@tangle-network/agent-eval 0.200.1 → 0.202.0

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 (241) hide show
  1. package/CHANGELOG.md +9 -0
  2. package/dist/adapters/http.d.ts +8 -9
  3. package/dist/adapters/http.d.ts.map +1 -1
  4. package/dist/adapters/http.js +2 -2
  5. package/dist/adapters/http.js.map +1 -1
  6. package/dist/agent-profile-DDfIjdyS.d.ts.map +1 -1
  7. package/dist/analyst/index.d.ts +76 -76
  8. package/dist/analyst/index.d.ts.map +1 -1
  9. package/dist/analyst/index.js +6 -6
  10. package/dist/{analyze-runs-DFbp2gB8.js → analyze-runs-BrIibNd0.js} +47 -42
  11. package/dist/{analyze-runs-DFbp2gB8.js.map → analyze-runs-BrIibNd0.js.map} +1 -1
  12. package/dist/{ast-CP9ae9B0.js → ast-DJ6Bc7ab.js} +5 -4
  13. package/dist/{ast-CP9ae9B0.js.map → ast-DJ6Bc7ab.js.map} +1 -1
  14. package/dist/authenticity/index.d.ts +14 -15
  15. package/dist/authenticity/index.d.ts.map +1 -1
  16. package/dist/{baseline-BC-eBZ7U.js → baseline-CVKsfWTg.js} +2 -2
  17. package/dist/{baseline-BC-eBZ7U.js.map → baseline-CVKsfWTg.js.map} +1 -1
  18. package/dist/{benchmark-BeQJyrRC.js → benchmark-BywPeAUo.js} +4 -3
  19. package/dist/{benchmark-BeQJyrRC.js.map → benchmark-BywPeAUo.js.map} +1 -1
  20. package/dist/{benchmark-command-EBmuMUBo.js → benchmark-command-DFs_i62N.js} +24 -23
  21. package/dist/{benchmark-command-EBmuMUBo.js.map → benchmark-command-DFs_i62N.js.map} +1 -1
  22. package/dist/benchmarks/index.d.ts +13 -13
  23. package/dist/benchmarks/index.d.ts.map +1 -1
  24. package/dist/benchmarks/index.js +3 -3
  25. package/dist/{bounded-process-BBZob7vl.js → bounded-process-Cj9YjXbE.js} +3 -3
  26. package/dist/{bounded-process-BBZob7vl.js.map → bounded-process-Cj9YjXbE.js.map} +1 -1
  27. package/dist/builder-eval/index.d.ts +16 -17
  28. package/dist/builder-eval/index.d.ts.map +1 -1
  29. package/dist/builder-eval/index.js +12 -10
  30. package/dist/builder-eval/index.js.map +1 -1
  31. package/dist/campaign/index.d.ts +2 -2
  32. package/dist/campaign/index.js +12 -12
  33. package/dist/{campaign-L391bmir.js → campaign-XXmOU0R0.js} +265 -27
  34. package/dist/campaign-XXmOU0R0.js.map +1 -0
  35. package/dist/{campaign-evidence-jL_U7Ikv.js → campaign-evidence-Dt_rWPk2.js} +4 -4
  36. package/dist/{campaign-evidence-jL_U7Ikv.js.map → campaign-evidence-Dt_rWPk2.js.map} +1 -1
  37. package/dist/{chat-client-CJORSG_J.js → chat-client-DUhWNQQS.js} +5 -4
  38. package/dist/{chat-client-CJORSG_J.js.map → chat-client-DUhWNQQS.js.map} +1 -1
  39. package/dist/{chat-json-call-DnlDjpco.js → chat-json-call-Dx_sHzqm.js} +2 -2
  40. package/dist/{chat-json-call-DnlDjpco.js.map → chat-json-call-Dx_sHzqm.js.map} +1 -1
  41. package/dist/{claim-V9Zqbq0J.js → claim-BXcZw-jo.js} +2 -2
  42. package/dist/{claim-V9Zqbq0J.js.map → claim-BXcZw-jo.js.map} +1 -1
  43. package/dist/cli.js +12 -11
  44. package/dist/cli.js.map +1 -1
  45. package/dist/contract/index.d.ts +3 -3
  46. package/dist/contract/index.js +22 -19
  47. package/dist/contract/index.js.map +1 -1
  48. package/dist/{default-registry-xMLX160E.js → default-registry-D4qXx2aG.js} +2 -2
  49. package/dist/{default-registry-xMLX160E.js.map → default-registry-D4qXx2aG.js.map} +1 -1
  50. package/dist/{define-agent-eval-7q_UwsTt.js → define-agent-eval-CJN9dTQw.js} +10 -9
  51. package/dist/{define-agent-eval-7q_UwsTt.js.map → define-agent-eval-CJN9dTQw.js.map} +1 -1
  52. package/dist/{define-agent-eval-1YuvJpOs.d.ts → define-agent-eval-DQXLPqWc.d.ts} +2 -2
  53. package/dist/define-agent-eval-DQXLPqWc.d.ts.map +1 -0
  54. package/dist/{descriptive-1V17A-qa.js → descriptive-D9LUFk1-.js} +2 -2
  55. package/dist/{descriptive-1V17A-qa.js.map → descriptive-D9LUFk1-.js.map} +1 -1
  56. package/dist/diagnosis/index.d.ts +7 -7
  57. package/dist/diagnosis/index.d.ts.map +1 -1
  58. package/dist/diagnosis/index.js +2 -2
  59. package/dist/diagnosis/index.js.map +1 -1
  60. package/dist/{dspy-rlm-engine-B5T43BkZ.js → dspy-rlm-engine-US8YsYGq.js} +5 -5
  61. package/dist/{dspy-rlm-engine-B5T43BkZ.js.map → dspy-rlm-engine-US8YsYGq.js.map} +1 -1
  62. package/dist/{eval-campaign-GapmPHIA.js → eval-campaign-ha_aWtk7.js} +14 -12
  63. package/dist/{eval-campaign-GapmPHIA.js.map → eval-campaign-ha_aWtk7.js.map} +1 -1
  64. package/dist/evaluation.d.ts +9 -10
  65. package/dist/evaluation.d.ts.map +1 -1
  66. package/dist/experiment/index.d.ts +29 -29
  67. package/dist/experiment/index.d.ts.map +1 -1
  68. package/dist/experiment/index.js +13 -12
  69. package/dist/experiment/index.js.map +1 -1
  70. package/dist/{exporters-DEcXbvgO.js → exporters-Coiv-qO2.js} +2 -2
  71. package/dist/{exporters-DEcXbvgO.js.map → exporters-Coiv-qO2.js.map} +1 -1
  72. package/dist/{external-optimizer-process-CEXcCKqd.js → external-optimizer-process-KLcw0Vnc.js} +2 -2
  73. package/dist/{external-optimizer-process-CEXcCKqd.js.map → external-optimizer-process-KLcw0Vnc.js.map} +1 -1
  74. package/dist/{external-optimizer-subprocess-laM1-UDV.js → external-optimizer-subprocess-CkyUUV7h.js} +4 -4
  75. package/dist/{external-optimizer-subprocess-laM1-UDV.js.map → external-optimizer-subprocess-CkyUUV7h.js.map} +1 -1
  76. package/dist/fuzz.d.ts +15 -15
  77. package/dist/fuzz.d.ts.map +1 -1
  78. package/dist/{hf-dataset-CRqtSjnR.js → hf-dataset-kjIfIK_e.js} +3 -3
  79. package/dist/{hf-dataset-CRqtSjnR.js.map → hf-dataset-kjIfIK_e.js.map} +1 -1
  80. package/dist/hosted/index.d.ts +8 -8
  81. package/dist/hosted/index.d.ts.map +1 -1
  82. package/dist/hosted/index.js +1 -1
  83. package/dist/{index-EkRNsLr4.d.ts → index-D1ZX28rV.d.ts} +3 -3
  84. package/dist/{index-EkRNsLr4.d.ts.map → index-D1ZX28rV.d.ts.map} +1 -1
  85. package/dist/{index-D4pC7Tvd2.d.ts → index-D4pC7Tvd.d.ts} +1 -1
  86. package/dist/index-D4pC7Tvd.d.ts.map +1 -0
  87. package/dist/{index-DNgf5gyG2.d.ts → index-DNgf5gyG.d.ts} +1 -1
  88. package/dist/index-DNgf5gyG.d.ts.map +1 -0
  89. package/dist/{index-DKmrHA02.d.ts → index-giYkbitB.d.ts} +47 -2
  90. package/dist/index-giYkbitB.d.ts.map +1 -0
  91. package/dist/index.d.ts +126 -126
  92. package/dist/index.d.ts.map +1 -1
  93. package/dist/index.js +52 -42
  94. package/dist/index.js.map +1 -1
  95. package/dist/{integrity-BFP4kjGj.js → integrity-BvnVARVl.js} +30 -29
  96. package/dist/{integrity-BFP4kjGj.js.map → integrity-BvnVARVl.js.map} +1 -1
  97. package/dist/{integrity-BvHANvnp.js → integrity-uwK7XC-8.js} +2 -2
  98. package/dist/{integrity-BvHANvnp.js.map → integrity-uwK7XC-8.js.map} +1 -1
  99. package/dist/{internal-BMFSR8Ns.js → internal-BOe3zmjQ.js} +2 -2
  100. package/dist/{internal-BMFSR8Ns.js.map → internal-BOe3zmjQ.js.map} +1 -1
  101. package/dist/{jev-review-Cu9xtIu4.js → jev-review-DZEjDzlF.js} +7 -5
  102. package/dist/{jev-review-Cu9xtIu4.js.map → jev-review-DZEjDzlF.js.map} +1 -1
  103. package/dist/jev.d.ts +10 -10
  104. package/dist/jev.d.ts.map +1 -1
  105. package/dist/jev.js +2 -2
  106. package/dist/{judge-calibration-DYmaBtJr.js → judge-calibration-C5zgtiJT.js} +2 -2
  107. package/dist/{judge-calibration-DYmaBtJr.js.map → judge-calibration-C5zgtiJT.js.map} +1 -1
  108. package/dist/kind-factory-CdPPHtGZ.js.map +1 -1
  109. package/dist/{llm-client-c_bKHTXz.js → llm-client-Bhqoglar.js} +6 -4
  110. package/dist/{llm-client-c_bKHTXz.js.map → llm-client-Bhqoglar.js.map} +1 -1
  111. package/dist/{llm-judge-CL6f3utD.js → llm-judge-C-0P1WiT.js} +26 -21
  112. package/dist/{llm-judge-CL6f3utD.js.map → llm-judge-C-0P1WiT.js.map} +1 -1
  113. package/dist/matrix/index.d.ts +1 -1
  114. package/dist/matrix/index.js +1 -1
  115. package/dist/{matrix-_NLSWsKM.d.ts → matrix--IZKROxD.d.ts} +2 -2
  116. package/dist/{matrix-_NLSWsKM.d.ts.map → matrix--IZKROxD.d.ts.map} +1 -1
  117. package/dist/{matrix-pWinRMPl.js → matrix-DnlHn5x1.js} +8 -6
  118. package/dist/{matrix-pWinRMPl.js.map → matrix-DnlHn5x1.js.map} +1 -1
  119. package/dist/meta-eval/index.d.ts +102 -59
  120. package/dist/meta-eval/index.d.ts.map +1 -1
  121. package/dist/meta-eval/index.js +89 -25
  122. package/dist/meta-eval/index.js.map +1 -1
  123. package/dist/{mint-iCD-W43s.js → mint-vlRckJEX.js} +2 -2
  124. package/dist/{mint-iCD-W43s.js.map → mint-vlRckJEX.js.map} +1 -1
  125. package/dist/multishot/golden/index.d.ts +23 -23
  126. package/dist/multishot/golden/index.d.ts.map +1 -1
  127. package/dist/multishot/golden/index.js +1 -1
  128. package/dist/multishot/index.d.ts +9 -9
  129. package/dist/multishot/index.d.ts.map +1 -1
  130. package/dist/multishot/index.js +1 -1
  131. package/dist/openapi.json +1 -1
  132. package/dist/{otel-spans-CQmSxtcS.js → otel-spans-C8YIJWy1.js} +9 -4
  133. package/dist/{otel-spans-CQmSxtcS.js.map → otel-spans-C8YIJWy1.js.map} +1 -1
  134. package/dist/{outcome-store-BXlkwMPR.js → outcome-store-DUcMj-3D.js} +2 -2
  135. package/dist/{outcome-store-BXlkwMPR.js.map → outcome-store-DUcMj-3D.js.map} +1 -1
  136. package/dist/package-version-D7lQHt_-.js.map +1 -1
  137. package/dist/{paired-arms-DKZFVPt8.js → paired-arms-DpS_h8ON.js} +8 -5
  138. package/dist/{paired-arms-DKZFVPt8.js.map → paired-arms-DpS_h8ON.js.map} +1 -1
  139. package/dist/{paired-binary-fzkwENI4.js → paired-binary-OMBVRtc-.js} +2 -2
  140. package/dist/{paired-binary-fzkwENI4.js.map → paired-binary-OMBVRtc-.js.map} +1 -1
  141. package/dist/{paired-promotion-decision-D5B1Cxc2.js → paired-promotion-decision-DhrmjdV8.js} +5 -4
  142. package/dist/{paired-promotion-decision-D5B1Cxc2.js.map → paired-promotion-decision-DhrmjdV8.js.map} +1 -1
  143. package/dist/{paired-tests-C8iCsioC.js → paired-tests-Ccr16gcW.js} +3 -3
  144. package/dist/{paired-tests-C8iCsioC.js.map → paired-tests-Ccr16gcW.js.map} +1 -1
  145. package/dist/pipelines/index.d.ts +19 -19
  146. package/dist/pipelines/index.d.ts.map +1 -1
  147. package/dist/pipelines/index.js +5 -4
  148. package/dist/pipelines/index.js.map +1 -1
  149. package/dist/{power-and-mde-K_CmO3dc.js → power-and-mde-bie6i965.js} +9 -6
  150. package/dist/{power-and-mde-K_CmO3dc.js.map → power-and-mde-bie6i965.js.map} +1 -1
  151. package/dist/{produced-state-DZMW8cuF.js → produced-state-CR2tAPLf.js} +8 -5
  152. package/dist/{produced-state-DZMW8cuF.js.map → produced-state-CR2tAPLf.js.map} +1 -1
  153. package/dist/{promotion-policy-C6FAJvmZ.js → promotion-policy-BI1RxzwD.js} +4 -4
  154. package/dist/{promotion-policy-C6FAJvmZ.js.map → promotion-policy-BI1RxzwD.js.map} +1 -1
  155. package/dist/{raw-provider-sink-CXuOhLvT.js → raw-provider-sink-rB2DYxr6.js} +2 -2
  156. package/dist/{raw-provider-sink-CXuOhLvT.js.map → raw-provider-sink-rB2DYxr6.js.map} +1 -1
  157. package/dist/{release-confidence-XQm_xjhR.js → release-confidence-CiFg1EPq.js} +3 -2
  158. package/dist/{release-confidence-XQm_xjhR.js.map → release-confidence-CiFg1EPq.js.map} +1 -1
  159. package/dist/{report-command-DKJ_yAIj.js → report-command-DzPpFgjB.js} +20 -18
  160. package/dist/{report-command-DKJ_yAIj.js.map → report-command-DzPpFgjB.js.map} +1 -1
  161. package/dist/reporting.js +5 -5
  162. package/dist/{reward-hacking-D0XwhVWE.js → reward-hacking-C-a__ShZ.js} +5 -3
  163. package/dist/{reward-hacking-D0XwhVWE.js.map → reward-hacking-C-a__ShZ.js.map} +1 -1
  164. package/dist/rl.d.ts +186 -186
  165. package/dist/rl.d.ts.map +1 -1
  166. package/dist/rl.js +21 -17
  167. package/dist/rl.js.map +1 -1
  168. package/dist/rollout/index.d.ts +1 -1
  169. package/dist/rollout/index.js +5 -5
  170. package/dist/{rollout-B-XBReTC.js → rollout-BUQixCoc.js} +5 -5
  171. package/dist/{rollout-B-XBReTC.js.map → rollout-BUQixCoc.js.map} +1 -1
  172. package/dist/{rubric-predictive-validity-CCK-1B7w.js → rubric-predictive-validity-B0Pg-3I5.js} +3 -3
  173. package/dist/{rubric-predictive-validity-CCK-1B7w.js.map → rubric-predictive-validity-B0Pg-3I5.js.map} +1 -1
  174. package/dist/{run-record-tNnlQIrD.js → run-record-BB290Y9A.js} +5 -3
  175. package/dist/{run-record-tNnlQIrD.js.map → run-record-BB290Y9A.js.map} +1 -1
  176. package/dist/{schema-C1aaAxTf.js → schema-DAMp77zB.js} +24 -20
  177. package/dist/{schema-C1aaAxTf.js.map → schema-DAMp77zB.js.map} +1 -1
  178. package/dist/search/index.d.ts +34 -34
  179. package/dist/search/index.d.ts.map +1 -1
  180. package/dist/search/index.js +2 -2
  181. package/dist/{search-kernel-AVgIjXaI.js → search-kernel-Dpniloks.js} +33 -26
  182. package/dist/{search-kernel-AVgIjXaI.js.map → search-kernel-Dpniloks.js.map} +1 -1
  183. package/dist/{search-ledger-CL0-mXQ8.js → search-ledger-CXQPwdWz.js} +4 -3
  184. package/dist/{search-ledger-CL0-mXQ8.js.map → search-ledger-CXQPwdWz.js.map} +1 -1
  185. package/dist/{search-shipper-Cs6K4km9.js → search-shipper-DhhofkDX.js} +3 -6
  186. package/dist/{search-shipper-Cs6K4km9.js.map → search-shipper-DhhofkDX.js.map} +1 -1
  187. package/dist/{search-summary-DvWk0hb3.js → search-summary-C1nAHgkB.js} +4 -4
