@tangle-network/agent-eval 0.95.0 → 0.95.1

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 (961) hide show
  1. package/CHANGELOG.md +17 -0
  2. package/dist/adapters/http.d.ts +17 -10
  3. package/dist/adapters/langchain.d.ts +14 -7
  4. package/dist/adapters/otel.d.ts +25 -13
  5. package/dist/{rl/adversarial.d.ts → adversarial-DIVcDoI_.d.ts} +7 -6
  6. package/dist/analyst/index.d.ts +236 -28
  7. package/dist/{trace-analyst/analyst.d.ts → analyst-C8HHvfJp.d.ts} +14 -12
  8. package/dist/{contract/analyze-runs.d.ts → analyze-runs-DtT6F_6T.d.ts} +10 -9
  9. package/dist/authenticity/index.d.ts +16 -15
  10. package/dist/{baseline.d.ts → baseline-Bbid3WoO.d.ts} +44 -8
  11. package/dist/belief-state/index.d.ts +605 -14
  12. package/dist/benchmarks/index.d.ts +5 -23
  13. package/dist/builder-eval/index.d.ts +250 -5
  14. package/dist/calibration-BPmzuVPk.d.ts +101 -0
  15. package/dist/campaign/index.d.ts +1452 -38
  16. package/dist/cli.d.ts +0 -2
  17. package/dist/{contract/intake/code-agent-session.d.ts → code-agent-session-CPHRCb4-.d.ts} +17 -15
  18. package/dist/contract/index.d.ts +688 -106
  19. package/dist/control-Doncu-B_.d.ts +259 -0
  20. package/dist/{control-runtime.d.ts → control-runtime-Acf9CGhw.d.ts} +26 -23
  21. package/dist/control.d.ts +10 -11
  22. package/dist/corpus-D4YW9UoJ.d.ts +560 -0
  23. package/dist/{cost-ledger.d.ts → cost-ledger-DuSqlw5B.d.ts} +11 -10
  24. package/dist/{counterfactual.d.ts → counterfactual-DlOz8PBx.d.ts} +13 -12
  25. package/dist/{dataset.d.ts → dataset-BbGkaN2I.d.ts} +14 -11
  26. package/dist/{analyst/registry.d.ts → default-registry-GyE8X5SP.d.ts} +37 -8
  27. package/dist/diagnose.d.ts +252 -1
  28. package/dist/{trace/emitter.d.ts → emitter-C2rqGH_l.d.ts} +12 -9
  29. package/dist/{errors.d.ts → errors-CzMUYo7b.d.ts} +11 -10
  30. package/dist/failure-cluster-DH9Flgcf.d.ts +76 -0
  31. package/dist/{feedback-trajectory.d.ts → feedback-trajectory-BxY0cKfs.d.ts} +38 -36
  32. package/dist/fuzz.d.ts +547 -1
  33. package/dist/gepa-C1NCIZ9o.d.ts +414 -0
  34. package/dist/governance/index.d.ts +135 -5
  35. package/dist/harness-optimizer-mOl9XX_O.d.ts +106 -0
  36. package/dist/hosted/index.d.ts +239 -10
  37. package/dist/index-_Y4oNOOb.d.ts +159 -0
  38. package/dist/index.d.ts +5660 -277
  39. package/dist/{contract/insight-report.d.ts → insight-report-BnRjTibG.d.ts} +19 -16
  40. package/dist/{trace/integrity.d.ts → integrity-D2t12mMw.d.ts} +14 -11
  41. package/dist/{judge-calibration.d.ts → judge-calibration-0p2QcWNE.d.ts} +17 -16
  42. package/dist/{analyst/kind-factory.d.ts → kind-factory-X3eDYbKn.d.ts} +61 -10
  43. package/dist/knowledge/index.d.ts +103 -3
  44. package/dist/{llm-client.d.ts → llm-client-Bj7g0rqu.d.ts} +23 -21
  45. package/dist/matrix/index.d.ts +30 -12
  46. package/dist/meta-eval/index.d.ts +182 -6
  47. package/dist/{multi-layer-verifier.d.ts → multi-layer-verifier-DUZXrPDA.d.ts} +15 -12
  48. package/dist/multishot/index.d.ts +290 -7
  49. package/dist/{rl/off-policy.d.ts → off-policy-DiwuKKg7.d.ts} +9 -8
  50. package/dist/openapi.json +1 -1
  51. package/dist/{meta-eval/outcome-store.d.ts → outcome-store-rnXLEqSn.d.ts} +8 -7
  52. package/dist/{pareto.d.ts → pareto-E-pembql.d.ts} +10 -9
  53. package/dist/perf/index.d.ts +119 -13
  54. package/dist/pipelines/index.d.ts +173 -8
  55. package/dist/pre-registration-nfUdc9EQ.d.ts +483 -0
  56. package/dist/prm/index.d.ts +104 -5
  57. package/dist/provenance-CncDq9qE.d.ts +426 -0
  58. package/dist/query-B7GGjRox.d.ts +32 -0
  59. package/dist/{trace/raw-provider-sink.d.ts → raw-provider-sink-C46HDghv.d.ts} +14 -13
  60. package/dist/{red-team.d.ts → red-team-BWdoyleI.d.ts} +16 -14
  61. package/dist/{trace/redact.d.ts → redact-B40YG2M_.d.ts} +8 -7
  62. package/dist/release-report-pidWUMZ2.d.ts +233 -0
  63. package/dist/reporting.d.ts +16 -15
  64. package/dist/{eval-campaign.d.ts → researcher-Jr8ME1dZ.d.ts} +156 -34
  65. package/dist/rl.d.ts +1193 -1
  66. package/dist/{prm/rubric.d.ts → rubric-Cc6UHvUb.d.ts} +13 -10
  67. package/dist/{meta-eval/rubric-predictive-validity.d.ts → rubric-predictive-validity-C2hDKM8Z.d.ts} +11 -8
  68. package/dist/run-critic-CmMf05uV.d.ts +56 -0
  69. package/dist/{run-record.d.ts → run-record-CP2ObebC.d.ts} +117 -15
  70. package/dist/runtime-trajectory-BOUUjI0y.d.ts +49 -0
  71. package/dist/{trace/schema.d.ts → schema-m0gsnbt3.d.ts} +30 -29
  72. package/dist/semantic-concept-judge-DSBB2Cfp.d.ts +624 -0
  73. package/dist/{sequential.d.ts → sequential-5iSVfzl2.d.ts} +10 -9
  74. package/dist/{series-convergence.d.ts → series-convergence-D5OWMBg6.d.ts} +5 -4
  75. package/dist/sink-fetch-B1Yg4Til.d.ts +101 -0
  76. package/dist/{statistics.d.ts → statistics-CCJpTGOS.d.ts} +48 -46
  77. package/dist/{trace/store.d.ts → store-BcFXE6LG.d.ts} +12 -21
  78. package/dist/{trace-analyst/types.d.ts → store-C1YxJDEK.d.ts} +74 -18
  79. package/dist/storyboard/index.d.ts +81 -16
  80. package/dist/{summary-report.d.ts → summary-report-CInXwsza.d.ts} +160 -23
  81. package/dist/telemetry/{sink-file.d.ts → file.d.ts} +7 -5
  82. package/dist/telemetry/index.d.ts +35 -17
  83. package/dist/{sandbox-harness.d.ts → test-graded-scenario-DeODGLra.d.ts} +60 -15
  84. package/dist/{locked-jsonl-appender.d.ts → testing-C21CHsq2.d.ts} +4 -3
  85. package/dist/testing.d.ts +1 -5
  86. package/dist/traces.d.ts +976 -4
  87. package/dist/{trajectory.d.ts → trajectory-2TkpSEVh.d.ts} +9 -8
  88. package/dist/{analyst/types.d.ts → types-B5x54y6n.d.ts} +112 -19
  89. package/dist/{matrix/types.d.ts → types-BUxNaJ8c.d.ts} +11 -9
  90. package/dist/{types.d.ts → types-C7DGg5ex.d.ts} +33 -31
  91. package/dist/{campaign/types.d.ts → types-DQRY8ZT-.d.ts} +47 -44
  92. package/dist/{verdict.d.ts → verdict-C9MlYujm.d.ts} +3 -2
  93. package/dist/wire/index.d.ts +570 -13
  94. package/dist/workflow/index.d.ts +496 -22
  95. package/package.json +2 -2
  96. package/dist/action-policy.d.ts +0 -24
  97. package/dist/action-policy.d.ts.map +0 -1
  98. package/dist/action-policy.test.d.ts +0 -2
  99. package/dist/action-policy.test.d.ts.map +0 -1
  100. package/dist/active-learning.d.ts +0 -41
  101. package/dist/active-learning.d.ts.map +0 -1
  102. package/dist/adapters/http.d.ts.map +0 -1
  103. package/dist/adapters/langchain.d.ts.map +0 -1
  104. package/dist/adapters/otel.d.ts.map +0 -1
  105. package/dist/agent-profile-cell.d.ts +0 -101
  106. package/dist/agent-profile-cell.d.ts.map +0 -1
  107. package/dist/agent-profile.d.ts +0 -27
  108. package/dist/agent-profile.d.ts.map +0 -1
  109. package/dist/agent-profile.test.d.ts +0 -2
  110. package/dist/agent-profile.test.d.ts.map +0 -1
  111. package/dist/analyst/adapters.d.ts +0 -62
  112. package/dist/analyst/adapters.d.ts.map +0 -1
  113. package/dist/analyst/analyst.test.d.ts +0 -2
  114. package/dist/analyst/analyst.test.d.ts.map +0 -1
  115. package/dist/analyst/ax-service.d.ts +0 -27
  116. package/dist/analyst/ax-service.d.ts.map +0 -1
  117. package/dist/analyst/behavioral-analyst.d.ts +0 -28
  118. package/dist/analyst/behavioral-analyst.d.ts.map +0 -1
  119. package/dist/analyst/chat-client.d.ts +0 -91
  120. package/dist/analyst/chat-client.d.ts.map +0 -1
  121. package/dist/analyst/default-registry.d.ts +0 -27
  122. package/dist/analyst/default-registry.d.ts.map +0 -1
  123. package/dist/analyst/default-registry.test.d.ts +0 -2
  124. package/dist/analyst/default-registry.test.d.ts.map +0 -1
  125. package/dist/analyst/finding-signature.d.ts +0 -48
  126. package/dist/analyst/finding-signature.d.ts.map +0 -1
  127. package/dist/analyst/finding-subject.d.ts +0 -146
  128. package/dist/analyst/finding-subject.d.ts.map +0 -1
  129. package/dist/analyst/finding-subject.test.d.ts +0 -2
  130. package/dist/analyst/finding-subject.test.d.ts.map +0 -1
  131. package/dist/analyst/findings-store.d.ts +0 -75
  132. package/dist/analyst/findings-store.d.ts.map +0 -1
  133. package/dist/analyst/index.d.ts.map +0 -1
  134. package/dist/analyst/kind-factory.d.ts.map +0 -1
  135. package/dist/analyst/kinds/failure-mode.d.ts +0 -19
  136. package/dist/analyst/kinds/failure-mode.d.ts.map +0 -1
  137. package/dist/analyst/kinds/improvement.d.ts +0 -23
  138. package/dist/analyst/kinds/improvement.d.ts.map +0 -1
  139. package/dist/analyst/kinds/index.d.ts +0 -22
  140. package/dist/analyst/kinds/index.d.ts.map +0 -1
  141. package/dist/analyst/kinds/kinds.test.d.ts +0 -2
  142. package/dist/analyst/kinds/kinds.test.d.ts.map +0 -1
  143. package/dist/analyst/kinds/knowledge-gap.d.ts +0 -28
  144. package/dist/analyst/kinds/knowledge-gap.d.ts.map +0 -1
  145. package/dist/analyst/kinds/knowledge-poisoning.d.ts +0 -22
  146. package/dist/analyst/kinds/knowledge-poisoning.d.ts.map +0 -1
  147. package/dist/analyst/kinds/skill-usage.d.ts +0 -84
  148. package/dist/analyst/kinds/skill-usage.d.ts.map +0 -1
  149. package/dist/analyst/kinds/skill-usage.test.d.ts +0 -2
  150. package/dist/analyst/kinds/skill-usage.test.d.ts.map +0 -1
  151. package/dist/analyst/parse-tolerant.d.ts +0 -26
  152. package/dist/analyst/parse-tolerant.d.ts.map +0 -1
  153. package/dist/analyst/parse-tolerant.test.d.ts +0 -2
  154. package/dist/analyst/parse-tolerant.test.d.ts.map +0 -1
  155. package/dist/analyst/registry.budget.test.d.ts +0 -2
  156. package/dist/analyst/registry.budget.test.d.ts.map +0 -1
  157. package/dist/analyst/registry.d.ts.map +0 -1
  158. package/dist/analyst/steer-firewall.d.ts +0 -35
  159. package/dist/analyst/steer-firewall.d.ts.map +0 -1
  160. package/dist/analyst/steer-firewall.test.d.ts +0 -2
  161. package/dist/analyst/steer-firewall.test.d.ts.map +0 -1
  162. package/dist/analyst/structure-findings.d.ts +0 -37
  163. package/dist/analyst/structure-findings.d.ts.map +0 -1
  164. package/dist/analyst/structure-findings.test.d.ts +0 -2
  165. package/dist/analyst/structure-findings.test.d.ts.map +0 -1
  166. package/dist/analyst/tool-groups.d.ts +0 -34
  167. package/dist/analyst/tool-groups.d.ts.map +0 -1
  168. package/dist/analyst/types.d.ts.map +0 -1
  169. package/dist/anti-slop.d.ts +0 -59
  170. package/dist/anti-slop.d.ts.map +0 -1
  171. package/dist/artifact-validator.d.ts +0 -74
  172. package/dist/artifact-validator.d.ts.map +0 -1
  173. package/dist/attestation.d.ts +0 -63
  174. package/dist/attestation.d.ts.map +0 -1
  175. package/dist/attestation.test.d.ts +0 -2
  176. package/dist/attestation.test.d.ts.map +0 -1
  177. package/dist/authenticity/index.d.ts.map +0 -1
  178. package/dist/authenticity/index.test.d.ts +0 -2
  179. package/dist/authenticity/index.test.d.ts.map +0 -1
  180. package/dist/auto-pr.d.ts +0 -120
  181. package/dist/auto-pr.d.ts.map +0 -1
  182. package/dist/baseline.d.ts.map +0 -1
  183. package/dist/behavior-dsl.d.ts +0 -73
  184. package/dist/behavior-dsl.d.ts.map +0 -1
  185. package/dist/belief-state/calibration.d.ts +0 -10
  186. package/dist/belief-state/calibration.d.ts.map +0 -1
  187. package/dist/belief-state/calibration.test.d.ts +0 -2
  188. package/dist/belief-state/calibration.test.d.ts.map +0 -1
  189. package/dist/belief-state/code-agent-corpus.d.ts +0 -66
  190. package/dist/belief-state/code-agent-corpus.d.ts.map +0 -1
  191. package/dist/belief-state/code-agent-corpus.test.d.ts +0 -2
  192. package/dist/belief-state/code-agent-corpus.test.d.ts.map +0 -1
  193. package/dist/belief-state/code-agent-evidence.d.ts +0 -22
  194. package/dist/belief-state/code-agent-evidence.d.ts.map +0 -1
  195. package/dist/belief-state/code-agent-evidence.test.d.ts +0 -2
