@tangle-network/agent-eval 0.93.0 → 0.95.0

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 (1087) hide show
  1. package/CHANGELOG.md +15 -0
  2. package/README.md +44 -30
  3. package/dist/action-policy.d.ts +24 -0
  4. package/dist/action-policy.d.ts.map +1 -0
  5. package/dist/action-policy.test.d.ts +2 -0
  6. package/dist/action-policy.test.d.ts.map +1 -0
  7. package/dist/active-learning.d.ts +41 -0
  8. package/dist/active-learning.d.ts.map +1 -0
  9. package/dist/adapters/http.d.ts +15 -21
  10. package/dist/adapters/http.d.ts.map +1 -0
  11. package/dist/adapters/http.js.map +1 -1
  12. package/dist/adapters/langchain.d.ts +7 -13
  13. package/dist/adapters/langchain.d.ts.map +1 -0
  14. package/dist/adapters/otel.d.ts +13 -24
  15. package/dist/adapters/otel.d.ts.map +1 -0
  16. package/dist/agent-profile-cell.d.ts +101 -0
  17. package/dist/agent-profile-cell.d.ts.map +1 -0
  18. package/dist/agent-profile.d.ts +27 -0
  19. package/dist/agent-profile.d.ts.map +1 -0
  20. package/dist/agent-profile.test.d.ts +2 -0
  21. package/dist/agent-profile.test.d.ts.map +1 -0
  22. package/dist/analyst/adapters.d.ts +62 -0
  23. package/dist/analyst/adapters.d.ts.map +1 -0
  24. package/dist/analyst/analyst.test.d.ts +2 -0
  25. package/dist/analyst/analyst.test.d.ts.map +1 -0
  26. package/dist/analyst/ax-service.d.ts +27 -0
  27. package/dist/analyst/ax-service.d.ts.map +1 -0
  28. package/dist/analyst/behavioral-analyst.d.ts +28 -0
  29. package/dist/analyst/behavioral-analyst.d.ts.map +1 -0
  30. package/dist/analyst/chat-client.d.ts +91 -0
  31. package/dist/analyst/chat-client.d.ts.map +1 -0
  32. package/dist/analyst/default-registry.d.ts +27 -0
  33. package/dist/analyst/default-registry.d.ts.map +1 -0
  34. package/dist/analyst/default-registry.test.d.ts +2 -0
  35. package/dist/analyst/default-registry.test.d.ts.map +1 -0
  36. package/dist/analyst/finding-signature.d.ts +48 -0
  37. package/dist/analyst/finding-signature.d.ts.map +1 -0
  38. package/dist/analyst/finding-subject.d.ts +146 -0
  39. package/dist/analyst/finding-subject.d.ts.map +1 -0
  40. package/dist/analyst/finding-subject.test.d.ts +2 -0
  41. package/dist/analyst/finding-subject.test.d.ts.map +1 -0
  42. package/dist/analyst/findings-store.d.ts +75 -0
  43. package/dist/analyst/findings-store.d.ts.map +1 -0
  44. package/dist/analyst/index.d.ts +28 -256
  45. package/dist/analyst/index.d.ts.map +1 -0
  46. package/dist/analyst/index.js +6 -66
  47. package/dist/analyst/index.js.map +1 -1
  48. package/dist/{kind-factory-5b7xXXOr.d.ts → analyst/kind-factory.d.ts} +11 -63
  49. package/dist/analyst/kind-factory.d.ts.map +1 -0
  50. package/dist/analyst/kinds/failure-mode.d.ts +19 -0
  51. package/dist/analyst/kinds/failure-mode.d.ts.map +1 -0
  52. package/dist/analyst/kinds/improvement.d.ts +23 -0
  53. package/dist/analyst/kinds/improvement.d.ts.map +1 -0
  54. package/dist/analyst/kinds/index.d.ts +22 -0
  55. package/dist/analyst/kinds/index.d.ts.map +1 -0
  56. package/dist/analyst/kinds/kinds.test.d.ts +2 -0
  57. package/dist/analyst/kinds/kinds.test.d.ts.map +1 -0
  58. package/dist/analyst/kinds/knowledge-gap.d.ts +28 -0
  59. package/dist/analyst/kinds/knowledge-gap.d.ts.map +1 -0
  60. package/dist/analyst/kinds/knowledge-poisoning.d.ts +22 -0
  61. package/dist/analyst/kinds/knowledge-poisoning.d.ts.map +1 -0
  62. package/dist/analyst/kinds/skill-usage.d.ts +84 -0
  63. package/dist/analyst/kinds/skill-usage.d.ts.map +1 -0
  64. package/dist/analyst/kinds/skill-usage.test.d.ts +2 -0
  65. package/dist/analyst/kinds/skill-usage.test.d.ts.map +1 -0
  66. package/dist/analyst/parse-tolerant.d.ts +26 -0
  67. package/dist/analyst/parse-tolerant.d.ts.map +1 -0
  68. package/dist/analyst/parse-tolerant.test.d.ts +2 -0
  69. package/dist/analyst/parse-tolerant.test.d.ts.map +1 -0
  70. package/dist/analyst/registry.budget.test.d.ts +2 -0
  71. package/dist/analyst/registry.budget.test.d.ts.map +1 -0
  72. package/dist/{default-registry-zoGHUQEH.d.ts → analyst/registry.d.ts} +8 -37
  73. package/dist/analyst/registry.d.ts.map +1 -0
  74. package/dist/analyst/steer-firewall.d.ts +35 -0
  75. package/dist/analyst/steer-firewall.d.ts.map +1 -0
  76. package/dist/analyst/steer-firewall.test.d.ts +2 -0
  77. package/dist/analyst/steer-firewall.test.d.ts.map +1 -0
  78. package/dist/analyst/structure-findings.d.ts +37 -0
  79. package/dist/analyst/structure-findings.d.ts.map +1 -0
  80. package/dist/analyst/structure-findings.test.d.ts +2 -0
  81. package/dist/analyst/structure-findings.test.d.ts.map +1 -0
  82. package/dist/analyst/tool-groups.d.ts +34 -0
  83. package/dist/analyst/tool-groups.d.ts.map +1 -0
  84. package/dist/{types-2VVIL04s.d.ts → analyst/types.d.ts} +19 -112
  85. package/dist/analyst/types.d.ts.map +1 -0
  86. package/dist/anti-slop.d.ts +59 -0
  87. package/dist/anti-slop.d.ts.map +1 -0
  88. package/dist/artifact-validator.d.ts +74 -0
  89. package/dist/artifact-validator.d.ts.map +1 -0
  90. package/dist/attestation.d.ts +63 -0
  91. package/dist/attestation.d.ts.map +1 -0
  92. package/dist/attestation.test.d.ts +2 -0
  93. package/dist/attestation.test.d.ts.map +1 -0
  94. package/dist/authenticity/index.d.ts +15 -16
  95. package/dist/authenticity/index.d.ts.map +1 -0
  96. package/dist/authenticity/index.test.d.ts +2 -0
  97. package/dist/authenticity/index.test.d.ts.map +1 -0
  98. package/dist/auto-pr.d.ts +120 -0
  99. package/dist/auto-pr.d.ts.map +1 -0
  100. package/dist/{baseline-DE36-Np7.d.ts → baseline.d.ts} +8 -44
  101. package/dist/baseline.d.ts.map +1 -0
  102. package/dist/behavior-dsl.d.ts +73 -0
  103. package/dist/behavior-dsl.d.ts.map +1 -0
  104. package/dist/belief-state/calibration.d.ts +10 -0
  105. package/dist/belief-state/calibration.d.ts.map +1 -0
  106. package/dist/belief-state/calibration.test.d.ts +2 -0
  107. package/dist/belief-state/calibration.test.d.ts.map +1 -0
  108. package/dist/belief-state/code-agent-corpus.d.ts +66 -0
  109. package/dist/belief-state/code-agent-corpus.d.ts.map +1 -0
  110. package/dist/belief-state/code-agent-corpus.test.d.ts +2 -0
  111. package/dist/belief-state/code-agent-corpus.test.d.ts.map +1 -0
  112. package/dist/belief-state/code-agent-evidence.d.ts +22 -0
  113. package/dist/belief-state/code-agent-evidence.d.ts.map +1 -0
  114. package/dist/belief-state/code-agent-evidence.test.d.ts +2 -0
  115. package/dist/belief-state/code-agent-evidence.test.d.ts.map +1 -0
  116. package/dist/belief-state/extract.d.ts +7 -0
  117. package/dist/belief-state/extract.d.ts.map +1 -0
  118. package/dist/belief-state/extract.test.d.ts +2 -0
  119. package/dist/belief-state/extract.test.d.ts.map +1 -0
  120. package/dist/belief-state/index.d.ts +14 -604
  121. package/dist/belief-state/index.d.ts.map +1 -0
  122. package/dist/belief-state/index.js +1 -1
  123. package/dist/belief-state/ope.d.ts +17 -0
  124. package/dist/belief-state/ope.d.ts.map +1 -0
  125. package/dist/belief-state/ope.test.d.ts +2 -0
  126. package/dist/belief-state/ope.test.d.ts.map +1 -0
  127. package/dist/belief-state/phase0-measurement.d.ts +55 -0
  128. package/dist/belief-state/phase0-measurement.d.ts.map +1 -0
  129. package/dist/belief-state/report.d.ts +17 -0
  130. package/dist/belief-state/report.d.ts.map +1 -0
  131. package/dist/belief-state/report.test.d.ts +2 -0
  132. package/dist/belief-state/report.test.d.ts.map +1 -0
  133. package/dist/belief-state/research-evidence.d.ts +23 -0
  134. package/dist/belief-state/research-evidence.d.ts.map +1 -0
  135. package/dist/belief-state/research-evidence.test.d.ts +2 -0
  136. package/dist/belief-state/research-evidence.test.d.ts.map +1 -0
  137. package/dist/belief-state/runtime-benchmark-corpus.d.ts +32 -0
  138. package/dist/belief-state/runtime-benchmark-corpus.d.ts.map +1 -0
  139. package/dist/belief-state/runtime-hooks.d.ts +87 -0
  140. package/dist/belief-state/runtime-hooks.d.ts.map +1 -0
  141. package/dist/belief-state/runtime-hooks.test.d.ts +2 -0
  142. package/dist/belief-state/runtime-hooks.test.d.ts.map +1 -0
  143. package/dist/belief-state/selective.d.ts +15 -0
  144. package/dist/belief-state/selective.d.ts.map +1 -0
  145. package/dist/belief-state/selective.test.d.ts +2 -0
  146. package/dist/belief-state/selective.test.d.ts.map +1 -0
  147. package/dist/belief-state/shadow-probe.d.ts +80 -0
  148. package/dist/belief-state/shadow-probe.d.ts.map +1 -0
  149. package/dist/belief-state/shadow-probe.test.d.ts +2 -0
  150. package/dist/belief-state/shadow-probe.test.d.ts.map +1 -0
  151. package/dist/belief-state/types.d.ts +195 -0
  152. package/dist/belief-state/types.d.ts.map +1 -0
  153. package/dist/belief-state/types.test.d.ts +2 -0
  154. package/dist/belief-state/types.test.d.ts.map +1 -0
  155. package/dist/benchmark.d.ts +14 -0
  156. package/dist/benchmark.d.ts.map +1 -0
  157. package/dist/benchmarks/index.d.ts +23 -4
  158. package/dist/benchmarks/index.d.ts.map +1 -0
  159. package/dist/benchmarks/routing/dataset.d.ts +34 -0
  160. package/dist/benchmarks/routing/dataset.d.ts.map +1 -0
  161. package/dist/benchmarks/routing/index.d.ts +34 -0
  162. package/dist/benchmarks/routing/index.d.ts.map +1 -0
  163. package/dist/benchmarks/types.d.ts +49 -0
  164. package/dist/benchmarks/types.d.ts.map +1 -0
  165. package/dist/bisector.d.ts +81 -0
  166. package/dist/bisector.d.ts.map +1 -0
  167. package/dist/budget-guard.d.ts +31 -0
  168. package/dist/budget-guard.d.ts.map +1 -0
  169. package/dist/builder-eval/builder-session.d.ts +111 -0
  170. package/dist/builder-eval/builder-session.d.ts.map +1 -0
  171. package/dist/builder-eval/correlation.d.ts +32 -0
  172. package/dist/builder-eval/correlation.d.ts.map +1 -0
  173. package/dist/builder-eval/index.d.ts +5 -250
  174. package/dist/builder-eval/index.d.ts.map +1 -0
  175. package/dist/builder-eval/index.js +2 -2
  176. package/dist/builder-eval/project-registry.d.ts +51 -0
  177. package/dist/builder-eval/project-registry.d.ts.map +1 -0
  178. package/dist/builder-eval/three-layer-eval.d.ts +55 -0
  179. package/dist/builder-eval/three-layer-eval.d.ts.map +1 -0
  180. package/dist/campaign/analyst-surface.d.ts +108 -0
  181. package/dist/campaign/analyst-surface.d.ts.map +1 -0
  182. package/dist/campaign/analyst-surface.test.d.ts +2 -0
  183. package/dist/campaign/analyst-surface.test.d.ts.map +1 -0
  184. package/dist/campaign/auto-pr.d.ts +46 -0
  185. package/dist/campaign/auto-pr.d.ts.map +1 -0
  186. package/dist/campaign/distillation/agreement-judge.d.ts +69 -0
  187. package/dist/campaign/distillation/agreement-judge.d.ts.map +1 -0
  188. package/dist/campaign/distillation/cli.d.ts +35 -0
  189. package/dist/campaign/distillation/cli.d.ts.map +1 -0
  190. package/dist/campaign/distillation/distillation.test.d.ts +2 -0
  191. package/dist/campaign/distillation/distillation.test.d.ts.map +1 -0
  192. package/dist/campaign/distillation/gold-scenarios.d.ts +54 -0
  193. package/dist/campaign/distillation/gold-scenarios.d.ts.map +1 -0
  194. package/dist/campaign/distillation/run-distillation.d.ts +119 -0
  195. package/dist/campaign/distillation/run-distillation.d.ts.map +1 -0
  196. package/dist/campaign/gates/compose.d.ts +12 -0
  197. package/dist/campaign/gates/compose.d.ts.map +1 -0
  198. package/dist/campaign/gates/default-production-gate.d.ts +58 -0
  199. package/dist/campaign/gates/default-production-gate.d.ts.map +1 -0
  200. package/dist/campaign/gates/heldout-gate.d.ts +12 -0
  201. package/dist/campaign/gates/heldout-gate.d.ts.map +1 -0
  202. package/dist/campaign/gates/promotion-policy.d.ts +125 -0
  203. package/dist/campaign/gates/promotion-policy.d.ts.map +1 -0
  204. package/dist/campaign/gates/promotion-policy.test.d.ts +2 -0
  205. package/dist/campaign/gates/promotion-policy.test.d.ts.map +1 -0
  206. package/dist/campaign/gates/sequential.d.ts +146 -0
  207. package/dist/campaign/gates/sequential.d.ts.map +1 -0
  208. package/dist/campaign/gates/sequential.test.d.ts +2 -0
  209. package/dist/campaign/gates/sequential.test.d.ts.map +1 -0
  210. package/dist/campaign/gates/statistical-heldout.d.ts +99 -0
  211. package/dist/campaign/gates/statistical-heldout.d.ts.map +1 -0
  212. package/dist/campaign/gates/statistical-heldout.test.d.ts +2 -0
