@openpond/evals 0.11.21 → 0.12.1

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 (311) hide show
  1. package/README.md +12 -0
  2. package/dist/artifact-verification.js +2 -0
  3. package/dist/connected-evidence/codex-command-observations.js +81 -0
  4. package/dist/connected-evidence/contracts.js +13 -0
  5. package/dist/connected-evidence/git-file-contracts.js +16 -0
  6. package/dist/connected-evidence/git-file-proof.js +86 -0
  7. package/dist/connected-evidence/git-starting-state.js +81 -0
  8. package/dist/connected-evidence/git.js +2 -0
  9. package/dist/connected-evidence/imports.js +2 -1
  10. package/dist/connected-evidence/index.js +1 -0
  11. package/dist/connected-evidence/normalize.js +1 -0
  12. package/dist/connected-evidence/sources/claude-branches.js +34 -1
  13. package/dist/connected-evidence/sources/claude-code.js +1 -1
  14. package/dist/connected-evidence/sources/codex-metadata.js +33 -0
  15. package/dist/connected-evidence/sources/codex.js +16 -2
  16. package/dist/graders.js +10 -4
  17. package/dist/javascript-verifier.js +1 -0
  18. package/dist/native-conversations/codex-window-inventory.js +6 -7
  19. package/dist/native-conversations/collector-acquisition.js +17 -6
  20. package/dist/native-conversations/collector-source-cache.js +1 -1
  21. package/dist/native-conversations/collector-store.js +27 -2
  22. package/dist/native-conversations/collector.js +4 -3
  23. package/dist/native-conversations/git-assets.js +61 -0
  24. package/dist/native-conversations/history.js +7 -0
  25. package/dist/native-conversations/index.js +2 -0
  26. package/dist/native-conversations/selection.js +57 -0
  27. package/dist/native-conversations/supervisor.js +2 -0
  28. package/dist/prepared-process-isolation.js +131 -0
  29. package/dist/rewards.js +17 -5
  30. package/dist/tasksets.js +7 -0
  31. package/dist/types/artifact-verification.d.ts.map +1 -1
  32. package/dist/types/benchmarks.d.ts +22 -22
  33. package/dist/types/benchmarks.d.ts.map +1 -1
  34. package/dist/types/builtin-benchmarks/harness-refiner.d.ts +40 -35
  35. package/dist/types/builtin-benchmarks/harness-refiner.d.ts.map +1 -1
  36. package/dist/types/compatibility.d.ts.map +1 -1
  37. package/dist/types/conformance.d.ts +98 -88
  38. package/dist/types/conformance.d.ts.map +1 -1
  39. package/dist/types/connected-evidence/codex-command-observations.d.ts +15 -0
  40. package/dist/types/connected-evidence/codex-command-observations.d.ts.map +1 -0
  41. package/dist/types/connected-evidence/collector-coverage-contracts.d.ts +6 -6
  42. package/dist/types/connected-evidence/collector-coverage-contracts.d.ts.map +1 -1
  43. package/dist/types/connected-evidence/contracts.d.ts +80 -58
  44. package/dist/types/connected-evidence/contracts.d.ts.map +1 -1
  45. package/dist/types/connected-evidence/git-file-contracts.d.ts +18 -0
  46. package/dist/types/connected-evidence/git-file-contracts.d.ts.map +1 -0
  47. package/dist/types/connected-evidence/git-file-proof.d.ts +13 -0
  48. package/dist/types/connected-evidence/git-file-proof.d.ts.map +1 -0
  49. package/dist/types/connected-evidence/git-starting-state.d.ts +28 -0
  50. package/dist/types/connected-evidence/git-starting-state.d.ts.map +1 -0
  51. package/dist/types/connected-evidence/git.d.ts +3 -0
  52. package/dist/types/connected-evidence/git.d.ts.map +1 -0
  53. package/dist/types/connected-evidence/imports.d.ts.map +1 -1
  54. package/dist/types/connected-evidence/index.d.ts +1 -0
  55. package/dist/types/connected-evidence/index.d.ts.map +1 -1
  56. package/dist/types/connected-evidence/normalize.d.ts +1 -0
  57. package/dist/types/connected-evidence/normalize.d.ts.map +1 -1
  58. package/dist/types/connected-evidence/python-correctness.d.ts +38 -38
  59. package/dist/types/connected-evidence/python-correctness.d.ts.map +1 -1
  60. package/dist/types/connected-evidence/sources/claude-branches.d.ts +1 -0
  61. package/dist/types/connected-evidence/sources/claude-branches.d.ts.map +1 -1
  62. package/dist/types/connected-evidence/sources/claude-code.d.ts +25 -15
  63. package/dist/types/connected-evidence/sources/claude-code.d.ts.map +1 -1
  64. package/dist/types/connected-evidence/sources/codex-authorization.d.ts.map +1 -1
  65. package/dist/types/connected-evidence/sources/codex-metadata.d.ts +10 -0