  196. package/dist/belief-state/code-agent-evidence.test.d.ts.map +0 -1
  197. package/dist/belief-state/extract.d.ts +0 -7
  198. package/dist/belief-state/extract.d.ts.map +0 -1
  199. package/dist/belief-state/extract.test.d.ts +0 -2
  200. package/dist/belief-state/extract.test.d.ts.map +0 -1
  201. package/dist/belief-state/index.d.ts.map +0 -1
  202. package/dist/belief-state/ope.d.ts +0 -17
  203. package/dist/belief-state/ope.d.ts.map +0 -1
  204. package/dist/belief-state/ope.test.d.ts +0 -2
  205. package/dist/belief-state/ope.test.d.ts.map +0 -1
  206. package/dist/belief-state/phase0-measurement.d.ts +0 -55
  207. package/dist/belief-state/phase0-measurement.d.ts.map +0 -1
  208. package/dist/belief-state/report.d.ts +0 -17
  209. package/dist/belief-state/report.d.ts.map +0 -1
  210. package/dist/belief-state/report.test.d.ts +0 -2
  211. package/dist/belief-state/report.test.d.ts.map +0 -1
  212. package/dist/belief-state/research-evidence.d.ts +0 -23
  213. package/dist/belief-state/research-evidence.d.ts.map +0 -1
  214. package/dist/belief-state/research-evidence.test.d.ts +0 -2
  215. package/dist/belief-state/research-evidence.test.d.ts.map +0 -1
  216. package/dist/belief-state/runtime-benchmark-corpus.d.ts +0 -32
  217. package/dist/belief-state/runtime-benchmark-corpus.d.ts.map +0 -1
  218. package/dist/belief-state/runtime-hooks.d.ts +0 -87
  219. package/dist/belief-state/runtime-hooks.d.ts.map +0 -1
  220. package/dist/belief-state/runtime-hooks.test.d.ts +0 -2
  221. package/dist/belief-state/runtime-hooks.test.d.ts.map +0 -1
  222. package/dist/belief-state/selective.d.ts +0 -15
  223. package/dist/belief-state/selective.d.ts.map +0 -1
  224. package/dist/belief-state/selective.test.d.ts +0 -2
  225. package/dist/belief-state/selective.test.d.ts.map +0 -1
  226. package/dist/belief-state/shadow-probe.d.ts +0 -80
  227. package/dist/belief-state/shadow-probe.d.ts.map +0 -1
  228. package/dist/belief-state/shadow-probe.test.d.ts +0 -2
  229. package/dist/belief-state/shadow-probe.test.d.ts.map +0 -1
  230. package/dist/belief-state/types.d.ts +0 -195
  231. package/dist/belief-state/types.d.ts.map +0 -1
  232. package/dist/belief-state/types.test.d.ts +0 -2
  233. package/dist/belief-state/types.test.d.ts.map +0 -1
  234. package/dist/benchmark.d.ts +0 -14
  235. package/dist/benchmark.d.ts.map +0 -1
  236. package/dist/benchmarks/index.d.ts.map +0 -1
  237. package/dist/benchmarks/routing/dataset.d.ts +0 -34
  238. package/dist/benchmarks/routing/dataset.d.ts.map +0 -1
  239. package/dist/benchmarks/routing/index.d.ts +0 -34
  240. package/dist/benchmarks/routing/index.d.ts.map +0 -1
  241. package/dist/benchmarks/types.d.ts +0 -49
  242. package/dist/benchmarks/types.d.ts.map +0 -1
  243. package/dist/bisector.d.ts +0 -81
  244. package/dist/bisector.d.ts.map +0 -1
  245. package/dist/budget-guard.d.ts +0 -31
  246. package/dist/budget-guard.d.ts.map +0 -1
  247. package/dist/builder-eval/builder-session.d.ts +0 -111
  248. package/dist/builder-eval/builder-session.d.ts.map +0 -1
  249. package/dist/builder-eval/correlation.d.ts +0 -32
  250. package/dist/builder-eval/correlation.d.ts.map +0 -1
  251. package/dist/builder-eval/index.d.ts.map +0 -1
  252. package/dist/builder-eval/project-registry.d.ts +0 -51
  253. package/dist/builder-eval/project-registry.d.ts.map +0 -1
  254. package/dist/builder-eval/three-layer-eval.d.ts +0 -55
  255. package/dist/builder-eval/three-layer-eval.d.ts.map +0 -1
  256. package/dist/campaign/analyst-surface.d.ts +0 -108
  257. package/dist/campaign/analyst-surface.d.ts.map +0 -1
  258. package/dist/campaign/analyst-surface.test.d.ts +0 -2
  259. package/dist/campaign/analyst-surface.test.d.ts.map +0 -1
  260. package/dist/campaign/auto-pr.d.ts +0 -46
  261. package/dist/campaign/auto-pr.d.ts.map +0 -1
  262. package/dist/campaign/distillation/agreement-judge.d.ts +0 -69
  263. package/dist/campaign/distillation/agreement-judge.d.ts.map +0 -1
  264. package/dist/campaign/distillation/cli.d.ts +0 -35
  265. package/dist/campaign/distillation/cli.d.ts.map +0 -1
  266. package/dist/campaign/distillation/distillation.test.d.ts +0 -2
  267. package/dist/campaign/distillation/distillation.test.d.ts.map +0 -1
  268. package/dist/campaign/distillation/gold-scenarios.d.ts +0 -54
  269. package/dist/campaign/distillation/gold-scenarios.d.ts.map +0 -1
  270. package/dist/campaign/distillation/run-distillation.d.ts +0 -119
  271. package/dist/campaign/distillation/run-distillation.d.ts.map +0 -1
  272. package/dist/campaign/gates/compose.d.ts +0 -12
  273. package/dist/campaign/gates/compose.d.ts.map +0 -1
  274. package/dist/campaign/gates/default-production-gate.d.ts +0 -58
  275. package/dist/campaign/gates/default-production-gate.d.ts.map +0 -1
  276. package/dist/campaign/gates/heldout-gate.d.ts +0 -12
  277. package/dist/campaign/gates/heldout-gate.d.ts.map +0 -1
  278. package/dist/campaign/gates/promotion-policy.d.ts +0 -125
  279. package/dist/campaign/gates/promotion-policy.d.ts.map +0 -1
  280. package/dist/campaign/gates/promotion-policy.test.d.ts +0 -2
  281. package/dist/campaign/gates/promotion-policy.test.d.ts.map +0 -1
  282. package/dist/campaign/gates/sequential.d.ts +0 -146
  283. package/dist/campaign/gates/sequential.d.ts.map +0 -1
  284. package/dist/campaign/gates/sequential.test.d.ts +0 -2
  285. package/dist/campaign/gates/sequential.test.d.ts.map +0 -1
  286. package/dist/campaign/gates/statistical-heldout.d.ts +0 -99
  287. package/dist/campaign/gates/statistical-heldout.d.ts.map +0 -1
  288. package/dist/campaign/gates/statistical-heldout.test.d.ts +0 -2
  289. package/dist/campaign/gates/statistical-heldout.test.d.ts.map +0 -1
  290. package/dist/campaign/index.d.ts.map +0 -1
  291. package/dist/campaign/labeled-store/fs-adapter.d.ts +0 -59
  292. package/dist/campaign/labeled-store/fs-adapter.d.ts.map +0 -1
  293. package/dist/campaign/presets/compare-proposers.d.ts +0 -146
  294. package/dist/campaign/presets/compare-proposers.d.ts.map +0 -1
  295. package/dist/campaign/presets/playback.d.ts +0 -120
  296. package/dist/campaign/presets/playback.d.ts.map +0 -1
  297. package/dist/campaign/presets/playback.test.d.ts +0 -2
  298. package/dist/campaign/presets/playback.test.d.ts.map +0 -1
  299. package/dist/campaign/presets/run-eval.d.ts +0 -14
  300. package/dist/campaign/presets/run-eval.d.ts.map +0 -1
  301. package/dist/campaign/presets/run-improvement-loop.d.ts +0 -63
  302. package/dist/campaign/presets/run-improvement-loop.d.ts.map +0 -1
  303. package/dist/campaign/presets/run-improvement-loop.test.d.ts +0 -2
  304. package/dist/campaign/presets/run-improvement-loop.test.d.ts.map +0 -1
  305. package/dist/campaign/presets/run-optimization.d.ts +0 -92
  306. package/dist/campaign/presets/run-optimization.d.ts.map +0 -1
  307. package/dist/campaign/presets/run-profile-matrix.d.ts +0 -151
  308. package/dist/campaign/presets/run-profile-matrix.d.ts.map +0 -1
  309. package/dist/campaign/presets/run-skill-opt.d.ts +0 -96
  310. package/dist/campaign/presets/run-skill-opt.d.ts.map +0 -1
  311. package/dist/campaign/proposers/_findings-text.d.ts +0 -22
  312. package/dist/campaign/proposers/_findings-text.d.ts.map +0 -1
  313. package/dist/campaign/proposers/ace.d.ts +0 -33
  314. package/dist/campaign/proposers/ace.d.ts.map +0 -1
  315. package/dist/campaign/proposers/ace.test.d.ts +0 -2
  316. package/dist/campaign/proposers/ace.test.d.ts.map +0 -1
  317. package/dist/campaign/proposers/analysis-edit.d.ts +0 -32
  318. package/dist/campaign/proposers/analysis-edit.d.ts.map +0 -1
  319. package/dist/campaign/proposers/evolutionary.d.ts +0 -20
  320. package/dist/campaign/proposers/evolutionary.d.ts.map +0 -1
  321. package/dist/campaign/proposers/fapo.d.ts +0 -120
  322. package/dist/campaign/proposers/fapo.d.ts.map +0 -1
  323. package/dist/campaign/proposers/gepa.d.ts +0 -86
  324. package/dist/campaign/proposers/gepa.d.ts.map +0 -1
  325. package/dist/campaign/proposers/halo.d.ts +0 -44
  326. package/dist/campaign/proposers/halo.d.ts.map +0 -1
  327. package/dist/campaign/proposers/halo.test.d.ts +0 -2
  328. package/dist/campaign/proposers/halo.test.d.ts.map +0 -1
  329. package/dist/campaign/proposers/memory.d.ts +0 -47
  330. package/dist/campaign/proposers/memory.d.ts.map +0 -1
  331. package/dist/campaign/proposers/memory.test.d.ts +0 -2
  332. package/dist/campaign/proposers/memory.test.d.ts.map +0 -1
  333. package/dist/campaign/proposers/skill-opt.d.ts +0 -88
  334. package/dist/campaign/proposers/skill-opt.d.ts.map +0 -1
  335. package/dist/campaign/proposers/trace-analyst.d.ts +0 -48
  336. package/dist/campaign/proposers/trace-analyst.d.ts.map +0 -1
  337. package/dist/campaign/proposers/trace-analyst.test.d.ts +0 -2
  338. package/dist/campaign/proposers/trace-analyst.test.d.ts.map +0 -1
  339. package/dist/campaign/provenance.d.ts +0 -185
  340. package/dist/campaign/provenance.d.ts.map +0 -1
  341. package/dist/campaign/run-campaign.d.ts +0 -90
  342. package/dist/campaign/run-campaign.d.ts.map +0 -1
  343. package/dist/campaign/score-utils.d.ts +0 -26
  344. package/dist/campaign/score-utils.d.ts.map +0 -1
  345. package/dist/campaign/skill-patch.d.ts +0 -62
  346. package/dist/campaign/skill-patch.d.ts.map +0 -1
  347. package/dist/campaign/storage.d.ts +0 -38
  348. package/dist/campaign/storage.d.ts.map +0 -1
  349. package/dist/campaign/types.d.ts.map +0 -1
  350. package/dist/campaign/worktree/index.d.ts +0 -53
  351. package/dist/campaign/worktree/index.d.ts.map +0 -1
  352. package/dist/canary.d.ts +0 -101
  353. package/dist/canary.d.ts.map +0 -1
  354. package/dist/causal-attribution.d.ts +0 -45
  355. package/dist/causal-attribution.d.ts.map +0 -1
  356. package/dist/ci-gate.d.ts +0 -44
  357. package/dist/ci-gate.d.ts.map +0 -1
  358. package/dist/cli.d.ts.map +0 -1
  359. package/dist/client.d.ts +0 -77
  360. package/dist/client.d.ts.map +0 -1
  361. package/dist/client.test.d.ts +0 -2
  362. package/dist/client.test.d.ts.map +0 -1
  363. package/dist/command-runner.d.ts +0 -74
  364. package/dist/command-runner.d.ts.map +0 -1
  365. package/dist/command-runner.test.d.ts +0 -2
  366. package/dist/command-runner.test.d.ts.map +0 -1
  367. package/dist/completion-verifier.d.ts +0 -147
  368. package/dist/completion-verifier.d.ts.map +0 -1
  369. package/dist/completion-verifier.test.d.ts +0 -9
  370. package/dist/completion-verifier.test.d.ts.map +0 -1
  371. package/dist/concurrency.d.ts +0 -23
  372. package/dist/concurrency.d.ts.map +0 -1
  373. package/dist/contamination-guard.d.ts +0 -81
  374. package/dist/contamination-guard.d.ts.map +0 -1
  375. package/dist/contract/analyze-runs.d.ts.map +0 -1
  376. package/dist/contract/define-agent-eval.d.ts +0 -52
  377. package/dist/contract/define-agent-eval.d.ts.map +0 -1
  378. package/dist/contract/diff.d.ts +0 -114
  379. package/dist/contract/diff.d.ts.map +0 -1
  380. package/dist/contract/index.d.ts.map +0 -1
  381. package/dist/contract/insight-report.d.ts.map +0 -1
  382. package/dist/contract/insight-types-fwd.d.ts +0 -7
  383. package/dist/contract/insight-types-fwd.d.ts.map +0 -1
  384. package/dist/contract/intake/agent-trace.d.ts +0 -97
  385. package/dist/contract/intake/agent-trace.d.ts.map +0 -1
  386. package/dist/contract/intake/code-agent-session.d.ts.map +0 -1
  387. package/dist/contract/intake/feedback-table.d.ts +0 -87
  388. package/dist/contract/intake/feedback-table.d.ts.map +0 -1
  389. package/dist/contract/intake/index.d.ts +0 -22