  213. package/dist/campaign/gates/statistical-heldout.test.d.ts.map +1 -0
  214. package/dist/campaign/index.d.ts +38 -1341
  215. package/dist/campaign/index.d.ts.map +1 -0
  216. package/dist/campaign/index.js +1865 -1318
  217. package/dist/campaign/index.js.map +1 -1
  218. package/dist/campaign/labeled-store/fs-adapter.d.ts +59 -0
  219. package/dist/campaign/labeled-store/fs-adapter.d.ts.map +1 -0
  220. package/dist/campaign/presets/compare-proposers.d.ts +146 -0
  221. package/dist/campaign/presets/compare-proposers.d.ts.map +1 -0
  222. package/dist/campaign/presets/playback.d.ts +120 -0
  223. package/dist/campaign/presets/playback.d.ts.map +1 -0
  224. package/dist/campaign/presets/playback.test.d.ts +2 -0
  225. package/dist/campaign/presets/playback.test.d.ts.map +1 -0
  226. package/dist/campaign/presets/run-eval.d.ts +14 -0
  227. package/dist/campaign/presets/run-eval.d.ts.map +1 -0
  228. package/dist/campaign/presets/run-improvement-loop.d.ts +63 -0
  229. package/dist/campaign/presets/run-improvement-loop.d.ts.map +1 -0
  230. package/dist/campaign/presets/run-improvement-loop.test.d.ts +2 -0
  231. package/dist/campaign/presets/run-improvement-loop.test.d.ts.map +1 -0
  232. package/dist/campaign/presets/run-optimization.d.ts +92 -0
  233. package/dist/campaign/presets/run-optimization.d.ts.map +1 -0
  234. package/dist/campaign/presets/run-profile-matrix.d.ts +151 -0
  235. package/dist/campaign/presets/run-profile-matrix.d.ts.map +1 -0
  236. package/dist/campaign/presets/run-skill-opt.d.ts +96 -0
  237. package/dist/campaign/presets/run-skill-opt.d.ts.map +1 -0
  238. package/dist/campaign/proposers/_findings-text.d.ts +22 -0
  239. package/dist/campaign/proposers/_findings-text.d.ts.map +1 -0
  240. package/dist/campaign/proposers/ace.d.ts +33 -0
  241. package/dist/campaign/proposers/ace.d.ts.map +1 -0
  242. package/dist/campaign/proposers/ace.test.d.ts +2 -0
  243. package/dist/campaign/proposers/ace.test.d.ts.map +1 -0
  244. package/dist/campaign/proposers/analysis-edit.d.ts +32 -0
  245. package/dist/campaign/proposers/analysis-edit.d.ts.map +1 -0
  246. package/dist/campaign/proposers/evolutionary.d.ts +20 -0
  247. package/dist/campaign/proposers/evolutionary.d.ts.map +1 -0
  248. package/dist/campaign/proposers/fapo.d.ts +120 -0
  249. package/dist/campaign/proposers/fapo.d.ts.map +1 -0
  250. package/dist/campaign/proposers/gepa.d.ts +86 -0
  251. package/dist/campaign/proposers/gepa.d.ts.map +1 -0
  252. package/dist/campaign/proposers/halo.d.ts +44 -0
  253. package/dist/campaign/proposers/halo.d.ts.map +1 -0
  254. package/dist/campaign/proposers/halo.test.d.ts +2 -0
  255. package/dist/campaign/proposers/halo.test.d.ts.map +1 -0
  256. package/dist/campaign/proposers/memory.d.ts +47 -0
  257. package/dist/campaign/proposers/memory.d.ts.map +1 -0
  258. package/dist/campaign/proposers/memory.test.d.ts +2 -0
  259. package/dist/campaign/proposers/memory.test.d.ts.map +1 -0
  260. package/dist/campaign/proposers/skill-opt.d.ts +88 -0
  261. package/dist/campaign/proposers/skill-opt.d.ts.map +1 -0
  262. package/dist/campaign/proposers/trace-analyst.d.ts +48 -0
  263. package/dist/campaign/proposers/trace-analyst.d.ts.map +1 -0
  264. package/dist/campaign/proposers/trace-analyst.test.d.ts +2 -0
  265. package/dist/campaign/proposers/trace-analyst.test.d.ts.map +1 -0
  266. package/dist/campaign/provenance.d.ts +185 -0
  267. package/dist/campaign/provenance.d.ts.map +1 -0
  268. package/dist/campaign/run-campaign.d.ts +90 -0
  269. package/dist/campaign/run-campaign.d.ts.map +1 -0
  270. package/dist/campaign/score-utils.d.ts +26 -0
  271. package/dist/campaign/score-utils.d.ts.map +1 -0
  272. package/dist/campaign/skill-patch.d.ts +62 -0
  273. package/dist/campaign/skill-patch.d.ts.map +1 -0
  274. package/dist/campaign/storage.d.ts +38 -0
  275. package/dist/campaign/storage.d.ts.map +1 -0
  276. package/dist/{types-BU-7W85F.d.ts → campaign/types.d.ts} +98 -99
  277. package/dist/campaign/types.d.ts.map +1 -0
  278. package/dist/campaign/worktree/index.d.ts +53 -0
  279. package/dist/campaign/worktree/index.d.ts.map +1 -0
  280. package/dist/canary.d.ts +101 -0
  281. package/dist/canary.d.ts.map +1 -0
  282. package/dist/causal-attribution.d.ts +45 -0
  283. package/dist/causal-attribution.d.ts.map +1 -0
  284. package/dist/{chunk-TWS7AZEY.js → chunk-2T4EZACH.js} +2 -2
  285. package/dist/chunk-2T4EZACH.js.map +1 -0
  286. package/dist/{chunk-LMZQ2Z4U.js → chunk-56GC6VYO.js} +126 -1
  287. package/dist/chunk-56GC6VYO.js.map +1 -0
  288. package/dist/{chunk-QG2OVF2D.js → chunk-77T4STFI.js} +154 -112
  289. package/dist/chunk-77T4STFI.js.map +1 -0
  290. package/dist/{chunk-ZFIBGEOL.js → chunk-7QTQKIDD.js} +179 -178
  291. package/dist/chunk-7QTQKIDD.js.map +1 -0
  292. package/dist/{chunk-UMMZHCPB.js → chunk-AQ5WQAIV.js} +26 -9
  293. package/dist/chunk-AQ5WQAIV.js.map +1 -0
  294. package/dist/{chunk-CVVHBFGN.js → chunk-CWNP4DV4.js} +53 -5
  295. package/dist/chunk-CWNP4DV4.js.map +1 -0
  296. package/dist/{chunk-QS3RBQPI.js → chunk-D5KBVULY.js} +2 -2
  297. package/dist/chunk-DJWX3GVS.js +81 -0
  298. package/dist/chunk-DJWX3GVS.js.map +1 -0
  299. package/dist/{chunk-S6OZEZQK.js → chunk-HMA63UEO.js} +37 -9
  300. package/dist/{chunk-S6OZEZQK.js.map → chunk-HMA63UEO.js.map} +1 -1
  301. package/dist/{chunk-UHMJT4T7.js → chunk-IZCEK2HR.js} +2 -2
  302. package/dist/{chunk-6SOJM3VR.js → chunk-KKWJD5E6.js} +21 -21
  303. package/dist/chunk-KKWJD5E6.js.map +1 -0
  304. package/dist/{chunk-KWRRMR3J.js → chunk-LO6IOIJ2.js} +10 -10
  305. package/dist/chunk-LO6IOIJ2.js.map +1 -0
  306. package/dist/{chunk-L3JOU6XM.js → chunk-NACM5RS7.js} +3 -3
  307. package/dist/{chunk-E4GH6USR.js → chunk-NZEQVRH5.js} +2 -2
  308. package/dist/chunk-NZEQVRH5.js.map +1 -0
  309. package/dist/{chunk-XY4DDNEG.js → chunk-PSWWQXHF.js} +14 -89
  310. package/dist/chunk-PSWWQXHF.js.map +1 -0
  311. package/dist/{chunk-4FBZZIYD.js → chunk-QTMYW64F.js} +2 -2
  312. package/dist/{chunk-D3V5B42D.js → chunk-S4SYLDFX.js} +7 -4
  313. package/dist/chunk-S4SYLDFX.js.map +1 -0
  314. package/dist/{chunk-47X6LRCE.js → chunk-UFSG7ACU.js} +10 -7
  315. package/dist/chunk-UFSG7ACU.js.map +1 -0
  316. package/dist/{chunk-CY6U5S3X.js → chunk-URWVKRCS.js} +2 -2
  317. package/dist/{chunk-GSH6QNNS.js → chunk-X74V6ESX.js} +102 -79
  318. package/dist/chunk-X74V6ESX.js.map +1 -0
  319. package/dist/{chunk-QAY5UIJO.js → chunk-YBIGNSCZ.js} +178 -59
  320. package/dist/chunk-YBIGNSCZ.js.map +1 -0
  321. package/dist/{chunk-Y47J2LJ3.js → chunk-Z6L6YSU6.js} +5 -5
  322. package/dist/{chunk-T375SUOZ.js → chunk-ZOPLCJSY.js} +86 -31
  323. package/dist/chunk-ZOPLCJSY.js.map +1 -0
  324. package/dist/ci-gate.d.ts +44 -0
  325. package/dist/ci-gate.d.ts.map +1 -0
  326. package/dist/cli.d.ts +2 -0
  327. package/dist/cli.d.ts.map +1 -0
  328. package/dist/cli.js +2 -2
  329. package/dist/client.d.ts +77 -0
  330. package/dist/client.d.ts.map +1 -0
  331. package/dist/client.test.d.ts +2 -0
  332. package/dist/client.test.d.ts.map +1 -0
  333. package/dist/command-runner.d.ts +74 -0
  334. package/dist/command-runner.d.ts.map +1 -0
  335. package/dist/command-runner.test.d.ts +2 -0
  336. package/dist/command-runner.test.d.ts.map +1 -0
  337. package/dist/completion-verifier.d.ts +147 -0
  338. package/dist/completion-verifier.d.ts.map +1 -0
  339. package/dist/completion-verifier.test.d.ts +9 -0
  340. package/dist/completion-verifier.test.d.ts.map +1 -0
  341. package/dist/concurrency.d.ts +23 -0
  342. package/dist/concurrency.d.ts.map +1 -0
  343. package/dist/contamination-guard.d.ts +81 -0
  344. package/dist/contamination-guard.d.ts.map +1 -0
  345. package/dist/{analyze-runs-DwCEkpO_.d.ts → contract/analyze-runs.d.ts} +9 -10
  346. package/dist/contract/analyze-runs.d.ts.map +1 -0
  347. package/dist/contract/define-agent-eval.d.ts +52 -0
  348. package/dist/contract/define-agent-eval.d.ts.map +1 -0
  349. package/dist/contract/diff.d.ts +114 -0
  350. package/dist/contract/diff.d.ts.map +1 -0
  351. package/dist/contract/index.d.ts +106 -640
  352. package/dist/contract/index.d.ts.map +1 -0
  353. package/dist/contract/index.js +131 -21
  354. package/dist/contract/index.js.map +1 -1
  355. package/dist/{insight-report-BBwvOh6x.d.ts → contract/insight-report.d.ts} +16 -19
  356. package/dist/contract/insight-report.d.ts.map +1 -0
  357. package/dist/contract/insight-types-fwd.d.ts +7 -0
  358. package/dist/contract/insight-types-fwd.d.ts.map +1 -0
  359. package/dist/contract/intake/agent-trace.d.ts +97 -0
  360. package/dist/contract/intake/agent-trace.d.ts.map +1 -0
  361. package/dist/{code-agent-session-BO8nCnv3.d.ts → contract/intake/code-agent-session.d.ts} +15 -17
  362. package/dist/contract/intake/code-agent-session.d.ts.map +1 -0
  363. package/dist/contract/intake/feedback-table.d.ts +87 -0
  364. package/dist/contract/intake/feedback-table.d.ts.map +1 -0
  365. package/dist/contract/intake/index.d.ts +22 -0
  366. package/dist/contract/intake/index.d.ts.map +1 -0
  367. package/dist/contract/intake/otel-spans.d.ts +33 -0
  368. package/dist/contract/intake/otel-spans.d.ts.map +1 -0
  369. package/dist/contract/self-improve.d.ts +284 -0
  370. package/dist/contract/self-improve.d.ts.map +1 -0
  371. package/dist/{control-runtime-DuFBYg7A.d.ts → control-runtime.d.ts} +23 -26
  372. package/dist/control-runtime.d.ts.map +1 -0
  373. package/dist/control-runtime.test.d.ts +2 -0
  374. package/dist/control-runtime.test.d.ts.map +1 -0
  375. package/dist/control.d.ts +11 -9
  376. package/dist/control.d.ts.map +1 -0
  377. package/dist/control.js +2 -2
  378. package/dist/convergence.d.ts +29 -0
  379. package/dist/convergence.d.ts.map +1 -0
  380. package/dist/{cost-ledger-DuSqlw5B.d.ts → cost-ledger.d.ts} +10 -11
  381. package/dist/cost-ledger.d.ts.map +1 -0
  382. package/dist/cost-ledger.test.d.ts +2 -0
  383. package/dist/cost-ledger.test.d.ts.map +1 -0
  384. package/dist/cost-report.d.ts +43 -0
  385. package/dist/cost-report.d.ts.map +1 -0
  386. package/dist/cost-report.test.d.ts +2 -0
  387. package/dist/cost-report.test.d.ts.map +1 -0
  388. package/dist/cost-tracker.d.ts +76 -0
  389. package/dist/cost-tracker.d.ts.map +1 -0
  390. package/dist/{counterfactual-Dwibr5IW.d.ts → counterfactual.d.ts} +12 -13
  391. package/dist/counterfactual.d.ts.map +1 -0
  392. package/dist/cross-trace-diff.d.ts +56 -0
  393. package/dist/cross-trace-diff.d.ts.map +1 -0
  394. package/dist/{dataset-BbGkaN2I.d.ts → dataset.d.ts} +11 -14
  395. package/dist/dataset.d.ts.map +1 -0
  396. package/dist/deploy-gate-layer.d.ts +125 -0
  397. package/dist/deploy-gate-layer.d.ts.map +1 -0
  398. package/dist/deploy-gate-layer.test.d.ts +2 -0
  399. package/dist/deploy-gate-layer.test.d.ts.map +1 -0
  400. package/dist/description-length-gate.d.ts +119 -0
  401. package/dist/description-length-gate.d.ts.map +1 -0
  402. package/dist/detectors/edge.test.d.ts +2 -0
  403. package/dist/detectors/edge.test.d.ts.map +1 -0
  404. package/dist/detectors/index.d.ts +81 -0
  405. package/dist/detectors/index.d.ts.map +1 -0
  406. package/dist/detectors/index.test.d.ts +2 -0
  407. package/dist/detectors/index.test.d.ts.map +1 -0
  408. package/dist/diagnose/causal-sweep.d.ts +100 -0
  409. package/dist/diagnose/causal-sweep.d.ts.map +1 -0
  410. package/dist/diagnose/index.d.ts +36 -0
  411. package/dist/diagnose/index.d.ts.map +1 -0
  412. package/dist/diagnose/remediation.d.ts +68 -0
  413. package/dist/diagnose/remediation.d.ts.map +1 -0
  414. package/dist/diagnose/repair.d.ts +77 -0
  415. package/dist/diagnose/repair.d.ts.map +1 -0
  416. package/dist/diagnose.d.ts +1 -251
  417. package/dist/diagnose.js +2 -2
  418. package/dist/discover-personas.d.ts +35 -0
  419. package/dist/discover-personas.d.ts.map +1 -0
  420. package/dist/driver.d.ts +95 -0
  421. package/dist/driver.d.ts.map +1 -0
  422. package/dist/driver.test.d.ts +8 -0
  423. package/dist/driver.test.d.ts.map +1 -0
  424. package/dist/dual-agent-bench.d.ts +81 -0
  425. package/dist/dual-agent-bench.d.ts.map +1 -0
  426. package/dist/error-count-extractor.d.ts +47 -0
  427. package/dist/error-count-extractor.d.ts.map +1 -0