  188. package/dist/{search-summary-DvWk0hb3.js.map → search-summary-C1nAHgkB.js.map} +1 -1
  189. package/dist/{semantic-concept-judge-2SE_kXGF.js → semantic-concept-judge-BcjYg37z.js} +4 -4
  190. package/dist/{semantic-concept-judge-2SE_kXGF.js.map → semantic-concept-judge-BcjYg37z.js.map} +1 -1
  191. package/dist/{sequential-Yz3OSskl.js → sequential-DWn_np6P.js} +2 -3
  192. package/dist/{sequential-Yz3OSskl.js.map → sequential-DWn_np6P.js.map} +1 -1
  193. package/dist/{server-D56r1lhp.js → server-B4jMhqQn.js} +2 -2
  194. package/dist/{server-D56r1lhp.js.map → server-B4jMhqQn.js.map} +1 -1
  195. package/dist/{skillopt-optimization-method-B2syNik0.js → skillopt-optimization-method-Bls9t-G_.js} +9 -9
  196. package/dist/{skillopt-optimization-method-B2syNik0.js.map → skillopt-optimization-method-Bls9t-G_.js.map} +1 -1
  197. package/dist/{statistical-heldout-BAFDCo07.js → statistical-heldout-CNLOvRq_.js} +3 -3
  198. package/dist/{statistical-heldout-BAFDCo07.js.map → statistical-heldout-CNLOvRq_.js.map} +1 -1
  199. package/dist/statistical-heldout-HswMvzbP.d.ts.map +1 -1
  200. package/dist/{store-DNe_Uv1Q.js → store-BhNuLdJw.js} +3 -5
  201. package/dist/{store-DNe_Uv1Q.js.map → store-BhNuLdJw.js.map} +1 -1
  202. package/dist/store-C4avFjIi.d.ts.map +1 -1
  203. package/dist/{store-otlp-BPowgPm6.js → store-otlp-yd9lDziN.js} +2 -2
  204. package/dist/{store-otlp-BPowgPm6.js.map → store-otlp-yd9lDziN.js.map} +1 -1
  205. package/dist/{store-tool-spans-BaGakFc2.js → store-tool-spans-Biio4_Nw.js} +4 -4
  206. package/dist/{store-tool-spans-BaGakFc2.js.map → store-tool-spans-Biio4_Nw.js.map} +1 -1
  207. package/dist/storyboard/index.d.ts +20 -21
  208. package/dist/storyboard/index.d.ts.map +1 -1
  209. package/dist/{student-t-BA-Uy51p.js → student-t-BBZYka0E.js} +4 -3
  210. package/dist/{student-t-BA-Uy51p.js.map → student-t-BBZYka0E.js.map} +1 -1
  211. package/dist/{summary-report-BeYnvddy.js → summary-report-DLzptLSA.js} +8 -7
  212. package/dist/{summary-report-BeYnvddy.js.map → summary-report-DLzptLSA.js.map} +1 -1
  213. package/dist/supervisor-run/index.d.ts +25 -25
  214. package/dist/supervisor-run/index.d.ts.map +1 -1
  215. package/dist/supervisor-run/index.js +2 -2
  216. package/dist/{tool-waste-CJDfxH45.js → tool-waste-B-u3rzCz.js} +11 -6
  217. package/dist/{tool-waste-CJDfxH45.js.map → tool-waste-B-u3rzCz.js.map} +1 -1
  218. package/dist/trace-repair/index.d.ts +96 -96
  219. package/dist/trace-repair/index.d.ts.map +1 -1
  220. package/dist/trace-repair/index.js +1 -1
  221. package/dist/trace-repair/index.js.map +1 -1
  222. package/dist/traces.d.ts +29 -29
  223. package/dist/traces.d.ts.map +1 -1
  224. package/dist/traces.js +9 -7
  225. package/dist/traces.js.map +1 -1
  226. package/dist/trajectory-replay/index.d.ts +35 -35
  227. package/dist/trajectory-replay/index.d.ts.map +1 -1
  228. package/dist/trajectory-replay/index.js +5 -5
  229. package/dist/trajectory-replay/index.js.map +1 -1
  230. package/dist/{tree-Ki_gW7z8.js → tree-D48IJUK0.js} +39 -31
  231. package/dist/{tree-Ki_gW7z8.js.map → tree-D48IJUK0.js.map} +1 -1
  232. package/dist/wire/index.d.ts +48 -48
  233. package/dist/wire/index.d.ts.map +1 -1
  234. package/dist/wire/index.js +1 -1
  235. package/docs/evaluator-construction.md +71 -0
  236. package/package.json +18 -36
  237. package/dist/campaign-L391bmir.js.map +0 -1
  238. package/dist/define-agent-eval-1YuvJpOs.d.ts.map +0 -1
  239. package/dist/index-D4pC7Tvd2.d.ts.map +0 -1
  240. package/dist/index-DKmrHA02.d.ts.map +0 -1
  241. package/dist/index-DNgf5gyG2.d.ts.map +0 -1
@@ -21,10 +21,10 @@ interface SemanticConceptJudgeAdapterOpts {
21
21
  /** Maximum post-cancellation wait for a provider receipt. Default 5 seconds. */
22
22
  settlementTimeoutMs?: number;
23
23
  }
24
- declare function createSemanticConceptJudgeAdapter(opts: SemanticConceptJudgeAdapterOpts): Analyst<SemanticConceptJudgeInput>;
24
+ export declare function createSemanticConceptJudgeAdapter(opts: SemanticConceptJudgeAdapterOpts): Analyst<SemanticConceptJudgeInput>;
25
25
  //#endregion
26
26
  //#region src/analyst/benchmark-agentrx-calibration.d.ts
27
- declare const AGENT_RX_UPSTREAM_REVISION = "f228165bfec60a801fd5fedd9d8ffe0f9de0c69d";
27
+ export declare const AGENT_RX_UPSTREAM_REVISION = "f228165bfec60a801fd5fedd9d8ffe0f9de0c69d";
28
28
  interface AgentRxCalibrationRunnerSummary {
29
29
  runnerId: string;
30
30
  selectedRuns: number;
@@ -53,8 +53,8 @@ interface AgentRxCalibrationSummary {
53
53
  rationale: string;
54
54
  runners: AgentRxCalibrationRunnerSummary[];
55
55
  }
56
- declare function summarizeAgentRxCalibration(result: AnalystBenchmarkResult, upstreamRevision: string): AgentRxCalibrationSummary;
57
- declare function renderAgentRxCalibrationMarkdown(summary: AgentRxCalibrationSummary): string;
56
+ export declare function summarizeAgentRxCalibration(result: AnalystBenchmarkResult, upstreamRevision: string): AgentRxCalibrationSummary;
57
+ export declare function renderAgentRxCalibrationMarkdown(summary: AgentRxCalibrationSummary): string;
58
58
  //#endregion
59
59
  //#region src/analyst/benchmark-dataset-types.d.ts
60
60
  type ExternalId = string | number;
@@ -159,20 +159,20 @@ interface UpstreamPredictionAdapterOptions extends StepLabelAdapterOptions {
159
159
  interface CodeTracerPredictionAdapterOptions extends UpstreamPredictionAdapterOptions, CodeTraceBenchLabelOptions {}
160
160
  //#endregion
161
161
  //#region src/analyst/benchmark-dataset-agentrx.d.ts
162
- declare function agentRxBenchmarkCase<TInput>(row: AgentRxRow, input: TInput, options?: AgentRxBenchmarkCaseOptions): AnalystBenchmarkCase<TInput>;
162
+ export declare function agentRxBenchmarkCase<TInput>(row: AgentRxRow, input: TInput, options?: AgentRxBenchmarkCaseOptions): AnalystBenchmarkCase<TInput>;
163
163
  /** Translate AgentRx `Report.to_dict()` output or its `failures` array into findings. */
164
- declare function agentRxPredictionsToFindings(trajectoryIdValue: ExternalId, output: unknown, options?: UpstreamPredictionAdapterOptions): AnalystFinding[];
165
- declare function normalizeAgentRxCategory(value: string): string;
164
+ export declare function agentRxPredictionsToFindings(trajectoryIdValue: ExternalId, output: unknown, options?: UpstreamPredictionAdapterOptions): AnalystFinding[];
165
+ export declare function normalizeAgentRxCategory(value: string): string;
166
166
  /** Match Python's round() behavior used by AgentRx for consensus steps. */
167
- declare function roundAgentRxStep(value: number): number;
167
+ export declare function roundAgentRxStep(value: number): number;
168
168
  //#endregion
169
169
  //#region src/analyst/benchmark-dataset-codetrace.d.ts
170
- declare function codeTraceBenchCase<TInput>(row: CodeTraceBenchRow, input: TInput, options?: CodeTraceBenchCaseOptions): AnalystBenchmarkCase<TInput>;
170
+ export declare function codeTraceBenchCase<TInput>(row: CodeTraceBenchRow, input: TInput, options?: CodeTraceBenchCaseOptions): AnalystBenchmarkCase<TInput>;
171
171
  /** Translate CodeTracer's `codetracer_labels.json` into shared findings. */
172
- declare function codeTracerPredictionsToFindings(trajectoryIdValue: string, predictions: CodeTracerPredictions, options?: CodeTracerPredictionAdapterOptions): AnalystFinding[];
172
+ export declare function codeTracerPredictionsToFindings(trajectoryIdValue: string, predictions: CodeTracerPredictions, options?: CodeTracerPredictionAdapterOptions): AnalystFinding[];
173
173
  //#endregion
174
174
  //#region src/analyst/benchmark-dataset-utils.d.ts
175
- declare function normalizeBenchmarkLabel(value: string): string;
175
+ export declare function normalizeBenchmarkLabel(value: string): string;
176
176
  //#endregion
177
177
  //#region src/analyst/benchmark-verification-outcome.d.ts
178
178
  type VerificationOutcomeStatus = 'passed' | 'failed' | 'unavailable';
@@ -198,10 +198,10 @@ interface VerificationResultFile {
198
198
  relativePath: string;
199
199
  content: string;
200
200
  }
201
- declare function parseVerificationOutcome(files: readonly VerificationResultFile[]): VerificationOutcome;
201
+ export declare function parseVerificationOutcome(files: readonly VerificationResultFile[]): VerificationOutcome;
202
202
  //#endregion
203
203
  //#region src/analyst/benchmark-verification-artifacts.d.ts
204
- declare const DEFAULT_MAX_VERIFICATION_ARTIFACT_BYTES: number;
204
+ export declare const DEFAULT_MAX_VERIFICATION_ARTIFACT_BYTES: number;
205
205
  type VerificationArtifactRole = 'final-test-output' | 'final-result' | 'final-metrics';
206
206
  interface VerificationArtifactFile {
207
207
  role: VerificationArtifactRole;
@@ -231,12 +231,12 @@ interface LoadedVerificationArtifacts {
231
231
  content: string;
232
232
  }>;
233
233
  }
234
- declare function loadCodeTraceVerificationArtifacts(options: {
234
+ export declare function loadCodeTraceVerificationArtifacts(options: {
235
235
  artifactDir: string;
236
236
  row: CodeTraceBenchRow;
237
237
  maxBytes?: number;
238
238
  }): Promise<LoadedVerificationArtifacts>;
239
- declare function appendVerificationArtifactsToOtlp(otlpText: string, traceId: string, artifacts: LoadedVerificationArtifacts, afterTimestamp: string): string;
239
+ export declare function appendVerificationArtifactsToOtlp(otlpText: string, traceId: string, artifacts: LoadedVerificationArtifacts, afterTimestamp: string): string;
240
240
  //#endregion
241
241
  //#region src/analyst/benchmark-public-types.d.ts
242
242
  type PublicAnalystBenchmarkDataset = 'agentrx' | 'codetracebench';
@@ -404,8 +404,8 @@ interface CodeTraceStepAssignment {
404
404
  step: number;
405
405
  block: CodeTraceFailureBlock;
406
406
  }
407
- declare function emptyPublicBenchmarkRunner(): AnalystBenchmarkRunner<AnalystRunInputs>;
408
- declare function adaptPublicBenchmarkFindings(options: {
407
+ export declare function emptyPublicBenchmarkRunner(): AnalystBenchmarkRunner<AnalystRunInputs>;
408
+ export declare function adaptPublicBenchmarkFindings(options: {
409
409
  dataset: PublicAnalystBenchmarkDataset;
410
410
  trajectoryId: string;
411
411
  findings: readonly AnalystFinding[];
@@ -425,7 +425,7 @@ declare function adaptPublicBenchmarkFindings(options: {
425
425
  * blocks never reach it: every runner reports blocks, and this function turns
426
426
  * them into the per-step findings the benchmark defines.
427
427
  */
428
- declare function expandCodeTraceFailureBlocks(options: {
428
+ export declare function expandCodeTraceFailureBlocks(options: {
429
429
  trajectoryId: string;
430
430
  blocks: readonly CodeTraceFailureBlock[];
431
431
  store: TraceAnalysisStore;
@@ -439,14 +439,14 @@ declare function expandCodeTraceFailureBlocks(options: {
439
439
  }>;
440
440
  //#endregion
441
441
  //#region src/analyst/benchmark-public-data.d.ts
442
- declare function loadPublicBenchmarkRows(path: string): Promise<Array<Record<string, unknown>>>;
443
- declare function selectPublicBenchmarkRows(dataset: PublicAnalystBenchmarkDataset, rows: readonly Record<string, unknown>[], options: {
442
+ export declare function loadPublicBenchmarkRows(path: string): Promise<Array<Record<string, unknown>>>;
443
+ export declare function selectPublicBenchmarkRows(dataset: PublicAnalystBenchmarkDataset, rows: readonly Record<string, unknown>[], options: {
444
444
  limit: number;
445
445
  seed: number;
446
446
  }): Array<Record<string, unknown>>;
447
- declare function publicBenchmarkDistributions(dataset: PublicAnalystBenchmarkDataset, rows: readonly Record<string, unknown>[]): PublicBenchmarkDistributions;
448
- declare function publicBenchmarkSelectionReport(dataset: PublicAnalystBenchmarkDataset, source: readonly Record<string, unknown>[], selected: readonly Record<string, unknown>[], seed: number): PublicBenchmarkSelectionReport;
449
- declare function preparePublicAnalystBenchmark(options: {
447
+ export declare function publicBenchmarkDistributions(dataset: PublicAnalystBenchmarkDataset, rows: readonly Record<string, unknown>[]): PublicBenchmarkDistributions;
448
+ export declare function publicBenchmarkSelectionReport(dataset: PublicAnalystBenchmarkDataset, source: readonly Record<string, unknown>[], selected: readonly Record<string, unknown>[], seed: number): PublicBenchmarkSelectionReport;
449
+ export declare function preparePublicAnalystBenchmark(options: {
450
450
  dataset: PublicAnalystBenchmarkDataset;
451
451
  labelsPath: string;
452
452
  traceDir: string;
@@ -646,14 +646,14 @@ interface AnalystDefinition<TRow = unknown, TAssignment = unknown, TBlock = unkn
646
646
  * message names the construct so an expressiveness gap is a loud, attributable
647
647
  * failure instead of a silently narrowed protocol.