  390. package/dist/contract/intake/index.d.ts.map +0 -1
  391. package/dist/contract/intake/otel-spans.d.ts +0 -33
  392. package/dist/contract/intake/otel-spans.d.ts.map +0 -1
  393. package/dist/contract/self-improve.d.ts +0 -284
  394. package/dist/contract/self-improve.d.ts.map +0 -1
  395. package/dist/control-runtime.d.ts.map +0 -1
  396. package/dist/control-runtime.test.d.ts +0 -2
  397. package/dist/control-runtime.test.d.ts.map +0 -1
  398. package/dist/control.d.ts.map +0 -1
  399. package/dist/convergence.d.ts +0 -29
  400. package/dist/convergence.d.ts.map +0 -1
  401. package/dist/cost-ledger.d.ts.map +0 -1
  402. package/dist/cost-ledger.test.d.ts +0 -2
  403. package/dist/cost-ledger.test.d.ts.map +0 -1
  404. package/dist/cost-report.d.ts +0 -43
  405. package/dist/cost-report.d.ts.map +0 -1
  406. package/dist/cost-report.test.d.ts +0 -2
  407. package/dist/cost-report.test.d.ts.map +0 -1
  408. package/dist/cost-tracker.d.ts +0 -76
  409. package/dist/cost-tracker.d.ts.map +0 -1
  410. package/dist/counterfactual.d.ts.map +0 -1
  411. package/dist/cross-trace-diff.d.ts +0 -56
  412. package/dist/cross-trace-diff.d.ts.map +0 -1
  413. package/dist/dataset.d.ts.map +0 -1
  414. package/dist/deploy-gate-layer.d.ts +0 -125
  415. package/dist/deploy-gate-layer.d.ts.map +0 -1
  416. package/dist/deploy-gate-layer.test.d.ts +0 -2
  417. package/dist/deploy-gate-layer.test.d.ts.map +0 -1
  418. package/dist/description-length-gate.d.ts +0 -119
  419. package/dist/description-length-gate.d.ts.map +0 -1
  420. package/dist/detectors/edge.test.d.ts +0 -2
  421. package/dist/detectors/edge.test.d.ts.map +0 -1
  422. package/dist/detectors/index.d.ts +0 -81
  423. package/dist/detectors/index.d.ts.map +0 -1
  424. package/dist/detectors/index.test.d.ts +0 -2
  425. package/dist/detectors/index.test.d.ts.map +0 -1
  426. package/dist/diagnose/causal-sweep.d.ts +0 -100
  427. package/dist/diagnose/causal-sweep.d.ts.map +0 -1
  428. package/dist/diagnose/index.d.ts +0 -36
  429. package/dist/diagnose/index.d.ts.map +0 -1
  430. package/dist/diagnose/remediation.d.ts +0 -68
  431. package/dist/diagnose/remediation.d.ts.map +0 -1
  432. package/dist/diagnose/repair.d.ts +0 -77
  433. package/dist/diagnose/repair.d.ts.map +0 -1
  434. package/dist/discover-personas.d.ts +0 -35
  435. package/dist/discover-personas.d.ts.map +0 -1
  436. package/dist/driver.d.ts +0 -95
  437. package/dist/driver.d.ts.map +0 -1
  438. package/dist/driver.test.d.ts +0 -8
  439. package/dist/driver.test.d.ts.map +0 -1
  440. package/dist/dual-agent-bench.d.ts +0 -81
  441. package/dist/dual-agent-bench.d.ts.map +0 -1
  442. package/dist/error-count-extractor.d.ts +0 -47
  443. package/dist/error-count-extractor.d.ts.map +0 -1
  444. package/dist/error-count-extractor.test.d.ts +0 -2
  445. package/dist/error-count-extractor.test.d.ts.map +0 -1
  446. package/dist/errors.d.ts.map +0 -1
  447. package/dist/eval-campaign.d.ts.map +0 -1
  448. package/dist/eval-campaign.test.d.ts +0 -2
  449. package/dist/eval-campaign.test.d.ts.map +0 -1
  450. package/dist/eval-tools.d.ts +0 -55
  451. package/dist/eval-tools.d.ts.map +0 -1
  452. package/dist/eval-trace-store.d.ts +0 -107
  453. package/dist/eval-trace-store.d.ts.map +0 -1
  454. package/dist/eval-trace-store.test.d.ts +0 -2
  455. package/dist/eval-trace-store.test.d.ts.map +0 -1
  456. package/dist/executor.d.ts +0 -38
  457. package/dist/executor.d.ts.map +0 -1
  458. package/dist/executor.test.d.ts +0 -10
  459. package/dist/executor.test.d.ts.map +0 -1
  460. package/dist/experiment-tracker.d.ts +0 -178
  461. package/dist/experiment-tracker.d.ts.map +0 -1
  462. package/dist/experiment-tracker.test.d.ts +0 -2
  463. package/dist/experiment-tracker.test.d.ts.map +0 -1
  464. package/dist/failure-taxonomy.d.ts +0 -38
  465. package/dist/failure-taxonomy.d.ts.map +0 -1
  466. package/dist/feedback-trajectory.d.ts.map +0 -1
  467. package/dist/feedback-trajectory.test.d.ts +0 -2
  468. package/dist/feedback-trajectory.test.d.ts.map +0 -1
  469. package/dist/flow-layer.d.ts +0 -90
  470. package/dist/flow-layer.d.ts.map +0 -1
  471. package/dist/flow-layer.test.d.ts +0 -2
  472. package/dist/flow-layer.test.d.ts.map +0 -1
  473. package/dist/fuzz/capsule.d.ts +0 -46
  474. package/dist/fuzz/capsule.d.ts.map +0 -1
  475. package/dist/fuzz/cube.d.ts +0 -36
  476. package/dist/fuzz/cube.d.ts.map +0 -1
  477. package/dist/fuzz/explorer-cost.test.d.ts +0 -2
  478. package/dist/fuzz/explorer-cost.test.d.ts.map +0 -1
  479. package/dist/fuzz/explorer.d.ts +0 -64
  480. package/dist/fuzz/explorer.d.ts.map +0 -1
  481. package/dist/fuzz/fuzz-agent.d.ts +0 -16
  482. package/dist/fuzz/fuzz-agent.d.ts.map +0 -1
  483. package/dist/fuzz/fuzz-agent.test.d.ts +0 -2
  484. package/dist/fuzz/fuzz-agent.test.d.ts.map +0 -1
  485. package/dist/fuzz/gates.d.ts +0 -33
  486. package/dist/fuzz/gates.d.ts.map +0 -1
  487. package/dist/fuzz/index.d.ts +0 -26
  488. package/dist/fuzz/index.d.ts.map +0 -1
  489. package/dist/fuzz/policies.d.ts +0 -28
  490. package/dist/fuzz/policies.d.ts.map +0 -1
  491. package/dist/fuzz/tools.d.ts +0 -20
  492. package/dist/fuzz/tools.d.ts.map +0 -1
  493. package/dist/fuzz/types.d.ts +0 -307
  494. package/dist/fuzz/types.d.ts.map +0 -1
  495. package/dist/golden-matcher.d.ts +0 -71
  496. package/dist/golden-matcher.d.ts.map +0 -1
  497. package/dist/governance/eu-ai-act.d.ts +0 -37
  498. package/dist/governance/eu-ai-act.d.ts.map +0 -1
  499. package/dist/governance/index.d.ts.map +0 -1
  500. package/dist/governance/nist-ai-rmf.d.ts +0 -15
  501. package/dist/governance/nist-ai-rmf.d.ts.map +0 -1
  502. package/dist/governance/soc2.d.ts +0 -12
  503. package/dist/governance/soc2.d.ts.map +0 -1
  504. package/dist/governance/types.d.ts +0 -66
  505. package/dist/governance/types.d.ts.map +0 -1
  506. package/dist/harness-optimizer.d.ts +0 -82
  507. package/dist/harness-optimizer.d.ts.map +0 -1
  508. package/dist/held-out-gate.d.ts +0 -135
  509. package/dist/held-out-gate.d.ts.map +0 -1
  510. package/dist/hosted/client.d.ts +0 -73
  511. package/dist/hosted/client.d.ts.map +0 -1
  512. package/dist/hosted/from-env.test.d.ts +0 -8
  513. package/dist/hosted/from-env.test.d.ts.map +0 -1
  514. package/dist/hosted/index.d.ts.map +0 -1
  515. package/dist/hosted/types.d.ts +0 -159
  516. package/dist/hosted/types.d.ts.map +0 -1
  517. package/dist/index.d.ts.map +0 -1
  518. package/dist/integrity/backend-integrity.d.ts +0 -71
  519. package/dist/integrity/backend-integrity.d.ts.map +0 -1
  520. package/dist/integrity/preflight.d.ts +0 -72
  521. package/dist/integrity/preflight.d.ts.map +0 -1
  522. package/dist/integrity/preflight.test.d.ts +0 -2
  523. package/dist/integrity/preflight.test.d.ts.map +0 -1
  524. package/dist/integrity/single-backend.d.ts +0 -67
  525. package/dist/integrity/single-backend.d.ts.map +0 -1
  526. package/dist/intent-match-judge.d.ts +0 -69
  527. package/dist/intent-match-judge.d.ts.map +0 -1
  528. package/dist/intent-match-judge.test.d.ts +0 -2
  529. package/dist/intent-match-judge.test.d.ts.map +0 -1
  530. package/dist/judge-calibration.d.ts.map +0 -1
  531. package/dist/judge-ensemble.d.ts +0 -66
  532. package/dist/judge-ensemble.d.ts.map +0 -1
  533. package/dist/judge-ensemble.test.d.ts +0 -8
  534. package/dist/judge-ensemble.test.d.ts.map +0 -1
  535. package/dist/judge-families.d.ts +0 -38
  536. package/dist/judge-families.d.ts.map +0 -1
  537. package/dist/judge-panel.d.ts +0 -65
  538. package/dist/judge-panel.d.ts.map +0 -1
  539. package/dist/judge-retry.d.ts +0 -70
  540. package/dist/judge-retry.d.ts.map +0 -1
  541. package/dist/judge-runner.d.ts +0 -36
  542. package/dist/judge-runner.d.ts.map +0 -1
  543. package/dist/judge-runner.test.d.ts +0 -2
  544. package/dist/judge-runner.test.d.ts.map +0 -1
  545. package/dist/judges.d.ts +0 -74
  546. package/dist/judges.d.ts.map +0 -1
  547. package/dist/keyword-coverage-judge.d.ts +0 -89
  548. package/dist/keyword-coverage-judge.d.ts.map +0 -1
  549. package/dist/keyword-coverage-judge.test.d.ts +0 -2
  550. package/dist/keyword-coverage-judge.test.d.ts.map +0 -1
  551. package/dist/knowledge/index.d.ts.map +0 -1
  552. package/dist/knowledge/readiness.d.ts +0 -26
  553. package/dist/knowledge/readiness.d.ts.map +0 -1
  554. package/dist/knowledge/types.d.ts +0 -75
  555. package/dist/knowledge/types.d.ts.map +0 -1
  556. package/dist/live-proof.d.ts +0 -62
  557. package/dist/live-proof.d.ts.map +0 -1
  558. package/dist/llm-client.d.ts.map +0 -1
  559. package/dist/llm-client.test.d.ts +0 -2
  560. package/dist/llm-client.test.d.ts.map +0 -1
  561. package/dist/locked-jsonl-appender.d.ts.map +0 -1
  562. package/dist/matrix/aggregation.d.ts +0 -16
  563. package/dist/matrix/aggregation.d.ts.map +0 -1
  564. package/dist/matrix/index.d.ts.map +0 -1
  565. package/dist/matrix/runner.d.ts +0 -15
  566. package/dist/matrix/runner.d.ts.map +0 -1
  567. package/dist/matrix/types.d.ts.map +0 -1
  568. package/dist/meta-eval/calibration.d.ts +0 -47
  569. package/dist/meta-eval/calibration.d.ts.map +0 -1
  570. package/dist/meta-eval/correlation-study.d.ts +0 -53
  571. package/dist/meta-eval/correlation-study.d.ts.map +0 -1
  572. package/dist/meta-eval/index.d.ts.map +0 -1
  573. package/dist/meta-eval/outcome-store.d.ts.map +0 -1
  574. package/dist/meta-eval/rubric-predictive-validity.d.ts.map +0 -1
  575. package/dist/meta-eval/sentinel.d.ts +0 -169
  576. package/dist/meta-eval/sentinel.d.ts.map +0 -1
  577. package/dist/metrics.d.ts +0 -63
  578. package/dist/metrics.d.ts.map +0 -1
  579. package/dist/model-seats.d.ts +0 -71
  580. package/dist/model-seats.d.ts.map +0 -1
  581. package/dist/model-seats.test.d.ts +0 -2
  582. package/dist/model-seats.test.d.ts.map +0 -1
  583. package/dist/muffled-gate-scanner.d.ts +0 -102
  584. package/dist/muffled-gate-scanner.d.ts.map +0 -1
  585. package/dist/multi-layer-verifier.d.ts.map +0 -1
  586. package/dist/multi-layer-verifier.test.d.ts +0 -2
  587. package/dist/multi-layer-verifier.test.d.ts.map +0 -1
  588. package/dist/multi-toolchain-layer.d.ts +0 -80
  589. package/dist/multi-toolchain-layer.d.ts.map +0 -1
  590. package/dist/multi-toolchain-layer.test.d.ts +0 -2
  591. package/dist/multi-toolchain-layer.test.d.ts.map +0 -1
  592. package/dist/multishot/default-tools.d.ts +0 -34
  593. package/dist/multishot/default-tools.d.ts.map +0 -1
  594. package/dist/multishot/index.d.ts.map +0 -1
  595. package/dist/multishot/judges.d.ts +0 -32
  596. package/dist/multishot/judges.d.ts.map +0 -1
  597. package/dist/multishot/matrix.d.ts +0 -107
  598. package/dist/multishot/matrix.d.ts.map +0 -1
  599. package/dist/multishot/multishot.d.ts +0 -23
  600. package/dist/multishot/multishot.d.ts.map +0 -1
  601. package/dist/multishot/router.d.ts +0 -37
  602. package/dist/multishot/router.d.ts.map +0 -1
  603. package/dist/multishot/types.d.ts +0 -60
  604. package/dist/multishot/types.d.ts.map +0 -1
  605. package/dist/observability.d.ts +0 -71
  606. package/dist/observability.d.ts.map +0 -1
  607. package/dist/oracle.d.ts +0 -55
  608. package/dist/oracle.d.ts.map +0 -1
  609. package/dist/orthogonality.d.ts +0 -35
  610. package/dist/orthogonality.d.ts.map +0 -1
  611. package/dist/otel-pipeline.d.ts +0 -31
  612. package/dist/otel-pipeline.d.ts.map +0 -1
  613. package/dist/paraphrase.d.ts +0 -107
  614. package/dist/paraphrase.d.ts.map +0 -1
  615. package/dist/pareto.d.ts.map +0 -1
  616. package/dist/partition-held-out.d.ts +0 -70
