,IAAI,CAAC,EAAE,IAAI,EAAE,4BAA4B,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,WAAW,EAAE,KAAK,CAAC,KAAK,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC;IACrH,IAAI,KAAK,CAAC,KAAK,KAAK,UAAU,EAAE,CAAC;QAC/B,MAAM,qBAAqB,CAAC,EAAE,aAAa,EAAE,KAAK,CAAC,aAAa,EAAE,OAAO,EAAE,KAAK,CAAC,QAAQ,EAAE,aAAa,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC;QACzH,IAAI,CAAC,EAAE,IAAI,EAAE,iBAAiB,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,WAAW,EAAE,KAAK,CAAC,KAAK,EAAE,KAAK,EAAE,kBAAkB,EAAE,CAAC,CAAC;QACzI,OAAO;IACT,CAAC;IACD,IAAI,KAAK,CAAC,KAAK,KAAK,SAAS,IAAI,KAAK,CAAC,KAAK,KAAK,sBAAsB,EAAE,CAAC;QACxE,MAAM,SAAS,CAAC,6BAA6B,KAAK,CAAC,QAAQ,OAAO,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC;IACnF,CAAC;IACD,MAAM,QAAQ,GAAG,MAAM,qBAAqB,CAAC,EAAE,aAAa,EAAE,KAAK,CAAC,aAAa,EAAE,OAAO,EAAE,KAAK,CAAC,QAAQ,EAAE,aAAa,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC;IAC1I,MAAM,QAAQ,CAAC,eAAe,CAAC,KAAK,CAAC,QAAQ,EAAE,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,KAAK,CAAC,CAAC;IAC5E,IAAI,CAAC,EAAE,IAAI,EAAE,qBAAqB,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,cAAc,EAAE,KAAK,CAAC,KAAK,EAAE,WAAW,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC;IAC1H,IAAI,gBAAyB,CAAC;IAC9B,IAAI,aAAa,GAAG,OAAO,CAAC,OAAO,EAAE,CAAC;IACtC,MAAM,KAAK,GAAG,WAAW,CAAC,GAAG,EAAE;QAC7B,aAAa,GAAG,aAAa,CAAC,IAAI,CAAC,KAAK,IAAI,EAAE;YAC5C,MAAM,uBAAuB,CAAC,EAAE,aAAa,EAAE,KAAK,CAAC,aAAa,EAAE,OAAO,EAAE,KAAK,CAAC,QAAQ,EAAE,aAAa,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC;QAChI,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE,GAAG,gBAAgB,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IACvD,CAAC,EAAE,oCAAoC,CAAC,CAAC;IACzC,KAAK,CAAC,KAAK,EAAE,CAAC;IACd,IAAI,CAAC;QACH,IAAI,KAAK,CAAC,eAAe,IAAI,KAAK,CAAC,KAAK,KAAK,SAAS;YAAE,MAAM,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,QAAQ,EAAE,QAAQ,CAAC,KAAK,CAAC,CAAC;QAC9G,MAAM,QAAQ,GAAG,MAAM,iCAAiC,CAAC,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,KAAK,CAAC,QAAQ,EAAE,KAAK,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC;QAC/H,MAAM,aAAa,CAAC;QACpB,IAAI,gBAAgB,KAAK,SAAS;YAAE,MAAM,gBAAgB,CAAC;QAC3D,IAAI,CAAC;YAAC,MAAM,oBAAoB,CAAC,KAAK,EAAE,QAAQ,EAAE,QAAQ,CAAC,iBAAiB,EAAE,IAAI,EAAE,cAAc,CAAC,CAAC;QAAC,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACpH,IAAI,CAAC,KAAK,CAAC,eAAe,IAAK,KAA2B,CAAC,IAAI,KAAK,6BAA6B;gBAAE,MAAM,KAAK,CAAC;QACjH,CAAC;QACD,IAAI,CAAC,KAAK,CAAC,eAAe,IAAI,QAAQ,CAAC,iBAAiB,KAAK,IAAI,IAAI,QAAQ,CAAC,iBAAiB,KAAK,CAAC,EAAE,CAAC;YACtG,MAAM,IAAI,UAAU,CAAC,6CAA6C,QAAQ,CAAC,iBAAiB,EAAE,EAAE,EAAE,IAAI,EAAE,eAAe,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC;QAC3I,CAAC;QACD,MAAM,QAAQ,CAAC,OAAO,CAAC,KAAK,CAAC,QAAQ,EAAE,QAAQ,CAAC,KAAK,CAAC,CAAC;QACvD,MAAM,qBAAqB,CAAC,EAAE,aAAa,EAAE,KAAK,CAAC,aAAa,EAAE,OAAO,EAAE,KAAK,CAAC,QAAQ,EAAE,aAAa,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC;QAC5H,IAAI,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,WAAW,EAAE,QAAQ,CAAC,KAAK,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,CAAC,CAAC;QAC5I,IAAI,CAAC,EAAE,IAAI,EAAE,sBAAsB,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,WAAW,EAAE,QAAQ,CAAC,KAAK,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;QACjH,IAAI,CAAC,EAAE,IAAI,EAAE,iBAAiB,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,WAAW,EAAE,QAAQ,CAAC,KAAK,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;IAC9G,CAAC;YAAS,CAAC;QACT,aAAa,CAAC,KAAK,CAAC,CAAC;QACrB,MAAM,aAAa,CAAC;IACtB,CAAC;AACH,CAAC;AAED,KAAK,UAAU,oBAAoB,CACjC,KAAuF,EACvF,KAAuB,EACvB,gBAAwB,EACxB,IAA8C,EAC9C,cAAwC;IAExC,MAAM,KAAK,GAAG,MAAM,mBAAmB,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;IAC7D,MAAM,IAAI,GAAG,KAAK,CAAC,aAAa,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,KAAK,KAAK,CAAC,OAAO,CAAC,CAAC;IAC3F,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,eAAe,KAAK,IAAI,IAAI,IAAI,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC;QAAE,MAAM,SAAS,CAAC,sDAAsD,CAAC,CAAC;IAClJ,MAAM,YAAY,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,gBAAgB,EAAE,KAAK,CAAC,CAAC;IAC7E,MAAM,SAAS,GAAG,MAAM,mBAAmB,CAAC,YAAY,CAAC,CAAC;IAC1D,IAAI,CAAC,SAAS;QAAE,MAAM,IAAI,UAAU,CAAC,oDAAoD,EAAE,EAAE,IAAI,EAAE,6BAA6B,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC;IAClJ,MAAM,gBAAgB,GAAG,MAAM,QAAQ,CAAmB,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,iBAAiB,CAAC,CAAC,CAAC;IAC7G,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAc,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,cAAc,CAAC,CAAC,CAAC;IAC5F,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,kBAAyD,CAAC;IACrF,MAAM,SAAS,GAAG,OAAO,OAAO,EAAE,UAAU,KAAK,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC;IAC3F,MAAM,IAAI,GAAG,MAAM,mBAAmB,CAAC;QACrC,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,MAAM,EAAE,KAAK,CAAC,MAAM;QACpB,aAAa,EAAE,KAAK,CAAC,aAAa;QAClC,kBAAkB,EAAE,YAAY;QAChC,eAAe,EAAE,IAAI,CAAC,eAAe;QACrC,OAAO;QACP,gBAAgB;QAChB,WAAW,EAAE,KAAK,CAAC,QAAQ,CAAC,YAAY;QACxC,iBAAiB,EAAE,KAAK,CAAC,QAAQ,CAAC,kBAAkB;QACpD,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACnC,GAAG,EAAE,KAAK,CAAC,GAAG,IAAI,OAAO,CAAC,GAAG;QAC7B,GAAG,CAAC,KAAK,CAAC,aAAa,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,cAAc,EAAE,IAAI,IAAI,KAAK,CAAC,aAAa,CAAC,aAAa,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7H,GAAG,CAAC,cAAc,EAAE,QAAQ,CAAC,CAAC,CAAC,EAAE,0BAA0B,EAAE,cAAc,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC5F,SAAS;KACV,EAAE,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;IAC1B,MAAM,QAAQ,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,OAAO,KAAK,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,GAAG,CAAC,OAAO,KAAK,IAAI,CAAC,eAAe,CAAC,CAAC;IAChH,qBAAqB,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC;IAC3C,MAAM,2BAA2B,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,MAAM,EAAE,QAAQ,EAAE;QACpE,WAAW,EAAE,KAAK,CAAC,QAAQ,CAAC,YAAY;QACxC,iBAAiB,EAAE,KAAK,CAAC,QAAQ,CAAC,kBAAkB;KACrD,CAAC,CAAC;IACH,IAAI,QAAQ,GAAG,KAAK,CAAC,QAAQ,CAAC;IAC9B,KAAK,MAAM,GAAG,IAAI,QAAQ;QAAE,QAAQ,GAAG,sBAAsB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;IAC7E,IAAI,QAAQ,CAAC,UAAU,KAAK,KAAK,CAAC,QAAQ,CAAC,UAAU;QAAE,MAAM,iBAAiB,CAAC,KAAK,CAAC,aAAa,EAAE,QAAQ,CAAC,CAAC;IAC9G,IAAI,CAAC,EAAE,IAAI,EAAE,0BAA0B,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC;AACvH,CAAC;AAED,KAAK,UAAU,8BAA8B,CAAC,KAK7C;IACC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC;QAAE,OAAO,SAAS,CAAC;IACtD,MAAM,KAAK,GAAG,MAAM,mBAAmB,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;IAC7D,MAAM,IAAI,GAAG,KAAK,CAAC,aAAa,CAAC,aAAa,CAAC;IAC/C,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,cAAc,KAAK,OAAO,IAAI,IAAI,CAAC,cAAc,KAAK,QAAQ;QAAE,OAAO,SAAS,CAAC;IACnG,IAAI,CAAC,MAAM,0BAA0B,CAAC,KAAK,CAAC,aAAa,EAAE,KAAK,CAAC,MAAM,EAAE,IAAI,CAAC,EAAE,CAAC;QAC/E,MAAM,SAAS,CAAC,4DAA4D,CAAC,CAAC;IAChF,CAAC;IACD,MAAM,OAAO,GAAG,MAAM,4BAA4B,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,aAAa,EAAE,KAAK,CAAC,aAAa,EAAE,GAAG,EAAE,KAAK,CAAC,GAAG,IAAI,OAAO,CAAC,GAAG,EAAE,CAAC,CAAC;IACtJ,IAAI,CAAC,OAAO,CAAC,KAAK,IAAI,CAAC,OAAO,CAAC,QAAQ,EAAE,CAAC;QACxC,MAAM,OAAO,CAAC,KAAK,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAC;QAC7C,MAAM,SAAS,CAAC,wDAAwD,CAAC,CAAC;IAC5E,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,SAAS,WAAW,CAAC,KAAuB;IAC1C,OAAO,KAAK,CAAC,KAAK,KAAK,SAAS,IAAI,KAAK,CAAC,KAAK,KAAK,UAAU,IAAI,KAAK,CAAC,KAAK,KAAK,SAAS,IAAI,KAAK,CAAC,KAAK,KAAK,WAAW,CAAC;AAC7H,CAAC;AAED,SAAS,SAAS,CAAC,OAAe;IAChC,OAAO,IAAI,UAAU,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,2BAA2B,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC;AACtF,CAAC"}
@@ -0,0 +1,99 @@
1
+ import { setTimeout as delay } from "node:timers/promises";
2
