agent-hitch 0.2.9 → 0.2.10

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 (410) hide show
  1. package/README.md +206 -137
  2. package/README.zh-CN.md +193 -119
  3. package/dist/scripts/canary-native-training.js +221 -0
  4. package/dist/scripts/canary-native-training.js.map +1 -0
  5. package/dist/scripts/canary-remote-training.js +246 -0
  6. package/dist/scripts/canary-remote-training.js.map +1 -0
  7. package/dist/scripts/canary-remote-verifier.js +124 -0
  8. package/dist/scripts/canary-remote-verifier.js.map +1 -0
  9. package/dist/scripts/canary-worker-host.js +113 -0
  10. package/dist/scripts/canary-worker-host.js.map +1 -0
  11. package/dist/scripts/canary-worker-peer.js +188 -0
  12. package/dist/scripts/canary-worker-peer.js.map +1 -0
  13. package/dist/scripts/check-architecture.js +4 -3
  14. package/dist/scripts/check-architecture.js.map +1 -1
  15. package/dist/src/adapters/catalog.js +2 -1
  16. package/dist/src/adapters/catalog.js.map +1 -1
  17. package/dist/src/adapters/providers/codex.js +65 -2
  18. package/dist/src/adapters/providers/codex.js.map +1 -1
  19. package/dist/src/adapters/providers/model-call.js +17 -1
  20. package/dist/src/adapters/providers/model-call.js.map +1 -1
  21. package/dist/src/adapters/providers/training-tool.js +29 -0
  22. package/dist/src/adapters/providers/training-tool.js.map +1 -0
  23. package/dist/src/artifacts/index.js +1 -1
  24. package/dist/src/artifacts/index.js.map +1 -1
  25. package/dist/src/artifacts/store.js +15 -10
  26. package/dist/src/artifacts/store.js.map +1 -1
  27. package/dist/src/backends/harbor/backend.js +24 -15
  28. package/dist/src/backends/harbor/backend.js.map +1 -1
  29. package/dist/src/backends/harbor/bridge-environment.js +12 -0
  30. package/dist/src/backends/harbor/bridge-environment.js.map +1 -0
  31. package/dist/src/backends/harbor/environment-config.js +2 -2
  32. package/dist/src/backends/harbor/environment-config.js.map +1 -1
  33. package/dist/src/backends/harbor/host-credential-helper.js +81 -0
  34. package/dist/src/backends/harbor/host-credential-helper.js.map +1 -0
  35. package/dist/src/backends/harbor/index.js +3 -0
  36. package/dist/src/backends/harbor/index.js.map +1 -1
  37. package/dist/src/backends/harbor/model-proxy-config.js +18 -1
  38. package/dist/src/backends/harbor/model-proxy-config.js.map +1 -1
  39. package/dist/src/backends/harbor/observation.js +61 -0
  40. package/dist/src/backends/harbor/observation.js.map +1 -0
  41. package/dist/src/backends/harbor/portable-regrade.js +150 -0
  42. package/dist/src/backends/harbor/portable-regrade.js.map +1 -0
  43. package/dist/src/backends/harbor/process.js +26 -5
  44. package/dist/src/backends/harbor/process.js.map +1 -1
  45. package/dist/src/backends/harbor/regrade.js +4 -1
  46. package/dist/src/backends/harbor/regrade.js.map +1 -1
  47. package/dist/src/backends/harbor/supervisor.js +22 -0
  48. package/dist/src/backends/harbor/supervisor.js.map +1 -1
  49. package/dist/src/backends/harbor/tools.js +2 -0
  50. package/dist/src/backends/harbor/tools.js.map +1 -1
  51. package/dist/src/backends/index.js +3 -0
  52. package/dist/src/backends/index.js.map +1 -1
  53. package/dist/src/cli/arguments.js +33 -0
  54. package/dist/src/cli/arguments.js.map +1 -1
  55. package/dist/src/cli/commands/capabilities.js +14 -1
  56. package/dist/src/cli/commands/capabilities.js.map +1 -1
  57. package/dist/src/cli/commands/daemon.js +19 -0
  58. package/dist/src/cli/commands/daemon.js.map +1 -1
  59. package/dist/src/cli/commands/eval.js +38 -10
  60. package/dist/src/cli/commands/eval.js.map +1 -1
  61. package/dist/src/cli/commands/local.js +130 -0
  62. package/dist/src/cli/commands/local.js.map +1 -0
  63. package/dist/src/cli/commands/model-node.js +26 -0
  64. package/dist/src/cli/commands/model-node.js.map +1 -0
  65. package/dist/src/cli/commands/models.js +72 -0
  66. package/dist/src/cli/commands/models.js.map +1 -0
  67. package/dist/src/cli/commands/run.js +36 -1
  68. package/dist/src/cli/commands/run.js.map +1 -1
  69. package/dist/src/cli/commands/training.js +32 -0
  70. package/dist/src/cli/commands/training.js.map +1 -0
  71. package/dist/src/cli/commands/worker.js +30 -2
  72. package/dist/src/cli/commands/worker.js.map +1 -1
  73. package/dist/src/cli/main.js +8 -0
  74. package/dist/src/cli/main.js.map +1 -1
  75. package/dist/src/cli/output.js +24 -2
  76. package/dist/src/cli/output.js.map +1 -1
  77. package/dist/src/control-plane/eval-records.js +2 -0
  78. package/dist/src/control-plane/eval-records.js.map +1 -1
  79. package/dist/src/control-plane/eval-recovery.js +7 -1
  80. package/dist/src/control-plane/eval-recovery.js.map +1 -1
  81. package/dist/src/control-plane/eval-scheduler.js +24 -50
  82. package/dist/src/control-plane/eval-scheduler.js.map +1 -1
  83. package/dist/src/control-plane/eval-submission-persist.js +30 -0
  84. package/dist/src/control-plane/eval-submission-persist.js.map +1 -0
  85. package/dist/src/control-plane/execution-observation-contract.js +60 -0
  86. package/dist/src/control-plane/execution-observation-contract.js.map +1 -0
  87. package/dist/src/control-plane/index.js +10 -0
  88. package/dist/src/control-plane/index.js.map +1 -1
  89. package/dist/src/control-plane/inference-manager.js +357 -0
  90. package/dist/src/control-plane/inference-manager.js.map +1 -0
  91. package/dist/src/control-plane/managed-gateway-receipt.js +26 -0
  92. package/dist/src/control-plane/managed-gateway-receipt.js.map +1 -0
  93. package/dist/src/control-plane/managed-service-recovery.js +158 -0
  94. package/dist/src/control-plane/managed-service-recovery.js.map +1 -0
  95. package/dist/src/control-plane/model-capture-planning.js +15 -1
  96. package/dist/src/control-plane/model-capture-planning.js.map +1 -1
  97. package/dist/src/control-plane/ordered-eval-control.js +221 -0
  98. package/dist/src/control-plane/ordered-eval-control.js.map +1 -0
  99. package/dist/src/control-plane/remote-backend-result.js +18 -0
  100. package/dist/src/control-plane/remote-backend-result.js.map +1 -0
  101. package/dist/src/control-plane/remote-execution-lease.js +15 -0
  102. package/dist/src/control-plane/remote-execution-lease.js.map +1 -0
  103. package/dist/src/control-plane/remote-execution-observation.js +108 -0
  104. package/dist/src/control-plane/remote-execution-observation.js.map +1 -0
  105. package/dist/src/control-plane/remote-model-routes.js +107 -0
  106. package/dist/src/control-plane/remote-model-routes.js.map +1 -0
  107. package/dist/src/control-plane/remote-observation-reporter.js +37 -0
  108. package/dist/src/control-plane/remote-observation-reporter.js.map +1 -0
  109. package/dist/src/control-plane/remote-offer-contract.js +48 -0
  110. package/dist/src/control-plane/remote-offer-contract.js.map +1 -0
  111. package/dist/src/control-plane/remote-rerun-journal.js +61 -0
  112. package/dist/src/control-plane/remote-rerun-journal.js.map +1 -0
  113. package/dist/src/control-plane/remote-result-transport.js +55 -9
  114. package/dist/src/control-plane/remote-result-transport.js.map +1 -1
  115. package/dist/src/control-plane/remote-verifier-import.js +52 -0
  116. package/dist/src/control-plane/remote-verifier-import.js.map +1 -0
  117. package/dist/src/control-plane/remote-verifier-result.js +61 -0
  118. package/dist/src/control-plane/remote-verifier-result.js.map +1 -0
  119. package/dist/src/control-plane/remote-verifier-source-transport.js +68 -0
  120. package/dist/src/control-plane/remote-verifier-source-transport.js.map +1 -0
  121. package/dist/src/control-plane/remote-work-coordinator.js +103 -47
  122. package/dist/src/control-plane/remote-work-coordinator.js.map +1 -1
  123. package/dist/src/control-plane/remote-work-inputs.js +26 -5
  124. package/dist/src/control-plane/remote-work-inputs.js.map +1 -1
  125. package/dist/src/control-plane/remote-work-recovery.js +123 -34
  126. package/dist/src/control-plane/remote-work-recovery.js.map +1 -1
  127. package/dist/src/control-plane/remote-worker-artifacts.js +10 -7
  128. package/dist/src/control-plane/remote-worker-artifacts.js.map +1 -1
  129. package/dist/src/control-plane/remote-worker-cleanup.js +104 -0
  130. package/dist/src/control-plane/remote-worker-cleanup.js.map +1 -0
  131. package/dist/src/control-plane/remote-worker-client.js +156 -14
  132. package/dist/src/control-plane/remote-worker-client.js.map +1 -1
  133. package/dist/src/control-plane/remote-worker-codecs.js +160 -0
  134. package/dist/src/control-plane/remote-worker-codecs.js.map +1 -0
  135. package/dist/src/control-plane/remote-worker-generation-recovery.js +67 -0
  136. package/dist/src/control-plane/remote-worker-generation-recovery.js.map +1 -0
  137. package/dist/src/control-plane/remote-worker-ownership.js +51 -0
  138. package/dist/src/control-plane/remote-worker-ownership.js.map +1 -0
  139. package/dist/src/control-plane/remote-worker-protocol.js +85 -119
  140. package/dist/src/control-plane/remote-worker-protocol.js.map +1 -1
  141. package/dist/src/control-plane/remote-worker-runner.js +95 -18
  142. package/dist/src/control-plane/remote-worker-runner.js.map +1 -1
  143. package/dist/src/control-plane/remote-workers.js +33 -5
  144. package/dist/src/control-plane/remote-workers.js.map +1 -1
  145. package/dist/src/control-plane/rerun-recovery.js +89 -0
  146. package/dist/src/control-plane/rerun-recovery.js.map +1 -0
  147. package/dist/src/control-plane/rerun-scheduler.js +70 -96
  148. package/dist/src/control-plane/rerun-scheduler.js.map +1 -1
  149. package/dist/src/control-plane/rerun-source.js +57 -0
  150. package/dist/src/control-plane/rerun-source.js.map +1 -0
  151. package/dist/src/control-plane/worker-model-relay.js +76 -0
  152. package/dist/src/control-plane/worker-model-relay.js.map +1 -0
  153. package/dist/src/controller-runtime/hash.js +3 -0
  154. package/dist/src/controller-runtime/hash.js.map +1 -1
  155. package/dist/src/controller-runtime/index.js +2 -0
  156. package/dist/src/controller-runtime/index.js.map +1 -1
  157. package/dist/src/controller-runtime/observation.js +29 -0
  158. package/dist/src/controller-runtime/observation.js.map +1 -0
  159. package/dist/src/controller-runtime/resident.js +22 -0
  160. package/dist/src/controller-runtime/resident.js.map +1 -0
  161. package/dist/src/daemon/client.js +34 -1
  162. package/dist/src/daemon/client.js.map +1 -1
  163. package/dist/src/daemon/execution-observation.js +38 -0
  164. package/dist/src/daemon/execution-observation.js.map +1 -0
  165. package/dist/src/daemon/http.js +57 -0
  166. package/dist/src/daemon/http.js.map +1 -0
  167. package/dist/src/daemon/inference-prepare.js +47 -0
  168. package/dist/src/daemon/inference-prepare.js.map +1 -0
  169. package/dist/src/daemon/scheduler.js +4 -1
  170. package/dist/src/daemon/scheduler.js.map +1 -1
  171. package/dist/src/daemon/server.js +80 -82
  172. package/dist/src/daemon/server.js.map +1 -1
  173. package/dist/src/daemon/worker-observation.js +34 -0
  174. package/dist/src/daemon/worker-observation.js.map +1 -0
  175. package/dist/src/daemon/worker-ownership.js +71 -0
  176. package/dist/src/daemon/worker-ownership.js.map +1 -0
  177. package/dist/src/daemon/worker-routes.js +83 -15
  178. package/dist/src/daemon/worker-routes.js.map +1 -1
  179. package/dist/src/domain/execution-observation.js +2 -0
  180. package/dist/src/domain/execution-observation.js.map +1 -0
  181. package/dist/src/domain/index.js +1 -0
  182. package/dist/src/domain/index.js.map +1 -1
  183. package/dist/src/domain/inference.js +2 -0
  184. package/dist/src/domain/inference.js.map +1 -0
  185. package/dist/src/domain/runs.js.map +1 -1
  186. package/dist/src/domain/training.js +2 -0
  187. package/dist/src/domain/training.js.map +1 -0
  188. package/dist/src/domain/validation.js +11 -0
  189. package/dist/src/domain/validation.js.map +1 -1
  190. package/dist/src/domain/verifier-source.js +2 -0
  191. package/dist/src/domain/verifier-source.js.map +1 -0
  192. package/dist/src/domain/worker-recovery.js +2 -0
  193. package/dist/src/domain/worker-recovery.js.map +1 -0
  194. package/dist/src/evals/eval-logical-plan.js +8 -1
  195. package/dist/src/evals/eval-logical-plan.js.map +1 -1
  196. package/dist/src/evals/execution-leases.js +15 -2
  197. package/dist/src/evals/execution-leases.js.map +1 -1
  198. package/dist/src/evals/execution-plan.js +11 -1
  199. package/dist/src/evals/execution-plan.js.map +1 -1
  200. package/dist/src/evals/harbor-bridge-error.js +8 -0
  201. package/dist/src/evals/harbor-bridge-error.js.map +1 -1
  202. package/dist/src/evals/index.js +19 -0
  203. package/dist/src/evals/index.js.map +1 -1
  204. package/dist/src/evals/model-capture-plan.js +12 -0
  205. package/dist/src/evals/model-capture-plan.js.map +1 -1
  206. package/dist/src/evals/model-capture-runtime.js +25 -0
  207. package/dist/src/evals/model-capture-runtime.js.map +1 -1
  208. package/dist/src/evals/physical-work-plan.js +57 -0
  209. package/dist/src/evals/physical-work-plan.js.map +1 -0
  210. package/dist/src/evals/preparation-rerun.js +6 -0
  211. package/dist/src/evals/preparation-rerun.js.map +1 -1
  212. package/dist/src/evals/recovery.js +8 -1
  213. package/dist/src/evals/recovery.js.map +1 -1
  214. package/dist/src/evals/regrade-evidence.js +4 -4
  215. package/dist/src/evals/regrade-evidence.js.map +1 -1
  216. package/dist/src/evals/remote-execution-ownership.js +50 -0
  217. package/dist/src/evals/remote-execution-ownership.js.map +1 -0
  218. package/dist/src/evals/remote-infrastructure-retry.js +2 -0
  219. package/dist/src/evals/remote-infrastructure-retry.js.map +1 -1
  220. package/dist/src/evals/remote-lease-release.js +58 -0
  221. package/dist/src/evals/remote-lease-release.js.map +1 -0
  222. package/dist/src/evals/remote-rerun-execution.js +85 -0
  223. package/dist/src/evals/remote-rerun-execution.js.map +1 -0
  224. package/dist/src/evals/remote-trial.js +6 -0
  225. package/dist/src/evals/remote-trial.js.map +1 -0
  226. package/dist/src/evals/remote-verifier-assessment.js +85 -0
  227. package/dist/src/evals/remote-verifier-assessment.js.map +1 -0
  228. package/dist/src/evals/remote-verifier-contract.js +43 -0
  229. package/dist/src/evals/remote-verifier-contract.js.map +1 -0
  230. package/dist/src/evals/remote-verifier-execution.js +105 -0
  231. package/dist/src/evals/remote-verifier-execution.js.map +1 -0
  232. package/dist/src/evals/remote-verifier-journal.js +21 -0
  233. package/dist/src/evals/remote-verifier-journal.js.map +1 -0
  234. package/dist/src/evals/remote-verifier-outcome.js +51 -0
  235. package/dist/src/evals/remote-verifier-outcome.js.map +1 -0
  236. package/dist/src/evals/remote-verifier-rerun.js +134 -0
  237. package/dist/src/evals/remote-verifier-rerun.js.map +1 -0
  238. package/dist/src/evals/remote-verifier-selection.js +61 -0
  239. package/dist/src/evals/remote-verifier-selection.js.map +1 -0
  240. package/dist/src/evals/remote-verifier-source.js +33 -0
  241. package/dist/src/evals/remote-verifier-source.js.map +1 -0
  242. package/dist/src/evals/remote-worker-cleanup.js +46 -0
  243. package/dist/src/evals/remote-worker-cleanup.js.map +1 -0
  244. package/dist/src/evals/request.js +58 -1
  245. package/dist/src/evals/request.js.map +1 -1
  246. package/dist/src/evals/rerun-completion.js +39 -0
  247. package/dist/src/evals/rerun-completion.js.map +1 -0
  248. package/dist/src/evals/rerun-resume.js +70 -0
  249. package/dist/src/evals/rerun-resume.js.map +1 -0