  617. package/dist/partition-held-out.d.ts.map +0 -1
  618. package/dist/partition-held-out.test.d.ts +0 -2
  619. package/dist/partition-held-out.test.d.ts.map +0 -1
  620. package/dist/perf/index.d.ts.map +0 -1
  621. package/dist/perf/integrity.d.ts +0 -30
  622. package/dist/perf/integrity.d.ts.map +0 -1
  623. package/dist/perf/journey.d.ts +0 -45
  624. package/dist/perf/journey.d.ts.map +0 -1
  625. package/dist/perf/ratchet.d.ts +0 -47
  626. package/dist/perf/ratchet.d.ts.map +0 -1
  627. package/dist/pipelines/budget-breach.d.ts +0 -31
  628. package/dist/pipelines/budget-breach.d.ts.map +0 -1
  629. package/dist/pipelines/budget-breach.test.d.ts +0 -2
  630. package/dist/pipelines/budget-breach.test.d.ts.map +0 -1
  631. package/dist/pipelines/failure-cluster.d.ts +0 -38
  632. package/dist/pipelines/failure-cluster.d.ts.map +0 -1
  633. package/dist/pipelines/failure-cluster.test.d.ts +0 -2
  634. package/dist/pipelines/failure-cluster.test.d.ts.map +0 -1
  635. package/dist/pipelines/first-divergence.d.ts +0 -26
  636. package/dist/pipelines/first-divergence.d.ts.map +0 -1
  637. package/dist/pipelines/first-divergence.test.d.ts +0 -2
  638. package/dist/pipelines/first-divergence.test.d.ts.map +0 -1
  639. package/dist/pipelines/index.d.ts.map +0 -1
  640. package/dist/pipelines/judge-agreement.d.ts +0 -26
  641. package/dist/pipelines/judge-agreement.d.ts.map +0 -1
  642. package/dist/pipelines/judge-agreement.test.d.ts +0 -2
  643. package/dist/pipelines/judge-agreement.test.d.ts.map +0 -1
  644. package/dist/pipelines/regression.d.ts +0 -23
  645. package/dist/pipelines/regression.d.ts.map +0 -1
  646. package/dist/pipelines/regression.test.d.ts +0 -2
  647. package/dist/pipelines/regression.test.d.ts.map +0 -1
  648. package/dist/pipelines/stuck-loop.d.ts +0 -32
  649. package/dist/pipelines/stuck-loop.d.ts.map +0 -1
  650. package/dist/pipelines/stuck-loop.test.d.ts +0 -2
  651. package/dist/pipelines/stuck-loop.test.d.ts.map +0 -1
  652. package/dist/pipelines/tool-waste.d.ts +0 -34
  653. package/dist/pipelines/tool-waste.d.ts.map +0 -1
  654. package/dist/pipelines/tool-waste.test.d.ts +0 -2
  655. package/dist/pipelines/tool-waste.test.d.ts.map +0 -1
  656. package/dist/playbook.d.ts +0 -16
  657. package/dist/playbook.d.ts.map +0 -1
  658. package/dist/pr-review-benchmark.d.ts +0 -88
  659. package/dist/pr-review-benchmark.d.ts.map +0 -1
  660. package/dist/pr-review-benchmark.test.d.ts +0 -2
  661. package/dist/pr-review-benchmark.test.d.ts.map +0 -1
  662. package/dist/pre-registration.d.ts +0 -125
  663. package/dist/pre-registration.d.ts.map +0 -1
  664. package/dist/prm/builtin-rubrics.d.ts +0 -33
  665. package/dist/prm/builtin-rubrics.d.ts.map +0 -1
  666. package/dist/prm/index.d.ts.map +0 -1
  667. package/dist/prm/inference.d.ts +0 -29
  668. package/dist/prm/inference.d.ts.map +0 -1
  669. package/dist/prm/inference.test.d.ts +0 -2
  670. package/dist/prm/inference.test.d.ts.map +0 -1
  671. package/dist/prm/rubric.d.ts.map +0 -1
  672. package/dist/prm/training-export.d.ts +0 -38
  673. package/dist/prm/training-export.d.ts.map +0 -1
  674. package/dist/produced-state.d.ts +0 -63
  675. package/dist/produced-state.d.ts.map +0 -1
  676. package/dist/produced-state.test.d.ts +0 -8
  677. package/dist/produced-state.test.d.ts.map +0 -1
  678. package/dist/profile/baselines.d.ts +0 -37
  679. package/dist/profile/baselines.d.ts.map +0 -1
  680. package/dist/profile/index.d.ts +0 -105
  681. package/dist/profile/index.d.ts.map +0 -1
  682. package/dist/promotion-gate.d.ts +0 -93
  683. package/dist/promotion-gate.d.ts.map +0 -1
  684. package/dist/prompt-registry.d.ts +0 -41
  685. package/dist/prompt-registry.d.ts.map +0 -1
  686. package/dist/propose-review-control.d.ts +0 -49
  687. package/dist/propose-review-control.d.ts.map +0 -1
  688. package/dist/propose-review-control.test.d.ts +0 -2
  689. package/dist/propose-review-control.test.d.ts.map +0 -1
  690. package/dist/propose-review.d.ts +0 -155
  691. package/dist/propose-review.d.ts.map +0 -1
  692. package/dist/red-team.d.ts.map +0 -1
  693. package/dist/reference-replay-steering.d.ts +0 -11
  694. package/dist/reference-replay-steering.d.ts.map +0 -1
  695. package/dist/reference-replay.d.ts +0 -176
  696. package/dist/reference-replay.d.ts.map +0 -1
  697. package/dist/reflective-mutation.d.ts +0 -79
  698. package/dist/reflective-mutation.d.ts.map +0 -1
  699. package/dist/registry.d.ts +0 -31
  700. package/dist/registry.d.ts.map +0 -1
  701. package/dist/release-confidence.d.ts +0 -128
  702. package/dist/release-confidence.d.ts.map +0 -1
  703. package/dist/release-report.d.ts +0 -11
  704. package/dist/release-report.d.ts.map +0 -1
  705. package/dist/replay.d.ts +0 -120
  706. package/dist/replay.d.ts.map +0 -1
  707. package/dist/reporter.d.ts +0 -14
  708. package/dist/reporter.d.ts.map +0 -1
  709. package/dist/reporting.d.ts.map +0 -1
  710. package/dist/researcher.d.ts +0 -140
  711. package/dist/researcher.d.ts.map +0 -1
  712. package/dist/reviewer.d.ts +0 -118
  713. package/dist/reviewer.d.ts.map +0 -1
  714. package/dist/reviewer.test.d.ts +0 -2
  715. package/dist/reviewer.test.d.ts.map +0 -1
  716. package/dist/reward-model-export.d.ts +0 -60
  717. package/dist/reward-model-export.d.ts.map +0 -1
  718. package/dist/rl/active-curriculum.d.ts +0 -110
  719. package/dist/rl/active-curriculum.d.ts.map +0 -1
  720. package/dist/rl/adaptation-eval.d.ts +0 -109
  721. package/dist/rl/adaptation-eval.d.ts.map +0 -1
  722. package/dist/rl/adversarial.d.ts.map +0 -1
  723. package/dist/rl/compute-curves.d.ts +0 -127
  724. package/dist/rl/compute-curves.d.ts.map +0 -1
  725. package/dist/rl/contamination.d.ts +0 -117
  726. package/dist/rl/contamination.d.ts.map +0 -1
  727. package/dist/rl/corpus.d.ts +0 -55
  728. package/dist/rl/corpus.d.ts.map +0 -1
  729. package/dist/rl/corpus.test.d.ts +0 -2
  730. package/dist/rl/corpus.test.d.ts.map +0 -1
  731. package/dist/rl/dataset.d.ts +0 -102
  732. package/dist/rl/dataset.d.ts.map +0 -1
  733. package/dist/rl/dataset.test.d.ts +0 -2
  734. package/dist/rl/dataset.test.d.ts.map +0 -1
  735. package/dist/rl/exporters.d.ts +0 -141
  736. package/dist/rl/exporters.d.ts.map +0 -1
  737. package/dist/rl/index.d.ts +0 -49
  738. package/dist/rl/index.d.ts.map +0 -1
  739. package/dist/rl/off-policy.d.ts.map +0 -1
  740. package/dist/rl/predictive-validity-researcher.d.ts +0 -69
  741. package/dist/rl/predictive-validity-researcher.d.ts.map +0 -1
  742. package/dist/rl/preferences.d.ts +0 -141
  743. package/dist/rl/preferences.d.ts.map +0 -1
  744. package/dist/rl/process-reward.d.ts +0 -122
  745. package/dist/rl/process-reward.d.ts.map +0 -1
  746. package/dist/rl/reward-hacking.d.ts +0 -104
  747. package/dist/rl/reward-hacking.d.ts.map +0 -1
  748. package/dist/rl/rl-campaign.d.ts +0 -85
  749. package/dist/rl/rl-campaign.d.ts.map +0 -1
  750. package/dist/rl/run-record-adapters.d.ts +0 -56
  751. package/dist/rl/run-record-adapters.d.ts.map +0 -1
  752. package/dist/rl/sim-fidelity.d.ts +0 -166
  753. package/dist/rl/sim-fidelity.d.ts.map +0 -1
  754. package/dist/rl/sim-fidelity.test.d.ts +0 -2
  755. package/dist/rl/sim-fidelity.test.d.ts.map +0 -1
  756. package/dist/rl/tournament.d.ts +0 -115
  757. package/dist/rl/tournament.d.ts.map +0 -1
  758. package/dist/rl/verifiable-reward.d.ts +0 -124
  759. package/dist/rl/verifiable-reward.d.ts.map +0 -1
  760. package/dist/run-critic.d.ts +0 -23
  761. package/dist/run-critic.d.ts.map +0 -1
  762. package/dist/run-evidence.d.ts +0 -32
  763. package/dist/run-evidence.d.ts.map +0 -1
  764. package/dist/run-record.d.ts.map +0 -1
  765. package/dist/run-record.test.d.ts +0 -2
  766. package/dist/run-record.test.d.ts.map +0 -1
  767. package/dist/run-score.d.ts +0 -31
  768. package/dist/run-score.d.ts.map +0 -1
  769. package/dist/runtime-trajectory.d.ts +0 -47
  770. package/dist/runtime-trajectory.d.ts.map +0 -1
  771. package/dist/sandbox-harness.d.ts.map +0 -1
  772. package/dist/sandbox-harness.test.d.ts +0 -2
  773. package/dist/sandbox-harness.test.d.ts.map +0 -1
  774. package/dist/sandbox-pool.d.ts +0 -74
  775. package/dist/sandbox-pool.d.ts.map +0 -1
  776. package/dist/sandbox-pool.test.d.ts +0 -2
  777. package/dist/sandbox-pool.test.d.ts.map +0 -1
  778. package/dist/scorecard.d.ts +0 -133
  779. package/dist/scorecard.d.ts.map +0 -1
  780. package/dist/scorecard.test.d.ts +0 -2
  781. package/dist/scorecard.test.d.ts.map +0 -1
  782. package/dist/self-play.d.ts +0 -69
  783. package/dist/self-play.d.ts.map +0 -1
  784. package/dist/semantic-concept-judge.d.ts +0 -135
  785. package/dist/semantic-concept-judge.d.ts.map +0 -1
  786. package/dist/semantic-concept-judge.test.d.ts +0 -2
  787. package/dist/semantic-concept-judge.test.d.ts.map +0 -1
  788. package/dist/sequential.d.ts.map +0 -1
  789. package/dist/series-convergence.d.ts.map +0 -1
  790. package/dist/slo.d.ts +0 -48
  791. package/dist/slo.d.ts.map +0 -1
  792. package/dist/state-continuity.d.ts +0 -47
  793. package/dist/state-continuity.d.ts.map +0 -1
  794. package/dist/statistics.d.ts.map +0 -1
  795. package/dist/statistics.test.d.ts +0 -2
  796. package/dist/statistics.test.d.ts.map +0 -1
  797. package/dist/steering-optimizer.d.ts +0 -58
  798. package/dist/steering-optimizer.d.ts.map +0 -1
  799. package/dist/steering.d.ts +0 -24
  800. package/dist/steering.d.ts.map +0 -1
  801. package/dist/storyboard/code-edit.d.ts +0 -64
  802. package/dist/storyboard/code-edit.d.ts.map +0 -1
  803. package/dist/storyboard/code-edit.test.d.ts +0 -2
  804. package/dist/storyboard/code-edit.test.d.ts.map +0 -1
  805. package/dist/storyboard/index.d.ts.map +0 -1
  806. package/dist/storyboard/index.test.d.ts +0 -2
  807. package/dist/storyboard/index.test.d.ts.map +0 -1
  808. package/dist/summary-report.d.ts.map +0 -1
  809. package/dist/telemetry/client.d.ts +0 -35
  810. package/dist/telemetry/client.d.ts.map +0 -1
  811. package/dist/telemetry/index.d.ts.map +0 -1
  812. package/dist/telemetry/schema.d.ts +0 -61
  813. package/dist/telemetry/schema.d.ts.map +0 -1
  814. package/dist/telemetry/sink-fetch.d.ts +0 -39
  815. package/dist/telemetry/sink-fetch.d.ts.map +0 -1
  816. package/dist/telemetry/sink-file.d.ts.map +0 -1
  817. package/dist/test-graded-scenario.d.ts +0 -42
  818. package/dist/test-graded-scenario.d.ts.map +0 -1
  819. package/dist/testing.d.ts.map +0 -1
  820. package/dist/tool-use-metrics.d.ts +0 -35
  821. package/dist/tool-use-metrics.d.ts.map +0 -1
  822. package/dist/trace/capture-fetch.d.ts +0 -48
  823. package/dist/trace/capture-fetch.d.ts.map +0 -1
  824. package/dist/trace/capture-fetch.test.d.ts +0 -2
  825. package/dist/trace/capture-fetch.test.d.ts.map +0 -1
  826. package/dist/trace/emitter.d.ts.map +0 -1
  827. package/dist/trace/extract-usage.d.ts +0 -46
  828. package/dist/trace/extract-usage.d.ts.map +0 -1
  829. package/dist/trace/extract-usage.test.d.ts +0 -2
  830. package/dist/trace/extract-usage.test.d.ts.map +0 -1
  831. package/dist/trace/index.d.ts +0 -15
  832. package/dist/trace/index.d.ts.map +0 -1
  833. package/dist/trace/integrity.d.ts.map +0 -1
  834. package/dist/trace/otel-bridge.d.ts +0 -29
  835. package/dist/trace/otel-bridge.d.ts.map +0 -1
  836. package/dist/trace/otel-export.d.ts +0 -52
  837. package/dist/trace/otel-export.d.ts.map +0 -1
  838. package/dist/trace/otel.d.ts +0 -57
  839. package/dist/trace/otel.d.ts.map +0 -1
  840. package/dist/trace/otlp-attributes.d.ts +0 -17