+ import { preparedArtifactForWorkItem } from "./work-item-artifacts.js";
3
+ import { HitchError, sha256JSON } from "../foundation/index.js";
4
+ import { retryableInfrastructureTrials } from "./infrastructure-retry.js";
5
+ import { replaceInvalidEvalProgressTrial, writeEvalProgress } from "./progress.js";
6
+ export async function runRemoteInfrastructureRetries(input) {
7
+ const executor = input.options.remoteWorkExecutor;
8
+ if (!executor)
9
+ throw new TypeError("remote work executor is unavailable");
10
+ let progress = input.progress;
11
+ let candidates = retryableInfrastructureTrials(input.initial.refs);
12
+ const runs = [];
13
+ for (let retry = 1; retry <= input.options.request.infrastructure_retries && candidates.length > 0; retry += 1) {
14
+ if (input.options.signal?.aborted)
15
+ break;
16
+ const triggers = [...candidates];
17
+ const backoffMs = input.options.request.infrastructure_retry_backoff_ms * retry;
18
+ input.options.sink.emit({
19
+ type: "eval.infrastructure-retry.scheduled", execution_kind: "physical-infrastructure-retry",
20
+ candidate_executes: true, attempt: input.item.logical_attempt, retry,
21
+ tasks: input.item.task_ids, backoff_ms: backoffMs, provider: input.item.provider,
22
+ });
23
+ if (backoffMs > 0)
24
+ await delay(backoffMs, undefined, input.options.signal ? { signal: input.options.signal } : undefined);
25
+ if (input.options.signal?.aborted)
26
+ break;
27
+ const work = retryWorkItem(input.item, retry);
28
+ const refs = [];
29
+ const publish = async (ref) => {
30
+ assertSelectedTrial(ref, input.item);
31
+ const existing = refs.find((entry) => entry.trial_id === ref.trial_id || entry.run_id === ref.run_id);
32
+ if (existing) {
33
+ if (JSON.stringify(existing) !== JSON.stringify(ref))
34
+ throw new Error(`remote retry trial identity changed: ${ref.trial_id}`);
35
+ return;
36
+ }
37
+ refs.push(ref);
38
+ if (ref.observation_status === "valid") {
39
+ progress = replaceInvalidEvalProgressTrial(progress, ref);
40
+ await writeEvalProgress(input.options.evalDirectory, progress);
41
+ }
42
+ input.options.sink.emit({
43
+ type: ref.observation_status === "valid" ? "eval.infrastructure-retry.repaired" : "eval.infrastructure-retry.failed",
44
+ attempt: input.item.logical_attempt, retry, task_id: ref.task_id, trial_id: ref.trial_id,
45
+ run_id: ref.run_id, observation_status: ref.observation_status,
46
+ ...(ref.invalid_reason ? { invalid_reason: ref.invalid_reason } : {}), generation: progress.generation,
47
+ });
48
+ };
49
+ const completed = await executor({
50
+ evalId: input.options.evalId, evalDirectory: input.options.evalDirectory, root: input.options.root,
51
+ request: input.options.request, plan: input.options.plan, workItem: work,
52
+ resolvedRevision: input.options.resolvedRevision, preparedArtifact: preparedArtifactForWorkItem(input.options, work),
53
+ runtimeDirectory: input.options.controllerRuntime.directory, runtimeId: input.options.controllerRuntime.runtime_id,
54
+ ...(input.initial.environmentImages ? { environmentImages: input.initial.environmentImages } : {}),
55
+ ...(input.options.plan.model_capture ? { modelCapturePlan: input.options.plan.model_capture } : {}),
56
+ publicationMode: "replace-invalid",
57
+ ...(input.options.signal ? { signal: input.options.signal } : {}),
58
+ emit: (event) => input.options.sink.emit({ ...event, execution_kind: "physical-infrastructure-retry", infrastructure_retry: retry }),
59
+ publish,
60
+ onLeaseState: (leaseId, state) => input.options.onWorkItemState?.(work.work_id, leaseId, state) ?? Promise.resolve(),
61
+ });
62
+ assertPublishedRefs(completed.refs, refs, input.item);
63
+ candidates = retryableInfrastructureTrials(refs);
64
+ runs.push({
65
+ attempt: input.item.logical_attempt, retry, tasks: [...input.item.task_ids], triggers,
66
+ refs, run: completed.run, leaseId: completed.leaseId, workId: work.work_id,
67
+ });
68
+ input.options.sink.emit({
69
+ type: "eval.infrastructure-retry.completed", execution_kind: "physical-infrastructure-retry",
70
+ candidate_executes: true, attempt: input.item.logical_attempt, retry, tasks: input.item.task_ids,
71
+ repaired_tasks: refs.filter((ref) => ref.observation_status === "valid").map((ref) => ref.task_id).sort(),
72
+ remaining_tasks: candidates.map((ref) => ref.task_id).sort(), provider: input.item.provider,
73
+ });
74
+ if (completed.run.backend.process_exit_code !== 0 || completed.run.rawResult === null)
75
+ break;
76
+ }
77
+ return { progress, runs };
78
+ }
79
+ function assertSelectedTrial(ref, item) {
80
+ if (ref.attempt !== item.logical_attempt || !item.task_ids.includes(ref.task_id)) {
81
+ throw new HitchError(`remote infrastructure retry returned an unselected trial: ${ref.task_id}#${ref.attempt}`, {
82
+ code: "eval_infrastructure_retry_trial_mismatch", exitCode: 12,
83
+ });
84
+ }
85
+ }
86
+ function assertPublishedRefs(returned, published, item) {
87
+ for (const ref of returned)
88
+ assertSelectedTrial(ref, item);
89
+ if (returned.length !== published.length || returned.some((ref) => !published.some((entry) => JSON.stringify(entry) === JSON.stringify(ref)))) {
90
+ throw new HitchError("remote infrastructure retry returned refs that were not durably published", {
91
+ code: "eval_infrastructure_retry_result_mismatch", exitCode: 12,
92
+ });
93
+ }
94
+ }
95
+ function retryWorkItem(item, retry) {
96
+ const identity = sha256JSON({ work_id: item.work_id, execution_kind: "physical-infrastructure-retry", retry });
97
+ return { ...item, work_id: `work_${identity.slice("sha256:".length, "sha256:".length + 32)}` };
98
+ }
99
+ //# sourceMappingURL=remote-infrastructure-retry.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"remote-infrastructure-retry.js","sourceRoot":"","sources":["../../../src/evals/remote-infrastructure-retry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,IAAI,KAAK,EAAE,MAAM,sBAAsB,CAAC;AAE3D,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAEhE,OAAO,EAAE,6BAA6B,EAAE,MAAM,2BAA2B,CAAC;AAE1E,OAAO,EAAE,+BAA+B,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAEnF,MAAM,CAAC,KAAK,UAAU,8BAA8B,CAAC,KAKpD;IACC,MAAM,QAAQ,GAAG,KAAK,CAAC,OAAO,CAAC,kBAAkB,CAAC;IAClD,IAAI,CAAC,QAAQ;QAAE,MAAM,IAAI,SAAS,CAAC,qCAAqC,CAAC,CAAC;IAC1E,IAAI,QAAQ,GAAG,KAAK,CAAC,QAAQ,CAAC;IAC9B,IAAI,UAAU,GAAG,6BAA6B,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IACnE,MAAM,IAAI,GAA6B,EAAE,CAAC;IAC1C,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,IAAI,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,sBAAsB,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QAC/G,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,OAAO;YAAE,MAAM;QACzC,MAAM,QAAQ,GAAG,CAAC,GAAG,UAAU,CAAC,CAAC;QACjC,MAAM,SAAS,GAAG,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,+BAA+B,GAAG,KAAK,CAAC;QAChF,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC;YACtB,IAAI,EAAE,qCAAqC,EAAE,cAAc,EAAE,+BAA+B;YAC5F,kBAAkB,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,eAAe,EAAE,KAAK;YACpE,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,QAAQ,EAAE,UAAU,EAAE,SAAS,EAAE,QAAQ,EAAE,KAAK,CAAC,IAAI,CAAC,QAAQ;SACjF,CAAC,CAAC;QACH,IAAI,SAAS,GAAG,CAAC;YAAE,MAAM,KAAK,CAAC,SAAS,EAAE,SAAS,EAAE,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;QAC1H,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,OAAO;YAAE,MAAM;QACzC,MAAM,IAAI,GAAG,aAAa,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;QAC9C,MAAM,IAAI,GAAqB,EAAE,CAAC;QAClC,MAAM,OAAO,GAAG,KAAK,EAAE,GAAmB,EAAiB,EAAE;YAC3D,mBAAmB,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;YACrC,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,GAAG,CAAC,QAAQ,IAAI,KAAK,CAAC,MAAM,KAAK,GAAG,CAAC,MAAM,CAAC,CAAC;YACtG,IAAI,QAAQ,EAAE,CAAC;gBACb,IAAI,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,KAAK,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC;oBAAE,MAAM,IAAI,KAAK,CAAC,wCAAwC,GAAG,CAAC,QAAQ,EAAE,CAAC,CAAC;gBAC9H,OAAO;YACT,CAAC;YACD,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YACf,IAAI,GAAG,CAAC,kBAAkB,KAAK,OAAO,EAAE,CAAC;gBACvC,QAAQ,GAAG,+BAA+B,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;gBAC1D,MAAM,iBAAiB,CAAC,KAAK,CAAC,OAAO,CAAC,aAAa,EAAE,QAAQ,CAAC,CAAC;YACjE,CAAC;YACD,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC;gBACtB,IAAI,EAAE,GAAG,CAAC,kBAAkB,KAAK,OAAO,CAAC,CAAC,CAAC,oCAAoC,CAAC,CAAC,CAAC,kCAAkC;gBACpH,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,eAAe,EAAE,KAAK,EAAE,OAAO,EAAE,GAAG,CAAC,OAAO,EAAE,QAAQ,EAAE,GAAG,CAAC,QAAQ;gBACxF,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,kBAAkB,EAAE,GAAG,CAAC,kBAAkB;gBAC9D,GAAG,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,cAAc,EAAE,GAAG,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,UAAU,EAAE,QAAQ,CAAC,UAAU;aACvG,CAAC,CAAC;QACL,CAAC,CAAC;QACF,MAAM,SAAS,GAAG,MAAM,QAAQ,CAAC;YAC/B,MAAM,EAAE,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,aAAa,EAAE,KAAK,CAAC,OAAO,CAAC,aAAa,EAAE,IAAI,EAAE,KAAK,CAAC,OAAO,CAAC,IAAI;YAClG,OAAO,EAAE,KAAK,CAAC,OAAO,CAAC,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,OAAO,CAAC,IAAI,EAAE,QAAQ,EAAE,IAAI;YACxE,gBAAgB,EAAE,KAAK,CAAC,OAAO,CAAC,gBAAgB,EAAE,gBAAgB,EAAE,2BAA2B,CAAC,KAAK,CAAC,OAAO,EAAE,IAAI,CAAC;YACpH,gBAAgB,EAAE,KAAK,CAAC,OAAO,CAAC,iBAAiB,CAAC,SAAS,EAAE,SAAS,EAAE,KAAK,CAAC,OAAO,CAAC,iBAAiB,CAAC,UAAU;YAClH,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,iBAAiB,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,KAAK,CAAC,OAAO,CAAC,iBAAiB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAClG,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,aAAa,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACnG,eAAe,EAAE,iBAAiB;YAClC,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACjE,IAAI,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,GAAG,KAAK,EAAE,cAAc,EAAE,+BAA+B,EAAE,oBAAoB,EAAE,KAAK,EAAE,CAAC;YACpI,OAAO;YACP,YAAY,EAAE,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,OAAO,CAAC,eAAe,EAAE,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,KAAK,CAAC,IAAI,OAAO,CAAC,OAAO,EAAE;SACrH,CAAC,CAAC;QACH,mBAAmB,CAAC,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;QACtD,UAAU,GAAG,6BAA6B,CAAC,IAAI,CAAC,CAAC;QACjD,IAAI,CAAC,IAAI,CAAC;YACR,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,eAAyB,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,GAAG,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,QAAQ;YAC/F,IAAI,EAAE,GAAG,EAAE,SAAS,CAAC,GAAG,EAAE,OAAO,EAAE,SAAS,CAAC,OAAO,EAAE,MAAM,EAAE,IAAI,CAAC,OAAO;SAC3E,CAAC,CAAC;QACH,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC;YACtB,IAAI,EAAE,qCAAqC,EAAE,cAAc,EAAE,+BAA+B;YAC5F,kBAAkB,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,eAAe,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,QAAQ;YAChG,cAAc,EAAE,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,kBAAkB,KAAK,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE;YACzG,eAAe,EAAE,UAAU,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE,QAAQ,EAAE,KAAK,CAAC,IAAI,CAAC,QAAQ;SAC5F,CAAC,CAAC;QACH,IAAI,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,iBAAiB,KAAK,CAAC,IAAI,SAAS,CAAC,GAAG,CAAC,SAAS,KAAK,IAAI;YAAE,MAAM;IAC/F,CAAC;IACD,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;AAC5B,CAAC;AAED,SAAS,mBAAmB,CAAC,GAAmB,EAAE,IAAuB;IACvE,IAAI,GAAG,CAAC,OAAO,KAAK,IAAI,CAAC,eAAe,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC;QACjF,MAAM,IAAI,UAAU,CAAC,6DAA6D,GAAG,CAAC,OAAO,IAAI,GAAG,CAAC,OAAO,EAAE,EAAE;YAC9G,IAAI,EAAE,0CAA0C,EAAE,QAAQ,EAAE,EAAE;SAC/D,CAAC,CAAC;IACL,CAAC;AACH,CAAC;AAED,SAAS,mBAAmB,CAAC,QAAmC,EAAE,SAAoC,EAAE,IAAuB;IAC7H,KAAK,MAAM,GAAG,IAAI,QAAQ;QAAE,mBAAmB,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;IAC3D,IAAI,QAAQ,CAAC,MAAM,KAAK,SAAS,CAAC,MAAM,IAAI,QAAQ,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,KAAK,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;QAC9I,MAAM,IAAI,UAAU,CAAC,2EAA2E,EAAE;YAChG,IAAI,EAAE,2CAA2C,EAAE,QAAQ,EAAE,EAAE;SAChE,CAAC,CAAC;IACL,CAAC;AACH,CAAC;AAED,SAAS,aAAa,CAAC,IAAuB,EAAE,KAAa;IAC3D,MAAM,QAAQ,GAAG,UAAU,CAAC,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,cAAc,EAAE,+BAA+B,EAAE,KAAK,EAAE,CAAC,CAAC;IAC/G,OAAO,EAAE,GAAG,IAAI,EAAE,OAAO,EAAE,QAAQ,QAAQ,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,EAAE,SAAS,CAAC,MAAM,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;AACjG,CAAC"}
@@ -1,5 +1,7 @@
1
1
  export const DEFAULT_EVAL_TIMEOUT_MS = 15 * 60 * 1_000;
2
2
  export const DEFAULT_EVAL_SETUP_TIMEOUT_MS = 30 * 60 * 1_000;
3
+ export const DEFAULT_INFRASTRUCTURE_RETRIES = 1;
4
+ export const DEFAULT_INFRASTRUCTURE_RETRY_BACKOFF_MS = 1_000;
3
5
  const HARBOR_EVAL_BYPASS_ARG = {
4
6
  codex: "--dangerously-bypass-approvals-and-sandbox",
5
7
  opencode: "--dangerously-skip-permissions",
@@ -17,7 +19,8 @@ export async function validateEvalRequest(input) {
17
19
  throw invalidInput("eval request must be a JSON object");
18
20
  const allowed = new Set([
19
21
  "schema_version", "backend", "dataset", "harness_ref", "model", "attempts",
20
- "max_concurrent", "timeout_ms", "setup_timeout_ms", "agent_args", "pass_env",
22
+ "max_concurrent", "infrastructure_retries", "infrastructure_retry_backoff_ms",
23
+ "timeout_ms", "setup_timeout_ms", "agent_args", "pass_env",
21
24
  ]);
22
25
  const unexpected = Object.keys(input).find((field) => !allowed.has(field));
23
26
  if (unexpected)
@@ -38,6 +41,8 @@ export async function validateEvalRequest(input) {
38
41
  const benchmark = await resolveBenchmarkReference(input.dataset.trim());
39
42
  const attempts = positiveInteger(input.attempts ?? 1, "attempts");
40
43
  const maxConcurrent = positiveInteger(input.max_concurrent ?? 4, "max_concurrent");
44
+ const infrastructureRetries = nonNegativeInteger(input.infrastructure_retries ?? DEFAULT_INFRASTRUCTURE_RETRIES, "infrastructure_retries");
45
+ const infrastructureRetryBackoff = nonNegativeNumber(input.infrastructure_retry_backoff_ms ?? DEFAULT_INFRASTRUCTURE_RETRY_BACKOFF_MS, "infrastructure_retry_backoff_ms");
41
46
  const timeout = nonNegativeNumber(input.timeout_ms ?? DEFAULT_EVAL_TIMEOUT_MS, "timeout_ms");
42
47
  const setupTimeout = nonNegativeNumber(input.setup_timeout_ms ?? DEFAULT_EVAL_SETUP_TIMEOUT_MS, "setup_timeout_ms");
43
48
  if (input.model !== undefined && typeof input.model !== "string")
@@ -60,6 +65,8 @@ export async function validateEvalRequest(input) {
60
65
  model: typeof input.model === "string" ? input.model : "",
61
66
  attempts,
62
67
  max_concurrent: maxConcurrent,
68
+ infrastructure_retries: infrastructureRetries,
69
+ infrastructure_retry_backoff_ms: infrastructureRetryBackoff,
63
70
  timeout_ms: timeout,
64
71
  setup_timeout_ms: setupTimeout,
65
72
  agent_args: agentArgs,
@@ -130,6 +137,12 @@ function positiveInteger(value, name) {
130
137
  throw invalidInput(`${name} must be a positive integer`);
131
138
  return parsed;
132
139
  }
140
+ function nonNegativeInteger(value, name) {
141
+ const parsed = Number(value);
142
+ if (!Number.isSafeInteger(parsed) || parsed < 0)
143
+ throw invalidInput(`${name} must be a non-negative integer`);
144
+ return parsed;
145
+ }
133
146
  function nonNegativeNumber(value, name) {
134
147
  const parsed = Number(value);
135
148
  if (!Number.isFinite(parsed) || parsed < 0)
@@ -1 +1 @@
1