  250. package/dist/src/evals/rerun-state.js +25 -0
  251. package/dist/src/evals/rerun-state.js.map +1 -0
  252. package/dist/src/evals/rerun-types.js.map +1 -1
  253. package/dist/src/evals/rerun.js +90 -46
  254. package/dist/src/evals/rerun.js.map +1 -1
  255. package/dist/src/evals/service.js +59 -5
  256. package/dist/src/evals/service.js.map +1 -1
  257. package/dist/src/evals/verifier-only-rerun.js +2 -16
  258. package/dist/src/evals/verifier-only-rerun.js.map +1 -1
  259. package/dist/src/evals/verifier-rerun-result.js +24 -0
  260. package/dist/src/evals/verifier-rerun-result.js.map +1 -0
  261. package/dist/src/evals/verifier-source-import.js +48 -0
  262. package/dist/src/evals/verifier-source-import.js.map +1 -0
  263. package/dist/src/evals/verifier-source.js +139 -0
  264. package/dist/src/evals/verifier-source.js.map +1 -0
  265. package/dist/src/foundation/base64.js +19 -0
  266. package/dist/src/foundation/base64.js.map +1 -0
  267. package/dist/src/foundation/config.js +10 -0
  268. package/dist/src/foundation/config.js.map +1 -1
  269. package/dist/src/foundation/credential-redaction.js +22 -9
  270. package/dist/src/foundation/credential-redaction.js.map +1 -1
  271. package/dist/src/foundation/index.js +1 -0
  272. package/dist/src/foundation/index.js.map +1 -1
  273. package/dist/src/foundation/process-identity.js +13 -9
  274. package/dist/src/foundation/process-identity.js.map +1 -1
  275. package/dist/src/inference/daemon-capacity.js +41 -0
  276. package/dist/src/inference/daemon-capacity.js.map +1 -0
  277. package/dist/src/inference/device-reservation.js +58 -0
  278. package/dist/src/inference/device-reservation.js.map +1 -0
  279. package/dist/src/inference/doctor.js +95 -0
  280. package/dist/src/inference/doctor.js.map +1 -0
  281. package/dist/src/inference/generation-recovery.js +83 -0
  282. package/dist/src/inference/generation-recovery.js.map +1 -0
  283. package/dist/src/inference/index.js +24 -0
  284. package/dist/src/inference/index.js.map +1 -0
  285. package/dist/src/inference/lock.js +331 -0
  286. package/dist/src/inference/lock.js.map +1 -0
  287. package/dist/src/inference/managed-launcher.js +35 -0
  288. package/dist/src/inference/managed-launcher.js.map +1 -0
  289. package/dist/src/inference/manifest.js +262 -0
  290. package/dist/src/inference/manifest.js.map +1 -0
  291. package/dist/src/inference/materialize.js +66 -0
  292. package/dist/src/inference/materialize.js.map +1 -0
  293. package/dist/src/inference/model-location.js +34 -0
  294. package/dist/src/inference/model-location.js.map +1 -0
  295. package/dist/src/inference/model-store.js +439 -0
  296. package/dist/src/inference/model-store.js.map +1 -0
  297. package/dist/src/inference/node-client.js +167 -0
  298. package/dist/src/inference/node-client.js.map +1 -0
  299. package/dist/src/inference/node-registry.js +79 -0
  300. package/dist/src/inference/node-registry.js.map +1 -0
  301. package/dist/src/inference/observation.js +79 -0
  302. package/dist/src/inference/observation.js.map +1 -0
  303. package/dist/src/inference/preflight.js +132 -0
  304. package/dist/src/inference/preflight.js.map +1 -0
  305. package/dist/src/inference/process-sglang.js +169 -0
  306. package/dist/src/inference/process-sglang.js.map +1 -0
  307. package/dist/src/inference/protocol-probe.js +29 -0
  308. package/dist/src/inference/protocol-probe.js.map +1 -0
  309. package/dist/src/inference/remote-model-store.js +24 -0
  310. package/dist/src/inference/remote-model-store.js.map +1 -0
  311. package/dist/src/inference/runtime-catalog.js +41 -0
  312. package/dist/src/inference/runtime-catalog.js.map +1 -0
  313. package/dist/src/inference/runtime-store.js +93 -0
  314. package/dist/src/inference/runtime-store.js.map +1 -0
  315. package/dist/src/inference/service-attachment.js +46 -0
  316. package/dist/src/inference/service-attachment.js.map +1 -0
  317. package/dist/src/inference/service-inspection.js +38 -0
  318. package/dist/src/inference/service-inspection.js.map +1 -0
  319. package/dist/src/inference/sglang.js +279 -0
  320. package/dist/src/inference/sglang.js.map +1 -0
  321. package/dist/src/inference/supervisor.js +460 -0
  322. package/dist/src/inference/supervisor.js.map +1 -0
  323. package/dist/src/model-access/environment.js +28 -0
  324. package/dist/src/model-access/environment.js.map +1 -0
  325. package/dist/src/model-access/index.js +4 -0
  326. package/dist/src/model-access/index.js.map +1 -1
  327. package/dist/src/model-access/local-gateway-request.js +139 -0
  328. package/dist/src/model-access/local-gateway-request.js.map +1 -0
  329. package/dist/src/model-access/local-gateway.js +323 -0
  330. package/dist/src/model-access/local-gateway.js.map +1 -0
  331. package/dist/src/model-access/proxy.js +90 -5
  332. package/dist/src/model-access/proxy.js.map +1 -1
  333. package/dist/src/model-access/remote-model.js +58 -0
  334. package/dist/src/model-access/remote-model.js.map +1 -0
  335. package/dist/src/model-access/training.js +115 -0
  336. package/dist/src/model-access/training.js.map +1 -0
  337. package/dist/src/runs/adapter-process.js +2 -2
  338. package/dist/src/runs/adapter-process.js.map +1 -1
  339. package/dist/src/runs/executor-types.js +2 -0
  340. package/dist/src/runs/executor-types.js.map +1 -0
  341. package/dist/src/runs/executor.js +54 -34
  342. package/dist/src/runs/executor.js.map +1 -1
  343. package/dist/src/runs/index.js +1 -0
  344. package/dist/src/runs/index.js.map +1 -1
  345. package/dist/src/runs/local-inference-environment.js +59 -0
  346. package/dist/src/runs/local-inference-environment.js.map +1 -0
  347. package/dist/src/runs/local-inference-run.js +53 -0
  348. package/dist/src/runs/local-inference-run.js.map +1 -0
  349. package/dist/src/runs/outcome.js +22 -0
  350. package/dist/src/runs/outcome.js.map +1 -1
  351. package/dist/src/runs/records.js +9 -0
  352. package/dist/src/runs/records.js.map +1 -1
  353. package/dist/src/runs/request.js +58 -3
  354. package/dist/src/runs/request.js.map +1 -1
  355. package/dist/src/runs/training-evidence.js +22 -0
  356. package/dist/src/runs/training-evidence.js.map +1 -0
  357. package/dist/src/runs/training-runtime.js +20 -0
  358. package/dist/src/runs/training-runtime.js.map +1 -0
  359. package/dist/src/workers/execution-observation.js +32 -0
  360. package/dist/src/workers/execution-observation.js.map +1 -0
  361. package/dist/src/workers/index.js +2 -0
  362. package/dist/src/workers/index.js.map +1 -1
  363. package/dist/src/workers/remote-harbor-ownership.js +266 -0
  364. package/dist/src/workers/remote-harbor-ownership.js.map +1 -0
  365. package/dist/src/workers/remote-harbor-verifier.js +76 -0
  366. package/dist/src/workers/remote-harbor-verifier.js.map +1 -0
  367. package/dist/src/workers/remote-harbor-work-spec.js +66 -10
  368. package/dist/src/workers/remote-harbor-work-spec.js.map +1 -1
  369. package/dist/src/workers/remote-harbor-worker.js +56 -49
  370. package/dist/src/workers/remote-harbor-worker.js.map +1 -1
  371. package/dist/src/workers/remote-worker-host.js +33 -0
  372. package/dist/src/workers/remote-worker-host.js.map +1 -0
  373. package/docs/local-model-inference-spec.zh-CN.md +666 -0
  374. package/docs/schemas/controller-runtime-observation.schema.json +21 -0
  375. package/docs/schemas/eval-execution-plan.schema.json +1 -0
  376. package/docs/schemas/eval-request.schema.json +3 -1
  377. package/docs/schemas/execution-lease.schema.json +95 -1
  378. package/docs/schemas/execution-provider-status.schema.json +162 -28
  379. package/docs/schemas/inference-execution-evidence.schema.json +487 -0
  380. package/docs/schemas/inference-lock.schema.json +487 -0
  381. package/docs/schemas/inference-runtime-manifest.schema.json +156 -0
  382. package/docs/schemas/inference-service-record.schema.json +227 -0
  383. package/docs/schemas/local-inference-selection.schema.json +98 -0
  384. package/docs/schemas/local-model-manifest.schema.json +43 -0
  385. package/docs/schemas/model-node-binding.schema.json +34 -0
  386. package/docs/schemas/model-node-generation-release.schema.json +46 -0
  387. package/docs/schemas/ordered-eval-control.schema.json +39 -0
  388. package/docs/schemas/remote-execution-lease-grant.schema.json +45 -0
  389. package/docs/schemas/remote-harbor-work-spec.schema.json +325 -0
  390. package/docs/schemas/remote-model-binding.schema.json +21 -0
  391. package/docs/schemas/remote-physical-execution.schema.json +108 -0
  392. package/docs/schemas/remote-result-envelope.schema.json +9 -3
  393. package/docs/schemas/remote-tree-envelope.schema.json +17 -3
  394. package/docs/schemas/remote-verifier-result.schema.json +203 -0
  395. package/docs/schemas/remote-verifier-source.schema.json +52 -0
  396. package/docs/schemas/remote-verifier-work.schema.json +125 -0
  397. package/docs/schemas/remote-work-offer.schema.json +170 -34
  398. package/docs/schemas/remote-worker-execution-admission.schema.json +214 -0
  399. package/docs/schemas/remote-worker-generation-cleanup.schema.json +199 -0
  400. package/docs/schemas/remote-worker-registration.schema.json +226 -25
  401. package/docs/schemas/run-manifest.schema.json +211 -44
  402. package/docs/schemas/run-request.schema.json +2 -1
  403. package/docs/schemas/training-external-binding.schema.json +26 -0
  404. package/integrations/harbor/hitch_harbor_agent.py +241 -13
  405. package/integrations/harbor/hitch_harbor_environment.py +22 -1
  406. package/integrations/harbor/hitch_host_credentials.py +354 -0
  407. package/integrations/harbor/hitch_private_exec.py +150 -0
  408. package/integrations/model-call/cli.js +13 -5
  409. package/integrations/training-tool/cli.js +72 -0
  410. package/package.json +8 -2
@@ -0,0 +1,105 @@
1
+ import { cp, mkdir, readFile } from "node:fs/promises";
2
+ import path from "node:path";
3
+ import { parse as parseTOML } from "smol-toml";
4
+ import { parsePortableHarborRegradeConfig, restorePortableHarborRegradeConfig, runHarborRegrade } from "../backends/index.js";
5
+ import { useControllerRuntimeDirectory } from "../controller-runtime/index.js";
6
+ import { HitchError, atomicWriteJSON, readJSON, sha256JSON, statePaths } from "../foundation/index.js";
7
+ import { dockerOwnershipLabelMap, dockerResourceOwnership } from "./docker-ownership.js";
8
+ import { startDockerResourceObserver } from "./docker-resource-observer.js";
9
+ import { runtimeResourcesForTask } from "./execution-plan-resources.js";
10
+ import { readExecutionLeases } from "./execution-leases.js";
11
+ import { regradeTreeDigest } from "./regrade-evidence.js";
12
+ import { portableRemoteTrial } from "./remote-trial.js";
13
+ import { parseRemoteVerifierSourceManifest, verifyRemoteVerifierSourceContents } from "./verifier-source.js";
14
+ import { verifierRuntimeRepair } from "./verifier-runtime.js";
15
+ /** Executes only the artifact regrade phase. The worker runner owns heartbeat, cancellation and proven resource release. */
16
+ export async function runRemoteVerifierWork(input) {
17
+ const source = parseRemoteVerifierSourceManifest(input.sourceManifest);
18
+ const checkLease = async () => {
19
+ const current = (await readExecutionLeases(path.join(statePaths(input.root).evals, input.lease.eval_id))).find(lease => lease.lease_id === input.lease.lease_id);
20
+ const fields = ["lease_id", "eval_id", "work_id", "worker_id", "provider", "collision_domain_id", "epoch", "reservation", "resource_epochs"];
21
+ if (!current || current.state !== "running" || Date.parse(current.expires_at) <= Date.now()
22
+ || fields.some(field => sha256JSON(current[field] ?? null) !== sha256JSON(input.lease[field] ?? null)))
23
+ throw unavailable("remote verifier has no current execution lease");
24
+ };
25
+ await checkLease();
26
+ if (source.status !== "available" || !source.regrade_config_digest || !source.original_result_digest
27
+ || !/^sha256:[a-f0-9]{64}$/.test(input.candidateResultDigest) || sha256JSON(input.candidateResult) !== input.candidateResultDigest
28
+ || input.candidateResult.run_id !== source.run_id || input.candidateResult.status !== "succeeded"
29
+ || sha256JSON(portableRemoteTrial(input.sourceTrial)) !== source.source_result_digest || input.sourceTrial.config !== undefined
30
+ || input.sourceTrial.trial_name !== source.trial_id
31
+ || !/^[A-Za-z0-9][A-Za-z0-9._-]{0,255}$/.test(input.trialName)
32
+ || !input.plan.slots.some(slot => slot.task_id === source.task_id) || input.plan.eval_id !== input.lease.eval_id
33
+ || input.plan.provider !== input.lease.provider)
34
+ throw unavailable("remote verifier source/work identity differs");
35
+ const metadata = input.sourceTrial.agent_result?.metadata;
36
+ if (metadata?.hitch_run_id !== source.run_id || metadata.controller_runtime_id !== source.controller_runtime_id)
37
+ throw unavailable("remote verifier candidate/runtime identity differs");
38
+ const task = parseTOML(await readFile(path.join(input.taskDirectory, "task.toml"), "utf8"));
39
+ const taskName = task.task?.name ?? path.basename(input.taskDirectory);
40
+ if (input.sourceTrial.task_name !== taskName)
41
+ throw unavailable("remote verifier task name differs");
42
+ const runtime = await useControllerRuntimeDirectory(input.runtimeDirectory, source.controller_runtime_id);
43
+ const verifierRuntime = await useControllerRuntimeDirectory(input.verifierRuntimeDirectory, input.verifierRuntimeId);
44
+ const repair = verifierRuntimeRepair(runtime, verifierRuntime);
45
+ const portable = parsePortableHarborRegradeConfig(await readJSON(path.join(input.sourceSnapshotDirectory, "regrade-config.json")));
46
+ if (sha256JSON(portable) !== source.regrade_config_digest || portable.source_config_digest !== source.source_config_digest)
47
+ throw unavailable("remote verifier config identity differs");
48
+ if (input.signal?.aborted)
49
+ throw input.signal.reason ?? new DOMException("Aborted", "AbortError");
50
+ await mkdir(input.directory, { recursive: false, mode: 0o700 });
51
+ const candidateDirectory = path.join(input.directory, "candidate");
52
+ await atomicWriteJSON(path.join(candidateDirectory, "result.json"), input.candidateResult);
53
+ const verifySource = () => verifyRemoteVerifierSourceContents({ manifest: source, snapshotDirectory: input.sourceSnapshotDirectory,
54
+ taskDirectory: input.taskDirectory, bundleDirectory: candidateDirectory });
55
+ await verifySource();
56
+ await checkLease();
57
+ if (input.signal?.aborted)
58
+ throw input.signal.reason ?? new DOMException("Aborted", "AbortError");
59
+ const replay = path.join(input.directory, "source-trial");
60
+ await cp(input.sourceSnapshotDirectory, replay, { recursive: true, force: false, errorOnExist: true, dereference: false });
61
+ if (await regradeTreeDigest(replay) !== source.snapshot_digest)
62
+ throw unavailable("remote verifier source changed while copying");
63
+ const ownership = dockerResourceOwnership(input.root, input.lease, source.task_id);
64
+ const { sourceConfig, regradeConfig } = restorePortableHarborRegradeConfig({ portable, sourceConfigDigest: source.source_config_digest,
65
+ taskDirectory: input.taskDirectory, sourceDirectory: replay, outputDirectory: path.join(input.directory, "trials"),
66
+ trialName: input.trialName, sourceResult: input.sourceTrial, ownershipLabels: dockerOwnershipLabelMap(ownership) });
67
+ await atomicWriteJSON(path.join(replay, "config.json"), sourceConfig);
68
+ // Harbor's TrialResult requires config for decoding; the private original stays on its original worker.