  841. package/dist/trace/otlp-attributes.d.ts.map +0 -1
  842. package/dist/trace/query.d.ts +0 -29
  843. package/dist/trace/query.d.ts.map +0 -1
  844. package/dist/trace/query.test.d.ts +0 -2
  845. package/dist/trace/query.test.d.ts.map +0 -1
  846. package/dist/trace/raw-provider-sink.d.ts.map +0 -1
  847. package/dist/trace/redact.d.ts.map +0 -1
  848. package/dist/trace/schema.d.ts.map +0 -1
  849. package/dist/trace/store-to-otlp.d.ts +0 -72
  850. package/dist/trace/store-to-otlp.d.ts.map +0 -1
  851. package/dist/trace/store-to-otlp.test.d.ts +0 -2
  852. package/dist/trace/store-to-otlp.test.d.ts.map +0 -1
  853. package/dist/trace/store.d.ts.map +0 -1
  854. package/dist/trace/store.test.d.ts +0 -2
  855. package/dist/trace/store.test.d.ts.map +0 -1
  856. package/dist/trace-analyst/analyst.d.ts.map +0 -1
  857. package/dist/trace-analyst/analyst.test.d.ts +0 -2
  858. package/dist/trace-analyst/analyst.test.d.ts.map +0 -1
  859. package/dist/trace-analyst/behavioral-metrics.d.ts +0 -40
  860. package/dist/trace-analyst/behavioral-metrics.d.ts.map +0 -1
  861. package/dist/trace-analyst/behavioral-metrics.test.d.ts +0 -2
  862. package/dist/trace-analyst/behavioral-metrics.test.d.ts.map +0 -1
  863. package/dist/trace-analyst/hook.d.ts +0 -55
  864. package/dist/trace-analyst/hook.d.ts.map +0 -1
  865. package/dist/trace-analyst/index.d.ts +0 -18
  866. package/dist/trace-analyst/index.d.ts.map +0 -1
  867. package/dist/trace-analyst/insights.d.ts +0 -71
  868. package/dist/trace-analyst/insights.d.ts.map +0 -1
  869. package/dist/trace-analyst/insights.test.d.ts +0 -2
  870. package/dist/trace-analyst/insights.test.d.ts.map +0 -1
  871. package/dist/trace-analyst/otlp-flatten.d.ts +0 -42
  872. package/dist/trace-analyst/otlp-flatten.d.ts.map +0 -1
  873. package/dist/trace-analyst/otlp-span.d.ts +0 -85
  874. package/dist/trace-analyst/otlp-span.d.ts.map +0 -1
  875. package/dist/trace-analyst/otlp-span.test.d.ts +0 -2
  876. package/dist/trace-analyst/otlp-span.test.d.ts.map +0 -1
  877. package/dist/trace-analyst/otlp-to-run-records.d.ts +0 -115
  878. package/dist/trace-analyst/otlp-to-run-records.d.ts.map +0 -1
  879. package/dist/trace-analyst/otlp-to-run-records.test.d.ts +0 -2
  880. package/dist/trace-analyst/otlp-to-run-records.test.d.ts.map +0 -1
  881. package/dist/trace-analyst/otlp-to-run-records.timestamps.test.d.ts +0 -2
  882. package/dist/trace-analyst/otlp-to-run-records.timestamps.test.d.ts.map +0 -1
  883. package/dist/trace-analyst/prompts.d.ts +0 -6
  884. package/dist/trace-analyst/prompts.d.ts.map +0 -1
  885. package/dist/trace-analyst/store-otlp.d.ts +0 -126
  886. package/dist/trace-analyst/store-otlp.d.ts.map +0 -1
  887. package/dist/trace-analyst/store-otlp.test.d.ts +0 -8
  888. package/dist/trace-analyst/store-otlp.test.d.ts.map +0 -1
  889. package/dist/trace-analyst/store-otlp.timestamps.test.d.ts +0 -2
  890. package/dist/trace-analyst/store-otlp.timestamps.test.d.ts.map +0 -1
  891. package/dist/trace-analyst/store.d.ts +0 -63
  892. package/dist/trace-analyst/store.d.ts.map +0 -1
  893. package/dist/trace-analyst/tools.d.ts +0 -44
  894. package/dist/trace-analyst/tools.d.ts.map +0 -1
  895. package/dist/trace-analyst/tools.test.d.ts +0 -10
  896. package/dist/trace-analyst/tools.test.d.ts.map +0 -1
  897. package/dist/trace-analyst/types.d.ts.map +0 -1
  898. package/dist/trace-contracts.d.ts +0 -180
  899. package/dist/trace-contracts.d.ts.map +0 -1
  900. package/dist/traced-analyst.d.ts +0 -26
  901. package/dist/traced-analyst.d.ts.map +0 -1
  902. package/dist/traced-judges.d.ts +0 -27
  903. package/dist/traced-judges.d.ts.map +0 -1
  904. package/dist/traces.d.ts.map +0 -1
  905. package/dist/trajectory.d.ts.map +0 -1
  906. package/dist/types.d.ts.map +0 -1
  907. package/dist/ui-finding.d.ts +0 -104
  908. package/dist/ui-finding.d.ts.map +0 -1
  909. package/dist/verdict-cache.d.ts +0 -78
  910. package/dist/verdict-cache.d.ts.map +0 -1
  911. package/dist/verdict-cache.test.d.ts +0 -2
  912. package/dist/verdict-cache.test.d.ts.map +0 -1
  913. package/dist/verdict.d.ts.map +0 -1
  914. package/dist/visual-diff.d.ts +0 -32
  915. package/dist/visual-diff.d.ts.map +0 -1
  916. package/dist/wire/handlers.d.ts +0 -54
  917. package/dist/wire/handlers.d.ts.map +0 -1
  918. package/dist/wire/index.d.ts.map +0 -1
  919. package/dist/wire/openapi.d.ts +0 -3
  920. package/dist/wire/openapi.d.ts.map +0 -1
  921. package/dist/wire/rpc.d.ts +0 -21
  922. package/dist/wire/rpc.d.ts.map +0 -1
  923. package/dist/wire/rubrics.d.ts +0 -34
  924. package/dist/wire/rubrics.d.ts.map +0 -1
  925. package/dist/wire/schemas.d.ts +0 -410
  926. package/dist/wire/schemas.d.ts.map +0 -1
  927. package/dist/wire/server.d.ts +0 -60
  928. package/dist/wire/server.d.ts.map +0 -1
  929. package/dist/workflow/event-schema.d.ts +0 -5
  930. package/dist/workflow/event-schema.d.ts.map +0 -1
  931. package/dist/workflow/feedback-pack.d.ts +0 -99
  932. package/dist/workflow/feedback-pack.d.ts.map +0 -1
  933. package/dist/workflow/index.d.ts.map +0 -1
  934. package/dist/workflow/intelligence-export.d.ts +0 -62
  935. package/dist/workflow/intelligence-export.d.ts.map +0 -1
  936. package/dist/workflow/partner-report.d.ts +0 -49
  937. package/dist/workflow/partner-report.d.ts.map +0 -1
  938. package/dist/workflow/phase-graph.d.ts +0 -43
  939. package/dist/workflow/phase-graph.d.ts.map +0 -1
  940. package/dist/workflow/promotion-gate.d.ts +0 -61
  941. package/dist/workflow/promotion-gate.d.ts.map +0 -1
  942. package/dist/workflow/run-record.d.ts +0 -12
  943. package/dist/workflow/run-record.d.ts.map +0 -1
  944. package/dist/workflow/runtime-adapter.d.ts +0 -20
  945. package/dist/workflow/runtime-adapter.d.ts.map +0 -1
  946. package/dist/workflow/sanitize.d.ts +0 -21
  947. package/dist/workflow/sanitize.d.ts.map +0 -1
  948. package/dist/workflow/schema.d.ts +0 -5
  949. package/dist/workflow/schema.d.ts.map +0 -1
  950. package/dist/workflow/summary.d.ts +0 -43
  951. package/dist/workflow/summary.d.ts.map +0 -1
  952. package/dist/workflow/trace-event-fields.d.ts +0 -6
  953. package/dist/workflow/trace-event-fields.d.ts.map +0 -1
  954. package/dist/workflow/trajectory.d.ts +0 -15
  955. package/dist/workflow/trajectory.d.ts.map +0 -1
  956. package/dist/workflow/types.d.ts +0 -68
  957. package/dist/workflow/types.d.ts.map +0 -1
  958. package/dist/workspace-inspector.d.ts +0 -67
  959. package/dist/workspace-inspector.d.ts.map +0 -1
  960. package/dist/wrangler-deploy-runner.test.d.ts +0 -2
  961. package/dist/wrangler-deploy-runner.test.d.ts.map +0 -1
@@ -1 +0,0 @@
1
- {"version":3,"file":"phase0-measurement.d.ts","sourceRoot":"","sources":["../../src/belief-state/phase0-measurement.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,eAAe,CAAA;AAChD,OAAO,EACL,KAAK,oCAAoC,EACzC,KAAK,gDAAgD,EAEtD,MAAM,qBAAqB,CAAA;AAC5B,OAAO,EACL,KAAK,0BAA0B,EAC/B,KAAK,sBAAsB,EAE5B,MAAM,iBAAiB,CAAA;AACxB,OAAO,KAAK,EAAE,qBAAqB,EAAE,mBAAmB,EAAE,MAAM,SAAS,CAAA;AAEzE,MAAM,WAAW,4BAA4B;IAC3C,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,QAAQ,EAAE,WAAW,CAAA;CACtB;AAED,MAAM,WAAW,0BAA0B;IACzC,UAAU,EAAE,MAAM,CAAA;IAClB,YAAY,EAAE,MAAM,CAAA;IACpB,OAAO,EAAE,qBAAqB,CAAA;IAC9B,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,IAAI,CAAC,EAAE,MAAM,GAAG,IAAI,CAAA;IACpB,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,QAAQ,CAAC,EAAE,WAAW,CAAA;IACtB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,0CACf,SAAQ,IAAI,CAAC,gDAAgD,EAAE,QAAQ,CAAC;IACxE,IAAI,EAAE,4BAA4B,EAAE,CAAA;IACpC,SAAS,EAAE,0BAA0B,EAAE,CAAA;IACvC,MAAM,CAAC,EAAE,sBAAsB,EAAE,CAAA;IACjC,MAAM,EAAE,0BAA0B,EAAE,CAAA;IACpC,gBAAgB,CAAC,EAAE,MAAM,CAAA;CAC1B;AAED,MAAM,WAAW,qCAAqC;IACpD,QAAQ,EAAE,MAAM,CAAA;IAChB,qBAAqB,EAAE,MAAM,CAAA;IAC7B,mBAAmB,EAAE,MAAM,CAAA;IAC3B,UAAU,EAAE,MAAM,CAAA;IAClB,mBAAmB,EAAE,MAAM,CAAA;IAC3B,WAAW,EAAE,MAAM,CAAA;IACnB,aAAa,EAAE,MAAM,CAAA;IACrB,qBAAqB,EAAE,MAAM,CAAA;IAC7B,iBAAiB,EAAE,MAAM,CAAA;IACzB,YAAY,EAAE,MAAM,CAAA;IACpB,WAAW,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,MAAM,CAAA;IACjB,gBAAgB,EAAE,MAAM,CAAA;IACxB,cAAc,EAAE,MAAM,CAAA;IACtB,gBAAgB,EAAE,MAAM,CAAA;IACxB,YAAY,EAAE,oCAAoC,CAAC,QAAQ,CAAC,CAAA;IAC5D,UAAU,EAAE,oCAAoC,CAAC,YAAY,CAAC,CAAA;CAC/D;AAED,MAAM,WAAW,8BAA8B;IAC7C,MAAM,EAAE,mBAAmB,EAAE,CAAA;IAC7B,MAAM,EAAE,oCAAoC,CAAA;IAC5C,OAAO,EAAE,qCAAqC,CAAA;IAC9C,WAAW,EAAE,MAAM,EAAE,CAAA;CACtB;AAID,wBAAgB,mCAAmC,CACjD,OAAO,EAAE,0CAA0C,GAClD,8BAA8B,CAoEhC"}
@@ -1,17 +0,0 @@
1
- import { type BeliefCalibrationOptions } from './calibration';
2
- import { type BeliefOpeOptions } from './ope';
3
- import { type EvaluateBeliefSelectivePolicyOptions } from './selective';
4
- import type { BeliefDecisionPoint, BeliefOpeTargetPolicy, BeliefPolicyEvaluationReport, BeliefSelectivePolicy } from './types';
5
- export interface AnalyzeBeliefPolicyOpeOptions extends BeliefOpeOptions {
6
- targetPolicy?: BeliefOpeTargetPolicy;
7
- }
8
- export interface AnalyzeBeliefPolicyOptions {
9
- points: BeliefDecisionPoint[];
10
- policy: BeliefSelectivePolicy;
11
- selective?: EvaluateBeliefSelectivePolicyOptions;
12
- calibration?: BeliefCalibrationOptions;
13
- ope?: AnalyzeBeliefPolicyOpeOptions;
14
- requireOpe?: boolean;
15
- }
16
- export declare function analyzeBeliefPolicy(options: AnalyzeBeliefPolicyOptions): BeliefPolicyEvaluationReport;
17
- //# sourceMappingURL=report.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"report.d.ts","sourceRoot":"","sources":["../../src/belief-state/report.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,wBAAwB,EAA4B,MAAM,eAAe,CAAA;AACvF,OAAO,EAAE,KAAK,gBAAgB,EAA2B,MAAM,OAAO,CAAA;AACtE,OAAO,EACL,KAAK,oCAAoC,EAE1C,MAAM,aAAa,CAAA;AACpB,OAAO,KAAK,EACV,mBAAmB,EAGnB,qBAAqB,EACrB,4BAA4B,EAC5B,qBAAqB,EACtB,MAAM,SAAS,CAAA;AAEhB,MAAM,WAAW,6BAA8B,SAAQ,gBAAgB;IACrE,YAAY,CAAC,EAAE,qBAAqB,CAAA;CACrC;AAED,MAAM,WAAW,0BAA0B;IACzC,MAAM,EAAE,mBAAmB,EAAE,CAAA;IAC7B,MAAM,EAAE,qBAAqB,CAAA;IAC7B,SAAS,CAAC,EAAE,oCAAoC,CAAA;IAChD,WAAW,CAAC,EAAE,wBAAwB,CAAA;IACtC,GAAG,CAAC,EAAE,6BAA6B,CAAA;IACnC,UAAU,CAAC,EAAE,OAAO,CAAA;CACrB;AAED,wBAAgB,mBAAmB,CACjC,OAAO,EAAE,0BAA0B,GAClC,4BAA4B,CA4C9B"}
@@ -1,2 +0,0 @@
1
- export {};
2
- //# sourceMappingURL=report.test.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"report.test.d.ts","sourceRoot":"","sources":["../../src/belief-state/report.test.ts"],"names":[],"mappings":""}
@@ -1,23 +0,0 @@
1
- import { type AnalyzeBeliefDecisionCorpusOptions, type BeliefDecisionCorpusEvaluation } from './code-agent-corpus';
2
- export type BeliefResearchClaimScope = 'selective' | 'counterfactual';
3
- export type BeliefResearchEvidenceStatus = 'supported' | 'blocked';
4
- export type BeliefResearchGateId = 'corpus' | 'selective' | 'calibration' | 'ope';
5
- export interface BeliefResearchEvidenceGate {
6
- id: BeliefResearchGateId;
7
- status: BeliefResearchEvidenceStatus;
8
- blockers: string[];
9
- caveats: string[];
10
- }
11
- export interface BeliefDecisionResearchEvidencePacket {
12
- claimScope: BeliefResearchClaimScope;
13
- status: BeliefResearchEvidenceStatus;
14
- analysis: BeliefDecisionCorpusEvaluation;
15
- gates: BeliefResearchEvidenceGate[];
16
- blockers: string[];
17
- caveats: string[];
18
- }
19
- export interface BuildBeliefDecisionResearchEvidencePacketOptions extends AnalyzeBeliefDecisionCorpusOptions {
20
- claimScope?: BeliefResearchClaimScope;
21
- }
22
- export declare function buildBeliefDecisionResearchEvidencePacket(options: BuildBeliefDecisionResearchEvidencePacketOptions): BeliefDecisionResearchEvidencePacket;