  66. package/dist/types/connected-evidence/sources/codex-metadata.d.ts.map +1 -0
  67. package/dist/types/connected-evidence/sources/codex.d.ts +25 -15
  68. package/dist/types/connected-evidence/sources/codex.d.ts.map +1 -1
  69. package/dist/types/connected-evidence/sources/grok-build.d.ts +25 -15
  70. package/dist/types/connected-evidence/sources/grok-build.d.ts.map +1 -1
  71. package/dist/types/connected-evidence/sources/history.d.ts +50 -30
  72. package/dist/types/connected-evidence/sources/history.d.ts.map +1 -1
  73. package/dist/types/connected-evidence/sources/opencode.d.ts +25 -15
  74. package/dist/types/connected-evidence/sources/opencode.d.ts.map +1 -1
  75. package/dist/types/connected-evidence/sources/pi.d.ts +25 -15
  76. package/dist/types/connected-evidence/sources/pi.d.ts.map +1 -1
  77. package/dist/types/connected-evidence/sources/usage.d.ts.map +1 -1
  78. package/dist/types/connected-evidence/usage-summary.d.ts.map +1 -1
  79. package/dist/types/deterministic-graders.d.ts.map +1 -1
  80. package/dist/types/environment-execution-services-contract.d.ts +19 -19
  81. package/dist/types/environment-execution-services-contract.d.ts.map +1 -1
  82. package/dist/types/environment-execution-services.d.ts.map +1 -1
  83. package/dist/types/evidence/authoring.d.ts.map +1 -1
  84. package/dist/types/evidence/conformance.d.ts +98 -98
  85. package/dist/types/evidence/conformance.d.ts.map +1 -1
  86. package/dist/types/evidence/contracts.d.ts +183 -183
  87. package/dist/types/evidence/contracts.d.ts.map +1 -1
  88. package/dist/types/evidence/eligibility.d.ts +160 -160
  89. package/dist/types/evidence/eligibility.d.ts.map +1 -1
  90. package/dist/types/evidence/index.d.ts.map +1 -1
  91. package/dist/types/execution-contracts.d.ts +118 -110
  92. package/dist/types/execution-contracts.d.ts.map +1 -1
  93. package/dist/types/execution-receipts.d.ts.map +1 -1
  94. package/dist/types/experiments.d.ts +96 -96
  95. package/dist/types/experiments.d.ts.map +1 -1
  96. package/dist/types/graders.d.ts +38 -35
  97. package/dist/types/graders.d.ts.map +1 -1
  98. package/dist/types/harness-experiment-source.d.ts.map +1 -1
  99. package/dist/types/harness.d.ts.map +1 -1
  100. package/dist/types/human-review/comparison.d.ts +189 -189
  101. package/dist/types/human-review/comparison.d.ts.map +1 -1
  102. package/dist/types/human-review/contracts.d.ts +178 -178
  103. package/dist/types/human-review/contracts.d.ts.map +1 -1
  104. package/dist/types/human-review/execution.d.ts +53 -53
  105. package/dist/types/human-review/execution.d.ts.map +1 -1
  106. package/dist/types/human-review/handoff.d.ts +43 -43
  107. package/dist/types/human-review/handoff.d.ts.map +1 -1
  108. package/dist/types/human-review/index.d.ts.map +1 -1
  109. package/dist/types/human-review/local-publication.d.ts +4 -4
  110. package/dist/types/human-review/local-publication.d.ts.map +1 -1
  111. package/dist/types/human-review/repository.d.ts.map +1 -1
  112. package/dist/types/human-review/results.d.ts +8 -8
  113. package/dist/types/human-review/results.d.ts.map +1 -1
  114. package/dist/types/human-review/service.d.ts +292 -292
  115. package/dist/types/human-review/service.d.ts.map +1 -1
  116. package/dist/types/human-review/task-proposals.d.ts +2 -2
  117. package/dist/types/human-review/task-proposals.d.ts.map +1 -1
  118. package/dist/types/index.d.ts.map +1 -1
  119. package/dist/types/javascript-environment-attempt.d.ts.map +1 -1
  120. package/dist/types/javascript-environment-contract.d.ts +39 -39
  121. package/dist/types/javascript-environment-contract.d.ts.map +1 -1
  122. package/dist/types/javascript-environment-node.d.ts.map +1 -1
  123. package/dist/types/javascript-environment.d.ts +14 -14
  124. package/dist/types/javascript-environment.d.ts.map +1 -1
  125. package/dist/types/javascript-isolate-node.d.ts.map +1 -1
  126. package/dist/types/javascript-isolate-process-entry.d.ts.map +1 -1
  127. package/dist/types/javascript-isolate-process.d.ts.map +1 -1
  128. package/dist/types/javascript-isolate.d.ts.map +1 -1
  129. package/dist/types/javascript-verifier-contract.d.ts.map +1 -1
  130. package/dist/types/javascript-verifier-node.d.ts.map +1 -1