- {"version":3,"file":"request.js","sourceRoot":"","sources":["../../../src/evals/request.ts"],"names":[],"mappings":"AAAA,MAAM,CAAC,MAAM,uBAAuB,GAAG,EAAE,GAAG,EAAE,GAAG,KAAK,CAAC;AACvD,MAAM,CAAC,MAAM,6BAA6B,GAAG,EAAE,GAAG,EAAE,GAAG,KAAK,CAAC;AAE7D,MAAM,sBAAsB,GAAqC;IAC/D,KAAK,EAAE,4CAA4C;IACnD,QAAQ,EAAE,gCAAgC;CAC3C,CAAC;AAEF,MAAM,UAAU,SAAS;IACvB,OAAO,QAAQ,UAAU,EAAE,CAAC,UAAU,CAAC,GAAG,EAAE,EAAE,CAAC,EAAY,CAAC;AAC9D,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,KAAa;IAC1C,IAAI,CAAC,qBAAqB,CAAC,IAAI,CAAC,KAAK,CAAC;QAAE,MAAM,YAAY,CAAC,oBAAoB,KAAK,EAAE,CAAC,CAAC;IACxF,OAAO,KAAe,CAAC;AACzB,CAAC;AAgBD,MAAM,CAAC,KAAK,UAAU,mBAAmB,CAAC,KAAuB;IAC/D,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,MAAM,YAAY,CAAC,oCAAoC,CAAC,CAAC;IAC1H,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC;QACtB,gBAAgB,EAAE,SAAS,EAAE,SAAS,EAAE,aAAa,EAAE,OAAO,EAAE,UAAU;QAC1E,gBAAgB,EAAE,YAAY,EAAE,kBAAkB,EAAE,YAAY,EAAE,UAAU;KAC7E,CAAC,CAAC;IACH,MAAM,UAAU,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC;IAC3E,IAAI,UAAU;QAAE,MAAM,YAAY,CAAC,+BAA+B,UAAU,EAAE,CAAC,CAAC;IAChF,IAAI,KAAK,CAAC,cAAc,KAAK,SAAS,IAAI,KAAK,CAAC,cAAc,KAAK,cAAc;QAAE,MAAM,YAAY,CAAC,+BAA+B,KAAK,CAAC,cAAc,EAAE,CAAC,CAAC;IAC7J,IAAI,CAAC,KAAK,CAAC,OAAO,IAAI,QAAQ,CAAC,KAAK,QAAQ;QAAE,MAAM,YAAY,CAAC,qDAAqD,CAAC,CAAC;IACxH,IAAI,OAAO,KAAK,CAAC,OAAO,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,EAAE;QAAE,MAAM,YAAY,CAAC,oCAAoC,CAAC,CAAC;IACzH,IAAI,OAAO,KAAK,CAAC,WAAW,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,WAAW,CAAC,IAAI,EAAE;QAAE,MAAM,YAAY,CAAC,wCAAwC,CAAC,CAAC;IACrI,MAAM,SAAS,GAAG,qBAAqB,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;IAC3D,IAAI,SAAS,CAAC,QAAQ,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;QAC5C,MAAM,YAAY,CAAC,yEAAyE,CAAC,CAAC;IAChG,CAAC;IACD,gCAAgC,CAAC,SAAS,CAAC,CAAC;IAC5C,MAAM,SAAS,GAAG,MAAM,yBAAyB,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;IACxE,MAAM,QAAQ,GAAG,eAAe,CAAC,KAAK,CAAC,QAAQ,IAAI,CAAC,EAAE,UAAU,CAAC,CAAC;IAClE,MAAM,aAAa,GAAG,eAAe,CAAC,KAAK,CAAC,cAAc,IAAI,CAAC,EAAE,gBAAgB,CAAC,CAAC;IACnF,MAAM,OAAO,GAAG,iBAAiB,CAAC,KAAK,CAAC,UAAU,IAAI,uBAAuB,EAAE,YAAY,CAAC,CAAC;IAC7F,MAAM,YAAY,GAAG,iBAAiB,CAAC,KAAK,CAAC,gBAAgB,IAAI,6BAA6B,EAAE,kBAAkB,CAAC,CAAC;IACpH,IAAI,KAAK,CAAC,KAAK,KAAK,SAAS,IAAI,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ;QAAE,MAAM,YAAY,CAAC,wBAAwB,CAAC,CAAC;IAC/G,IAAI,KAAK,CAAC,UAAU,KAAK,SAAS,IAAI,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,UAAU,CAAC,IAAI,KAAK,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,EAAE,CAAC;QACxI,MAAM,YAAY,CAAC,wCAAwC,CAAC,CAAC;IAC/D,CAAC;IACD,IAAI,KAAK,CAAC,QAAQ,KAAK,SAAS,IAAI,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,EAAE,CAAC;QAClI,MAAM,YAAY,CAAC,sCAAsC,CAAC,CAAC;IAC7D,CAAC;IACD,MAAM,SAAS,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,UAAU,CAAa,CAAC,CAAC,CAAC,EAAE,CAAC;IAC3F,MAAM,SAAS,GAAG,sBAAsB,CAAC,SAAS,CAAC,UAAU,CAAC,CAAC;IAC/D,IAAI,SAAS,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC;QAAE,SAAS,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;IAC9E,OAAO;QACL,cAAc,EAAE,cAAc;QAC9B,OAAO,EAAE,QAAQ;QACjB,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE;QACrC,WAAW,EAAE,SAAS,CAAC,SAAS;QAChC,KAAK,EAAE,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE;QACzD,QAAQ;QACR,cAAc,EAAE,aAAa;QAC7B,UAAU,EAAE,OAAO;QACnB,gBAAgB,EAAE,YAAY;QAC9B,UAAU,EAAE,SAAS;QACrB,QAAQ,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,QAAoB,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QACvF,YAAY,EAAE,SAAS,CAAC,YAAY;QACpC,kBAAkB,EAAE,SAAS,CAAC,kBAAkB;KACjD,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,OAAe;IAC7D,MAAM,GAAG,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC;IAC3B,MAAM,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IAChC,IAAI,CAAC;QACH,IAAI,CAAC,MAAM,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,WAAW,EAAE,EAAE,CAAC;YACtC,OAAO;gBACL,YAAY,EAAE,SAAS,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE;gBAC7C,kBAAkB,EAAE,MAAM,eAAe,CAAC,KAAK,EAAE,EAAE,gBAAgB,EAAE,IAAI,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC;aAC1F,CAAC;QACJ,CAAC;IACH,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,MAAM,KAAK,CAAC;IACtE,CAAC;IACD,MAAM,SAAS,GAAG,GAAG,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC;IACvC,IAAI,SAAS,IAAI,CAAC,IAAI,SAAS,KAAK,GAAG,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACnD,MAAM,YAAY,CAAC,wEAAwE,CAAC,CAAC;IAC/F,CAAC;IACD,MAAM,WAAW,GAAG,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC;IAC5C,MAAM,QAAQ,GAAG,GAAG,CAAC,KAAK,CAAC,SAAS,GAAG,CAAC,CAAC,CAAC;IAC1C,IAAI,QAAQ,CAAC,WAAW,EAAE,KAAK,QAAQ;QAAE,MAAM,YAAY,CAAC,mCAAmC,CAAC,CAAC;IACjG,OAAO,EAAE,YAAY,EAAE,WAAW,EAAE,kBAAkB,EAAE,QAAQ,EAAE,CAAC;AACrE,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,0BAA0B,CAAC,OAAe;IAC9D,MAAM,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;IAC3C,IAAI,CAAC;QACH,IAAI,CAAC,CAAC,MAAM,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,WAAW,EAAE;YAAE,OAAO,IAAI,CAAC;IACtD,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,IAAI,CAAC;QACpE,MAAM,KAAK,CAAC;IACd,CAAC;IACD,IAAI,MAAM,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,EAAE,WAAW,CAAC,CAAC;QAAE,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC;IACtF,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,KAAK,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;IAC9D,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,KAAK,MAAM,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;QACvF,IAAI,KAAK,CAAC,WAAW,EAAE,IAAI,MAAM,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,EAAE,KAAK,CAAC,IAAI,EAAE,WAAW,CAAC,CAAC;YAAE,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IACpH,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,KAAK,UAAU,aAAa,CAAC,IAAY;IACvC,IAAI,CAAC;QACH,OAAO,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;IACrC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,KAAK,CAAC;QACrE,MAAM,KAAK,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,eAAe,CAAC,KAAc,EAAE,IAAY;IACnD,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC;IAC7B,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,MAAM,IAAI,CAAC;QAAE,MAAM,YAAY,CAAC,GAAG,IAAI,6BAA6B,CAAC,CAAC;IACvG,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAc,EAAE,IAAY;IACrD,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC;IAC7B,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,MAAM,GAAG,CAAC;QAAE,MAAM,YAAY,CAAC,GAAG,IAAI,gCAAgC,CAAC,CAAC;IACxG,OAAO,MAAM,CAAC;AAChB,CAAC;AACD,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,kBAAkB,CAAC;AACjD,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AACtE,OAAO,EAAE,gCAAgC,EAAE,qBAAqB,EAAE,MAAM,uBAAuB,CAAC;AAChG,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC"}
1
