agent-hitch 0.2.6 → 0.2.8

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 (562) hide show
  1. package/README.md +30 -3
  2. package/README.zh-CN.md +14 -2
  3. package/dist/bin/hitch.js +25 -3
  4. package/dist/bin/hitch.js.map +1 -1
  5. package/dist/scripts/canary-benchmark.js +54 -0
  6. package/dist/scripts/canary-benchmark.js.map +1 -0
  7. package/dist/scripts/canary-buildkit-secrets.js +166 -0
  8. package/dist/scripts/canary-buildkit-secrets.js.map +1 -0
  9. package/dist/scripts/canary-collision-domains.js +77 -0
  10. package/dist/scripts/canary-collision-domains.js.map +1 -0
  11. package/dist/scripts/canary-docker-image-observation.js +181 -0
  12. package/dist/scripts/canary-docker-image-observation.js.map +1 -0
  13. package/dist/scripts/canary-eval-scheduler-throughput.js +47 -0
  14. package/dist/scripts/canary-eval-scheduler-throughput.js.map +1 -0
  15. package/dist/scripts/canary-gpu-hardware.js +95 -0
  16. package/dist/scripts/canary-gpu-hardware.js.map +1 -0
  17. package/dist/scripts/canary-harbor-load.js +365 -0
  18. package/dist/scripts/canary-harbor-load.js.map +1 -0
  19. package/dist/scripts/canary-packaged-harness.js +75 -0
  20. package/dist/scripts/canary-packaged-harness.js.map +1 -0
  21. package/dist/scripts/canary-resource-load.js +188 -0
  22. package/dist/scripts/canary-resource-load.js.map +1 -0
  23. package/dist/scripts/check-architecture.js +10 -5
  24. package/dist/scripts/check-architecture.js.map +1 -1
  25. package/dist/scripts/docker-reaper-crash-child.js +21 -0
  26. package/dist/scripts/docker-reaper-crash-child.js.map +1 -0
  27. package/dist/src/adapters/catalog.js +3 -1
  28. package/dist/src/adapters/catalog.js.map +1 -1
  29. package/dist/src/adapters/discovery.js +1 -1
  30. package/dist/src/adapters/discovery.js.map +1 -1
  31. package/dist/src/adapters/providers/claude.js +6 -0
  32. package/dist/src/adapters/providers/claude.js.map +1 -1
  33. package/dist/src/adapters/providers/codex-auth.js +29 -0
  34. package/dist/src/adapters/providers/codex-auth.js.map +1 -0
  35. package/dist/src/adapters/providers/codex.js +29 -6
  36. package/dist/src/adapters/providers/codex.js.map +1 -1
  37. package/dist/src/adapters/providers/deepseek.js +6 -0
  38. package/dist/src/adapters/providers/deepseek.js.map +1 -1
  39. package/dist/src/adapters/providers/model-call.js +26 -0
  40. package/dist/src/adapters/providers/model-call.js.map +1 -0
  41. package/dist/src/adapters/providers/opencode.js +6 -0
  42. package/dist/src/adapters/providers/opencode.js.map +1 -1
  43. package/dist/src/adapters/providers/pi.js +6 -0
  44. package/dist/src/adapters/providers/pi.js.map +1 -1
  45. package/dist/src/artifacts/index.js +1 -0
  46. package/dist/src/artifacts/index.js.map +1 -1
  47. package/dist/src/artifacts/preparer.js +7 -2
  48. package/dist/src/artifacts/preparer.js.map +1 -1
  49. package/dist/src/backends/harbor/agent-budget.js +31 -0
  50. package/dist/src/backends/harbor/agent-budget.js.map +1 -0
  51. package/dist/src/backends/harbor/backend.js +59 -107
  52. package/dist/src/backends/harbor/backend.js.map +1 -1
  53. package/dist/src/backends/harbor/environment-config.js +92 -0
  54. package/dist/src/backends/harbor/environment-config.js.map +1 -0
  55. package/dist/src/backends/harbor/index.js +6 -1
  56. package/dist/src/backends/harbor/index.js.map +1 -1
  57. package/dist/src/backends/harbor/model-proxy-config.js +23 -0
  58. package/dist/src/backends/harbor/model-proxy-config.js.map +1 -0
  59. package/dist/src/backends/harbor/process.js +177 -0
  60. package/dist/src/backends/harbor/process.js.map +1 -0
  61. package/dist/src/backends/harbor/regrade.js +77 -0
  62. package/dist/src/backends/harbor/regrade.js.map +1 -0
  63. package/dist/src/backends/harbor/runtime-toolchain.js +30 -0
  64. package/dist/src/backends/harbor/runtime-toolchain.js.map +1 -0
  65. package/dist/src/backends/harbor/supervisor.js +74 -0
  66. package/dist/src/backends/harbor/supervisor.js.map +1 -0
  67. package/dist/src/backends/harbor/task-resources.js +206 -0
  68. package/dist/src/backends/harbor/task-resources.js.map +1 -0
  69. package/dist/src/backends/harbor/tools.js +6 -29
  70. package/dist/src/backends/harbor/tools.js.map +1 -1
  71. package/dist/src/backends/index.js +5 -1
  72. package/dist/src/backends/index.js.map +1 -1
  73. package/dist/src/benchmarks/index.js +3 -0
  74. package/dist/src/benchmarks/index.js.map +1 -0
  75. package/dist/src/benchmarks/loader.js +271 -0
  76. package/dist/src/benchmarks/loader.js.map +1 -0
  77. package/dist/src/benchmarks/metrics.js +16 -0
  78. package/dist/src/benchmarks/metrics.js.map +1 -0
  79. package/dist/src/benchmarks/toml.js +13 -0
  80. package/dist/src/benchmarks/toml.js.map +1 -0
  81. package/dist/src/benchmarks/validation.js +242 -0
  82. package/dist/src/benchmarks/validation.js.map +1 -0
  83. package/dist/src/cli/arguments.js +6 -4
  84. package/dist/src/cli/arguments.js.map +1 -1
  85. package/dist/src/cli/commands/benchmark.js +22 -0
  86. package/dist/src/cli/commands/benchmark.js.map +1 -0
  87. package/dist/src/cli/commands/capabilities.js +18 -0
  88. package/dist/src/cli/commands/capabilities.js.map +1 -0
  89. package/dist/src/cli/commands/daemon.js +135 -8
  90. package/dist/src/cli/commands/daemon.js.map +1 -1
  91. package/dist/src/cli/commands/eval.js +228 -6
  92. package/dist/src/cli/commands/eval.js.map +1 -1
  93. package/dist/src/cli/commands/images.js +40 -0
  94. package/dist/src/cli/commands/images.js.map +1 -0
  95. package/dist/src/cli/commands/run.js +42 -17
  96. package/dist/src/cli/commands/run.js.map +1 -1
  97. package/dist/src/cli/commands/runs.js +25 -3
  98. package/dist/src/cli/commands/runs.js.map +1 -1
  99. package/dist/src/cli/commands/trajectory.js +96 -5
  100. package/dist/src/cli/commands/trajectory.js.map +1 -1
  101. package/dist/src/cli/commands/verifier.js +36 -0
  102. package/dist/src/cli/commands/verifier.js.map +1 -0
  103. package/dist/src/cli/commands/worker.js +79 -0
  104. package/dist/src/cli/commands/worker.js.map +1 -0
  105. package/dist/src/cli/main.js +10 -0
  106. package/dist/src/cli/main.js.map +1 -1
  107. package/dist/src/cli/output.js +84 -4
  108. package/dist/src/cli/output.js.map +1 -1
  109. package/dist/src/control-plane/build-admission.js +69 -0
  110. package/dist/src/control-plane/build-admission.js.map +1 -0
  111. package/dist/src/control-plane/build-records.js +6 -0
  112. package/dist/src/control-plane/build-records.js.map +1 -0
  113. package/dist/src/control-plane/collisions.js +40 -0
  114. package/dist/src/control-plane/collisions.js.map +1 -0
  115. package/dist/src/control-plane/eval-control-state.js +22 -0
  116. package/dist/src/control-plane/eval-control-state.js.map +1 -0
  117. package/dist/src/control-plane/eval-control-work.js +44 -0
  118. package/dist/src/control-plane/eval-control-work.js.map +1 -0
  119. package/dist/src/control-plane/eval-image-resolution.js +52 -0
  120. package/dist/src/control-plane/eval-image-resolution.js.map +1 -0
  121. package/dist/src/control-plane/eval-image-services.js +30 -0
  122. package/dist/src/control-plane/eval-image-services.js.map +1 -0
  123. package/dist/src/control-plane/eval-records.js +252 -0
  124. package/dist/src/control-plane/eval-records.js.map +1 -0
  125. package/dist/src/control-plane/eval-recovery.js +181 -0
  126. package/dist/src/control-plane/eval-recovery.js.map +1 -0
  127. package/dist/src/control-plane/eval-scheduler.js +450 -0
  128. package/dist/src/control-plane/eval-scheduler.js.map +1 -0
  129. package/dist/src/control-plane/index.js +20 -0
  130. package/dist/src/control-plane/index.js.map +1 -0
  131. package/dist/src/control-plane/local-worker.js +49 -0
  132. package/dist/src/control-plane/local-worker.js.map +1 -0
  133. package/dist/src/control-plane/model-capture-planning.js +12 -0
  134. package/dist/src/control-plane/model-capture-planning.js.map +1 -0
  135. package/dist/src/control-plane/remote-result-transport.js +175 -0
  136. package/dist/src/control-plane/remote-result-transport.js.map +1 -0
  137. package/dist/src/control-plane/remote-work-coordinator.js +333 -0
  138. package/dist/src/control-plane/remote-work-coordinator.js.map +1 -0
  139. package/dist/src/control-plane/remote-work-inputs.js +266 -0
  140. package/dist/src/control-plane/remote-work-inputs.js.map +1 -0
  141. package/dist/src/control-plane/remote-work-item.js +62 -0
  142. package/dist/src/control-plane/remote-work-item.js.map +1 -0
  143. package/dist/src/control-plane/remote-work-recovery.js +220 -0
  144. package/dist/src/control-plane/remote-work-recovery.js.map +1 -0
  145. package/dist/src/control-plane/remote-worker-artifacts.js +135 -0
  146. package/dist/src/control-plane/remote-worker-artifacts.js.map +1 -0
  147. package/dist/src/control-plane/remote-worker-client.js +202 -0
  148. package/dist/src/control-plane/remote-worker-client.js.map +1 -0
  149. package/dist/src/control-plane/remote-worker-configuration.js +34 -0
  150. package/dist/src/control-plane/remote-worker-configuration.js.map +1 -0
  151. package/dist/src/control-plane/remote-worker-credentials.js +58 -0
  152. package/dist/src/control-plane/remote-worker-credentials.js.map +1 -0
  153. package/dist/src/control-plane/remote-worker-protocol.js +489 -0
  154. package/dist/src/control-plane/remote-worker-protocol.js.map +1 -0
  155. package/dist/src/control-plane/remote-worker-runner.js +214 -0
  156. package/dist/src/control-plane/remote-worker-runner.js.map +1 -0
  157. package/dist/src/control-plane/remote-workers.js +377 -0
  158. package/dist/src/control-plane/remote-workers.js.map +1 -0
  159. package/dist/src/control-plane/rerun-scheduler.js +448 -0
  160. package/dist/src/control-plane/rerun-scheduler.js.map +1 -0
  161. package/dist/src/control-plane/rerun-submission.js +62 -0
  162. package/dist/src/control-plane/rerun-submission.js.map +1 -0
  163. package/dist/src/control-plane/resources.js +182 -0
  164. package/dist/src/control-plane/resources.js.map +1 -0
  165. package/dist/src/control-plane/scheduler-eval-entry.js +32 -0
  166. package/dist/src/control-plane/scheduler-eval-entry.js.map +1 -0
  167. package/dist/src/control-plane/synthetic-result.js +21 -0
