@tangle-network/agent-eval 1.0.23 → 1.0.24

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 (178) hide show
  1. package/CHANGELOG.md +41 -0
  2. package/dist/adapters/http.d.ts +2 -2
  3. package/dist/{agent-profile-C80hNPYb.d.ts → agent-profile-DC_v4p1f.d.ts} +3 -3
  4. package/dist/{agent-profile-C80hNPYb.d.ts.map → agent-profile-DC_v4p1f.d.ts.map} +1 -1
  5. package/dist/analyst/index.d.ts +14 -14
  6. package/dist/analyst/index.js +9 -9
  7. package/dist/{analyze-runs-DbHQHvDC.d.ts → analyze-runs-DGr7mpC3.d.ts} +4 -4
  8. package/dist/{analyze-runs-DbHQHvDC.d.ts.map → analyze-runs-DGr7mpC3.d.ts.map} +1 -1
  9. package/dist/app-grade/index.d.ts +4 -4
  10. package/dist/app-grade/index.js +3 -3
  11. package/dist/{benchmark-command-DkUpd9Q6.js → benchmark-command-DRAHkn4D.js} +8 -9
  12. package/dist/{benchmark-command-DkUpd9Q6.js.map → benchmark-command-DRAHkn4D.js.map} +1 -1
  13. package/dist/benchmarks/index.d.ts +4 -4
  14. package/dist/benchmarks/index.js +3 -3
  15. package/dist/campaign/index.d.ts +7 -7
  16. package/dist/campaign/index.js +4 -4
  17. package/dist/{campaign-D00Tbey-.js → campaign-C9QC27C2.js} +6 -6
  18. package/dist/{campaign-D00Tbey-.js.map → campaign-C9QC27C2.js.map} +1 -1
  19. package/dist/{chat-client-Zz97O3NB.js → chat-client-9XLgWrQ7.js} +5 -3
  20. package/dist/chat-client-9XLgWrQ7.js.map +1 -0
  21. package/dist/{chat-json-call-Dgi7-RH-.js → chat-json-call-CaNO3k0A.js} +2 -2
  22. package/dist/{chat-json-call-Dgi7-RH-.js.map → chat-json-call-CaNO3k0A.js.map} +1 -1
  23. package/dist/cli.js +3 -3
  24. package/dist/{cluster-Cd0MRbr9.d.ts → cluster-CkzP4h3T.d.ts} +4 -2
  25. package/dist/{cluster-Cd0MRbr9.d.ts.map → cluster-CkzP4h3T.d.ts.map} +1 -1
  26. package/dist/{completion-verifier-DFyYFezn.js → completion-verifier-BYLOcQt7.js} +5 -4
  27. package/dist/completion-verifier-BYLOcQt7.js.map +1 -0
  28. package/dist/{completion-verifier-BEpsr3N4.d.ts → completion-verifier-CJVDPv2T.d.ts} +3 -2
  29. package/dist/{completion-verifier-BEpsr3N4.d.ts.map → completion-verifier-CJVDPv2T.d.ts.map} +1 -1
  30. package/dist/contract/index.d.ts +10 -10
  31. package/dist/contract/index.js +7 -8
  32. package/dist/contract/index.js.map +1 -1
  33. package/dist/{cost-ledger-DfhStGcZ.js → cost-ledger-1CwUnVOd.js} +145 -3
  34. package/dist/cost-ledger-1CwUnVOd.js.map +1 -0
  35. package/dist/{decision-point-Di2LATNH.js → decision-point-CQOczIIH.js} +2 -2
  36. package/dist/{decision-point-Di2LATNH.js.map → decision-point-CQOczIIH.js.map} +1 -1
  37. package/dist/{default-registry-CjgRr-l7.d.ts → default-registry-C_Akmpui.d.ts} +7 -7
  38. package/dist/{default-registry-CjgRr-l7.d.ts.map → default-registry-C_Akmpui.d.ts.map} +1 -1
  39. package/dist/{default-registry-CdFbxPnr.js → default-registry-D-xd7KxV.js} +5 -5
  40. package/dist/{default-registry-CdFbxPnr.js.map → default-registry-D-xd7KxV.js.map} +1 -1
  41. package/dist/{define-agent-eval-BV16x8C0.js → define-agent-eval-BLVewqJm.js} +3 -3
  42. package/dist/{define-agent-eval-BV16x8C0.js.map → define-agent-eval-BLVewqJm.js.map} +1 -1
  43. package/dist/{define-agent-eval-BQBwKHwF.d.ts → define-agent-eval-nNPC7cD5.d.ts} +7 -7