+ {"version":3,"file":"request.js","sourceRoot":"","sources":["../../../src/evals/request.ts"],"names":[],"mappings":"AAAA,MAAM,CAAC,MAAM,uBAAuB,GAAG,EAAE,GAAG,EAAE,GAAG,KAAK,CAAC;AACvD,MAAM,CAAC,MAAM,6BAA6B,GAAG,EAAE,GAAG,EAAE,GAAG,KAAK,CAAC;AAC7D,MAAM,CAAC,MAAM,8BAA8B,GAAG,CAAC,CAAC;AAChD,MAAM,CAAC,MAAM,uCAAuC,GAAG,KAAK,CAAC;AAE7D,MAAM,sBAAsB,GAAqC;IAC/D,KAAK,EAAE,4CAA4C;IACnD,QAAQ,EAAE,gCAAgC;CAC3C,CAAC;AAEF,MAAM,UAAU,SAAS;IACvB,OAAO,QAAQ,UAAU,EAAE,CAAC,UAAU,CAAC,GAAG,EAAE,EAAE,CAAC,EAAY,CAAC;AAC9D,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,KAAa;IAC1C,IAAI,CAAC,qBAAqB,CAAC,IAAI,CAAC,KAAK,CAAC;QAAE,MAAM,YAAY,CAAC,oBAAoB,KAAK,EAAE,CAAC,CAAC;IACxF,OAAO,KAAe,CAAC;AACzB,CAAC;AAkBD,MAAM,CAAC,KAAK,UAAU,mBAAmB,CAAC,KAAuB;IAC/D,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,MAAM,YAAY,CAAC,oCAAoC,CAAC,CAAC;IAC1H,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC;QACtB,gBAAgB,EAAE,SAAS,EAAE,SAAS,EAAE,aAAa,EAAE,OAAO,EAAE,UAAU;QAC1E,gBAAgB,EAAE,wBAAwB,EAAE,iCAAiC;QAC7E,YAAY,EAAE,kBAAkB,EAAE,YAAY,EAAE,UAAU;KAC3D,CAAC,CAAC;IACH,MAAM,UAAU,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC;IAC3E,IAAI,UAAU;QAAE,MAAM,YAAY,CAAC,+BAA+B,UAAU,EAAE,CAAC,CAAC;IAChF,IAAI,KAAK,CAAC,cAAc,KAAK,SAAS,IAAI,KAAK,CAAC,cAAc,KAAK,cAAc;QAAE,MAAM,YAAY,CAAC,+BAA+B,KAAK,CAAC,cAAc,EAAE,CAAC,CAAC;IAC7J,IAAI,CAAC,KAAK,CAAC,OAAO,IAAI,QAAQ,CAAC,KAAK,QAAQ;QAAE,MAAM,YAAY,CAAC,qDAAqD,CAAC,CAAC;IACxH,IAAI,OAAO,KAAK,CAAC,OAAO,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,EAAE;QAAE,MAAM,YAAY,CAAC,oCAAoC,CAAC,CAAC;IACzH,IAAI,OAAO,KAAK,CAAC,WAAW,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,WAAW,CAAC,IAAI,EAAE;QAAE,MAAM,YAAY,CAAC,wCAAwC,CAAC,CAAC;IACrI,MAAM,SAAS,GAAG,qBAAqB,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;IAC3D,IAAI,SAAS,CAAC,QAAQ,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;QAC5C,MAAM,YAAY,CAAC,yEAAyE,CAAC,CAAC;IAChG,CAAC;IACD,gCAAgC,CAAC,SAAS,CAAC,CAAC;IAC5C,MAAM,SAAS,GAAG,MAAM,yBAAyB,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;IACxE,MAAM,QAAQ,GAAG,eAAe,CAAC,KAAK,CAAC,QAAQ,IAAI,CAAC,EAAE,UAAU,CAAC,CAAC;IAClE,MAAM,aAAa,GAAG,eAAe,CAAC,KAAK,CAAC,cAAc,IAAI,CAAC,EAAE,gBAAgB,CAAC,CAAC;IACnF,MAAM,qBAAqB,GAAG,kBAAkB,CAC9C,KAAK,CAAC,sBAAsB,IAAI,8BAA8B,EAC9D,wBAAwB,CACzB,CAAC;IACF,MAAM,0BAA0B,GAAG,iBAAiB,CAClD,KAAK,CAAC,+BAA+B,IAAI,uCAAuC,EAChF,iCAAiC,CAClC,CAAC;IACF,MAAM,OAAO,GAAG,iBAAiB,CAAC,KAAK,CAAC,UAAU,IAAI,uBAAuB,EAAE,YAAY,CAAC,CAAC;IAC7F,MAAM,YAAY,GAAG,iBAAiB,CAAC,KAAK,CAAC,gBAAgB,IAAI,6BAA6B,EAAE,kBAAkB,CAAC,CAAC;IACpH,IAAI,KAAK,CAAC,KAAK,KAAK,SAAS,IAAI,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ;QAAE,MAAM,YAAY,CAAC,wBAAwB,CAAC,CAAC;IAC/G,IAAI,KAAK,CAAC,UAAU,KAAK,SAAS,IAAI,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,UAAU,CAAC,IAAI,KAAK,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,EAAE,CAAC;QACxI,MAAM,YAAY,CAAC,wCAAwC,CAAC,CAAC;IAC/D,CAAC;IACD,IAAI,KAAK,CAAC,QAAQ,KAAK,SAAS,IAAI,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,EAAE,CAAC;QAClI,MAAM,YAAY,CAAC,sCAAsC,CAAC,CAAC;IAC7D,CAAC;IACD,MAAM,SAAS,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,UAAU,CAAa,CAAC,CAAC,CAAC,EAAE,CAAC;IAC3F,MAAM,SAAS,GAAG,sBAAsB,CAAC,SAAS,CAAC,UAAU,CAAC,CAAC;IAC/D,IAAI,SAAS,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC;QAAE,SAAS,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;IAC9E,OAAO;QACL,cAAc,EAAE,cAAc;QAC9B,OAAO,EAAE,QAAQ;QACjB,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE;QACrC,WAAW,EAAE,SAAS,CAAC,SAAS;QAChC,KAAK,EAAE,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE;QACzD,QAAQ;QACR,cAAc,EAAE,aAAa;QAC7B,sBAAsB,EAAE,qBAAqB;QAC7C,+BAA+B,EAAE,0BAA0B;QAC3D,UAAU,EAAE,OAAO;QACnB,gBAAgB,EAAE,YAAY;QAC9B,UAAU,EAAE,SAAS;QACrB,QAAQ,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,QAAoB,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QACvF,YAAY,EAAE,SAAS,CAAC,YAAY;QACpC,kBAAkB,EAAE,SAAS,CAAC,kBAAkB;KACjD,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,OAAe;IAC7D,MAAM,GAAG,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC;IAC3B,MAAM,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IAChC,IAAI,CAAC;QACH,IAAI,CAAC,MAAM,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,WAAW,EAAE,EAAE,CAAC;YACtC,OAAO;gBACL,YAAY,EAAE,SAAS,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE;gBAC7C,kBAAkB,EAAE,MAAM,eAAe,CAAC,KAAK,EAAE,EAAE,gBAAgB,EAAE,IAAI,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC;aAC1F,CAAC;QACJ,CAAC;IACH,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,MAAM,KAAK,CAAC;IACtE,CAAC;IACD,MAAM,SAAS,GAAG,GAAG,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC;IACvC,IAAI,SAAS,IAAI,CAAC,IAAI,SAAS,KAAK,GAAG,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACnD,MAAM,YAAY,CAAC,wEAAwE,CAAC,CAAC;IAC/F,CAAC;IACD,MAAM,WAAW,GAAG,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC;IAC5C,MAAM,QAAQ,GAAG,GAAG,CAAC,KAAK,CAAC,SAAS,GAAG,CAAC,CAAC,CAAC;IAC1C,IAAI,QAAQ,CAAC,WAAW,EAAE,KAAK,QAAQ;QAAE,MAAM,YAAY,CAAC,mCAAmC,CAAC,CAAC;IACjG,OAAO,EAAE,YAAY,EAAE,WAAW,EAAE,kBAAkB,EAAE,QAAQ,EAAE,CAAC;AACrE,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,0BAA0B,CAAC,OAAe;IAC9D,MAAM,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;IAC3C,IAAI,CAAC;QACH,IAAI,CAAC,CAAC,MAAM,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,WAAW,EAAE;YAAE,OAAO,IAAI,CAAC;IACtD,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,IAAI,CAAC;QACpE,MAAM,KAAK,CAAC;IACd,CAAC;IACD,IAAI,MAAM,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,EAAE,WAAW,CAAC,CAAC;QAAE,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC;IACtF,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,KAAK,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;IAC9D,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,KAAK,MAAM,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC;QACvF,IAAI,KAAK,CAAC,WAAW,EAAE,IAAI,MAAM,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,EAAE,KAAK,CAAC,IAAI,EAAE,WAAW,CAAC,CAAC;YAAE,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IACpH,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,KAAK,UAAU,aAAa,CAAC,IAAY;IACvC,IAAI,CAAC;QACH,OAAO,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;IACrC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,KAAK,CAAC;QACrE,MAAM,KAAK,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,eAAe,CAAC,KAAc,EAAE,IAAY;IACnD,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC;IAC7B,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,MAAM,IAAI,CAAC;QAAE,MAAM,YAAY,CAAC,GAAG,IAAI,6BAA6B,CAAC,CAAC;IACvG,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,SAAS,kBAAkB,CAAC,KAAc,EAAE,IAAY;IACtD,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC;IAC7B,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,MAAM,GAAG,CAAC;QAAE,MAAM,YAAY,CAAC,GAAG,IAAI,iCAAiC,CAAC,CAAC;IAC9G,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAc,EAAE,IAAY;IACrD,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC;IAC7B,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,MAAM,GAAG,CAAC;QAAE,MAAM,YAAY,CAAC,GAAG,IAAI,gCAAgC,CAAC,CAAC;IACxG,OAAO,MAAM,CAAC;AAChB,CAAC;AACD,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,kBAAkB,CAAC;AACjD,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AACtE,OAAO,EAAE,gCAAgC,EAAE,qBAAqB,EAAE,MAAM,uBAAuB,CAAC;AAChG,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC"}
@@ -0,0 +1,89 @@
1
+ import path from "node:path";
2
+ import { loadPreparedArtifact, preparedArtifactDirectory } from "../artifacts/index.js";
3
+ import { validateLocalGitTransportManifest, verifyLocalGitTransport } from "../backends/index.js";
4
+ import { HitchError, readJSON } from "../foundation/index.js";
5
+ import { loadHarborArtifact } from "./harbor-artifact-builder.js";
6
+ export async function loadRerunResolvedRevision(evalDirectory, plan) {
7
+ const resolution = await readJSON(path.join(evalDirectory, "resolution.json"));
8
+ if (!resolution || resolution.identity !== requiredString(plan.candidate.revision_identity, "candidate revision identity")
9
+ || resolution.harness_id !== requiredString(plan.candidate.harness_id, "candidate harness id")) {
10
+ throw unavailable("eval resolution identity changed");
11
+ }
12
+ return resolution;
13
+ }
14
+ export async function loadRerunPreparedArtifacts(root, plan) {
15
+ const artifacts = new Map();
16
+ for (const summary of plan.preparedArtifacts) {
17
+ const artifact = await loadRerunPreparedArtifactSummary(root, summary);
18
+ artifacts.set(artifact.artifact_id, artifact);
19
+ }
20
+ if (artifacts.size === 0)
21
+ throw unavailable("eval has no prepared artifact assignments");
22
+ return artifacts;
23
+ }
24
+ async function loadRerunPreparedArtifactSummary(root, summary) {
