@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
@@ -1 +1 @@
1
- {"version":3,"file":"benchmarks.d.ts","sourceRoot":"","sources":["../../../../src/benchmarks.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAYxB,OAAO,EAAE,KAAK,cAAc,EAAE,KAAK,gBAAgB,EAAE,MAAM,WAAW,CAAC;AAGvE,OAAO,EACL,6BAA6B,EAC7B,8BAA8B,GAC/B,MAAM,yCAAyC,CAAC;AAEjD,eAAO,MAAM,qBAAqB;;;;;EAKhC,CAAC;AACH,eAAO,MAAM,uBAAuB;;;EAAoC,CAAC;AAEzE,eAAO,MAAM,uBAAuB;;;;;;;;;;;;;;kBASzB,CAAC;AAEZ,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAe3B,CAAC;AAEZ,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;;;;;;;;kBAS3B,CAAC;AAQZ,eAAO,MAAM,gCAAgC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAkBlC,CAAC;AACZ,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAE3B,CAAC;AAEZ,eAAO,MAAM,gCAAgC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAelC,CAAC;AACZ,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAE3B,CAAC;AAEZ,wBAAgB,yBAAyB,CACvC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,GAC/C,mBAAmB,CAErB;AAED,wBAAgB,yBAAyB,CAAC,KAAK,EAAE;IAC/C,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,iBAAiB,CAAC;IACzB,UAAU,EAAE,gBAAgB,CAAC;IAC7B,QAAQ,EAAE,cAAc,EAAE,CAAC;IAC3B,eAAe,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,QAAQ,EAAE,iBAAiB,CAAC;IAC5B,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC,GAAG,mBAAmB,CAuCtB;AAED,wBAAgB,oBAAoB,CAAC,KAAK,EAAE;IAC1C,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,mBAAmB,CAAC;IAC9B,SAAS,EAAE,mBAAmB,CAAC;IAC/B,aAAa,EAAE,eAAe,CAAC;IAC/B,WAAW,EAAE,MAAM,GAAG,gBAAgB,GAAG,UAAU,CAAC;IACpD,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC,GAAG,mBAAmB,CAuDtB;AAkDD,MAAM,MAAM,eAAe,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,qBAAqB,CAAC,CAAC;AACpE,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;AACxE,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC"}
1
+ {"version":3,"file":"benchmarks.d.ts","sourceRoot":"","sources":["../../src/benchmarks.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAYxB,OAAO,EAAE,KAAK,cAAc,EAAE,KAAK,gBAAgB,EAAE,MAAM,WAAW,CAAC;AAGvE,OAAO,EACL,6BAA6B,EAC7B,8BAA8B,GAC/B,MAAM,yCAAyC,CAAC;AAEjD,eAAO,MAAM,qBAAqB;;;;;EAKhC,CAAC;AACH,eAAO,MAAM,uBAAuB;;;EAAoC,CAAC;AAEzE,eAAO,MAAM,uBAAuB;;;;;;;;;;;;;;kBASzB,CAAC;AAEZ,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAe3B,CAAC;AAEZ,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;;;;;;;;kBAS3B,CAAC;AAQZ,eAAO,MAAM,gCAAgC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAkBlC,CAAC;AACZ,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAE3B,CAAC;AAEZ,eAAO,MAAM,gCAAgC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAelC,CAAC;AACZ,eAAO,MAAM,yBAAyB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAE3B,CAAC;AAEZ,wBAAgB,yBAAyB,CACvC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,GAC/C,mBAAmB,CAErB;AAED,wBAAgB,yBAAyB,CAAC,KAAK,EAAE;IAC/C,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,iBAAiB,CAAC;IACzB,UAAU,EAAE,gBAAgB,CAAC;IAC7B,QAAQ,EAAE,cAAc,EAAE,CAAC;IAC3B,eAAe,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,QAAQ,EAAE,iBAAiB,CAAC;IAC5B,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC,GAAG,mBAAmB,CAuCtB;AAED,wBAAgB,oBAAoB,CAAC,KAAK,EAAE;IAC1C,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,mBAAmB,CAAC;IAC9B,SAAS,EAAE,mBAAmB,CAAC;IAC/B,aAAa,EAAE,eAAe,CAAC;IAC/B,WAAW,EAAE,MAAM,GAAG,gBAAgB,GAAG,UAAU,CAAC;IACpD,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC,GAAG,mBAAmB,CAuDtB;AAkDD,MAAM,MAAM,eAAe,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,qBAAqB,CAAC,CAAC;AACpE,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;AACxE,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC"}