  131. package/dist/types/javascript-verifier-worker.d.ts.map +1 -1
  132. package/dist/types/javascript-verifier.d.ts +1 -0
  133. package/dist/types/javascript-verifier.d.ts.map +1 -1
  134. package/dist/types/learned-preference.d.ts +10 -10
  135. package/dist/types/learned-preference.d.ts.map +1 -1
  136. package/dist/types/learning/acceptance-group.d.ts +44 -44
  137. package/dist/types/learning/acceptance-group.d.ts.map +1 -1
  138. package/dist/types/learning/acceptance-plan.d.ts +8 -8
  139. package/dist/types/learning/acceptance-plan.d.ts.map +1 -1
  140. package/dist/types/learning/admission.d.ts +38 -36
  141. package/dist/types/learning/admission.d.ts.map +1 -1
  142. package/dist/types/learning/assets.d.ts +2 -2
  143. package/dist/types/learning/assets.d.ts.map +1 -1
  144. package/dist/types/learning/authoring-service.d.ts +11 -9
  145. package/dist/types/learning/authoring-service.d.ts.map +1 -1
  146. package/dist/types/learning/authoring.d.ts +46 -38
  147. package/dist/types/learning/authoring.d.ts.map +1 -1
  148. package/dist/types/learning/batch-service.d.ts +75 -68
  149. package/dist/types/learning/batch-service.d.ts.map +1 -1
  150. package/dist/types/learning/budgeted-reward-fixtures.d.ts.map +1 -1
  151. package/dist/types/learning/contracts.d.ts +314 -290
  152. package/dist/types/learning/contracts.d.ts.map +1 -1
  153. package/dist/types/learning/errors.d.ts.map +1 -1
  154. package/dist/types/learning/grade-judge-budget.d.ts.map +1 -1
  155. package/dist/types/learning/grade-worker.d.ts.map +1 -1
  156. package/dist/types/learning/index.d.ts.map +1 -1
  157. package/dist/types/learning/intake-contracts.d.ts +10 -10
  158. package/dist/types/learning/intake-contracts.d.ts.map +1 -1
  159. package/dist/types/learning/intake-csv.d.ts.map +1 -1
  160. package/dist/types/learning/intake-history.d.ts.map +1 -1
  161. package/dist/types/learning/intake-service.d.ts.map +1 -1
  162. package/dist/types/learning/intake-text.d.ts.map +1 -1
  163. package/dist/types/learning/intake.d.ts.map +1 -1
  164. package/dist/types/learning/iteration-budget.d.ts.map +1 -1
  165. package/dist/types/learning/iteration-candidate-decision.d.ts +16 -16
  166. package/dist/types/learning/iteration-candidate-decision.d.ts.map +1 -1
  167. package/dist/types/learning/iteration-dispatch-commands.d.ts.map +1 -1
  168. package/dist/types/learning/iteration-dispatch-contracts.d.ts +4 -4
  169. package/dist/types/learning/iteration-dispatch-contracts.d.ts.map +1 -1
  170. package/dist/types/learning/iteration-dispatch-worker.d.ts.map +1 -1
  171. package/dist/types/learning/iteration-eligibility.d.ts +1 -1
  172. package/dist/types/learning/iteration-eligibility.d.ts.map +1 -1
  173. package/dist/types/learning/iteration-reservation-cancellation.d.ts.map +1 -1
  174. package/dist/types/learning/iteration-reservation-contracts.d.ts.map +1 -1
  175. package/dist/types/learning/iteration-reservation-service.d.ts.map +1 -1
  176. package/dist/types/learning/iteration-training-parent.d.ts.map +1 -1
  177. package/dist/types/learning/judge-budget.d.ts +1 -1
  178. package/dist/types/learning/judge-budget.d.ts.map +1 -1
  179. package/dist/types/learning/model-judge.d.ts.map +1 -1
  180. package/dist/types/learning/nightly-schedule.d.ts.map +1 -1
  181. package/dist/types/learning/operations.d.ts +312 -298
  182. package/dist/types/learning/operations.d.ts.map +1 -1
  183. package/dist/types/learning/policy-inspection.d.ts.map +1 -1
  184. package/dist/types/learning/rating-reward-fixture.d.ts.map +1 -1
  185. package/dist/types/learning/repository.d.ts +197 -180
  186. package/dist/types/learning/repository.d.ts.map +1 -1
  187. package/dist/types/learning/reward-authoring.d.ts.map +1 -1
  188. package/dist/types/learning/reward-calibration-closure.d.ts +252 -244
  189. package/dist/types/learning/reward-calibration-closure.d.ts.map +1 -1
  190. package/dist/types/learning/reward-calibration.d.ts +62 -62
  191. package/dist/types/learning/reward-calibration.d.ts.map +1 -1
  192. package/dist/types/learning/reward-check-budget.d.ts.map +1 -1
  193. package/dist/types/learning/reward-check-service.d.ts.map +1 -1
  194. package/dist/types/learning/reward-check-worker.d.ts.map +1 -1
  195. package/dist/types/learning/reward-checks.d.ts +30 -28