  428. package/dist/error-count-extractor.test.d.ts +2 -0
  429. package/dist/error-count-extractor.test.d.ts.map +1 -0
  430. package/dist/{errors-CzMUYo7b.d.ts → errors.d.ts} +10 -11
  431. package/dist/errors.d.ts.map +1 -0
  432. package/dist/{researcher-DE6Gpnb4.d.ts → eval-campaign.d.ts} +34 -156
  433. package/dist/eval-campaign.d.ts.map +1 -0
  434. package/dist/eval-campaign.test.d.ts +2 -0
  435. package/dist/eval-campaign.test.d.ts.map +1 -0
  436. package/dist/eval-tools.d.ts +55 -0
  437. package/dist/eval-tools.d.ts.map +1 -0
  438. package/dist/eval-trace-store.d.ts +107 -0
  439. package/dist/eval-trace-store.d.ts.map +1 -0
  440. package/dist/eval-trace-store.test.d.ts +2 -0
  441. package/dist/eval-trace-store.test.d.ts.map +1 -0
  442. package/dist/executor.d.ts +38 -0
  443. package/dist/executor.d.ts.map +1 -0
  444. package/dist/executor.test.d.ts +10 -0
  445. package/dist/executor.test.d.ts.map +1 -0
  446. package/dist/experiment-tracker.d.ts +178 -0
  447. package/dist/experiment-tracker.d.ts.map +1 -0
  448. package/dist/experiment-tracker.test.d.ts +2 -0
  449. package/dist/experiment-tracker.test.d.ts.map +1 -0
  450. package/dist/failure-taxonomy.d.ts +38 -0
  451. package/dist/failure-taxonomy.d.ts.map +1 -0
  452. package/dist/{feedback-trajectory-D9OVLrg9.d.ts → feedback-trajectory.d.ts} +36 -38
  453. package/dist/feedback-trajectory.d.ts.map +1 -0
  454. package/dist/feedback-trajectory.test.d.ts +2 -0
  455. package/dist/feedback-trajectory.test.d.ts.map +1 -0
  456. package/dist/flow-layer.d.ts +90 -0
  457. package/dist/flow-layer.d.ts.map +1 -0
  458. package/dist/flow-layer.test.d.ts +2 -0
  459. package/dist/flow-layer.test.d.ts.map +1 -0
  460. package/dist/fuzz/capsule.d.ts +46 -0
  461. package/dist/fuzz/capsule.d.ts.map +1 -0
  462. package/dist/fuzz/cube.d.ts +36 -0
  463. package/dist/fuzz/cube.d.ts.map +1 -0
  464. package/dist/fuzz/explorer-cost.test.d.ts +2 -0
  465. package/dist/fuzz/explorer-cost.test.d.ts.map +1 -0
  466. package/dist/fuzz/explorer.d.ts +64 -0
  467. package/dist/fuzz/explorer.d.ts.map +1 -0
  468. package/dist/fuzz/fuzz-agent.d.ts +16 -0
  469. package/dist/fuzz/fuzz-agent.d.ts.map +1 -0
  470. package/dist/fuzz/fuzz-agent.test.d.ts +2 -0
  471. package/dist/fuzz/fuzz-agent.test.d.ts.map +1 -0
  472. package/dist/fuzz/gates.d.ts +33 -0
  473. package/dist/fuzz/gates.d.ts.map +1 -0
  474. package/dist/fuzz/index.d.ts +26 -0
  475. package/dist/fuzz/index.d.ts.map +1 -0
  476. package/dist/fuzz/policies.d.ts +28 -0
  477. package/dist/fuzz/policies.d.ts.map +1 -0
  478. package/dist/fuzz/tools.d.ts +20 -0
  479. package/dist/fuzz/tools.d.ts.map +1 -0
  480. package/dist/fuzz/types.d.ts +307 -0
  481. package/dist/fuzz/types.d.ts.map +1 -0
  482. package/dist/fuzz.d.ts +1 -547
  483. package/dist/golden-matcher.d.ts +71 -0
  484. package/dist/golden-matcher.d.ts.map +1 -0
  485. package/dist/governance/eu-ai-act.d.ts +37 -0
  486. package/dist/governance/eu-ai-act.d.ts.map +1 -0
  487. package/dist/governance/index.d.ts +5 -135
  488. package/dist/governance/index.d.ts.map +1 -0
  489. package/dist/governance/nist-ai-rmf.d.ts +15 -0
  490. package/dist/governance/nist-ai-rmf.d.ts.map +1 -0
  491. package/dist/governance/soc2.d.ts +12 -0
  492. package/dist/governance/soc2.d.ts.map +1 -0
  493. package/dist/governance/types.d.ts +66 -0
  494. package/dist/governance/types.d.ts.map +1 -0
  495. package/dist/harness-optimizer.d.ts +82 -0
  496. package/dist/harness-optimizer.d.ts.map +1 -0
  497. package/dist/held-out-gate.d.ts +135 -0
  498. package/dist/held-out-gate.d.ts.map +1 -0
  499. package/dist/hosted/client.d.ts +73 -0
  500. package/dist/hosted/client.d.ts.map +1 -0
  501. package/dist/hosted/from-env.test.d.ts +8 -0
  502. package/dist/hosted/from-env.test.d.ts.map +1 -0
  503. package/dist/hosted/index.d.ts +10 -238
  504. package/dist/hosted/index.d.ts.map +1 -0
  505. package/dist/hosted/types.d.ts +159 -0
  506. package/dist/hosted/types.d.ts.map +1 -0
  507. package/dist/index.d.ts +277 -5643
  508. package/dist/index.d.ts.map +1 -0
  509. package/dist/index.js +317 -108
  510. package/dist/index.js.map +1 -1
  511. package/dist/integrity/backend-integrity.d.ts +71 -0
  512. package/dist/integrity/backend-integrity.d.ts.map +1 -0
  513. package/dist/integrity/preflight.d.ts +72 -0
  514. package/dist/integrity/preflight.d.ts.map +1 -0
  515. package/dist/integrity/preflight.test.d.ts +2 -0
  516. package/dist/integrity/preflight.test.d.ts.map +1 -0
  517. package/dist/integrity/single-backend.d.ts +67 -0
  518. package/dist/integrity/single-backend.d.ts.map +1 -0
  519. package/dist/intent-match-judge.d.ts +69 -0
  520. package/dist/intent-match-judge.d.ts.map +1 -0
  521. package/dist/intent-match-judge.test.d.ts +2 -0
  522. package/dist/intent-match-judge.test.d.ts.map +1 -0
  523. package/dist/{judge-calibration-0p2QcWNE.d.ts → judge-calibration.d.ts} +16 -17
  524. package/dist/judge-calibration.d.ts.map +1 -0
  525. package/dist/judge-ensemble.d.ts +66 -0
  526. package/dist/judge-ensemble.d.ts.map +1 -0
  527. package/dist/judge-ensemble.test.d.ts +8 -0
  528. package/dist/judge-ensemble.test.d.ts.map +1 -0
  529. package/dist/judge-families.d.ts +38 -0
  530. package/dist/judge-families.d.ts.map +1 -0
  531. package/dist/judge-panel.d.ts +65 -0
  532. package/dist/judge-panel.d.ts.map +1 -0
  533. package/dist/judge-retry.d.ts +70 -0
  534. package/dist/judge-retry.d.ts.map +1 -0
  535. package/dist/judge-runner.d.ts +36 -0
  536. package/dist/judge-runner.d.ts.map +1 -0
  537. package/dist/judge-runner.test.d.ts +2 -0
  538. package/dist/judge-runner.test.d.ts.map +1 -0
  539. package/dist/judges.d.ts +74 -0
  540. package/dist/judges.d.ts.map +1 -0
  541. package/dist/keyword-coverage-judge.d.ts +89 -0
  542. package/dist/keyword-coverage-judge.d.ts.map +1 -0
  543. package/dist/keyword-coverage-judge.test.d.ts +2 -0
  544. package/dist/keyword-coverage-judge.test.d.ts.map +1 -0
  545. package/dist/knowledge/index.d.ts +3 -103
  546. package/dist/knowledge/index.d.ts.map +1 -0
  547. package/dist/knowledge/readiness.d.ts +26 -0
  548. package/dist/knowledge/readiness.d.ts.map +1 -0
  549. package/dist/knowledge/types.d.ts +75 -0
  550. package/dist/knowledge/types.d.ts.map +1 -0
  551. package/dist/live-proof.d.ts +62 -0
  552. package/dist/live-proof.d.ts.map +1 -0
  553. package/dist/{llm-client-BeEcAokY.d.ts → llm-client.d.ts} +52 -23
  554. package/dist/llm-client.d.ts.map +1 -0
  555. package/dist/llm-client.test.d.ts +2 -0
  556. package/dist/llm-client.test.d.ts.map +1 -0
  557. package/dist/locked-jsonl-appender.d.ts +19 -0
  558. package/dist/locked-jsonl-appender.d.ts.map +1 -0
  559. package/dist/matrix/aggregation.d.ts +16 -0
  560. package/dist/matrix/aggregation.d.ts.map +1 -0
  561. package/dist/matrix/index.d.ts +12 -30
  562. package/dist/matrix/index.d.ts.map +1 -0
  563. package/dist/matrix/runner.d.ts +15 -0
  564. package/dist/matrix/runner.d.ts.map +1 -0
  565. package/dist/{types-mn5Aqk7x.d.ts → matrix/types.d.ts} +11 -15
  566. package/dist/matrix/types.d.ts.map +1 -0
  567. package/dist/meta-eval/calibration.d.ts +47 -0
  568. package/dist/meta-eval/calibration.d.ts.map +1 -0
  569. package/dist/meta-eval/correlation-study.d.ts +53 -0
  570. package/dist/meta-eval/correlation-study.d.ts.map +1 -0
  571. package/dist/meta-eval/index.d.ts +6 -181
  572. package/dist/meta-eval/index.d.ts.map +1 -0
  573. package/dist/meta-eval/index.js +1 -1
  574. package/dist/{outcome-store-rnXLEqSn.d.ts → meta-eval/outcome-store.d.ts} +7 -8
  575. package/dist/meta-eval/outcome-store.d.ts.map +1 -0
  576. package/dist/{rubric-predictive-validity-Cy_W-hWZ.d.ts → meta-eval/rubric-predictive-validity.d.ts} +8 -11
  577. package/dist/meta-eval/rubric-predictive-validity.d.ts.map +1 -0
  578. package/dist/meta-eval/sentinel.d.ts +169 -0
  579. package/dist/meta-eval/sentinel.d.ts.map +1 -0
  580. package/dist/metrics.d.ts +63 -0
  581. package/dist/metrics.d.ts.map +1 -0
  582. package/dist/model-seats.d.ts +71 -0
  583. package/dist/model-seats.d.ts.map +1 -0
  584. package/dist/model-seats.test.d.ts +2 -0
  585. package/dist/model-seats.test.d.ts.map +1 -0
  586. package/dist/muffled-gate-scanner.d.ts +102 -0
  587. package/dist/muffled-gate-scanner.d.ts.map +1 -0
  588. package/dist/{multi-layer-verifier-DUZXrPDA.d.ts → multi-layer-verifier.d.ts} +12 -15
  589. package/dist/multi-layer-verifier.d.ts.map +1 -0
  590. package/dist/multi-layer-verifier.test.d.ts +2 -0
  591. package/dist/multi-layer-verifier.test.d.ts.map +1 -0
  592. package/dist/multi-toolchain-layer.d.ts +80 -0
  593. package/dist/multi-toolchain-layer.d.ts.map +1 -0
  594. package/dist/multi-toolchain-layer.test.d.ts +2 -0
  595. package/dist/multi-toolchain-layer.test.d.ts.map +1 -0
  596. package/dist/multishot/default-tools.d.ts +34 -0
  597. package/dist/multishot/default-tools.d.ts.map +1 -0
  598. package/dist/multishot/index.d.ts +7 -290
  599. package/dist/multishot/index.d.ts.map +1 -0
  600. package/dist/multishot/index.js.map +1 -1
  601. package/dist/multishot/judges.d.ts +32 -0
  602. package/dist/multishot/judges.d.ts.map +1 -0
  603. package/dist/multishot/matrix.d.ts +107 -0
  604. package/dist/multishot/matrix.d.ts.map +1 -0
  605. package/dist/multishot/multishot.d.ts +23 -0
  606. package/dist/multishot/multishot.d.ts.map +1 -0
  607. package/dist/multishot/router.d.ts +37 -0
  608. package/dist/multishot/router.d.ts.map +1 -0
  609. package/dist/multishot/types.d.ts +60 -0
  610. package/dist/multishot/types.d.ts.map +1 -0
  611. package/dist/observability.d.ts +71 -0
  612. package/dist/observability.d.ts.map +1 -0
  613. package/dist/openapi.json +1 -1
  614. package/dist/oracle.d.ts +55 -0
  615. package/dist/oracle.d.ts.map +1 -0
  616. package/dist/orthogonality.d.ts +35 -0
  617. package/dist/orthogonality.d.ts.map +1 -0
  618. package/dist/otel-pipeline.d.ts +31 -0
  619. package/dist/otel-pipeline.d.ts.map +1 -0
  620. package/dist/paraphrase.d.ts +107 -0
  621. package/dist/paraphrase.d.ts.map +1 -0
  622. package/dist/{pareto-E-pembql.d.ts → pareto.d.ts} +9 -10
  623. package/dist/pareto.d.ts.map +1 -0
  624. package/dist/partition-held-out.d.ts +70 -0
  625. package/dist/partition-held-out.d.ts.map +1 -0
  626. package/dist/partition-held-out.test.d.ts +2 -0
  627. package/dist/partition-held-out.test.d.ts.map +1 -0
  628. package/dist/perf/index.d.ts +13 -119
  629. package/dist/perf/index.d.ts.map +1 -0
  630. package/dist/perf/integrity.d.ts +30 -0
  631. package/dist/perf/integrity.d.ts.map +1 -0
  632. package/dist/perf/journey.d.ts +45 -0
  633. package/dist/perf/journey.d.ts.map +1 -0
  634. package/dist/perf/ratchet.d.ts +47 -0
  635. package/dist/perf/ratchet.d.ts.map +1 -0
  636. package/dist/pipelines/budget-breach.d.ts +31 -0
  637. package/dist/pipelines/budget-breach.d.ts.map +1 -0
  638. package/dist/pipelines/budget-breach.test.d.ts +2 -0
  639. package/dist/pipelines/budget-breach.test.d.ts.map +1 -0
  640. package/dist/pipelines/failure-cluster.d.ts +38 -0
  641. package/dist/pipelines/failure-cluster.d.ts.map +1 -0
  642. package/dist/pipelines/failure-cluster.test.d.ts +2 -0
  643. package/dist/pipelines/failure-cluster.test.d.ts.map +1 -0
  644. package/dist/pipelines/first-divergence.d.ts +26 -0
  645. package/dist/pipelines/first-divergence.d.ts.map +1 -0
  646. package/dist/pipelines/first-divergence.test.d.ts +2 -0
  647. package/dist/pipelines/first-divergence.test.d.ts.map +1 -0
  648. package/dist/pipelines/index.d.ts +8 -173
  649. package/dist/pipelines/index.d.ts.map +1 -0
  650. package/dist/pipelines/index.js +40 -15
  651. package/dist/pipelines/index.js.map +1 -1
  652. package/dist/pipelines/judge-agreement.d.ts +26 -0
  653. package/dist/pipelines/judge-agreement.d.ts.map +1 -0
  654. package/dist/pipelines/judge-agreement.test.d.ts +2 -0
  655. package/dist/pipelines/judge-agreement.test.d.ts.map +1 -0
  656. package/dist/pipelines/regression.d.ts +23 -0
  657. package/dist/pipelines/regression.d.ts.map +1 -0
  658. package/dist/pipelines/regression.test.d.ts +2 -0
  659. package/dist/pipelines/regression.test.d.ts.map +1 -0
  660. package/dist/pipelines/stuck-loop.d.ts +32 -0