  168. package/dist/src/control-plane/synthetic-result.js.map +1 -0
  169. package/dist/src/control-plane/work-admission.js +23 -0
  170. package/dist/src/control-plane/work-admission.js.map +1 -0
  171. package/dist/src/control-plane/work-dispatcher.js +237 -0
  172. package/dist/src/control-plane/work-dispatcher.js.map +1 -0
  173. package/dist/src/controller-runtime/hash.js +19 -3
  174. package/dist/src/controller-runtime/hash.js.map +1 -1
  175. package/dist/src/controller-runtime/index.js +1 -1
  176. package/dist/src/controller-runtime/index.js.map +1 -1
  177. package/dist/src/controller-runtime/store.js +15 -6
  178. package/dist/src/controller-runtime/store.js.map +1 -1
  179. package/dist/src/daemon/auth.js +5 -0
  180. package/dist/src/daemon/auth.js.map +1 -1
  181. package/dist/src/daemon/client.js.map +1 -1
  182. package/dist/src/daemon/event-stream.js +46 -0
  183. package/dist/src/daemon/event-stream.js.map +1 -0
  184. package/dist/src/daemon/health.js +66 -0
  185. package/dist/src/daemon/health.js.map +1 -0
  186. package/dist/src/daemon/launcher.js +28 -1
  187. package/dist/src/daemon/launcher.js.map +1 -1
  188. package/dist/src/daemon/logging.js +23 -0
  189. package/dist/src/daemon/logging.js.map +1 -0
  190. package/dist/src/daemon/scheduler.js +26 -5
  191. package/dist/src/daemon/scheduler.js.map +1 -1
  192. package/dist/src/daemon/server.js +255 -98
  193. package/dist/src/daemon/server.js.map +1 -1
  194. package/dist/src/daemon/telemetry.js +163 -0
  195. package/dist/src/daemon/telemetry.js.map +1 -0
  196. package/dist/src/daemon/worker-routes.js +219 -0
  197. package/dist/src/daemon/worker-routes.js.map +1 -0
  198. package/dist/src/domain/benchmarks.js +2 -0
  199. package/dist/src/domain/benchmarks.js.map +1 -0
  200. package/dist/src/domain/bundles.js +2 -0
  201. package/dist/src/domain/bundles.js.map +1 -0
  202. package/dist/src/domain/control-plane.js +2 -0
  203. package/dist/src/domain/control-plane.js.map +1 -0
  204. package/dist/src/domain/execution-evidence.js +2 -0
  205. package/dist/src/domain/execution-evidence.js.map +1 -0
  206. package/dist/src/domain/execution-plan.js +2 -0
  207. package/dist/src/domain/execution-plan.js.map +1 -0
  208. package/dist/src/domain/images.js +2 -0
  209. package/dist/src/domain/images.js.map +1 -0
  210. package/dist/src/domain/index.js +13 -0
  211. package/dist/src/domain/index.js.map +1 -1
  212. package/dist/src/domain/interactions.js +2 -0
  213. package/dist/src/domain/interactions.js.map +1 -0
  214. package/dist/src/domain/providers.js +2 -0
  215. package/dist/src/domain/providers.js.map +1 -0
  216. package/dist/src/domain/resources.js +2 -0
  217. package/dist/src/domain/resources.js.map +1 -0
  218. package/dist/src/domain/runs.js.map +1 -1
  219. package/dist/src/domain/validation.js +11 -4
  220. package/dist/src/domain/validation.js.map +1 -1
  221. package/dist/src/domain/verifier-evidence-validation.js +214 -0
  222. package/dist/src/domain/verifier-evidence-validation.js.map +1 -0
  223. package/dist/src/domain/verifier-evidence.js +3 -0
  224. package/dist/src/domain/verifier-evidence.js.map +1 -0
  225. package/dist/src/domain/verifier-score-contract.js +232 -0
  226. package/dist/src/domain/verifier-score-contract.js.map +1 -0
  227. package/dist/src/domain/workers.js +2 -0
  228. package/dist/src/domain/workers.js.map +1 -0
  229. package/dist/src/evals/benchmark-adapter-manifest.js +219 -0
  230. package/dist/src/evals/benchmark-adapter-manifest.js.map +1 -0
  231. package/dist/src/evals/benchmark-candidate.js +36 -0
  232. package/dist/src/evals/benchmark-candidate.js.map +1 -0
  233. package/dist/src/evals/benchmark-run.js +166 -0
  234. package/dist/src/evals/benchmark-run.js.map +1 -0
  235. package/dist/src/evals/collect-only-rerun.js +183 -0
  236. package/dist/src/evals/collect-only-rerun.js.map +1 -0
  237. package/dist/src/evals/directory.js +30 -0
  238. package/dist/src/evals/directory.js.map +1 -0
  239. package/dist/src/evals/docker-engine-stats.js +89 -0
  240. package/dist/src/evals/docker-engine-stats.js.map +1 -0
  241. package/dist/src/evals/docker-ownership.js +51 -0
  242. package/dist/src/evals/docker-ownership.js.map +1 -0
  243. package/dist/src/evals/docker-reaper.js +144 -0
  244. package/dist/src/evals/docker-reaper.js.map +1 -0
  245. package/dist/src/evals/docker-resource-observer.js +207 -0
  246. package/dist/src/evals/docker-resource-observer.js.map +1 -0
  247. package/dist/src/evals/duration-estimator.js +126 -0
  248. package/dist/src/evals/duration-estimator.js.map +1 -0
  249. package/dist/src/evals/environment-image-planning.js +164 -0
  250. package/dist/src/evals/environment-image-planning.js.map +1 -0
  251. package/dist/src/evals/eval-artifact-planning.js +56 -0
  252. package/dist/src/evals/eval-artifact-planning.js.map +1 -0
  253. package/dist/src/evals/eval-finalization.js +10 -0
  254. package/dist/src/evals/eval-finalization.js.map +1 -0
  255. package/dist/src/evals/eval-lifecycle-events.js +29 -0
  256. package/dist/src/evals/eval-lifecycle-events.js.map +1 -0
  257. package/dist/src/evals/eval-logical-plan.js +51 -0
  258. package/dist/src/evals/eval-logical-plan.js.map +1 -0
  259. package/dist/src/evals/eval-result-builder.js +81 -0
  260. package/dist/src/evals/eval-result-builder.js.map +1 -0
  261. package/dist/src/evals/events.js +53 -5
  262. package/dist/src/evals/events.js.map +1 -1
  263. package/dist/src/evals/evolution-baseline.js +191 -0
  264. package/dist/src/evals/evolution-baseline.js.map +1 -0
  265. package/dist/src/evals/execution-evidence.js +116 -0
  266. package/dist/src/evals/execution-evidence.js.map +1 -0
  267. package/dist/src/evals/execution-leases.js +342 -0
  268. package/dist/src/evals/execution-leases.js.map +1 -0
  269. package/dist/src/evals/execution-plan-artifacts.js +72 -0
  270. package/dist/src/evals/execution-plan-artifacts.js.map +1 -0
  271. package/dist/src/evals/execution-plan-images.js +69 -0
  272. package/dist/src/evals/execution-plan-images.js.map +1 -0
  273. package/dist/src/evals/execution-plan-resources.js +174 -0
  274. package/dist/src/evals/execution-plan-resources.js.map +1 -0
  275. package/dist/src/evals/execution-plan.js +451 -0
  276. package/dist/src/evals/execution-plan.js.map +1 -0
  277. package/dist/src/evals/failure-classifier.js +61 -0
  278. package/dist/src/evals/failure-classifier.js.map +1 -0
  279. package/dist/src/evals/harbor-artifact-builder-image.js +99 -0
  280. package/dist/src/evals/harbor-artifact-builder-image.js.map +1 -0
  281. package/dist/src/evals/harbor-artifact-builder.js +358 -0
  282. package/dist/src/evals/harbor-artifact-builder.js.map +1 -0
  283. package/dist/src/evals/harbor-bridge-error.js +58 -3
  284. package/dist/src/evals/harbor-bridge-error.js.map +1 -1
  285. package/dist/src/evals/harbor-node-runtime.js +146 -0
  286. package/dist/src/evals/harbor-node-runtime.js.map +1 -0
  287. package/dist/src/evals/harbor-phase-timings.js +60 -0
  288. package/dist/src/evals/harbor-phase-timings.js.map +1 -0
  289. package/dist/src/evals/index.js +40 -2
  290. package/dist/src/evals/index.js.map +1 -1
  291. package/dist/src/evals/infrastructure-retry.js +208 -84
  292. package/dist/src/evals/infrastructure-retry.js.map +1 -1
  293. package/dist/src/evals/interaction-capture-import.js +49 -0
  294. package/dist/src/evals/interaction-capture-import.js.map +1 -0
  295. package/dist/src/evals/local-docker-provider.js +372 -0
  296. package/dist/src/evals/local-docker-provider.js.map +1 -0
  297. package/dist/src/evals/local-eval-planning.js +80 -0
  298. package/dist/src/evals/local-eval-planning.js.map +1 -0
  299. package/dist/src/evals/model-capture-plan.js +22 -0
  300. package/dist/src/evals/model-capture-plan.js.map +1 -0
  301. package/dist/src/evals/model-capture-runtime.js +114 -0
  302. package/dist/src/evals/model-capture-runtime.js.map +1 -0
  303. package/dist/src/evals/model-proxy-runtime-state.js +53 -0
  304. package/dist/src/evals/model-proxy-runtime-state.js.map +1 -0
  305. package/dist/src/evals/native-phase-evidence.js +440 -0
  306. package/dist/src/evals/native-phase-evidence.js.map +1 -0
  307. package/dist/src/evals/physical-retry-work.js +31 -0
  308. package/dist/src/evals/physical-retry-work.js.map +1 -0
  309. package/dist/src/evals/planned-execution-support.js +66 -0
  310. package/dist/src/evals/planned-execution-support.js.map +1 -0
  311. package/dist/src/evals/planned-execution.js +469 -0
  312. package/dist/src/evals/planned-execution.js.map +1 -0
  313. package/dist/src/evals/planned-progress-publisher.js +62 -0
  314. package/dist/src/evals/planned-progress-publisher.js.map +1 -0
  315. package/dist/src/evals/planned-remote-execution.js +40 -0
  316. package/dist/src/evals/planned-remote-execution.js.map +1 -0
  317. package/dist/src/evals/planned-retry-execution.js +174 -0
  318. package/dist/src/evals/planned-retry-execution.js.map +1 -0
  319. package/dist/src/evals/planned-retry-lifecycle.js +188 -0
  320. package/dist/src/evals/planned-retry-lifecycle.js.map +1 -0
  321. package/dist/src/evals/preparation-rerun.js +225 -0
  322. package/dist/src/evals/preparation-rerun.js.map +1 -0
  323. package/dist/src/evals/prepared-harness.js +76 -0
  324. package/dist/src/evals/prepared-harness.js.map +1 -0
  325. package/dist/src/evals/progress.js +52 -12
  326. package/dist/src/evals/progress.js.map +1 -1
  327. package/dist/src/evals/records.js +20 -3
  328. package/dist/src/evals/records.js.map +1 -1
  329. package/dist/src/evals/recovery.js +242 -0
  330. package/dist/src/evals/recovery.js.map +1 -0
  331. package/dist/src/evals/regrade-evidence.js +77 -0
  332. package/dist/src/evals/regrade-evidence.js.map +1 -0
  333. package/dist/src/evals/remote-infrastructure-retry.js +165 -0
  334. package/dist/src/evals/remote-infrastructure-retry.js.map +1 -0
  335. package/dist/src/evals/request.js +21 -3
  336. package/dist/src/evals/request.js.map +1 -1
  337. package/dist/src/evals/rerun-inputs.js +108 -0
  338. package/dist/src/evals/rerun-inputs.js.map +1 -0
  339. package/dist/src/evals/rerun-slots.js +4 -3
  340. package/dist/src/evals/rerun-slots.js.map +1 -1
  341. package/dist/src/evals/rerun-types.js +45 -0