  196. package/dist/types/learning/reward-checks.d.ts.map +1 -1
  197. package/dist/types/learning/reward-fixture-executor.d.ts.map +1 -1
  198. package/dist/types/learning/schedule-contracts.d.ts +3 -3
  199. package/dist/types/learning/schedule-contracts.d.ts.map +1 -1
  200. package/dist/types/learning/schedule-service.d.ts +2 -2
  201. package/dist/types/learning/schedule-service.d.ts.map +1 -1
  202. package/dist/types/learning/schedule-worker.d.ts.map +1 -1
  203. package/dist/types/learning/service.d.ts +3 -3
  204. package/dist/types/learning/service.d.ts.map +1 -1
  205. package/dist/types/learning/task-queue-inspection.d.ts.map +1 -1
  206. package/dist/types/learning/training-handoff.d.ts +9 -9
  207. package/dist/types/learning/training-handoff.d.ts.map +1 -1
  208. package/dist/types/learning/transport.d.ts +2203 -2179
  209. package/dist/types/learning/transport.d.ts.map +1 -1
  210. package/dist/types/metric-policy.d.ts +6 -6
  211. package/dist/types/metric-policy.d.ts.map +1 -1
  212. package/dist/types/metrics-node.d.ts.map +1 -1
  213. package/dist/types/metrics.d.ts.map +1 -1
  214. package/dist/types/model-improvement-qualification.d.ts +17 -17
  215. package/dist/types/model-improvement-qualification.d.ts.map +1 -1
  216. package/dist/types/native-conversations/appimage-runtime.d.ts.map +1 -1
  217. package/dist/types/native-conversations/codex-window-inventory.d.ts.map +1 -1
  218. package/dist/types/native-conversations/collector-acquisition.d.ts.map +1 -1
  219. package/dist/types/native-conversations/collector-branches.d.ts.map +1 -1
  220. package/dist/types/native-conversations/collector-calendar-gate.d.ts.map +1 -1
  221. package/dist/types/native-conversations/collector-contracts.d.ts +3 -0
  222. package/dist/types/native-conversations/collector-contracts.d.ts.map +1 -1
  223. package/dist/types/native-conversations/collector-coverage.d.ts +2 -2
  224. package/dist/types/native-conversations/collector-coverage.d.ts.map +1 -1
  225. package/dist/types/native-conversations/collector-cron.d.ts.map +1 -1
  226. package/dist/types/native-conversations/collector-destinations.d.ts.map +1 -1
  227. package/dist/types/native-conversations/collector-errors.d.ts.map +1 -1
  228. package/dist/types/native-conversations/collector-progress.d.ts.map +1 -1
  229. package/dist/types/native-conversations/collector-schedule.d.ts.map +1 -1
  230. package/dist/types/native-conversations/collector-service-files.d.ts.map +1 -1
  231. package/dist/types/native-conversations/collector-snapshots.d.ts.map +1 -1
  232. package/dist/types/native-conversations/collector-source-cache.d.ts +2 -2
  233. package/dist/types/native-conversations/collector-source-cache.d.ts.map +1 -1
  234. package/dist/types/native-conversations/collector-store.d.ts.map +1 -1
  235. package/dist/types/native-conversations/collector-sync-requests.d.ts.map +1 -1
  236. package/dist/types/native-conversations/collector.d.ts.map +1 -1
  237. package/dist/types/native-conversations/contracts.d.ts.map +1 -1
  238. package/dist/types/native-conversations/database.d.ts.map +1 -1
  239. package/dist/types/native-conversations/discovery.d.ts.map +1 -1
  240. package/dist/types/native-conversations/git-assets.d.ts +32 -0
  241. package/dist/types/native-conversations/git-assets.d.ts.map +1 -0
  242. package/dist/types/native-conversations/history.d.ts +1 -0
  243. package/dist/types/native-conversations/history.d.ts.map +1 -1
  244. package/dist/types/native-conversations/index.d.ts +2 -0
  245. package/dist/types/native-conversations/index.d.ts.map +1 -1
  246. package/dist/types/native-conversations/location.d.ts.map +1 -1
  247. package/dist/types/native-conversations/openclaw-cold-archive.d.ts.map +1 -1
  248. package/dist/types/native-conversations/openclaw-database.d.ts.map +1 -1
  249. package/dist/types/native-conversations/readiness.d.ts.map +1 -1
  250. package/dist/types/native-conversations/selection.d.ts +32 -0
  251. package/dist/types/native-conversations/selection.d.ts.map +1 -0
  252. package/dist/types/native-conversations/source-locations.d.ts.map +1 -1
  253. package/dist/types/native-conversations/supervisor.d.ts.map +1 -1
  254. package/dist/types/native-conversations/titles.d.ts.map +1 -1
  255. package/dist/types/native-harness-experiment-attempt.d.ts.map +1 -1