  661. package/dist/pipelines/stuck-loop.d.ts.map +1 -0
  662. package/dist/pipelines/stuck-loop.test.d.ts +2 -0
  663. package/dist/pipelines/stuck-loop.test.d.ts.map +1 -0
  664. package/dist/pipelines/tool-waste.d.ts +34 -0
  665. package/dist/pipelines/tool-waste.d.ts.map +1 -0
  666. package/dist/pipelines/tool-waste.test.d.ts +2 -0
  667. package/dist/pipelines/tool-waste.test.d.ts.map +1 -0
  668. package/dist/playbook.d.ts +16 -0
  669. package/dist/playbook.d.ts.map +1 -0
  670. package/dist/pr-review-benchmark.d.ts +88 -0
  671. package/dist/pr-review-benchmark.d.ts.map +1 -0
  672. package/dist/pr-review-benchmark.test.d.ts +2 -0
  673. package/dist/pr-review-benchmark.test.d.ts.map +1 -0
  674. package/dist/pre-registration.d.ts +125 -0
  675. package/dist/pre-registration.d.ts.map +1 -0
  676. package/dist/prm/builtin-rubrics.d.ts +33 -0
  677. package/dist/prm/builtin-rubrics.d.ts.map +1 -0
  678. package/dist/prm/index.d.ts +5 -100
  679. package/dist/prm/index.d.ts.map +1 -0
  680. package/dist/prm/index.js +55 -12
  681. package/dist/prm/index.js.map +1 -1
  682. package/dist/prm/inference.d.ts +29 -0
  683. package/dist/prm/inference.d.ts.map +1 -0
  684. package/dist/prm/inference.test.d.ts +2 -0
  685. package/dist/prm/inference.test.d.ts.map +1 -0
  686. package/dist/{rubric-BOfxn4ja.d.ts → prm/rubric.d.ts} +10 -13
  687. package/dist/prm/rubric.d.ts.map +1 -0
  688. package/dist/prm/training-export.d.ts +38 -0
  689. package/dist/prm/training-export.d.ts.map +1 -0
  690. package/dist/produced-state.d.ts +63 -0
  691. package/dist/produced-state.d.ts.map +1 -0
  692. package/dist/produced-state.test.d.ts +8 -0
  693. package/dist/produced-state.test.d.ts.map +1 -0
  694. package/dist/profile/baselines.d.ts +37 -0
  695. package/dist/profile/baselines.d.ts.map +1 -0
  696. package/dist/profile/index.d.ts +105 -0
  697. package/dist/profile/index.d.ts.map +1 -0
  698. package/dist/promotion-gate.d.ts +93 -0
  699. package/dist/promotion-gate.d.ts.map +1 -0
  700. package/dist/prompt-registry.d.ts +41 -0
  701. package/dist/prompt-registry.d.ts.map +1 -0
  702. package/dist/propose-review-control.d.ts +49 -0
  703. package/dist/propose-review-control.d.ts.map +1 -0
  704. package/dist/propose-review-control.test.d.ts +2 -0
  705. package/dist/propose-review-control.test.d.ts.map +1 -0
  706. package/dist/propose-review.d.ts +155 -0
  707. package/dist/propose-review.d.ts.map +1 -0
  708. package/dist/{red-team-BXHil6c8.d.ts → red-team.d.ts} +14 -16
  709. package/dist/red-team.d.ts.map +1 -0
  710. package/dist/reference-replay-steering.d.ts +11 -0
  711. package/dist/reference-replay-steering.d.ts.map +1 -0
  712. package/dist/reference-replay.d.ts +176 -0
  713. package/dist/reference-replay.d.ts.map +1 -0
  714. package/dist/reflective-mutation.d.ts +79 -0
  715. package/dist/reflective-mutation.d.ts.map +1 -0
  716. package/dist/registry.d.ts +31 -0
  717. package/dist/registry.d.ts.map +1 -0
  718. package/dist/release-confidence.d.ts +128 -0
  719. package/dist/release-confidence.d.ts.map +1 -0
  720. package/dist/release-report.d.ts +11 -0
  721. package/dist/release-report.d.ts.map +1 -0
  722. package/dist/replay.d.ts +120 -0
  723. package/dist/replay.d.ts.map +1 -0
  724. package/dist/reporter.d.ts +14 -0
  725. package/dist/reporter.d.ts.map +1 -0
  726. package/dist/reporting.d.ts +15 -15
  727. package/dist/reporting.d.ts.map +1 -0
  728. package/dist/reporting.js +3 -3
  729. package/dist/researcher.d.ts +140 -0
  730. package/dist/researcher.d.ts.map +1 -0
  731. package/dist/reviewer.d.ts +118 -0
  732. package/dist/reviewer.d.ts.map +1 -0
  733. package/dist/reviewer.test.d.ts +2 -0
  734. package/dist/reviewer.test.d.ts.map +1 -0
  735. package/dist/reward-model-export.d.ts +60 -0
  736. package/dist/reward-model-export.d.ts.map +1 -0
  737. package/dist/rl/active-curriculum.d.ts +110 -0
  738. package/dist/rl/active-curriculum.d.ts.map +1 -0
  739. package/dist/rl/adaptation-eval.d.ts +109 -0
  740. package/dist/rl/adaptation-eval.d.ts.map +1 -0
  741. package/dist/{adversarial-DIVcDoI_.d.ts → rl/adversarial.d.ts} +6 -7
  742. package/dist/rl/adversarial.d.ts.map +1 -0
  743. package/dist/rl/compute-curves.d.ts +127 -0
  744. package/dist/rl/compute-curves.d.ts.map +1 -0
  745. package/dist/rl/contamination.d.ts +117 -0
  746. package/dist/rl/contamination.d.ts.map +1 -0
  747. package/dist/rl/corpus.d.ts +55 -0
  748. package/dist/rl/corpus.d.ts.map +1 -0
  749. package/dist/rl/corpus.test.d.ts +2 -0
  750. package/dist/rl/corpus.test.d.ts.map +1 -0
  751. package/dist/rl/dataset.d.ts +102 -0
  752. package/dist/rl/dataset.d.ts.map +1 -0
  753. package/dist/rl/dataset.test.d.ts +2 -0
  754. package/dist/rl/dataset.test.d.ts.map +1 -0
  755. package/dist/rl/exporters.d.ts +141 -0
  756. package/dist/rl/exporters.d.ts.map +1 -0
  757. package/dist/rl/index.d.ts +49 -0
  758. package/dist/rl/index.d.ts.map +1 -0
  759. package/dist/{off-policy-DiwuKKg7.d.ts → rl/off-policy.d.ts} +8 -9
  760. package/dist/rl/off-policy.d.ts.map +1 -0
  761. package/dist/rl/predictive-validity-researcher.d.ts +69 -0
  762. package/dist/rl/predictive-validity-researcher.d.ts.map +1 -0
  763. package/dist/rl/preferences.d.ts +141 -0
  764. package/dist/rl/preferences.d.ts.map +1 -0
  765. package/dist/rl/process-reward.d.ts +122 -0
  766. package/dist/rl/process-reward.d.ts.map +1 -0
  767. package/dist/rl/reward-hacking.d.ts +104 -0
  768. package/dist/rl/reward-hacking.d.ts.map +1 -0
  769. package/dist/rl/rl-campaign.d.ts +85 -0
  770. package/dist/rl/rl-campaign.d.ts.map +1 -0
  771. package/dist/rl/run-record-adapters.d.ts +56 -0
  772. package/dist/rl/run-record-adapters.d.ts.map +1 -0
  773. package/dist/rl/sim-fidelity.d.ts +166 -0
  774. package/dist/rl/sim-fidelity.d.ts.map +1 -0
  775. package/dist/rl/sim-fidelity.test.d.ts +2 -0
  776. package/dist/rl/sim-fidelity.test.d.ts.map +1 -0
  777. package/dist/rl/tournament.d.ts +115 -0
  778. package/dist/rl/tournament.d.ts.map +1 -0
  779. package/dist/rl/verifiable-reward.d.ts +124 -0
  780. package/dist/rl/verifiable-reward.d.ts.map +1 -0
  781. package/dist/rl.d.ts +1 -1192
  782. package/dist/rl.js +615 -615
  783. package/dist/rl.js.map +1 -1
  784. package/dist/{run-campaign-RDGAM5KJ.js → run-campaign-WXY7KI67.js} +3 -3
  785. package/dist/run-critic.d.ts +23 -0
  786. package/dist/run-critic.d.ts.map +1 -0
  787. package/dist/run-evidence.d.ts +32 -0
  788. package/dist/run-evidence.d.ts.map +1 -0
  789. package/dist/{run-record-e7vj1uZQ.d.ts → run-record.d.ts} +16 -122
  790. package/dist/run-record.d.ts.map +1 -0
  791. package/dist/run-record.test.d.ts +2 -0
  792. package/dist/run-record.test.d.ts.map +1 -0
  793. package/dist/run-score.d.ts +31 -0
  794. package/dist/run-score.d.ts.map +1 -0
  795. package/dist/runtime-trajectory.d.ts +47 -0
  796. package/dist/runtime-trajectory.d.ts.map +1 -0
  797. package/dist/{test-graded-scenario-BdVaPyHT.d.ts → sandbox-harness.d.ts} +34 -60
  798. package/dist/sandbox-harness.d.ts.map +1 -0
  799. package/dist/sandbox-harness.test.d.ts +2 -0
  800. package/dist/sandbox-harness.test.d.ts.map +1 -0
  801. package/dist/sandbox-pool.d.ts +74 -0
  802. package/dist/sandbox-pool.d.ts.map +1 -0
  803. package/dist/sandbox-pool.test.d.ts +2 -0
  804. package/dist/sandbox-pool.test.d.ts.map +1 -0
  805. package/dist/scorecard.d.ts +133 -0
  806. package/dist/scorecard.d.ts.map +1 -0
  807. package/dist/scorecard.test.d.ts +2 -0
  808. package/dist/scorecard.test.d.ts.map +1 -0
  809. package/dist/self-play.d.ts +69 -0
  810. package/dist/self-play.d.ts.map +1 -0
  811. package/dist/semantic-concept-judge.d.ts +135 -0
  812. package/dist/semantic-concept-judge.d.ts.map +1 -0
  813. package/dist/semantic-concept-judge.test.d.ts +2 -0
  814. package/dist/semantic-concept-judge.test.d.ts.map +1 -0
  815. package/dist/{sequential-5iSVfzl2.d.ts → sequential.d.ts} +9 -10
  816. package/dist/sequential.d.ts.map +1 -0
  817. package/dist/{series-convergence-D5OWMBg6.d.ts → series-convergence.d.ts} +4 -5
  818. package/dist/series-convergence.d.ts.map +1 -0
  819. package/dist/slo.d.ts +48 -0
  820. package/dist/slo.d.ts.map +1 -0
  821. package/dist/state-continuity.d.ts +47 -0
  822. package/dist/state-continuity.d.ts.map +1 -0
  823. package/dist/{statistics-C7PozGrZ.d.ts → statistics.d.ts} +152 -39
  824. package/dist/statistics.d.ts.map +1 -0
  825. package/dist/statistics.test.d.ts +2 -0
  826. package/dist/statistics.test.d.ts.map +1 -0
  827. package/dist/steering-optimizer.d.ts +58 -0
  828. package/dist/steering-optimizer.d.ts.map +1 -0
  829. package/dist/steering.d.ts +24 -0
  830. package/dist/steering.d.ts.map +1 -0
  831. package/dist/storyboard/code-edit.d.ts +64 -0
  832. package/dist/storyboard/code-edit.d.ts.map +1 -0
  833. package/dist/storyboard/code-edit.test.d.ts +2 -0
  834. package/dist/storyboard/code-edit.test.d.ts.map +1 -0
  835. package/dist/storyboard/index.d.ts +16 -81
  836. package/dist/storyboard/index.d.ts.map +1 -0
  837. package/dist/storyboard/index.test.d.ts +2 -0
  838. package/dist/storyboard/index.test.d.ts.map +1 -0
  839. package/dist/{summary-report-DGmUucwQ.d.ts → summary-report.d.ts} +23 -160
  840. package/dist/summary-report.d.ts.map +1 -0
  841. package/dist/telemetry/client.d.ts +35 -0
  842. package/dist/telemetry/client.d.ts.map +1 -0
  843. package/dist/telemetry/index.d.ts +17 -35
  844. package/dist/telemetry/index.d.ts.map +1 -0
  845. package/dist/telemetry/schema.d.ts +61 -0
  846. package/dist/telemetry/schema.d.ts.map +1 -0
  847. package/dist/telemetry/sink-fetch.d.ts +39 -0
  848. package/dist/telemetry/sink-fetch.d.ts.map +1 -0
  849. package/dist/telemetry/{file.d.ts → sink-file.d.ts} +5 -7
  850. package/dist/telemetry/sink-file.d.ts.map +1 -0
  851. package/dist/test-graded-scenario.d.ts +42 -0
  852. package/dist/test-graded-scenario.d.ts.map +1 -0
  853. package/dist/testing.d.ts +5 -0
  854. package/dist/testing.d.ts.map +1 -0
  855. package/dist/testing.js +8 -0
  856. package/dist/testing.js.map +1 -0
  857. package/dist/tool-use-metrics.d.ts +35 -0
  858. package/dist/tool-use-metrics.d.ts.map +1 -0
  859. package/dist/trace/capture-fetch.d.ts +48 -0
  860. package/dist/trace/capture-fetch.d.ts.map +1 -0
  861. package/dist/trace/capture-fetch.test.d.ts +2 -0
  862. package/dist/trace/capture-fetch.test.d.ts.map +1 -0
  863. package/dist/{emitter-DEZwY14K.d.ts → trace/emitter.d.ts} +9 -12
  864. package/dist/trace/emitter.d.ts.map +1 -0
  865. package/dist/trace/extract-usage.d.ts +46 -0
  866. package/dist/trace/extract-usage.d.ts.map +1 -0
  867. package/dist/trace/extract-usage.test.d.ts +2 -0
  868. package/dist/trace/extract-usage.test.d.ts.map +1 -0
  869. package/dist/trace/index.d.ts +15 -0
  870. package/dist/trace/index.d.ts.map +1 -0
  871. package/dist/{integrity-VJ9A7aST.d.ts → trace/integrity.d.ts} +11 -14
  872. package/dist/trace/integrity.d.ts.map +1 -0
  873. package/dist/trace/otel-bridge.d.ts +29 -0
  874. package/dist/trace/otel-bridge.d.ts.map +1 -0
  875. package/dist/trace/otel-export.d.ts +52 -0
  876. package/dist/trace/otel-export.d.ts.map +1 -0
  877. package/dist/trace/otel.d.ts +57 -0
  878. package/dist/trace/otel.d.ts.map +1 -0
  879. package/dist/trace/otlp-attributes.d.ts +17 -0
  880. package/dist/trace/otlp-attributes.d.ts.map +1 -0
  881. package/dist/trace/query.d.ts +29 -0
  882. package/dist/trace/query.d.ts.map +1 -0
  883. package/dist/trace/query.test.d.ts +2 -0
  884. package/dist/trace/query.test.d.ts.map +1 -0
  885. package/dist/{raw-provider-sink-C46HDghv.d.ts → trace/raw-provider-sink.d.ts} +13 -14
  886. package/dist/trace/raw-provider-sink.d.ts.map +1 -0
  887. package/dist/{redact-B40YG2M_.d.ts → trace/redact.d.ts} +7 -8
  888. package/dist/trace/redact.d.ts.map +1 -0
  889. package/dist/{schema-m0gsnbt3.d.ts → trace/schema.d.ts} +29 -30
  890. package/dist/trace/schema.d.ts.map +1 -0
  891. package/dist/trace/store-to-otlp.d.ts +72 -0
  892. package/dist/trace/store-to-otlp.d.ts.map +1 -0
  893. package/dist/trace/store-to-otlp.test.d.ts +2 -0
  894. package/dist/trace/store-to-otlp.test.d.ts.map +1 -0
  895. package/dist/{store-CKUAgsJz.d.ts → trace/store.d.ts} +37 -13