69
+ await atomicWriteJSON(path.join(replay, "result.json"), { ...input.sourceTrial, config: sourceConfig });
70
+ const replayDigest = await regradeTreeDigest(replay);
71
+ const limits = runtimeResourcesForTask(input.plan, source.task_id, input.lease.reservation);
72
+ const environment = regradeConfig.environment, kwargs = environment.kwargs;
73
+ if (environment.cpu_enforcement_policy !== "limit" || environment.memory_enforcement_policy !== "limit"
74
+ || environment.override_cpus !== limits.mainLimits.cpu_millis / 1000 || environment.override_memory_mb !== limits.mainLimits.memory_bytes / 1024 ** 2
75
+ || (kwargs.hitch_main_gpu_count ?? 0) !== (limits.mainLimits.gpu_count ?? 0)
76
+ || environment.override_gpus != null && environment.override_gpus !== (limits.mainLimits.gpu_count ?? 0)
77
+ || sha256JSON(kwargs.hitch_service_resource_limits ?? {}) !== sha256JSON(limits.sidecarLimits))
78
+ throw unavailable("remote verifier resource settings differ from its lease plan");
79
+ const observer = startDockerResourceObserver({ ownership, workerId: input.lease.worker_id, collisionDomainId: input.lease.collision_domain_id,
80
+ reservation: input.lease.reservation, mainLimits: limits.mainLimits, sidecarLimits: limits.sidecarLimits, env: input.env,
81
+ ...(input.signal ? { signal: input.signal } : {}) });
82
+ let outcome;
83
+ try {
84
+ outcome = await runHarborRegrade({ root: input.root, directory: input.directory, config: regradeConfig,
85
+ runtimeDirectory: verifierRuntime.directory, trustedResult: input.candidateResult, env: input.env,
86
+ ...(input.processHooks ? { processHooks: input.processHooks } : {}),
87
+ ...(input.harborExecutable ? { harborExecutable: input.harborExecutable } : {}), ...(input.signal ? { signal: input.signal } : {}) });
88
+ }
89
+ finally {
90
+ await atomicWriteJSON(path.join(input.directory, "execution.json"), await observer.stop());
91
+ }
92
+ await verifySource();
93
+ await checkLease();
94
+ if (input.signal?.aborted)
95
+ throw input.signal.reason ?? new DOMException("Aborted", "AbortError");
96
+ if (await regradeTreeDigest(replay) !== replayDigest || outcome.trial.trial_name !== input.trialName || outcome.trial.task_name !== taskName
97
+ || sha256JSON(outcome.trial.agent_result) !== sha256JSON(input.sourceTrial.agent_result))
98
+ throw unavailable("remote verifier mutated its source candidate");
99
+ return { trial: portableRemoteTrial(outcome.trial), backend: outcome.backend, execution: await readJSON(path.join(input.directory, "execution.json")),
100
+ source_manifest_digest: sha256JSON(source), candidate_result_digest: input.candidateResultDigest,
101
+ config_digest: sha256JSON(regradeConfig), controller_runtime_id: verifierRuntime.runtime_id,
102
+ ...(repair ? { runtime_repair: repair } : {}), trial_directory: path.join(input.directory, "trials", input.trialName) };
103
+ }
104
+ function unavailable(message) { return new HitchError(message, { code: "eval_verifier_only_unavailable", exitCode: 2 }); }
105
+ //# sourceMappingURL=remote-verifier-execution.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"remote-verifier-execution.js","sourceRoot":"","sources":["../../../src/evals/remote-verifier-execution.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AACvD,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,EAAE,KAAK,IAAI,SAAS,EAAE,MAAM,WAAW,CAAC;AAC/C,OAAO,EAAE,gCAAgC,EAAE,kCAAkC,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AAC9H,OAAO,EAAE,6BAA6B,EAAE,MAAM,gCAAgC,CAAC;AAE/E,OAAO,EAAE,UAAU,EAAE,eAAe,EAAE,QAAQ,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AACvG,OAAO,EAAE,uBAAuB,EAAE,uBAAuB,EAAE,MAAM,uBAAuB,CAAC;AACzF,OAAO,EAAE,2BAA2B,EAAE,MAAM,+BAA+B,CAAC;AAC5E,OAAO,EAAE,uBAAuB,EAAE,MAAM,+BAA+B,CAAC;AACxE,OAAO,EAAE,mBAAmB,EAAE,MAAM,uBAAuB,CAAC;AAC5D,OAAO,EAAE,iBAAiB,EAAE,MAAM,uBAAuB,CAAC;AAC1D,OAAO,EAAE,mBAAmB,EAAE,MAAM,mBAAmB,CAAC;AACxD,OAAO,EAAE,iCAAiC,EAAE,kCAAkC,EAAE,MAAM,sBAAsB,CAAC;AAC7G,OAAO,EAAE,qBAAqB,EAAE,MAAM,uBAAuB,CAAC;AAE9D,4HAA4H;AAC5H,MAAM,CAAC,KAAK,UAAU,qBAAqB,CAAC,KAQ3C;IACC,MAAM,MAAM,GAAG,iCAAiC,CAAC,KAAK,CAAC,cAAc,CAAC,CAAC;IACvE,MAAM,UAAU,GAAG,KAAK,IAAI,EAAE;QAC5B,MAAM,OAAO,GAAG,CAAC,MAAM,mBAAmB,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,KAAK,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,KAAK,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;QACjK,MAAM,MAAM,GAAG,CAAC,UAAU,EAAE,SAAS,EAAE,SAAS,EAAE,WAAW,EAAE,UAAU,EAAE,qBAAqB,EAAE,OAAO,EAAE,aAAa,EAAE,iBAAiB,CAAU,CAAC;QACtJ,IAAI,CAAC,OAAO,IAAI,OAAO,CAAC,KAAK,KAAK,SAAS,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,UAAU,CAAC,IAAI,IAAI,CAAC,GAAG,EAAE;eACtF,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,CAAC,KAAK,CAAC,IAAI,IAAI,CAAC,KAAK,UAAU,CAAC,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,IAAI,CAAC,CAAC;YAAE,MAAM,WAAW,CAAC,gDAAgD,CAAC,CAAC;IAChL,CAAC,CAAC;IACF,MAAM,UAAU,EAAE,CAAC;IACnB,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW,IAAI,CAAC,MAAM,CAAC,qBAAqB,IAAI,CAAC,MAAM,CAAC,sBAAsB;WAC/F,CAAC,uBAAuB,CAAC,IAAI,CAAC,KAAK,CAAC,qBAAqB,CAAC,IAAI,UAAU,CAAC,KAAK,CAAC,eAAe,CAAC,KAAK,KAAK,CAAC,qBAAqB;WAC/H,KAAK,CAAC,eAAe,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM,IAAI,KAAK,CAAC,eAAe,CAAC,MAAM,KAAK,WAAW;WAC9F,UAAU,CAAC,mBAAmB,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC,KAAK,MAAM,CAAC,oBAAoB,IAAI,KAAK,CAAC,WAAW,CAAC,MAAM,KAAK,SAAS;WAC5H,KAAK,CAAC,WAAW,CAAC,UAAU,KAAK,MAAM,CAAC,QAAQ;WAChD,CAAC,oCAAoC,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC;WAC3D,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,OAAO,KAAK,MAAM,CAAC,OAAO,CAAC,IAAI,KAAK,CAAC,IAAI,CAAC,OAAO,KAAK,KAAK,CAAC,KAAK,CAAC,OAAO;WAC7G,KAAK,CAAC,IAAI,CAAC,QAAQ,KAAK,KAAK,CAAC,KAAK,CAAC,QAAQ;QAAE,MAAM,WAAW,CAAC,8CAA8C,CAAC,CAAC;IACrH,MAAM,QAAQ,GAAI,KAAK,CAAC,WAAW,CAAC,YAAmE,EAAE,QAAQ,CAAC;IAClH,IAAI,QAAQ,EAAE,YAAY,KAAK,MAAM,CAAC,MAAM,IAAI,QAAQ,CAAC,qBAAqB,KAAK,MAAM,CAAC,qBAAqB;QAAE,MAAM,WAAW,CAAC,oDAAoD,CAAC,CAAC;IACzL,MAAM,IAAI,GAAG,SAAS,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,WAAW,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;IAC5F,MAAM,QAAQ,GAAI,IAAI,CAAC,IAA4C,EAAE,IAAI,IAAI,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;IAChH,IAAI,KAAK,CAAC,WAAW,CAAC,SAAS,KAAK,QAAQ;QAAE,MAAM,WAAW,CAAC,mCAAmC,CAAC,CAAC;IACrG,MAAM,OAAO,GAAG,MAAM,6BAA6B,CAAC,KAAK,CAAC,gBAAgB,EAAE,MAAM,CAAC,qBAAqB,CAAC,CAAC;IAC1G,MAAM,eAAe,GAAG,MAAM,6BAA6B,CAAC,KAAK,CAAC,wBAAwB,EAAE,KAAK,CAAC,iBAAiB,CAAC,CAAC;IACrH,MAAM,MAAM,GAAG,qBAAqB,CAAC,OAAO,EAAE,eAAe,CAAC,CAAC;IAC/D,MAAM,QAAQ,GAAG,gCAAgC,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,uBAAuB,EAAE,qBAAqB,CAAC,CAAC,CAAC,CAAC;IACnI,IAAI,UAAU,CAAC,QAAQ,CAAC,KAAK,MAAM,CAAC,qBAAqB,IAAI,QAAQ,CAAC,oBAAoB,KAAK,MAAM,CAAC,oBAAoB;QAAE,MAAM,WAAW,CAAC,yCAAyC,CAAC,CAAC;IACzL,IAAI,KAAK,CAAC,MAAM,EAAE,OAAO;QAAE,MAAM,KAAK,CAAC,MAAM,CAAC,MAAM,IAAI,IAAI,YAAY,CAAC,SAAS,EAAE,YAAY,CAAC,CAAC;IAClG,MAAM,KAAK,CAAC,KAAK,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC;IAChE,MAAM,kBAAkB,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,WAAW,CAAC,CAAC;IACnE,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,kBAAkB,EAAE,aAAa,CAAC,EAAE,KAAK,CAAC,eAAe,CAAC,CAAC;IAC3F,MAAM,YAAY,GAAG,GAAG,EAAE,CAAC,kCAAkC,CAAC,EAAE,QAAQ,EAAE,MAAM,EAAE,iBAAiB,EAAE,KAAK,CAAC,uBAAuB;QAChI,aAAa,EAAE,KAAK,CAAC,aAAa,EAAE,eAAe,EAAE,kBAAkB,EAAE,CAAC,CAAC;IAC7E,MAAM,YAAY,EAAE,CAAC;IACrB,MAAM,UAAU,EAAE,CAAC;IACnB,IAAI,KAAK,CAAC,MAAM,EAAE,OAAO;QAAE,MAAM,KAAK,CAAC,MAAM,CAAC,MAAM,IAAI,IAAI,YAAY,CAAC,SAAS,EAAE,YAAY,CAAC,CAAC;IAClG,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,cAAc,CAAC,CAAC;IAC1D,MAAM,EAAE,CAAC,KAAK,CAAC,uBAAuB,EAAE,MAAM,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,KAAK,EAAE,YAAY,EAAE,IAAI,EAAE,WAAW,EAAE,KAAK,EAAE,CAAC,CAAC;IAC3H,IAAI,MAAM,iBAAiB,CAAC,MAAM,CAAC,KAAK,MAAM,CAAC,eAAe;QAAE,MAAM,WAAW,CAAC,8CAA8C,CAAC,CAAC;IAClI,MAAM,SAAS,GAAG,uBAAuB,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,KAAK,EAAE,MAAM,CAAC,OAAO,CAAC,CAAC;IACnF,MAAM,EAAE,YAAY,EAAE,aAAa,EAAE,GAAG,kCAAkC,CAAC,EAAE,QAAQ,EAAE,kBAAkB,EAAE,MAAM,CAAC,oBAAoB;QACpI,aAAa,EAAE,KAAK,CAAC,aAAa,EAAE,eAAe,EAAE,MAAM,EAAE,eAAe,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,QAAQ,CAAC;QAClH,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,YAAY,EAAE,KAAK,CAAC,WAAW,EAAE,eAAe,EAAE,uBAAuB,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC;IACtH,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,aAAa,CAAC,EAAE,YAAY,CAAC,CAAC;IACtE,wGAAwG;IACxG,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,aAAa,CAAC,EAAE,EAAE,GAAG,KAAK,CAAC,WAAW,EAAE,MAAM,EAAE,YAAY,EAAE,CAAC,CAAC;IACxG,MAAM,YAAY,GAAG,MAAM,iBAAiB,CAAC,MAAM,CAAC,CAAC;IACrD,MAAM,MAAM,GAAG,uBAAuB,CAAC,KAAK,CAAC,IAAI,EAAE,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC;IAC5F,MAAM,WAAW,GAAG,aAAa,CAAC,WAAsC,EAAE,MAAM,GAAG,WAAW,CAAC,MAAiC,CAAC;IACjI,IAAI,WAAW,CAAC,sBAAsB,KAAK,OAAO,IAAI,WAAW,CAAC,yBAAyB,KAAK,OAAO;WAClG,WAAW,CAAC,aAAa,KAAK,MAAM,CAAC,UAAU,CAAC,UAAU,GAAG,IAAI,IAAI,WAAW,CAAC,kBAAkB,KAAK,MAAM,CAAC,UAAU,CAAC,YAAY,GAAG,IAAI,IAAI,CAAC;WAClJ,CAAC,MAAM,CAAC,oBAAoB,IAAI,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,UAAU,CAAC,SAAS,IAAI,CAAC,CAAC;WACzE,WAAW,CAAC,aAAa,IAAI,IAAI,IAAI,WAAW,CAAC,aAAa,KAAK,CAAC,MAAM,CAAC,UAAU,CAAC,SAAS,IAAI,CAAC,CAAC;WACrG,UAAU,CAAC,MAAM,CAAC,6BAA6B,IAAI,EAAE,CAAC,KAAK,UAAU,CAAC,MAAM,CAAC,aAAa,CAAC;QAAE,MAAM,WAAW,CAAC,8DAA8D,CAAC,CAAC;IACpL,MAAM,QAAQ,GAAG,2BAA2B,CAAC,EAAE,SAAS,EAAE,QAAQ,EAAE,KAAK,CAAC,KAAK,CAAC,SAAS,EAAE,iBAAiB,EAAE,KAAK,CAAC,KAAK,CAAC,mBAAmB;QAC3I,WAAW,EAAE,KAAK,CAAC,KAAK,CAAC,WAAW,EAAE,UAAU,EAAE,MAAM,CAAC,UAAU,EAAE,aAAa,EAAE,MAAM,CAAC,aAAa,EAAE,GAAG,EAAE,KAAK,CAAC,GAAG;QACxH,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;IACvD,IAAI,OAAqD,CAAC;IAC1D,IAAI,CAAC;QACH,OAAO,GAAG,MAAM,gBAAgB,CAAC,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,MAAM,EAAE,aAAa;YACpG,gBAAgB,EAAE,eAAe,CAAC,SAAS,EAAE,aAAa,EAAE,KAAK,CAAC,eAAe,EAAE,GAAG,EAAE,KAAK,CAAC,GAAG;YACjG,GAAG,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,KAAK,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACnE,GAAG,CAAC,KAAK,CAAC,gBAAgB,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,KAAK,CAAC,gBAAgB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;IAC1I,CAAC;YAAS,CAAC;QAAC,MAAM,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,gBAAgB,CAAC,EAAE,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC;IAAC,CAAC;IACzG,MAAM,YAAY,EAAE,CAAC;IACrB,MAAM,UAAU,EAAE,CAAC;IACnB,IAAI,KAAK,CAAC,MAAM,EAAE,OAAO;QAAE,MAAM,KAAK,CAAC,MAAM,CAAC,MAAM,IAAI,IAAI,YAAY,CAAC,SAAS,EAAE,YAAY,CAAC,CAAC;IAClG,IAAI,MAAM,iBAAiB,CAAC,MAAM,CAAC,KAAK,YAAY,IAAI,OAAO,CAAC,KAAK,CAAC,UAAU,KAAK,KAAK,CAAC,SAAS,IAAI,OAAO,CAAC,KAAK,CAAC,SAAS,KAAK,QAAQ;WACvI,UAAU,CAAC,OAAO,CAAC,KAAK,CAAC,YAAY,CAAC,KAAK,UAAU,CAAC,KAAK,CAAC,WAAW,CAAC,YAAY,CAAC;QAAE,MAAM,WAAW,CAAC,8CAA8C,CAAC,CAAC;IAC9J,OAAO,EAAE,KAAK,EAAE,mBAAmB,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,OAAO,EAAE,OAAO,CAAC,OAAO,EAAE,SAAS,EAAE,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,gBAAgB,CAAC,CAAC;QACnJ,sBAAsB,EAAE,UAAU,CAAC,MAAM,CAAC,EAAE,uBAAuB,EAAE,KAAK,CAAC,qBAAqB;QAChG,aAAa,EAAE,UAAU,CAAC,aAAa,CAAC,EAAE,qBAAqB,EAAE,eAAe,CAAC,UAAU;QAC3F,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,cAAc,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,eAAe,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,QAAQ,EAAE,KAAK,CAAC,SAAS,CAAC,EAAE,CAAC;AAC5H,CAAC;AACD,SAAS,WAAW,CAAC,OAAe,IAAgB,OAAO,IAAI,UAAU,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,gCAAgC,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC"}