648
648
  */
649
- declare class AnalystExpressivenessError extends Error {}
649
+ export declare class AnalystExpressivenessError extends Error {}
650
650
  /**
651
651
  * Digest of everything a definition can send to its model. An inline
652
652
  * definition hashes under the historical prime-protocol domain, so its digest
653
653
  * equals the digest its bespoke arm always recorded; other projections hash
654
654
  * under the definition domain.
655
655
  */
656
- declare function analystDefinitionProtocolSha256<TRow, TAssignment, TBlock>(definition: AnalystDefinition<TRow, TAssignment, TBlock>): string;
656
+ export declare function analystDefinitionProtocolSha256<TRow, TAssignment, TBlock>(definition: AnalystDefinition<TRow, TAssignment, TBlock>): string;
657
657
  /** One definition's declared difference from the compared set. */
658
658
  interface AnalystDefinitionAsymmetry {
659
659
  readonly id: string;
@@ -682,27 +682,27 @@ interface AnalystDefinitionAsymmetryReport {
682
682
  * arm, because a retry is a second sample. Projection, reasoning effort, and
683
683
  * budget differences are declared and reported, never hidden.
684
684
  */
685
- declare function analystDefinitionAsymmetries(definitions: ReadonlyArray<AnalystDefinition<unknown, unknown, unknown>>): AnalystDefinitionAsymmetryReport;
685
+ export declare function analystDefinitionAsymmetries(definitions: ReadonlyArray<AnalystDefinition<unknown, unknown, unknown>>): AnalystDefinitionAsymmetryReport;
686
686
  //#endregion
687
687
  //#region src/analyst/benchmark-public-prompt.d.ts
688
688
  /** Widest contiguous failure block a model may report. The published corpus's
689
689
  * widest labeled block is 8 steps and its widest stage span is 9, so this bound
690
690
  * never binds honest enumeration; it caps how far one over-wide block can push
691
691
  * unlabeled steps into the precision denominator. */
692
- declare const MAX_INCORRECT_BLOCK_STEPS = 12;
692
+ export declare const MAX_INCORRECT_BLOCK_STEPS = 12;
693
693
  /** Most blocks a model may report for one trajectory. The published corpus's
694
694
  * densest case carries 4 disjoint labeled blocks. Together with the per-block
695
695
  * cap this bounds one case at 192 predicted steps without a second ceiling. */
696
- declare const MAX_INCORRECT_BLOCKS = 16;
697
- declare const CODE_TRACE_BENCH_ANALYST_PROMPT = "Analyze exactly one coding-agent trajectory and its attached final verification.\nYour task is the CodeTraceBench incorrect-step task: identify every incorrect step, defined as a wrong state-changing intervention given the evidence — a mislocalized edit, a wrong hypothesis that drives an action, a regression, an irrelevant change, or an incorrect dependency or configuration choice.\nIf the final verification failed, the trajectory MUST contain at least one incorrect step. Never return an empty findings array on a failing trajectory; trace backward until you find the root cause.\nWork backward, the way this benchmark was annotated, never by scanning forward for suspicious steps: start from the final verification outcome or the latest observed failure evidence, identify the immediately preceding step whose action or output produced that observed error, then recursively ask which earlier decision led to each intermediate failure, until the preceding steps contain no error or the cause is unrelated to the trajectory's own decisions.\nEach backward chain terminates at an error-critical step — the earliest decision that triggered the downstream cascade — and that step is the block's first_step: the step that committed the mistake, not the step that planned it and not a later step that repeats it.\nA block is a maximal contiguous sequence of strictly incorrect steps. A step belongs in the block ONLY if it introduces, propagates, or compounds the error. \nDo NOT include steps that merely \"act on\", diagnose, or react to the error. A diagnostic command, a test run exposing the bug, or a correct exploratory read is a CORRECT step. \nIf an incorrect step is followed by a correct diagnostic step and then another incorrect step, you MUST emit two separate blocks. NEVER bridge correct steps by grouping them into a single block with incorrect steps. Over-blocking drastically hurts your precision.\nAfter identifying first_step, extend last_step forward ONLY through consecutive steps that independently introduce, propagate, or compound the mistake. A cascade of repeated failed attempts at the same wrong approach is one maximal block, provided EVERY step is independently incorrect.\nDo not end a block merely because the agent tried a variation of the same wrong approach; a variation that still carries the error stays inside the block.\nA partially correct or ambiguous fix still counts as incorrect; the block ends only at the first step free of the error — a clean diagnostic read, the corrective action that closes the issue and needs no further rework, or a genuine abandonment of the wrong approach.\nBlock extent follows the traced chain and this forward extension, nothing else.\nReport each failure block as exactly one finding whose first_step is the block's first incorrect step and whose last_step is its last, covering every consecutive step between them.\nEvery step inside a block is scored on its own: naming a correct step costs exactly as much as missing an incorrect one, and naming only the first step of a longer block forfeits every unnamed step. Because of this, carefully verify every step between first_step and last_step. Only include steps that introduce, propagate, or compound the error.\nReport blocks separated by at least one correct step as separate findings, and never let two blocks overlap. If there are multiple separate failure cascades, emit a separate finding for each one.\nPrefer anchored blocks: a block whose chain traces back from observed failure evidence — a failing command or verification, an error observation, a regression, or, on a solved trajectory, a later step that reverts or supersedes it — outranks one without.\nWhen an action is clearly wrong on its own evidence but you cannot trace such an anchor, report the block anyway with proportionally lower confidence.\nA solved trajectory still carries every mistake made along the way: inspect its final patching and verification stages for a state-changing action that a later step reverted, superseded, or corrected — a wrong edit just before the final fix is incorrect even when every test ends green.\nBefore emitting a candidate block, check its boundaries.\nNeighbor check: ask whether the accusation fits one step earlier (the decision rather than its consequence) or one step later (the next step still acts on or reworks the same error) better than where you placed it, and move the boundary when it does; a boundary off by one step scores zero at that step.\nCompleteness check: a block must cover the maximal contiguous sequence of incorrect steps. If an agent fails at step 10, tries to fix it at 11, fails, and tries again at 12 and 13, all four steps are incorrect and must be included in the block. Never truncate a cascade. If you miss the later steps of a cascade, your recall drops to zero for them.\nCounterfactual check: ask which step's correct execution would have made the downstream failure or rework disappear, and move first_step onto that step; use this check only to move a boundary, never to delete a block.\nWidth check: inspect the step immediately before first_step and the step immediately after last_step; when either neighbor commits to, compounds, or propagates the same mistake, it belongs inside the block — move that boundary outward and repeat until both neighbors are free of the error.\nFor each block you keep, name as consequence_step the step number whose action or observation shows the damage — a failing command, a wrong file state, a repeated failure, or rework the agent had to do because of this block; that step is the block's own last step when its observation already shows the damage, and a later step otherwise.\nWhen you cannot name that step number from the trace you were given, drop the block; a plausible story about why a step looks wrong is not evidence that it was.\nA passing final verification is not evidence that a block caused nothing, and a failing final verification is not evidence that any particular block caused it.\nFor every block, decide whether the agent escaped the failure.\nMark escape_status \"escaped\" only when you can name the single later step that fully reversed the block, the agent needed no other step to recover, and nothing after that step revisits the same file, command, or hypothesis; write that step number in the rationale. If the agent required multiple steps to fix the issue, or if you are unsure, it is not escaped.\nMark escape_status \"unescaped\" in every other case, including whenever you are unsure.\nA passing final verification never makes a block escaped.\nDo not label a diagnostic probe or test run merely because its output exposes an earlier defect.\nDo not label a redundant but correct read or search; CodeTraceBench scores unuseful steps separately, and this run scores incorrect steps only. Never include a correct step in a block just to bridge two incorrect steps; instead, emit two separate blocks.\nDo not label a step solely because final verification failed.\nWhen final verification is unavailable, trace backward from the latest failure evidence inside the trajectory itself.\nEvery step in a reported block MUST be the positive integer n from an existing assistant LLM span named step-<n>.\nNever select an EVALUATOR, TOOL, CHAIN, final-verification, benchmark-verification, or message-<n> span.\nBefore emitting a finding, inspect every covered span's attributes.content and describe only the actions shown there.\nReport at most 16 blocks and at most 12 steps in one block; when more candidates than that exist, report the ones whose chains carry the clearest downstream evidence.\nWhen the trajectory has no incorrect steps — its final verification passed and the final-stage sweep found no reverted, superseded, or corrected action — return an empty findings array.";
696
+ export declare const MAX_INCORRECT_BLOCKS = 16;
697
+ export declare const CODE_TRACE_BENCH_ANALYST_PROMPT = "Analyze exactly one coding-agent trajectory and its attached final verification.\nYour task is the CodeTraceBench incorrect-step task: identify every incorrect step, defined as a wrong state-changing intervention given the evidence — a mislocalized edit, a wrong hypothesis that drives an action, a regression, an irrelevant change, or an incorrect dependency or configuration choice.\nIf the final verification failed, the trajectory MUST contain at least one incorrect step. Never return an empty findings array on a failing trajectory; trace backward until you find the root cause.\nWork backward, the way this benchmark was annotated, never by scanning forward for suspicious steps: start from the final verification outcome or the latest observed failure evidence, identify the immediately preceding step whose action or output produced that observed error, then recursively ask which earlier decision led to each intermediate failure, until the preceding steps contain no error or the cause is unrelated to the trajectory's own decisions.\nEach backward chain terminates at an error-critical step — the earliest decision that triggered the downstream cascade — and that step is the block's first_step: the step that committed the mistake, not the step that planned it and not a later step that repeats it.\nA block is a maximal contiguous sequence of strictly incorrect steps. A step belongs in the block ONLY if it introduces, propagates, or compounds the error. \nDo NOT include steps that merely \"act on\", diagnose, or react to the error. A diagnostic command, a test run exposing the bug, or a correct exploratory read is a CORRECT step. \nIf an incorrect step is followed by a correct diagnostic step and then another incorrect step, you MUST emit two separate blocks. NEVER bridge correct steps by grouping them into a single block with incorrect steps. Over-blocking drastically hurts your precision.\nAfter identifying first_step, extend last_step forward ONLY through consecutive steps that independently introduce, propagate, or compound the mistake. A cascade of repeated failed attempts at the same wrong approach is one maximal block, provided EVERY step is independently incorrect.\nDo not end a block merely because the agent tried a variation of the same wrong approach; a variation that still carries the error stays inside the block.\nA partially correct or ambiguous fix still counts as incorrect; the block ends only at the first step free of the error — a clean diagnostic read, the corrective action that closes the issue and needs no further rework, or a genuine abandonment of the wrong approach.\nBlock extent follows the traced chain and this forward extension, nothing else.\nReport each failure block as exactly one finding whose first_step is the block's first incorrect step and whose last_step is its last, covering every consecutive step between them.\nEvery step inside a block is scored on its own: naming a correct step costs exactly as much as missing an incorrect one, and naming only the first step of a longer block forfeits every unnamed step. Because of this, carefully verify every step between first_step and last_step. Only include steps that introduce, propagate, or compound the error.\nReport blocks separated by at least one correct step as separate findings, and never let two blocks overlap. If there are multiple separate failure cascades, emit a separate finding for each one.\nPrefer anchored blocks: a block whose chain traces back from observed failure evidence — a failing command or verification, an error observation, a regression, or, on a solved trajectory, a later step that reverts or supersedes it — outranks one without.\nWhen an action is clearly wrong on its own evidence but you cannot trace such an anchor, report the block anyway with proportionally lower confidence.\nA solved trajectory still carries every mistake made along the way: inspect its final patching and verification stages for a state-changing action that a later step reverted, superseded, or corrected — a wrong edit just before the final fix is incorrect even when every test ends green.\nBefore emitting a candidate block, check its boundaries.\nNeighbor check: ask whether the accusation fits one step earlier (the decision rather than its consequence) or one step later (the next step still acts on or reworks the same error) better than where you placed it, and move the boundary when it does; a boundary off by one step scores zero at that step.\nCompleteness check: a block must cover the maximal contiguous sequence of incorrect steps. If an agent fails at step 10, tries to fix it at 11, fails, and tries again at 12 and 13, all four steps are incorrect and must be included in the block. Never truncate a cascade. If you miss the later steps of a cascade, your recall drops to zero for them.\nCounterfactual check: ask which step's correct execution would have made the downstream failure or rework disappear, and move first_step onto that step; use this check only to move a boundary, never to delete a block.\nWidth check: inspect the step immediately before first_step and the step immediately after last_step; when either neighbor commits to, compounds, or propagates the same mistake, it belongs inside the block — move that boundary outward and repeat until both neighbors are free of the error.\nFor each block you keep, name as consequence_step the step number whose action or observation shows the damage — a failing command, a wrong file state, a repeated failure, or rework the agent had to do because of this block; that step is the block's own last step when its observation already shows the damage, and a later step otherwise.\nWhen you cannot name that step number from the trace you were given, drop the block; a plausible story about why a step looks wrong is not evidence that it was.\nA passing final verification is not evidence that a block caused nothing, and a failing final verification is not evidence that any particular block caused it.\nFor every block, decide whether the agent escaped the failure.\nMark escape_status \"escaped\" only when you can name the single later step that fully reversed the block, the agent needed no other step to recover, and nothing after that step revisits the same file, command, or hypothesis; write that step number in the rationale. If the agent required multiple steps to fix the issue, or if you are unsure, it is not escaped.\nMark escape_status \"unescaped\" in every other case, including whenever you are unsure.\nA passing final verification never makes a block escaped.\nDo not label a diagnostic probe or test run merely because its output exposes an earlier defect.\nDo not label a redundant but correct read or search; CodeTraceBench scores unuseful steps separately, and this run scores incorrect steps only. Never include a correct step in a block just to bridge two incorrect steps; instead, emit two separate blocks.\nDo not label a step solely because final verification failed.\nWhen final verification is unavailable, trace backward from the latest failure evidence inside the trajectory itself.\nEvery step in a reported block MUST be the positive integer n from an existing assistant LLM span named step-<n>.\nNever select an EVALUATOR, TOOL, CHAIN, final-verification, benchmark-verification, or message-<n> span.\nBefore emitting a finding, inspect every covered span's attributes.content and describe only the actions shown there.\nReport at most 16 blocks and at most 12 steps in one block; when more candidates than that exist, report the ones whose chains carry the clearest downstream evidence.\nWhen the trajectory has no incorrect steps — its final verification passed and the final-stage sweep found no reverted, superseded, or corrected action — return an empty findings array.";
698
698
  /** One-shot JSON transport prompt for the direct runner. */
699
- declare function publicBenchmarkSystemPrompt(dataset: PublicAnalystBenchmarkDataset): string;
699
+ export declare function publicBenchmarkSystemPrompt(dataset: PublicAnalystBenchmarkDataset): string;
700
700
  /** Tool-loop prompt for the recursive runner. Same task, subject-encoded block. */
701
- declare function publicBenchmarkRlmInstructions(dataset: PublicAnalystBenchmarkDataset): string;
701
+ export declare function publicBenchmarkRlmInstructions(dataset: PublicAnalystBenchmarkDataset): string;
702
702
  /** Digest of every prompt a runner can send plus the shared transport limits.