  896. package/dist/trace/store.d.ts.map +1 -0
  897. package/dist/trace/store.test.d.ts +2 -0
  898. package/dist/trace/store.test.d.ts.map +1 -0
  899. package/dist/{analyst-C8HHvfJp.d.ts → trace-analyst/analyst.d.ts} +12 -14
  900. package/dist/trace-analyst/analyst.d.ts.map +1 -0
  901. package/dist/trace-analyst/analyst.test.d.ts +2 -0
  902. package/dist/trace-analyst/analyst.test.d.ts.map +1 -0
  903. package/dist/trace-analyst/behavioral-metrics.d.ts +40 -0
  904. package/dist/trace-analyst/behavioral-metrics.d.ts.map +1 -0
  905. package/dist/trace-analyst/behavioral-metrics.test.d.ts +2 -0
  906. package/dist/trace-analyst/behavioral-metrics.test.d.ts.map +1 -0
  907. package/dist/trace-analyst/hook.d.ts +55 -0
  908. package/dist/trace-analyst/hook.d.ts.map +1 -0
  909. package/dist/trace-analyst/index.d.ts +18 -0
  910. package/dist/trace-analyst/index.d.ts.map +1 -0
  911. package/dist/trace-analyst/insights.d.ts +71 -0
  912. package/dist/trace-analyst/insights.d.ts.map +1 -0
  913. package/dist/trace-analyst/insights.test.d.ts +2 -0
  914. package/dist/trace-analyst/insights.test.d.ts.map +1 -0
  915. package/dist/trace-analyst/otlp-flatten.d.ts +42 -0
  916. package/dist/trace-analyst/otlp-flatten.d.ts.map +1 -0
  917. package/dist/trace-analyst/otlp-span.d.ts +85 -0
  918. package/dist/trace-analyst/otlp-span.d.ts.map +1 -0
  919. package/dist/trace-analyst/otlp-span.test.d.ts +2 -0
  920. package/dist/trace-analyst/otlp-span.test.d.ts.map +1 -0
  921. package/dist/trace-analyst/otlp-to-run-records.d.ts +115 -0
  922. package/dist/trace-analyst/otlp-to-run-records.d.ts.map +1 -0
  923. package/dist/trace-analyst/otlp-to-run-records.test.d.ts +2 -0
  924. package/dist/trace-analyst/otlp-to-run-records.test.d.ts.map +1 -0
  925. package/dist/trace-analyst/otlp-to-run-records.timestamps.test.d.ts +2 -0
  926. package/dist/trace-analyst/otlp-to-run-records.timestamps.test.d.ts.map +1 -0
  927. package/dist/trace-analyst/prompts.d.ts +6 -0
  928. package/dist/trace-analyst/prompts.d.ts.map +1 -0
  929. package/dist/trace-analyst/store-otlp.d.ts +126 -0
  930. package/dist/trace-analyst/store-otlp.d.ts.map +1 -0
  931. package/dist/trace-analyst/store-otlp.test.d.ts +8 -0
  932. package/dist/trace-analyst/store-otlp.test.d.ts.map +1 -0
  933. package/dist/trace-analyst/store-otlp.timestamps.test.d.ts +2 -0
  934. package/dist/trace-analyst/store-otlp.timestamps.test.d.ts.map +1 -0
  935. package/dist/trace-analyst/store.d.ts +63 -0
  936. package/dist/trace-analyst/store.d.ts.map +1 -0
  937. package/dist/trace-analyst/tools.d.ts +44 -0
  938. package/dist/trace-analyst/tools.d.ts.map +1 -0
  939. package/dist/trace-analyst/tools.test.d.ts +10 -0
  940. package/dist/trace-analyst/tools.test.d.ts.map +1 -0
  941. package/dist/{store-C1YxJDEK.d.ts → trace-analyst/types.d.ts} +18 -74
  942. package/dist/trace-analyst/types.d.ts.map +1 -0
  943. package/dist/trace-contracts.d.ts +180 -0
  944. package/dist/trace-contracts.d.ts.map +1 -0
  945. package/dist/traced-analyst.d.ts +26 -0
  946. package/dist/traced-analyst.d.ts.map +1 -0
  947. package/dist/traced-judges.d.ts +27 -0
  948. package/dist/traced-judges.d.ts.map +1 -0
  949. package/dist/traces.d.ts +4 -947
  950. package/dist/traces.d.ts.map +1 -0
  951. package/dist/traces.js +42 -12
  952. package/dist/{trajectory-GEdXJCL5.d.ts → trajectory.d.ts} +8 -9
  953. package/dist/trajectory.d.ts.map +1 -0
  954. package/dist/{types-Croy5h7V.d.ts → types.d.ts} +33 -33
  955. package/dist/types.d.ts.map +1 -0
  956. package/dist/ui-finding.d.ts +104 -0
  957. package/dist/ui-finding.d.ts.map +1 -0
  958. package/dist/verdict-cache.d.ts +78 -0
  959. package/dist/verdict-cache.d.ts.map +1 -0
  960. package/dist/verdict-cache.test.d.ts +2 -0
  961. package/dist/verdict-cache.test.d.ts.map +1 -0
  962. package/dist/{verdict-C9MlYujm.d.ts → verdict.d.ts} +2 -3
  963. package/dist/verdict.d.ts.map +1 -0
  964. package/dist/visual-diff.d.ts +32 -0
  965. package/dist/visual-diff.d.ts.map +1 -0
  966. package/dist/wire/handlers.d.ts +54 -0
  967. package/dist/wire/handlers.d.ts.map +1 -0
  968. package/dist/wire/index.d.ts +13 -570
  969. package/dist/wire/index.d.ts.map +1 -0
  970. package/dist/wire/index.js +2 -2
  971. package/dist/wire/openapi.d.ts +3 -0
  972. package/dist/wire/openapi.d.ts.map +1 -0
  973. package/dist/wire/rpc.d.ts +21 -0
  974. package/dist/wire/rpc.d.ts.map +1 -0
  975. package/dist/wire/rubrics.d.ts +34 -0
  976. package/dist/wire/rubrics.d.ts.map +1 -0
  977. package/dist/wire/schemas.d.ts +410 -0
  978. package/dist/wire/schemas.d.ts.map +1 -0
  979. package/dist/wire/server.d.ts +60 -0
  980. package/dist/wire/server.d.ts.map +1 -0
  981. package/dist/workflow/event-schema.d.ts +5 -0
  982. package/dist/workflow/event-schema.d.ts.map +1 -0
  983. package/dist/workflow/feedback-pack.d.ts +99 -0
  984. package/dist/workflow/feedback-pack.d.ts.map +1 -0
  985. package/dist/workflow/index.d.ts +22 -495
  986. package/dist/workflow/index.d.ts.map +1 -0
  987. package/dist/workflow/index.js +2 -2
  988. package/dist/workflow/intelligence-export.d.ts +62 -0
  989. package/dist/workflow/intelligence-export.d.ts.map +1 -0
  990. package/dist/workflow/partner-report.d.ts +49 -0
  991. package/dist/workflow/partner-report.d.ts.map +1 -0
  992. package/dist/workflow/phase-graph.d.ts +43 -0
  993. package/dist/workflow/phase-graph.d.ts.map +1 -0
  994. package/dist/workflow/promotion-gate.d.ts +61 -0
  995. package/dist/workflow/promotion-gate.d.ts.map +1 -0
  996. package/dist/workflow/run-record.d.ts +12 -0
  997. package/dist/workflow/run-record.d.ts.map +1 -0
  998. package/dist/workflow/runtime-adapter.d.ts +20 -0
  999. package/dist/workflow/runtime-adapter.d.ts.map +1 -0
  1000. package/dist/workflow/sanitize.d.ts +21 -0
  1001. package/dist/workflow/sanitize.d.ts.map +1 -0
  1002. package/dist/workflow/schema.d.ts +5 -0
  1003. package/dist/workflow/schema.d.ts.map +1 -0
  1004. package/dist/workflow/summary.d.ts +43 -0
  1005. package/dist/workflow/summary.d.ts.map +1 -0
  1006. package/dist/workflow/trace-event-fields.d.ts +6 -0
  1007. package/dist/workflow/trace-event-fields.d.ts.map +1 -0
  1008. package/dist/workflow/trajectory.d.ts +15 -0
  1009. package/dist/workflow/trajectory.d.ts.map +1 -0
  1010. package/dist/workflow/types.d.ts +68 -0
  1011. package/dist/workflow/types.d.ts.map +1 -0
  1012. package/dist/workspace-inspector.d.ts +67 -0
  1013. package/dist/workspace-inspector.d.ts.map +1 -0
  1014. package/dist/wrangler-deploy-runner.test.d.ts +2 -0
  1015. package/dist/wrangler-deploy-runner.test.d.ts.map +1 -0
  1016. package/docs/campaign-proposers.md +170 -0
  1017. package/docs/concepts.md +8 -4
  1018. package/docs/customer-journeys.md +15 -13
  1019. package/docs/design/loop-taxonomy.md +34 -66
  1020. package/docs/distributed-driver.md +14 -14
  1021. package/docs/feature-guide.md +1 -1
  1022. package/docs/hosted-ingest-spec.md +2 -3
  1023. package/docs/multi-shot-optimization.md +8 -8
  1024. package/docs/product-eval-adoption.md +1 -1
  1025. package/docs/self-improvement-map.md +33 -29
  1026. package/package.json +9 -15
  1027. package/dist/calibration-Cpr3WaX3.d.ts +0 -101
  1028. package/dist/chunk-47X6LRCE.js.map +0 -1
  1029. package/dist/chunk-6SOJM3VR.js.map +0 -1
  1030. package/dist/chunk-CVVHBFGN.js.map +0 -1
  1031. package/dist/chunk-D3V5B42D.js.map +0 -1
  1032. package/dist/chunk-E4GH6USR.js.map +0 -1
  1033. package/dist/chunk-GSH6QNNS.js.map +0 -1
  1034. package/dist/chunk-KWRRMR3J.js.map +0 -1
  1035. package/dist/chunk-LMZQ2Z4U.js.map +0 -1
  1036. package/dist/chunk-QAY5UIJO.js.map +0 -1
  1037. package/dist/chunk-QG2OVF2D.js.map +0 -1
  1038. package/dist/chunk-T375SUOZ.js.map +0 -1
  1039. package/dist/chunk-TWS7AZEY.js.map +0 -1
  1040. package/dist/chunk-UMMZHCPB.js.map +0 -1
  1041. package/dist/chunk-XY4DDNEG.js.map +0 -1
  1042. package/dist/chunk-ZFIBGEOL.js.map +0 -1
  1043. package/dist/control-_Qb7skHX.d.ts +0 -259
  1044. package/dist/corpus-BoR-041R.d.ts +0 -560
  1045. package/dist/failure-cluster-CL7IVgkJ.d.ts +0 -76
  1046. package/dist/harness-optimizer-EnEnQPsr.d.ts +0 -106
  1047. package/dist/index-Bx3gZ8xl.d.ts +0 -159
  1048. package/dist/pre-registration-mAnCugl9.d.ts +0 -523
  1049. package/dist/provenance-LnqRT0sS.d.ts +0 -441
  1050. package/dist/query-CqTxMwDw.d.ts +0 -31
  1051. package/dist/release-report-euXIV_Sk.d.ts +0 -233
  1052. package/dist/run-critic-BAIjX99r.d.ts +0 -56
  1053. package/dist/run-improvement-loop-5z_l5zDz.d.ts +0 -427
  1054. package/dist/runtime-trajectory-BDgfGZSr.d.ts +0 -49
  1055. package/dist/semantic-concept-judge-Dn8Z6KEG.d.ts +0 -624
  1056. package/dist/sink-fetch-B1Yg4Til.d.ts +0 -101
  1057. package/docs/design/external-agent-wedge.md +0 -89
  1058. package/docs/design/phase-d-rfc.md +0 -125
  1059. package/docs/design/phase4-consumer-migration.md +0 -70
  1060. package/docs/design/primitives-integration-spec.md +0 -393
  1061. package/docs/design/product-self-improvement-loop.md +0 -146
  1062. package/docs/design/self-improvement-engine.md +0 -140
  1063. package/docs/design/self-improvement-protocol.md +0 -223
  1064. package/docs/design/self-improvement-roadmap.md +0 -106
  1065. package/docs/design/substrate-gaps.md +0 -118
  1066. package/docs/phase-b-pairing-kit.md +0 -188
  1067. package/docs/phase-b-runbook.md +0 -176
  1068. package/docs/pilot/README.md +0 -62
  1069. package/docs/pilot/customer-checklist.md +0 -90
  1070. package/docs/pilot/integration-foreign-stack.md +0 -296
  1071. package/docs/pilot/integration-tangle-stack.md +0 -248
  1072. package/docs/pilot/one-pager.md +0 -161
  1073. package/docs/pilot/sample-insight-report.json +0 -172
  1074. package/docs/quickstart-external.md +0 -229
  1075. package/docs/research/belief-state-agent-eval-roadmap.md +0 -593
  1076. package/docs/research/research-roadmap.md +0 -205
  1077. package/docs/specs/driver-honest-spec.md +0 -251
  1078. package/docs/specs/hermes-self-improvement-audit.md +0 -93
  1079. package/docs/specs/profile-versioning.md +0 -291
  1080. package/docs/three-package-architecture.md +0 -168
  1081. /package/dist/{chunk-QS3RBQPI.js.map → chunk-D5KBVULY.js.map} +0 -0
  1082. /package/dist/{chunk-UHMJT4T7.js.map → chunk-IZCEK2HR.js.map} +0 -0
  1083. /package/dist/{chunk-L3JOU6XM.js.map → chunk-NACM5RS7.js.map} +0 -0
  1084. /package/dist/{chunk-4FBZZIYD.js.map → chunk-QTMYW64F.js.map} +0 -0
  1085. /package/dist/{chunk-CY6U5S3X.js.map → chunk-URWVKRCS.js.map} +0 -0
  1086. /package/dist/{chunk-Y47J2LJ3.js.map → chunk-Z6L6YSU6.js.map} +0 -0
  1087. /package/dist/{run-campaign-RDGAM5KJ.js.map → run-campaign-WXY7KI67.js.map} +0 -0
@@ -1,181 +1,6 @@
1
- export { C as CalibrationBin, a as CalibrationOptions, b as CalibrationPair, c as CalibrationReport, d as CorrelationResult, e as CorrelationStudyOptions, f as CorrelationStudyResult, E as EvalMetricSpec, O as OutcomePair, g as calibrationCurve, h as calibrationFromPairs, i as correlationStudy } from '../calibration-Cpr3WaX3.js';
2
- export { D as DeploymentOutcome, F as FileSystemOutcomeStore, a as FileSystemOutcomeStoreOptions, I as InMemoryOutcomeStore, O as OutcomeFilter, b as OutcomeStore } from '../outcome-store-rnXLEqSn.js';
3
- export { R as RubricOutcomePair, a as RubricPredictiveValidityInput, b as RubricPredictiveValidityReport, c as RubricRanking, r as rubricPredictiveValidity } from '../rubric-predictive-validity-Cy_W-hWZ.js';
4
- import { C as ContinuousAgreement, a as CalibrationResult, b as ContinuousCalibrationResult, c as CandidateScore, G as GoldenItem } from '../judge-calibration-0p2QcWNE.js';
5
- import { S as SeriesConvergenceOptions, a as SeriesConvergenceResult } from '../series-convergence-D5OWMBg6.js';
6
- import { C as CorpusAgreementReport } from '../statistics-C7PozGrZ.js';
7
- import '../store-CKUAgsJz.js';
8
- import '../schema-m0gsnbt3.js';
9
- import '../run-record-e7vj1uZQ.js';
10
- import '../errors-CzMUYo7b.js';
11
- import '../types-Croy5h7V.js';
12
- import '@tangle-network/tcloud';
13
-
14
- /**
15
- * Judge sentinel — eval trustworthiness as a continuously measured,
16
- * alarmed trend.
17
- *
18
- * Judges are models; models change underneath us; calibration decays.