@@ -0,0 +1,21 @@
1
+ import path from "node:path";
2
+ import { HitchError, readJSON, sha256JSON } from "../foundation/index.js";
3
+ import { assertRemoteVerifierWork } from "./remote-verifier-contract.js";
4
+ export function remoteVerifierWorkIdentity(input) {
5
+ assertRemoteVerifierWork({ verifier: input.verifier, plan: input.plan, work: input.work, physical: input.physical,
6
+ runtimeId: input.verifier.source_manifest.controller_runtime_id });
7
+ if (input.physical.kind !== "verifier-only" || input.request.training_binding)
8
+ throw ambiguous();
9
+ return { schema_version: "2", execution_kind: "verifier-only", rerun_id: input.physical.rerun_id, source_work_id: input.physical.source_work_id,
10
+ work_id: input.work.work_id, plan_digest: sha256JSON(input.plan), request_digest: sha256JSON(input.request), verifier_work_digest: sha256JSON(input.verifier) };
11
+ }
12
+ /** A scoring dispatch must already have an immutable journal for crash recovery. */
13
+ export async function assertRemoteVerifierDispatch(input) {
14
+ const identity = remoteVerifierWorkIdentity(input);
15
+ const file = path.join(input.evalDirectory, "reruns", identity.rerun_id, "remote-work", `${identity.work_id}.json`);
16
+ const record = await readJSON(file, null);
17
+ if (!record || record.state !== "dispatching" || record.result || sha256JSON(record.identity) !== sha256JSON(identity))
18
+ throw ambiguous();
19
+ }
20
+ function ambiguous() { return new HitchError("remote verifier dispatch requires its frozen durable work identity", { code: "execution_state_ambiguous", exitCode: 12 }); }
21
+ //# sourceMappingURL=remote-verifier-journal.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"remote-verifier-journal.js","sourceRoot":"","sources":["../../../src/evals/remote-verifier-journal.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAC1E,OAAO,EAAE,wBAAwB,EAAE,MAAM,+BAA+B,CAAC;AAEzE,MAAM,UAAU,0BAA0B,CAAC,KAE1C;IACC,wBAAwB,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ;QAC/G,SAAS,EAAE,KAAK,CAAC,QAAQ,CAAC,eAAe,CAAC,qBAAqB,EAAE,CAAC,CAAC;IACrE,IAAI,KAAK,CAAC,QAAQ,CAAC,IAAI,KAAK,eAAe,IAAI,KAAK,CAAC,OAAO,CAAC,gBAAgB;QAAE,MAAM,SAAS,EAAE,CAAC;IACjG,OAAO,EAAE,cAAc,EAAE,GAAG,EAAE,cAAc,EAAE,eAAe,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,CAAC,QAAQ,EAAE,cAAc,EAAE,KAAK,CAAC,QAAQ,CAAC,cAAc;QAC7I,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,OAAO,EAAE,WAAW,EAAE,UAAU,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,cAAc,EAAE,UAAU,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,oBAAoB,EAAE,UAAU,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC;AACpK,CAAC;AAED,oFAAoF;AACpF,MAAM,CAAC,KAAK,UAAU,4BAA4B,CAAC,KAAmF;IACpI,MAAM,QAAQ,GAAG,0BAA0B,CAAC,KAAK,CAAC,CAAC;IACnD,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,QAAQ,EAAE,QAAQ,CAAC,QAAQ,EAAE,aAAa,EAAE,GAAG,QAAQ,CAAC,OAAO,OAAO,CAAC,CAAC;IACpH,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAgE,IAAI,EAAE,IAAI,CAAC,CAAC;IACzG,IAAI,CAAC,MAAM,IAAI,MAAM,CAAC,KAAK,KAAK,aAAa,IAAI,MAAM,CAAC,MAAM,IAAI,UAAU,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,UAAU,CAAC,QAAQ,CAAC;QAAE,MAAM,SAAS,EAAE,CAAC;AAC5I,CAAC;AACD,SAAS,SAAS,KAAiB,OAAO,IAAI,UAAU,CAAC,oEAAoE,EAAE,EAAE,IAAI,EAAE,2BAA2B,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC"}
@@ -0,0 +1,51 @@
1
+ import { sha256JSON } from "../foundation/index.js";
2
+ import { parseExecutionEvidence } from "./execution-evidence.js";
3
+ import { runtimeResourcesForTask } from "./execution-plan-resources.js";
4
+ const SHA = /^sha256:[a-f0-9]{64}$/;
5
+ export function parseRemoteVerifierOutcome(value) {
6
+ if (!object(value))
7
+ throw invalid();
8
+ const keys = ["trial", "backend", "execution", "config_digest", "controller_runtime_id", ...(value.runtime_repair === undefined ? [] : ["runtime_repair"])];
9
+ if (Object.keys(value).length !== keys.length || Object.keys(value).some(key => !keys.includes(key))
10
+ || !object(value.trial) || Object.hasOwn(value.trial, "config")
11
+ || value.trial.agent_setup != null || value.trial.agent_execution != null
12
+ || typeof value.trial.trial_name !== "string" || !/^[A-Za-z0-9][A-Za-z0-9._-]{0,255}$/.test(value.trial.trial_name)
13
+ || !object(value.backend) || Object.keys(value.backend).sort().join(",") !== "process_exit_code,signal"
14
+ || !(value.backend.process_exit_code === null || Number.isSafeInteger(value.backend.process_exit_code)
15
+ && Number(value.backend.process_exit_code) >= 0 && Number(value.backend.process_exit_code) <= 255)
16
+ || !(value.backend.signal === null || typeof value.backend.signal === "string" && /^SIG[A-Z0-9]{1,24}$/.test(value.backend.signal))
17
+ || ![value.config_digest, value.controller_runtime_id].every(v => typeof v === "string" && SHA.test(v))
18
+ || value.runtime_repair !== undefined && !validRepair(value.runtime_repair, value.controller_runtime_id))
19
+ throw invalid();
20
+ return { ...structuredClone(value), execution: parseExecutionEvidence(value.execution) };
21
+ }
22
+ export function assertRemoteVerifierOutcome(input) {
23
+ const o = parseRemoteVerifierOutcome(input.outcome), v = input.verifier, e = o.execution, l = input.lease;
24
+ const limits = runtimeResourcesForTask(input.plan, v.source_ref.task_id, input.work.reservation);
25
+ if (input.work.eval_id !== l.eval_id || input.work.work_id !== l.work_id || input.work.provider !== l.provider
26
+ || e.eval_id !== l.eval_id || e.work_id !== l.work_id || e.lease_id !== l.lease_id || e.lease_epoch !== l.epoch
27
+ || e.worker_id !== l.worker_id || e.provider !== l.provider || e.collision_domain_id !== l.collision_domain_id
28
+ || e.task_id !== v.source_ref.task_id || sha256JSON(e.reservation) !== sha256JSON(input.work.reservation)
29
+ || sha256JSON(l.reservation) !== sha256JSON(e.reservation)
30
+ || sha256JSON(e.enforced.main_limits) !== sha256JSON(limits.mainLimits) || sha256JSON(e.enforced.sidecar_limits) !== sha256JSON(limits.sidecarLimits)
31
+ || o.controller_runtime_id !== v.verifier_runtime_id || o.trial.task_name !== v.source_trial.task_name
32
+ || o.trial.trial_name !== remoteVerifierTrialName(v.assessment_id)
33
+ || sha256JSON(o.trial.agent_result) !== sha256JSON(v.source_trial.agent_result))
34
+ throw invalid();
35
+ }
36
+ export function remoteVerifierTrialName(assessmentId) {
37
+ if (!/^assessment_[a-f0-9]{32}$/.test(assessmentId))
38
+ throw invalid();
39
+ return `regrade__${assessmentId}`;
40
+ }
41
+ function object(value) { return !!value && typeof value === "object" && !Array.isArray(value); }
42
+ function validRepair(value, runtimeId) {
43
+ const keys = ["schema_version", "kind", "source_runtime_id", "replacement_runtime_id", "path", "source_sha256", "replacement_sha256", "unchanged_file_count"];
44
+ return object(value) && Object.keys(value).length === keys.length && Object.keys(value).every(key => keys.includes(key))
45
+ && value.schema_version === "1" && value.kind === "verifier-environment-runtime-repair"
46
+ && value.path === "integrations/harbor/hitch_harbor_environment.py" && value.replacement_runtime_id === runtimeId
47
+ && [value.source_runtime_id, value.replacement_runtime_id, value.source_sha256, value.replacement_sha256].every(v => typeof v === "string" && SHA.test(v))
48
+ && Number.isSafeInteger(value.unchanged_file_count) && Number(value.unchanged_file_count) >= 0;
49
+ }
50
+ function invalid() { return new TypeError("remote verifier outcome differs from its scoring work, candidate or execution lease"); }
51