23
- //# sourceMappingURL=research-evidence.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"research-evidence.d.ts","sourceRoot":"","sources":["../../src/belief-state/research-evidence.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,KAAK,kCAAkC,EAEvC,KAAK,8BAA8B,EACpC,MAAM,qBAAqB,CAAA;AAE5B,MAAM,MAAM,wBAAwB,GAAG,WAAW,GAAG,gBAAgB,CAAA;AACrE,MAAM,MAAM,4BAA4B,GAAG,WAAW,GAAG,SAAS,CAAA;AAClE,MAAM,MAAM,oBAAoB,GAAG,QAAQ,GAAG,WAAW,GAAG,aAAa,GAAG,KAAK,CAAA;AAEjF,MAAM,WAAW,0BAA0B;IACzC,EAAE,EAAE,oBAAoB,CAAA;IACxB,MAAM,EAAE,4BAA4B,CAAA;IACpC,QAAQ,EAAE,MAAM,EAAE,CAAA;IAClB,OAAO,EAAE,MAAM,EAAE,CAAA;CAClB;AAED,MAAM,WAAW,oCAAoC;IACnD,UAAU,EAAE,wBAAwB,CAAA;IACpC,MAAM,EAAE,4BAA4B,CAAA;IACpC,QAAQ,EAAE,8BAA8B,CAAA;IACxC,KAAK,EAAE,0BAA0B,EAAE,CAAA;IACnC,QAAQ,EAAE,MAAM,EAAE,CAAA;IAClB,OAAO,EAAE,MAAM,EAAE,CAAA;CAClB;AAED,MAAM,WAAW,gDACf,SAAQ,kCAAkC;IAC1C,UAAU,CAAC,EAAE,wBAAwB,CAAA;CACtC;AAED,wBAAgB,yCAAyC,CACvD,OAAO,EAAE,gDAAgD,GACxD,oCAAoC,CA4BtC"}
@@ -1,2 +0,0 @@
1
- export {};
2
- //# sourceMappingURL=research-evidence.test.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"research-evidence.test.d.ts","sourceRoot":"","sources":["../../src/belief-state/research-evidence.test.ts"],"names":[],"mappings":""}
@@ -1,32 +0,0 @@
1
- import { type ProjectRuntimeTrajectoryEvidenceOptions, type RuntimeTrajectoryEvidenceProjection, type RuntimeTrajectoryRecord } from '../runtime-trajectory';
2
- import { type BuildRuntimeBeliefPhase0MeasurementOptions, type RuntimeBeliefDecisionLabel, type RuntimeBeliefPhase0Measurement, type RuntimeBeliefPhase0RunRecord } from './phase0-measurement';
3
- import type { RuntimeBeliefDecisionPoint, RuntimeBeliefHookEvent } from './runtime-hooks';
4
- type RuntimeBenchmarkTrajectoryRecord = RuntimeTrajectoryRecord & {
5
- benchmark?: unknown;
6
- condition?: unknown;
7
- instanceId?: unknown;
8
- runtimeDecisionPoints?: unknown;
9
- };
10
- export interface BuildRuntimeBenchmarkBeliefPhase0MeasurementOptions extends Omit<BuildRuntimeBeliefPhase0MeasurementOptions, 'runs' | 'events' | 'decisions' | 'labels'> {
11
- records: RuntimeBenchmarkTrajectoryRecord[];
12
- decisions?: RuntimeBeliefDecisionPoint[];
13
- defaultSplitTag?: ProjectRuntimeTrajectoryEvidenceOptions['defaultSplitTag'];
14
- labels?: RuntimeBeliefDecisionLabel[];
15
- }
16
- export interface RuntimeBenchmarkBeliefPhase0Summary {
17
- decisionCount: number;
18
- labelCount: number;
19
- }
20
- export interface RuntimeBenchmarkBeliefPhase0Measurement {
21
- runs: RuntimeBeliefPhase0RunRecord[];
22
- events: RuntimeBeliefHookEvent[];
23
- decisions: RuntimeBeliefDecisionPoint[];
24
- labels: RuntimeBeliefDecisionLabel[];
25
- trajectory: RuntimeTrajectoryEvidenceProjection;
26
- measurement: RuntimeBeliefPhase0Measurement;
27
- summary: RuntimeBenchmarkBeliefPhase0Summary;
28
- diagnostics: string[];
29
- }
30
- export declare function buildRuntimeBenchmarkBeliefPhase0Measurement(options: BuildRuntimeBenchmarkBeliefPhase0MeasurementOptions): RuntimeBenchmarkBeliefPhase0Measurement;
31
- export {};
32
- //# sourceMappingURL=runtime-benchmark-corpus.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"runtime-benchmark-corpus.d.ts","sourceRoot":"","sources":["../../src/belief-state/runtime-benchmark-corpus.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,KAAK,uCAAuC,EAE5C,KAAK,mCAAmC,EACxC,KAAK,uBAAuB,EAC7B,MAAM,uBAAuB,CAAA;AAC9B,OAAO,EACL,KAAK,0CAA0C,EAE/C,KAAK,0BAA0B,EAC/B,KAAK,8BAA8B,EACnC,KAAK,4BAA4B,EAClC,MAAM,sBAAsB,CAAA;AAC7B,OAAO,KAAK,EAAE,0BAA0B,EAAE,sBAAsB,EAAE,MAAM,iBAAiB,CAAA;AAEzF,KAAK,gCAAgC,GAAG,uBAAuB,GAAG;IAChE,SAAS,CAAC,EAAE,OAAO,CAAA;IACnB,SAAS,CAAC,EAAE,OAAO,CAAA;IACnB,UAAU,CAAC,EAAE,OAAO,CAAA;IACpB,qBAAqB,CAAC,EAAE,OAAO,CAAA;CAChC,CAAA;AAmBD,MAAM,WAAW,mDACf,SAAQ,IAAI,CACV,0CAA0C,EAC1C,MAAM,GAAG,QAAQ,GAAG,WAAW,GAAG,QAAQ,CAC3C;IACD,OAAO,EAAE,gCAAgC,EAAE,CAAA;IAC3C,SAAS,CAAC,EAAE,0BAA0B,EAAE,CAAA;IACxC,eAAe,CAAC,EAAE,uCAAuC,CAAC,iBAAiB,CAAC,CAAA;IAC5E,MAAM,CAAC,EAAE,0BAA0B,EAAE,CAAA;CACtC;AAED,MAAM,WAAW,mCAAmC;IAClD,aAAa,EAAE,MAAM,CAAA;IACrB,UAAU,EAAE,MAAM,CAAA;CACnB;AAED,MAAM,WAAW,uCAAuC;IACtD,IAAI,EAAE,4BAA4B,EAAE,CAAA;IACpC,MAAM,EAAE,sBAAsB,EAAE,CAAA;IAChC,SAAS,EAAE,0BAA0B,EAAE,CAAA;IACvC,MAAM,EAAE,0BAA0B,EAAE,CAAA;IACpC,UAAU,EAAE,mCAAmC,CAAA;IAC/C,WAAW,EAAE,8BAA8B,CAAA;IAC3C,OAAO,EAAE,mCAAmC,CAAA;IAC5C,WAAW,EAAE,MAAM,EAAE,CAAA;CACtB;AAED,wBAAgB,4CAA4C,CAC1D,OAAO,EAAE,mDAAmD,GAC3D,uCAAuC,CA2CzC"}
@@ -1,87 +0,0 @@
1
- import type { BeliefShadowProbeInput } from './shadow-probe';
2
- import type { BeliefDecisionKind, BeliefDecisionOutcome, BeliefDecisionPoint, BeliefEvidenceQuality } from './types';
3
- export interface RuntimeBeliefDecisionEvidenceRef {
4
- source: string;
5
- id: string;
6
- detail?: string;
7
- quality?: BeliefEvidenceQuality;
8
- metadata?: Record<string, unknown>;
9
- }
10
- export interface RuntimeBeliefDecisionPoint {
11
- id: string;
12
- runId: string;
13
- scenarioId?: string;
14
- stepIndex: number;
15
- kind: string;
16
- candidateActions?: string[];
17
- context?: string;
18
- evidence?: RuntimeBeliefDecisionEvidenceRef[];
19
- metadata?: Record<string, unknown>;
20
- }
21
- export interface RuntimeBeliefHookEvent {
22
- id: string;
23
- runId: string;
24
- scenarioId?: string;
25
- target: string;
26
- phase: string;
27
- timestamp: number;
28
- stepIndex?: number;
29
- parentId?: string;
30
- payload?: unknown;
31
- metadata?: Record<string, unknown>;
32
- }
33
- export interface RuntimeBeliefHookContext {
34
- signal?: AbortSignal;
35
- }
36
- export interface RuntimeBeliefHooks {
37
- onEvent?: (event: RuntimeBeliefHookEvent, context: RuntimeBeliefHookContext) => void | Promise<void>;
38
- onDecisionPoint?: (point: RuntimeBeliefDecisionPoint, context: RuntimeBeliefHookContext) => void | Promise<void>;
39
- }
40
- export interface RuntimeBeliefConversionDiagnostic {
41
- decisionId: string;
42
- severity: 'warning' | 'error';
43
- reason: string;
44
- }
45
- export interface RuntimeBeliefShadowProbeInputOptions {
46
- probeId: string;
47
- decisionKind?: BeliefDecisionKind;
48
- includeEvidenceDetail?: boolean;
49
- includeLifecycleEvidence?: boolean;
50
- lifecycleEvents?: RuntimeBeliefHookEvent[];
51
- maxContextChars?: number;
52
- }
53
- export interface RuntimeBeliefDecisionPointOptions {
54
- chosenAction?: string;
55
- decisionKind?: BeliefDecisionKind;
56
- confidence?: number;
57
- behaviorProb?: number;
58
- targetProb?: number;
59
- qHat?: number | null;
60
- costUsd?: number;
61
- outcome?: BeliefDecisionOutcome;
62
- metadata?: Record<string, unknown>;
63
- includeLifecycleEvidence?: boolean;
64
- lifecycleEvents?: RuntimeBeliefHookEvent[];
65
- }
66
- export interface RuntimeBeliefShadowProbeInputReport {
67
- input?: BeliefShadowProbeInput;
68
- diagnostics: RuntimeBeliefConversionDiagnostic[];
69
- }
70
- export interface RuntimeBeliefDecisionPointReport {
71
- point?: BeliefDecisionPoint;
72
- diagnostics: RuntimeBeliefConversionDiagnostic[];
73
- }
74
- export interface BeliefRuntimeHookCollector {
75
- hooks: RuntimeBeliefHooks;
76
- decisions: RuntimeBeliefDecisionPoint[];
77
- events: RuntimeBeliefHookEvent[];
78
- toShadowProbeInputs(options?: Partial<RuntimeBeliefShadowProbeInputOptions>): {
79
- inputs: BeliefShadowProbeInput[];
80
- diagnostics: RuntimeBeliefConversionDiagnostic[];
81
- };
82
- clear(): void;
83
- }
84
- export declare function runtimeDecisionPointToBeliefShadowProbeInput(point: RuntimeBeliefDecisionPoint, options: RuntimeBeliefShadowProbeInputOptions): RuntimeBeliefShadowProbeInputReport;
85
- export declare function runtimeDecisionPointToBeliefDecisionPoint(point: RuntimeBeliefDecisionPoint, options: RuntimeBeliefDecisionPointOptions): RuntimeBeliefDecisionPointReport;
86
- export declare function createBeliefRuntimeHookCollector(defaults: RuntimeBeliefShadowProbeInputOptions): BeliefRuntimeHookCollector;
87
- //# sourceMappingURL=runtime-hooks.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"runtime-hooks.d.ts","sourceRoot":"","sources":["../../src/belief-state/runtime-hooks.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,sBAAsB,EAAE,MAAM,gBAAgB,CAAA;AAC5D,OAAO,KAAK,EACV,kBAAkB,EAClB,qBAAqB,EACrB,mBAAmB,EACnB,qBAAqB,EAEtB,MAAM,SAAS,CAAA;AAGhB,MAAM,WAAW,gCAAgC;IAC/C,MAAM,EAAE,MAAM,CAAA;IACd,EAAE,EAAE,MAAM,CAAA;IACV,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,OAAO,CAAC,EAAE,qBAAqB,CAAA;IAC/B,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,0BAA0B;IACzC,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,MAAM,CAAA;IACjB,IAAI,EAAE,MAAM,CAAA;IACZ,gBAAgB,CAAC,EAAE,MAAM,EAAE,CAAA;IAC3B,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,QAAQ,CAAC,EAAE,gCAAgC,EAAE,CAAA;IAC7C,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,sBAAsB;IACrC,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,MAAM,EAAE,MAAM,CAAA;IACd,KAAK,EAAE,MAAM,CAAA;IACb,SAAS,EAAE,MAAM,CAAA;IACjB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,OAAO,CAAC,EAAE,OAAO,CAAA;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,wBAAwB;IACvC,MAAM,CAAC,EAAE,WAAW,CAAA;CACrB;AAED,MAAM,WAAW,kBAAkB;IACjC,OAAO,CAAC,EAAE,CACR,KAAK,EAAE,sBAAsB,EAC7B,OAAO,EAAE,wBAAwB,KAC9B,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IACzB,eAAe,CAAC,EAAE,CAChB,KAAK,EAAE,0BAA0B,EACjC,OAAO,EAAE,wBAAwB,KAC9B,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;CAC1B;AAED,MAAM,WAAW,iCAAiC;IAChD,UAAU,EAAE,MAAM,CAAA;IAClB,QAAQ,EAAE,SAAS,GAAG,OAAO,CAAA;IAC7B,MAAM,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,oCAAoC;IACnD,OAAO,EAAE,MAAM,CAAA;IACf,YAAY,CAAC,EAAE,kBAAkB,CAAA;IACjC,qBAAqB,CAAC,EAAE,OAAO,CAAA;IAC/B,wBAAwB,CAAC,EAAE,OAAO,CAAA;IAClC,eAAe,CAAC,EAAE,sBAAsB,EAAE,CAAA;IAC1C,eAAe,CAAC,EAAE,MAAM,CAAA;CACzB;AAED,MAAM,WAAW,iCAAiC;IAChD,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,YAAY,CAAC,EAAE,kBAAkB,CAAA;IACjC,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,IAAI,CAAC,EAAE,MAAM,GAAG,IAAI,CAAA;IACpB,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,OAAO,CAAC,EAAE,qBAAqB,CAAA;IAC/B,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IAClC,wBAAwB,CAAC,EAAE,OAAO,CAAA;IAClC,eAAe,CAAC,EAAE,sBAAsB,EAAE,CAAA;CAC3C;AAED,MAAM,WAAW,mCAAmC;IAClD,KAAK,CAAC,EAAE,sBAAsB,CAAA;IAC9B,WAAW,EAAE,iCAAiC,EAAE,CAAA;CACjD;AAED,MAAM,WAAW,gCAAgC;IAC/C,KAAK,CAAC,EAAE,mBAAmB,CAAA;IAC3B,WAAW,EAAE,iCAAiC,EAAE,CAAA;CACjD;AAED,MAAM,WAAW,0BAA0B;IACzC,KAAK,EAAE,kBAAkB,CAAA;IACzB,SAAS,EAAE,0BAA0B,EAAE,CAAA;IACvC,MAAM,EAAE,sBAAsB,EAAE,CAAA;IAChC,mBAAmB,CAAC,OAAO,CAAC,EAAE,OAAO,CAAC,oCAAoC,CAAC,GAAG;QAC5E,MAAM,EAAE,sBAAsB,EAAE,CAAA;QAChC,WAAW,EAAE,iCAAiC,EAAE,CAAA;KACjD,CAAA;IACD,KAAK,IAAI,IAAI,CAAA;CACd;AAKD,wBAAgB,4CAA4C,CAC1D,KAAK,EAAE,0BAA0B,EACjC,OAAO,EAAE,oCAAoC,GAC5C,mCAAmC,CA2BrC;AAED,wBAAgB,yCAAyC,CACvD,KAAK,EAAE,0BAA0B,EACjC,OAAO,EAAE,iCAAiC,GACzC,gCAAgC,CAiDlC;AAED,wBAAgB,gCAAgC,CAC9C,QAAQ,EAAE,oCAAoC,GAC7C,0BAA0B,CAwC5B"}
@@ -1,2 +0,0 @@
1
- export {};
2
- //# sourceMappingURL=runtime-hooks.test.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"runtime-hooks.test.d.ts","sourceRoot":"","sources":["../../src/belief-state/runtime-hooks.test.ts"],"names":[],"mappings":""}
@@ -1,15 +0,0 @@
1
- import type { BeliefDecisionPoint, BeliefPolicyAction, BeliefSelectivePolicy, BeliefSelectivePolicyMetrics, BeliefUtilityOptions } from './types';
2
- export interface EvaluateBeliefSelectivePolicyOptions {
3
- utility?: BeliefUtilityOptions;
4
- minN?: number;
5
- minAccepted?: number;
6
- minUtilityDelta?: number;
7
- seed?: number;
8
- }
9
- export declare function thresholdSelectivePolicy(options: {
10
- id?: string;
11
- confidenceThreshold: number;