19
- * This module composes three existing instruments into one loop:
20
- *
21
- * snapshot — adapters turn real instrument outputs (`calibrateJudge` /
22
- * `calibrateJudgeContinuous`, `corpusInterRaterAgreement` /
23
- * `continuousAgreement`, a rerun sentinel golden set) into
24
- * `SentinelSnapshot`s
25
- * series — `analyzeSeries` (src/series-convergence) classifies each
26
- * judge×metric history as stabilized / drifting / noisy
27
- * alarm — `judgeSentinelReport` turns trends + thresholds into named
28
- * alarms; `evalHealthStamp` is the tiny object a campaign
29
- * attaches to its verdicts
30
- *
31
- * Alarm conditions:
32
- * - convergence state `drifting-down` on any tracked metric
33
- * - irr below `minIrr`
34
- * - calibrationKappa / sentinelPassRate dropped vs the series baseline
35
- * beyond `maxKappaDrop` / `maxSentinelDrop`
36
- * - judge model changed with no post-change golden-grounded snapshot
37
- * (the silent-upgrade trap: judges agreeing with each other after a
38
- * model swap proves nothing — only re-measuring against gold does)
39
- * - newest snapshot older than `staleAfterDays` relative to `asOf`
40
- *
41
- * Wire-in: run the snapshot adapters from a post-campaign hook or a
42
- * nightly job, append to a `SentinelStore`, then compute
43
- * `judgeSentinelReport(await store.history(), { asOf })` with `asOf` =
44
- * the campaign timestamp. Attach `evalHealthStamp(report)` to campaign
45
- * verdicts: an alarmed sentinel means every downstream verdict carries
46
- * an integrity warning until the judge is recalibrated.
47
- *
48
- * The module is clock-free — every timestamp (`at`, `asOf`) is
49
- * caller-supplied ISO-8601, so reports are reproducible.
50
- */
51
-
52
- declare const SENTINEL_METRIC_NAMES: readonly ["irr", "calibrationKappa", "sentinelPassRate"];
53
- type SentinelMetricName = (typeof SENTINEL_METRIC_NAMES)[number];
54
- interface SentinelMetrics {
55
- /** Inter-rater reliability — ICC(2,1) from agreement instruments. */
56
- irr?: number;
57
- /** Weighted κ vs the human golden set (`calibrateJudge*`). */
58
- calibrationKappa?: number;
59
- /** Fraction of a fixed sentinel golden set the judge still scores within tolerance. */
60
- sentinelPassRate?: number;
61
- }
62
- interface SentinelSnapshot {
63
- /** Caller-supplied ISO-8601 timestamp — the module never reads a clock. */
64
- at: string;
65
- judgeId: string;
66
- /** Exact model identity behind the judge (pin the full version string). */
67
- judgeModel: string;
68
- /**
69
- * Metrics measured at `at`. May be empty: an empty-metrics snapshot is a
70
- * model-change marker — it records the new `judgeModel` immediately and
71
- * the silent-upgrade alarm stays raised until a golden-grounded snapshot
72
- * (calibrationKappa or sentinelPassRate) follows.
73
- */
74
- metrics: SentinelMetrics;
75
- }
76
- /** Identity + timestamp the caller supplies alongside an instrument output. */
77
- interface SnapshotMeta {
78
- at: string;
79
- judgeId: string;
80
- judgeModel: string;
81
- }
82
- /** Shape guard for snapshots crossing a boundary (store append, JSONL load, report input). */
83
- declare function validateSentinelSnapshot(snapshot: SentinelSnapshot, source?: string): void;
84
- /**
85
- * From `calibrateJudge` / `calibrateJudgeContinuous` output. Prefers the
86
- * un-rounded continuous κ_w when the report carries one (integer κ
87
- * discards information for [0,1] judges).
88
- */
89
- declare function snapshotFromCalibration(report: CalibrationResult | ContinuousCalibrationResult, meta: SnapshotMeta): SentinelSnapshot;
90
- /**
91
- * From `corpusInterRaterAgreement` (statistics.ts) or `continuousAgreement`
92
- * (judge-calibration.ts) output. IRR = ICC(2,1) — the reliability
93
- * coefficient both instruments compute. For ensemble-level agreement,
94
- * `meta.judgeId` names the ensemble and `meta.judgeModel` pins its
95
- * composition (e.g. a joined list of member model strings).
96
- */
97
- declare function snapshotFromAgreement(report: CorpusAgreementReport | ContinuousAgreement, meta: SnapshotMeta): SentinelSnapshot;
98
- interface SentinelSetOptions {
99
- /** Max |judge − human| for an item to count as passed. Default 0.1 (tuned for [0,1] scores). */
100
- tolerance?: number;
101
- }
102
- /**
103
- * From a rerun of a fixed sentinel golden set: the same `GoldenItem`s the
104
- * judge was originally calibrated on, re-scored periodically. Pass rate =
105
- * fraction of joined items where the judge stays within `tolerance` of the
106
- * human score. Items the judge didn't score are excluded from the
107
- * denominator; zero joined items is an error, not a 0% pass rate.
108
- */
109
- declare function snapshotFromSentinelSet(scores: CandidateScore[], golden: GoldenItem[], meta: SnapshotMeta, options?: SentinelSetOptions): SentinelSnapshot;
110
- interface SentinelStore {
111
- append(snapshot: SentinelSnapshot): Promise<void>;
112
- /** All snapshots (optionally for one judge), in append order. */
113
- history(judgeId?: string): Promise<SentinelSnapshot[]>;
114
- }
115
- declare function inMemorySentinelStore(initial?: SentinelSnapshot[]): SentinelStore;
116
- /**
117
- * JSONL store — one snapshot per line, appended atomically per call.
118
- * A corrupt or shape-invalid line is a loud error naming the file and
119
- * line number: a sentinel history that silently drops records would
120
- * defeat the drift detection it exists to provide.
121
- */
122
- declare function fileSentinelStore(path: string): SentinelStore;
123
- interface SentinelThresholds {
124
- /** Floor for irr — below it the judge ensemble is unreliable regardless of trend. Default 0.6. */
125
- minIrr?: number;
126
- /** Max tolerated calibrationKappa drop vs the series baseline. Default 0.15. */
127
- maxKappaDrop?: number;
128
- /** Max tolerated sentinelPassRate drop vs the series baseline. Default 0.1. */
129
- maxSentinelDrop?: number;
130
- /** Newest snapshot older than this (days, vs asOf) = the sentinel itself went blind. Default 30. */
131
- staleAfterDays?: number;
132
- }
133
- interface JudgeSentinelOptions {
134
- /** Caller-supplied ISO timestamp staleness is measured against. */
135
- asOf: string;
136
- thresholds?: SentinelThresholds;
137
- /** Forwarded to `analyzeSeries` (window, stableCv, driftRun). */
138
- convergence?: SeriesConvergenceOptions;
139
- }
140
- interface SentinelTrend {
141
- judgeId: string;
142
- metric: SentinelMetricName;
143
- /** Verbatim `analyzeSeries` state for this judge×metric history. */
144
- state: SeriesConvergenceResult['state'];
145
- /** Newest value in the series. */
146
- current: number;
147
- /** Oldest value in the series — the reference the drop thresholds compare against. */
148
- baseline: number;
149
- /** current − baseline (signed; negative = decayed). */
150
- drift: number;
151
- alarmed: boolean;
152
- reason?: string;
153
- }
154
- interface SentinelReport {
155
- perJudge: SentinelTrend[];
156
- alarms: string[];
157
- healthy: boolean;
158
- /**
159
- * `judgeId:metric` pairs with too few snapshots for the convergence
160
- * machine. Named, not counted — a blind spot you can't see is a blind
161
- * spot you won't fix. Floor/drop checks still apply to thin series, so
162
- * insufficient history alone never masks an alarm.
163
- */
164
- insufficientHistory: string[];
165
- }
166
- declare function judgeSentinelReport(history: SentinelSnapshot[], opts: JudgeSentinelOptions): SentinelReport;
167
- interface EvalHealthStamp {
168
- healthy: boolean;
169
- alarms: string[];
170
- }
171
- /**
172
- * The object a campaign attaches to its verdicts. Compute it from the
173
- * sentinel report in a post-campaign hook (or a nightly job feeding the
174
- * next day's campaigns) and store it alongside the verdict payload.
175
- * `healthy: false` means the judges that produced those verdicts have an
176
- * unresolved drift / decay / silent-upgrade / staleness alarm — treat the
177
- * verdicts as carrying an integrity warning until recalibration clears it.
178
- */
179
- declare function evalHealthStamp(report: SentinelReport): EvalHealthStamp;
180
-
181
- export { type EvalHealthStamp, type JudgeSentinelOptions, SENTINEL_METRIC_NAMES, type SentinelMetricName, type SentinelMetrics, type SentinelReport, type SentinelSetOptions, type SentinelSnapshot, type SentinelStore, type SentinelThresholds, type SentinelTrend, type SnapshotMeta, evalHealthStamp, fileSentinelStore, inMemorySentinelStore, judgeSentinelReport, snapshotFromAgreement, snapshotFromCalibration, snapshotFromSentinelSet, validateSentinelSnapshot };
1
+ export * from './calibration';
2
+ export * from './correlation-study';
3
+ export * from './outcome-store';
4
+ export * from './rubric-predictive-validity';
5
+ export * from './sentinel';
6
+ //# sourceMappingURL=index.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/meta-eval/index.ts"],"names":[],"mappings":"AAAA,cAAc,eAAe,CAAA;AAC7B,cAAc,qBAAqB,CAAA;AACnC,cAAc,iBAAiB,CAAA;AAC/B,cAAc,8BAA8B,CAAA;AAC5C,cAAc,YAAY,CAAA"}
@@ -15,7 +15,7 @@ import {
15
15
  import {
16
16
  aggregateLlm,
17
17
  llmSpans
18
- } from "../chunk-47X6LRCE.js";
18
+ } from "../chunk-UFSG7ACU.js";
19
19
  import "../chunk-5BKGXME7.js";
20
20
  import {
21
21
  ValidationError
@@ -10,7 +10,7 @@
10
10
  * The whole point of this module is to make the meta-eval correlation
11
11
  * question computable: `correlate(evalMetric, outcomeMetric) → r, ρ, n, CI`.
12
12
  */
13
- interface DeploymentOutcome {
13
+ export interface DeploymentOutcome {
14
14
  runId: string;
15
15
  capturedAt: number;
16
16
  /** Numeric outcomes keyed by name — retention_7d, csat, revenue_usd, etc. */
@@ -20,7 +20,7 @@ interface DeploymentOutcome {
20
20
  /** Free-form provenance (source system, pipeline version). */
21
21
  source?: string;
22
22
  }
23
- interface OutcomeFilter {
23
+ export interface OutcomeFilter {
24
24
  runIds?: string[];
25
25
  since?: number;
26
26
  until?: number;
@@ -30,24 +30,24 @@ interface OutcomeFilter {
30
30
  };
31
31
  source?: string;
32
32
  }
33
- interface OutcomeStore {
33
+ export interface OutcomeStore {
34
34
  append(outcome: DeploymentOutcome): Promise<void>;
35
35
  /** All outcomes attached to this run (a single run can have many — multiple
36
36
  * capture windows over deployment time). */
37
37
  forRun(runId: string): Promise<DeploymentOutcome[]>;
38
38
  list(filter?: OutcomeFilter): Promise<DeploymentOutcome[]>;
39
39
  }
40
- declare class InMemoryOutcomeStore implements OutcomeStore {
40
+ export declare class InMemoryOutcomeStore implements OutcomeStore {
41
41
  private items;
42
42
  append(outcome: DeploymentOutcome): Promise<void>;
43
43
  forRun(runId: string): Promise<DeploymentOutcome[]>;
44
44
  list(filter?: OutcomeFilter): Promise<DeploymentOutcome[]>;
45
45
  }
46
- interface FileSystemOutcomeStoreOptions {
46
+ export interface FileSystemOutcomeStoreOptions {
47
47
  dir: string;
48
48
  maxBytes?: number;
49
49
  }
50
- declare class FileSystemOutcomeStore implements OutcomeStore {
50
+ export declare class FileSystemOutcomeStore implements OutcomeStore {
51
51
  private dir;
52
52
  private maxBytes;
53
53
  private memo?;
@@ -59,5 +59,4 @@ declare class FileSystemOutcomeStore implements OutcomeStore {
59
59
  forRun(runId: string): Promise<DeploymentOutcome[]>;
60
60
  list(filter?: OutcomeFilter): Promise<DeploymentOutcome[]>;
61
61
  }
62
-
63
- export { type DeploymentOutcome as D, FileSystemOutcomeStore as F, InMemoryOutcomeStore as I, type OutcomeFilter as O, type FileSystemOutcomeStoreOptions as a, type OutcomeStore as b };
62
+ //# sourceMappingURL=outcome-store.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"outcome-store.d.ts","sourceRoot":"","sources":["../../src/meta-eval/outcome-store.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,MAAM,WAAW,iBAAiB;IAChC,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,EAAE,MAAM,CAAA;IAClB,6EAA6E;IAC7E,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC/B,yEAAyE;IACzE,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC/B,8DAA8D;IAC9D,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,aAAa;IAC5B,MAAM,CAAC,EAAE,MAAM,EAAE,CAAA;IACjB,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,KAAK,CAAC,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAA;IACtC,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,YAAY;IAC3B,MAAM,CAAC,OAAO,EAAE,iBAAiB,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IACjD;iDAC6C;IAC7C,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,iBAAiB,EAAE,CAAC,CAAA;IACnD,IAAI,CAAC,MAAM,CAAC,EAAE,aAAa,GAAG,OAAO,CAAC,iBAAiB,EAAE,CAAC,CAAA;CAC3D;AAED,qBAAa,oBAAqB,YAAW,YAAY;IACvD,OAAO,CAAC,KAAK,CAA0B;IAEjC,MAAM,CAAC,OAAO,EAAE,iBAAiB,GAAG,OAAO,CAAC,IAAI,CAAC;IAIjD,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,iBAAiB,EAAE,CAAC;IAInD,IAAI,CAAC,MAAM,GAAE,aAAkB,GAAG,OAAO,CAAC,iBAAiB,EAAE,CAAC;CAGrE;AAED,MAAM,WAAW,6BAA6B;IAC5C,GAAG,EAAE,MAAM,CAAA;IACX,QAAQ,CAAC,EAAE,MAAM,CAAA;CAClB;AAED,qBAAa,sBAAuB,YAAW,YAAY;IACzD,OAAO,CAAC,GAAG,CAAQ;IACnB,OAAO,CAAC,QAAQ,CAAQ;IACxB,OAAO,CAAC,IAAI,CAAC,CAAsB;IACnC,OAAO,CAAC,MAAM,CAAQ;gBAEV,OAAO,EAAE,6BAA6B;YAKpC,SAAS;IAKjB,MAAM,CAAC,OAAO,EAAE,iBAAiB,GAAG,OAAO,CAAC,IAAI,CAAC;YAiBzC,IAAI;IAuBZ,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,iBAAiB,EAAE,CAAC;IAInD,IAAI,CAAC,MAAM,CAAC,EAAE,aAAa,GAAG,OAAO,CAAC,iBAAiB,EAAE,CAAC;CAGjE"}
@@ -1,6 +1,3 @@
1
- import { R as RunRecord } from './run-record-e7vj1uZQ.js';
2
- import { b as OutcomeStore } from './outcome-store-rnXLEqSn.js';
3
-
4
1
  /**
5
2
  * Rubric predictive validity — does our eval rubric predict deployment
6
3
  * outcomes?
@@ -32,8 +29,9 @@ import { b as OutcomeStore } from './outcome-store-rnXLEqSn.js';
32
29
  * decorative rubrics, re-weight composite scores, or trigger a
33
30
  * recalibration sweep when predictive validity drops below a threshold.
34
31
  */
35
-
36
- interface RubricPredictiveValidityInput {
32
+ import type { RunRecord } from '../run-record';
33
+ import type { OutcomeStore } from './outcome-store';
34
+ export interface RubricPredictiveValidityInput {
37
35
  /**
38
36
  * Canonical campaign output. Each record's `outcome.raw[<rubricId>]`
39
37
  * provides the eval score; missing keys are silently skipped per pair.