+ //# sourceMappingURL=remote-verifier-outcome.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"remote-verifier-outcome.js","sourceRoot":"","sources":["../../../src/evals/remote-verifier-outcome.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AACpD,OAAO,EAAE,sBAAsB,EAAE,MAAM,yBAAyB,CAAC;AACjE,OAAO,EAAE,uBAAuB,EAAE,MAAM,+BAA+B,CAAC;AAExE,MAAM,GAAG,GAAG,uBAAuB,CAAC;AACpC,MAAM,UAAU,0BAA0B,CAAC,KAAc;IACvD,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC;QAAE,MAAM,OAAO,EAAE,CAAC;IACpC,MAAM,IAAI,GAAG,CAAC,OAAO,EAAE,SAAS,EAAE,WAAW,EAAE,eAAe,EAAE,uBAAuB,EAAE,GAAG,CAAC,KAAK,CAAC,cAAc,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC;IAC5J,IAAI,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,KAAK,IAAI,CAAC,MAAM,IAAI,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;WAC/F,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC;WAC5D,KAAK,CAAC,KAAK,CAAC,WAAW,IAAI,IAAI,IAAI,KAAK,CAAC,KAAK,CAAC,eAAe,IAAI,IAAI;WACtE,OAAO,KAAK,CAAC,KAAK,CAAC,UAAU,KAAK,QAAQ,IAAI,CAAC,oCAAoC,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,UAAU,CAAC;WAChH,CAAC,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,0BAA0B;WACpG,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,iBAAiB,KAAK,IAAI,IAAI,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,OAAO,CAAC,iBAAiB,CAAC;eACjG,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,iBAAiB,CAAC,IAAI,CAAC,IAAI,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,iBAAiB,CAAC,IAAI,GAAG,CAAC;WACjG,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,KAAK,IAAI,IAAI,OAAO,KAAK,CAAC,OAAO,CAAC,MAAM,KAAK,QAAQ,IAAI,qBAAqB,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;WAChI,CAAC,CAAC,KAAK,CAAC,aAAa,EAAE,KAAK,CAAC,qBAAqB,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,IAAI,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;WACpG,KAAK,CAAC,cAAc,KAAK,SAAS,IAAI,CAAC,WAAW,CAAC,KAAK,CAAC,cAAc,EAAE,KAAK,CAAC,qBAAqB,CAAC;QAAE,MAAM,OAAO,EAAE,CAAC;IAC5H,OAAO,EAAE,GAAG,eAAe,CAAC,KAAK,CAAC,EAAE,SAAS,EAAE,sBAAsB,CAAC,KAAK,CAAC,SAAS,CAAC,EAAwC,CAAC;AACjI,CAAC;AAED,MAAM,UAAU,2BAA2B,CAAC,KAE3C;IACC,MAAM,CAAC,GAAG,0BAA0B,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,QAAQ,EAAE,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,CAAC,GAAG,KAAK,CAAC,KAAK,CAAC;IAC1G,MAAM,MAAM,GAAG,uBAAuB,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,UAAU,CAAC,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;IACjG,IAAI,KAAK,CAAC,IAAI,CAAC,OAAO,KAAK,CAAC,CAAC,OAAO,IAAI,KAAK,CAAC,IAAI,CAAC,OAAO,KAAK,CAAC,CAAC,OAAO,IAAI,KAAK,CAAC,IAAI,CAAC,QAAQ,KAAK,CAAC,CAAC,QAAQ;WACzG,CAAC,CAAC,OAAO,KAAK,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,OAAO,KAAK,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,QAAQ,KAAK,CAAC,CAAC,QAAQ,IAAI,CAAC,CAAC,WAAW,KAAK,CAAC,CAAC,KAAK;WAC5G,CAAC,CAAC,SAAS,KAAK,CAAC,CAAC,SAAS,IAAI,CAAC,CAAC,QAAQ,KAAK,CAAC,CAAC,QAAQ,IAAI,CAAC,CAAC,mBAAmB,KAAK,CAAC,CAAC,mBAAmB;WAC3G,CAAC,CAAC,OAAO,KAAK,CAAC,CAAC,UAAU,CAAC,OAAO,IAAI,UAAU,CAAC,CAAC,CAAC,WAAW,CAAC,KAAK,UAAU,CAAC,KAAK,CAAC,IAAI,CAAC,WAAW,CAAC;WACtG,UAAU,CAAC,CAAC,CAAC,WAAW,CAAC,KAAK,UAAU,CAAC,CAAC,CAAC,WAAW,CAAC;WACvD,UAAU,CAAC,CAAC,CAAC,QAAQ,CAAC,WAAW,CAAC,KAAK,UAAU,CAAC,MAAM,CAAC,UAAU,CAAC,IAAI,UAAU,CAAC,CAAC,CAAC,QAAQ,CAAC,cAAc,CAAC,KAAK,UAAU,CAAC,MAAM,CAAC,aAAa,CAAC;WAClJ,CAAC,CAAC,qBAAqB,KAAK,CAAC,CAAC,mBAAmB,IAAI,CAAC,CAAC,KAAK,CAAC,SAAS,KAAK,CAAC,CAAC,YAAY,CAAC,SAAS;WACnG,CAAC,CAAC,KAAK,CAAC,UAAU,KAAK,uBAAuB,CAAC,CAAC,CAAC,aAAa,CAAC;WAC/D,UAAU,CAAC,CAAC,CAAC,KAAK,CAAC,YAAY,CAAC,KAAK,UAAU,CAAC,CAAC,CAAC,YAAY,CAAC,YAAY,CAAC;QAAE,MAAM,OAAO,EAAE,CAAC;AACrG,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,YAAoB;IAC1D,IAAI,CAAC,2BAA2B,CAAC,IAAI,CAAC,YAAY,CAAC;QAAE,MAAM,OAAO,EAAE,CAAC;IACrE,OAAO,YAAY,YAAY,EAAE,CAAC;AACpC,CAAC;AACD,SAAS,MAAM,CAAC,KAAc,IAAsC,OAAO,CAAC,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;AAC3I,SAAS,WAAW,CAAC,KAAc,EAAE,SAAkB;IACrD,MAAM,IAAI,GAAG,CAAC,gBAAgB,EAAE,MAAM,EAAE,mBAAmB,EAAE,wBAAwB,EAAE,MAAM,EAAE,eAAe,EAAE,oBAAoB,EAAE,sBAAsB,CAAC,CAAC;IAC9J,OAAO,MAAM,CAAC,KAAK,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,KAAK,IAAI,CAAC,MAAM,IAAI,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;WACnH,KAAK,CAAC,cAAc,KAAK,GAAG,IAAI,KAAK,CAAC,IAAI,KAAK,qCAAqC;WACpF,KAAK,CAAC,IAAI,KAAK,iDAAiD,IAAI,KAAK,CAAC,sBAAsB,KAAK,SAAS;WAC9G,CAAC,KAAK,CAAC,iBAAiB,EAAE,KAAK,CAAC,sBAAsB,EAAE,KAAK,CAAC,aAAa,EAAE,KAAK,CAAC,kBAAkB,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,IAAI,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;WACvJ,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,oBAAoB,CAAC,IAAI,MAAM,CAAC,KAAK,CAAC,oBAAoB,CAAC,IAAI,CAAC,CAAC;AACnG,CAAC;AACD,SAAS,OAAO,KAAgB,OAAO,IAAI,SAAS,CAAC,qFAAqF,CAAC,CAAC,CAAC,CAAC"}
@@ -0,0 +1,134 @@
1
+ import path from "node:path";
2
+ import { useControllerRuntimeById } from "../controller-runtime/index.js";
3
+ import { HitchError, atomicWriteJSON, readJSON, sha256JSON, statePaths } from "../foundation/index.js";
4
+ import { validateEvalId } from "./request.js";
5
+ import { EvalEventSink } from "./events.js";
6
+ import { readExecutionLeases } from "./execution-leases.js";
7
+ import { parseEvalExecutionPlan } from "./execution-plan.js";
8
+ import { replaceInvalidEvalProgressTrial, writeEvalProgress } from "./progress.js";
9
+ import { readRegradeObservation } from "./regrade-evidence.js";
10
+ import { remoteVerifierWorkIdentity } from "./remote-verifier-journal.js";
11
+ import { assertRemoteVerifierPublication, freezeRemoteVerifierSelection } from "./remote-verifier-selection.js";
12
+ import { loadRerunPreparedArtifacts, loadRerunResolvedRevision, parseRerunPlan } from "./rerun-inputs.js";
13
+ import { validateEvalTrialReferences } from "./trial-import.js";
14
+ import { frozenRerunBenchmark } from "./verifier-only-rerun.js";
15
+ import { finishVerifierRerun } from "./verifier-rerun-result.js";
16
+ import { verifierRuntimeRepair } from "./verifier-runtime.js";
17
+ import { preparedArtifactForWorkItem } from "./work-item-artifacts.js";
18
+ /** Execute only scoring, with a frozen source selection and a durable per-work journal. */
19
+ export async function remoteVerifierOnlyEvalRerun(input) {
20
+ const evalId = validateEvalId(input.evalId);
21
+ const execution = parseEvalExecutionPlan(await readJSON(path.join(input.evalDirectory, "execution-plan.json")));
22
+ const benchmark = await frozenRerunBenchmark(input.evalDirectory);
23
+ if (!benchmark || execution.provider === "local-docker" || execution.membership !== "known" || execution.eval_id !== input.evalId
24
+ || benchmark.id !== input.request.benchmark_id || benchmark.revision !== input.request.benchmark_revision || input.request.training_binding
25
+ || !input.remoteWorkExecutor || input.executionWorker?.provider !== execution.provider)
26
+ throw ambiguous("remote scoring requires the frozen benchmark, provider and executor");
27
+ const plan = parseRerunPlan(await readJSON(path.join(input.evalDirectory, "plan.json")), input.evalId, input.request);
28
+ const resolution = await loadRerunResolvedRevision(input.evalDirectory, plan), artifacts = await loadRerunPreparedArtifacts(input.root, plan);
29
+ const runtimeId = String(plan.controllerRuntime.runtime_id);
30
+ const runtime = await useControllerRuntimeById(statePaths(input.root), runtimeId.replace(/^sha256:/, ""));
31
+ if (runtime.runtime_id !== runtimeId || runtime.manifest_digest !== plan.controllerRuntime.manifest_digest)
32
+ throw ambiguous("source controller runtime changed");
33
+ const verifierRuntime = input.verifierRuntimeId ? await useControllerRuntimeById(statePaths(input.root), input.verifierRuntimeId.slice(7)) : runtime;
34
+ verifierRuntimeRepair(runtime, verifierRuntime);
35
+ const selection = await freezeRemoteVerifierSelection({ ...input, plan: execution, slots: input.selectedTrials,
36
+ taskRoot: benchmark.tasks, sourceRuntimeId: runtimeId, verifierRuntimeId: verifierRuntime.runtime_id });
37
+ const artifactFor = (work) => {
38
+ if (!work.artifact_id && artifacts.size !== 1)
39
+ throw ambiguous("scoring source has no unique frozen artifact");
40
+ return preparedArtifactForWorkItem({ preparedArtifact: artifacts.values().next().value, preparedArtifacts: artifacts }, work);
41
+ };
42
+ for (const item of selection)
43
+ artifactFor(item.work);
44
+ let progress = input.progress;
45
+ const repaired = [], sources = [];
46
+ for (const item of selection) {
47
+ input.signal?.throwIfAborted();
48
+ const { work, physical, source, record: { descriptor, slot } } = item;
49
+ const identity = remoteVerifierWorkIdentity({ verifier: descriptor, physical, plan: execution, request: input.request, work });
50
+ const file = path.join(input.rerunDirectory, "remote-work", `${work.work_id}.json`);
51
+ const previous = await readJSON(file, null);
52
+ const released = async (leaseId) => {
53
+ if (!(await readExecutionLeases(input.evalDirectory)).some(lease => lease.lease_id === leaseId && lease.work_id === work.work_id
54
+ && lease.provider === execution.provider && lease.state === "released"))
55
+ throw ambiguous("scoring work requires confirmed release of its own lease");
56
+ };
57
+ if (previous && sha256JSON(previous.identity) !== sha256JSON(identity))
58
+ throw ambiguous("scoring work journal identity changed");
59
+ if (previous) {
60
+ if (previous.state === "not-started" && !previous.result)
61
+ await released(previous.lease_id);
62
+ else if (previous.state === "completed" && previous.result)
63
+ await released(previous.result.leaseId);
64
+ else
65
+ throw ambiguous("scoring work was dispatched and requires lease reconciliation");
66
+ }
67
+ const published = [];
68
+ const publish = async (ref) => {
69
+ assertRemoteVerifierPublication(descriptor.source_ref, ref, descriptor.assessment_id);
70
+ if (published.length && sha256JSON(published[0]) !== sha256JSON(ref))
71
+ throw ambiguous("scoring work changed its acknowledged result");
72
+ if (published.length)
73
+ return;
74
+ if (ref.observation_status === "valid") {
75
+ await validateEvalTrialReferences(input.root, input.evalId, [ref], { benchmarkId: benchmark.id, benchmarkRevision: benchmark.revision });
76
+ progress = replaceInvalidEvalProgressTrial(progress, ref);
77
+ await writeEvalProgress(input.evalDirectory, progress);
78
+ }
79
+ published.push(ref);
80
+ };
81
+ let result = previous?.state === "completed" ? previous.result : undefined;
82
+ if (!result) {
83
+ await atomicWriteJSON(file, { identity, state: "dispatching" });
84
+ const sink = new EvalEventSink(path.join(input.rerunDirectory, "remote-work", work.work_id), evalId);
85
+ await sink.open();
86
+ result = await input.remoteWorkExecutor({ root: input.root, evalId, evalDirectory: input.evalDirectory,
87
+ request: input.request, plan: execution, workItem: work, physicalExecution: physical,
88
+ preparedArtifact: artifactFor(work), resolvedRevision: resolution, runtimeDirectory: runtime.directory, runtimeId,
89
+ verifierOnly: { descriptor, sourceSnapshotDirectory: source.sourceSnapshotDirectory, verifierRuntimeDirectory: verifierRuntime.directory },
90
+ publicationMode: "replace-invalid", ...(input.signal ? { signal: input.signal } : {}), publish,
91
+ emit: event => sink.emit({ ...event, rerun_id: input.rerunId, execution_kind: "verifier-only", source_work_id: physical.source_work_id }),
92
+ onLeaseState: async (leaseId, state) => {
93
+ const current = await readJSON(file);
94
+ if (sha256JSON(current.identity) !== sha256JSON(identity) || current.lease_id && current.lease_id !== leaseId
95
+ || current.result && current.result.leaseId !== leaseId)
96
+ throw ambiguous("scoring lease callback changed journal ownership");
97
+ // The coordinator collects before release. Its terminal callback must never erase the result.