@@ -10,14 +10,18 @@ export declare const harnessRefinerBenchmarkRelease: {
10
10
  };
11
11
  environment: {
12
12
  protocolVersion: "openpond.environment.v1";
13
- kind: "text" | "agent" | "work" | "custom_program";
13
+ kind: "agent" | "custom_program" | "text" | "work";
14
14
  entrypoint: string;
15
15
  stateful: boolean;
16
16
  deterministicSeeds: boolean;
17
- lifecycle: ("step" | "create" | "reset" | "collect" | "destroy")[];
18
- networkPolicy: "none" | "declared_read_only" | "declared_scoped";
17
+ lifecycle: ("collect" | "create" | "destroy" | "reset" | "step")[];
18
+ networkPolicy: "declared_read_only" | "declared_scoped" | "none";
19
19
  defaultTimeoutMs: number;
20
20
  };
21
+ environmentRelease?: {
22
+ id: string;
23
+ contentHash: string;
24
+ } | undefined;
21
25
  tools: {
22
26
  name: string;
23
27
  description: string;
@@ -30,12 +34,12 @@ export declare const harnessRefinerBenchmarkRelease: {
30
34
  id: string;
31
35
  required: boolean;
32
36
  scopes: string[];
33
- portability: "portable" | "host_adapter" | "local_only" | "hosted_only";
37
+ portability: "host_adapter" | "hosted_only" | "local_only" | "portable";
34
38
  }[];
35
39
  tasks: {
36
40
  id: string;
37
41
  clusterKey: string;
38
- split: "train" | "validation" | "test" | "frozen_eval";
42
+ split: "frozen_eval" | "test" | "train" | "validation";
39
43
  input: Record<string, unknown>;
40
44
  expectedOutput: Record<string, unknown> | null;
41
45
  policyVisibleContext: Record<string, unknown>;
@@ -46,9 +50,8 @@ export declare const harnessRefinerBenchmarkRelease: {
46
50
  contentHash: string;
47
51
  sizeBytes: number;
48
52
  mediaType: string;
49
- visibility: "policy" | "verifier" | "host_private";
53
+ visibility: "host_private" | "policy" | "verifier";
50
54
  }[];
51
- tags: string[];
52
55
  requiredOutputs?: {
53
56
  path: string;
54
57
  mediaType: string;
@@ -58,23 +61,27 @@ export declare const harnessRefinerBenchmarkRelease: {
58
61
  contentHash: string;
59
62
  sizeBytes: number;
60
63
  mediaType: string;
61
- visibility: "policy" | "verifier" | "host_private";
64
+ visibility: "host_private" | "policy" | "verifier";
62
65
  } | null;
66
+ submissionMode?: "structured_json" | undefined;
63
67
  maxBytes: number | null;
64
68
  metadata: Record<string, unknown>;
65
69
  }[] | undefined;
70
+ tags: string[];
66
71
  }[];
67
72
  graders: ({
68
73
  id: string;
74
+ taskIds?: string[] | undefined;
69
75
  version: string;
70
76
  weight: number;
71
77
  hardGate: boolean;
72
78
  rewardEligible: boolean;
73
79
  privileged: boolean;
74
- kind: "content" | "schema" | "artifact" | "runtime_event" | "state";
80
+ kind: "artifact" | "content" | "runtime_event" | "schema" | "state";
75
81
  config: Record<string, unknown>;
76
82
  } | {
77
83
  id: string;
84
+ taskIds?: string[] | undefined;
78
85
  version: string;
79
86
  weight: number;
80
87
  hardGate: boolean;
@@ -87,9 +94,9 @@ export declare const harnessRefinerBenchmarkRelease: {
87
94
  contentHash: string;
88
95
  sizeBytes: number;
89
96
  mediaType: string;
90
- visibility: "policy" | "verifier" | "host_private";
97
+ visibility: "host_private" | "policy" | "verifier";
91
98
  };
92
- calibrationStatus: "pending" | "passed" | "failed";