  342. package/dist/src/evals/rerun-types.js.map +1 -0
  343. package/dist/src/evals/rerun.js +133 -88
  344. package/dist/src/evals/rerun.js.map +1 -1
  345. package/dist/src/evals/result-helpers.js +78 -0
  346. package/dist/src/evals/result-helpers.js.map +1 -0
  347. package/dist/src/evals/resume-state.js +36 -0
  348. package/dist/src/evals/resume-state.js.map +1 -0
  349. package/dist/src/evals/retry-backoff.js +20 -0
  350. package/dist/src/evals/retry-backoff.js.map +1 -0
  351. package/dist/src/evals/retry-state.js +195 -0
  352. package/dist/src/evals/retry-state.js.map +1 -0
  353. package/dist/src/evals/scheduler-metrics.js +95 -0
  354. package/dist/src/evals/scheduler-metrics.js.map +1 -0
  355. package/dist/src/evals/scheduler-trace-replay.js +107 -0
  356. package/dist/src/evals/scheduler-trace-replay.js.map +1 -0
  357. package/dist/src/evals/service-types.js +2 -0
  358. package/dist/src/evals/service-types.js.map +1 -0
  359. package/dist/src/evals/service.js +331 -347
  360. package/dist/src/evals/service.js.map +1 -1
  361. package/dist/src/evals/task-resources.js +209 -0
  362. package/dist/src/evals/task-resources.js.map +1 -0
  363. package/dist/src/evals/trial-environment-evidence.js +87 -0
  364. package/dist/src/evals/trial-environment-evidence.js.map +1 -0
  365. package/dist/src/evals/trial-execution-evidence.js +13 -0
  366. package/dist/src/evals/trial-execution-evidence.js.map +1 -0
  367. package/dist/src/evals/trial-import-identity.js +31 -0
  368. package/dist/src/evals/trial-import-identity.js.map +1 -0
  369. package/dist/src/evals/trial-import.js +183 -188
  370. package/dist/src/evals/trial-import.js.map +1 -1
  371. package/dist/src/evals/trial-publication-recovery.js +144 -0
  372. package/dist/src/evals/trial-publication-recovery.js.map +1 -0
  373. package/dist/src/evals/trial-publication.js +25 -0
  374. package/dist/src/evals/trial-publication.js.map +1 -0
  375. package/dist/src/evals/trial-reference-validation.js +49 -0
  376. package/dist/src/evals/trial-reference-validation.js.map +1 -0
  377. package/dist/src/evals/verifier-artifacts.js +317 -0
  378. package/dist/src/evals/verifier-artifacts.js.map +1 -0
  379. package/dist/src/evals/verifier-eligibility.js +25 -0
  380. package/dist/src/evals/verifier-eligibility.js.map +1 -0
  381. package/dist/src/evals/verifier-only-rerun.js +210 -0
  382. package/dist/src/evals/verifier-only-rerun.js.map +1 -0
  383. package/dist/src/evals/verifier-runtime.js +61 -0
  384. package/dist/src/evals/verifier-runtime.js.map +1 -0
  385. package/dist/src/evals/verifier-score-artifacts.js +119 -0
  386. package/dist/src/evals/verifier-score-artifacts.js.map +1 -0
  387. package/dist/src/evals/work-item-artifacts.js +23 -0
  388. package/dist/src/evals/work-item-artifacts.js.map +1 -0
  389. package/dist/src/foundation/config.js +17 -0
  390. package/dist/src/foundation/config.js.map +1 -1
  391. package/dist/src/foundation/contained-file.js +85 -0
  392. package/dist/src/foundation/contained-file.js.map +1 -0
  393. package/dist/src/foundation/credential-redaction.js +162 -0
  394. package/dist/src/foundation/credential-redaction.js.map +1 -0
  395. package/dist/src/foundation/environment-image-references.js +59 -0
  396. package/dist/src/foundation/environment-image-references.js.map +1 -0
  397. package/dist/src/foundation/executable.js +45 -5
  398. package/dist/src/foundation/executable.js.map +1 -1
  399. package/dist/src/foundation/fs.js +2 -1
  400. package/dist/src/foundation/fs.js.map +1 -1
  401. package/dist/src/foundation/index.js +6 -2
  402. package/dist/src/foundation/index.js.map +1 -1
  403. package/dist/src/foundation/process-identity.js +69 -0
  404. package/dist/src/foundation/process-identity.js.map +1 -0
  405. package/dist/src/images/context.js +47 -0
  406. package/dist/src/images/context.js.map +1 -0
  407. package/dist/src/images/docker-buildkit.js +137 -0
  408. package/dist/src/images/docker-buildkit.js.map +1 -0
  409. package/dist/src/images/dockerfile.js +67 -0
  410. package/dist/src/images/dockerfile.js.map +1 -0
  411. package/dist/src/images/gc.js +261 -0
  412. package/dist/src/images/gc.js.map +1 -0
  413. package/dist/src/images/index.js +10 -0
  414. package/dist/src/images/index.js.map +1 -0
  415. package/dist/src/images/manifest.js +109 -0
  416. package/dist/src/images/manifest.js.map +1 -0
  417. package/dist/src/images/ownership.js +5 -0
  418. package/dist/src/images/ownership.js.map +1 -0
  419. package/dist/src/images/records.js +32 -0
  420. package/dist/src/images/records.js.map +1 -0
  421. package/dist/src/images/registry.js +154 -0
  422. package/dist/src/images/registry.js.map +1 -0
  423. package/dist/src/images/service.js +224 -0
  424. package/dist/src/images/service.js.map +1 -0
  425. package/dist/src/model-access/capture.js +255 -0
  426. package/dist/src/model-access/capture.js.map +1 -0
  427. package/dist/src/model-access/index.js +5 -0
  428. package/dist/src/model-access/index.js.map +1 -0
  429. package/dist/src/model-access/policy.js +49 -0
  430. package/dist/src/model-access/policy.js.map +1 -0
  431. package/dist/src/model-access/proxy.js +336 -0
  432. package/dist/src/model-access/proxy.js.map +1 -0
  433. package/dist/src/model-access/records.js +167 -0
  434. package/dist/src/model-access/records.js.map +1 -0
  435. package/dist/src/runs/adapter-process.js +12 -0
  436. package/dist/src/runs/adapter-process.js.map +1 -0
  437. package/dist/src/runs/bundle.js +450 -0
  438. package/dist/src/runs/bundle.js.map +1 -0
  439. package/dist/src/runs/events.js +12 -1
  440. package/dist/src/runs/events.js.map +1 -1
  441. package/dist/src/runs/executor.js +46 -45
  442. package/dist/src/runs/executor.js.map +1 -1
  443. package/dist/src/runs/finalizer.js +17 -0
  444. package/dist/src/runs/finalizer.js.map +1 -1
  445. package/dist/src/runs/index.js +8 -1
  446. package/dist/src/runs/index.js.map +1 -1
  447. package/dist/src/runs/manifest.js +5 -5
  448. package/dist/src/runs/manifest.js.map +1 -1
  449. package/dist/src/runs/phase-bundle.js +86 -0
  450. package/dist/src/runs/phase-bundle.js.map +1 -0
  451. package/dist/src/runs/phase-cancellation.js +107 -0
  452. package/dist/src/runs/phase-cancellation.js.map +1 -0
  453. package/dist/src/runs/phase-group.js +113 -0
  454. package/dist/src/runs/phase-group.js.map +1 -0
  455. package/dist/src/runs/query.js +1 -1
  456. package/dist/src/runs/query.js.map +1 -1
  457. package/dist/src/runs/records.js +3 -0
  458. package/dist/src/runs/records.js.map +1 -1
  459. package/dist/src/runs/request.js +10 -3
  460. package/dist/src/runs/request.js.map +1 -1
  461. package/dist/src/runs/training-candidate.js +161 -0
  462. package/dist/src/runs/training-candidate.js.map +1 -0
  463. package/dist/src/runs/verifier-evidence-redaction.js +61 -0
  464. package/dist/src/runs/verifier-evidence-redaction.js.map +1 -0
  465. package/dist/src/runs/verifier-evidence.js +459 -0
  466. package/dist/src/runs/verifier-evidence.js.map +1 -0
  467. package/dist/src/runs/verifier-structured-evidence.js +83 -0
  468. package/dist/src/runs/verifier-structured-evidence.js.map +1 -0
  469. package/dist/src/trajectories/analysis.js +288 -0
  470. package/dist/src/trajectories/analysis.js.map +1 -0
  471. package/dist/src/trajectories/chunk-projection.js +148 -0
  472. package/dist/src/trajectories/chunk-projection.js.map +1 -0
  473. package/dist/src/trajectories/content-projection.js +389 -0
  474. package/dist/src/trajectories/content-projection.js.map +1 -0
  475. package/dist/src/trajectories/dsh-chunk-contract.js +176 -0
  476. package/dist/src/trajectories/dsh-chunk-contract.js.map +1 -0
  477. package/dist/src/trajectories/dsh-contract.js +284 -0
  478. package/dist/src/trajectories/dsh-contract.js.map +1 -0
  479. package/dist/src/trajectories/events-chunk-drill.js +95 -0
  480. package/dist/src/trajectories/events-chunk-drill.js.map +1 -0
  481. package/dist/src/trajectories/events-page.js +367 -0
  482. package/dist/src/trajectories/events-page.js.map +1 -0
  483. package/dist/src/trajectories/index.js +3 -0
  484. package/dist/src/trajectories/index.js.map +1 -1
  485. package/dist/src/trajectories/provider-capture.js +15 -24
  486. package/dist/src/trajectories/provider-capture.js.map +1 -1
  487. package/dist/src/trajectories/providers/deepseek.js +3 -3
  488. package/dist/src/trajectories/providers/deepseek.js.map +1 -1
  489. package/dist/src/trajectories/request-attempt.js +47 -0
  490. package/dist/src/trajectories/request-attempt.js.map +1 -0
  491. package/dist/src/trajectories/stream-reader.js +311 -0
  492. package/dist/src/trajectories/stream-reader.js.map +1 -0
  493. package/dist/src/trajectories/surface-fold.js +126 -0
  494. package/dist/src/trajectories/surface-fold.js.map +1 -0
  495. package/dist/src/workers/index.js +3 -0
  496. package/dist/src/workers/index.js.map +1 -0
  497. package/dist/src/workers/remote-harbor-work-spec.js +97 -0
  498. package/dist/src/workers/remote-harbor-work-spec.js.map +1 -0
  499. package/dist/src/workers/remote-harbor-worker.js +223 -0
  500. package/dist/src/workers/remote-harbor-worker.js.map +1 -0
  501. package/docs/schemas/benchmark-hook-request-v1.schema.json +65 -0
  502. package/docs/schemas/benchmark-hook-response-v1.schema.json +75 -0
  503. package/docs/schemas/benchmark-hook-v1.schema.json +33 -0
  504. package/docs/schemas/benchmark-lock-v1.schema.json +351 -0
  505. package/docs/schemas/benchmark-metric-v1.schema.json +37 -0
  506. package/docs/schemas/benchmark-package-v1.schema.json +167 -0
  507. package/docs/schemas/benchmark-phase-group.schema.json +41 -0
  508. package/docs/schemas/benchmark-profile-v1.schema.json +141 -0
  509. package/docs/schemas/benchmark-task-v1.schema.json +384 -0
  510. package/docs/schemas/benchmark-tool-result-v1.schema.json +30 -0
  511. package/docs/schemas/environment-build-index.schema.json +13 -0
  512. package/docs/schemas/environment-build-record.schema.json +25 -0
  513. package/docs/schemas/environment-image-manifest.schema.json +62 -0
  514. package/docs/schemas/error.schema.json +22 -0
  515. package/docs/schemas/eval-control.schema.json +36 -0
  516. package/docs/schemas/eval-execution-plan.schema.json +216 -0