25
+ const artifactId = requiredString(summary.artifact_id, "prepared artifact id");
26
+ if (summary.storage === "harbor-artifact-cache-v2") {
27
+ return loadHarborArtifact(root, {
28
+ artifact_id: artifactId,
29
+ artifact_integrity: requiredString(summary.artifact_integrity, "prepared artifact integrity"),
30
+ entrypoint_integrity: requiredString(summary.entrypoint_integrity, "prepared artifact entrypoint integrity"),
31
+ harness_id: requiredString(summary.harness_id, "prepared artifact harness id"),
32
+ revision_identity: requiredString(summary.revision_identity, "prepared artifact revision identity"),
33
+ platform: requiredString(summary.platform, "prepared artifact platform"),
34
+ });
35
+ }
36
+ const artifact = await loadPreparedArtifact(preparedArtifactDirectory(root, artifactId), {
37
+ artifact_id: artifactId,
38
+ artifact_integrity: requiredString(summary.artifact_integrity, "prepared artifact integrity"),
39
+ entrypoint_integrity: requiredString(summary.entrypoint_integrity, "prepared artifact entrypoint integrity"),
40
+ harness_id: requiredString(summary.harness_id, "prepared artifact harness id"),
41
+ revision_identity: requiredString(summary.revision_identity, "prepared artifact revision identity"),
42
+ platform: requiredString(summary.platform, "prepared artifact platform"),
43
+ });
44
+ return {
45
+ directory: preparedArtifactDirectory(root, artifact.artifact_id),
46
+ artifact_id: artifact.artifact_id,
47
+ artifact_integrity: requiredString(artifact.artifact_integrity, "artifact integrity"),
48
+ entrypoint_integrity: requiredString(artifact.entrypoint_integrity, "artifact entrypoint integrity"),
49
+ harness_id: artifact.harness_id,
50
+ revision_identity: artifact.revision_identity,
51
+ adapter_version: artifact.adapter_version,
52
+ recipe_version: artifact.recipe_version,
53
+ platform: artifact.platform,
54
+ node_version: artifact.toolchain.node || process.version,
55
+ source_type: artifact.source_type,
56
+ };
57
+ }
58
+ export async function loadRerunLocalTransport(evalDirectory, plan, resolution, env, signal) {
59
+ if (plan.localSourceTransport === undefined)
60
+ return undefined;
61
+ const directory = path.join(evalDirectory, "local-source");
62
+ const manifestPath = path.join(directory, "manifest.json");
63
+ const use = {
64
+ directory,
65
+ manifestPath,
66
+ payloadPath: path.join(directory, "payload.pack"),
67
+ resolutionPath: path.join(directory, "resolution.json"),
68
+ manifest: validateLocalGitTransportManifest(await readJSON(manifestPath)),
69
+ };
70
+ await verifyLocalGitTransport(use, {
71
+ expected: {
72
+ harnessId: resolution.harness_id,
73
+ resolutionIdentity: resolution.identity,
74
+ commit: requiredString(resolution.revision.commit, "local source commit"),
75
+ },
76
+ env: env ?? process.env,
77
+ ...(signal ? { signal } : {}),
78
+ });
79
+ return use;
80
+ }
81
+ function requiredString(value, label) {
82
+ if (typeof value !== "string" || value.length === 0)
83
+ throw unavailable(`${label} is missing`);
84
+ return value;
85
+ }
86
+ function unavailable(message) {
87
+ return new HitchError(message, { code: "eval_rerun_unavailable", exitCode: 2 });
88
+ }
89
+ //# sourceMappingURL=rerun-inputs.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"rerun-inputs.js","sourceRoot":"","sources":["../../../src/evals/rerun-inputs.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,EAAE,oBAAoB,EAAE,yBAAyB,EAAE,MAAM,uBAAuB,CAAC;AAExF,OAAO,EAAE,iCAAiC,EAAE,uBAAuB,EAAE,MAAM,sBAAsB,CAAC;AAElG,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,wBAAwB,CAAC;AAC9D,OAAO,EAAE,kBAAkB,EAAE,MAAM,8BAA8B,CAAC;AAQlE,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAAC,aAAqB,EAAE,IAAoB;IACzF,MAAM,UAAU,GAAG,MAAM,QAAQ,CAAmB,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,iBAAiB,CAAC,CAAC,CAAC;IACjG,IAAI,CAAC,UAAU,IAAI,UAAU,CAAC,QAAQ,KAAK,cAAc,CAAC,IAAI,CAAC,SAAS,CAAC,iBAAiB,EAAE,6BAA6B,CAAC;WACrH,UAAU,CAAC,UAAU,KAAK,cAAc,CAAC,IAAI,CAAC,SAAS,CAAC,UAAU,EAAE,sBAAsB,CAAC,EAAE,CAAC;QACjG,MAAM,WAAW,CAAC,kCAAkC,CAAC,CAAC;IACxD,CAAC;IACD,OAAO,UAAU,CAAC;AACpB,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,0BAA0B,CAAC,IAAY,EAAE,IAAoB;IACjF,MAAM,SAAS,GAAG,IAAI,GAAG,EAAqC,CAAC;IAC/D,KAAK,MAAM,OAAO,IAAI,IAAI,CAAC,iBAAiB,EAAE,CAAC;QAC7C,MAAM,QAAQ,GAAG,MAAM,gCAAgC,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;QACvE,SAAS,CAAC,GAAG,CAAC,QAAQ,CAAC,WAAW,EAAE,QAAQ,CAAC,CAAC;IAChD,CAAC;IACD,IAAI,SAAS,CAAC,IAAI,KAAK,CAAC;QAAE,MAAM,WAAW,CAAC,2CAA2C,CAAC,CAAC;IACzF,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,KAAK,UAAU,gCAAgC,CAAC,IAAY,EAAE,OAAgC;IAC5F,MAAM,UAAU,GAAG,cAAc,CAAC,OAAO,CAAC,WAAW,EAAE,sBAAsB,CAAC,CAAC;IAC/E,IAAI,OAAO,CAAC,OAAO,KAAK,0BAA0B,EAAE,CAAC;QACnD,OAAO,kBAAkB,CAAC,IAAI,EAAE;YAC9B,WAAW,EAAE,UAAU;YACvB,kBAAkB,EAAE,cAAc,CAAC,OAAO,CAAC,kBAAkB,EAAE,6BAA6B,CAAC;YAC7F,oBAAoB,EAAE,cAAc,CAAC,OAAO,CAAC,oBAAoB,EAAE,wCAAwC,CAAC;YAC5G,UAAU,EAAE,cAAc,CAAC,OAAO,CAAC,UAAU,EAAE,8BAA8B,CAAC;YAC9E,iBAAiB,EAAE,cAAc,CAAC,OAAO,CAAC,iBAAiB,EAAE,qCAAqC,CAAC;YACnG,QAAQ,EAAE,cAAc,CAAC,OAAO,CAAC,QAAQ,EAAE,4BAA4B,CAAC;SACzE,CAAC,CAAC;IACL,CAAC;IACD,MAAM,QAAQ,GAAG,MAAM,oBAAoB,CAAC,yBAAyB,CAAC,IAAI,EAAE,UAAU,CAAC,EAAE;QACvF,WAAW,EAAE,UAAU;QACvB,kBAAkB,EAAE,cAAc,CAAC,OAAO,CAAC,kBAAkB,EAAE,6BAA6B,CAAC;QAC7F,oBAAoB,EAAE,cAAc,CAAC,OAAO,CAAC,oBAAoB,EAAE,wCAAwC,CAAC;QAC5G,UAAU,EAAE,cAAc,CAAC,OAAO,CAAC,UAAU,EAAE,8BAA8B,CAAC;QAC9E,iBAAiB,EAAE,cAAc,CAAC,OAAO,CAAC,iBAAiB,EAAE,qCAAqC,CAAC;QACnG,QAAQ,EAAE,cAAc,CAAC,OAAO,CAAC,QAAQ,EAAE,4BAA4B,CAAC;KACzE,CAAC,CAAC;IACH,OAAO;QACL,SAAS,EAAE,yBAAyB,CAAC,IAAI,EAAE,QAAQ,CAAC,WAAW,CAAC;QAChE,WAAW,EAAE,QAAQ,CAAC,WAAW;QACjC,kBAAkB,EAAE,cAAc,CAAC,QAAQ,CAAC,kBAAkB,EAAE,oBAAoB,CAAC;QACrF,oBAAoB,EAAE,cAAc,CAAC,QAAQ,CAAC,oBAAoB,EAAE,+BAA+B,CAAC;QACpG,UAAU,EAAE,QAAQ,CAAC,UAAU;QAC/B,iBAAiB,EAAE,QAAQ,CAAC,iBAAiB;QAC7C,eAAe,EAAE,QAAQ,CAAC,eAAe;QACzC,cAAc,EAAE,QAAQ,CAAC,cAAc;QACvC,QAAQ,EAAE,QAAQ,CAAC,QAAQ;QAC3B,YAAY,EAAE,QAAQ,CAAC,SAAS,CAAC,IAAI,IAAI,OAAO,CAAC,OAAO;QACxD,WAAW,EAAE,QAAQ,CAAC,WAAW;KAClC,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,uBAAuB,CAC3C,aAAqB,EACrB,IAAoB,EACpB,UAA4B,EAC5B,GAAkC,EAClC,MAA+B;IAE/B,IAAI,IAAI,CAAC,oBAAoB,KAAK,SAAS;QAAE,OAAO,SAAS,CAAC;IAC9D,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,cAAc,CAAC,CAAC;IAC3D,MAAM,YAAY,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,eAAe,CAAC,CAAC;IAC3D,MAAM,GAAG,GAAyB;QAChC,SAAS;QACT,YAAY;QACZ,WAAW,EAAE,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,cAAc,CAAC;QACjD,cAAc,EAAE,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,iBAAiB,CAAC;QACvD,QAAQ,EAAE,iCAAiC,CAAC,MAAM,QAAQ,CAAC,YAAY,CAAC,CAAC;KAC1E,CAAC;IACF,MAAM,uBAAuB,CAAC,GAAG,EAAE;QACjC,QAAQ,EAAE;YACR,SAAS,EAAE,UAAU,CAAC,UAAU;YAChC,kBAAkB,EAAE,UAAU,CAAC,QAAQ;YACvC,MAAM,EAAE,cAAc,CAAC,UAAU,CAAC,QAAQ,CAAC,MAAM,EAAE,qBAAqB,CAAC;SAC1E;QACD,GAAG,EAAE,GAAG,IAAI,OAAO,CAAC,GAAG;QACvB,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC9B,CAAC,CAAC;IACH,OAAO,GAAG,CAAC;AACb,CAAC;AAED,SAAS,cAAc,CAAC,KAAc,EAAE,KAAa;IACnD,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,MAAM,WAAW,CAAC,GAAG,KAAK,aAAa,CAAC,CAAC;IAC9F,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,WAAW,CAAC,OAAe;IAClC,OAAO,IAAI,UAAU,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,wBAAwB,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;AAClF,CAAC"}
@@ -3,7 +3,7 @@ import { evalTrialKey } from "./progress.js";
3
3
  export function selectRerunTasks(tasks, progress, selector, attempts = 1) {
4
4
  return uniqueTasks(selectRerunTrialSlots(tasks, attempts, progress, selector));
5
5
  }
6
- export function selectRerunTrialSlots(tasks, attempts, progress, selector) {