99
+ calibrationStatus: "failed" | "passed" | "pending";
93
100
  model?: {
94
101
  providerId: string;
95
102
  modelId: string;
@@ -98,26 +105,28 @@ export declare const harnessRefinerBenchmarkRelease: {
98
105
  temperature?: number | undefined;
99
106
  } | {
100
107
  id: string;
108
+ taskIds?: string[] | undefined;
101
109
  version: string;
102
110
  weight: number;
103
111
  hardGate: boolean;
104
112
  rewardEligible: boolean;
105
113
  privileged: boolean;
106
114
  kind: "custom_verifier";
115
+ runtime?: "isolated_javascript" | "sandbox_process" | undefined;
107
116
  verifierRef: {
108
117
  id: string;
109
118
  path: string;
110
119
  contentHash: string;
111
120
  sizeBytes: number;
112
121
  mediaType: string;
113
- visibility: "policy" | "verifier" | "host_private";
122
+ visibility: "host_private" | "policy" | "verifier";
114
123
  };
124
+ exportName?: string | undefined;
115
125
  timeoutMs: number;
116
126
  networkPolicy: "none";
117
- runtime?: "isolated_javascript" | "sandbox_process" | undefined;
118
- exportName?: string | undefined;
119
127
  } | {
120
128
  id: string;
129
+ taskIds?: string[] | undefined;
121
130
  version: string;
122
131
  weight: number;
123
132
  hardGate: boolean;
@@ -130,7 +139,7 @@ export declare const harnessRefinerBenchmarkRelease: {
130
139
  contentHash: string;
131
140
  sizeBytes: number;
132
141
  mediaType: string;
133
- visibility: "policy" | "verifier" | "host_private";
142
+ visibility: "host_private" | "policy" | "verifier";
134
143
  };
135
144
  reviewerRole: string;
136
145
  form?: {
@@ -143,77 +152,71 @@ export declare const harnessRefinerBenchmarkRelease: {
143
152
  instructions: string;
144
153
  required: boolean;
145
154
  allowAbstain: boolean;
146
- kind: "score";
147
- minimum: number;
148
- maximum: number;
149
- step: number;
150
155
  requireRationale?: boolean | undefined;
151
156
  anchors?: {
152
157
  value: string | number | boolean;
153
158
  description: string;
154
159
  }[] | undefined;
160
+ kind: "score";
161
+ minimum: number;
162
+ maximum: number;
163
+ step: number;
155
164
  } | {
156
165
  id: string;
157
166
  label: string;
158
167
  instructions: string;
159
168
  required: boolean;
160
169
  allowAbstain: boolean;
161
- kind: "category";
162
- options: string[];
163
170
  requireRationale?: boolean | undefined;
164
171
  anchors?: {
165
172
  value: string | number | boolean;
166
173
  description: string;
167
174
  }[] | undefined;
175
+ kind: "category";
176
+ options: string[];
168
177
  } | {
169
178
  id: string;
170
179
  label: string;
171
180
  instructions: string;
172
181
  required: boolean;
173
182
  allowAbstain: boolean;
174
- kind: "boolean";
175
183
  requireRationale?: boolean | undefined;
176
184
  anchors?: {
177
185
  value: string | number | boolean;
178
186
  description: string;
179
187
  }[] | undefined;
188
+ kind: "boolean";
180
189
  } | {
181
190
  id: string;
182
191
  label: string;
183
192
  instructions: string;
184
193
  required: boolean;
185
194
  allowAbstain: boolean;
186
- kind: "text";
187
- maximumLength: number;
188
195
  requireRationale?: boolean | undefined;
189
196
  anchors?: {
190
197
  value: string | number | boolean;
191
198
  description: string;
192
199
  }[] | undefined;
200
+ kind: "text";
193
201
  minimumLength?: number | undefined;
202
+ maximumLength: number;
194
203
  } | {
195
204
  id: string;
196
205
  label: string;