703
703
  * Both runner contracts are hashed so an edit to either one changes the digest
704
704
  * a run records, whichever runner executed. */
705
- declare function publicBenchmarkProtocolSha256(dataset: PublicAnalystBenchmarkDataset): string;
705
+ export declare function publicBenchmarkProtocolSha256(dataset: PublicAnalystBenchmarkDataset): string;
706
706
  //#endregion
707
707
  //#region src/analyst/benchmark-public-model.d.ts
708
708
  /**
@@ -724,9 +724,9 @@ interface PublicDirectDefinitionArgs {
724
724
  maxCostUsd: number;
725
725
  }
726
726
  /** The direct arm as a declarative unit for one public dataset. */
727
- declare function publicDirectAnalystDefinition(dataset: PublicAnalystBenchmarkDataset, args: PublicDirectDefinitionArgs): AnalystDefinition<PublicBenchmarkModelPrediction>;
727
+ export declare function publicDirectAnalystDefinition(dataset: PublicAnalystBenchmarkDataset, args: PublicDirectDefinitionArgs): AnalystDefinition<PublicBenchmarkModelPrediction>;
728
728
  /** Thin shell: validate config, declare the definition, run the chunked strategy. */
729
- declare function createPublicBenchmarkDirectRunner(dataset: PublicAnalystBenchmarkDataset, config: PublicAnalystBenchmarkModelConfig): AnalystBenchmarkRunner<AnalystRunInputs>;
729
+ export declare function createPublicBenchmarkDirectRunner(dataset: PublicAnalystBenchmarkDataset, config: PublicAnalystBenchmarkModelConfig): AnalystBenchmarkRunner<AnalystRunInputs>;
730
730
  declare const AgentRxPredictionSchema: z.ZodObject<{
731
731
  step: z.ZodNumber;
732
732
  severity: z.ZodEnum<{
@@ -805,11 +805,11 @@ interface PublicRlmDefinitionArgs {
805
805
  engineLimits: TraceAnalystLimits;
806
806
  }
807
807
  /** The dspy-rlm arm as a declarative unit for one public dataset. */
808
- declare function publicRlmAnalystDefinition(dataset: PublicAnalystBenchmarkDataset, args: PublicRlmDefinitionArgs): AnalystDefinition<RawAnalystFinding, CodeTraceStepAssignment>;
808
+ export declare function publicRlmAnalystDefinition(dataset: PublicAnalystBenchmarkDataset, args: PublicRlmDefinitionArgs): AnalystDefinition<RawAnalystFinding, CodeTraceStepAssignment>;
809
809
  /** Thin shell: validate config, declare the definition, run the repl-variable strategy. */
810
- declare function createPublicBenchmarkRlmRunner(dataset: PublicAnalystBenchmarkDataset, config: PublicAnalystBenchmarkModelConfig): AnalystBenchmarkRunner<AnalystRunInputs>;
810
+ export declare function createPublicBenchmarkRlmRunner(dataset: PublicAnalystBenchmarkDataset, config: PublicAnalystBenchmarkModelConfig): AnalystBenchmarkRunner<AnalystRunInputs>;
811
811
  /** Engine iteration limits with this arm's defaults applied. */
812
- declare function rlmEngineLimits(config: PublicAnalystBenchmarkModelConfig): TraceAnalystLimits;
812
+ export declare function rlmEngineLimits(config: PublicAnalystBenchmarkModelConfig): TraceAnalystLimits;
813
813
  //#endregion
814
814
  //#region src/analyst/benchmark-runner-prime.d.ts
815
815
  interface PrimeBenchmarkRunnerOptions {
@@ -830,7 +830,7 @@ interface PrimeBenchmarkRunnerOptions {
830
830
  * Digest of everything this arm can send to the bridge, recorded per
831
831
  * observation so a prime result names the exact contract that produced it.
832
832
  */
833
- declare function primeAnalystProtocolSha256(): string;
833
+ export declare function primeAnalystProtocolSha256(): string;
834
834
  interface PrimeCodeTraceDefinitionArgs {
835
835
  /** Deadline for one bridge call. */
836
836
  timeoutMs: number;
@@ -841,9 +841,9 @@ interface PrimeCodeTraceDefinitionArgs {
841
841
  * The prime arm as a declarative unit. CodeTraceBench-only: the question,
842
842
  * task text, and block grammar speak its incorrect-step definition.
843
843
  */
844
- declare function primeCodeTraceAnalystDefinition(args: PrimeCodeTraceDefinitionArgs): AnalystDefinition<CodeTraceFailureBlock>;
844
+ export declare function primeCodeTraceAnalystDefinition(args: PrimeCodeTraceDefinitionArgs): AnalystDefinition<CodeTraceFailureBlock>;
845
845
  /** Thin shell: validate options, declare the definition, run the inline strategy. */
846
- declare function createPrimeBenchmarkRunner(options: PrimeBenchmarkRunnerOptions): AnalystBenchmarkRunner<AnalystRunInputs>;
846
+ export declare function createPrimeBenchmarkRunner(options: PrimeBenchmarkRunnerOptions): AnalystBenchmarkRunner<AnalystRunInputs>;
847
847
  //#endregion
848
848
  //#region src/analyst/benchmark-comparison.d.ts
849
849
  /**
@@ -904,7 +904,7 @@ interface AnalystRunnerComparison {
904
904
  candidateRunnerId: string;
905
905
  metrics: AnalystMetricComparison[];
906
906
  }
907
- declare function compareAnalystRunners(result: AnalystBenchmarkResult, options: {
907
+ export declare function compareAnalystRunners(result: AnalystBenchmarkResult, options: {
908
908
  baselineRunnerId: string;
909
909
  candidateRunnerId: string;
910
910
  confidence?: number;
@@ -942,8 +942,8 @@ interface CodeTraceCalibrationSummary {
942
942
  rationale: string;
943
943
  runners: CodeTraceCalibrationRunnerSummary[];
944
944
  }
945
- declare function summarizeCodeTraceCalibration(result: AnalystBenchmarkResult): CodeTraceCalibrationSummary;
946
- declare function renderCodeTraceCalibrationMarkdown(summary: CodeTraceCalibrationSummary): string;
945
+ export declare function summarizeCodeTraceCalibration(result: AnalystBenchmarkResult): CodeTraceCalibrationSummary;
946
+ export declare function renderCodeTraceCalibrationMarkdown(summary: CodeTraceCalibrationSummary): string;
947
947
  //#endregion
948
948
  //#region src/analyst/benchmark-command-artifact.d.ts
949
949
  interface AnalystBenchmarkArtifact {
@@ -1122,13 +1122,13 @@ interface AnalystBenchmarkProgressRow {
1122
1122
  observation: AnalystBenchmarkObservation;
1123
1123
  rowSha256: string;
1124
1124
  }
1125
- declare const ANALYST_BENCHMARK_MANIFEST_FILE = "manifest.json";
1126
- declare const ANALYST_BENCHMARK_OBSERVATIONS_FILE = "observations.jsonl";
1127
- declare const ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE = "run.local.json";
1128
- declare const ANALYST_BENCHMARK_COST_LEDGER_FILE = "cost-ledger.jsonl";
1125
+ export declare const ANALYST_BENCHMARK_MANIFEST_FILE = "manifest.json";
1126
+ export declare const ANALYST_BENCHMARK_OBSERVATIONS_FILE = "observations.jsonl";
1127
+ export declare const ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE = "run.local.json";
1128
+ export declare const ANALYST_BENCHMARK_COST_LEDGER_FILE = "cost-ledger.jsonl";
1129
1129
  //#endregion
1130
1130
  //#region src/analyst/benchmark-command-result.d.ts
1131
- declare function readAnalystBenchmarkArtifact(path: string): Promise<AnalystBenchmarkArtifact>;
1131
+ export declare function readAnalystBenchmarkArtifact(path: string): Promise<AnalystBenchmarkArtifact>;
1132
1132
  //#endregion
1133
1133
  //#region src/analyst/benchmark-command.d.ts
1134
1134
  interface AnalystBenchmarkCommandDependencies {
@@ -1176,27 +1176,27 @@ interface AnalystBenchmarkCommandConfig {
1176
1176
  command: string;
1177
1177
  resume: boolean;
1178
1178
  }
1179
- declare function runAnalystBenchmarkCommand(argv: readonly string[], env?: NodeJS.ProcessEnv, dependencies?: AnalystBenchmarkCommandDependencies): Promise<number>;
1179
+ export declare function runAnalystBenchmarkCommand(argv: readonly string[], env?: NodeJS.ProcessEnv, dependencies?: AnalystBenchmarkCommandDependencies): Promise<number>;
1180
1180
  //#endregion
1181
1181
  //#region src/analyst/benchmark-implementation.d.ts
1182
- declare const ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM = "sha256-canonical-source-manifest";
1182
+ export declare const ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM = "sha256-canonical-source-manifest";
1183
1183
  /** The dependency files' manifest digest with the package's own version
1184
1184
  * fields blanked, so a release that only bumps the version keeps the pin. */
1185
- declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM = "sha256-canonical-file-manifest-without-own-version";
1186
- declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES: readonly string[];
1187
- declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 = "007041f0de3b376aa3def1d3efe8221a724030a22ac0a7988921db12c4624c09";
1188
- declare const ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256 = "1e03f2daed356d60316aabefb407ec1e437ac94d408d61eea4ae096e9c6fbb5b";
1189
- declare const ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256 = "4dba263b6256a30d56c7fdb2d992d3a953c0035d731f359b704db806f68f75ac";
1190
- declare const ANALYST_BENCHMARK_IMPLEMENTATION_FILES: readonly string[];
1191
- declare const ANALYST_BENCHMARK_IMPLEMENTATION_SHA256 = "8bb4bdae57e4c5596ffe77d80c7a16e67eef7b2fc5087c2af369a417ed423a26";
1192
- declare function analystBenchmarkImplementationDigest(): string;
1193
- declare function analystBenchmarkDependencyLockDigest(): string;
1185
+ export declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM = "sha256-canonical-file-manifest-without-own-version";
1186
+ export declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES: readonly string[];
1187
+ export declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 = "e57c71c31ecb3ec79d6cf08dc6f1603aa8109bafee4208e63f0e3a1234619d5a";
1188
+ export declare const ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256 = "1e03f2daed356d60316aabefb407ec1e437ac94d408d61eea4ae096e9c6fbb5b";
1189
+ export declare const ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256 = "4dba263b6256a30d56c7fdb2d992d3a953c0035d731f359b704db806f68f75ac";
1190
+ export declare const ANALYST_BENCHMARK_IMPLEMENTATION_FILES: readonly string[];
1191
+ export declare const ANALYST_BENCHMARK_IMPLEMENTATION_SHA256 = "86541c7a96fc25a05247656739bced9fbfd89625ce67b56e484266f564447663";
1192
+ export declare function analystBenchmarkImplementationDigest(): string;
1193
+ export declare function analystBenchmarkDependencyLockDigest(): string;
1194
1194
  //#endregion
1195
1195
  //#region src/analyst/benchmark-instructions-override.d.ts
1196
1196
  /** Build an override from instruction text. Blank text is a caller error. */
1197
- declare function analystInstructionsOverrideFromText(text: string): AnalystInstructionsOverride;
1197
+ export declare function analystInstructionsOverrideFromText(text: string): AnalystInstructionsOverride;
1198
1198
  /** Read override instructions from a file. Any read failure is fatal. */
1199
- declare function readAnalystInstructionsOverride(path: string): AnalystInstructionsOverride;
1199
+ export declare function readAnalystInstructionsOverride(path: string): AnalystInstructionsOverride;
1200
1200
  /**
1201
1201
  * Protocol digest of the run as executed.
1202
1202
  *
@@ -1207,13 +1207,13 @@ declare function readAnalystInstructionsOverride(path: string): AnalystInstructi
1207
1207
  * prompt) to the exact override text, so the recorded digest always hashes the
1208
1208
  * instructions that actually ran.
1209
1209
  */
1210
- declare function effectiveAnalystProtocolSha256(dataset: PublicAnalystBenchmarkDataset, override?: Pick<AnalystInstructionsOverride, 'sha256'>): string;
1210
+ export declare function effectiveAnalystProtocolSha256(dataset: PublicAnalystBenchmarkDataset, override?: Pick<AnalystInstructionsOverride, 'sha256'>): string;
1211
1211
  //#endregion
1212
1212
  //#region src/analyst/benchmark-report.d.ts
1213
- declare function renderAnalystBenchmarkMarkdown(result: AnalystBenchmarkResult, comparisons?: readonly AnalystRunnerComparison[]): string;
1213
+ export declare function renderAnalystBenchmarkMarkdown(result: AnalystBenchmarkResult, comparisons?: readonly AnalystRunnerComparison[]): string;
1214
1214
  //#endregion
1215
1215
  //#region src/analyst/benchmark-summary.d.ts
1216
- declare function summarizeAnalystBenchmarkRunner(runnerId: string, observations: readonly AnalystBenchmarkObservation[]): AnalystBenchmarkSummary;
1216
+ export declare function summarizeAnalystBenchmarkRunner(runnerId: string, observations: readonly AnalystBenchmarkObservation[]): AnalystBenchmarkSummary;
1217
1217
  //#endregion
1218
1218
  //#region src/analyst/bind.d.ts
1219
1219
  /** Execution half of a binding: who actually reaches the model. */
@@ -1233,7 +1233,7 @@ type AnalystTransportBinding = {
1233
1233
  * Compile a definition plus a transport binding into a runnable arm. Dispatch
1234
1234
  * is total over the expressible pairs; everything else is a loud refusal.
1235
1235
  */
1236
- declare function bindAnalyst<TRow, TAssignment, TBlock>(definition: AnalystDefinition<TRow, TAssignment, TBlock>, transports: AnalystTransportBinding): AnalystBenchmarkRunner<AnalystRunInputs>;
1236
+ export declare function bindAnalyst<TRow, TAssignment, TBlock>(definition: AnalystDefinition<TRow, TAssignment, TBlock>, transports: AnalystTransportBinding): AnalystBenchmarkRunner<AnalystRunInputs>;
1237
1237
  //#endregion
1238
1238
  //#region src/analyst/define.d.ts
1239
1239
  /**
@@ -1243,7 +1243,7 @@ declare function bindAnalyst<TRow, TAssignment, TBlock>(definition: AnalystDefin
1243
1243
  * it to any TraceAnalysisEngine with `runTraceAnalyst` or
1244
1244
  * `createTraceAnalyst`.
1245
1245
  */
1246
- declare function defineTraceAnalyst(definition: TraceAnalystDefinition): TraceAnalystDefinition;
1246
+ export declare function defineTraceAnalyst(definition: TraceAnalystDefinition): TraceAnalystDefinition;
1247
1247
  interface DefineCustomAnalystOptions {
1248
1248
  id: string;
1249
1249
  description: string;
@@ -1256,8 +1256,8 @@ interface DefineExactCustomAnalystOptions extends DefineCustomAnalystOptions {
1256
1256
  executionConfig: Readonly<Record<string, unknown>>;
1257
1257
  }
1258
1258
  /** Construct a registrable analyst from a hand-written analyze function. */
1259
- declare function defineCustomAnalyst(options: DefineExactCustomAnalystOptions): ExactCapableAnalyst<TraceAnalysisStore>;
1260
- declare function defineCustomAnalyst(options: DefineCustomAnalystOptions): Analyst<TraceAnalysisStore>;
1259
+ export declare function defineCustomAnalyst(options: DefineExactCustomAnalystOptions): ExactCapableAnalyst<TraceAnalysisStore>;
1260
+ export declare function defineCustomAnalyst(options: DefineCustomAnalystOptions): Analyst<TraceAnalysisStore>;
1261
1261
  //#endregion
1262
1262
  //#region src/analyst/kinds/skill-usage.d.ts
1263
1263
  type SkillKind = 'public' | 'private';
@@ -1307,9 +1307,9 @@ interface SkillUsageScanConfig {
1307
1307
  maxTranscriptsPerDir?: number;
1308
1308
  }
1309
1309
  /** Scan the corpus + skill roots into a {@link SkillUsageReport}. Deterministic. */
1310
- declare function buildSkillUsageReport(config: SkillUsageScanConfig): SkillUsageReport;
1310
+ export declare function buildSkillUsageReport(config: SkillUsageScanConfig): SkillUsageReport;
1311
1311
  /** Pure rule pass over a report → findings. Exported for direct/unit use. */
1312
- declare function emitSkillUsageFindings(report: SkillUsageReport, producedAt: string): AnalystFinding[];
1312
+ export declare function emitSkillUsageFindings(report: SkillUsageReport, producedAt: string): AnalystFinding[];
1313
1313
  declare class SkillUsageAnalyst implements ExactCapableAnalyst<SkillUsageReport> {
1314
1314
  readonly id = "skill-usage";
1315
1315
  readonly description = "Deterministic multi-signal skill-usage analysis: flags dead skills, measurement-invisible (orchestrated) usage, discovery gaps, public-repo leaks, bloat, missing evals, and missing run-logging.";
@@ -1326,7 +1326,7 @@ declare class SkillUsageAnalyst implements ExactCapableAnalyst<SkillUsageReport>
1326
1326
  };
1327
1327
  analyze(input: SkillUsageReport, ctx: AnalystContext): Promise<AnalystFinding[]>;
1328
1328
  }
1329
- declare const SKILL_USAGE_ANALYST: SkillUsageAnalyst;
1329
+ export declare const SKILL_USAGE_ANALYST: SkillUsageAnalyst;
1330
1330
  //#endregion
1331
1331
  //#region src/analyst/parse-tolerant.d.ts
1332
1332
  /**
@@ -1341,22 +1341,22 @@ declare const SKILL_USAGE_ANALYST: SkillUsageAnalyst;
1341
1341
  * Pure + deterministic. No model, no network.