@@ -64,7 +62,7 @@ interface RubricPredictiveValidityInput {
64
62
  */
65
63
  reduction?: 'latest' | 'mean' | 'max';
66
64
  }
67
- interface RubricOutcomePair {
65
+ export interface RubricOutcomePair {
68
66
  rubric: string;
69
67
  outcome: string;
70
68
  n: number;
@@ -82,7 +80,7 @@ interface RubricOutcomePair {
82
80
  */
83
81
  verdict: 'load_bearing' | 'informative' | 'decorative';
84
82
  }
85
- interface RubricRanking {
83
+ export interface RubricRanking {
86
84
  rubric: string;
87
85
  /** Outcome metric this rubric correlated best with. */
88
86
  bestOutcome: string;
@@ -91,7 +89,7 @@ interface RubricRanking {
91
89
  n: number;
92
90
  verdict: RubricOutcomePair['verdict'];
93
91
  }
94
- interface RubricPredictiveValidityReport {
92
+ export interface RubricPredictiveValidityReport {
95
93
  pairs: RubricOutcomePair[];
96
94
  /** Per-rubric best pair, sorted descending by |spearman|. */
97
95
  ranked: RubricRanking[];
@@ -100,6 +98,5 @@ interface RubricPredictiveValidityReport {
100
98
  /** Rubrics that were declared but never produced a usable score. */
101
99
  rubricsWithoutData: string[];
102
100
  }
103
- declare function rubricPredictiveValidity(input: RubricPredictiveValidityInput): Promise<RubricPredictiveValidityReport>;
104
-
105
- export { type RubricOutcomePair as R, type RubricPredictiveValidityInput as a, type RubricPredictiveValidityReport as b, type RubricRanking as c, rubricPredictiveValidity as r };
101
+ export declare function rubricPredictiveValidity(input: RubricPredictiveValidityInput): Promise<RubricPredictiveValidityReport>;
102
+ //# sourceMappingURL=rubric-predictive-validity.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"rubric-predictive-validity.d.ts","sourceRoot":"","sources":["../../src/meta-eval/rubric-predictive-validity.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AAEH,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,eAAe,CAAA;AAC9C,OAAO,KAAK,EAAqB,YAAY,EAAE,MAAM,iBAAiB,CAAA;AAEtE,MAAM,WAAW,6BAA6B;IAC5C;;;OAGG;IACH,IAAI,EAAE,SAAS,EAAE,CAAA;IACjB,QAAQ,EAAE,YAAY,CAAA;IACtB;;;;OAIG;IACH,cAAc,EAAE,MAAM,EAAE,CAAA;IACxB;;;;OAIG;IACH,OAAO,CAAC,EAAE,MAAM,EAAE,CAAA;IAClB,wEAAwE;IACxE,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,+CAA+C;IAC/C,kBAAkB,CAAC,EAAE,MAAM,CAAA;IAC3B,+EAA+E;IAC/E,IAAI,CAAC,EAAE,MAAM,CAAA;IACb;;;OAGG;IACH,SAAS,CAAC,EAAE,QAAQ,GAAG,MAAM,GAAG,KAAK,CAAA;CACtC;AAED,MAAM,WAAW,iBAAiB;IAChC,MAAM,EAAE,MAAM,CAAA;IACd,OAAO,EAAE,MAAM,CAAA;IACf,CAAC,EAAE,MAAM,CAAA;IACT,OAAO,EAAE,MAAM,CAAA;IACf,QAAQ,EAAE,MAAM,CAAA;IAChB,IAAI,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAA;IACnC;;;;;OAKG;IACH,OAAO,EAAE,cAAc,GAAG,aAAa,GAAG,YAAY,CAAA;CACvD;AAED,MAAM,WAAW,aAAa;IAC5B,MAAM,EAAE,MAAM,CAAA;IACd,uDAAuD;IACvD,WAAW,EAAE,MAAM,CAAA;IACnB,QAAQ,EAAE,MAAM,CAAA;IAChB,OAAO,EAAE,MAAM,CAAA;IACf,CAAC,EAAE,MAAM,CAAA;IACT,OAAO,EAAE,iBAAiB,CAAC,SAAS,CAAC,CAAA;CACtC;AAED,MAAM,WAAW,8BAA8B;IAC7C,KAAK,EAAE,iBAAiB,EAAE,CAAA;IAC1B,6DAA6D;IAC7D,MAAM,EAAE,aAAa,EAAE,CAAA;IACvB,aAAa,EAAE,MAAM,CAAA;IACrB,WAAW,EAAE,MAAM,CAAA;IACnB,oEAAoE;IACpE,kBAAkB,EAAE,MAAM,EAAE,CAAA;CAC7B;AAED,wBAAsB,wBAAwB,CAC5C,KAAK,EAAE,6BAA6B,GACnC,OAAO,CAAC,8BAA8B,CAAC,CAyGzC"}
@@ -0,0 +1,169 @@
1
+ /**
2
+ * Judge sentinel — eval trustworthiness as a continuously measured,
3
+ * alarmed trend.
4
+ *
5
+ * Judges are models; models change underneath us; calibration decays.
6
+ * This module composes three existing instruments into one loop:
7
+ *
8
+ * snapshot — adapters turn real instrument outputs (`calibrateJudge` /
9
+ * `calibrateJudgeContinuous`, `corpusInterRaterAgreement` /
10
+ * `continuousAgreement`, a rerun sentinel golden set) into
11
+ * `SentinelSnapshot`s
12
+ * series — `analyzeSeries` (src/series-convergence) classifies each
13
+ * judge×metric history as stabilized / drifting / noisy
14
+ * alarm — `judgeSentinelReport` turns trends + thresholds into named
15
+ * alarms; `evalHealthStamp` is the tiny object a campaign
16
+ * attaches to its verdicts
17
+ *
18
+ * Alarm conditions:
19
+ * - convergence state `drifting-down` on any tracked metric
20
+ * - irr below `minIrr`
21
+ * - calibrationKappa / sentinelPassRate dropped vs the series baseline
22
+ * beyond `maxKappaDrop` / `maxSentinelDrop`
23
+ * - judge model changed with no post-change golden-grounded snapshot
24
+ * (the silent-upgrade trap: judges agreeing with each other after a
25
+ * model swap proves nothing — only re-measuring against gold does)
26
+ * - newest snapshot older than `staleAfterDays` relative to `asOf`
27
+ *
28
+ * Wire-in: run the snapshot adapters from a post-campaign hook or a
29
+ * nightly job, append to a `SentinelStore`, then compute
30
+ * `judgeSentinelReport(await store.history(), { asOf })` with `asOf` =
31
+ * the campaign timestamp. Attach `evalHealthStamp(report)` to campaign
32
+ * verdicts: an alarmed sentinel means every downstream verdict carries
33
+ * an integrity warning until the judge is recalibrated.
34
+ *
35
+ * The module is clock-free — every timestamp (`at`, `asOf`) is
36
+ * caller-supplied ISO-8601, so reports are reproducible.
37
+ */
38
+ import type { CalibrationResult, CandidateScore, ContinuousAgreement, ContinuousCalibrationResult, GoldenItem } from '../judge-calibration';
39
+ import { type SeriesConvergenceOptions, type SeriesConvergenceResult } from '../series-convergence';
40
+ import type { CorpusAgreementReport } from '../statistics';
41
+ export declare const SENTINEL_METRIC_NAMES: readonly ["irr", "calibrationKappa", "sentinelPassRate"];
42
+ export type SentinelMetricName = (typeof SENTINEL_METRIC_NAMES)[number];
43
+ export interface SentinelMetrics {
44
+ /** Inter-rater reliability — ICC(2,1) from agreement instruments. */
45
+ irr?: number;
46
+ /** Weighted κ vs the human golden set (`calibrateJudge*`). */
47
+ calibrationKappa?: number;
48
+ /** Fraction of a fixed sentinel golden set the judge still scores within tolerance. */
49
+ sentinelPassRate?: number;
50
+ }
51
+ export interface SentinelSnapshot {
52
+ /** Caller-supplied ISO-8601 timestamp — the module never reads a clock. */
53
+ at: string;
54
+ judgeId: string;
55
+ /** Exact model identity behind the judge (pin the full version string). */
56
+ judgeModel: string;
57
+ /**
58
+ * Metrics measured at `at`. May be empty: an empty-metrics snapshot is a
59
+ * model-change marker — it records the new `judgeModel` immediately and
60
+ * the silent-upgrade alarm stays raised until a golden-grounded snapshot
61
+ * (calibrationKappa or sentinelPassRate) follows.
62
+ */
63
+ metrics: SentinelMetrics;
64
+ }
65
+ /** Identity + timestamp the caller supplies alongside an instrument output. */
66
+ export interface SnapshotMeta {
67
+ at: string;
68
+ judgeId: string;
69
+ judgeModel: string;
70
+ }
71
+ /** Shape guard for snapshots crossing a boundary (store append, JSONL load, report input). */
72
+ export declare function validateSentinelSnapshot(snapshot: SentinelSnapshot, source?: string): void;
73
+ /**
74
+ * From `calibrateJudge` / `calibrateJudgeContinuous` output. Prefers the
75
+ * un-rounded continuous κ_w when the report carries one (integer κ
76
+ * discards information for [0,1] judges).
77
+ */
78
+ export declare function snapshotFromCalibration(report: CalibrationResult | ContinuousCalibrationResult, meta: SnapshotMeta): SentinelSnapshot;
79
+ /**
80
+ * From `corpusInterRaterAgreement` (statistics.ts) or `continuousAgreement`
81
+ * (judge-calibration.ts) output. IRR = ICC(2,1) — the reliability
82
+ * coefficient both instruments compute. For ensemble-level agreement,
83
+ * `meta.judgeId` names the ensemble and `meta.judgeModel` pins its
84
+ * composition (e.g. a joined list of member model strings).
85
+ */
86
+ export declare function snapshotFromAgreement(report: CorpusAgreementReport | ContinuousAgreement, meta: SnapshotMeta): SentinelSnapshot;
87
+ export interface SentinelSetOptions {
88
+ /** Max |judge − human| for an item to count as passed. Default 0.1 (tuned for [0,1] scores). */
89
+ tolerance?: number;
90
+ }
91
+ /**
92
+ * From a rerun of a fixed sentinel golden set: the same `GoldenItem`s the
93
+ * judge was originally calibrated on, re-scored periodically. Pass rate =
94
+ * fraction of joined items where the judge stays within `tolerance` of the
95
+ * human score. Items the judge didn't score are excluded from the
96
+ * denominator; zero joined items is an error, not a 0% pass rate.
97
+ */
98
+ export declare function snapshotFromSentinelSet(scores: CandidateScore[], golden: GoldenItem[], meta: SnapshotMeta, options?: SentinelSetOptions): SentinelSnapshot;
99
+ export interface SentinelStore {
100
+ append(snapshot: SentinelSnapshot): Promise<void>;
101
+ /** All snapshots (optionally for one judge), in append order. */
102
+ history(judgeId?: string): Promise<SentinelSnapshot[]>;
103
+ }
104
+ export declare function inMemorySentinelStore(initial?: SentinelSnapshot[]): SentinelStore;
105
+ /**
106
+ * JSONL store — one snapshot per line, appended atomically per call.
107
+ * A corrupt or shape-invalid line is a loud error naming the file and
108
+ * line number: a sentinel history that silently drops records would
109
+ * defeat the drift detection it exists to provide.
110
+ */
111
+ export declare function fileSentinelStore(path: string): SentinelStore;
112
+ export interface SentinelThresholds {
113
+ /** Floor for irr — below it the judge ensemble is unreliable regardless of trend. Default 0.6. */
114
+ minIrr?: number;
115
+ /** Max tolerated calibrationKappa drop vs the series baseline. Default 0.15. */
116
+ maxKappaDrop?: number;
117
+ /** Max tolerated sentinelPassRate drop vs the series baseline. Default 0.1. */
118
+ maxSentinelDrop?: number;
119
+ /** Newest snapshot older than this (days, vs asOf) = the sentinel itself went blind. Default 30. */
120
+ staleAfterDays?: number;
121
+ }
122
+ export interface JudgeSentinelOptions {
123
+ /** Caller-supplied ISO timestamp staleness is measured against. */
124
+ asOf: string;
125
+ thresholds?: SentinelThresholds;
126
+ /** Forwarded to `analyzeSeries` (window, stableCv, driftRun). */
127
+ convergence?: SeriesConvergenceOptions;
128
+ }
129
+ export interface SentinelTrend {
130
+ judgeId: string;
131
+ metric: SentinelMetricName;
132
+ /** Verbatim `analyzeSeries` state for this judge×metric history. */
133
+ state: SeriesConvergenceResult['state'];
134
+ /** Newest value in the series. */
135
+ current: number;
136
+ /** Oldest value in the series — the reference the drop thresholds compare against. */
137
+ baseline: number;
138
+ /** current − baseline (signed; negative = decayed). */
139
+ drift: number;
140
+ alarmed: boolean;
141
+ reason?: string;
142
+ }
143
+ export interface SentinelReport {
144
+ perJudge: SentinelTrend[];
145
+ alarms: string[];
146
+ healthy: boolean;
147
+ /**
148
+ * `judgeId:metric` pairs with too few snapshots for the convergence
149
+ * machine. Named, not counted — a blind spot you can't see is a blind
150
+ * spot you won't fix. Floor/drop checks still apply to thin series, so
151
+ * insufficient history alone never masks an alarm.
152
+ */
153
+ insufficientHistory: string[];
154
+ }
155
+ export declare function judgeSentinelReport(history: SentinelSnapshot[], opts: JudgeSentinelOptions): SentinelReport;
156
+ export interface EvalHealthStamp {
157
+ healthy: boolean;
158
+ alarms: string[];
159
+ }
160
+ /**
161
+ * The object a campaign attaches to its verdicts. Compute it from the
162
+ * sentinel report in a post-campaign hook (or a nightly job feeding the
163
+ * next day's campaigns) and store it alongside the verdict payload.
164
+ * `healthy: false` means the judges that produced those verdicts have an
165
+ * unresolved drift / decay / silent-upgrade / staleness alarm — treat the
166
+ * verdicts as carrying an integrity warning until recalibration clears it.