197
206
  instructions: string;
198
207
  required: boolean;
199
208
  allowAbstain: boolean;
200
- kind: "preference";
201
- allowTie: boolean;
202
209
  requireRationale?: boolean | undefined;
203
210
  anchors?: {
204
211
  value: string | number | boolean;
205
212
  description: string;
206
213
  }[] | undefined;
214
+ kind: "preference";
207
215
  requireStrength?: boolean | undefined;
216
+ allowTie: boolean;
208
217
  })[];
209
218
  } | undefined;
210
219
  })[];
211
- metadata: Record<string, unknown>;
212
- contentHash: string;
213
- environmentRelease?: {
214
- id: string;
215
- contentHash: string;
216
- } | undefined;
217
220
  verifierSetRelease?: {
218
221
  id: string;
219
222
  contentHash: string;
@@ -221,8 +224,9 @@ export declare const harnessRefinerBenchmarkRelease: {
221
224
  metrics?: {
222
225
  schemaVersion: "openpond.tasksetMetricPolicy.v1";
223
226
  primaryMetric: string;
224
- aggregation: "mean_score" | "pass_rate" | "weighted_mean" | "custom";
225
- missingReward: "zero" | "exclude";
227
+ aggregation: "custom" | "mean_score" | "pass_rate" | "weighted_mean";
228
+ missingReward: "exclude" | "zero";
229
+ taskWeights?: Record<string, number> | undefined;
226
230
  customAggregator: {
227
231
  module: string;
228
232
  exportName: string;
@@ -230,8 +234,9 @@ export declare const harnessRefinerBenchmarkRelease: {
230
234
  timeoutMs: number;
231
235
  networkPolicy: "none";
232
236
  } | null;
233
- taskWeights?: Record<string, number> | undefined;
234
237
  } | undefined;
238
+ metadata: Record<string, unknown>;
239
+ contentHash: string;
235
240
  };
236
241
  export declare const harnessRefinerBenchmarkAssets: Readonly<Record<string, string>>;
237
242
  //# sourceMappingURL=harness-refiner.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"harness-refiner.d.ts","sourceRoot":"","sources":["../../../../../src/builtin-benchmarks/harness-refiner.ts"],"names":[],"mappings":"AAGA,eAAO,MAAM,8BAA8B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAqqD1C,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CASzE,CAAC"}
1
+ {"version":3,"file":"harness-refiner.d.ts","sourceRoot":"","sources":["../../../src/builtin-benchmarks/harness-refiner.ts"],"names":[],"mappings":"AAGA,eAAO,MAAM,8BAA8B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAqqD1C,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CASzE,CAAC"}
@@ -1 +1 @@
1
- {"version":3,"file":"compatibility.d.ts","sourceRoot":"","sources":["../../../../src/compatibility.ts"],"names":[],"mappings":"AACA,OAAO,EAEL,KAAK,cAAc,EACpB,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EAEL,KAAK,2BAA2B,EACjC,MAAM,WAAW,CAAC;AACnB,OAAO,EAEL,KAAK,cAAc,EACpB,MAAM,eAAe,CAAC;AAEvB,wBAAgB,yCAAyC,CAAC,KAAK,EAAE;IAC/D,EAAE,EAAE,MAAM,CAAC;IACX,kBAAkB,EAAE,cAAc,CAAC;IACnC,uBAAuB,EAAE,cAAc,CAAC;IACxC,cAAc,EAAE,cAAc,CAAC;IAC/B,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC,GAAG,2BAA2B,CA4C9B"}
1
+ {"version":3,"file":"compatibility.d.ts","sourceRoot":"","sources":["../../src/compatibility.ts"],"names":[],"mappings":"AACA,OAAO,EAEL,KAAK,cAAc,EACpB,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EAEL,KAAK,2BAA2B,EACjC,MAAM,WAAW,CAAC;AACnB,OAAO,EAEL,KAAK,cAAc,EACpB,MAAM,eAAe,CAAC;AAEvB,wBAAgB,yCAAyC,CAAC,KAAK,EAAE;IAC/D,EAAE,EAAE,MAAM,CAAC;IACX,kBAAkB,EAAE,cAAc,CAAC;IACnC,uBAAuB,EAAE,cAAc,CAAC;IACxC,cAAc,EAAE,cAAc,CAAC;IAC/B,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC,GAAG,2BAA2B,CA4C9B"}