1342
1342
  */
1343
1343
  /** Strip a ```lang ... ``` (or bare ``` ... ```) code fence, if the string is one. */
1344
- declare function stripCodeFences(text: string): string;
1344
+ export declare function stripCodeFences(text: string): string;
1345
1345
  /**
1346
1346
  * Best-effort parse of a string into JSON. De-fences, drops trailing commas,
1347
1347
  * then `JSON.parse`. Returns `undefined` (never throws) when unrecoverable.
1348
1348
  */
1349
- declare function coerceJson(text: string): unknown;
1349
+ export declare function coerceJson(text: string): unknown;
1350
1350
  //#endregion
1351
1351
  //#region src/analyst/proposal-findings.d.ts
1352
1352
  /** True when a finding names a source candidate generation may learn from. */
1353
- declare function isProposalFinding(finding: unknown): finding is ProposalFinding;
1353
+ export declare function isProposalFinding(finding: unknown): finding is ProposalFinding;
1354
1354
  /**
1355
1355
  * Reject findings whose source has not been explicitly admitted for candidate
1356
1356
  * generation. Search feedback and observed production behavior are allowed;
1357
1357
  * final evaluation data has no allowed origin.
1358
1358
  */
1359
- declare function assertProposalFindings(findings: unknown, context?: string): ReadonlyArray<ProposalFinding>;
1359
+ export declare function assertProposalFindings(findings: unknown, context?: string): ReadonlyArray<ProposalFinding>;
1360
1360
  //#endregion
1361
- export { AGENT_RX_UPSTREAM_REVISION, ANALYST_BENCHMARK_COST_LEDGER_FILE, ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM, ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES, ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256, ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM, ANALYST_BENCHMARK_IMPLEMENTATION_FILES, ANALYST_BENCHMARK_IMPLEMENTATION_SHA256, ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, ANALYST_BENCHMARK_MANIFEST_FILE, ANALYST_BENCHMARK_OBSERVATIONS_FILE, type AgentRxBenchmarkCaseOptions, type AgentRxCalibrationRunnerSummary, type AgentRxCalibrationSummary, type AgentRxFailure, type AgentRxPrediction, type AgentRxPredictionReport, type AgentRxRow, type Analyst, type AnalystBenchmarkArtifact, type AnalystBenchmarkCase, type AnalystBenchmarkCommandConfig, type AnalystBenchmarkCommandDependencies, type AnalystBenchmarkDatasetRef, type AnalystBenchmarkDescriptor, type AnalystBenchmarkError, type AnalystBenchmarkLabelState, type AnalystBenchmarkLocalRunReceipt, type AnalystBenchmarkObservation, type AnalystBenchmarkOutput, type AnalystBenchmarkProgressRow, type AnalystBenchmarkProvenance, type AnalystBenchmarkResult, type AnalystBenchmarkRunIdentity, type AnalystBenchmarkRunManifest, type AnalystBenchmarkRunner, type AnalystBenchmarkSummary, type AnalystBudgetDeclaration, type AnalystComparisonMetric, type AnalystContext, type AnalystCost, type AnalystDefinition, type AnalystDefinitionAsymmetry, type AnalystDefinitionAsymmetryReport, type AnalystEvidenceBinding, type AnalystEvidenceExpectation, type AnalystEvidenceResolution, type AnalystEvidenceResolutionError, type AnalystEvidenceResolver, AnalystExpressivenessError, type AnalystFinding, type AnalystFindingScore, type AnalystHooks, type AnalystInputKind, type AnalystInstructionsOverride, type AnalystIssueExpectation, type AnalystLatencyDistribution, type AnalystMetricComparison, type AnalystProfileFragment, AnalystRegistry, type AnalystRegistryOptions, type AnalystRepairDeclaration, type AnalystRequirements, type AnalystRowExpansion, type AnalystRunEvent, type AnalystRunInputs, type AnalystRunResult, type AnalystRunSummary, type AnalystRunnerComparison, type AnalystSeverity, type AnalystTransportBinding, type AnalystUsageReceipt, type AskTraceQuestionsOptions, type BehavioralAnalystOptions, type BudgetPolicy, CODE_TRACE_BENCH_ANALYST_PROMPT, CONTROL_INTEGRITY_ANALYST, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTraceEngineOptions, type ChatTransport, type ChunkedEvidenceBinding, type CodeTraceBenchCaseOptions, type CodeTraceBenchLabelOptions, type CodeTraceBenchLabelSet, type CodeTraceBenchRow, type CodeTraceBlockDiagnostics, type CodeTraceCalibrationRunnerSummary, type CodeTraceCalibrationSummary, type CodeTraceFailureBlock, type CodeTraceStageAnnotation, type CodeTracerLabelGroup, type CodeTracerPredictionAdapterOptions, type CodeTracerPredictions, type CodeTracerStepLabel, ControlIntegrityAnalyst, type CreateChatClientOpts, type CreateTraceAnalystOptions, type CustomTransportOpts, DEFAULT_MAX_VERIFICATION_ARTIFACT_BYTES, DEFAULT_TRACE_ANALYST_KINDS, DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, type DecodedReply, type DefaultAnalystRegistryOptions, type DefineCustomAnalystOptions, type DefineExactCustomAnalystOptions, type DiffPolicy, type DspyRlmTraceEngineOptions, type EvidenceProjection, type EvidenceRef, type ExactAnalystBudgetPolicy, type ExactAnalystBudgetSnapshot, type ExactAnalystExecutionPlanSnapshot, type ExactAnalystRunCompletion, type ExactAnalystRunEvent, ExactAnalystRunExecutionError, type ExactAnalystRunPolicySnapshot, type ExactAnalystRunResult, type ExactAnalystRunSummary, type ExactAnalystSnapshot, type ExactCapableAnalyst, type ExactExecutionComponentIdentity, type ExactExecutionComponentSnapshot, type ExactRegistryRunOpts, type ExecutionProbe, type ExecutionProbeOutcome, type ExecutionProbeRequest, type ExpandRowsArgs, FAILURE_MODE_KIND_SPEC, FINDING_SUBJECT_KINDS, FINDING_SUBJECT_SYNTAX, type FindingSubject, type FindingSubjectKind, type FindingsDiff, FindingsStore, IMPROVEMENT_KIND_SPEC, type InlineEvidenceBinding, KIND_EXPECTED_SUBJECTS, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, type LoadedVerificationArtifacts, MAX_INCORRECT_BLOCKS, MAX_INCORRECT_BLOCK_STEPS, type MockTransportOpts, type OpenAiCompatibleTransportOpts, type PersistedFinding, type PreparedPublicAnalystBenchmark, type PrimeBenchmarkRunnerOptions, type PrimeBridgeTransport, type PrimeBridgeTransportRequest, type PrimeBridgeTransportResult, type PrimeCodeTraceDefinitionArgs, type PrimeExchangeOptions, type PrimeExchangeOutcome, type PrimeFailure, type PrimeProjectionDelivery, type PrimeProjectionOutcome, type PrimeProjectionSource, type PrimePromptSpec, type PrimeProtocolIdentity, type PrimeRawUsage, type PrimeRejectedRow, type PrimeRepairPromptSpec, type PrimeRepairState, type PrimeReplyContract, type PrimeRowDecoded, type PrimeTurnRecord, type ProposalFinding, type ProposalFindingOrigin, type PublicAnalystBenchmarkDataset, type PublicAnalystBenchmarkModelConfig, type PublicAnalystBenchmarkModelOwner, type PublicAnalystBenchmarkModelSettings, type PublicBenchmarkDistributions, type PublicBenchmarkModelPrediction, type PublicBenchmarkSelectionReport, type PublicBenchmarkValueDistribution, type PublicDirectDefinitionArgs, type PublicRlmDefinitionArgs, RAW_FINDING_SCHEMA_PROMPT, type RawAnalystEvidence, type RawAnalystFinding, RawAnalystFindingSchema, type RegistryRunOpts, type ReplVariableConsensusPort, type ReplVariableEvidenceBinding, type ReplyContract, type ReplyEnvelope, type ReplyRowDecoded, type ReplyRowRejection, type RunAnalystBenchmarkOptions, SKILL_USAGE_ANALYST, type SandboxSdkTransportOpts, type SemanticConceptJudgeAdapterOpts, type SkillUsageRecord, type SkillUsageReport, type SkillUsageScanConfig, type StepLabelAdapterOptions, type TraceAnalysisEngine, type TraceAnalysisEngineRequest, type TraceAnalysisEngineResult, type TraceAnalystDefinition, type TraceAnalystLimits, type TraceQuestionOutcome, type TraceQuestionSpec, type TraceToolGroupName, type UpstreamPredictionAdapterOptions, type VerificationArtifactFile, type VerificationArtifactManifest, type VerificationArtifactRole, type VerificationAvailabilitySummary, type VerificationOutcome, type VerificationOutcomeSource, type VerificationOutcomeStatus, type VerificationResultFile, type VerifiedTraceFinding, adaptPublicBenchmarkFindings, agentRxBenchmarkCase, agentRxPredictionsToFindings, analystBenchmarkDependencyLockDigest, analystBenchmarkImplementationDigest, analystDefinitionAsymmetries, analystDefinitionProtocolSha256, analystInstructionsOverrideFromText, analystUsageReceiptFromPrimeUsage, appendVerificationArtifactsToOtlp, askTraceQuestions, assertProposalFindings, behavioralAnalyst, bindAnalyst, buildDefaultAnalystRegistry, buildPrimePrompt, buildPrimeRepairPrompt, buildSkillUsageReport, buildTraceToolsForGroup, citedSpansOnly, codeTraceBenchCase, codeTracerPredictionsToFindings, coerceJson, compareAnalystRunners, computeFindingId, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createPrimeBenchmarkRunner, createPublicBenchmarkDirectRunner, createPublicBenchmarkRlmRunner, createSemanticConceptJudgeAdapter, createTraceAnalyst, decodeReplyRows, defaultIsMaterial, defineCustomAnalyst, defineTraceAnalyst, deriveEfficiencyFindings, diffFindings, effectiveAnalystProtocolSha256, emitControlIntegrityFindings, emitSkillUsageFindings, emptyPrimeRawUsage, emptyPublicBenchmarkRunner, evidenceRefsFromRawFinding, expandCodeTraceFailureBlocks, extractPrimeJsonObject, findingSubjectGrammarPromptFor, isProposalFinding, loadCodeTraceVerificationArtifacts, loadPublicBenchmarkRows, makeFinding, makeProposalFinding, mergePrimeRawUsage, nodeHttpPrimeBridgeTransport, normalizeAgentRxCategory, normalizeBenchmarkLabel, normalizePrimeUsage, parseFindingSubject, parseRawFinding, parseVerificationOutcome, preparePublicAnalystBenchmark, primeAnalystProtocolSha256, primeCodeTraceAnalystDefinition, primeProtocolSha256, primeReplyDefect, projectPrimeTrajectory, publicBenchmarkDistributions, publicBenchmarkProtocolSha256, publicBenchmarkRlmInstructions, publicBenchmarkSelectionReport, publicBenchmarkSystemPrompt, publicDirectAnalystDefinition, publicRlmAnalystDefinition, readAnalystBenchmarkArtifact, readAnalystInstructionsOverride, registryBenchmarkRunner, renderAgentRxCalibrationMarkdown, renderAnalystBenchmarkMarkdown, renderCodeTraceCalibrationMarkdown, renderFindingSubject, renderPriorFindings, resolveTraceAnalystLimits, rlmEngineLimits, roundAgentRxStep, runAnalystBenchmark, runAnalystBenchmarkCommand, runPrimeExchange, runTraceAnalyst, scoreAnalystFindings, selectPublicBenchmarkRows, stripCodeFences, summarizeAgentRxCalibration, summarizeAnalystBenchmarkRunner, summarizeCodeTraceCalibration, traceStoreEvidenceResolver };
1361