98
+ if (state === "terminal" && ["collected", "completed"].includes(current.state) && current.result)
99
+ return;
100
+ if (state === "running" && current.state !== "dispatching" || state === "terminal" && !["running", "terminal"].includes(current.state))
101
+ throw ambiguous("scoring lease callback regressed durable state");
102
+ await atomicWriteJSON(file, { ...current, lease_id: leaseId, state });
103
+ },
104
+ }).finally(() => sink.close());
105
+ await released(result.leaseId);
106
+ if (sha256JSON(result.refs) !== sha256JSON(published))
107
+ throw ambiguous("scoring result differs from acknowledged references");
108
+ const current = await readJSON(file);
109
+ if (current.result && sha256JSON(current.result) !== sha256JSON(result))
110
+ throw ambiguous("scoring executor changed its collected result");
111
+ await atomicWriteJSON(file, { identity, state: "completed", lease_id: result.leaseId, result });
112
+ }
113
+ if (!result.refs.length && result.run.rawResult === null && !result.assessments?.length) {
114
+ throw new HitchError("remote scoring failed before producing an assessment", { code: "eval_rerun_harbor_failed", exitCode: 13 });
115
+ }
116
+ const ref = result.refs[0], assessment = result.assessments?.[0];
117
+ if (result.refs.length !== 1 || result.assessments?.length !== 1 || !ref || assessment?.id !== descriptor.assessment_id)
118
+ throw ambiguous("scoring result has no unique assessment");
119
+ assertRemoteVerifierPublication(descriptor.source_ref, ref, descriptor.assessment_id);
120
+ await readRegradeObservation(input.root, input.evalId, { ...descriptor.source_ref, assessment });
121
+ const manifest = await readJSON(path.join(input.evalDirectory, "assessments", assessment.id, "assessment.json"));
122
+ if (manifest.rerun_id !== input.rerunId || manifest.remote_verifier_work_digest !== sha256JSON(descriptor)
123
+ || ref.observation_status === "valid" && sha256JSON(ref.assessment) !== sha256JSON(assessment))
124
+ throw ambiguous("scoring assessment belongs to different work");
125
+ await publish(ref);
126
+ if (ref.observation_status === "valid")
127
+ repaired.push(slot);
128
+ sources.push({ source_trial_id: descriptor.source_ref.trial_id, source_run_id: descriptor.source_ref.run_id, source_work_id: source.execution.work_id,
129
+ source_backend_directory: path.relative(input.evalDirectory, path.dirname(path.dirname(source.trialDirectory))).split(path.sep).join("/") });
130
+ }
131
+ return finishVerifierRerun({ ...input, progress, repaired, sources, remote: true });
132
+ }
133
+ function ambiguous(message) { return new HitchError(message, { code: "execution_state_ambiguous", exitCode: 12 }); }
134
+ //# sourceMappingURL=remote-verifier-rerun.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"remote-verifier-rerun.js","sourceRoot":"","sources":["../../../src/evals/remote-verifier-rerun.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,EAAE,wBAAwB,EAAE,MAAM,gCAAgC,CAAC;AAC1E,OAAO,EAAE,UAAU,EAAE,eAAe,EAAE,QAAQ,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AACvG,OAAO,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AAC9C,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAC5C,OAAO,EAAE,mBAAmB,EAAE,MAAM,uBAAuB,CAAC;AAC5D,OAAO,EAAE,sBAAsB,EAAE,MAAM,qBAAqB,CAAC;AAC7D,OAAO,EAAE,+BAA+B,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AACnF,OAAO,EAAE,sBAAsB,EAAE,MAAM,uBAAuB,CAAC;AAC/D,OAAO,EAAE,0BAA0B,EAAE,MAAM,8BAA8B,CAAC;AAC1E,OAAO,EAAE,+BAA+B,EAAE,6BAA6B,EAAE,MAAM,gCAAgC,CAAC;AAChH,OAAO,EAAE,0BAA0B,EAAE,yBAAyB,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AAI1G,OAAO,EAAE,2BAA2B,EAAE,MAAM,mBAAmB,CAAC;AAChE,OAAO,EAAE,oBAAoB,EAAE,MAAM,0BAA0B,CAAC;AAEhE,OAAO,EAAE,mBAAmB,EAAE,MAAM,4BAA4B,CAAC;AACjE,OAAO,EAAE,qBAAqB,EAAE,MAAM,uBAAuB,CAAC;AAC9D,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AAIvE,2FAA2F;AAC3F,MAAM,CAAC,KAAK,UAAU,2BAA2B,CAAC,KAAkD;IAClG,MAAM,MAAM,GAAG,cAAc,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IAC5C,MAAM,SAAS,GAAG,sBAAsB,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,qBAAqB,CAAC,CAAC,CAAC,CAAC;IAChH,MAAM,SAAS,GAAG,MAAM,oBAAoB,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;IAClE,IAAI,CAAC,SAAS,IAAI,SAAS,CAAC,QAAQ,KAAK,cAAc,IAAI,SAAS,CAAC,UAAU,KAAK,OAAO,IAAI,SAAS,CAAC,OAAO,KAAK,KAAK,CAAC,MAAM;WAC5H,SAAS,CAAC,EAAE,KAAK,KAAK,CAAC,OAAO,CAAC,YAAY,IAAI,SAAS,CAAC,QAAQ,KAAK,KAAK,CAAC,OAAO,CAAC,kBAAkB,IAAI,KAAK,CAAC,OAAO,CAAC,gBAAgB;WACxI,CAAC,KAAK,CAAC,kBAAkB,IAAI,KAAK,CAAC,eAAe,EAAE,QAAQ,KAAK,SAAS,CAAC,QAAQ;QAAE,MAAM,SAAS,CAAC,qEAAqE,CAAC,CAAC;IACjL,MAAM,IAAI,GAAG,cAAc,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,WAAW,CAAC,CAAC,EAAE,KAAK,CAAC,MAAM,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC;IACtH,MAAM,UAAU,GAAG,MAAM,yBAAyB,CAAC,KAAK,CAAC,aAAa,EAAE,IAAI,CAAC,EAAE,SAAS,GAAG,MAAM,0BAA0B,CAAC,KAAK,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;IAC9I,MAAM,SAAS,GAAG,MAAM,CAAC,IAAI,CAAC,iBAAiB,CAAC,UAAU,CAAC,CAAC;IAC5D,MAAM,OAAO,GAAG,MAAM,wBAAwB,CAAC,UAAU,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,SAAS,CAAC,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC,CAAC;IAC1G,IAAI,OAAO,CAAC,UAAU,KAAK,SAAS,IAAI,OAAO,CAAC,eAAe,KAAK,IAAI,CAAC,iBAAiB,CAAC,eAAe;QAAE,MAAM,SAAS,CAAC,mCAAmC,CAAC,CAAC;IACjK,MAAM,eAAe,GAAG,KAAK,CAAC,iBAAiB,CAAC,CAAC,CAAC,MAAM,wBAAwB,CAAC,UAAU,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,KAAK,CAAC,iBAAiB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;IACrJ,qBAAqB,CAAC,OAAO,EAAE,eAAe,CAAC,CAAC;IAChD,MAAM,SAAS,GAAG,MAAM,6BAA6B,CAAC,EAAE,GAAG,KAAK,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,EAAE,KAAK,CAAC,cAAc;QAC5G,QAAQ,EAAE,SAAS,CAAC,KAAK,EAAE,eAAe,EAAE,SAAS,EAAE,iBAAiB,EAAE,eAAe,CAAC,UAAU,EAAE,CAAC,CAAC;IAC1G,MAAM,WAAW,GAAG,CAAC,IAAyC,EAAE,EAAE;QAChE,IAAI,CAAC,IAAI,CAAC,WAAW,IAAI,SAAS,CAAC,IAAI,KAAK,CAAC;YAAE,MAAM,SAAS,CAAC,8CAA8C,CAAC,CAAC;QAC/G,OAAO,2BAA2B,CAAC,EAAE,gBAAgB,EAAE,SAAS,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,KAAM,EAAE,iBAAiB,EAAE,SAAS,EAAE,EAAE,IAAI,CAAC,CAAC;IACjI,CAAC,CAAC;IACF,KAAK,MAAM,IAAI,IAAI,SAAS;QAAE,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACrD,IAAI,QAAQ,GAAG,KAAK,CAAC,QAAQ,CAAC;IAC9B,MAAM,QAAQ,GAAoB,EAAE,EAAE,OAAO,GAA4C,EAAE,CAAC;IAC5F,KAAK,MAAM,IAAI,IAAI,SAAS,EAAE,CAAC;QAC7B,KAAK,CAAC,MAAM,EAAE,cAAc,EAAE,CAAC;QAC/B,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,MAAM,EAAE,EAAE,UAAU,EAAE,IAAI,EAAE,EAAE,GAAG,IAAI,CAAC;QACtE,MAAM,QAAQ,GAAG,0BAA0B,CAAC,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;QAC/H,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,cAAc,EAAE,aAAa,EAAE,GAAG,IAAI,CAAC,OAAO,OAAO,CAAC,CAAC;QACpF,MAAM,QAAQ,GAAG,MAAM,QAAQ,CAAiB,IAAI,EAAE,IAAI,CAAC,CAAC;QAC5D,MAAM,QAAQ,GAAG,KAAK,EAAE,OAA2B,EAAiB,EAAE;YACpE,IAAI,CAAC,CAAC,MAAM,mBAAmB,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,OAAO,IAAI,KAAK,CAAC,OAAO,KAAK,IAAI,CAAC,OAAO;mBAC3H,KAAK,CAAC,QAAQ,KAAK,SAAS,CAAC,QAAQ,IAAI,KAAK,CAAC,KAAK,KAAK,UAAU,CAAC;gBAAE,MAAM,SAAS,CAAC,0DAA0D,CAAC,CAAC;QACzJ,CAAC,CAAC;QACF,IAAI,QAAQ,IAAI,UAAU,CAAC,QAAQ,CAAC,QAAQ,CAAC,KAAK,UAAU,CAAC,QAAQ,CAAC;YAAE,MAAM,SAAS,CAAC,uCAAuC,CAAC,CAAC;QACjI,IAAI,QAAQ,EAAE,CAAC;YACb,IAAI,QAAQ,CAAC,KAAK,KAAK,aAAa,IAAI,CAAC,QAAQ,CAAC,MAAM;gBAAE,MAAM,QAAQ,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;iBACvF,IAAI,QAAQ,CAAC,KAAK,KAAK,WAAW,IAAI,QAAQ,CAAC,MAAM;gBAAE,MAAM,QAAQ,CAAC,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;;gBAC/F,MAAM,SAAS,CAAC,+DAA+D,CAAC,CAAC;QACxF,CAAC;QACD,MAAM,SAAS,GAA0C,EAAE,CAAC;QAC5D,MAAM,OAAO,GAAG,KAAK,EAAE,GAAkD,EAAiB,EAAE;YAC1F,+BAA+B,CAAC,UAAU,CAAC,UAAU,EAAE,GAAG,EAAE,UAAU,CAAC,aAAa,CAAC,CAAC;YACtF,IAAI,SAAS,CAAC,MAAM,IAAI,UAAU,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,KAAK,UAAU,CAAC,GAAG,CAAC;gBAAE,MAAM,SAAS,CAAC,8CAA8C,CAAC,CAAC;YACtI,IAAI,SAAS,CAAC,MAAM;gBAAE,OAAO;YAC7B,IAAI,GAAG,CAAC,kBAAkB,KAAK,OAAO,EAAE,CAAC;gBACvC,MAAM,2BAA2B,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,EAAE,EAAE,WAAW,EAAE,SAAS,CAAC,EAAE,EAAE,iBAAiB,EAAE,SAAS,CAAC,QAAQ,EAAE,CAAC,CAAC;gBACzI,QAAQ,GAAG,+BAA+B,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;gBAC1D,MAAM,iBAAiB,CAAC,KAAK,CAAC,aAAa,EAAE,QAAQ,CAAC,CAAC;YACzD,CAAC;YACD,SAAS,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACtB,CAAC,CAAC;QACF,IAAI,MAAM,GAAG,QAAQ,EAAE,KAAK,KAAK,WAAW,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC;QAC3E,IAAI,CAAC,MAAM,EAAE,CAAC;YACZ,MAAM,eAAe,CAAC,IAAI,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,aAAa,EAAE,CAAC,CAAC;YAChE,MAAM,IAAI,GAAG,IAAI,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,cAAc,EAAE,aAAa,EAAE,IAAI,CAAC,OAAO,CAAC,EAAE,MAAM,CAAC,CAAC;YACrG,MAAM,IAAI,CAAC,IAAI,EAAE,CAAC;YAClB,MAAM,GAAG,MAAM,KAAK,CAAC,kBAAkB,CAAC,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,MAAM,EAAE,aAAa,EAAE,KAAK,CAAC,aAAa;gBACpG,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,QAAQ,EAAE,IAAI,EAAE,iBAAiB,EAAE,QAAQ;gBACpF,gBAAgB,EAAE,WAAW,CAAC,IAAI,CAAC,EAAE,gBAAgB,EAAE,UAAU,EAAE,gBAAgB,EAAE,OAAO,CAAC,SAAS,EAAE,SAAS;gBACjH,YAAY,EAAE,EAAE,UAAU,EAAE,uBAAuB,EAAE,MAAM,CAAC,uBAAuB,EAAE,wBAAwB,EAAE,eAAe,CAAC,SAAS,EAAE;gBAC1I,eAAe,EAAE,iBAAiB,EAAE,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,OAAO;gBAC9F,IAAI,EAAE,KAAK,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,GAAG,KAAK,EAAE,QAAQ,EAAE,KAAK,CAAC,OAAO,EAAE,cAAc,EAAE,eAAe,EAAE,cAAc,EAAE,QAAQ,CAAC,cAAc,EAAE,CAAC;gBACzI,YAAY,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE;oBACrC,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAU,IAAI,CAAC,CAAC;oBAC9C,IAAI,UAAU,CAAC,OAAO,CAAC,QAAQ,CAAC,KAAK,UAAU,CAAC,QAAQ,CAAC,IAAI,OAAO,CAAC,QAAQ,IAAI,OAAO,CAAC,QAAQ,KAAK,OAAO;2BACxG,OAAO,CAAC,MAAM,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,KAAK,OAAO;wBAAE,MAAM,SAAS,CAAC,kDAAkD,CAAC,CAAC;oBAC/H,8FAA8F;oBAC9F,IAAI,KAAK,KAAK,UAAU,IAAI,CAAC,WAAW,EAAE,WAAW,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,KAAK,CAAC,IAAI,OAAO,CAAC,MAAM;wBAAE,OAAO;oBACzG,IAAI,KAAK,KAAK,SAAS,IAAI,OAAO,CAAC,KAAK,KAAK,aAAa,IAAI,KAAK,KAAK,UAAU,IAAI,CAAC,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,KAAK,CAAC;wBAAE,MAAM,SAAS,CAAC,gDAAgD,CAAC,CAAC;oBAC1M,MAAM,eAAe,CAAC,IAAI,EAAE,EAAE,GAAG,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC,CAAC;gBACxE,CAAC;aACF,CAAC,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;YAC/B,MAAM,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YAC/B,IAAI,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,UAAU,CAAC,SAAS,CAAC;gBAAE,MAAM,SAAS,CAAC,qDAAqD,CAAC,CAAC;YAC9H,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAU,IAAI,CAAC,CAAC;YAC9C,IAAI,OAAO,CAAC,MAAM,IAAI,UAAU,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,UAAU,CAAC,MAAM,CAAC;gBAAE,MAAM,SAAS,CAAC,+CAA+C,CAAC,CAAC;YAC1I,MAAM,eAAe,CAAC,IAAI,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,WAAW,EAAE,QAAQ,EAAE,MAAM,CAAC,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;QAClG,CAAC;QACD,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,IAAI,MAAM,CAAC,GAAG,CAAC,SAAS,KAAK,IAAI,IAAI,CAAC,MAAM,CAAC,WAAW,EAAE,MAAM,EAAE,CAAC;YACxF,MAAM,IAAI,UAAU,CAAC,sDAAsD,EAAE,EAAE,IAAI,EAAE,0BAA0B,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC;QACnI,CAAC;QACD,MAAM,GAAG,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,UAAU,GAAG,MAAM,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,CAAC;QACjE,IAAI,MAAM,CAAC,IAAI,CAAC,MAAM,KAAK,CAAC,IAAI,MAAM,CAAC,WAAW,EAAE,MAAM,KAAK,CAAC,IAAI,CAAC,GAAG,IAAI,UAAU,EAAE,EAAE,KAAK,UAAU,CAAC,aAAa;YAAE,MAAM,SAAS,CAAC,yCAAyC,CAAC,CAAC;QACpL,+BAA+B,CAAC,UAAU,CAAC,UAAU,EAAE,GAAG,EAAE,UAAU,CAAC,aAAa,CAAC,CAAC;QACtF,MAAM,sBAAsB,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,MAAM,EAAE,EAAE,GAAG,UAAU,CAAC,UAAU,EAAE,UAAU,EAAE,CAAC,CAAC;QACjG,MAAM,QAAQ,GAAG,MAAM,QAAQ,CAA0B,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,aAAa,EAAE,UAAU,CAAC,EAAE,EAAE,iBAAiB,CAAC,CAAC,CAAC;QAC1I,IAAI,QAAQ,CAAC,QAAQ,KAAK,KAAK,CAAC,OAAO,IAAI,QAAQ,CAAC,2BAA2B,KAAK,UAAU,CAAC,UAAU,CAAC;eACrG,GAAG,CAAC,kBAAkB,KAAK,OAAO,IAAI,UAAU,CAAC,GAAG,CAAC,UAAU,CAAC,KAAK,UAAU,CAAC,UAAU,CAAC;YAAE,MAAM,SAAS,CAAC,8CAA8C,CAAC,CAAC;QAClK,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC;QACnB,IAAI,GAAG,CAAC,kBAAkB,KAAK,OAAO;YAAE,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC5D,OAAO,CAAC,IAAI,CAAC,EAAE,eAAe,EAAE,UAAU,CAAC,UAAU,CAAC,QAAQ,EAAE,aAAa,EAAE,UAAU,CAAC,UAAU,CAAC,MAAO,EAAE,cAAc,EAAE,MAAM,CAAC,SAAS,CAAC,OAAO;YACpJ,wBAAwB,EAAE,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,aAAa,EAAE,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;IACjJ,CAAC;IACD,OAAO,mBAAmB,CAAC,EAAE,GAAG,KAAK,EAAE,QAAQ,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC;AACtF,CAAC;AACD,SAAS,SAAS,CAAC,OAAe,IAAgB,OAAO,IAAI,UAAU,CAAC,OAAO,EAAE,EAAE,IAAI,EAAE,2BAA2B,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC"}