6
+ export function selectRerunTrialSlots(tasks, attempts, progress, selector, options = {}) {
7
7
  const planned = new Set(tasks);
8
8
  if (planned.size !== tasks.length || tasks.some((task) => typeof task !== "string" || task.length === 0)) {
9
9
  throw unavailable("eval task plan is invalid");
@@ -12,7 +12,7 @@ export function selectRerunTrialSlots(tasks, attempts, progress, selector) {
12
12
  throw unavailable("eval attempts plan is invalid");
13
13
  const invalid = invalidTrialSlots(tasks, attempts, progress);
14
14
  if (selector.mode === "invalid")
15
- return invalid;
15
+ return options.allowVerifierFailures ? invalid : rejectVerifierOnlyReruns(progress, invalid);
16
16
  const requested = [...new Set(selector.taskNames)].sort();
17
17
  if (requested.length === 0)
18
18
  throw invalidInput("eval rerun requires at least one --task");
@@ -24,7 +24,16 @@ export function selectRerunTrialSlots(tasks, attempts, progress, selector) {
24
24
  }
25
25
  }
26
26
  const requestedSet = new Set(requested);
27
- return invalid.filter((slot) => requestedSet.has(slot.task_id));
27
+ const selected = invalid.filter((slot) => requestedSet.has(slot.task_id));
28
+ return options.allowVerifierFailures ? selected : rejectVerifierOnlyReruns(progress, selected);
29
+ }
30
+ function rejectVerifierOnlyReruns(progress, slots) {
31
+ const selected = new Set(slots.map(slotKey));
32
+ const verifierOnly = progress.trials.filter((trial) => selected.has(evalTrialKey(trial))
33
+ && (trial.invalid_reason === "verifier_infrastructure_failure" || trial.invalid_reason === "verifier_result_missing"));
34
+ if (verifierOnly.length === 0)
35
+ return slots;
36
+ throw new HitchError(`verifier-only retry is unavailable after the original trial environment closed: ${verifierOnly.map((trial) => `${trial.task_id}#${trial.attempt}`).join(", ")}; the Candidate Agent will not be rerun`, { code: "eval_verifier_only_rerun_unavailable", exitCode: 2 });
28
37
  }
29
38
  export function invalidTrialSlots(tasks, attempts, progress) {
30
39
  const byKey = new Map();
@@ -1 +1 @@
1
- {"version":3,"file":"rerun-slots.js","sourceRoot":"","sources":["../../../src/evals/rerun-slots.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AAClE,OAAO,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AAW7C,MAAM,UAAU,gBAAgB,CAC9B,KAAwB,EACxB,QAAwB,EACxB,QAAuB,EACvB,QAAQ,GAAG,CAAC;IAEZ,OAAO,WAAW,CAAC,qBAAqB,CAAC,KAAK,EAAE,QAAQ,EAAE,QAAQ,EAAE,QAAQ,CAAC,CAAC,CAAC;AACjF,CAAC;AAED,MAAM,UAAU,qBAAqB,CACnC,KAAwB,EACxB,QAAgB,EAChB,QAAwB,EACxB,QAAuB;IAEvB,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,CAAC;IAC/B,IAAI,OAAO,CAAC,IAAI,KAAK,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,CAAC,EAAE,CAAC;QACzG,MAAM,WAAW,CAAC,2BAA2B,CAAC,CAAC;IACjD,CAAC;IACD,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,QAAQ,CAAC,IAAI,QAAQ,GAAG,CAAC;QAAE,MAAM,WAAW,CAAC,+BAA+B,CAAC,CAAC;IACxG,MAAM,OAAO,GAAG,iBAAiB,CAAC,KAAK,EAAE,QAAQ,EAAE,QAAQ,CAAC,CAAC;IAC7D,IAAI,QAAQ,CAAC,IAAI,KAAK,SAAS;QAAE,OAAO,OAAO,CAAC;IAChD,MAAM,SAAS,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IAC1D,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;QAAE,MAAM,YAAY,CAAC,yCAAyC,CAAC,CAAC;IAC1F,KAAK,MAAM,IAAI,IAAI,SAAS,EAAE,CAAC;QAC7B,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC;YAAE,MAAM,IAAI,UAAU,CAAC,iCAAiC,IAAI,EAAE,EAAE,EAAE,IAAI,EAAE,yBAAyB,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;QACxI,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,KAAK,IAAI,CAAC,EAAE,CAAC;YACnD,MAAM,IAAI,UAAU,CAAC,+BAA+B,IAAI,EAAE,EAAE,EAAE,IAAI,EAAE,yBAAyB,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;QAChH,CAAC;IACH,CAAC;IACD,MAAM,YAAY,GAAG,IAAI,GAAG,CAAC,SAAS,CAAC,CAAC;IACxC,OAAO,OAAO,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC;AAClE,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,KAAwB,EAAE,QAAgB,EAAE,QAAwB;IACpG,MAAM,KAAK,GAAG,IAAI,GAAG,EAA0B,CAAC;IAChD,KAAK,MAAM,KAAK,IAAI,QAAQ,CAAC,MAAM,EAAE,CAAC;QACpC,MAAM,GAAG,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC;QAChC,IAAI,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC;YAAE,MAAM,WAAW,CAAC,qCAAqC,KAAK,CAAC,OAAO,YAAY,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC;QACrH,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;IACxB,CAAC;IACD,MAAM,OAAO,GAAoB,EAAE,CAAC;IACpC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,QAAQ,EAAE,OAAO,IAAI,CAAC,EAAE,CAAC;YACxD,IAAI,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC,EAAE,kBAAkB,KAAK,OAAO,EAAE,CAAC;gBACnF,OAAO,CAAC,IAAI,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;YAC3C,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC,OAAO,CAAC,CAAC;AAC5B,CAAC;AAED,MAAM,UAAU,oBAAoB,CAClC,QAAwB,EACxB,IAAoD,EACpD,OAAoB,EACpB,MAAc;IAEd,IAAI,QAAQ,CAAC,OAAO,KAAK,MAAM;QAAE,MAAM,WAAW,CAAC,mCAAmC,CAAC,CAAC;IACxF,IAAI,QAAQ,CAAC,YAAY,KAAK,OAAO,CAAC,YAAY,IAAI,QAAQ,CAAC,kBAAkB,KAAK,OAAO,CAAC,kBAAkB,EAAE,CAAC;QACjH,MAAM,WAAW,CAAC,0CAA0C,CAAC,CAAC;IAChE,CAAC;IACD,IAAI,QAAQ,CAAC,aAAa,KAAK,IAAI,CAAC,KAAK,CAAC,MAAM,IAAI,QAAQ,CAAC,cAAc,KAAK,IAAI,CAAC,KAAK,CAAC,MAAM,GAAG,IAAI,CAAC,QAAQ,EAAE,CAAC;QAClH,MAAM,WAAW,CAAC,2DAA2D,CAAC,CAAC;IACjF,CAAC;IACD,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IACpC,KAAK,MAAM,KAAK,IAAI,QAAQ,CAAC,MAAM,EAAE,CAAC;QACpC,IAAI,KAAK,CAAC,OAAO,GAAG,IAAI,CAAC,QAAQ,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC;YACjE,MAAM,WAAW,CAAC,8CAA8C,KAAK,CAAC,OAAO,YAAY,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC;QAC5G,CAAC;IACH,CAAC;IACD,iBAAiB,CAAC,IAAI,CAAC,KAAK,EAAE,IAAI,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;AACzD,CAAC;AAED,MAAM,UAAU,OAAO,CAAC,IAAmB;IACzC,OAAO,GAAG,IAAI,CAAC,OAAO,SAAS,IAAI,CAAC,OAAO,EAAE,CAAC;AAChD,CAAC;AAED,MAAM,UAAU,SAAS,CAAC,KAA+B;IACvD,OAAO,CAAC,GAAG,KAAK,CAAC;SACd,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;SACjE,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC;AACtG,CAAC;AAED,MAAM,UAAU,WAAW,CAAC,KAA+B;IACzD,OAAO,CAAC,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;AAChE,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,KAA+B;IACjE,MAAM,OAAO,GAAG,IAAI,GAAG,EAA2B,CAAC;IACnD,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC;QAChD,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACnB,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;IACrC,CAAC;IACD,OAAO,CAAC,GAAG,OAAO,CAAC,OAAO,EAAE,CAAC;SAC1B,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC,IAAI,GAAG,KAAK,CAAC;SACvC,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,MAAM,CAAC,EAAE,EAAE,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;AAC9D,CAAC;AAED,MAAM,UAAU,oBAAoB,CAAC,OAAe;IAClD,OAAO,WAAW,MAAM,CAAC,OAAO,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC;AACvD,CAAC;AAED,MAAM,UAAU,UAAU,CAAC,IAAmB;IAC5C,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;AAC3C,CAAC;AAED,SAAS,WAAW,CAAC,OAAe;IAClC,OAAO,IAAI,UAAU,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,wBAAwB,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;AAClF,CAAC"}
1