  256. package/dist/types/preferences.d.ts +22 -22
  257. package/dist/types/preferences.d.ts.map +1 -1
  258. package/dist/types/prepared-process-isolation.d.ts +5 -0
  259. package/dist/types/prepared-process-isolation.d.ts.map +1 -0
  260. package/dist/types/profile-evaluation-comparisons.d.ts +18 -18
  261. package/dist/types/profile-evaluation-comparisons.d.ts.map +1 -1
  262. package/dist/types/profile-evaluation-execution.d.ts.map +1 -1
  263. package/dist/types/profile-evaluation-reports.d.ts +2 -2
  264. package/dist/types/profile-evaluation-reports.d.ts.map +1 -1
  265. package/dist/types/profile-evaluation-suites.d.ts.map +1 -1
  266. package/dist/types/profile-evaluation-target.d.ts.map +1 -1
  267. package/dist/types/profile-evaluations.d.ts +15 -15
  268. package/dist/types/profile-evaluations.d.ts.map +1 -1
  269. package/dist/types/profile-external-evaluation.d.ts +84 -84
  270. package/dist/types/profile-external-evaluation.d.ts.map +1 -1
  271. package/dist/types/profile-source-candidate.d.ts.map +1 -1
  272. package/dist/types/review-conformance.d.ts +58 -58
  273. package/dist/types/review-conformance.d.ts.map +1 -1
  274. package/dist/types/rewards.d.ts +110 -101
  275. package/dist/types/rewards.d.ts.map +1 -1
  276. package/dist/types/rollouts.d.ts +16 -16
  277. package/dist/types/rollouts.d.ts.map +1 -1
  278. package/dist/types/runs.d.ts +22 -22
  279. package/dist/types/runs.d.ts.map +1 -1
  280. package/dist/types/sql-execution-contract.d.ts +3 -3
  281. package/dist/types/sql-execution-contract.d.ts.map +1 -1
  282. package/dist/types/sql-execution-engine.d.ts.map +1 -1
  283. package/dist/types/sql-execution-node.d.ts.map +1 -1
  284. package/dist/types/sql-execution-process-entry.d.ts.map +1 -1
  285. package/dist/types/task-schema-validation.d.ts.map +1 -1
  286. package/dist/types/task-schema.d.ts.map +1 -1
  287. package/dist/types/task-value-validator.d.ts.map +1 -1
  288. package/dist/types/taskset-run-contract.d.ts +26 -26
  289. package/dist/types/taskset-run-contract.d.ts.map +1 -1
  290. package/dist/types/tasksets.d.ts +89 -69
  291. package/dist/types/tasksets.d.ts.map +1 -1
  292. package/dist/types/telemetry/index.d.ts.map +1 -1
  293. package/dist/types/telemetry-analysis.d.ts +6 -6
  294. package/dist/types/telemetry-analysis.d.ts.map +1 -1
  295. package/dist/types/telemetry-bundle.d.ts +42 -42
  296. package/dist/types/telemetry-bundle.d.ts.map +1 -1
  297. package/dist/types/telemetry-catalog.d.ts +67 -67
  298. package/dist/types/telemetry-catalog.d.ts.map +1 -1
  299. package/dist/types/telemetry.d.ts +44 -44
  300. package/dist/types/telemetry.d.ts.map +1 -1
  301. package/package.json +6 -1
  302. package/schemas/learning/v1/binding.schema.json +8 -0
  303. package/schemas/learning/v1/command-request.schema.json +12 -0
  304. package/schemas/learning/v1/decision.schema.json +5 -1
  305. package/schemas/learning/v1/definition.schema.json +4 -0
  306. package/schemas/learning/v1/draft.schema.json +8 -0
  307. package/schemas/learning/v1/evidence.schema.json +4 -0
  308. package/schemas/learning/v1/example-submission.schema.json +4 -0
  309. package/schemas/learning/v1/grade.schema.json +5 -1
  310. package/schemas/learning/v1/package.schema.json +117 -93
  311. package/schemas/learning/v1/reward_check.schema.json +2 -1
@@ -8,26 +8,26 @@ export declare const TelemetryVisibilitySchema: z.ZodEnum<{
8
8
  team_visible: "team_visible";
9
9
  }>;
10
10
  export declare const TelemetrySourceSchema: z.ZodEnum<{
11
- grader: "grader";
12
- runtime: "runtime";
11
+ control_plane: "control_plane";
13
12
  environment: "environment";
14
13
  evaluation: "evaluation";
14
+ grader: "grader";
15
15
  optimizer: "optimizer";
16
- control_plane: "control_plane";
16
+ runtime: "runtime";
17
17
  }>;
18
18
  export declare const TelemetryEventTypeSchema: z.ZodEnum<{
19
- cleanup_completed: "cleanup_completed";
20
- run_started: "run_started";
21
- run_state_changed: "run_state_changed";
22
- rollout_group_started: "rollout_group_started";
23
19
  attempt_completed: "attempt_completed";
24
- grader_completed: "grader_completed";
25
- reward_composed: "reward_composed";
26
- optimizer_step_completed: "optimizer_step_completed";
27
20