@@ -0,0 +1,61 @@
1
+ import { readdir } from "node:fs/promises";
2
+ import path from "node:path";
3
+ import { HitchError, atomicWriteJSON, readJSON, sha256JSON } from "../foundation/index.js";
4
+ import { verifierOnlyWorkItem } from "./physical-work-plan.js";
5
+ import { assertRemoteVerifierWork, parseRemoteVerifierWork } from "./remote-verifier-contract.js";
6
+ import { loadRemoteVerifierSource } from "./remote-verifier-source.js";
7
+ export async function freezeRemoteVerifierSelection(input) {
8
+ const identity = { schema_version: "2", eval_id: input.evalId, rerun_id: input.rerunId,
9
+ plan_digest: sha256JSON(input.plan), request_digest: sha256JSON(input.request), selected_slots_digest: sha256JSON(input.slots),
10
+ source_runtime_id: input.sourceRuntimeId, verifier_runtime_id: input.verifierRuntimeId };
11
+ const file = path.join(input.rerunDirectory, "verifier-selection.json");
12
+ const saved = await readJSON(file, null);
13
+ if (saved && (sha256JSON(saved.identity) !== sha256JSON(identity) || !Array.isArray(saved.items) || saved.items.length !== input.slots.length))
14
+ throw ambiguous();
15
+ if (!saved && (await readdir(path.join(input.rerunDirectory, "remote-work")).catch(error => {
16
+ if (error.code === "ENOENT")
17
+ return [];
18
+ throw error;
19
+ })).length)
20
+ throw ambiguous();
21
+ const items = [];
22
+ for (const [index, slot] of input.slots.entries()) {
23
+ const frozen = saved?.items[index];
24
+ if (saved && (!frozen || sha256JSON(frozen.slot) !== sha256JSON(slot)))
25
+ throw ambiguous();
26
+ const original = frozen ? parseRemoteVerifierWork(frozen.descriptor).source_ref
27
+ : input.progress.trials.find(ref => ref.task_id === slot.task_id && ref.attempt === slot.attempt);
28
+ if (!original)
29
+ throw ambiguous();
30
+ const source = await loadRemoteVerifierSource({ ...input, sourceRef: original, provider: input.plan.provider,
31
+ taskDirectory: path.join(input.taskRoot, slot.task_id), benchmarkId: input.request.benchmark_id, benchmarkRevision: input.request.benchmark_revision });
32
+ const logical = input.plan.slots.find(s => s.task_id === slot.task_id && s.attempt === slot.attempt);
33
+ const planned = logical && input.plan.work_items.find(w => w.slots.includes(logical.slot_id));
34
+ if (!planned)
35
+ throw ambiguous();
36
+ const assessmentId = `assessment_${sha256JSON({ rerun_id: input.rerunId, source_ref: original,
37
+ source_manifest: source.manifest, verifier_runtime_id: input.verifierRuntimeId }).slice(7, 39)}`;
38
+ const descriptor = parseRemoteVerifierWork({ schema_version: "2", kind: "verifier-only", assessment_id: assessmentId,
39
+ source_ref: original, source_manifest: source.manifest, source_trial: source.sourceTrial, candidate_result: source.candidateResult,
40
+ candidate_result_digest: source.candidateResultDigest, canonical_bundle_digest: source.canonicalBundleDigest, verifier_runtime_id: input.verifierRuntimeId });
41
+ const work = verifierOnlyWorkItem(planned, input.rerunId, assessmentId);
42
+ const physical = { schema_version: "2", kind: "verifier-only", source_work_id: planned.work_id, rerun_id: input.rerunId, assessment_id: assessmentId };
43
+ assertRemoteVerifierWork({ verifier: descriptor, plan: input.plan, work, physical, runtimeId: input.sourceRuntimeId });
44
+ const record = { slot, source_work_id: planned.work_id, work_id: work.work_id, descriptor };
45
+ if (frozen && sha256JSON(frozen) !== sha256JSON(record))
46
+ throw ambiguous();
47
+ items.push({ record, work, physical, source });
48
+ }
49
+ // Seal the entire selection before the first offer, including undispatched slots.
50
+ if (!saved)
51
+ await atomicWriteJSON(file, { identity, items: items.map(item => item.record) });
52
+ return items;
53
+ }
54
+ export function assertRemoteVerifierPublication(original, ref, assessmentId) {
55
+ if (ref.task_id !== original.task_id || ref.attempt !== original.attempt || ref.trial_id !== original.trial_id || ref.run_id !== original.run_id
56
+ || ref.observation_status === "valid" && ref.assessment?.id !== assessmentId
57
+ || ref.observation_status !== "valid" && sha256JSON(ref) !== sha256JSON(original))
58
+ throw ambiguous();
59
+ }
60
+ function ambiguous() { return new HitchError("remote verifier selection or original candidate changed", { code: "execution_state_ambiguous", exitCode: 12 }); }
61
+ //# sourceMappingURL=remote-verifier-selection.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"remote-verifier-selection.js","sourceRoot":"","sources":["../../../src/evals/remote-verifier-selection.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,kBAAkB,CAAC;AAC3C,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,UAAU,EAAE,eAAe,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAC3F,OAAO,EAAE,oBAAoB,EAAE,MAAM,yBAAyB,CAAC;AAC/D,OAAO,EAAE,wBAAwB,EAAE,uBAAuB,EAAE,MAAM,+BAA+B,CAAC;AAClG,OAAO,EAAE,wBAAwB,EAAE,MAAM,6BAA6B,CAAC;AAGvE,MAAM,CAAC,KAAK,UAAU,6BAA6B,CAAC,KAInD;IACC,MAAM,QAAQ,GAAG,EAAE,cAAc,EAAE,GAAG,EAAE,OAAO,EAAE,KAAK,CAAC,MAAM,EAAE,QAAQ,EAAE,KAAK,CAAC,OAAO;QACpF,WAAW,EAAE,UAAU,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,cAAc,EAAE,UAAU,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,qBAAqB,EAAE,UAAU,CAAC,KAAK,CAAC,KAAK,CAAC;QAC9H,iBAAiB,EAAE,KAAK,CAAC,eAAe,EAAE,mBAAmB,EAAE,KAAK,CAAC,iBAAiB,EAAE,CAAC;IAC3F,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,cAAc,EAAE,yBAAyB,CAAC,CAAC;IACxE,MAAM,KAAK,GAAG,MAAM,QAAQ,CAAiJ,IAAI,EAAE,IAAI,CAAC,CAAC;IACzL,IAAI,KAAK,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,QAAQ,CAAC,KAAK,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC,KAAK,CAAC,MAAM,KAAK,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC;QAAE,MAAM,SAAS,EAAE,CAAC;IAClK,IAAI,CAAC,KAAK,IAAI,CAAC,MAAM,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,cAAc,EAAE,aAAa,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,EAAE;QACzF,IAAK,KAA+B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,EAAE,CAAC;QAClE,MAAM,KAAK,CAAC;IACd,CAAC,CAAC,CAAC,CAAC,MAAM;QAAE,MAAM,SAAS,EAAE,CAAC;IAC9B,MAAM,KAAK,GAAG,EAAE,CAAC;IACjB,KAAK,MAAM,CAAC,KAAK,EAAE,IAAI,CAAC,IAAI,KAAK,CAAC,KAAK,CAAC,OAAO,EAAE,EAAE,CAAC;QAClD,MAAM,MAAM,GAAG,KAAK,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC;QACnC,IAAI,KAAK,IAAI,CAAC,CAAC,MAAM,IAAI,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,UAAU,CAAC,IAAI,CAAC,CAAC;YAAE,MAAM,SAAS,EAAE,CAAC;QAC1F,MAAM,QAAQ,GAAG,MAAM,CAAC,CAAC,CAAC,uBAAuB,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,UAAU;YAC7E,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,OAAO,KAAK,IAAI,CAAC,OAAO,IAAI,GAAG,CAAC,OAAO,KAAK,IAAI,CAAC,OAAO,CAAC,CAAC;QACpG,IAAI,CAAC,QAAQ;YAAE,MAAM,SAAS,EAAE,CAAC;QACjC,MAAM,MAAM,GAAG,MAAM,wBAAwB,CAAC,EAAE,GAAG,KAAK,EAAE,SAAS,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,CAAC,IAAI,CAAC,QAAQ;YAC1G,aAAa,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,QAAQ,EAAE,IAAI,CAAC,OAAO,CAAC,EAAE,WAAW,EAAE,KAAK,CAAC,OAAO,CAAC,YAAY,EAAE,iBAAiB,EAAE,KAAK,CAAC,OAAO,CAAC,kBAAkB,EAAE,CAAC,CAAC;QAC1J,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,OAAO,KAAK,IAAI,CAAC,OAAO,IAAI,CAAC,CAAC,OAAO,KAAK,IAAI,CAAC,OAAO,CAAC,CAAC;QACrG,MAAM,OAAO,GAAG,OAAO,IAAI,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC;QAC9F,IAAI,CAAC,OAAO;YAAE,MAAM,SAAS,EAAE,CAAC;QAChC,MAAM,YAAY,GAAG,cAAc,UAAU,CAAC,EAAE,QAAQ,EAAE,KAAK,CAAC,OAAO,EAAE,UAAU,EAAE,QAAQ;YAC3F,eAAe,EAAE,MAAM,CAAC,QAAQ,EAAE,mBAAmB,EAAE,KAAK,CAAC,iBAAiB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,CAAC;QACnG,MAAM,UAAU,GAAG,uBAAuB,CAAC,EAAE,cAAc,EAAE,GAAG,EAAE,IAAI,EAAE,eAAe,EAAE,aAAa,EAAE,YAAY;YAClH,UAAU,EAAE,QAAQ,EAAE,eAAe,EAAE,MAAM,CAAC,QAAQ,EAAE,YAAY,EAAE,MAAM,CAAC,WAAW,EAAE,gBAAgB,EAAE,MAAM,CAAC,eAAe;YAClI,uBAAuB,EAAE,MAAM,CAAC,qBAAqB,EAAE,uBAAuB,EAAE,MAAM,CAAC,qBAAqB,EAAE,mBAAmB,EAAE,KAAK,CAAC,iBAAiB,EAAE,CAAC,CAAC;QAChK,MAAM,IAAI,GAAG,oBAAoB,CAAC,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;QACxE,MAAM,QAAQ,GAAG,EAAE,cAAc,EAAE,GAAY,EAAE,IAAI,EAAE,eAAwB,EAAE,cAAc,EAAE,OAAO,CAAC,OAAO,EAAE,QAAQ,EAAE,KAAK,CAAC,OAAO,EAAE,aAAa,EAAE,YAAY,EAAE,CAAC;QACzK,wBAAwB,CAAC,EAAE,QAAQ,EAAE,UAAU,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,IAAI,EAAE,QAAQ,EAAE,SAAS,EAAE,KAAK,CAAC,eAAe,EAAE,CAAC,CAAC;QACvH,MAAM,MAAM,GAAG,EAAE,IAAI,EAAE,cAAc,EAAE,OAAO,CAAC,OAAO,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,UAAU,EAAE,CAAC;QAC5F,IAAI,MAAM,IAAI,UAAU,CAAC,MAAM,CAAC,KAAK,UAAU,CAAC,MAAM,CAAC;YAAE,MAAM,SAAS,EAAE,CAAC;QAC3E,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC,CAAC;IACjD,CAAC;IACD,kFAAkF;IAClF,IAAI,CAAC,KAAK;QAAE,MAAM,eAAe,CAAC,IAAI,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;IAC7F,OAAO,KAAK,CAAC;AACf,CAAC;AAED,MAAM,UAAU,+BAA+B,CAAC,QAAwB,EAAE,GAAmB,EAAE,YAAoB;IACjH,IAAI,GAAG,CAAC,OAAO,KAAK,QAAQ,CAAC,OAAO,IAAI,GAAG,CAAC,OAAO,KAAK,QAAQ,CAAC,OAAO,IAAI,GAAG,CAAC,QAAQ,KAAK,QAAQ,CAAC,QAAQ,IAAI,GAAG,CAAC,MAAM,KAAK,QAAQ,CAAC,MAAM;WAC3I,GAAG,CAAC,kBAAkB,KAAK,OAAO,IAAI,GAAG,CAAC,UAAU,EAAE,EAAE,KAAK,YAAY;WACzE,GAAG,CAAC,kBAAkB,KAAK,OAAO,IAAI,UAAU,CAAC,GAAG,CAAC,KAAK,UAAU,CAAC,QAAQ,CAAC;QAAE,MAAM,SAAS,EAAE,CAAC;AACzG,CAAC;AACD,SAAS,SAAS,KAAiB,OAAO,IAAI,UAAU,CAAC,yDAAyD,EAAE,EAAE,IAAI,EAAE,2BAA2B,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC"}
@@ -0,0 +1,33 @@
1
+ import path from "node:path";
2
+ import { HitchError, readJSON, sha256JSON, statePaths } from "../foundation/index.js";
3
+ import { loadRunRecord, verifyResultBundleIndex } from "../runs/index.js";
4
+ import { parseExecutionEvidence } from "./execution-evidence.js";
5
+ import { parseEvalTrialRef } from "./progress.js";
6
+ import { validateEvalTrialReferences } from "./trial-import.js";
7
+ import { verifyImportedRemoteVerifierSource } from "./verifier-source-import.js";
8
+ /** Locate imported artifacts using the canonical candidate's actual execution, including candidate-restart work. */
9
+ export async function loadRemoteVerifierSource(input) {
10
+ const original = parseEvalTrialRef(input.sourceRef);
11
+ if (original.run_group || original.assessment || original.observation_status !== "invalid"
12
+ || !["verifier_infrastructure_failure", "verifier_result_missing"].includes(original.invalid_reason ?? "")
13
+ || !/^[A-Za-z0-9][A-Za-z0-9._-]{0,255}$/.test(original.trial_id))
14
+ throw invalid();
15
+ await validateEvalTrialReferences(input.root, input.evalId, [original], { benchmarkId: input.benchmarkId, benchmarkRevision: input.benchmarkRevision });
16