12
- belowThresholdAction?: Exclude<BeliefPolicyAction, 'accept'>;
13
- }): BeliefSelectivePolicy;
14
- export declare function evaluateBeliefSelectivePolicy(points: BeliefDecisionPoint[], policy: BeliefSelectivePolicy, options?: EvaluateBeliefSelectivePolicyOptions): BeliefSelectivePolicyMetrics;
15
- //# sourceMappingURL=selective.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"selective.d.ts","sourceRoot":"","sources":["../../src/belief-state/selective.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EACV,mBAAmB,EACnB,kBAAkB,EAClB,qBAAqB,EACrB,4BAA4B,EAC5B,oBAAoB,EACrB,MAAM,SAAS,CAAA;AAEhB,MAAM,WAAW,oCAAoC;IACnD,OAAO,CAAC,EAAE,oBAAoB,CAAA;IAC9B,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,IAAI,CAAC,EAAE,MAAM,CAAA;CACd;AAaD,wBAAgB,wBAAwB,CAAC,OAAO,EAAE;IAChD,EAAE,CAAC,EAAE,MAAM,CAAA;IACX,mBAAmB,EAAE,MAAM,CAAA;IAC3B,oBAAoB,CAAC,EAAE,OAAO,CAAC,kBAAkB,EAAE,QAAQ,CAAC,CAAA;CAC7D,GAAG,qBAAqB,CAsBxB;AAED,wBAAgB,6BAA6B,CAC3C,MAAM,EAAE,mBAAmB,EAAE,EAC7B,MAAM,EAAE,qBAAqB,EAC7B,OAAO,GAAE,oCAAyC,GACjD,4BAA4B,CA+D9B"}
@@ -1,2 +0,0 @@
1
- export {};
2
- //# sourceMappingURL=selective.test.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"selective.test.d.ts","sourceRoot":"","sources":["../../src/belief-state/selective.test.ts"],"names":[],"mappings":""}
@@ -1,80 +0,0 @@
1
- import type { BeliefDecisionKind, BeliefDecisionOutcome, BeliefDecisionPoint, BeliefEvidenceQuality } from './types';
2
- export interface BeliefShadowProbeInput {
3
- probeId: string;
4
- decisionId: string;
5
- runId: string;
6
- scenarioId?: string;
7
- stepIndex: number;
8
- decisionKind: BeliefDecisionKind;
9
- candidateActions: string[];
10
- observedAction?: string;
11
- evidence: BeliefShadowProbeEvidenceRef[];
12
- context?: string;
13
- metadata?: Record<string, unknown>;
14
- }
15
- export interface BeliefShadowProbeEvidenceRef {
16
- id: string;
17
- source: string;
18
- detail?: string;
19
- quality?: BeliefEvidenceQuality;
20
- }
21
- export interface BeliefShadowProbeResponse {
22
- predictedAction: string;
23
- confidence: number;
24
- beliefSummary?: string;
25
- uncertainty?: string[];
26
- evidenceRefs?: string[];
27
- wouldChangeMindIf?: string[];
28
- targetProb?: number;
29
- qHat?: number | null;
30
- metadata?: Record<string, unknown>;
31
- }
32
- export interface BeliefShadowProbeRecord extends BeliefShadowProbeResponse {
33
- probeId: string;
34
- decisionId: string;
35
- runId: string;
36
- scenarioId?: string;
37
- stepIndex: number;
38
- decisionKind: BeliefDecisionKind;
39
- candidateActions: string[];
40
- observedAction: string;
41
- agreesWithObservedAction: boolean;
42
- outcome?: BeliefDecisionOutcome;
43
- }
44
- export interface BeliefShadowProbeDiagnostic {
45
- decisionId: string;
46
- severity: 'warning' | 'error';
47
- reason: string;
48
- }
49
- export interface BeliefShadowProbeSummary {
50
- attempted: number;
51
- completed: number;
52
- dropped: number;
53
- withOutcome: number;
54
- withTargetProb: number;
55
- meanConfidence: number | null;
56
- observedAgreementRate: number | null;
57
- }
58
- export interface BeliefShadowProbeRun {
59
- probeId: string;
60
- records: BeliefShadowProbeRecord[];
61
- diagnostics: BeliefShadowProbeDiagnostic[];
62
- summary: BeliefShadowProbeSummary;
63
- }
64
- export interface RunBeliefShadowProbeOptions {
65
- probeId: string;
66
- points: BeliefDecisionPoint[];
67
- probe: (input: BeliefShadowProbeInput) => BeliefShadowProbeResponse | Promise<BeliefShadowProbeResponse>;
68
- contextOf?: (point: BeliefDecisionPoint) => string | undefined | Promise<string | undefined>;
69
- metadataOf?: (point: BeliefDecisionPoint) => Record<string, unknown> | undefined | Promise<Record<string, unknown> | undefined>;
70
- includeObservedAction?: boolean;
71
- includeEvidenceDetail?: boolean;
72
- includeOutcomeInRecord?: boolean;
73
- requireCandidateActions?: boolean;
74
- allowOutOfSetActions?: boolean;
75
- concurrency?: number;
76
- maxContextChars?: number;
77
- }
78
- export declare function runBeliefShadowProbe(options: RunBeliefShadowProbeOptions): Promise<BeliefShadowProbeRun>;
79
- export declare function formatBeliefShadowProbePrompt(input: BeliefShadowProbeInput): string;
80
- //# sourceMappingURL=shadow-probe.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"shadow-probe.d.ts","sourceRoot":"","sources":["../../src/belief-state/shadow-probe.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,kBAAkB,EAClB,qBAAqB,EACrB,mBAAmB,EACnB,qBAAqB,EACtB,MAAM,SAAS,CAAA;AAEhB,MAAM,WAAW,sBAAsB;IACrC,OAAO,EAAE,MAAM,CAAA;IACf,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,MAAM,CAAA;IACjB,YAAY,EAAE,kBAAkB,CAAA;IAChC,gBAAgB,EAAE,MAAM,EAAE,CAAA;IAC1B,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,QAAQ,EAAE,4BAA4B,EAAE,CAAA;IACxC,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,4BAA4B;IAC3C,EAAE,EAAE,MAAM,CAAA;IACV,MAAM,EAAE,MAAM,CAAA;IACd,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,OAAO,CAAC,EAAE,qBAAqB,CAAA;CAChC;AAED,MAAM,WAAW,yBAAyB;IACxC,eAAe,EAAE,MAAM,CAAA;IACvB,UAAU,EAAE,MAAM,CAAA;IAClB,aAAa,CAAC,EAAE,MAAM,CAAA;IACtB,WAAW,CAAC,EAAE,MAAM,EAAE,CAAA;IACtB,YAAY,CAAC,EAAE,MAAM,EAAE,CAAA;IACvB,iBAAiB,CAAC,EAAE,MAAM,EAAE,CAAA;IAC5B,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,IAAI,CAAC,EAAE,MAAM,GAAG,IAAI,CAAA;IACpB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,uBAAwB,SAAQ,yBAAyB;IACxE,OAAO,EAAE,MAAM,CAAA;IACf,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,MAAM,CAAA;IACjB,YAAY,EAAE,kBAAkB,CAAA;IAChC,gBAAgB,EAAE,MAAM,EAAE,CAAA;IAC1B,cAAc,EAAE,MAAM,CAAA;IACtB,wBAAwB,EAAE,OAAO,CAAA;IACjC,OAAO,CAAC,EAAE,qBAAqB,CAAA;CAChC;AAED,MAAM,WAAW,2BAA2B;IAC1C,UAAU,EAAE,MAAM,CAAA;IAClB,QAAQ,EAAE,SAAS,GAAG,OAAO,CAAA;IAC7B,MAAM,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,wBAAwB;IACvC,SAAS,EAAE,MAAM,CAAA;IACjB,SAAS,EAAE,MAAM,CAAA;IACjB,OAAO,EAAE,MAAM,CAAA;IACf,WAAW,EAAE,MAAM,CAAA;IACnB,cAAc,EAAE,MAAM,CAAA;IACtB,cAAc,EAAE,MAAM,GAAG,IAAI,CAAA;IAC7B,qBAAqB,EAAE,MAAM,GAAG,IAAI,CAAA;CACrC;AAED,MAAM,WAAW,oBAAoB;IACnC,OAAO,EAAE,MAAM,CAAA;IACf,OAAO,EAAE,uBAAuB,EAAE,CAAA;IAClC,WAAW,EAAE,2BAA2B,EAAE,CAAA;IAC1C,OAAO,EAAE,wBAAwB,CAAA;CAClC;AAED,MAAM,WAAW,2BAA2B;IAC1C,OAAO,EAAE,MAAM,CAAA;IACf,MAAM,EAAE,mBAAmB,EAAE,CAAA;IAC7B,KAAK,EAAE,CACL,KAAK,EAAE,sBAAsB,KAC1B,yBAAyB,GAAG,OAAO,CAAC,yBAAyB,CAAC,CAAA;IACnE,SAAS,CAAC,EAAE,CAAC,KAAK,EAAE,mBAAmB,KAAK,MAAM,GAAG,SAAS,GAAG,OAAO,CAAC,MAAM,GAAG,SAAS,CAAC,CAAA;IAC5F,UAAU,CAAC,EAAE,CACX,KAAK,EAAE,mBAAmB,KACvB,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,SAAS,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,SAAS,CAAC,CAAA;IACvF,qBAAqB,CAAC,EAAE,OAAO,CAAA;IAC/B,qBAAqB,CAAC,EAAE,OAAO,CAAA;IAC/B,sBAAsB,CAAC,EAAE,OAAO,CAAA;IAChC,uBAAuB,CAAC,EAAE,OAAO,CAAA;IACjC,oBAAoB,CAAC,EAAE,OAAO,CAAA;IAC9B,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,eAAe,CAAC,EAAE,MAAM,CAAA;CACzB;AAKD,wBAAsB,oBAAoB,CACxC,OAAO,EAAE,2BAA2B,GACnC,OAAO,CAAC,oBAAoB,CAAC,CA2B/B;AAED,wBAAgB,6BAA6B,CAAC,KAAK,EAAE,sBAAsB,GAAG,MAAM,CAwBnF"}
@@ -1,2 +0,0 @@
1
- export {};
2
- //# sourceMappingURL=shadow-probe.test.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"shadow-probe.test.d.ts","sourceRoot":"","sources":["../../src/belief-state/shadow-probe.test.ts"],"names":[],"mappings":""}
@@ -1,195 +0,0 @@
1
- import type { CalibrationReport } from '../meta-eval/calibration';
2
- import type { OffPolicyEstimate } from '../rl/off-policy';
3
- export declare const BELIEF_DECISION_KINDS: readonly ["continue", "verify", "ask", "retry", "stop", "memory-write", "memory-read", "tool-select", "skill-select", "workflow-select", "surface-promote"];
4
- export type BeliefDecisionKind = (typeof BELIEF_DECISION_KINDS)[number];
5
- export declare const BELIEF_EVIDENCE_SOURCES: readonly ["run", "span", "event", "finding", "memory", "knowledge", "policy"];
6
- export type BeliefEvidenceSource = (typeof BELIEF_EVIDENCE_SOURCES)[number];
7
- export declare const BELIEF_EVIDENCE_QUALITIES: readonly ["direct", "derived", "self-reported", "unverified", "stale", "contradicted"];
8
- export type BeliefEvidenceQuality = (typeof BELIEF_EVIDENCE_QUALITIES)[number];
9
- export declare const BELIEF_EVALUATION_CRITERIA: readonly [{
10
- readonly id: "capture-integrity";
11
- readonly label: "Capture integrity";
12
- readonly reasonCodes: readonly ["trace-missing", "run-record-missing", "backend-integrity-missing"];
13
- }, {
14
- readonly id: "decision-completeness";
15
- readonly label: "Decision completeness";
16
- readonly reasonCodes: readonly ["candidate-actions-missing", "chosen-action-missing", "decision-evidence-missing"];
17
- }, {
18
- readonly id: "evidence-quality";
19
- readonly label: "Evidence quality";
20
- readonly reasonCodes: readonly ["evidence-stale", "evidence-contradictory", "evidence-unverified", "evidence-self-reported"];
21
- }, {
22
- readonly id: "outcome-quality";
23
- readonly label: "Outcome quality";
24
- readonly reasonCodes: readonly ["outcome-missing", "outcome-delayed", "cost-missing"];
25
- }, {
26
- readonly id: "calibration";
27
- readonly label: "Calibration";
28
- readonly reasonCodes: readonly ["confidence-missing", "calibration-unsupported", "calibration-gap-high"];
29
- }, {
30
- readonly id: "accepted-region-risk";
31
- readonly label: "Accepted-region risk";
32
- readonly reasonCodes: readonly ["accepted-error-high", "coverage-too-low"];
33
- }, {
34
- readonly id: "policy-value";
35
- readonly label: "Policy value";
36
- readonly reasonCodes: readonly ["utility-lift-missing", "baseline-dominates", "cost-too-high"];
37
- }, {
38
- readonly id: "ope-support";
39
- readonly label: "OPE support";
40
- readonly reasonCodes: readonly ["behavior-propensity-missing", "behavior-propensity-invalid", "target-propensity-missing", "target-propensity-invalid", "effective-sample-size-low", "importance-weight-high"];
41
- }, {
42
- readonly id: "memory-health";
43
- readonly label: "Memory health";
44
- readonly reasonCodes: readonly ["memory-stale", "memory-poisoning-risk", "context-bloat", "memory-write-unverified"];
45
- }, {
46
- readonly id: "surface-attribution";
47
- readonly label: "Surface attribution";
48
- readonly reasonCodes: readonly ["surface-claim-unsupported", "causal-attribution-missing"];
49
- }, {
50
- readonly id: "generalization";
51
- readonly label: "Generalization";
52
- readonly reasonCodes: readonly ["split-missing", "holdout-regression", "task-family-coverage-low", "leakage-risk"];
53
- }, {
54
- readonly id: "promotion";
55
- readonly label: "Promotion";
56
- readonly reasonCodes: readonly ["negative-control-failed", "promotion-gate-failed", "human-review-required"];
57
- }];
58
- export type BeliefEvaluationCriterionId = (typeof BELIEF_EVALUATION_CRITERIA)[number]['id'];
59
- export type BeliefDecisionReasonCode = (typeof BELIEF_EVALUATION_CRITERIA)[number]['reasonCodes'][number];
60
- export interface BeliefDecisionReason {
61
- code: BeliefDecisionReasonCode;
62
- criterion?: BeliefEvaluationCriterionId;
63
- detail?: string;
64
- evidenceIds?: string[];
65
- metadata?: Record<string, unknown>;
66
- }
67
- export declare function isBeliefDecisionKind(value: unknown): value is BeliefDecisionKind;
68
- export declare function isBeliefEvidenceSource(value: unknown): value is BeliefEvidenceSource;
69
- export interface BeliefEvidenceRef {
70
- source: BeliefEvidenceSource;
71
- id: string;
72
- runId?: string;
73
- spanId?: string;
74
- eventId?: string;
75
- detail?: string;
76
- quality?: BeliefEvidenceQuality;
77
- observedAt?: string;
78
- metadata?: Record<string, unknown>;
79
- }