  checkpoint_committed: "checkpoint_committed";
21
+ cleanup_completed: "cleanup_completed";
28
22
  evaluation_completed: "evaluation_completed";
23
+ grader_completed: "grader_completed";
24
+ optimizer_step_completed: "optimizer_step_completed";
25
+ reward_composed: "reward_composed";
26
+ rollout_group_started: "rollout_group_started";
29
27
  run_completed: "run_completed";
30
28
  run_failed: "run_failed";
29
+ run_started: "run_started";
30
+ run_state_changed: "run_state_changed";
31
31
  }>;
32
32
  export declare const RunTelemetryLineageSchema: z.ZodObject<{
33
33
  modelProjectId: z.ZodString;
@@ -48,26 +48,26 @@ export declare const RunTelemetryEventSchema: z.ZodObject<{
48
48
  sequence: z.ZodNumber;
49
49
  occurredAt: z.ZodString;
50
50
  source: z.ZodEnum<{
51
- grader: "grader";
52
- runtime: "runtime";
51
+ control_plane: "control_plane";
53
52
  environment: "environment";
54
53
  evaluation: "evaluation";
54
+ grader: "grader";
55
55
  optimizer: "optimizer";
56
- control_plane: "control_plane";
56
+ runtime: "runtime";
57
57
  }>;
58
58
  type: z.ZodEnum<{
59
- cleanup_completed: "cleanup_completed";
60
- run_started: "run_started";
61
- run_state_changed: "run_state_changed";
62
- rollout_group_started: "rollout_group_started";
63
59
  attempt_completed: "attempt_completed";
64
- grader_completed: "grader_completed";
65
- reward_composed: "reward_composed";
66
- optimizer_step_completed: "optimizer_step_completed";
67
60
  checkpoint_committed: "checkpoint_committed";
61
+ cleanup_completed: "cleanup_completed";
68
62
  evaluation_completed: "evaluation_completed";
63
+ grader_completed: "grader_completed";
64
+ optimizer_step_completed: "optimizer_step_completed";
65
+ reward_composed: "reward_composed";
66
+ rollout_group_started: "rollout_group_started";
69
67
  run_completed: "run_completed";
70
68
  run_failed: "run_failed";
69
+ run_started: "run_started";
70
+ run_state_changed: "run_state_changed";
71
71
  }>;
72
72
  visibility: z.ZodEnum<{
73
73
  host_private: "host_private";
@@ -90,19 +90,19 @@ export declare const RunTelemetryEventSchema: z.ZodObject<{
90
90
  attributes: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
91
91
  }, z.core.$strict>;
92
92
  export declare const MetricValueTypeSchema: z.ZodEnum<{
93
- gauge: "gauge";
94
93
  counter: "counter";
95
94
  distribution: "distribution";
95
+ gauge: "gauge";
96
96
  }>;
97
97
  export declare const MetricUnitSchema: z.ZodEnum<{
98
- ratio: "ratio";
98
+ bytes: "bytes";
99
99
  count: "count";
100
- seconds: "seconds";
101
100
  milliseconds: "milliseconds";
101
+ ratio: "ratio";
102
+ scalar: "scalar";
103
+ seconds: "seconds";
102
104
  tokens: "tokens";
103
- bytes: "bytes";
104
105
  usd: "usd";
105
- scalar: "scalar";
106
106
  }>;
107
107
  export declare const MetricDirectionSchema: z.ZodEnum<{
108
108
  higher: "higher";
@@ -110,13 +110,13 @@ export declare const MetricDirectionSchema: z.ZodEnum<{
110
110
  neutral: "neutral";
111
111
  }>;
112
112
  export declare const MetricAggregationSchema: z.ZodEnum<{
113
- max: "max";
114
113
  last: "last";
115
- sum: "sum";
114
+ max: "max";
116
115
  mean: "mean";
117
116
  min: "min";
118
117
  p50: "p50";
119
118
  p95: "p95";
119
+ sum: "sum";
120
120
  }>;
121
121
  export declare const MetricDefinitionSchema: z.ZodObject<{
122
122
  schemaVersion: z.ZodLiteral<"openpond.metricDefinition.v1">;
@@ -124,19 +124,19 @@ export declare const MetricDefinitionSchema: z.ZodObject<{
124
124
  displayName: z.ZodString;
125
125
  description: z.ZodString;
126
126
  valueType: z.ZodEnum<{
127
- gauge: "gauge";
128
127
  counter: "counter";
129
128
  distribution: "distribution";
129
+ gauge: "gauge";
130
130
  }>;
131
131
  unit: z.ZodEnum<{
132
- ratio: "ratio";
132
+ bytes: "bytes";
133
133
  count: "count";
134
- seconds: "seconds";
135
134
  milliseconds: "milliseconds";
135
+ ratio: "ratio";
136
+ scalar: "scalar";
137
+ seconds: "seconds";
136
138
  tokens: "tokens";
137
- bytes: "bytes";
138
139
  usd: "usd";
139
- scalar: "scalar";
140
140
  }>;
141
141
  direction: z.ZodEnum<{
142
142
  higher: "higher";
@@ -144,13 +144,13 @@ export declare const MetricDefinitionSchema: z.ZodObject<{
144
144
  neutral: "neutral";
145
145
  }>;
146
146
  aggregation: z.ZodEnum<{
147
- max: "max";