  517. package/docs/schemas/eval-progress.schema.json +1 -15
  518. package/docs/schemas/eval-rerun-result.schema.json +58 -0
  519. package/docs/schemas/eval-rerun-state.schema.json +61 -0
  520. package/docs/schemas/eval-rerun-submission.schema.json +60 -0
  521. package/docs/schemas/eval-result.schema.json +32 -13
  522. package/docs/schemas/eval-submission-input.schema.json +14 -0
  523. package/docs/schemas/eval-submission.schema.json +68 -0
  524. package/docs/schemas/eval-trial-publication.schema.json +15 -0
  525. package/docs/schemas/eval-trial-reference.schema.json +129 -0
  526. package/docs/schemas/execution-evidence.schema.json +96 -0
  527. package/docs/schemas/execution-lease.schema.json +50 -0
  528. package/docs/schemas/execution-provider-status.schema.json +65 -0
  529. package/docs/schemas/execution-worker.schema.json +53 -0
  530. package/docs/schemas/harbor-process-exit.schema.json +14 -0
  531. package/docs/schemas/interaction-capture-ref.schema.json +45 -0
  532. package/docs/schemas/local-provider-execution.schema.json +33 -0
  533. package/docs/schemas/model-interaction.schema.json +41 -0
  534. package/docs/schemas/regrade-assessment-reference.schema.json +11 -0
  535. package/docs/schemas/remote-result-envelope.schema.json +32 -0
  536. package/docs/schemas/remote-tree-envelope.schema.json +25 -0
  537. package/docs/schemas/remote-work-offer.schema.json +58 -0
  538. package/docs/schemas/remote-work-receipt.schema.json +50 -0
  539. package/docs/schemas/remote-worker-artifact.schema.json +17 -0
  540. package/docs/schemas/remote-worker-event.schema.json +18 -0
  541. package/docs/schemas/remote-worker-heartbeat.schema.json +37 -0
  542. package/docs/schemas/remote-worker-registration.schema.json +52 -0
  543. package/docs/schemas/result-bundle-index.schema.json +133 -0
  544. package/docs/schemas/run-context.schema.json +15 -0
  545. package/docs/schemas/run-environment-images.schema.json +39 -0
  546. package/docs/schemas/run-manifest.schema.json +4 -0
  547. package/docs/schemas/run-request.schema.json +4 -0
  548. package/docs/schemas/training-data-candidate.schema.json +46 -0
  549. package/docs/schemas/trajectory-analysis.schema.json +258 -0
  550. package/docs/schemas/trajectory-events-page.schema.json +40 -0
  551. package/docs/schemas/verifier-assessment.schema.json +77 -0
  552. package/docs/schemas/verifier-evidence.schema.json +276 -0
  553. package/integrations/harbor/hitch_benchmark.py +252 -0
  554. package/integrations/harbor/hitch_candidate_recycle.py +262 -0
  555. package/integrations/harbor/hitch_harbor_agent.py +698 -568
  556. package/integrations/harbor/hitch_harbor_environment.py +469 -0
  557. package/integrations/harbor/hitch_harbor_task_resources.py +275 -0
  558. package/integrations/harbor/hitch_harbor_verifier.py +83 -2
  559. package/integrations/harbor/hitch_phase_supervisor.py +452 -0
  560. package/integrations/harbor/hitch_tool_client.mjs +91 -0
  561. package/integrations/model-call/cli.js +57 -0
  562. package/package.json +30 -6
@@ -0,0 +1,242 @@
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 { replaceInvalidEvalProgressTrial } from "./progress.js";
9
+ import { assertBackendTrialSet } from "./result-helpers.js";
10
+ import { importEvalTrialRuns, validateEvalTrialReferences } from "./trial-import.js";
11
+ import { startEvalModelCaptureRuntime } from "./model-capture-runtime.js";
12
+ import { readModelProxyRuntimeState } from "./model-proxy-runtime-state.js";
13
+ import { ensurePhysicalRetryDecision, readEvalRetryState, resolveRetryWork, transitionRetryDecision } from "./retry-state.js";
14
+ import { classifyTrialFailure, physicalRetryAllowed } from "./failure-classifier.js";
15
+ import { physicalRetryWorkItem } from "./physical-retry-work.js";
16
+ import { retryBackoffMs } from "./retry-backoff.js";
17
+ export async function recoverLocalDockerEvalLeases(input) {
18
+ const emit = input.emit ?? (() => { });
19
+ const recovered = [];
20
+ let failure;
21
+ let captureRuntime;
22
+ try {
23
+ captureRuntime = await restoreModelCaptureForRecovery(input);
24
+ for (const lease of input.leases) {
25
+ if (!activeLease(lease))
26
+ continue;
27
+ try {
28
+ await recoverLease(input, lease, emit, captureRuntime);
29
+ recovered.push(lease.lease_id);
30
+ }
31
+ catch (error) {
32
+ const typed = error instanceof HitchError;
33
+ failure ??= {
34
+ code: typed ? error.code : "execution_state_ambiguous",
35
+ message: error?.message || String(error),
36
+ };
37
+ const current = (await readExecutionLeases(input.evalDirectory)).find((entry) => entry.lease_id === lease.lease_id);
38
+ if (current && activeLease(current))
39
+ await markExecutionLeaseLost({
40
+ evalDirectory: input.evalDirectory,
41
+ leaseId: current.lease_id,
42
+ expectedEpoch: current.epoch,
43
+ });
44
+ emit({ type: "eval.lease.recovery-failed", lease_id: lease.lease_id, code: failure.code });
45
+ emit({ type: "eval.work.lost", work_id: lease.work_id, lease_id: lease.lease_id, code: failure.code });
46
+ }
47
+ }
48
+ }
49
+ catch (error) {
50
+ failure = { code: "execution_state_ambiguous", message: error?.message || String(error) };
51
+ for (const lease of input.leases.filter(activeLease)) {
52
+ const current = (await readExecutionLeases(input.evalDirectory)).find((entry) => entry.lease_id === lease.lease_id);
53
+ if (current && activeLease(current))
54
+ await markExecutionLeaseLost({ evalDirectory: input.evalDirectory, leaseId: current.lease_id, expectedEpoch: current.epoch });
55
+ emit({ type: "eval.lease.recovery-failed", lease_id: lease.lease_id, code: failure.code });
56
+ emit({ type: "eval.work.lost", work_id: lease.work_id, lease_id: lease.lease_id, code: failure.code });
57
+ }
58
+ }
59
+ finally {
60
+ await captureRuntime?.close().catch(() => undefined);
61
+ }
62
+ return failure
63
+ ? { status: "ambiguous", recovered_lease_ids: recovered, ...failure }
64
+ : { status: "resumable", recovered_lease_ids: recovered };
65
+ }
66
+ async function recoverLease(input, lease, emit, captureRuntime) {
67
+ if (Date.parse(lease.expires_at) <= Date.now())
68
+ emit({ type: "lease.expired", work_id: lease.work_id, lease_id: lease.lease_id, lease_epoch: lease.epoch, worker_id: lease.worker_id });
69
+ if (lease.provider !== "local-docker")
70
+ throw ambiguous(`unsupported recovery provider: ${lease.provider}`);
71
+ const provider = new LocalDockerExecutionProvider({
72
+ root: input.root,
73
+ workerId: lease.worker_id,
74
+ status: () => { throw new Error("recovery does not inspect provider capacity"); },
75
+ });
76
+ const record = await readLocalDockerProcessRecordByLease({ root: input.root, evalId: input.evalId, leaseId: lease.lease_id }).catch(() => {
77
+ throw ambiguous(`local provider has no durable process identity for ${lease.lease_id}`);
78
+ });
79
+ if (record.lease_epoch === lease.epoch - 1)
80
+ await provider.adoptLeaseEpoch(lease.lease_id, record.lease_epoch, lease.epoch);
81
+ else if (record.lease_epoch !== lease.epoch)
82
+ throw ambiguous(`local provider epoch does not match ${lease.lease_id}`);
83
+ const probe = await provider.recover(lease);
84
+ emit({ type: "eval.lease.recovery-probed", lease_id: lease.lease_id, lease_epoch: lease.epoch, state: probe.state });
85
+ if (probe.state === "released") {
86
+ await releaseExecutionLease({ evalDirectory: input.evalDirectory, leaseId: lease.lease_id, expectedEpoch: lease.epoch });
87
+ emit({ type: "lease.recovered", work_id: lease.work_id, lease_id: lease.lease_id, lease_epoch: lease.epoch, state: "already-released" });
88
+ return;
89
+ }
90
+ if (probe.state !== "running" && probe.state !== "terminal-uncollected") {
91
+ throw ambiguous(`local provider classified ${lease.lease_id} as ${probe.state}`);
92
+ }
93
+ const reissued = await reissueExecutionLease({ evalDirectory: input.evalDirectory, leaseId: lease.lease_id, expectedEpoch: lease.epoch });
94
+ await provider.adoptLeaseEpoch(lease.lease_id, lease.epoch, reissued.epoch);
95
+ emit({ type: "eval.lease.reissued", lease_id: lease.lease_id, previous_epoch: lease.epoch, lease_epoch: reissued.epoch });
96
+ let heartbeatFailure;
97
+ let heartbeatTail = Promise.resolve();
98
+ const timer = setInterval(() => {
99
+ heartbeatTail = heartbeatTail.then(async () => {
100
+ await heartbeatExecutionLease({ evalDirectory: input.evalDirectory, leaseId: lease.lease_id, expectedEpoch: reissued.epoch });
101
+ }).catch((error) => { heartbeatFailure ??= error; });
102
+ }, DEFAULT_EXECUTION_LEASE_HEARTBEAT_MS);
103
+ timer.unref();
104
+ try {
105
+ if (input.cancelRequested && probe.state === "running")
106
+ await provider.cancel(lease.lease_id, reissued.epoch);
107
+ const terminal = await waitForLocalDockerProcessTerminal({ root: input.root, leaseId: lease.lease_id, epoch: reissued.epoch });
108
+ await heartbeatTail;
109
+ if (heartbeatFailure !== undefined)
110
+ throw heartbeatFailure;
111
+ try {
112
+ await collectRecoveredWork(input, reissued, terminal.backend_directory, emit, captureRuntime);
113
+ }
114
+ catch (error) {
115
+ if (!input.cancelRequested || error.code !== "recovery_collection_missing")
116
+ throw error;
117
+ }
118
+ if (!input.cancelRequested && terminal.process_exit_code !== null && terminal.process_exit_code !== 0) {
119
+ throw new HitchError(`recovered Harbor process exited with code ${terminal.process_exit_code}`, { code: "harbor_failed", exitCode: 13 });
120
+ }
121
+ await provider.release(lease.lease_id, reissued.epoch);
122
+ await releaseExecutionLease({ evalDirectory: input.evalDirectory, leaseId: lease.lease_id, expectedEpoch: reissued.epoch });
123
+ emit({ type: "lease.released", work_id: lease.work_id, lease_id: lease.lease_id, lease_epoch: reissued.epoch, worker_id: lease.worker_id });
124
+ emit({ type: "eval.lease.recovered", lease_id: lease.lease_id, lease_epoch: reissued.epoch, state: "released" });
125
+ emit({ type: "lease.recovered", lease_id: lease.lease_id, lease_epoch: reissued.epoch, state: "released" });
126
+ }
127
+ finally {
128
+ clearInterval(timer);
129
+ await heartbeatTail;
130
+ }
131
+ }
132
+ async function collectRecoveredWork(input, lease, backendDirectory, emit, captureRuntime) {
133
+ const state = await loadEvalResumeState(input.evalDirectory);