167
+ */
168
+ export declare function evalHealthStamp(report: SentinelReport): EvalHealthStamp;
169
+ //# sourceMappingURL=sentinel.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"sentinel.d.ts","sourceRoot":"","sources":["../../src/meta-eval/sentinel.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoCG;AAGH,OAAO,KAAK,EACV,iBAAiB,EACjB,cAAc,EACd,mBAAmB,EACnB,2BAA2B,EAC3B,UAAU,EACX,MAAM,sBAAsB,CAAA;AAC7B,OAAO,EAEL,KAAK,wBAAwB,EAC7B,KAAK,uBAAuB,EAC7B,MAAM,uBAAuB,CAAA;AAC9B,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,eAAe,CAAA;AAE1D,eAAO,MAAM,qBAAqB,0DAA2D,CAAA;AAE7F,MAAM,MAAM,kBAAkB,GAAG,CAAC,OAAO,qBAAqB,CAAC,CAAC,MAAM,CAAC,CAAA;AAEvE,MAAM,WAAW,eAAe;IAC9B,qEAAqE;IACrE,GAAG,CAAC,EAAE,MAAM,CAAA;IACZ,8DAA8D;IAC9D,gBAAgB,CAAC,EAAE,MAAM,CAAA;IACzB,uFAAuF;IACvF,gBAAgB,CAAC,EAAE,MAAM,CAAA;CAC1B;AAED,MAAM,WAAW,gBAAgB;IAC/B,2EAA2E;IAC3E,EAAE,EAAE,MAAM,CAAA;IACV,OAAO,EAAE,MAAM,CAAA;IACf,2EAA2E;IAC3E,UAAU,EAAE,MAAM,CAAA;IAClB;;;;;OAKG;IACH,OAAO,EAAE,eAAe,CAAA;CACzB;AAED,+EAA+E;AAC/E,MAAM,WAAW,YAAY;IAC3B,EAAE,EAAE,MAAM,CAAA;IACV,OAAO,EAAE,MAAM,CAAA;IACf,UAAU,EAAE,MAAM,CAAA;CACnB;AAYD,8FAA8F;AAC9F,wBAAgB,wBAAwB,CAAC,QAAQ,EAAE,gBAAgB,EAAE,MAAM,SAAa,GAAG,IAAI,CAwB9F;AAID;;;;GAIG;AACH,wBAAgB,uBAAuB,CACrC,MAAM,EAAE,iBAAiB,GAAG,2BAA2B,EACvD,IAAI,EAAE,YAAY,GACjB,gBAAgB,CAUlB;AAED;;;;;;GAMG;AACH,wBAAgB,qBAAqB,CACnC,MAAM,EAAE,qBAAqB,GAAG,mBAAmB,EACnD,IAAI,EAAE,YAAY,GACjB,gBAAgB,CAUlB;AAED,MAAM,WAAW,kBAAkB;IACjC,gGAAgG;IAChG,SAAS,CAAC,EAAE,MAAM,CAAA;CACnB;AAED;;;;;;GAMG;AACH,wBAAgB,uBAAuB,CACrC,MAAM,EAAE,cAAc,EAAE,EACxB,MAAM,EAAE,UAAU,EAAE,EACpB,IAAI,EAAE,YAAY,EAClB,OAAO,GAAE,kBAAuB,GAC/B,gBAAgB,CAwClB;AAID,MAAM,WAAW,aAAa;IAC5B,MAAM,CAAC,QAAQ,EAAE,gBAAgB,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IACjD,iEAAiE;IACjE,OAAO,CAAC,OAAO,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,gBAAgB,EAAE,CAAC,CAAA;CACvD;AAED,wBAAgB,qBAAqB,CAAC,OAAO,GAAE,gBAAgB,EAAO,GAAG,aAAa,CAarF;AAED;;;;;GAKG;AACH,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,GAAG,aAAa,CAsC7D;AAID,MAAM,WAAW,kBAAkB;IACjC,kGAAkG;IAClG,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,gFAAgF;IAChF,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,+EAA+E;IAC/E,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,oGAAoG;IACpG,cAAc,CAAC,EAAE,MAAM,CAAA;CACxB;AAED,MAAM,WAAW,oBAAoB;IACnC,mEAAmE;IACnE,IAAI,EAAE,MAAM,CAAA;IACZ,UAAU,CAAC,EAAE,kBAAkB,CAAA;IAC/B,iEAAiE;IACjE,WAAW,CAAC,EAAE,wBAAwB,CAAA;CACvC;AAED,MAAM,WAAW,aAAa;IAC5B,OAAO,EAAE,MAAM,CAAA;IACf,MAAM,EAAE,kBAAkB,CAAA;IAC1B,oEAAoE;IACpE,KAAK,EAAE,uBAAuB,CAAC,OAAO,CAAC,CAAA;IACvC,kCAAkC;IAClC,OAAO,EAAE,MAAM,CAAA;IACf,sFAAsF;IACtF,QAAQ,EAAE,MAAM,CAAA;IAChB,uDAAuD;IACvD,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,EAAE,OAAO,CAAA;IAChB,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,cAAc;IAC7B,QAAQ,EAAE,aAAa,EAAE,CAAA;IACzB,MAAM,EAAE,MAAM,EAAE,CAAA;IAChB,OAAO,EAAE,OAAO,CAAA;IAChB;;;;;OAKG;IACH,mBAAmB,EAAE,MAAM,EAAE,CAAA;CAC9B;AAID,wBAAgB,mBAAmB,CACjC,OAAO,EAAE,gBAAgB,EAAE,EAC3B,IAAI,EAAE,oBAAoB,GACzB,cAAc,CAyGhB;AAID,MAAM,WAAW,eAAe;IAC9B,OAAO,EAAE,OAAO,CAAA;IAChB,MAAM,EAAE,MAAM,EAAE,CAAA;CACjB;AAED;;;;;;;GAOG;AACH,wBAAgB,eAAe,CAAC,MAAM,EAAE,cAAc,GAAG,eAAe,CAEvE"}
@@ -0,0 +1,63 @@
1
+ import type { ProductClient } from './client';
2
+ import type { DriverState, TurnMetrics } from './types';
3
+ interface TokenPrice {
4
+ input: number;
5
+ output: number;
6
+ }
7
+ /** Per-1K token pricing for exact model ids. */
8
+ export declare const MODEL_PRICING: Record<string, TokenPrice>;
9
+ /** Resolve pricing for a model id: exact table, then family fallback.
10
+ * Returns null when the id matches nothing (caller decides — never a
11
+ * silent-zero masquerading as a real $0 cost). */
12
+ export declare function resolveModelPricing(model: string): TokenPrice | null;
13
+ /** True when `model` has known pricing (exact or family). Lets cost-aware
14
+ * callers distinguish a real $0 from an unpriced model. */
15
+ export declare function isModelPriced(model: string): boolean;
16
+ /** Estimate token count from string length (chars / 4 approximation) */
17
+ export declare function estimateTokens(text: string): number;
18
+ /** Calculate cost in USD from token counts and model. Unknown models warn
19
+ * once (not a silent zero) and return 0 so callers that ignore pricing keep
20
+ * working; cost-sensitive callers should gate on {@link isModelPriced}. */
21
+ export declare function estimateCost(inputTokens: number, outputTokens: number, model: string): number;
22
+ /**
23
+ * TokenCounter — accumulates token usage and cost across turns.
24
+ */
25
+ export declare class TokenCounter {
26
+ private totalInput;
27
+ private totalOutput;
28
+ private totalCost;
29
+ private model;
30
+ constructor(model?: string);
31
+ /** Record tokens for a turn, returns per-turn cost */
32
+ record(inputTokens: number, outputTokens: number): number;
33
+ /** Estimate and record from raw text */
34
+ recordFromText(inputText: string, outputText: string): {
35
+ inputTokens: number;
36
+ outputTokens: number;
37
+ cost: number;
38
+ };
39
+ getTotalInput(): number;
40
+ getTotalOutput(): number;
41
+ getTotalCost(): number;
42
+ }
43
+ /**
44
+ * MetricsCollector — collects per-turn metrics from the product.
45
+ *
46
+ * After each turn, queries the product's APIs to measure state changes.
47
+ */
48
+ export declare class MetricsCollector {
49
+ private client;
50
+ private workspaceId;
51
+ private metrics;
52
+ constructor(client: ProductClient, workspaceId: string);
53
+ /** Collect metrics after a turn completes */
54
+ collect(turn: number, responseLatencyMs: number, responseChars: number, codeBlocksProduced: number, blocksExtracted: number, completionCriteriaMet: number, completionCriteriaTotal: number, qualityScore?: number, inputTokens?: number, outputTokens?: number, estimatedCostUsd?: number): Promise<TurnMetrics>;
55
+ /** Get current product state */
56
+ getState(): Promise<DriverState>;
57
+ /** Get all collected metrics */
58
+ getMetrics(): TurnMetrics[];
59
+ /** Get convergence curve (completion% over turns) */
60
+ getConvergenceCurve(): number[];
61
+ }
62
+ export {};
63
+ //# sourceMappingURL=metrics.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"metrics.d.ts","sourceRoot":"","sources":["../src/metrics.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,UAAU,CAAA;AAC7C,OAAO,KAAK,EAAE,WAAW,EAAE,WAAW,EAAE,MAAM,SAAS,CAAA;AAEvD,UAAU,UAAU;IAClB,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,EAAE,MAAM,CAAA;CACf;AAED,gDAAgD;AAChD,eAAO,MAAM,aAAa,EAAE,MAAM,CAAC,MAAM,EAAE,UAAU,CAOpD,CAAA;AAgCD;;mDAEmD;AACnD,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,GAAG,UAAU,GAAG,IAAI,CAQpE;AAED;4DAC4D;AAC5D,wBAAgB,aAAa,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAEpD;AAID,wEAAwE;AACxE,wBAAgB,cAAc,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAEnD;AAED;;4EAE4E;AAC5E,wBAAgB,YAAY,CAAC,WAAW,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,MAAM,CAa7F;AAED;;GAEG;AACH,qBAAa,YAAY;IACvB,OAAO,CAAC,UAAU,CAAI;IACtB,OAAO,CAAC,WAAW,CAAI;IACvB,OAAO,CAAC,SAAS,CAAI;IACrB,OAAO,CAAC,KAAK,CAAQ;gBAET,KAAK,SAAW;IAI5B,sDAAsD;IACtD,MAAM,CAAC,WAAW,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,GAAG,MAAM;IAQzD,wCAAwC;IACxC,cAAc,CACZ,SAAS,EAAE,MAAM,EACjB,UAAU,EAAE,MAAM,GACjB;QAAE,WAAW,EAAE,MAAM,CAAC;QAAC,YAAY,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE;IAO9D,aAAa,IAAI,MAAM;IAGvB,cAAc,IAAI,MAAM;IAGxB,YAAY,IAAI,MAAM;CAGvB;AAED;;;;GAIG;AACH,qBAAa,gBAAgB;IAC3B,OAAO,CAAC,MAAM,CAAe;IAC7B,OAAO,CAAC,WAAW,CAAQ;IAC3B,OAAO,CAAC,OAAO,CAAoB;gBACvB,MAAM,EAAE,aAAa,EAAE,WAAW,EAAE,MAAM;IAKtD,6CAA6C;IACvC,OAAO,CACX,IAAI,EAAE,MAAM,EACZ,iBAAiB,EAAE,MAAM,EACzB,aAAa,EAAE,MAAM,EACrB,kBAAkB,EAAE,MAAM,EAC1B,eAAe,EAAE,MAAM,EACvB,qBAAqB,EAAE,MAAM,EAC7B,uBAAuB,EAAE,MAAM,EAC/B,YAAY,CAAC,EAAE,MAAM,EACrB,WAAW,SAAI,EACf,YAAY,SAAI,EAChB,gBAAgB,SAAI,GACnB,OAAO,CAAC,WAAW,CAAC;IA2BvB,gCAAgC;IAC1B,QAAQ,IAAI,OAAO,CAAC,WAAW,CAAC;IAsBtC,gCAAgC;IAChC,UAAU,IAAI,WAAW,EAAE;IAI3B,qDAAqD;IACrD,mBAAmB,IAAI,MAAM,EAAE;CAGhC"}
@@ -0,0 +1,71 @@
1
+ /**
2
+ * ModelSeats — the program's model seating chart.
3
+ *
4
+ * One object names which model fills each role in an eval program: the worker
5
+ * under evaluation, the judge panel, the analyst, the reflection/driver model,
6
+ * and the verifier. Re-tiering an entire program (economy ↔ frontier) is one
7
+ * swapped object instead of a hunt through call sites.
8
+ *
9
+ * Wiring points — consumers thread seats; this module implements none of them
10
+ * (those files belong to other surfaces):
11
+ * - `judges` → `ensembleJudge({ models: seats.judges, … })` (src/judge-panel.ts)
12
+ * and the `JudgeConfig`s handed to `makeEvalTools({ judges })`
13
+ * (src/eval-tools.ts).
14
+ * - `reflection` → `selfImprove({ llm: { model: seats.reflection } })` — the
15
+ * `gepaProposer` reflection model (src/contract/self-improve.ts);
16
+ * same seat for any custom `SurfaceProposer`'s LLM.
17
+ * - `worker` → the dispatch model the agent itself calls — the model an
18
+ * `AgentProfile` declares.
19
+ * - `analyst` → the LLM behind `analyzeRuns` / analyst-registry kinds.
20
+ * - `verifier` → completion-verifier / objective-checker model.
21
+ * - campaign cells thread `judges` + driver models the same way; that wiring
22
+ * lands with the campaign surface, not here.
23
+ *
24
+ * `resolveSeat` is the only read path: an unset seat with no explicit fallback
25
+ * throws — a model id is a budget decision, never a silent default.
26
+ */
27
+ import { ConfigError } from './errors';
28
+ export interface ModelSeats {
29
+ /** The model under evaluation — what the agent itself dispatches with. */
30
+ worker?: string;
31
+ /** Judge-panel model ids — thread into `ensembleJudge({ models })`. */
32
+ judges?: string[];
33
+ /** Analyst model — `analyzeRuns` / analyst-registry LLM calls. */
34
+ analyst?: string;
35
+ /** Reflection/proposer model — `gepaProposer` mutation proposals. */
36
+ reflection?: string;
37
+ /** Verifier model — completion/objective checking. */
38
+ verifier?: string;
39
+ }
40
+ export type SeatName = keyof ModelSeats;
41
+ export type SeatPresetName = keyof typeof seatPresets;
42
+ /**
43
+ * Tier presets — plain data, swap or spread freely.
44
+ *
45
+ * `economy` uses the fleet-policy ids: every id resolves through the
46
+ * substrate's family pricing (no costUnknown axis) and the judge trio spans
47
+ * three provider families (moonshot / deepseek / openai), so it passes
48
+ * `assertCrossFamily` as-is.
49
+ *
50
+ * `frontier` is deliberately EMPTY: entitled frontier ids vary per router
51
+ * account, and a hardcoded claude/gpt-5 id 401s on keys that lack it. Supply
52
+ * your own: `{ ...seatPresets.frontier, worker: '<your-frontier-id>', … }` —
53
+ * `resolveSeat` throws on every seat you haven't filled.
54
+ */
55
+ export declare const seatPresets: Record<'economy' | 'frontier', ModelSeats>;
56
+ /** Thrown by `resolveSeat` when a seat is unset and no fallback was given. */
57
+ export declare class SeatUnsetError extends ConfigError {
58
+ readonly seat: SeatName;
59
+ constructor(seat: SeatName);
60
+ }
61
+ /**
62
+ * Read one seat. Blank strings and empty arrays count as unset (env-var
63
+ * plumbing produces them); malformed values (non-string seat, non-array or
64
+ * blank-entry `judges`) throw `ValidationError`. When the seat is unset, an
65
+ * explicit `fallback` is returned (`[fallback]` for `judges` — a one-model
66
+ * panel); without one, `SeatUnsetError`.
67
+ */
68
+ export declare function resolveSeat(seats: ModelSeats, seat: 'judges', fallback?: string): string[];
69
+ export declare function resolveSeat(seats: ModelSeats, seat: Exclude<SeatName, 'judges'>, fallback?: string): string;
70
+ export declare function resolveSeat(seats: ModelSeats, seat: SeatName, fallback?: string): string | string[];
71
+ //# sourceMappingURL=model-seats.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"model-seats.d.ts","sourceRoot":"","sources":["../src/model-seats.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AAEH,OAAO,EAAE,WAAW,EAAmB,MAAM,UAAU,CAAA;AAEvD,MAAM,WAAW,UAAU;IACzB,0EAA0E;IAC1E,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,uEAAuE;IACvE,MAAM,CAAC,EAAE,MAAM,EAAE,CAAA;IACjB,kEAAkE;IAClE,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,qEAAqE;IACrE,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,sDAAsD;IACtD,QAAQ,CAAC,EAAE,MAAM,CAAA;CAClB;AAED,MAAM,MAAM,QAAQ,GAAG,MAAM,UAAU,CAAA;AAEvC,MAAM,MAAM,cAAc,GAAG,MAAM,OAAO,WAAW,CAAA;AAErD;;;;;;;;;;;;GAYG;AACH,eAAO,MAAM,WAAW,EAAE,MAAM,CAAC,SAAS,GAAG,UAAU,EAAE,UAAU,CASlE,CAAA;AAED,8EAA8E;AAC9E,qBAAa,cAAe,SAAQ,WAAW;aACjB,IAAI,EAAE,QAAQ;gBAAd,IAAI,EAAE,QAAQ;CAM3C;AAED;;;;;;GAMG;AACH,wBAAgB,WAAW,CAAC,KAAK,EAAE,UAAU,EAAE,IAAI,EAAE,QAAQ,EAAE,QAAQ,CAAC,EAAE,MAAM,GAAG,MAAM,EAAE,CAAA;AAC3F,wBAAgB,WAAW,CACzB,KAAK,EAAE,UAAU,EACjB,IAAI,EAAE,OAAO,CAAC,QAAQ,EAAE,QAAQ,CAAC,EACjC,QAAQ,CAAC,EAAE,MAAM,GAChB,MAAM,CAAA;AACT,wBAAgB,WAAW,CAAC,KAAK,EAAE,UAAU,EAAE,IAAI,EAAE,QAAQ,EAAE,QAAQ,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,MAAM,EAAE,CAAA"}
@@ -0,0 +1,2 @@
1
+ export {};
2
+ //# sourceMappingURL=model-seats.test.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"model-seats.test.d.ts","sourceRoot":"","sources":["../src/model-seats.test.ts"],"names":[],"mappings":""}