148
147
  last: "last";
149
- sum: "sum";
148
+ max: "max";
150
149
  mean: "mean";
151
150
  min: "min";
152
151
  p50: "p50";
153
152
  p95: "p95";
153
+ sum: "sum";
154
154
  }>;
155
155
  visibility: z.ZodEnum<{
156
156
  host_private: "host_private";
@@ -190,26 +190,26 @@ export declare const RunTelemetryBatchSchema: z.ZodObject<{
190
190
  sequence: z.ZodNumber;
191
191
  occurredAt: z.ZodString;
192
192
  source: z.ZodEnum<{
193
- grader: "grader";
194
- runtime: "runtime";
193
+ control_plane: "control_plane";
195
194
  environment: "environment";
196
195
  evaluation: "evaluation";
196
+ grader: "grader";
197
197
  optimizer: "optimizer";
198
- control_plane: "control_plane";
198
+ runtime: "runtime";
199
199
  }>;
200
200
  type: z.ZodEnum<{
201
- cleanup_completed: "cleanup_completed";
202
- run_started: "run_started";
203
- run_state_changed: "run_state_changed";
204
- rollout_group_started: "rollout_group_started";
205
201
  attempt_completed: "attempt_completed";
206
- grader_completed: "grader_completed";
207
- reward_composed: "reward_composed";
208
- optimizer_step_completed: "optimizer_step_completed";
209
202
  checkpoint_committed: "checkpoint_committed";
203
+ cleanup_completed: "cleanup_completed";
210
204
  evaluation_completed: "evaluation_completed";
205
+ grader_completed: "grader_completed";
206
+ optimizer_step_completed: "optimizer_step_completed";
207
+ reward_composed: "reward_composed";
208
+ rollout_group_started: "rollout_group_started";
211
209
  run_completed: "run_completed";
212
210
  run_failed: "run_failed";
211
+ run_started: "run_started";
212
+ run_state_changed: "run_state_changed";
213
213
  }>;
214
214
  visibility: z.ZodEnum<{
215
215
  host_private: "host_private";
@@ -1 +1 @@
1
- {"version":3,"file":"telemetry.d.ts","sourceRoot":"","sources":["../../../../src/telemetry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AASxB,eAAO,MAAM,4BAA4B,EAAG,+BAAwC,CAAC;AACrF,eAAO,MAAM,gCAAgC,EAAG,8BAAuC,CAAC;AACxF,eAAO,MAAM,iCAAiC,EAAG,+BAAwC,CAAC;AAE1F,eAAO,MAAM,yBAAyB;;;;EAIpC,CAAC;AAEH,eAAO,MAAM,qBAAqB;;;;;;;EAOhC,CAAC;AAEH,eAAO,MAAM,wBAAwB;;;;;;;;;;;;;EAanC,CAAC;AAEH,eAAO,MAAM,yBAAyB;;;;;;;;;;;;kBAY3B,CAAC;AAEZ,eAAO,MAAM,uBAAuB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAUzB,CAAC;AAEZ,eAAO,MAAM,qBAAqB;;;;EAA+C,CAAC;AAClF,eAAO,MAAM,gBAAgB;;;;;;;;;EAS3B,CAAC;AACH,eAAO,MAAM,qBAAqB;;;;EAAyC,CAAC;AAC5E,eAAO,MAAM,uBAAuB;;;;;;;;EAA8D,CAAC;AAEnG,eAAO,MAAM,sBAAsB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAWxB,CAAC;AAEZ,eAAO,MAAM,uBAAuB;;;;;;;;;;;;;;;;;;;;;;kBAUzB,CAAC;AAEZ,eAAO,MAAM,uBAAuB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAmBlC,CAAC;AAEH,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,MAAM,MAAM,eAAe,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,qBAAqB,CAAC,CAAC;AACpE,MAAM,MAAM,kBAAkB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,wBAAwB,CAAC,CAAC;AAC1E,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,MAAM,MAAM,iBAAiB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,uBAAuB,CAAC,CAAC;AACxE,MAAM,MAAM,gBAAgB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,sBAAsB,CAAC,CAAC;AACtE,MAAM,MAAM,iBAAiB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,uBAAuB,CAAC,CAAC;AACxE,MAAM,MAAM,iBAAiB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,uBAAuB,CAAC,CAAC"}
1
+ {"version":3,"file":"telemetry.d.ts","sourceRoot":"","sources":["../../src/telemetry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AASxB,eAAO,MAAM,4BAA4B,EAAG,+BAAwC,CAAC;AACrF,eAAO,MAAM,gCAAgC,EAAG,8BAAuC,CAAC;AACxF,eAAO,MAAM,iCAAiC,EAAG,+BAAwC,CAAC;AAE1F,eAAO,MAAM,yBAAyB;;;;EAIpC,CAAC;AAEH,eAAO,MAAM,qBAAqB;;;;;;;EAOhC,CAAC;AAEH,eAAO,MAAM,wBAAwB;;;;;;;;;;;;;EAanC,CAAC;AAEH,eAAO,MAAM,yBAAyB;;;;;;;;;;;;kBAY3B,CAAC;AAEZ,eAAO,MAAM,uBAAuB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAUzB,CAAC;AAEZ,eAAO,MAAM,qBAAqB;;;;EAA+C,CAAC;AAClF,eAAO,MAAM,gBAAgB;;;;;;;;;EAS3B,CAAC;AACH,eAAO,MAAM,qBAAqB;;;;EAAyC,CAAC;AAC5E,eAAO,MAAM,uBAAuB;;;;;;;;EAA8D,CAAC;AAEnG,eAAO,MAAM,sBAAsB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAWxB,CAAC;AAEZ,eAAO,MAAM,uBAAuB;;;;;;;;;;;;;;;;;;;;;;kBAUzB,CAAC;AAEZ,eAAO,MAAM,uBAAuB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAmBlC,CAAC;AAEH,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,MAAM,MAAM,eAAe,