80
- export interface BeliefDecisionOutcome {
81
- success?: boolean;
82
- score?: number;
83
- reward?: number;
84
- costUsd?: number;
85
- observedAt?: string;
86
- metadata?: Record<string, unknown>;
87
- }
88
- export interface BeliefDecisionPoint {
89
- id: string;
90
- runId: string;
91
- scenarioId?: string;
92
- stepIndex: number;
93
- kind: BeliefDecisionKind;
94
- chosenAction: string;
95
- candidateActions?: string[];
96
- confidence?: number;
97
- behaviorProb?: number;
98
- targetProb?: number;
99
- qHat?: number | null;
100
- costUsd?: number;
101
- evidence: BeliefEvidenceRef[];
102
- outcome?: BeliefDecisionOutcome;
103
- reasons?: BeliefDecisionReason[];
104
- metadata?: Record<string, unknown>;
105
- }
106
- export interface BeliefDecisionExtractionDiagnostic {
107
- runId: string;
108
- eventId?: string;
109
- severity: 'info' | 'warning' | 'error';
110
- reason: string;
111
- }
112
- export interface BeliefDecisionExtractionReport {
113
- decisions: BeliefDecisionPoint[];
114
- diagnostics: BeliefDecisionExtractionDiagnostic[];
115
- }
116
- export type BeliefPolicyAction = 'accept' | 'defer' | 'verify' | 'ask' | 'retry' | 'stop';
117
- export interface BeliefPolicyDecision {
118
- action: BeliefPolicyAction;
119
- confidence?: number;
120
- targetProb?: number;
121
- qHat?: number | null;
122
- reason?: string;
123
- reasons?: BeliefDecisionReason[];
124
- }
125
- export interface BeliefSelectivePolicy {
126
- id: string;
127
- decide(point: BeliefDecisionPoint): BeliefPolicyDecision;
128
- }
129
- export interface BeliefOpeTargetPolicy {
130
- id: string;
131
- targetProbOf(point: BeliefDecisionPoint): number | null | undefined;
132
- qHatOf?(point: BeliefDecisionPoint): number | null | undefined;
133
- }
134
- export interface BeliefUtilityOptions {
135
- successUtility?: number;
136
- failureUtility?: number;
137
- deferUtility?: number;
138
- verifyCost?: number;
139
- askCost?: number;
140
- retryCost?: number;
141
- stopUtility?: number;
142
- costWeight?: number;
143
- }
144
- export interface BeliefSelectivePolicyMetrics {
145
- policyId: string;
146
- n: number;
147
- accepted: number;
148
- rejected: number;
149
- coverage: number;
150
- acceptedErrorRate: number;
151
- baselineUtility: number;
152
- policyUtility: number;
153
- utilityDelta: number;
154
- utilityCi95: {
155
- mean: number;
156
- lower: number;
157
- upper: number;
158
- };
159
- rejectedMeanReward: number | null;
160
- recommendation: 'ship' | 'hold' | 'need_more_data';
161
- reasons: string[];
162
- }
163
- export interface BeliefOpeSupportDiagnostics {
164
- supported: boolean;
165
- n: number;
166
- dropped: number;
167
- effectiveSampleSize: number;
168
- effectiveSampleRatio: number;
169
- maxImportanceWeight: number;
170
- reasons: string[];
171
- }
172
- export interface BeliefOpeReport {
173
- targetPolicyId: string;
174
- ips: OffPolicyEstimate;
175
- snips: OffPolicyEstimate;
176
- dr: OffPolicyEstimate;
177
- support: BeliefOpeSupportDiagnostics;
178
- }
179
- export type BeliefEvaluationStatus = 'ship' | 'hold' | 'need_more_data';
180
- export type BeliefCalibrationStatus = 'supported' | 'unsupported';
181
- export type BeliefOpeStatus = 'supported' | 'unsupported' | 'not_requested';
182
- export interface BeliefPolicyEvaluationReport {
183
- policyId: string;
184
- n: number;
185
- status: BeliefEvaluationStatus;
186
- selectiveStatus: BeliefEvaluationStatus;
187
- calibrationStatus: BeliefCalibrationStatus;
188
- opeStatus: BeliefOpeStatus;
189
- opeTargetPolicyId?: string;
190
- selective: BeliefSelectivePolicyMetrics;
191
- calibration?: CalibrationReport;
192
- ope?: BeliefOpeReport;
193
- diagnostics: string[];
194
- }
195
- //# sourceMappingURL=types.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/belief-state/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,0BAA0B,CAAA;AACjE,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAA;AAEzD,eAAO,MAAM,qBAAqB,6JAYxB,CAAA;AAEV,MAAM,MAAM,kBAAkB,GAAG,CAAC,OAAO,qBAAqB,CAAC,CAAC,MAAM,CAAC,CAAA;AAEvE,eAAO,MAAM,uBAAuB,+EAQ1B,CAAA;AAEV,MAAM,MAAM,oBAAoB,GAAG,CAAC,OAAO,uBAAuB,CAAC,CAAC,MAAM,CAAC,CAAA;AAE3E,eAAO,MAAM,yBAAyB,wFAO5B,CAAA;AAEV,MAAM,MAAM,qBAAqB,GAAG,CAAC,OAAO,yBAAyB,CAAC,CAAC,MAAM,CAAC,CAAA;AAE9E,eAAO,MAAM,0BAA0B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAuF7B,CAAA;AAEV,MAAM,MAAM,2BAA2B,GAAG,CAAC,OAAO,0BAA0B,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAA;AAC3F,MAAM,MAAM,wBAAwB,GAClC,CAAC,OAAO,0BAA0B,CAAC,CAAC,MAAM,CAAC,CAAC,aAAa,CAAC,CAAC,MAAM,CAAC,CAAA;AAEpE,MAAM,WAAW,oBAAoB;IACnC,IAAI,EAAE,wBAAwB,CAAA;IAC9B,SAAS,CAAC,EAAE,2BAA2B,CAAA;IACvC,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,WAAW,CAAC,EAAE,MAAM,EAAE,CAAA;IACtB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,wBAAgB,oBAAoB,CAAC,KAAK,EAAE,OAAO,GAAG,KAAK,IAAI,kBAAkB,CAEhF;AAED,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,OAAO,GAAG,KAAK,IAAI,oBAAoB,CAIpF;AAED,MAAM,WAAW,iBAAiB;IAChC,MAAM,EAAE,oBAAoB,CAAA;IAC5B,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,OAAO,CAAC,EAAE,qBAAqB,CAAA;IAC/B,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,qBAAqB;IACpC,OAAO,CAAC,EAAE,OAAO,CAAA;IACjB,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,mBAAmB;IAClC,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,MAAM,CAAA;IACjB,IAAI,EAAE,kBAAkB,CAAA;IACxB,YAAY,EAAE,MAAM,CAAA;IACpB,gBAAgB,CAAC,EAAE,MAAM,EAAE,CAAA;IAC3B,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,IAAI,CAAC,EAAE,MAAM,GAAG,IAAI,CAAA;IACpB,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,QAAQ,EAAE,iBAAiB,EAAE,CAAA;IAC7B,OAAO,CAAC,EAAE,qBAAqB,CAAA;IAC/B,OAAO,CAAC,EAAE,oBAAoB,EAAE,CAAA;IAChC,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,kCAAkC;IACjD,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,QAAQ,EAAE,MAAM,GAAG,SAAS,GAAG,OAAO,CAAA;IACtC,MAAM,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,8BAA8B;IAC7C,SAAS,EAAE,mBAAmB,EAAE,CAAA;IAChC,WAAW,EAAE,kCAAkC,EAAE,CAAA;CAClD;AAED,MAAM,MAAM,kBAAkB,GAAG,QAAQ,GAAG,OAAO,GAAG,QAAQ,GAAG,KAAK,GAAG,OAAO,GAAG,MAAM,CAAA;AAEzF,MAAM,WAAW,oBAAoB;IACnC,MAAM,EAAE,kBAAkB,CAAA;IAC1B,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,IAAI,CAAC,EAAE,MAAM,GAAG,IAAI,CAAA;IACpB,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,OAAO,CAAC,EAAE,oBAAoB,EAAE,CAAA;CACjC;AAED,MAAM,WAAW,qBAAqB;IACpC,EAAE,EAAE,MAAM,CAAA;IACV,MAAM,CAAC,KAAK,EAAE,mBAAmB,GAAG,oBAAoB,CAAA;CACzD;AAED,MAAM,WAAW,qBAAqB;IACpC,EAAE,EAAE,MAAM,CAAA;IACV,YAAY,CAAC,KAAK,EAAE,mBAAmB,GAAG,MAAM,GAAG,IAAI,GAAG,SAAS,CAAA;IACnE,MAAM,CAAC,CAAC,KAAK,EAAE,mBAAmB,GAAG,MAAM,GAAG,IAAI,GAAG,SAAS,CAAA;CAC/D;AAED,MAAM,WAAW,oBAAoB;IACnC,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,UAAU,CAAC,EAAE,MAAM,CAAA;CACpB;AAED,MAAM,WAAW,4BAA4B;IAC3C,QAAQ,EAAE,MAAM,CAAA;IAChB,CAAC,EAAE,MAAM,CAAA;IACT,QAAQ,EAAE,MAAM,CAAA;IAChB,QAAQ,EAAE,MAAM,CAAA;IAChB,QAAQ,EAAE,MAAM,CAAA;IAChB,iBAAiB,EAAE,MAAM,CAAA;IACzB,eAAe,EAAE,MAAM,CAAA;IACvB,aAAa,EAAE,MAAM,CAAA;IACrB,YAAY,EAAE,MAAM,CAAA;IACpB,WAAW,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAA;IAC3D,kBAAkB,EAAE,MAAM,GAAG,IAAI,CAAA;IACjC,cAAc,EAAE,MAAM,GAAG,MAAM,GAAG,gBAAgB,CAAA;IAClD,OAAO,EAAE,MAAM,EAAE,CAAA;CAClB;AAED,MAAM,WAAW,2BAA2B;IAC1C,SAAS,EAAE,OAAO,CAAA;IAClB,CAAC,EAAE,MAAM,CAAA;IACT,OAAO,EAAE,MAAM,CAAA;IACf,mBAAmB,EAAE,MAAM,CAAA;IAC3B,oBAAoB,EAAE,MAAM,CAAA;IAC5B,mBAAmB,EAAE,MAAM,CAAA;IAC3B,OAAO,EAAE,MAAM,EAAE,CAAA;CAClB;AAED,MAAM,WAAW,eAAe;IAC9B,cAAc,EAAE,MAAM,CAAA;IACtB,GAAG,EAAE,iBAAiB,CAAA;IACtB,KAAK,EAAE,iBAAiB,CAAA;IACxB,EAAE,EAAE,iBAAiB,CAAA;IACrB,OAAO,EAAE,2BAA2B,CAAA;CACrC;AAED,MAAM,MAAM,sBAAsB,GAAG,MAAM,GAAG,MAAM,GAAG,gBAAgB,CAAA;AACvE,MAAM,MAAM,uBAAuB,GAAG,WAAW,GAAG,aAAa,CAAA;AACjE,MAAM,MAAM,eAAe,GAAG,WAAW,GAAG,aAAa,GAAG,eAAe,CAAA;AAE3E,MAAM,WAAW,4BAA4B;IAC3C,QAAQ,EAAE,MAAM,CAAA;IAChB,CAAC,EAAE,MAAM,CAAA;IACT,MAAM,EAAE,sBAAsB,CAAA;IAC9B,eAAe,EAAE,sBAAsB,CAAA;IACvC,iBAAiB,EAAE,uBAAuB,CAAA;IAC1C,SAAS,EAAE,eAAe,CAAA;IAC1B,iBAAiB,CAAC,EAAE,MAAM,CAAA;IAC1B,SAAS,EAAE,4BAA4B,CAAA;IACvC,WAAW,CAAC,EAAE,iBAAiB,CAAA;IAC/B,GAAG,CAAC,EAAE,eAAe,CAAA;IACrB,WAAW,EAAE,MAAM,EAAE,CAAA;CACtB"}
@@ -1,2 +0,0 @@
1
- export {};
2
- //# sourceMappingURL=types.test.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"types.test.d.ts","sourceRoot":"","sources":["../../src/belief-state/types.test.ts"],"names":[],"mappings":""}
@@ -1,14 +0,0 @@
1
- import type { TCloud } from '@tangle-network/tcloud';
2
- import type { BenchmarkReport, BenchmarkRunnerConfig, Scenario } from './types';
3
- /**
4
- * BenchmarkRunner — orchestrates scenarios, executor, judges, and scoring.
5
- *
6
- * Domain-agnostic. Each agent provides its own scenarios, judges, and system prompt.
7
- */
8
- export declare class BenchmarkRunner {
9
- private tc;
10
- private config;
11
- constructor(tc: TCloud, config: BenchmarkRunnerConfig);
12
- run(scenarios?: Scenario[]): Promise<BenchmarkReport>;
13
- }
14
- //# sourceMappingURL=benchmark.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"benchmark.d.ts","sourceRoot":"","sources":["../src/benchmark.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,wBAAwB,CAAA;AAEpD,OAAO,KAAK,EAAE,eAAe,EAAE,qBAAqB,EAAE,QAAQ,EAAkB,MAAM,SAAS,CAAA;AAE/F;;;;GAIG;AACH,qBAAa,eAAe;IAC1B,OAAO,CAAC,EAAE,CAAQ;IAClB,OAAO,CAAC,MAAM,CAAuB;gBAEzB,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,qBAAqB;IAK/C,GAAG,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,GAAG,OAAO,CAAC,eAAe,CAAC;CAgK5D"}
@@ -1 +0,0 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/benchmarks/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,KAAK,OAAO,MAAM,iBAAiB,CAAA;AAC1C,YAAY,EACV,gBAAgB,EAChB,oBAAoB,EACpB,mBAAmB,GACpB,MAAM,SAAS,CAAA;AAChB,OAAO,EAAE,oBAAoB,EAAE,kBAAkB,EAAE,MAAM,SAAS,CAAA"}
@@ -1,34 +0,0 @@
1
- /**
2
- * Synthetic routing dataset. 16 tasks across 4 categories. Used as a
3
- * deterministic, dependency-free benchmark for any router that maps a
4
- * natural-language request to one of a fixed set of route labels.
5
- *
6
- * Format (see `routing/README.md` for prose):
7
- *
8
- * {
9
- * id: stable per-task ID (matches across processes).
10
- * category: one of the four route labels.
11
- * prompt: the user-facing request the router must classify.
12
- * route: the ground-truth route the router should pick.
13
- * synonyms: other strings that count as a correct answer.
14
- * hardNegatives:close-but-wrong route labels — used to detect the
15
- * "always picks the popular route" failure mode.
16
- * }
17
- *
18
- * The four categories are intentionally cross-domain (file ops,
19
- * math, search, conversation) so a router that collapses to one
20
- * category is easy to spot.
21
- */
22
- export interface RoutingItem {
23
- id: string;
24
- category: 'file' | 'math' | 'search' | 'chat';
25
- prompt: string;
26
- /** Canonical correct route label. */
27
- route: string;
28
- /** Alternate route labels that also count as correct. */
29
- synonyms: string[];
30
- /** Wrong-but-tempting route labels (for analysis, not grading). */
31
- hardNegatives: string[];
32
- }
33
- export declare const ROUTING_DATASET: RoutingItem[];
34
- //# sourceMappingURL=dataset.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"dataset.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/routing/dataset.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAEH,MAAM,WAAW,WAAW;IAC1B,EAAE,EAAE,MAAM,CAAA;IACV,QAAQ,EAAE,MAAM,GAAG,MAAM,GAAG,QAAQ,GAAG,MAAM,CAAA;IAC7C,MAAM,EAAE,MAAM,CAAA;IACd,qCAAqC;IACrC,KAAK,EAAE,MAAM,CAAA;IACb,yDAAyD;IACzD,QAAQ,EAAE,MAAM,EAAE,CAAA;IAClB,mEAAmE;IACnE,aAAa,EAAE,MAAM,EAAE,CAAA;CACxB;AAED,eAAO,MAAM,eAAe,EAAE,WAAW,EAiIxC,CAAA"}