+ const runDirectory = path.join(statePaths(input.root).runs, original.run_id);
17
+ const bundle = await verifyResultBundleIndex(runDirectory), candidate = await loadRunRecord(runDirectory, { verifyTrajectory: true });
18
+ if (candidate.record.status !== "succeeded" || candidate.record_status !== "valid" || candidate.trajectory_status !== "valid")
19
+ throw invalid();
20
+ const execution = parseExecutionEvidence(await readJSON(path.join(runDirectory, "execution.json")));
21
+ if (execution.eval_id !== input.evalId || execution.provider !== input.provider || execution.task_id !== original.task_id)
22
+ throw invalid();
23
+ const trialDirectory = path.join(input.evalDirectory, "harbor", "work-items", execution.work_id, `epoch-${String(execution.lease_epoch).padStart(6, "0")}`, "job", original.trial_id);
24
+ const manifest = await verifyImportedRemoteVerifierSource({ root: input.root, runId: original.run_id, trialDirectory, taskDirectory: input.taskDirectory });
25
+ if (!manifest.original_result_digest || !manifest.regrade_config_digest)
26
+ throw invalid();
27
+ const sourceTrial = await readJSON(path.join(trialDirectory, "result.json"));
28
+ const candidateResult = await readJSON(path.join(runDirectory, "result.json"));
29
+ return { original, manifest, sourceTrial, candidateResult, execution, trialDirectory, sourceSnapshotDirectory: path.join(trialDirectory, "verifier-source"),
30
+ canonicalBundleDigest: sha256JSON(bundle), candidateResultDigest: sha256JSON(candidateResult) };
31
+ }
32
+ function invalid() { return new HitchError("remote verifier source has no intact original candidate and import receipt", { code: "eval_verifier_only_unavailable", exitCode: 2 }); }
33
+ //# sourceMappingURL=remote-verifier-source.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"remote-verifier-source.js","sourceRoot":"","sources":["../../../src/evals/remote-verifier-source.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AACtF,OAAO,EAAE,aAAa,EAAE,uBAAuB,EAAE,MAAM,kBAAkB,CAAC;AAC1E,OAAO,EAAE,sBAAsB,EAAE,MAAM,yBAAyB,CAAC;AACjE,OAAO,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AAClD,OAAO,EAAE,2BAA2B,EAAE,MAAM,mBAAmB,CAAC;AAChE,OAAO,EAAE,kCAAkC,EAAE,MAAM,6BAA6B,CAAC;AAEjF,oHAAoH;AACpH,MAAM,CAAC,KAAK,UAAU,wBAAwB,CAAC,KAG9C;IACC,MAAM,QAAQ,GAAG,iBAAiB,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;IACpD,IAAI,QAAQ,CAAC,SAAS,IAAI,QAAQ,CAAC,UAAU,IAAI,QAAQ,CAAC,kBAAkB,KAAK,SAAS;WACrF,CAAC,CAAC,iCAAiC,EAAE,yBAAyB,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,cAAc,IAAI,EAAE,CAAC;WACvG,CAAC,oCAAoC,CAAC,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC;QAAE,MAAM,OAAO,EAAE,CAAC;IACpF,MAAM,2BAA2B,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,EAAE,EAAE,WAAW,EAAE,KAAK,CAAC,WAAW,EAAE,iBAAiB,EAAE,KAAK,CAAC,iBAAiB,EAAE,CAAC,CAAC;IACxJ,MAAM,YAAY,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;IAC7E,MAAM,MAAM,GAAG,MAAM,uBAAuB,CAAC,YAAY,CAAC,EAAE,SAAS,GAAG,MAAM,aAAa,CAAC,YAAY,EAAE,EAAE,gBAAgB,EAAE,IAAI,EAAE,CAAC,CAAC;IACtI,IAAI,SAAS,CAAC,MAAM,CAAC,MAAM,KAAK,WAAW,IAAI,SAAS,CAAC,aAAa,KAAK,OAAO,IAAI,SAAS,CAAC,iBAAiB,KAAK,OAAO;QAAE,MAAM,OAAO,EAAE,CAAC;IAC/I,MAAM,SAAS,GAAG,sBAAsB,CAAC,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,gBAAgB,CAAC,CAAC,CAAC,CAAC;IACpG,IAAI,SAAS,CAAC,OAAO,KAAK,KAAK,CAAC,MAAM,IAAI,SAAS,CAAC,QAAQ,KAAK,KAAK,CAAC,QAAQ,IAAI,SAAS,CAAC,OAAO,KAAK,QAAQ,CAAC,OAAO;QAAE,MAAM,OAAO,EAAE,CAAC;IAC3I,MAAM,cAAc,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,QAAQ,EAAE,YAAY,EAAE,SAAS,CAAC,OAAO,EAC7F,SAAS,MAAM,CAAC,SAAS,CAAC,WAAW,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,EAAE,KAAK,EAAE,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACvF,MAAM,QAAQ,GAAG,MAAM,kCAAkC,CAAC,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,KAAK,EAAE,QAAQ,CAAC,MAAM,EAAE,cAAc,EAAE,aAAa,EAAE,KAAK,CAAC,aAAa,EAAE,CAAC,CAAC;IAC5J,IAAI,CAAC,QAAQ,CAAC,sBAAsB,IAAI,CAAC,QAAQ,CAAC,qBAAqB;QAAE,MAAM,OAAO,EAAE,CAAC;IACzF,MAAM,WAAW,GAAG,MAAM,QAAQ,CAA0B,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,aAAa,CAAC,CAAC,CAAC;IACtG,MAAM,eAAe,GAAG,MAAM,QAAQ,CAA0B,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,aAAa,CAAC,CAAC,CAAC;IACxG,OAAO,EAAE,QAAQ,EAAE,QAAQ,EAAE,WAAW,EAAE,eAAe,EAAE,SAAS,EAAE,cAAc,EAAE,uBAAuB,EAAE,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,iBAAiB,CAAC;QACzJ,qBAAqB,EAAE,UAAU,CAAC,MAAM,CAAC,EAAE,qBAAqB,EAAE,UAAU,CAAC,eAAe,CAAC,EAAE,CAAC;AACpG,CAAC;AACD,SAAS,OAAO,KAAiB,OAAO,IAAI,UAAU,CAAC,4EAA4E,EAAE,EAAE,IAAI,EAAE,gCAAgC,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC"}
@@ -0,0 +1,46 @@
1
+ import { HitchError, sha256JSON } from "../foundation/index.js";
2
+ import { parseRemoteExecutionAdmission, parseRemoteExecutionOwnership, parseRemoteWorkerHostIdentity } from "./remote-execution-ownership.js";
3
+ export function parseRemoteWorkerCleanupChallenge(value, offer) {
4
+ const record = exact(value, ["schema_version", "cleanup_id", "nonce", "worker_id", "generation", "issued_at", "expires_at", "admission"]);
5
+ const admission = parseRemoteExecutionAdmission(record.admission, offer);
6
+ if (record.schema_version !== "2" || typeof record.cleanup_id !== "string" || !/^cleanup_[a-f0-9]{32}$/.test(record.cleanup_id)
7
+ || typeof record.nonce !== "string" || !/^[a-f0-9]{32}$/.test(record.nonce) || record.worker_id !== offer.worker_id
8
+ || !Number.isSafeInteger(record.generation) || record.generation <= offer.generation
9
+ || !timestamp(record.issued_at) || !timestamp(record.expires_at) || Date.parse(record.expires_at) <= Date.parse(record.issued_at)
10
+ || Date.parse(record.expires_at) - Date.parse(record.issued_at) > 5 * 60_000
11
+ || !offer.accepted_at || !offer.accept_receipt_digest)
12
+ throw invalid();
13
+ return { ...record, admission };
14
+ }
15
+ export function parseRemoteWorkerCleanupReceipt(value, offer) {
16
+ const record = exact(value, ["schema_version", "challenge", "observed_at", "observation"]);
17
+ const challenge = parseRemoteWorkerCleanupChallenge(record.challenge, offer);
18
+ const reboot = record.schema_version === "3";
19
+ const observation = exact(record.observation, ["ownership", "execution_process", "worker_status", "docker_resources_empty", reboot ? "host_identity" : "process_group_empty"]);
20
+ const ownership = parseRemoteExecutionOwnership(observation.ownership, offer);
21
+ if (record.schema_version !== "2" && !reboot || !timestamp(record.observed_at) || Date.parse(record.observed_at) < Date.parse(challenge.issued_at)
22
+ || Date.parse(record.observed_at) > Date.parse(challenge.expires_at)
23
+ || sha256JSON(ownership) !== challenge.admission.ownership_digest
24
+ || sha256JSON(observation.execution_process) !== sha256JSON(challenge.admission.execution_process)
25
+ || observation.docker_resources_empty !== true)
26
+ throw invalid();
27
+ if (reboot) {
28
+ const host = parseRemoteWorkerHostIdentity(observation.host_identity);
29
+ if (ownership.schema_version !== "3" || observation.worker_status !== "previous-boot"
30
+ || host.platform !== ownership.host_identity.platform || host.host_id !== ownership.host_identity.host_id
31
+ || host.boot_id === ownership.host_identity.boot_id)
32
+ throw invalid();
33
+ }
34
+ else if (!["terminal", "identity-mismatch"].includes(String(observation.worker_status)) || observation.process_group_empty !== true)
35
+ throw invalid();
36
+ return { ...record, challenge, observation: { ...observation, ownership } };
37
+ }
38
+ function exact(value, fields) {
39
+ if (!value || typeof value !== "object" || Array.isArray(value) || Object.keys(value).length !== fields.length
40
+ || Object.keys(value).some(key => !fields.includes(key)))
41
+ throw invalid();
42
+ return value;
43
+ }
44
+ function timestamp(value) { return typeof value === "string" && Number.isFinite(Date.parse(value)); }
45
+ function invalid() { return new HitchError("worker generation cleanup does not match the original execution", { code: "remote_worker_cleanup_ambiguous", exitCode: 12 }); }
46
+ //# sourceMappingURL=remote-worker-cleanup.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"remote-worker-cleanup.js","sourceRoot":"","sources":["../../../src/evals/remote-worker-cleanup.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,wBAAwB,CAAC;AAChE,OAAO,EAAE,6BAA6B,EAAE,6BAA6B,EAAE,6BAA6B,EAAE,MAAM,iCAAiC,CAAC;AAE9I,MAAM,UAAU,iCAAiC,CAAC,KAAc,EAAE,KAAwB;IACxF,MAAM,MAAM,GAAG,KAAK,CAAC,KAAK,EAAE,CAAC,gBAAgB,EAAE,YAAY,EAAE,OAAO,EAAE,WAAW,EAAE,YAAY,EAAE,WAAW,EAAE,YAAY,EAAE,WAAW,CAAC,CAAC,CAAC;IAC1I,MAAM,SAAS,GAAG,6BAA6B,CAAC,MAAM,CAAC,SAAS,EAAE,KAAK,CAAC,CAAC;IACzE,IAAI,MAAM,CAAC,cAAc,KAAK,GAAG,IAAI,OAAO,MAAM,CAAC,UAAU,KAAK,QAAQ,IAAI,CAAC,wBAAwB,CAAC,IAAI,CAAC,MAAM,CAAC,UAAU,CAAC;WAC1H,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,IAAI,CAAC,gBAAgB,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,MAAM,CAAC,SAAS,KAAK,KAAK,CAAC,SAAS;WAChH,CAAC,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,UAAU,CAAC,IAAK,MAAM,CAAC,UAAqB,IAAI,KAAK,CAAC,UAAU;WAC7F,CAAC,SAAS,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,UAAU,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,UAAU,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,SAAS,CAAC;WAC9H,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,UAAU,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,MAAM;WACzE,CAAC,KAAK,CAAC,WAAW,IAAI,CAAC,KAAK,CAAC,qBAAqB;QAAE,MAAM,OAAO,EAAE,CAAC;IACzE,OAAO,EAAE,GAAG,MAAM,EAAE,SAAS,EAA+C,CAAC;AAC/E,CAAC;AAED,MAAM,UAAU,+BAA+B,CAAC,KAAc,EAAE,KAAwB;IACtF,MAAM,MAAM,GAAG,KAAK,CAAC,KAAK,EAAE,CAAC,gBAAgB,EAAE,WAAW,EAAE,aAAa,EAAE,aAAa,CAAC,CAAC,CAAC;IAC3F,MAAM,SAAS,GAAG,iCAAiC,CAAC,MAAM,CAAC,SAAS,EAAE,KAAK,CAAC,CAAC;IAC7E,MAAM,MAAM,GAAG,MAAM,CAAC,cAAc,KAAK,GAAG,CAAC;IAC7C,MAAM,WAAW,GAAG,KAAK,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC,WAAW,EAAE,mBAAmB,EAAE,eAAe,EAAE,wBAAwB,EAAE,MAAM,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,qBAAqB,CAAC,CAAC,CAAC;IAC/K,MAAM,SAAS,GAAG,6BAA6B,CAAC,WAAW,CAAC,SAAS,EAAE,KAAK,CAAC,CAAC;IAC9E,IAAI,MAAM,CAAC,cAAc,KAAK,GAAG,IAAI,CAAC,MAAM,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,WAAW,CAAC,IAAI,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,WAAW,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,SAAS,CAAC;WAC7I,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,WAAW,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,UAAU,CAAC;WACjE,UAAU,CAAC,SAAS,CAAC,KAAK,SAAS,CAAC,SAAS,CAAC,gBAAgB;WAC9D,UAAU,CAAC,WAAW,CAAC,iBAAiB,CAAC,KAAK,UAAU,CAAC,SAAS,CAAC,SAAS,CAAC,iBAAiB,CAAC;WAC/F,WAAW,CAAC,sBAAsB,KAAK,IAAI;QAAE,MAAM,OAAO,EAAE,CAAC;IAClE,IAAI,MAAM,EAAE,CAAC;QACX,MAAM,IAAI,GAAG,6BAA6B,CAAC,WAAW,CAAC,aAAa,CAAC,CAAC;QACtE,IAAI,SAAS,CAAC,cAAc,KAAK,GAAG,IAAI,WAAW,CAAC,aAAa,KAAK,eAAe;eAChF,IAAI,CAAC,QAAQ,KAAK,SAAS,CAAC,aAAa,CAAC,QAAQ,IAAI,IAAI,CAAC,OAAO,KAAK,SAAS,CAAC,aAAa,CAAC,OAAO;eACtG,IAAI,CAAC,OAAO,KAAK,SAAS,CAAC,aAAa,CAAC,OAAO;YAAE,MAAM,OAAO,EAAE,CAAC;IACzE,CAAC;SAAM,IAAI,CAAC,CAAC,UAAU,EAAE,mBAAmB,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,WAAW,CAAC,aAAa,CAAC,CAAC,IAAI,WAAW,CAAC,mBAAmB,KAAK,IAAI;QAAE,MAAM,OAAO,EAAE,CAAC;IACvJ,OAAO,EAAE,GAAG,MAAM,EAAE,SAAS,EAAE,WAAW,EAAE,EAAE,GAAG,WAAW,EAAE,SAAS,EAAE,EAA2C,CAAC;AACvH,CAAC;AAED,SAAS,KAAK,CAAC,KAAc,EAAE,MAAgB;IAC7C,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM;WACzG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;QAAE,MAAM,OAAO,EAAE,CAAC;IAC5E,OAAO,KAAgC,CAAC;AAC1C,CAAC;AACD,SAAS,SAAS,CAAC,KAAc,IAAqB,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;AAC/H,SAAS,OAAO,KAAiB,OAAO,IAAI,UAAU,CAAC,iEAAiE,EAAE,EAAE,IAAI,EAAE,iCAAiC,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC"}