+ {"version":3,"file":"rerun-slots.js","sourceRoot":"","sources":["../../../src/evals/rerun-slots.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AAClE,OAAO,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AAW7C,MAAM,UAAU,gBAAgB,CAC9B,KAAwB,EACxB,QAAwB,EACxB,QAAuB,EACvB,QAAQ,GAAG,CAAC;IAEZ,OAAO,WAAW,CAAC,qBAAqB,CAAC,KAAK,EAAE,QAAQ,EAAE,QAAQ,EAAE,QAAQ,CAAC,CAAC,CAAC;AACjF,CAAC;AAED,MAAM,UAAU,qBAAqB,CACnC,KAAwB,EACxB,QAAgB,EAChB,QAAwB,EACxB,QAAuB,EACvB,UAA+C,EAAE;IAEjD,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,CAAC;IAC/B,IAAI,OAAO,CAAC,IAAI,KAAK,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,CAAC,EAAE,CAAC;QACzG,MAAM,WAAW,CAAC,2BAA2B,CAAC,CAAC;IACjD,CAAC;IACD,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,QAAQ,CAAC,IAAI,QAAQ,GAAG,CAAC;QAAE,MAAM,WAAW,CAAC,+BAA+B,CAAC,CAAC;IACxG,MAAM,OAAO,GAAG,iBAAiB,CAAC,KAAK,EAAE,QAAQ,EAAE,QAAQ,CAAC,CAAC;IAC7D,IAAI,QAAQ,CAAC,IAAI,KAAK,SAAS;QAAE,OAAO,OAAO,CAAC,qBAAqB,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,wBAAwB,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IAC9H,MAAM,SAAS,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IAC1D,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;QAAE,MAAM,YAAY,CAAC,yCAAyC,CAAC,CAAC;IAC1F,KAAK,MAAM,IAAI,IAAI,SAAS,EAAE,CAAC;QAC7B,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC;YAAE,MAAM,IAAI,UAAU,CAAC,iCAAiC,IAAI,EAAE,EAAE,EAAE,IAAI,EAAE,yBAAyB,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;QACxI,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,KAAK,IAAI,CAAC,EAAE,CAAC;YACnD,MAAM,IAAI,UAAU,CAAC,+BAA+B,IAAI,EAAE,EAAE,EAAE,IAAI,EAAE,yBAAyB,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;QAChH,CAAC;IACH,CAAC;IACD,MAAM,YAAY,GAAG,IAAI,GAAG,CAAC,SAAS,CAAC,CAAC;IACxC,MAAM,QAAQ,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC;IAC1E,OAAO,OAAO,CAAC,qBAAqB,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,wBAAwB,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;AACjG,CAAC;AAED,SAAS,wBAAwB,CAAC,QAAwB,EAAE,KAAsB;IAChF,MAAM,QAAQ,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC;IAC7C,MAAM,YAAY,GAAG,QAAQ,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,QAAQ,CAAC,GAAG,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC;WACnF,CAAC,KAAK,CAAC,cAAc,KAAK,iCAAiC,IAAI,KAAK,CAAC,cAAc,KAAK,yBAAyB,CAAC,CAAC,CAAC;IACzH,IAAI,YAAY,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IAC5C,MAAM,IAAI,UAAU,CAClB,mFAAmF,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,GAAG,KAAK,CAAC,OAAO,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,yCAAyC,EACvM,EAAE,IAAI,EAAE,sCAAsC,EAAE,QAAQ,EAAE,CAAC,EAAE,CAC9D,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,KAAwB,EAAE,QAAgB,EAAE,QAAwB;IACpG,MAAM,KAAK,GAAG,IAAI,GAAG,EAA0B,CAAC;IAChD,KAAK,MAAM,KAAK,IAAI,QAAQ,CAAC,MAAM,EAAE,CAAC;QACpC,MAAM,GAAG,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC;QAChC,IAAI,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC;YAAE,MAAM,WAAW,CAAC,qCAAqC,KAAK,CAAC,OAAO,YAAY,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC;QACrH,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;IACxB,CAAC;IACD,MAAM,OAAO,GAAoB,EAAE,CAAC;IACpC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,QAAQ,EAAE,OAAO,IAAI,CAAC,EAAE,CAAC;YACxD,IAAI,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC,EAAE,kBAAkB,KAAK,OAAO,EAAE,CAAC;gBACnF,OAAO,CAAC,IAAI,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC;YAC3C,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC,OAAO,CAAC,CAAC;AAC5B,CAAC;AAED,MAAM,UAAU,oBAAoB,CAClC,QAAwB,EACxB,IAAoD,EACpD,OAAoB,EACpB,MAAc;IAEd,IAAI,QAAQ,CAAC,OAAO,KAAK,MAAM;QAAE,MAAM,WAAW,CAAC,mCAAmC,CAAC,CAAC;IACxF,IAAI,QAAQ,CAAC,YAAY,KAAK,OAAO,CAAC,YAAY,IAAI,QAAQ,CAAC,kBAAkB,KAAK,OAAO,CAAC,kBAAkB,EAAE,CAAC;QACjH,MAAM,WAAW,CAAC,0CAA0C,CAAC,CAAC;IAChE,CAAC;IACD,IAAI,QAAQ,CAAC,aAAa,KAAK,IAAI,CAAC,KAAK,CAAC,MAAM,IAAI,QAAQ,CAAC,cAAc,KAAK,IAAI,CAAC,KAAK,CAAC,MAAM,GAAG,IAAI,CAAC,QAAQ,EAAE,CAAC;QAClH,MAAM,WAAW,CAAC,2DAA2D,CAAC,CAAC;IACjF,CAAC;IACD,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IACpC,KAAK,MAAM,KAAK,IAAI,QAAQ,CAAC,MAAM,EAAE,CAAC;QACpC,IAAI,KAAK,CAAC,OAAO,GAAG,IAAI,CAAC,QAAQ,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC;YACjE,MAAM,WAAW,CAAC,8CAA8C,KAAK,CAAC,OAAO,YAAY,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC;QAC5G,CAAC;IACH,CAAC;IACD,iBAAiB,CAAC,IAAI,CAAC,KAAK,EAAE,IAAI,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;AACzD,CAAC;AAED,MAAM,UAAU,OAAO,CAAC,IAAmB;IACzC,OAAO,GAAG,IAAI,CAAC,OAAO,SAAS,IAAI,CAAC,OAAO,EAAE,CAAC;AAChD,CAAC;AAED,MAAM,UAAU,SAAS,CAAC,KAA+B;IACvD,OAAO,CAAC,GAAG,KAAK,CAAC;SACd,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;SACjE,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC;AACtG,CAAC;AAED,MAAM,UAAU,WAAW,CAAC,KAA+B;IACzD,OAAO,CAAC,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;AAChE,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,KAA+B;IACjE,MAAM,OAAO,GAAG,IAAI,GAAG,EAA2B,CAAC;IACnD,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC;QAChD,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACnB,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;IACrC,CAAC;IACD,OAAO,CAAC,GAAG,OAAO,CAAC,OAAO,EAAE,CAAC;SAC1B,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC,IAAI,GAAG,KAAK,CAAC;SACvC,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,MAAM,CAAC,EAAE,EAAE,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;AAC9D,CAAC;AAED,MAAM,UAAU,oBAAoB,CAAC,OAAe;IAClD,OAAO,WAAW,MAAM,CAAC,OAAO,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC;AACvD,CAAC;AAED,MAAM,UAAU,UAAU,CAAC,IAAmB;IAC5C,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;AAC3C,CAAC;AAED,SAAS,WAAW,CAAC,OAAe;IAClC,OAAO,IAAI,UAAU,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,wBAAwB,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC;AAClF,CAAC"}
@@ -0,0 +1,49 @@
1
+ import { HitchError, invalidInput } from "../foundation/index.js";
2
+ export const EVAL_RERUN_TYPES = [
3
+ "candidate-restart",
4
+ "candidate-resume",
5
+ "trajectory-replay",
6
+ "verifier-only",
7
+ "collect-only",
8
+ ];
9
+ export function parseEvalRerunType(value) {
10
+ if (typeof value !== "string" || !EVAL_RERUN_TYPES.includes(value)) {
11
+ throw invalidInput(`eval rerun --type must be one of: ${EVAL_RERUN_TYPES.join(", ")}`);
12
+ }
13
+ return value;
14
+ }
15
+ export function evalRerunSemantics(type) {
16
+ switch (type) {
17
+ case "candidate-restart":
18
+ return { candidate_action: "restart", conversation_source: "original-instruction", sandbox_source: "clean", candidate_executes: true };
19
+ case "candidate-resume":
20
+ return { candidate_action: "resume", conversation_source: "native-session", sandbox_source: "checkpoint", candidate_executes: true };
21
+ case "trajectory-replay":
22
+ return { candidate_action: "replay", conversation_source: "canonical-trajectory", sandbox_source: "checkpoint", candidate_executes: true };
23
+ case "verifier-only":
24
+ return { candidate_action: "none", conversation_source: "none", sandbox_source: "retained", candidate_executes: false };
25
+ case "collect-only":
26
+ return { candidate_action: "none", conversation_source: "none", sandbox_source: "none", candidate_executes: false };
27
+ }
28
+ }
29
+ export function assertEvalRerunTypeSupported(type) {
30
+ if (type === "candidate-restart" || type === "collect-only")
31
+ return;
32
+ const unavailable = {
33
+ "candidate-resume": {
34
+ code: "eval_candidate_resume_unavailable",
35
+ message: "candidate-resume requires both a restorable sandbox checkpoint and adapter-native session resume; the current Harbor execution does not retain both",
36
+ },
37
+ "trajectory-replay": {
38
+ code: "eval_trajectory_replay_unavailable",
39
+ message: "trajectory-replay requires a verified trajectory, a restorable sandbox checkpoint, and explicit adapter replay support; trajectory evidence alone cannot restore tool or process state",
40
+ },
41
+ "verifier-only": {
42
+ code: "eval_verifier_only_rerun_unavailable",
43
+ message: "verifier-only requires the original retained sandbox; current verifier retries run only while that sandbox is live",
44
+ },
45
+ };
46
+ const failure = unavailable[type];
47
+ throw new HitchError(failure.message, { code: failure.code, exitCode: 2 });
48
+ }
49
+ //# sourceMappingURL=rerun-types.js.map