+ export { type AgentRxBenchmarkCaseOptions, type AgentRxCalibrationRunnerSummary, type AgentRxCalibrationSummary, type AgentRxFailure, type AgentRxPrediction, type AgentRxPredictionReport, type AgentRxRow, type Analyst, type AnalystBenchmarkArtifact, type AnalystBenchmarkCase, type AnalystBenchmarkCommandConfig, type AnalystBenchmarkCommandDependencies, type AnalystBenchmarkDatasetRef, type AnalystBenchmarkDescriptor, type AnalystBenchmarkError, type AnalystBenchmarkLabelState, type AnalystBenchmarkLocalRunReceipt, type AnalystBenchmarkObservation, type AnalystBenchmarkOutput, type AnalystBenchmarkProgressRow, type AnalystBenchmarkProvenance, type AnalystBenchmarkResult, type AnalystBenchmarkRunIdentity, type AnalystBenchmarkRunManifest, type AnalystBenchmarkRunner, type AnalystBenchmarkSummary, type AnalystBudgetDeclaration, type AnalystComparisonMetric, type AnalystContext, type AnalystCost, type AnalystDefinition, type AnalystDefinitionAsymmetry, type AnalystDefinitionAsymmetryReport, type AnalystEvidenceBinding, type AnalystEvidenceExpectation, type AnalystEvidenceResolution, type AnalystEvidenceResolutionError, type AnalystEvidenceResolver, type AnalystFinding, type AnalystFindingScore, type AnalystHooks, type AnalystInputKind, type AnalystInstructionsOverride, type AnalystIssueExpectation, type AnalystLatencyDistribution, type AnalystMetricComparison, type AnalystProfileFragment, AnalystRegistry, type AnalystRegistryOptions, type AnalystRepairDeclaration, type AnalystRequirements, type AnalystRowExpansion, type AnalystRunEvent, type AnalystRunInputs, type AnalystRunResult, type AnalystRunSummary, type AnalystRunnerComparison, type AnalystSeverity, type AnalystTransportBinding, type AnalystUsageReceipt, type AskTraceQuestionsOptions, type BehavioralAnalystOptions, type BudgetPolicy, CONTROL_INTEGRITY_ANALYST, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTraceEngineOptions, type ChatTransport, type ChunkedEvidenceBinding, type CodeTraceBenchCaseOptions, type CodeTraceBenchLabelOptions, type CodeTraceBenchLabelSet, type CodeTraceBenchRow, type CodeTraceBlockDiagnostics, type CodeTraceCalibrationRunnerSummary, type CodeTraceCalibrationSummary, type CodeTraceFailureBlock, type CodeTraceStageAnnotation, type CodeTracerLabelGroup, type CodeTracerPredictionAdapterOptions, type CodeTracerPredictions, type CodeTracerStepLabel, ControlIntegrityAnalyst, type CreateChatClientOpts, type CreateTraceAnalystOptions, type CustomTransportOpts, DEFAULT_TRACE_ANALYST_KINDS, DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, type DecodedReply, type DefaultAnalystRegistryOptions, type DefineCustomAnalystOptions, type DefineExactCustomAnalystOptions, type DiffPolicy, type DspyRlmTraceEngineOptions, type EvidenceProjection, type EvidenceRef, type ExactAnalystBudgetPolicy, type ExactAnalystBudgetSnapshot, type ExactAnalystExecutionPlanSnapshot, type ExactAnalystRunCompletion, type ExactAnalystRunEvent, ExactAnalystRunExecutionError, type ExactAnalystRunPolicySnapshot, type ExactAnalystRunResult, type ExactAnalystRunSummary, type ExactAnalystSnapshot, type ExactCapableAnalyst, type ExactExecutionComponentIdentity, type ExactExecutionComponentSnapshot, type ExactRegistryRunOpts, type ExecutionProbe, type ExecutionProbeOutcome, type ExecutionProbeRequest, type ExpandRowsArgs, FAILURE_MODE_KIND_SPEC, FINDING_SUBJECT_KINDS, FINDING_SUBJECT_SYNTAX, type FindingSubject, type FindingSubjectKind, type FindingsDiff, FindingsStore, IMPROVEMENT_KIND_SPEC, type InlineEvidenceBinding, KIND_EXPECTED_SUBJECTS, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, type LoadedVerificationArtifacts, type MockTransportOpts, type OpenAiCompatibleTransportOpts, type PersistedFinding, type PreparedPublicAnalystBenchmark, type PrimeBenchmarkRunnerOptions, type PrimeBridgeTransport, type PrimeBridgeTransportRequest, type PrimeBridgeTransportResult, type PrimeCodeTraceDefinitionArgs, type PrimeExchangeOptions, type PrimeExchangeOutcome, type PrimeFailure, type PrimeProjectionDelivery, type PrimeProjectionOutcome, type PrimeProjectionSource, type PrimePromptSpec, type PrimeProtocolIdentity, type PrimeRawUsage, type PrimeRejectedRow, type PrimeRepairPromptSpec, type PrimeRepairState, type PrimeReplyContract, type PrimeRowDecoded, type PrimeTurnRecord, type ProposalFinding, type ProposalFindingOrigin, type PublicAnalystBenchmarkDataset, type PublicAnalystBenchmarkModelConfig, type PublicAnalystBenchmarkModelOwner, type PublicAnalystBenchmarkModelSettings, type PublicBenchmarkDistributions, type PublicBenchmarkModelPrediction, type PublicBenchmarkSelectionReport, type PublicBenchmarkValueDistribution, type PublicDirectDefinitionArgs, type PublicRlmDefinitionArgs, RAW_FINDING_SCHEMA_PROMPT, type RawAnalystEvidence, type RawAnalystFinding, RawAnalystFindingSchema, type RegistryRunOpts, type ReplVariableConsensusPort, type ReplVariableEvidenceBinding, type ReplyContract, type ReplyEnvelope, type ReplyRowDecoded, type ReplyRowRejection, type RunAnalystBenchmarkOptions, type SandboxSdkTransportOpts, type SemanticConceptJudgeAdapterOpts, type SkillUsageRecord, type SkillUsageReport, type SkillUsageScanConfig, type StepLabelAdapterOptions, type TraceAnalysisEngine, type TraceAnalysisEngineRequest, type TraceAnalysisEngineResult, type TraceAnalystDefinition, type TraceAnalystLimits, type TraceQuestionOutcome, type TraceQuestionSpec, type TraceToolGroupName, type UpstreamPredictionAdapterOptions, type VerificationArtifactFile, type VerificationArtifactManifest, type VerificationArtifactRole, type VerificationAvailabilitySummary, type VerificationOutcome, type VerificationOutcomeSource, type VerificationOutcomeStatus, type VerificationResultFile, type VerifiedTraceFinding, analystUsageReceiptFromPrimeUsage, askTraceQuestions, behavioralAnalyst, buildDefaultAnalystRegistry, buildPrimePrompt, buildPrimeRepairPrompt, buildTraceToolsForGroup, citedSpansOnly, computeFindingId, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createTraceAnalyst, decodeReplyRows, defaultIsMaterial, deriveEfficiencyFindings, diffFindings, emitControlIntegrityFindings, emptyPrimeRawUsage, evidenceRefsFromRawFinding, extractPrimeJsonObject, findingSubjectGrammarPromptFor, makeFinding, makeProposalFinding, mergePrimeRawUsage, nodeHttpPrimeBridgeTransport, normalizePrimeUsage, parseFindingSubject, parseRawFinding, primeProtocolSha256, primeReplyDefect, projectPrimeTrajectory, registryBenchmarkRunner, renderFindingSubject, renderPriorFindings, resolveTraceAnalystLimits, runAnalystBenchmark, runPrimeExchange, runTraceAnalyst, scoreAnalystFindings, traceStoreEvidenceResolver };
1362
1362
  //# sourceMappingURL=index.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","names":[],"sources":["../../src/analyst/adapters.ts","../../src/analyst/benchmark-agentrx-calibration.ts","../../src/analyst/benchmark-dataset-types.ts","../../src/analyst/benchmark-dataset-agentrx.ts","../../src/analyst/benchmark-dataset-codetrace.ts","../../src/analyst/benchmark-dataset-utils.ts","../../src/analyst/benchmark-verification-outcome.ts","../../src/analyst/benchmark-verification-artifacts.ts","../../src/analyst/benchmark-public-types.ts","../../src/analyst/benchmark-public-adapters.ts","../../src/analyst/benchmark-public-data.ts","../../src/analyst/definition.ts","../../src/analyst/benchmark-public-prompt.ts","../../src/analyst/benchmark-public-model.ts","../../src/analyst/benchmark-public-rlm.ts","../../src/analyst/benchmark-runner-prime.ts","../../src/analyst/benchmark-comparison.ts","../../src/analyst/benchmark-public-calibration.ts","../../src/analyst/benchmark-command-artifact.ts","../../src/analyst/benchmark-command-result.ts","../../src/analyst/benchmark-command.ts","../../src/analyst/benchmark-implementation.ts","../../src/analyst/benchmark-instructions-override.ts","../../src/analyst/benchmark-report.ts","../../src/analyst/benchmark-summary.ts","../../src/analyst/bind.ts","../../src/analyst/define.ts","../../src/analyst/kinds/skill-usage.ts","../../src/analyst/parse-tolerant.ts","../../src/analyst/proposal-findings.ts"],"mappings":";;;;;;;;;;;;;;;UA8CiB;EACf;EACA;;EAEA,SAAS,KAAK;;EAEd;;iBAGc,kCACd,MAAM,kCACL,QAAQ;;;cCtDE;UAEI;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA,SAAS;;iBAGK,4BACd,QAAQ,wBACR,2BACC;iBAkBa,iCAAiC,SAAS;;;KCtD9C;UAEK;EACf,YAAY;EACZ;EACA;EACA;EACA;EACA;;UAGe;EACf,eAAe;EACf,mBAAmB;EACnB;IAAe,YAAY;IAAY;;EACvC,wBAAwB;EACxB;EACA;EACA;;UAGe;EACf,UAAU;EACV;EACA;EACA;EACA;;UAGe;EACf,UAAU;EACV,mBAAmB;EACnB;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA,iBAAiB;;KAGP,0CAEC,sCACA,iCACA;UAEI;EACf;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA,oCAAoC;;UAGrB;EACf,eAAe;EACf,WAAW,sBAAsB;;KAGvB;UAEK;;;;;EAKf,WAAW;;UAGI,kCACP,yBACN;UAEa,oCAAoC;EACnD;;EAEA;;UAGe,yCAAyC;EACxD;EACA;EACA;EACA;;UAGe,2CACP,kCACN;;;iBCtGY,qBAAqB,QACnC,KAAK,YACL,OAAO,QACP,UAAS,8BACR,qBAAqB;;iBAsGR,6BACd,mBAAmB,YACnB,iBACA,UAAS,mCACR;iBA6Ea,yBAAyB;;iBAoNzB,iBAAiB;;;iBC/YjB,mBAAmB,QACjC,KAAK,mBACL,OAAO,QACP,UAAS,4BACR,qBAAqB;;iBAwFR,gCACd,2BACA,aAAa,uBACb,UAAS,qCACR;;;iBClHa,wBAAwB;;;KCA5B;UAEK;EACf;EACA;EACA,QAAQ;;UAGO;EACf,QAAQ;EACR;EAKA;IAAe;IAAe;;EAC9B,SAAS;EACT;EACA;EACA;EACA;;UAGe;EACf;EACA;;iBA8Ec,yBACd,gBAAgB,2BACf;;;cC/FU;KAED;UAEK;EACf,MAAM;EACN;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;EACT;EACA;EACA;EACA;EACA;EACA,OAAO;EACP,cAAc;EACd,UAAU,OAAO;;UAGF;EACf,UAAU;EACV,SAAS;EACT,OAAO,MAAM;IAA6B;;;iBAatB,mCAAmC;EACvD;EACA,KAAK;EACL;IACE,QAAQ;iBAwHI,kCACd,kBACA,iBACA,WAAW,6BACX;;;KC7KU;;;;;;;;UASK;;WAEN;;WAEA;;;UAIM;EACf,MAAM;EACN;EACA,kBAAkB,aAAa;;EAE/B,UAAU;;UAGK;;EAEf,MAAM;;EAEN;;EAEA,kBAAkB,aAAa;EAC/B;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,UAAU;;EAEV;;EAEA,uBAAuB;EACvB;IACE,SAAS;IACT;IACA;IACA;IACA;IACA;IACA;IACA;IACA;;;;;;IAMA;;EAEF,aAAa;EACb;IACE;IACA;;;;;;;;;KAUQ,sCAAsC,KAChD,iEAGA,QAAQ,KAAK;UAEE;EACf,OAAO,qBAAqB;EAC5B;EACA;EACA;EACA,YAAY;IACV;IACA;IACA;;EAEF,uBAAuB;EACvB,WAAW;;UAGI;EACf;EACA;EACA,QAAQ;;UAGO;EACf,OAAO;EACP,OAAO;EACP,OAAO;EACP,YAAY;EACZ,QAAQ;;UAGO;EACf;EACA;EACA;EACA;EACA;EACA;EACA,QAAQ;EACR,UAAU;;;;;;;;;;UChHK;EACf;EACA;;EAEA;EACA;EACA,UAAU;EACV;EACA;EACA;EACA;EACA,WAAW;;;;;;;;;UAUI;EACf;EACA;;EAEA,kCAAkC;;EAElC;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;UASe;EACf;EACA,OAAO;;iBAGO,8BAA8B,uBAAuB;iBAiB/C,6BAA6B;EACjD,SAAS;EACT;EACA,mBAAmB;EACnB;EACA,OAAO;EACP,SAAS;IACP;EACF,UAAU;EACV,aAAa;;EAEb,aAAa;;;;;;;;;iBA4MO,6BAA6B;EACjD;EACA,iBAAiB;EACjB,OAAO;EACP;EACA;EACA,SAAS;IACP;EACF,UAAU;EACV,aAAa;EACb,YAAY;;;;iBCrQQ,wBACpB,eACC,QAAQ,MAAM;iBA2BD,0BACd,SAAS,+BACT,eAAe,2BACf;EAAW;EAAe;IACzB,MAAM;iBAwBO,6BACd,SAAS,+BACT,eAAe,4BACd;iBAyCa,+BACd,SAAS,+BACT,iBAAiB,2BACjB,mBAAmB,2BACnB,eACC;iBAcmB,8BAA8B;EAClD,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;IACE,QAAQ;;;;;;;;KCnIA,yBAAyB,KAAK;;;;;;KAS9B;WAEG;;WAEA;;;;;;WAMA;;WAGA;;WAEA;;;WAIA;WACA,WAAW;;;WAIX;WACA,WAAW;;UAKT;;WAEN;;WAEA;;WAEA;;WAEA,eAAe;;UAGT;;;;;WAKN;;UAKM;EACf,UAAU;;EAEV;;UAGe,eAAe;;EAE9B;EACA,eAAe;EACf,OAAO;EACP;EACA;;EAEA;EACA,SAAS;;;UAIM,sBAAsB;WAC5B;EACT,kBAAkB;;WAET,cAAc,SAAS;;;;;EAKhC,OAAO,iBAAiB,gBAAgB;;EAExC,QAAQ,iBAAiB,gBAAgB;EACzC,WAAW,MAAM,eAAe,QAAQ,QAAQ;;;UAIjC,uBAAuB;WAC7B;EACT,kBAAkB;WACT,cAAc,SAAS;;WAEvB;;WAEA;;EAET,YAAY;EACZ,WAAW,MAAM,eAAe,QAAQ,QAAQ;;EAEhD,gBAAgB;IACd;IACA,mBAAmB;IACnB,OAAO;IACP,SAAS;MACP;;;UAIW,0BAA0B,aAAa;;EAEtD,KAAK,SAAS,uBAAuB;IACnC,iBAAiB;IACjB;;;EAGF,OAAO;IACL;IACA,iBAAiB;IACjB,OAAO;IACP;IACA;IACA,SAAS;MACP,QAAQ;;EAEZ,aAAa,sBAAsB,gBAAgB;;;UAIpC,4BAA4B,uBAAuB;WACzD;;WAEA;EACT,kBAAkB;WACT,cAAc,SAAS;;WAEvB,qBAAqB,SAAS;;WAE9B;;EAET,qBAAqB,8BAA8B;;EAEnD,MAAM;IACJ;IACA,mBAAmB;IACnB;IACA,OAAO;IACP,SAAS;MACP;IAAU,UAAU;IAAkB,aAAa;IAAe;;;EAEtE,YAAY,0BAA0B,aAAa;;EAEnD,mBACE,QAAQ,oCACP,uBAAuB;;KAGhB,uBAAuB,MAAM,uBAAuB,oBAC5D,sBAAsB,QACtB,uBAAuB,QACvB,4BAA4B,aAAa;UAI5B,kBAAkB,gBAAgB,uBAAuB;;WAE/D;WACA;WACA;;WAEA;WACA,SAAS;;WAET;;WAEA;WACA,YAAY;WACZ,eAAe,cAAc;;;;;;WAM7B,gBAAgB,SAAS;WACzB,QAAQ;WACR,QAAQ;;;;;;WAMR;WACA,SAAS,uBAAuB,MAAM,aAAa;;;;;;;;;cAUjD,mCAAmC;;;;;;;iBAUhC,gCAAgC,MAAM,aAAa,QACjE,YAAY,kBAAkB,MAAM,aAAa;;UAwClC;WACN;WACA,gBAAgB;WAChB,iBAAiB,YAAY;WAC7B;WACA;WACA;;WAEA;;WAEA;;UAGM;WACN;;WAEA;;WAEA,sBAAsB;WACtB,sBAAsB;;;;;;;;;iBAUjB,6BACd,aAAa,cAAc,gDAC1B;;;;;;;cCzUU;;;;cAKA;cAMA;;iBA8GG,4BAA4B,SAAS;;iBASrC,+BAA+B,SAAS;;;;iBAcxC,8BAA8B,SAAS;;;;;;;;;;;;;UC5EtC;;EAEf;;EAEA;;EAEA;;;iBAIc,8BACd,SAAS,+BACT,MAAM,6BACL,kBAAkB;;iBAsEL,kCACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;cAwhBpB,yBAAuB,EAAA;;;;;;;;;;;;;GASlB,EAAA,KAAA;cACL,gCAA8B,EAAA;;;;;;;;;;;;;;;;;;;GAsChC,EAAA,KAAA;cACE,uBAAqB,EAAA;;;;;;;;;;;;KA2BtB,yBAAyB,EAAE,aAAa;EAC3C,WAAW,EAAE,aAAa;;KAEvB,2BAA2B,EAAE,aAAa;KACnC,iCAAiC,yBAAyB;;;;;;;;;;;;;UCjtBrD;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,cAAc;;;iBAIA,2BACd,SAAS,+BACT,MAAM,0BACL,kBAAkB,mBAAmB;;iBA+GxB,+BACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;;iBAwBV,gBAAgB,QAAQ,oCAAoC;;;UCtI3D;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA,UAAU;;EAEV,YAAY;;;;;;iBA8EE;UAIC;;EAEf;;EAEA;;;;;;iBAOc,gCACd,MAAM,+BACL,kBAAkB;;iBAyDL,2BACd,SAAS,8BACR,uBAAuB;;;;;;;;;;cC5OpB;WACQ;WACC;WACK;WACd;WACkB;WACJ;WACO;WACD;WACJ;WACK;WACd;WACJ;WACM;WACC;WACG;WACH;WACI;WACT;;KAGC,uCAAuC;UAgBlC;EACf,QAAQ;EACR;;EAEA;;EAEA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;;iBASK,sBACd,QAAQ,wBACR;EACE;EACA;EACA;EACA;EACA;IAED;;;UC3Fc;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;;iBAGK,8BACd,QAAQ,yBACP;iBAca,mCAAmC,SAAS;;;UCpC3C;EACf;EACA;EACA;IACE,SAAS;IACT;IACA;IACA;IACA;IACA,YAAY;MAAQ;MAAiB;MAAsB;;IAC3D,uBAAuB;IACvB,0BAA0B;IAC1B;MACE;MACA;MACA;MACA,QAAQ;;IAEV;MACE;MACA;;MAEA;MACA;;MAEA;MACA;MACA;MACA;MACA;MACA;MACA;MACA,UAAU;MACV;QACE;QACA;QACA;QACA;QACA;QACA;QACA;QACA;;MAEF;QACE;QACA;QACA;;MAEF;MACA;MACA;;MAEA;MACA;MACA;;;EAGJ,QAAQ;EACR,aAAa;EACb,uBAAuB;EACvB,qBAAqB;;UAGN;EACf;EACA;EACA;EACA;IACE;IACA;IACA;;;UAIa;EACf;IACE,SAAS;IACT;IACA;IACA;MACE;MACA;MACA;MACA;MACA;MACA;MACA;MACA;MACA,SAAS;MACT;QACE;QACA;QACA;QACA;QACA;QACA;QACA;QACA;;MAEF;QACE;QACA;QACA;;;IAGJ;IACA;IACA;IACA;;IAEA;IACA;IACA;IACA;;IAEA;IACA;IACA;IACA;;EAEF;IACE;IACA;IACA;IACA,YAAY;MAAQ;MAAiB;MAAsB;;IAC3D;IACA;;;UAIa;EACf;EACA;EACA;EACA;EACA,UAAU;;UAGK;EACf;EACA;EACA;EACA;IACE;IACA;IACA;IACA;;IAEA;;EAEF;EACA;IACE;IACA;IACA;;EAEF;IACE;IACA;IACA;IACA;IACA;IACA;;;UAIa;EACf;EACA;EACA;EACA,aAAa;EACb;;cAGW;cACA;cACA;cACA;;;iBC5KS,6BACpB,eACC,QAAQ;;;UCuEM;EACf,uBACE,SAAS,+BACT,QAAQ,wCACL,uBAAuB;EAC5B,2BACE,mBACA;IACE;IACA,aAAa,SAAS,OAAO;QAE5B,QAAQ;;;;;;;;;KAUH;;UAGK;EACf;EACA;;UAGe;EACf,SAAS;EACT,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;EACA,OAAO;EACP;EACA;EACA;EACA;;EAEA;EACA;EACA;;EAEA;;EAEA,QAAQ;EACR;EACA;;iBAKoB,2BACpB,yBACA,MAAK,OAAO,YACZ,eAAc,sCACb;;;cCzJU;;;cAIA;cAGA;cAOA;cAYA;cAGA;cAGA;cAiHA;iBAGG;iBAIA;;;;iBC/IA,oCAAoC,eAAe;;iBAQnD,gCAAgC,eAAe;;;;;;;;;;;iBAyB/C,+BACd,SAAS,+BACT,WAAW,KAAK;;;iBCzCF,+BACd,QAAQ,wBACR,uBAAsB;;;iBCIR,gCACd,kBACA,uBAAuB,gCACtB;;;;KCcS;;WAGG;WACA;WACA;WACA,YAAY;WACZ,UAAU;;;WAIV;WACA,QAAQ;;;;;;iBAOP,YAAY,MAAM,aAAa,QAC7C,YAAY,kBAAkB,MAAM,aAAa,SACjD,YAAY,0BACX,uBAAuB;;;;;;;;;;iBCpCV,mBAAmB,YAAY,yBAAyB;UAkBvD;EACf;EACA;EACA;EACA,MAAM;EACN,SAAS,QAAQ;;UAGF,wCAAwC;;EAEvD,iBAAiB,SAAS;;;iBAIZ,oBACd,SAAS,kCACR,oBAAoB;iBACP,oBACd,SAAS,6BACR,QAAQ;;;KCpBC;;UAGK;EACf;EACA,MAAM;;EAEN;EACA;;EAEA;;EAEA;;;EAGA;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;UAGe;EACf;EACA,SAAS;;UAGM;;EAEf;;EAEA;IAAc;IAAc,MAAM;;;EAElC;;;EAGA,kBAAkB;;EAElB;;;iBA2Ec,sBAAsB,QAAQ,uBAAuB;;iBAiHrD,uBACd,QAAQ,kBACR,qBACC;cA2HG,6BAA6B,oBAAoB;WAC5C;WACA;WAEA;WACA;IAAS;IAAgC;;WACzC;WACA;aACP;aACA;aACA;;EAGI,QAAQ,OAAO,kBAAkB,KAAK,iBAAiB,QAAQ;;cAS1D,qBAAmB;;;;;;;;;;;;;;;iBC5YhB,gBAAgB;;;;;iBAgBhB,WAAW;;;;iBCIX,kBAAkB,mBAAmB,WAAW;;;;;;iBAShD,uBACd,mBACA,mBACC,cAAc"}
1
+ {"version":3,"file":"index.d.ts","names":[],"sources":["../../src/analyst/adapters.ts","../../src/analyst/benchmark-agentrx-calibration.ts","../../src/analyst/benchmark-dataset-types.ts","../../src/analyst/benchmark-dataset-agentrx.ts","../../src/analyst/benchmark-dataset-codetrace.ts","../../src/analyst/benchmark-dataset-utils.ts","../../src/analyst/benchmark-verification-outcome.ts","../../src/analyst/benchmark-verification-artifacts.ts","../../src/analyst/benchmark-public-types.ts","../../src/analyst/benchmark-public-adapters.ts","../../src/analyst/benchmark-public-data.ts","../../src/analyst/definition.ts","../../src/analyst/benchmark-public-prompt.ts","../../src/analyst/benchmark-public-model.ts","../../src/analyst/benchmark-public-rlm.ts","../../src/analyst/benchmark-runner-prime.ts","../../src/analyst/benchmark-comparison.ts","../../src/analyst/benchmark-public-calibration.ts","../../src/analyst/benchmark-command-artifact.ts","../../src/analyst/benchmark-command-result.ts","../../src/analyst/benchmark-command.ts","../../src/analyst/benchmark-implementation.ts","../../src/analyst/benchmark-instructions-override.ts","../../src/analyst/benchmark-report.ts","../../src/analyst/benchmark-summary.ts","../../src/analyst/bind.ts","../../src/analyst/define.ts","../../src/analyst/kinds/skill-usage.ts","../../src/analyst/parse-tolerant.ts","../../src/analyst/proposal-findings.ts"],"mappings":";;;;;;;;;;;;;;;UA8CiB;EACf;EACA;;EAEA,SAAS,KAAK;;EAEd;;wBAGc,kCACd,MAAM,kCACL,QAAQ;;;qBCtDE;UAEI;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA,SAAS;;wBAGK,4BACd,QAAQ,wBACR,2BACC;wBAkBa,iCAAiC,SAAS;;;KCtD9C;UAEK;EACf,YAAY;EACZ;EACA;EACA;EACA;EACA;;UAGe;EACf,eAAe;EACf,mBAAmB;EACnB;IAAe,YAAY;IAAY;;EACvC,wBAAwB;EACxB;EACA;EACA;;UAGe;EACf,UAAU;EACV;EACA;EACA;EACA;;UAGe;EACf,UAAU;EACV,mBAAmB;EACnB;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA,iBAAiB;;KAGP,0CAEC,sCACA,iCACA;UAEI;EACf;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA,oCAAoC;;UAGrB;EACf,eAAe;EACf,WAAW,sBAAsB;;KAGvB;UAEK;;;;;EAKf,WAAW;;UAGI,kCACP,yBACN;UAEa,oCAAoC;EACnD;;EAEA;;UAGe,yCAAyC;EACxD;EACA;EACA;EACA;;UAGe,2CACP,kCACN;;;wBCtGY,qBAAqB,QACnC,KAAK,YACL,OAAO,QACP,UAAS,8BACR,qBAAqB;;wBAsGR,6BACd,mBAAmB,YACnB,iBACA,UAAS,mCACR;wBA6Ea,yBAAyB;;wBAoNzB,iBAAiB;;;wBC/YjB,mBAAmB,QACjC,KAAK,mBACL,OAAO,QACP,UAAS,4BACR,qBAAqB;;wBAwFR,gCACd,2BACA,aAAa,uBACb,UAAS,qCACR;;;wBClHa,wBAAwB;;;KCA5B;UAEK;EACf;EACA;EACA,QAAQ;;UAGO;EACf,QAAQ;EACR;EAKA;IAAe;IAAe;;EAC9B,SAAS;EACT;EACA;EACA;EACA;;UAGe;EACf;EACA;;wBA8Ec,yBACd,gBAAgB,2BACf;;;qBC/FU;KAED;UAEK;EACf,MAAM;EACN;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;EACT;EACA;EACA;EACA;EACA;EACA,OAAO;EACP,cAAc;EACd,UAAU,OAAO;;UAGF;EACf,UAAU;EACV,SAAS;EACT,OAAO,MAAM;IAA6B;;;wBAatB,mCAAmC;EACvD;EACA,KAAK;EACL;IACE,QAAQ;wBAwHI,kCACd,kBACA,iBACA,WAAW,6BACX;;;KC7KU;;;;;;;;UASK;;WAEN;;WAEA;;;UAIM;EACf,MAAM;EACN;EACA,kBAAkB,aAAa;;EAE/B,UAAU;;UAGK;;EAEf,MAAM;;EAEN;;EAEA,kBAAkB,aAAa;EAC/B;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,UAAU;;EAEV;;EAEA,uBAAuB;EACvB;IACE,SAAS;IACT;IACA;IACA;IACA;IACA;IACA;IACA;IACA;;;;;;IAMA;;EAEF,aAAa;EACb;IACE;IACA;;;;;;;;;KAUQ,sCAAsC,KAChD,iEAGA,QAAQ,KAAK;UAEE;EACf,OAAO,qBAAqB;EAC5B;EACA;EACA;EACA,YAAY;IACV;IACA;IACA;;EAEF,uBAAuB;EACvB,WAAW;;UAGI;EACf;EACA;EACA,QAAQ;;UAGO;EACf,OAAO;EACP,OAAO;EACP,OAAO;EACP,YAAY;EACZ,QAAQ;;UAGO;EACf;EACA;EACA;EACA;EACA;EACA;EACA,QAAQ;EACR,UAAU;;;;;;;;;;UChHK;EACf;EACA;;EAEA;EACA;EACA,UAAU;EACV;EACA;EACA;EACA;EACA,WAAW;;;;;;;;;UAUI;EACf;EACA;;EAEA,kCAAkC;;EAElC;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;UASe;EACf;EACA,OAAO;;wBAGO,8BAA8B,uBAAuB;wBAiB/C,6BAA6B;EACjD,SAAS;EACT;EACA,mBAAmB;EACnB;EACA,OAAO;EACP,SAAS;IACP;EACF,UAAU;EACV,aAAa;;EAEb,aAAa;;;;;;;;;wBA4MO,6BAA6B;EACjD;EACA,iBAAiB;EACjB,OAAO;EACP;EACA;EACA,SAAS;IACP;EACF,UAAU;EACV,aAAa;EACb,YAAY;;;;wBCrQQ,wBACpB,eACC,QAAQ,MAAM;wBA2BD,0BACd,SAAS,+BACT,eAAe,2BACf;EAAW;EAAe;IACzB,MAAM;wBAwBO,6BACd,SAAS,+BACT,eAAe,4BACd;wBAyCa,+BACd,SAAS,+BACT,iBAAiB,2BACjB,mBAAmB,2BACnB,eACC;wBAcmB,8BAA8B;EAClD,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;IACE,QAAQ;;;;;;;;KCnIA,yBAAyB,KAAK;;;;;;KAS9B;WAEG;;WAEA;;;;;;WAMA;;WAGA;;WAEA;;;WAIA;WACA,WAAW;;;WAIX;WACA,WAAW;;UAKT;;WAEN;;WAEA;;WAEA;;WAEA,eAAe;;UAGT;;;;;WAKN;;UAKM;EACf,UAAU;;EAEV;;UAGe,eAAe;;EAE9B;EACA,eAAe;EACf,OAAO;EACP;EACA;;EAEA;EACA,SAAS;;;UAIM,sBAAsB;WAC5B;EACT,kBAAkB;;WAET,cAAc,SAAS;;;;;EAKhC,OAAO,iBAAiB,gBAAgB;;EAExC,QAAQ,iBAAiB,gBAAgB;EACzC,WAAW,MAAM,eAAe,QAAQ,QAAQ;;;UAIjC,uBAAuB;WAC7B;EACT,kBAAkB;WACT,cAAc,SAAS;;WAEvB;;WAEA;;EAET,YAAY;EACZ,WAAW,MAAM,eAAe,QAAQ,QAAQ;;EAEhD,gBAAgB;IACd;IACA,mBAAmB;IACnB,OAAO;IACP,SAAS;MACP;;;UAIW,0BAA0B,aAAa;;EAEtD,KAAK,SAAS,uBAAuB;IACnC,iBAAiB;IACjB;;;EAGF,OAAO;IACL;IACA,iBAAiB;IACjB,OAAO;IACP;IACA;IACA,SAAS;MACP,QAAQ;;EAEZ,aAAa,sBAAsB,gBAAgB;;;UAIpC,4BAA4B,uBAAuB;WACzD;;WAEA;EACT,kBAAkB;WACT,cAAc,SAAS;;WAEvB,qBAAqB,SAAS;;WAE9B;;EAET,qBAAqB,8BAA8B;;EAEnD,MAAM;IACJ;IACA,mBAAmB;IACnB;IACA,OAAO;IACP,SAAS;MACP;IAAU,UAAU;IAAkB,aAAa;IAAe;;;EAEtE,YAAY,0BAA0B,aAAa;;EAEnD,mBACE,QAAQ,oCACP,uBAAuB;;KAGhB,uBAAuB,MAAM,uBAAuB,oBAC5D,sBAAsB,QACtB,uBAAuB,QACvB,4BAA4B,aAAa;UAI5B,kBAAkB,gBAAgB,uBAAuB;;WAE/D;WACA;WACA;;WAEA;WACA,SAAS;;WAET;;WAEA;WACA,YAAY;WACZ,eAAe,cAAc;;;;;;WAM7B,gBAAgB,SAAS;WACzB,QAAQ;WACR,QAAQ;;;;;;WAMR;WACA,SAAS,uBAAuB,MAAM,aAAa;;;;;;;;;qBAUjD,mCAAmC;;;;;;;wBAUhC,gCAAgC,MAAM,aAAa,QACjE,YAAY,kBAAkB,MAAM,aAAa;;UAwClC;WACN;WACA,gBAAgB;WAChB,iBAAiB,YAAY;WAC7B;WACA;WACA;;WAEA;;WAEA;;UAGM;WACN;;WAEA;;WAEA,sBAAsB;WACtB,sBAAsB;;;;;;;;;wBAUjB,6BACd,aAAa,cAAc,gDAC1B;;;;;;;qBCzUU;;;;qBAKA;qBAMA;;wBA8GG,4BAA4B,SAAS;;wBASrC,+BAA+B,SAAS;;;;wBAcxC,8BAA8B,SAAS;;;;;;;;;;;;;UC5EtC;;EAEf;;EAEA;;EAEA;;;wBAIc,8BACd,SAAS,+BACT,MAAM,6BACL,kBAAkB;;wBAsEL,kCACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;cAwhBpB,yBAAuB,EAAA;;;;;;;;;;;;;GASlB,EAAA,KAAA;cACL,gCAA8B,EAAA;;;;;;;;;;;;;;;;;;;GAsChC,EAAA,KAAA;cACE,uBAAqB,EAAA;;;;;;;;;;;;KA2BtB,yBAAyB,EAAE,aAAa;EAC3C,WAAW,EAAE,aAAa;;KAEvB,2BAA2B,EAAE,aAAa;KACnC,iCAAiC,yBAAyB;;;;;;;;;;;;;UCjtBrD;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,cAAc;;;wBAIA,2BACd,SAAS,+BACT,MAAM,0BACL,kBAAkB,mBAAmB;;wBA+GxB,+BACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;;wBAwBV,gBAAgB,QAAQ,oCAAoC;;;UCtI3D;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA,UAAU;;EAEV,YAAY;;;;;;wBA8EE;UAIC;;EAEf;;EAEA;;;;;;wBAOc,gCACd,MAAM,+BACL,kBAAkB;;wBAyDL,2BACd,SAAS,8BACR,uBAAuB;;;;;;;;;;cC5OpB;WACQ;WACC;WACK;WACd;WACkB;WACJ;WACO;WACD;WACJ;WACK;WACd;WACJ;WACM;WACC;WACG;WACH;WACI;WACT;;KAGC,uCAAuC;UAgBlC;EACf,QAAQ;EACR;;EAEA;;EAEA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;;wBASK,sBACd,QAAQ,wBACR;EACE;EACA;EACA;EACA;EACA;IAED;;;UC3Fc;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;;wBAGK,8BACd,QAAQ,yBACP;wBAca,mCAAmC,SAAS;;;UCpC3C;EACf;EACA;EACA;IACE,SAAS;IACT;IACA;IACA;IACA;IACA,YAAY;MAAQ;MAAiB;MAAsB;;IAC3D,uBAAuB;IACvB,0BAA0B;IAC1B;MACE;MACA;MACA;MACA,QAAQ;;IAEV;MACE;MACA;;MAEA;MACA;;MAEA;MACA;MACA;MACA;MACA;MACA;MACA;MACA,UAAU;MACV;QACE;QACA;QACA;QACA;QACA;QACA;QACA;QACA;;MAEF;QACE;QACA;QACA;;MAEF;MACA;MACA;;MAEA;MACA;MACA;;;EAGJ,QAAQ;EACR,aAAa;EACb,uBAAuB;EACvB,qBAAqB;;UAGN;EACf;EACA;EACA;EACA;IACE;IACA;IACA;;;UAIa;EACf;IACE,SAAS;IACT;IACA;IACA;MACE;MACA;MACA;MACA;MACA;MACA;MACA;MACA;MACA,SAAS;MACT;QACE;QACA;QACA;QACA;QACA;QACA;QACA;QACA;;MAEF;QACE;QACA;QACA;;;IAGJ;IACA;IACA;IACA;;IAEA;IACA;IACA;IACA;;IAEA;IACA;IACA;IACA;;EAEF;IACE;IACA;IACA;IACA,YAAY;MAAQ;MAAiB;MAAsB;;IAC3D;IACA;;;UAIa;EACf;EACA;EACA;EACA;EACA,UAAU;;UAGK;EACf;EACA;EACA;EACA;IACE;IACA;IACA;IACA;;IAEA;;EAEF;EACA;IACE;IACA;IACA;;EAEF;IACE;IACA;IACA;IACA;IACA;IACA;;;UAIa;EACf;EACA;EACA;EACA,aAAa;EACb;;qBAGW;qBACA;qBACA;qBACA;;;wBC5KS,6BACpB,eACC,QAAQ;;;UCuEM;EACf,uBACE,SAAS,+BACT,QAAQ,wCACL,uBAAuB;EAC5B,2BACE,mBACA;IACE;IACA,aAAa,SAAS,OAAO;QAE5B,QAAQ;;;;;;;;;KAUH;;UAGK;EACf;EACA;;UAGe;EACf,SAAS;EACT,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;EACA,OAAO;EACP;EACA;EACA;EACA;;EAEA;EACA;EACA;;EAEA;;EAEA,QAAQ;EACR;EACA;;wBAKoB,2BACpB,yBACA,MAAK,OAAO,YACZ,eAAc,sCACb;;;qBCzJU;;;qBAIA;qBAGA;qBAOA;qBAYA;qBAGA;qBAGA;qBAiHA;wBAGG;wBAIA;;;;wBC/IA,oCAAoC,eAAe;;wBAQnD,gCAAgC,eAAe;;;;;;;;;;;wBAyB/C,+BACd,SAAS,+BACT,WAAW,KAAK;;;wBCzCF,+BACd,QAAQ,wBACR,uBAAsB;;;wBCIR,gCACd,kBACA,uBAAuB,gCACtB;;;;KCcS;;WAGG;WACA;WACA;WACA,YAAY;WACZ,UAAU;;;WAIV;WACA,QAAQ;;;;;;wBAOP,YAAY,MAAM,aAAa,QAC7C,YAAY,kBAAkB,MAAM,aAAa,SACjD,YAAY,0BACX,uBAAuB;;;;;;;;;;wBCpCV,mBAAmB,YAAY,yBAAyB;UAkBvD;EACf;EACA;EACA;EACA,MAAM;EACN,SAAS,QAAQ;;UAGF,wCAAwC;;EAEvD,iBAAiB,SAAS;;;wBAIZ,oBACd,SAAS,kCACR,oBAAoB;wBACP,oBACd,SAAS,6BACR,QAAQ;;;KCpBC;;UAGK;EACf;EACA,MAAM;;EAEN;EACA;;EAEA;;EAEA;;;EAGA;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;UAGe;EACf;EACA,SAAS;;UAGM;;EAEf;;EAEA;IAAc;IAAc,MAAM;;;EAElC;;;EAGA,kBAAkB;;EAElB;;;wBA2Ec,sBAAsB,QAAQ,uBAAuB;;wBAiHrD,uBACd,QAAQ,kBACR,qBACC;cA2HG,6BAA6B,oBAAoB;WAC5C;WACA;WAEA;WACA;IAAS;IAAgC;;WACzC;WACA;aACP;aACA;aACA;;EAGI,QAAQ,OAAO,kBAAkB,KAAK,iBAAiB,QAAQ;;qBAS1D,qBAAmB;;;;;;;;;;;;;;;wBC5YhB,gBAAgB;;;;;wBAgBhB,WAAW;;;;wBCIX,kBAAkB,mBAAmB,WAAW;;;;;;wBAShD,uBACd,mBACA,mBACC,cAAc"}
@@ -2,13 +2,13 @@ import { i as CostLedger } from "../cost-ledger-_IbkejCF.js";
2
2
  import { n as assertProposalFindings, r as isProposalFinding } from "../abort-signal-CtzAM_sJ.js";
3
3
  import { B as FINDING_SUBJECT_SYNTAX, F as parseRawFinding, G as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, H as findingSubjectGrammarPromptFor, I as parseTraceSpanEvidenceUri, K as resolveTraceAnalystLimits, L as coerceJson, M as RAW_FINDING_SCHEMA_PROMPT, N as RawAnalystFindingSchema, P as evidenceRefsFromRawFinding, R as stripCodeFences, U as parseFindingSubject, V as KIND_EXPECTED_SUBJECTS, W as renderFindingSubject, i as buildTraceToolsForGroup, n as renderPriorFindings, r as runTraceAnalyst, t as createTraceAnalyst, z as FINDING_SUBJECT_KINDS } from "../kind-factory-CdPPHtGZ.js";
4
4
  import { i as validateUsageSettlementTimeout, n as settleUsageReceiptFromCostLedger } from "../usage-receipt-Cx3xNeyB.js";
5
- import { a as KNOWLEDGE_POISONING_KIND_SPEC, c as FAILURE_MODE_KIND_SPEC, d as emitControlIntegrityFindings, f as behavioralAnalyst, i as DEFAULT_TRACE_ANALYST_KINDS, l as CONTROL_INTEGRITY_ANALYST, n as AnalystRegistry, o as KNOWLEDGE_GAP_KIND_SPEC, p as deriveEfficiencyFindings, r as ExactAnalystRunExecutionError, s as IMPROVEMENT_KIND_SPEC, t as buildDefaultAnalystRegistry, u as ControlIntegrityAnalyst } from "../default-registry-xMLX160E.js";
5
+ import { a as KNOWLEDGE_POISONING_KIND_SPEC, c as FAILURE_MODE_KIND_SPEC, d as emitControlIntegrityFindings, f as behavioralAnalyst, i as DEFAULT_TRACE_ANALYST_KINDS, l as CONTROL_INTEGRITY_ANALYST, n as AnalystRegistry, o as KNOWLEDGE_GAP_KIND_SPEC, p as deriveEfficiencyFindings, r as ExactAnalystRunExecutionError, s as IMPROVEMENT_KIND_SPEC, t as buildDefaultAnalystRegistry, u as ControlIntegrityAnalyst } from "../default-registry-D4qXx2aG.js";
6
6
  import { i as makeProposalFinding, n as computeFindingId, r as makeFinding } from "../types-CW5b90JH.js";
7
- import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-B5T43BkZ.js";
8
- import { a as diffFindings, i as defaultIsMaterial, n as runSemanticConceptJudge, o as createChatTraceEngine, r as FindingsStore, t as SEMANTIC_CONCEPT_JUDGE_VERSION } from "../semantic-concept-judge-2SE_kXGF.js";
9
- import { t as createChatClient } from "../chat-client-CJORSG_J.js";
10
- import { a as scoreAnalystFindings, i as summarizeAnalystBenchmarkRunner, n as runAnalystBenchmark, r as traceStoreEvidenceResolver, t as registryBenchmarkRunner } from "../benchmark-BeQJyrRC.js";
11
- import { $ as ANALYST_BENCHMARK_MANIFEST_FILE, A as readAnalystInstructionsOverride, B as parseVerificationOutcome, C as emptyPublicBenchmarkRunner, D as summarizeCodeTraceCalibration, E as renderCodeTraceCalibrationMarkdown, F as publicBenchmarkRlmInstructions, G as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, H as ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES, I as publicBenchmarkSystemPrompt, J as ANALYST_BENCHMARK_IMPLEMENTATION_SHA256, K as ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM, L as DEFAULT_MAX_VERIFICATION_ARTIFACT_BYTES, M as MAX_INCORRECT_BLOCKS, N as MAX_INCORRECT_BLOCK_STEPS, O as analystInstructionsOverrideFromText, P as publicBenchmarkProtocolSha256, Q as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, R as appendVerificationArtifactsToOtlp, S as adaptPublicBenchmarkFindings, T as readAnalystBenchmarkArtifact, U as ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256, V as ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM, W as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, X as analystBenchmarkImplementationDigest, Y as analystBenchmarkDependencyLockDigest, Z as ANALYST_BENCHMARK_COST_LEDGER_FILE, _ as publicDirectAnalystDefinition, a as primeCodeTraceAnalystDefinition, at as codeTraceBenchCase, b as analystDefinitionAsymmetries, c as preparePublicAnalystBenchmark, ct as agentRxPredictionsToFindings, d as selectPublicBenchmarkRows, dt as normalizeBenchmarkLabel, et as ANALYST_BENCHMARK_OBSERVATIONS_FILE, f as createPublicBenchmarkRlmRunner, g as createPublicBenchmarkDirectRunner, h as runReplVariableAnalystDefinition, i as primeAnalystProtocolSha256, it as summarizeAgentRxCalibration, j as CODE_TRACE_BENCH_ANALYST_PROMPT, k as effectiveAnalystProtocolSha256, l as publicBenchmarkDistributions, lt as normalizeAgentRxCategory, m as rlmEngineLimits, n as renderAnalystBenchmarkMarkdown, nt as AGENT_RX_UPSTREAM_REVISION, o as runInlineAnalystDefinition, ot as codeTracerPredictionsToFindings, p as publicRlmAnalystDefinition, q as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, r as createPrimeBenchmarkRunner, rt as renderAgentRxCalibrationMarkdown, s as loadPublicBenchmarkRows, st as agentRxBenchmarkCase, t as runAnalystBenchmarkCommand, tt as compareAnalystRunners, u as publicBenchmarkSelectionReport, ut as roundAgentRxStep, v as runChunkedAnalystDefinition, w as expandCodeTraceFailureBlocks, x as analystDefinitionProtocolSha256, y as AnalystExpressivenessError, z as loadCodeTraceVerificationArtifacts } from "../benchmark-command-EBmuMUBo.js";
7
+ import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-US8YsYGq.js";
8
+ import { a as diffFindings, i as defaultIsMaterial, n as runSemanticConceptJudge, o as createChatTraceEngine, r as FindingsStore, t as SEMANTIC_CONCEPT_JUDGE_VERSION } from "../semantic-concept-judge-BcjYg37z.js";
9
+ import { t as createChatClient } from "../chat-client-DUhWNQQS.js";
10
+ import { a as scoreAnalystFindings, i as summarizeAnalystBenchmarkRunner, n as runAnalystBenchmark, r as traceStoreEvidenceResolver, t as registryBenchmarkRunner } from "../benchmark-BywPeAUo.js";
11
+ import { $ as ANALYST_BENCHMARK_MANIFEST_FILE, A as readAnalystInstructionsOverride, B as parseVerificationOutcome, C as emptyPublicBenchmarkRunner, D as summarizeCodeTraceCalibration, E as renderCodeTraceCalibrationMarkdown, F as publicBenchmarkRlmInstructions, G as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, H as ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES, I as publicBenchmarkSystemPrompt, J as ANALYST_BENCHMARK_IMPLEMENTATION_SHA256, K as ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM, L as DEFAULT_MAX_VERIFICATION_ARTIFACT_BYTES, M as MAX_INCORRECT_BLOCKS, N as MAX_INCORRECT_BLOCK_STEPS, O as analystInstructionsOverrideFromText, P as publicBenchmarkProtocolSha256, Q as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, R as appendVerificationArtifactsToOtlp, S as adaptPublicBenchmarkFindings, T as readAnalystBenchmarkArtifact, U as ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256, V as ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM, W as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, X as analystBenchmarkImplementationDigest, Y as analystBenchmarkDependencyLockDigest, Z as ANALYST_BENCHMARK_COST_LEDGER_FILE, _ as publicDirectAnalystDefinition, a as primeCodeTraceAnalystDefinition, at as codeTraceBenchCase, b as analystDefinitionAsymmetries, c as preparePublicAnalystBenchmark, ct as agentRxPredictionsToFindings, d as selectPublicBenchmarkRows, dt as normalizeBenchmarkLabel, et as ANALYST_BENCHMARK_OBSERVATIONS_FILE, f as createPublicBenchmarkRlmRunner, g as createPublicBenchmarkDirectRunner, h as runReplVariableAnalystDefinition, i as primeAnalystProtocolSha256, it as summarizeAgentRxCalibration, j as CODE_TRACE_BENCH_ANALYST_PROMPT, k as effectiveAnalystProtocolSha256, l as publicBenchmarkDistributions, lt as normalizeAgentRxCategory, m as rlmEngineLimits, n as renderAnalystBenchmarkMarkdown, nt as AGENT_RX_UPSTREAM_REVISION, o as runInlineAnalystDefinition, ot as codeTracerPredictionsToFindings, p as publicRlmAnalystDefinition, q as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, r as createPrimeBenchmarkRunner, rt as renderAgentRxCalibrationMarkdown, s as loadPublicBenchmarkRows, st as agentRxBenchmarkCase, t as runAnalystBenchmarkCommand, tt as compareAnalystRunners, u as publicBenchmarkSelectionReport, ut as roundAgentRxStep, v as runChunkedAnalystDefinition, w as expandCodeTraceFailureBlocks, x as analystDefinitionProtocolSha256, y as AnalystExpressivenessError, z as loadCodeTraceVerificationArtifacts } from "../benchmark-command-DFs_i62N.js";
12
12
  import { a as extractPrimeJsonObject, c as primeProtocolSha256, d as runPrimeExchange, f as decodeReplyRows, i as emptyPrimeRawUsage, l as primeReplyDefect, n as buildPrimePrompt, o as mergePrimeRawUsage, r as buildPrimeRepairPrompt, s as normalizePrimeUsage, t as analystUsageReceiptFromPrimeUsage, u as projectPrimeTrajectory } from "../prime-protocol-DcAK9pC8.js";
13
13
  import { t as nodeHttpPrimeBridgeTransport } from "../prime-bridge-transport-Z3cpKGoV.js";
14
14
  import { existsSync, readFileSync, readdirSync, statSync } from "node:fs";