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,qBAAqB,CAAC,CAAC;AACpE,MAAM,MAAM,kBAAkB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,wBAAwB,CAAC,CAAC;AAC1E,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,MAAM,MAAM,iBAAiB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,uBAAuB,CAAC,CAAC;AACxE,MAAM,MAAM,gBAAgB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,sBAAsB,CAAC,CAAC;AACtE,MAAM,MAAM,iBAAiB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,uBAAuB,CAAC,CAAC;AACxE,MAAM,MAAM,iBAAiB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,uBAAuB,CAAC,CAAC"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@openpond/evals",
3
- "version": "0.11.21",
3
+ "version": "0.12.1",
4
4
  "description": "Portable Tasksets, graders, evaluation runs, receipts, conformance fixtures, and Work evidence for OpenPond Harnesses",
5
5
  "license": "MIT",
6
6
  "type": "module",
@@ -19,6 +19,10 @@
19
19
  "LICENSE"
20
20
  ],
21
21
  "exports": {
22
+ "./connected-evidence/git": {
23
+ "types": "./dist/types/connected-evidence/git.d.ts",
24
+ "import": "./dist/connected-evidence/git.js"
25
+ },
22
26
  "./native-conversations": {
23
27
  "types": "./dist/types/native-conversations/index.d.ts",
24
28
  "import": "./dist/native-conversations/index.js"
@@ -177,6 +181,7 @@
177
181
  "dependencies": {
178
182
  "@jitl/quickjs-singlefile-cjs-release-sync": "0.32.0",
179
183
  "@sqlite.org/sqlite-wasm": "3.53.0-build1",
184
+ "acorn": "8.15.0",
180
185
  "ajv": "^8.20.0",
181
186
  "cron-parser": "^5.6.2",
182
187
  "quickjs-emscripten-core": "0.32.0",
@@ -93,6 +93,14 @@
93
93
  "graderId": {
94
94
  "$ref": "#/$defs/__schema0"
95
95
  },
96
+ "taskIds": {
97
+ "minItems": 1,
98
+ "maxItems": 10000,
99
+ "type": "array",
100
+ "items": {
101
+ "$ref": "#/$defs/__schema0"
102
+ }
103
+ },
96
104
  "reward": {
97
105
  "$ref": "#/$defs/__schema1"
98
106
  },
@@ -1482,6 +1482,14 @@
1482
1482
  "graderId": {
1483
1483
  "$ref": "#/$defs/__schema0"
1484
1484
  },
1485
+ "taskIds": {
1486
+ "minItems": 1,
1487
+ "maxItems": 10000,
1488
+ "type": "array",
1489
+ "items": {
1490
+ "$ref": "#/$defs/__schema0"
1491
+ }
1492
+ },
1485
1493
  "reward": {
1486
1494
  "$ref": "#/$defs/__schema1"
1487
1495
  },
@@ -1985,6 +1993,10 @@
1985
1993
  }
1986
1994
  ]
1987
1995
  },
1996
+ "submissionMode": {
1997
+ "type": "string",
1998
+ "const": "structured_json"
1999
+ },
1988
2000
  "maxBytes": {
1989
2001
  "default": null,
1990
2002
  "anyOf": [
@@ -236,6 +236,9 @@
236
236
  "taskHash": {
237
237
  "$ref": "#/$defs/__schema2"
238
238
  },
239
+ "taskId": {
240
+ "$ref": "#/$defs/__schema0"
241
+ },
239
242
  "outputHash": {
240
243
  "$ref": "#/$defs/__schema2"
241
244
  },
@@ -368,7 +371,8 @@
368
371
  "scored",
369
372
  "pending",
370
373
  "unavailable",
371
- "failed"
374
+ "failed",
375
+ "not_applicable"
372
376
  ]
373
377
  },
374
378
  "rawScore": {
@@ -368,6 +368,10 @@
368
368
  }
369
369
  ]
370
370
  },
371
+ "submissionMode": {
372
+ "type": "string",
373
+ "const": "structured_json"
374
+ },
371
375
  "maxBytes": {
372
376
  "default": null,
373
377
  "anyOf": [
@@ -439,6 +439,14 @@
439
439
  "graderId": {
440
440
  "$ref": "#/$defs/__schema0"
441
441
  },
442
+ "taskIds": {
443
+ "minItems": 1,
444
+ "maxItems": 10000,
445
+ "type": "array",
446
+ "items": {
447
+ "$ref": "#/$defs/__schema0"
448
+ }
449
+ },
442
450
  "reward": {
443
451
  "$ref": "#/$defs/__schema2"
444
452
  },
@@ -342,6 +342,10 @@
342
342
  }
343
343
  ]
344
344
  },
345
+ "submissionMode": {
346
+ "type": "string",
347
+ "const": "structured_json"
348
+ },
345
349
  "maxBytes": {
346
350
  "default": null,
347
351
  "anyOf": [
@@ -278,6 +278,10 @@
278
278
  }
279
279
  ]
280
280
  },
281
+ "submissionMode": {
282
+ "type": "string",
283
+ "const": "structured_json"
284
+ },
281
285
  "maxBytes": {
282
286
  "default": null,
283
287
  "anyOf": [
@@ -62,6 +62,9 @@
62
62
  "taskHash": {
63
63
  "$ref": "#/$defs/__schema2"
64
64
  },
65
+ "taskId": {
66
+ "$ref": "#/$defs/__schema0"
67
+ },
65
68
  "outputHash": {
66
69
  "$ref": "#/$defs/__schema2"
67
70
  },
@@ -345,7 +348,8 @@
345
348
  "scored",
346
349
  "pending",
347
350
  "unavailable",
348
- "failed"
351
+ "failed",
352
+ "not_applicable"
349
353
  ]
350
354
  },
351
355
  "rawScore": {