134
+ const retryState = await readEvalRetryState(input.evalDirectory, input.evalId);
135
+ const dynamic = resolveRetryWork(state.executionPlan, retryState, lease.work_id);
136
+ const work = state.executionPlan.work_items.find((item) => item.work_id === lease.work_id) ?? dynamic?.item;
137
+ if (!work || work.logical_attempt === null || work.task_ids.length !== 1)
138
+ throw ambiguous("recovered lease does not match one planned task slot");
139
+ const existing = state.progress.trials.find((trial) => trial.task_id === work.task_ids[0] && trial.attempt === work.logical_attempt);
140
+ if (dynamic && existing?.observation_status === "valid") {
141
+ for (const decision of dynamic.decisions)
142
+ await settleRecoveredRetryDecision(input.evalDirectory, input.evalId, decision.decision_id, decision.state, "repaired");
143
+ emit({ type: "eval.retry.recovered", work_id: work.work_id, lease_id: lease.lease_id, state: "repaired", collection: "already-published" });
144
+ return;
145
+ }
146
+ const jobDirectory = path.join(input.evalDirectory, backendDirectory, "job");
147
+ const rawResult = await readHarborRawResult(jobDirectory);
148
+ if (!rawResult)
149
+ throw new HitchError("recovered Harbor work item has no aggregate result", { code: "recovery_collection_missing", exitCode: 12 });
150
+ const resolvedRevision = await readJSON(path.join(input.evalDirectory, "resolution.json"));
151
+ const request = await readJSON(path.join(input.evalDirectory, "request.json"));
152
+ const runtime = state.plan.controller_runtime;
153
+ const runtimeId = typeof runtime?.runtime_id === "string" ? runtime.runtime_id : undefined;
154
+ const refs = await importEvalTrialRuns({
155
+ root: input.root,
156
+ evalId: input.evalId,
157
+ evalDirectory: input.evalDirectory,
158
+ harborJobDirectory: jobDirectory,
159
+ expectedAttempt: work.logical_attempt,
160
+ request,
161
+ resolvedRevision,
162
+ benchmarkId: state.progress.benchmark_id,
163
+ benchmarkRevision: state.progress.benchmark_revision,
164
+ ...(runtimeId ? { runtimeId } : {}),
165
+ env: input.env ?? process.env,
166
+ ...(state.executionPlan.model_capture ? { modelCapturePlan: captureRuntime?.plan ?? state.executionPlan.model_capture } : {}),
167
+ ...(captureRuntime?.exporter ? { interactionCaptureExporter: captureRuntime.exporter } : {}),
168
+ ...(dynamic ? { publicationMode: "replace-invalid" } : {}),
169
+ rawResult,
170
+ }, dynamic ? [] : state.progress.trials);
171
+ const workRefs = refs.filter((ref) => ref.task_id === work.task_ids[0] && ref.attempt === work.logical_attempt);
172
+ assertBackendTrialSet(rawResult, workRefs);
173
+ await validateEvalTrialReferences(input.root, input.evalId, workRefs, {
174
+ benchmarkId: state.progress.benchmark_id,
175
+ benchmarkRevision: state.progress.benchmark_revision,
176
+ });
177
+ let progress = state.progress;
178
+ for (const ref of workRefs) {
179
+ if (dynamic) {
180
+ if (ref.observation_status === "valid")
181
+ progress = replaceInvalidEvalProgressTrial(progress, ref);
182
+ }
183
+ else
184
+ progress = mergeEvalProgressTrial(progress, ref);
185
+ }
186
+ if (progress.generation !== state.progress.generation)
187
+ await writeEvalProgress(input.evalDirectory, progress);
188
+ if (dynamic) {
189
+ const repaired = workRefs.some((ref) => ref.observation_status === "valid");
190
+ const retryable = workRefs.filter((ref) => physicalRetryAllowed(classifyTrialFailure(ref)));
191
+ const retryIndex = dynamic.decisions[0]?.retry_index ?? 0;
192
+ if (!repaired && retryable.length > 0 && retryIndex < state.executionPlan.retry_policy.infrastructure_retries) {
193
+ const origin = state.executionPlan.work_items.find((item) => item.slots.includes(dynamic.decisions[0].slot_id));
194
+ if (!origin)
195
+ throw ambiguous("recovered retry origin is absent from execution plan");
196
+ const nextWork = physicalRetryWorkItem(origin, retryIndex + 1, retryable);
197
+ for (const trigger of retryable)
198
+ await ensurePhysicalRetryDecision({
199
+ evalDirectory: input.evalDirectory, evalId: input.evalId, item: origin, retryIndex: retryIndex + 1, trigger,
200
+ notBefore: new Date(Date.now() + retryBackoffMs(state.executionPlan.retry_policy.infrastructure_retry_backoff_ms, retryIndex + 1, nextWork.work_id)).toISOString(),
201
+ });
202
+ }
203
+ const target = repaired ? "repaired" : retryIndex >= state.executionPlan.retry_policy.infrastructure_retries ? "exhausted" : "invalid";
204
+ for (const decision of dynamic.decisions)
205
+ await settleRecoveredRetryDecision(input.evalDirectory, input.evalId, decision.decision_id, decision.state, target);
206
+ }
207
+ emit({ type: "eval.work-item.recovered", work_id: work.work_id, lease_id: lease.lease_id, trials: workRefs.length });
208
+ }
209
+ async function settleRecoveredRetryDecision(evalDirectory, evalId, decisionId, current, target) {
210
+ if (current === target)
211
+ return;
212
+ if (current === "planned") {
213
+ await transitionRetryDecision({ evalDirectory, evalId, decisionId, state: "running" });
214
+ current = "running";
215
+ }
216
+ if (current === "running")
217
+ await transitionRetryDecision({ evalDirectory, evalId, decisionId, state: target });
218
+ }
219
+ async function restoreModelCaptureForRecovery(input) {
220
+ if (!input.leases.some(activeLease))
221
+ return undefined;
222
+ const state = await loadEvalResumeState(input.evalDirectory);
223
+ const plan = state.executionPlan.model_capture;
224
+ if (!plan || plan.effective_mode !== "proxy" && plan.effective_mode !== "hybrid")
225
+ return undefined;
226
+ if (!await readModelProxyRuntimeState(input.evalDirectory, input.evalId, plan)) {
227
+ throw ambiguous("recoverable model proxy has no persisted endpoint identity");
228
+ }
229
+ const runtime = await startEvalModelCaptureRuntime({ plan, evalId: input.evalId, evalDirectory: input.evalDirectory, env: input.env ?? process.env });
230
+ if (!runtime.route || !runtime.exporter) {
231
+ await runtime.close().catch(() => undefined);
232
+ throw ambiguous("recoverable model proxy endpoint could not be restored");
233
+ }
234
+ return runtime;
235
+ }
236
+ function activeLease(lease) {
237
+ return lease.state === "offered" || lease.state === "accepted" || lease.state === "running" || lease.state === "releasing";
238
+ }
239
+ function ambiguous(message) {
240
+ return new HitchError(message, { code: "execution_state_ambiguous", exitCode: 12 });
241
+ }
242
+ //# 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,+BAA+B,EAAE,MAAM,eAAe,CAAC;AAChE,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;AAC5E,OAAO,EAAE,2BAA2B,EAAE,kBAAkB,EAAE,gBAAgB,EAAE,uBAAuB,EAAE,MAAM,kBAAkB,CAAC;AAC9H,OAAO,EAAE,oBAAoB,EAAE,oBAAoB,EAAE,MAAM,yBAAyB,CAAC;AACrF,OAAO,EAAE,qBAAqB,EAAE,MAAM,0BAA0B,CAAC;AACjE,OAAO,EAAE,cAAc,EAAE,MAAM,oBAAoB,CAAC;AASpD,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,UAAU,GAAG,MAAM,kBAAkB,CAAC,KAAK,CAAC,aAAa,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IAC/E,MAAM,OAAO,GAAG,gBAAgB,CAAC,KAAK,CAAC,aAAa,EAAE,UAAU,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;IACjF,MAAM,IAAI,GAAG,KAAK,CAAC,aAAa,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,OAAO,KAAK,KAAK,CAAC,OAAO,CAAC,IAAI,OAAO,EAAE,IAAI,CAAC;IAC5G,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,QAAQ,GAAG,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,OAAO,KAAK,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,OAAO,KAAK,IAAI,CAAC,eAAe,CAAC,CAAC;IACrI,IAAI,OAAO,IAAI,QAAQ,EAAE,kBAAkB,KAAK,OAAO,EAAE,CAAC;QACxD,KAAK,MAAM,QAAQ,IAAI,OAAO,CAAC,SAAS;YAAE,MAAM,4BAA4B,CAAC,KAAK,CAAC,aAAa,EAAE,KAAK,CAAC,MAAM,EAAE,QAAQ,CAAC,WAAW,EAAE,QAAQ,CAAC,KAAK,EAAE,UAAU,CAAC,CAAC;QAClK,IAAI,CAAC,EAAE,IAAI,EAAE,sBAAsB,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,KAAK,EAAE,UAAU,EAAE,UAAU,EAAE,mBAAmB,EAAE,CAAC,CAAC;QAC5I,OAAO;IACT,CAAC;IACD,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,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,iBAA0B,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACnE,SAAS;KACV,EAAE,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;IACzC,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,EAAE,CAAC;QAC3B,IAAI,OAAO,EAAE,CAAC;YACZ,IAAI,GAAG,CAAC,kBAAkB,KAAK,OAAO;gBAAE,QAAQ,GAAG,+BAA+B,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;QACpG,CAAC;;YAAM,QAAQ,GAAG,sBAAsB,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;IAC1D,CAAC;IACD,IAAI,QAAQ,CAAC,UAAU,KAAK,KAAK,CAAC,QAAQ,CAAC,UAAU;QAAE,MAAM,iBAAiB,CAAC,KAAK,CAAC,aAAa,EAAE,QAAQ,CAAC,CAAC;IAC9G,IAAI,OAAO,EAAE,CAAC;QACZ,MAAM,QAAQ,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,kBAAkB,KAAK,OAAO,CAAC,CAAC;QAC5E,MAAM,SAAS,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,oBAAoB,CAAC,oBAAoB,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QAC5F,MAAM,UAAU,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,WAAW,IAAI,CAAC,CAAC;QAC1D,IAAI,CAAC,QAAQ,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC,IAAI,UAAU,GAAG,KAAK,CAAC,aAAa,CAAC,YAAY,CAAC,sBAAsB,EAAE,CAAC;YAC9G,MAAM,MAAM,GAAG,KAAK,CAAC,aAAa,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAE,CAAC,OAAO,CAAC,CAAC,CAAC;YACjH,IAAI,CAAC,MAAM;gBAAE,MAAM,SAAS,CAAC,sDAAsD,CAAC,CAAC;YACrF,MAAM,QAAQ,GAAG,qBAAqB,CAAC,MAAM,EAAE,UAAU,GAAG,CAAC,EAAE,SAAS,CAAC,CAAC;YAC1E,KAAK,MAAM,OAAO,IAAI,SAAS;gBAAE,MAAM,2BAA2B,CAAC;oBACjE,aAAa,EAAE,KAAK,CAAC,aAAa,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,UAAU,EAAE,UAAU,GAAG,CAAC,EAAE,OAAO;oBAC3G,SAAS,EAAE,IAAI,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,cAAc,CAAC,KAAK,CAAC,aAAa,CAAC,YAAY,CAAC,+BAA+B,EAAE,UAAU,GAAG,CAAC,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,WAAW,EAAE;iBACnK,CAAC,CAAC;QACL,CAAC;QACD,MAAM,MAAM,GAAG,QAAQ,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,UAAU,IAAI,KAAK,CAAC,aAAa,CAAC,YAAY,CAAC,sBAAsB,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;QACvI,KAAK,MAAM,QAAQ,IAAI,OAAO,CAAC,SAAS;YAAE,MAAM,4BAA4B,CAAC,KAAK,CAAC,aAAa,EAAE,KAAK,CAAC,MAAM,EAAE,QAAQ,CAAC,WAAW,EAAE,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;IAChK,CAAC;IACD,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,4BAA4B,CACzC,aAAqB,EACrB,MAAc,EACd,UAAkB,EAClB,OAAiF,EACjF,MAA4C;IAE5C,IAAI,OAAO,KAAK,MAAM;QAAE,OAAO;IAC/B,IAAI,OAAO,KAAK,SAAS,EAAE,CAAC;QAC1B,MAAM,uBAAuB,CAAC,EAAE,aAAa,EAAE,MAAM,EAAE,UAAU,EAAE,KAAK,EAAE,SAAS,EAAE,CAAC,CAAC;QACvF,OAAO,GAAG,SAAS,CAAC;IACtB,CAAC;IACD,IAAI,OAAO,KAAK,SAAS;QAAE,MAAM,uBAAuB,CAAC,EAAE,aAAa,EAAE,MAAM,EAAE,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;AACjH,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 +1,77 @@
1
+ import { createHash } from "node:crypto";
2
+ import { createReadStream } from "node:fs";
3
+ import { lstat, readFile, readdir } from "node:fs/promises";
4
+ import path from "node:path";
5
+ import { validateRunObservation } from "../domain/index.js";
6
+ import { atomicWriteJSON, sha256Bytes, sha256JSON, statePaths } from "../foundation/index.js";
7
+ import { verifyResultBundleIndex } from "../runs/index.js";
8
+ /** A bounded inventory includes empty directories and rejects symlinks. */
9
+ export async function regradeTreeDigest(directory) {
10
+ const files = [];
11
+ let total = 0;
12
+ async function visit(relative) {
13
+ const absolute = path.join(directory, relative);
14
+ const info = await lstat(absolute);
15
+ if (info.isSymbolicLink() || (!info.isFile() && !info.isDirectory()))
16
+ throw new Error("regrade input contains a symlink or special file");
17
+ if (files.length >= 100_000 || (total += info.isFile() ? info.size : 0) > 4 * 1024 ** 3)
18
+ throw new Error("regrade input exceeds inventory limits");
19
+ if (info.isDirectory()) {
20
+ files.push({ path: relative, type: "directory" });
21
+ for (const name of (await readdir(absolute)).sort())
22
+ await visit(relative ? `${relative}/${name}` : name);
23
+ }
24
+ else {
25
+ const hash = createHash("sha256");
26
+ for await (const chunk of createReadStream(absolute))
27
+ hash.update(chunk);
28
+ files.push({ path: relative, bytes: info.size, mode: info.mode & 0o777, sha256: `sha256:${hash.digest("hex")}` });
29
+ }
30
+ }
31
+ await visit("");
32
+ return sha256JSON(files);
33
+ }
34
+ export async function sealRegradeAssessment(directory, record) {
35
+ const evidenceDigest = await regradeTreeDigest(path.join(directory, "evidence"));
36
+ const manifest = { ...record, schema_version: "1", kind: "verifier-only-assessment", candidate_executes: false, evidence_digest: evidenceDigest };
37
+ await atomicWriteJSON(path.join(directory, "assessment.json"), manifest);
38
+ return { id: path.basename(directory), digest: sha256Bytes(await readFile(path.join(directory, "assessment.json"))) };
39
+ }
40
+ export async function readRegradeObservation(root, evalId, trial) {
41
+ if (trial.run_group)
42
+ throw new Error("phase groups require their native phase assessment");
43
+ const reference = trial.assessment;
44
+ if (!reference || !/^assessment_[a-f0-9]{32}$/.test(reference.id) || !/^sha256:[a-f0-9]{64}$/.test(reference.digest))
45
+ throw new Error("invalid regrade assessment reference");
46
+ const directory = path.join(statePaths(root).evals, evalId, "assessments", reference.id);
47
+ const raw = await readFile(path.join(directory, "assessment.json"));
48
+ if (sha256Bytes(raw) !== reference.digest)
49
+ throw new Error("regrade assessment digest mismatch");
50
+ const record = JSON.parse(raw.toString());
51
+ const source = record.source;
52
+ if (record.schema_version !== "1" || record.kind !== "verifier-only-assessment" || record.candidate_executes !== false
53
+ || record.eval_id !== evalId || record.task_id !== trial.task_id || record.attempt !== trial.attempt
54
+ || source?.trial_id !== trial.trial_id || source.run_id !== trial.run_id)
55
+ throw new Error("regrade assessment source identity mismatch");
56
+ if (record.evidence_digest !== await regradeTreeDigest(path.join(directory, "evidence")))
57
+ throw new Error("regrade assessment evidence changed");
58
+ if (record.runtime_repair !== undefined) {
59
+ const repair = record.runtime_repair;
60
+ const receipt = JSON.parse(await readFile(path.join(directory, "evidence/runtime-repair.json"), "utf8"));
61
+ if (!repair || typeof repair !== "object" || repair.schema_version !== "1" || repair.kind !== "verifier-environment-runtime-repair"
62
+ || repair.path !== "integrations/harbor/hitch_harbor_environment.py"
63
+ || repair.source_runtime_id !== source.controller_runtime_id || repair.replacement_runtime_id !== record.controller_runtime_id
64
+ || ![repair.source_runtime_id, repair.replacement_runtime_id, repair.source_sha256, repair.replacement_sha256].every(value => typeof value === "string" && /^sha256:[a-f0-9]{64}$/.test(value))
65
+ || !Number.isSafeInteger(repair.unchanged_file_count) || Number(repair.unchanged_file_count) < 0
66
+ || sha256JSON(repair) !== sha256JSON(receipt))
67
+ throw new Error("verifier runtime repair provenance mismatch");
68
+ }
69
+ else if (source.controller_runtime_id !== undefined && source.controller_runtime_id !== record.controller_runtime_id) {
70
+ throw new Error("verifier controller runtime changed without repair provenance");
71
+ }
72
+ const sourceIndex = await verifyResultBundleIndex(path.join(statePaths(root).runs, trial.run_id));
73
+ if (source.bundle_index_digest !== sha256JSON(sourceIndex))
74
+ throw new Error("regrade source run bundle changed");
75
+ return validateRunObservation(record.observation);
76
+ }
77
+ //# sourceMappingURL=regrade-evidence.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"regrade-evidence.js","sourceRoot":"","sources":["../../../src/evals/regrade-evidence.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,gBAAgB,EAAE,MAAM,SAAS,CAAC;AAC3C,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,kBAAkB,CAAC;AAC5D,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,sBAAsB,EAAE,MAAM,oBAAoB,CAAC;AAC5D,OAAO,EAAE,eAAe,EAAE,WAAW,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAC9F,OAAO,EAAE,uBAAuB,EAAE,MAAM,kBAAkB,CAAC;AAE3D,2EAA2E;AAC3E,MAAM,CAAC,KAAK,UAAU,iBAAiB,CAAC,SAAiB;IACvD,MAAM,KAAK,GAAc,EAAE,CAAC;IAC5B,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,UAAU,KAAK,CAAC,QAAgB;QACnC,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC;QAChD,MAAM,IAAI,GAAG,MAAM,KAAK,CAAC,QAAQ,CAAC,CAAC;QACnC,IAAI,IAAI,CAAC,cAAc,EAAE,IAAI,CAAC,CAAC,IAAI,CAAC,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,kDAAkD,CAAC,CAAC;QAC1I,IAAI,KAAK,CAAC,MAAM,IAAI,OAAO,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,IAAI,IAAI,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,wCAAwC,CAAC,CAAC;QACnJ,IAAI,IAAI,CAAC,WAAW,EAAE,EAAE,CAAC;YACvB,KAAK,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,WAAW,EAAE,CAAC,CAAC;YAClD,KAAK,MAAM,IAAI,IAAI,CAAC,MAAM,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,EAAE;gBAAE,MAAM,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,GAAG,QAAQ,IAAI,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;QAC5G,CAAC;aAAM,CAAC;YACN,MAAM,IAAI,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC;YAClC,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,gBAAgB,CAAC,QAAQ,CAAC;gBAAE,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YACzE,KAAK,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,KAAK,EAAE,IAAI,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,GAAG,KAAK,EAAE,MAAM,EAAE,UAAU,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC;QACpH,CAAC;IACH,CAAC;IACD,MAAM,KAAK,CAAC,EAAE,CAAC,CAAC;IAChB,OAAO,UAAU,CAAC,KAAK,CAAC,CAAC;AAC3B,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,qBAAqB,CAAC,SAAiB,EAAE,MAA+B;IAC5F,MAAM,cAAc,GAAG,MAAM,iBAAiB,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC,CAAC;IACjF,MAAM,QAAQ,GAAG,EAAE,GAAG,MAAM,EAAE,cAAc,EAAE,GAAG,EAAE,IAAI,EAAE,0BAA0B,EAAE,kBAAkB,EAAE,KAAK,EAAE,eAAe,EAAE,cAAc,EAAE,CAAC;IAClJ,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,iBAAiB,CAAC,EAAE,QAAQ,CAAC,CAAC;IACzE,OAAO,EAAE,EAAE,EAAE,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,EAAE,WAAW,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,iBAAiB,CAAC,CAAC,CAAC,EAAE,CAAC;AACxH,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,sBAAsB,CAAC,IAAY,EAAE,MAAc,EAAE,KAAqB;IAC9F,IAAI,KAAK,CAAC,SAAS;QAAE,MAAM,IAAI,KAAK,CAAC,oDAAoD,CAAC,CAAC;IAC3F,MAAM,SAAS,GAAG,KAAK,CAAC,UAAU,CAAC;IACnC,IAAI,CAAC,SAAS,IAAI,CAAC,2BAA2B,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,IAAI,CAAC,uBAAuB,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,sCAAsC,CAAC,CAAC;IAC9K,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,aAAa,EAAE,SAAS,CAAC,EAAE,CAAC,CAAC;IACzF,MAAM,GAAG,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,iBAAiB,CAAC,CAAC,CAAC;IACpE,IAAI,WAAW,CAAC,GAAG,CAAC,KAAK,SAAS,CAAC,MAAM;QAAE,MAAM,IAAI,KAAK,CAAC,oCAAoC,CAAC,CAAC;IACjG,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,EAAE,CAA4B,CAAC;IACrE,MAAM,MAAM,GAAG,MAAM,CAAC,MAA6C,CAAC;IACpE,IAAI,MAAM,CAAC,cAAc,KAAK,GAAG,IAAI,MAAM,CAAC,IAAI,KAAK,0BAA0B,IAAI,MAAM,CAAC,kBAAkB,KAAK,KAAK;WACjH,MAAM,CAAC,OAAO,KAAK,MAAM,IAAI,MAAM,CAAC,OAAO,KAAK,KAAK,CAAC,OAAO,IAAI,MAAM,CAAC,OAAO,KAAK,KAAK,CAAC,OAAO;WACjG,MAAM,EAAE,QAAQ,KAAK,KAAK,CAAC,QAAQ,IAAI,MAAM,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM;QAAE,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;IAC3I,IAAI,MAAM,CAAC,eAAe,KAAK,MAAM,iBAAiB,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,qCAAqC,CAAC,CAAC;IACjJ,IAAI,MAAM,CAAC,cAAc,KAAK,SAAS,EAAE,CAAC;QACxC,MAAM,MAAM,GAAG,MAAM,CAAC,cAAyC,CAAC;QAChE,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,8BAA8B,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;QACzG,IAAI,CAAC,MAAM,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,CAAC,cAAc,KAAK,GAAG,IAAI,MAAM,CAAC,IAAI,KAAK,qCAAqC;eAC9H,MAAM,CAAC,IAAI,KAAK,iDAAiD;eACjE,MAAM,CAAC,iBAAiB,KAAK,MAAM,CAAC,qBAAqB,IAAI,MAAM,CAAC,sBAAsB,KAAK,MAAM,CAAC,qBAAqB;eAC3H,CAAC,CAAC,MAAM,CAAC,iBAAiB,EAAE,MAAM,CAAC,sBAAsB,EAAE,MAAM,CAAC,aAAa,EAAE,MAAM,CAAC,kBAAkB,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,EAAE,CAAC,OAAO,KAAK,KAAK,QAAQ,IAAI,uBAAuB,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;eAC5L,CAAC,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,oBAAoB,CAAC,IAAI,MAAM,CAAC,MAAM,CAAC,oBAAoB,CAAC,GAAG,CAAC;eAC7F,UAAU,CAAC,MAAM,CAAC,KAAK,UAAU,CAAC,OAAO,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;IAClH,CAAC;SAAM,IAAI,MAAM,CAAC,qBAAqB,KAAK,SAAS,IAAI,MAAM,CAAC,qBAAqB,KAAK,MAAM,CAAC,qBAAqB,EAAE,CAAC;QACvH,MAAM,IAAI,KAAK,CAAC,+DAA+D,CAAC,CAAC;IACnF,CAAC;IACD,MAAM,WAAW,GAAG,MAAM,uBAAuB,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC;IAClG,IAAI,MAAM,CAAC,mBAAmB,KAAK,UAAU,CAAC,WAAW,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,mCAAmC,CAAC,CAAC;IACjH,OAAO,sBAAsB,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC;AACpD,CAAC"}
@@ -0,0 +1,165 @@
1
+ import { setTimeout as delay } from "node:timers/promises";
2
+ import { preparedArtifactForWorkItem } from "./work-item-artifacts.js";
3
+ import { HitchError } from "../foundation/index.js";
4
+ import { retryableInfrastructureTrials } from "./infrastructure-retry.js";
5
+ import { replaceInvalidEvalProgressTrial, writeEvalProgress } from "./progress.js";
6
+ import { physicalRetryWorkItem } from "./physical-retry-work.js";
7
+ import { ensurePhysicalRetryDecision, transitionRetryDecision } from "./retry-state.js";
8
+ import { retryBackoffMs } from "./retry-backoff.js";
9
+ export async function runRemoteInfrastructureRetries(input) {
10
+ const executor = input.options.remoteWorkExecutor;
11
+ if (!executor)
12
+ throw new TypeError("remote work executor is unavailable");
13
+ let progress = input.progress;
14
+ let candidates = retryableInfrastructureTrials(input.initial.refs);
15
+ const runs = [];
16
+ const firstRetryIndex = input.firstRetryIndex ?? 1;
17
+ if (!Number.isSafeInteger(firstRetryIndex) || firstRetryIndex < 1)
18
+ throw new TypeError("first infrastructure retry index is invalid");
19
+ for (let retry = firstRetryIndex; retry <= input.options.request.infrastructure_retries && candidates.length > 0; retry += 1) {
20
+ if (input.options.signal?.aborted)
21
+ break;
22
+ const triggers = [...candidates];
23
+ const work = physicalRetryWorkItem(input.item, retry, triggers);
24
+ const backoffMs = retryBackoffMs(input.options.request.infrastructure_retry_backoff_ms, retry, work.work_id);
25
+ const notBefore = new Date(Date.now() + backoffMs).toISOString();
26
+ const decisions = await Promise.all(triggers.map((trigger) => ensurePhysicalRetryDecision({
27
+ evalDirectory: input.options.evalDirectory, evalId: input.options.evalId, item: input.item,
28
+ retryIndex: retry, trigger, notBefore,
29
+ })));
30
+ if (decisions.some((decision) => decision.retry_work_id !== work.work_id))
31
+ throw new Error(`persisted retry work identity conflicts with execution: ${work.work_id}`);
32
+ await input.options.onWorkItemQueued?.(work.work_id);
33
+ input.options.sink.emit({
34
+ type: "eval.retry.decision", execution_kind: "physical-infrastructure-retry", work_id: work.work_id,
35
+ decision_ids: decisions.map((decision) => decision.decision_id),
36
+ candidate_executes: true, attempt: input.item.logical_attempt, retry,
37
+ tasks: input.item.task_ids, backoff_ms: backoffMs, provider: input.item.provider,
38
+ });
39
+ const remainingBackoffMs = Math.max(0, Date.parse(decisions[0]?.not_before ?? notBefore) - Date.now());
40
+ input.metrics?.addBlocked("backoff", remainingBackoffMs);
41
+ if (remainingBackoffMs > 0)
42
+ await delay(remainingBackoffMs, undefined, input.options.signal ? { signal: input.options.signal } : undefined);
43
+ if (input.options.signal?.aborted)
44
+ break;
45
+ input.options.sink.emit({
46
+ type: "eval.retry.ready", work_id: work.work_id, decision_ids: decisions.map((decision) => decision.decision_id),
47
+ retry, tasks: input.item.task_ids, priority: input.item.scheduling?.remaining_path_ms ?? 0,
48
+ });
49
+ const refs = [];
50
+ let decisionsSettled = false;
51
+ const settleDecisions = async () => {
52
+ if (decisionsSettled || refs.length === 0)
53
+ return;
54
+ candidates = retryableInfrastructureTrials(refs);
55
+ if (retry < input.options.request.infrastructure_retries && candidates.length > 0) {
56
+ const nextWork = physicalRetryWorkItem(input.item, retry + 1, candidates);
57
+ const nextNotBefore = new Date(Date.now() + retryBackoffMs(input.options.request.infrastructure_retry_backoff_ms, retry + 1, nextWork.work_id)).toISOString();
58
+ await Promise.all(candidates.map((trigger) => ensurePhysicalRetryDecision({
59
+ evalDirectory: input.options.evalDirectory, evalId: input.options.evalId, item: input.item,
60
+ retryIndex: retry + 1, trigger, notBefore: nextNotBefore,
61
+ })));
62
+ await input.options.onWorkItemQueued?.(nextWork.work_id);
63
+ }
64
+ const retryState = refs.some((ref) => ref.observation_status === "valid")
65
+ ? "repaired"
66
+ : retry < input.options.request.infrastructure_retries && candidates.length > 0 ? "invalid" : "exhausted";
67
+ for (const decision of decisions)
68
+ await transitionRetryDecision({
69
+ evalDirectory: input.options.evalDirectory, evalId: input.options.evalId, decisionId: decision.decision_id, state: retryState,
70
+ });
71
+ decisionsSettled = true;
72
+ };
73
+ const publish = async (ref) => {
74
+ assertSelectedTrial(ref, input.item);
75
+ const existing = refs.find((entry) => entry.trial_id === ref.trial_id || (entry.run_group ? entry.run_group.run_group_id === ref.run_group?.run_group_id : entry.run_id === ref.run_id));
76
+ if (existing) {
77
+ if (JSON.stringify(existing) !== JSON.stringify(ref))
78
+ throw new Error(`remote retry trial identity changed: ${ref.trial_id}`);
79
+ return;
80
+ }
81
+ refs.push(ref);
82
+ if (ref.invalid_reason === "candidate_evidence_unavailable")
83
+ input.options.sink.emit({
84
+ type: "eval.verifier.skipped", work_id: work.work_id, trial_id: ref.trial_id, task_id: ref.task_id,
85
+ reason: "candidate_evidence_unavailable", candidate_executes: true, verifier_executes: false,
86
+ });
87
+ if (ref.observation_status === "valid") {
88
+ if (input.replaceProgressTrial) {
89
+ await input.replaceProgressTrial(ref, work.work_id);
90
+ progress = input.currentProgress?.() ?? progress;
91
+ }
92
+ else {
93
+ progress = replaceInvalidEvalProgressTrial(progress, ref);
94
+ await writeEvalProgress(input.options.evalDirectory, progress);
95
+ }
96
+ }
97
+ input.options.sink.emit({
98
+ type: ref.observation_status === "valid" ? "eval.infrastructure-retry.repaired" : "eval.infrastructure-retry.failed",
99
+ attempt: input.item.logical_attempt, retry, task_id: ref.task_id, trial_id: ref.trial_id,
100
+ run_id: ref.run_id, observation_status: ref.observation_status,
101
+ ...(ref.invalid_reason ? { invalid_reason: ref.invalid_reason } : {}), generation: (input.currentProgress?.() ?? progress).generation,
102
+ });
103
+ };
104
+ const completed = await executor({
105
+ evalId: input.options.evalId, evalDirectory: input.options.evalDirectory, root: input.options.root,
106
+ request: input.options.request, plan: input.options.plan, workItem: work,
107
+ resolvedRevision: input.options.resolvedRevision, preparedArtifact: preparedArtifactForWorkItem(input.options, work),
108
+ runtimeDirectory: input.options.controllerRuntime.directory, runtimeId: input.options.controllerRuntime.runtime_id,
109
+ ...(input.initial.environmentImages ? { environmentImages: input.initial.environmentImages } : {}),
110
+ ...(input.options.plan.model_capture ? { modelCapturePlan: input.options.plan.model_capture } : {}),
111
+ publicationMode: "replace-invalid",
112
+ ...(input.options.signal ? { signal: input.options.signal } : {}),
113
+ emit: (event) => input.options.sink.emit({ ...event, execution_kind: "physical-infrastructure-retry", infrastructure_retry: retry }),
114
+ publish,
115
+ onLeaseState: async (leaseId, state) => {
116
+ if (state === "running") {
117
+ input.metrics?.startWork(work.work_id, "retry");
118
+ for (const decision of decisions)
119
+ await transitionRetryDecision({
120
+ evalDirectory: input.options.evalDirectory, evalId: input.options.evalId, decisionId: decision.decision_id, state: "running",
121
+ });
122
+ input.options.sink.emit({ type: "eval.retry.admitted", work_id: work.work_id, decision_ids: decisions.map((decision) => decision.decision_id) });
123
+ }
124
+ if (state === "terminal") {
125
+ await settleDecisions();
126
+ input.metrics?.finishWork(work.work_id);
127
+ }
128
+ await input.options.onWorkItemState?.(work.work_id, leaseId, state);
129
+ },
130
+ });
131
+ assertPublishedRefs(completed.refs, refs, input.item);
132
+ await settleDecisions();
133
+ input.metrics?.finishWork(work.work_id);
134
+ runs.push({
135
+ attempt: input.item.logical_attempt, retry, tasks: [...input.item.task_ids], triggers,
136
+ refs, run: completed.run, leaseId: completed.leaseId, workId: work.work_id,
137
+ });
138
+ input.options.sink.emit({
139
+ type: "eval.infrastructure-retry.completed", execution_kind: "physical-infrastructure-retry",
140
+ candidate_executes: true, attempt: input.item.logical_attempt, retry, tasks: input.item.task_ids,
141
+ repaired_tasks: refs.filter((ref) => ref.observation_status === "valid").map((ref) => ref.task_id).sort(),
142
+ remaining_tasks: candidates.map((ref) => ref.task_id).sort(), provider: input.item.provider,
143
+ });
144
+ if (completed.run.backend.process_exit_code !== 0 || completed.run.rawResult === null)
145
+ break;
146
+ }
147
+ return { progress: input.currentProgress?.() ?? progress, runs };
148
+ }
149
+ function assertSelectedTrial(ref, item) {
150
+ if (ref.attempt !== item.logical_attempt || !item.task_ids.includes(ref.task_id)) {
151
+ throw new HitchError(`remote infrastructure retry returned an unselected trial: ${ref.task_id}#${ref.attempt}`, {
152
+ code: "eval_infrastructure_retry_trial_mismatch", exitCode: 12,
153
+ });
154
+ }
155
+ }
156
+ function assertPublishedRefs(returned, published, item) {
157
+ for (const ref of returned)
158
+ assertSelectedTrial(ref, item);
159
+ if (returned.length !== published.length || returned.some((ref) => !published.some((entry) => JSON.stringify(entry) === JSON.stringify(ref)))) {
160
+ throw new HitchError("remote infrastructure retry returned refs that were not durably published", {
161
+ code: "eval_infrastructure_retry_result_mismatch", exitCode: 12,
162
+ });
163
+ }
164
+ }
165
+ //# 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,MAAM,wBAAwB,CAAC;AAEpD,OAAO,EAAE,6BAA6B,EAAE,MAAM,2BAA2B,CAAC;AAE1E,OAAO,EAAE,+BAA+B,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AACnF,OAAO,EAAE,qBAAqB,EAAE,MAAM,0BAA0B,CAAC;AACjE,OAAO,EAAE,2BAA2B,EAAE,uBAAuB,EAAE,MAAM,kBAAkB,CAAC;AACxF,OAAO,EAAE,cAAc,EAAE,MAAM,oBAAoB,CAAC;AAGpD,MAAM,CAAC,KAAK,UAAU,8BAA8B,CAAC,KASpD;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,MAAM,eAAe,GAAG,KAAK,CAAC,eAAe,IAAI,CAAC,CAAC;IACnD,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,eAAe,CAAC,IAAI,eAAe,GAAG,CAAC;QAAE,MAAM,IAAI,SAAS,CAAC,6CAA6C,CAAC,CAAC;IACtI,KAAK,IAAI,KAAK,GAAG,eAAe,EAAE,KAAK,IAAI,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,sBAAsB,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QAC7H,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,OAAO;YAAE,MAAM;QACzC,MAAM,QAAQ,GAAG,CAAC,GAAG,UAAU,CAAC,CAAC;QACjC,MAAM,IAAI,GAAG,qBAAqB,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,EAAE,QAAQ,CAAC,CAAC;QAChE,MAAM,SAAS,GAAG,cAAc,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,+BAA+B,EAAE,KAAK,EAAE,IAAI,CAAC,OAAO,CAAC,CAAC;QAC7G,MAAM,SAAS,GAAG,IAAI,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC,CAAC,WAAW,EAAE,CAAC;QACjE,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,2BAA2B,CAAC;YACxF,aAAa,EAAE,KAAK,CAAC,OAAO,CAAC,aAAa,EAAE,MAAM,EAAE,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI;YAC1F,UAAU,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS;SACtC,CAAC,CAAC,CAAC,CAAC;QACL,IAAI,SAAS,CAAC,IAAI,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,aAAa,KAAK,IAAI,CAAC,OAAO,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,2DAA2D,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;QACtK,MAAM,KAAK,CAAC,OAAO,CAAC,gBAAgB,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QACrD,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC;YACtB,IAAI,EAAE,qBAAqB,EAAE,cAAc,EAAE,+BAA+B,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO;YACnG,YAAY,EAAE,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,WAAW,CAAC;YAC/D,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,MAAM,kBAAkB,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,UAAU,IAAI,SAAS,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC;QACvG,KAAK,CAAC,OAAO,EAAE,UAAU,CAAC,SAAS,EAAE,kBAAkB,CAAC,CAAC;QACzD,IAAI,kBAAkB,GAAG,CAAC;YAAE,MAAM,KAAK,CAAC,kBAAkB,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;QAC5I,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,OAAO;YAAE,MAAM;QACzC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC;YACtB,IAAI,EAAE,kBAAkB,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,YAAY,EAAE,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,WAAW,CAAC;YAChH,KAAK,EAAE,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,QAAQ,EAAE,QAAQ,EAAE,KAAK,CAAC,IAAI,CAAC,UAAU,EAAE,iBAAiB,IAAI,CAAC;SAC3F,CAAC,CAAC;QACH,MAAM,IAAI,GAAqB,EAAE,CAAC;QAClC,IAAI,gBAAgB,GAAG,KAAK,CAAC;QAC7B,MAAM,eAAe,GAAG,KAAK,IAAmB,EAAE;YAChD,IAAI,gBAAgB,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC;gBAAE,OAAO;YAClD,UAAU,GAAG,6BAA6B,CAAC,IAAI,CAAC,CAAC;YACjD,IAAI,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,sBAAsB,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBAClF,MAAM,QAAQ,GAAG,qBAAqB,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,GAAG,CAAC,EAAE,UAAU,CAAC,CAAC;gBAC1E,MAAM,aAAa,GAAG,IAAI,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,cAAc,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,+BAA+B,EAAE,KAAK,GAAG,CAAC,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC;gBAC9J,MAAM,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,2BAA2B,CAAC;oBACxE,aAAa,EAAE,KAAK,CAAC,OAAO,CAAC,aAAa,EAAE,MAAM,EAAE,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI;oBAC1F,UAAU,EAAE,KAAK,GAAG,CAAC,EAAE,OAAO,EAAE,SAAS,EAAE,aAAa;iBACzD,CAAC,CAAC,CAAC,CAAC;gBACL,MAAM,KAAK,CAAC,OAAO,CAAC,gBAAgB,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;YAC3D,CAAC;YACD,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,kBAAkB,KAAK,OAAO,CAAC;gBACvE,CAAC,CAAC,UAAU;gBACZ,CAAC,CAAC,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,sBAAsB,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC;YAC5G,KAAK,MAAM,QAAQ,IAAI,SAAS;gBAAE,MAAM,uBAAuB,CAAC;oBAC9D,aAAa,EAAE,KAAK,CAAC,OAAO,CAAC,aAAa,EAAE,MAAM,EAAE,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,UAAU,EAAE,QAAQ,CAAC,WAAW,EAAE,KAAK,EAAE,UAAU;iBAC9H,CAAC,CAAC;YACH,gBAAgB,GAAG,IAAI,CAAC;QAC1B,CAAC,CAAC;QACF,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,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,YAAY,KAAK,GAAG,CAAC,SAAS,EAAE,YAAY,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,KAAK,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC;YACzL,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,cAAc,KAAK,gCAAgC;gBAAE,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC;oBACnF,IAAI,EAAE,uBAAuB,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,GAAG,CAAC,QAAQ,EAAE,OAAO,EAAE,GAAG,CAAC,OAAO;oBAClG,MAAM,EAAE,gCAAgC,EAAE,kBAAkB,EAAE,IAAI,EAAE,iBAAiB,EAAE,KAAK;iBAC7F,CAAC,CAAC;YACH,IAAI,GAAG,CAAC,kBAAkB,KAAK,OAAO,EAAE,CAAC;gBACvC,IAAI,KAAK,CAAC,oBAAoB,EAAE,CAAC;oBAC/B,MAAM,KAAK,CAAC,oBAAoB,CAAC,GAAG,EAAE,IAAI,CAAC,OAAO,CAAC,CAAC;oBACpD,QAAQ,GAAG,KAAK,CAAC,eAAe,EAAE,EAAE,IAAI,QAAQ,CAAC;gBACnD,CAAC;qBAAM,CAAC;oBACN,QAAQ,GAAG,+BAA+B,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;oBAC1D,MAAM,iBAAiB,CAAC,KAAK,CAAC,OAAO,CAAC,aAAa,EAAE,QAAQ,CAAC,CAAC;gBACjE,CAAC;YACH,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,CAAC,KAAK,CAAC,eAAe,EAAE,EAAE,IAAI,QAAQ,CAAC,CAAC,UAAU;aACtI,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,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE;gBACrC,IAAI,KAAK,KAAK,SAAS,EAAE,CAAC;oBACxB,KAAK,CAAC,OAAO,EAAE,SAAS,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;oBAChD,KAAK,MAAM,QAAQ,IAAI,SAAS;wBAAE,MAAM,uBAAuB,CAAC;4BAC9D,aAAa,EAAE,KAAK,CAAC,OAAO,CAAC,aAAa,EAAE,MAAM,EAAE,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,UAAU,EAAE,QAAQ,CAAC,WAAW,EAAE,KAAK,EAAE,SAAS;yBAC7H,CAAC,CAAC;oBACH,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,qBAAqB,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,YAAY,EAAE,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,WAAW,CAAC,EAAE,CAAC,CAAC;gBACnJ,CAAC;gBACD,IAAI,KAAK,KAAK,UAAU,EAAE,CAAC;oBACzB,MAAM,eAAe,EAAE,CAAC;oBACxB,KAAK,CAAC,OAAO,EAAE,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;gBAC1C,CAAC;gBACD,MAAM,KAAK,CAAC,OAAO,CAAC,eAAe,EAAE,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,KAAK,CAAC,CAAC;YACtE,CAAC;SACF,CAAC,CAAC;QACH,mBAAmB,CAAC,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;QACtD,MAAM,eAAe,EAAE,CAAC;QACxB,KAAK,CAAC,OAAO,EAAE,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QACxC,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,KAAK,CAAC,eAAe,EAAE,EAAE,IAAI,QAAQ,EAAE,IAAI,EAAE,CAAC;AACnE,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"}
@@ -38,12 +38,12 @@ export async function validateEvalRequest(input) {
38
38
  throw invalidInput("eval requires an immutable harness ref: version:<exact> or commit:<sha>");
39
39
  }
40
40
  assertExactLocalGitEvalReference(reference);
41
- const benchmark = await resolveBenchmarkReference(input.dataset.trim());
41
+ const benchmark = await resolveDatasetReference(input.dataset.trim());
42
42
  const attempts = positiveInteger(input.attempts ?? 1, "attempts");
43
43
  const maxConcurrent = positiveInteger(input.max_concurrent ?? 4, "max_concurrent");
44
44
  const infrastructureRetries = nonNegativeInteger(input.infrastructure_retries ?? DEFAULT_INFRASTRUCTURE_RETRIES, "infrastructure_retries");
45
45
  const infrastructureRetryBackoff = nonNegativeNumber(input.infrastructure_retry_backoff_ms ?? DEFAULT_INFRASTRUCTURE_RETRY_BACKOFF_MS, "infrastructure_retry_backoff_ms");
46
- const timeout = nonNegativeNumber(input.timeout_ms ?? DEFAULT_EVAL_TIMEOUT_MS, "timeout_ms");
46
+ const timeout = nonNegativeNumber(input.timeout_ms ?? (benchmark.manifest ? 0 : DEFAULT_EVAL_TIMEOUT_MS), "timeout_ms");
47
47
  const setupTimeout = nonNegativeNumber(input.setup_timeout_ms ?? DEFAULT_EVAL_SETUP_TIMEOUT_MS, "setup_timeout_ms");
48
48
  if (input.model !== undefined && typeof input.model !== "string")
49
49
  throw invalidInput("model must be a string");
@@ -54,6 +54,9 @@ export async function validateEvalRequest(input) {
54
54
  throw invalidInput("pass_env must be an array of strings");
55
55
  }
56
56
  const agentArgs = Array.isArray(input.agent_args) ? [...input.agent_args] : [];
57
+ if (benchmark.manifest) {
58
+ await assertStandardBenchmarkCandidate(input.dataset.trim(), benchmark.manifest, reference.harness_id, agentArgs);
59
+ }
57
60
  const bypassArg = HARBOR_EVAL_BYPASS_ARG[reference.harness_id];
58
61
  if (bypassArg && !agentArgs.includes(bypassArg))
59
62
  agentArgs.unshift(bypassArg);
@@ -76,13 +79,26 @@ export async function validateEvalRequest(input) {
76
79
  };
77
80
  }
78
81
  export async function resolveBenchmarkReference(dataset) {
82
+ const { benchmark_id, benchmark_revision } = await resolveDatasetReference(dataset);
83
+ return { benchmark_id, benchmark_revision };
84
+ }
85
+ async function resolveDatasetReference(dataset) {
79
86
  const raw = dataset.trim();
80
87
  const local = path.resolve(raw);
81
88
  try {
82
89
  if ((await stat(local)).isDirectory()) {
90
+ const manifest = await loadBenchmarkAdapterManifest(local);
91
+ if (manifest) {
92
+ return {
93
+ benchmark_id: manifest.benchmark.id,
94
+ benchmark_revision: manifest.dataset_digest,
95
+ manifest,
96
+ };
97
+ }
83
98
  return {
84
99
  benchmark_id: `local:${path.basename(local)}`,
85
100
  benchmark_revision: await workspaceDigest(local, { excludedTopLevel: new Set([".git"]) }),
101
+ manifest: null,
86
102
  };
87
103
  }
88
104
  }
@@ -98,7 +114,7 @@ export async function resolveBenchmarkReference(dataset) {
98
114
  const revision = raw.slice(separator + 1);
99
115
  if (revision.toLowerCase() === "latest")
100
116
  throw invalidInput("dataset revision cannot be latest");
101
- return { benchmark_id: benchmarkId, benchmark_revision: revision };
117
+ return { benchmark_id: benchmarkId, benchmark_revision: revision, manifest: null };
102
118
  }
103
119
  export async function resolveLocalDatasetTaskIds(dataset) {
104
120
  const local = path.resolve(dataset.trim());
@@ -155,4 +171,6 @@ import path from "node:path";
155
171
  import { SCHEMA_VERSION, invalidInput } from "../foundation/index.js";
156
172
  import { assertExactLocalGitEvalReference, parseHarnessReference } from "../revisions/index.js";
157
173
  import { workspaceDigest } from "../workspaces/index.js";
174
+ import { loadBenchmarkAdapterManifest } from "./benchmark-adapter-manifest.js";
175
+ import { assertStandardBenchmarkCandidate } from "./benchmark-candidate.js";
158
176
  //# sourceMappingURL=request.js.map