  44. package/dist/{define-agent-eval-BQBwKHwF.d.ts.map → define-agent-eval-nNPC7cD5.d.ts.map} +1 -1
  45. package/dist/diagnosis/index.d.ts +2 -2
  46. package/dist/{dspy-rlm-engine-cWyn8G1a.js → dspy-rlm-engine-BS_jxE0P.js} +4 -4
  47. package/dist/{dspy-rlm-engine-cWyn8G1a.js.map → dspy-rlm-engine-BS_jxE0P.js.map} +1 -1
  48. package/dist/{engine-s3t6uYH9.d.ts → engine-B0gr15-l.d.ts} +2 -2
  49. package/dist/{engine-s3t6uYH9.d.ts.map → engine-B0gr15-l.d.ts.map} +1 -1
  50. package/dist/{evaluation-bU_bhlv-.js → evaluation-C_GJvEEG.js} +2 -2
  51. package/dist/{evaluation-bU_bhlv-.js.map → evaluation-C_GJvEEG.js.map} +1 -1
  52. package/dist/{evaluation-DRgHKEVV.d.ts → evaluation-mk3u_ZJC.d.ts} +4 -4
  53. package/dist/{evaluation-DRgHKEVV.d.ts.map → evaluation-mk3u_ZJC.d.ts.map} +1 -1
  54. package/dist/evaluation.d.ts +1 -1
  55. package/dist/evaluation.js +1 -1
  56. package/dist/{exact-types-DTlZSygs.d.ts → exact-types-Blq7YvsK.d.ts} +2 -2
  57. package/dist/{exact-types-DTlZSygs.d.ts.map → exact-types-Blq7YvsK.d.ts.map} +1 -1
  58. package/dist/experiment/index.d.ts +3 -3
  59. package/dist/{external-optimizer-process-2ImGH_LR.js → external-optimizer-process-m9xln7zN.js} +2 -2
  60. package/dist/{external-optimizer-process-2ImGH_LR.js.map → external-optimizer-process-m9xln7zN.js.map} +1 -1
  61. package/dist/{external-optimizer-subprocess-_C_2yriN.js → external-optimizer-subprocess-GO6s84Ov.js} +3 -3
  62. package/dist/{external-optimizer-subprocess-_C_2yriN.js.map → external-optimizer-subprocess-GO6s84Ov.js.map} +1 -1
  63. package/dist/{feedback-trajectory-DrKk79NU.d.ts → feedback-trajectory-D3oISeGl.d.ts} +3 -3
  64. package/dist/{feedback-trajectory-DrKk79NU.d.ts.map → feedback-trajectory-D3oISeGl.d.ts.map} +1 -1
  65. package/dist/{findings-store-BSuaV3dN.js → findings-store-CTy71Lnr.js} +5 -5
  66. package/dist/{findings-store-BSuaV3dN.js.map → findings-store-CTy71Lnr.js.map} +1 -1
  67. package/dist/fuzz.js +1 -1
  68. package/dist/{index-DNXFzSpc.d.ts → index-BN6htPVi.d.ts} +2 -2
  69. package/dist/{index-DNXFzSpc.d.ts.map → index-BN6htPVi.d.ts.map} +1 -1
  70. package/dist/{index-DAnkXnwF.d.ts → index-BdFlQP54.d.ts} +3 -3
  71. package/dist/{index-DAnkXnwF.d.ts.map → index-BdFlQP54.d.ts.map} +1 -1
  72. package/dist/{index-DtX456kD.d.ts → index-CxuK3sxA.d.ts} +9 -9
  73. package/dist/{index-DtX456kD.d.ts.map → index-CxuK3sxA.d.ts.map} +1 -1
  74. package/dist/{index-DZ_aqB7v.d.ts → index-Dodw93oW.d.ts} +23 -23
  75. package/dist/{index-DZ_aqB7v.d.ts.map → index-Dodw93oW.d.ts.map} +1 -1
  76. package/dist/{index-DjrXBogW.d.ts → index-IE8O0HNr.d.ts} +8 -8
  77. package/dist/{index-DjrXBogW.d.ts.map → index-IE8O0HNr.d.ts.map} +1 -1
  78. package/dist/{index-DwscLCQX.d.ts → index-KAYaJtMA.d.ts} +3 -3
  79. package/dist/{index-DwscLCQX.d.ts.map → index-KAYaJtMA.d.ts.map} +1 -1
  80. package/dist/{index-DLZp90hM.d.ts → index-cp5gkylN.d.ts} +10 -10
  81. package/dist/{index-DLZp90hM.d.ts.map → index-cp5gkylN.d.ts.map} +1 -1
  82. package/dist/index.d.ts +27 -27
  83. package/dist/index.js +62 -21
  84. package/dist/index.js.map +1 -1
  85. package/dist/{integrity-CrmnGB2T.d.ts → integrity-zeWMeTQt.d.ts} +2 -2
  86. package/dist/{integrity-CrmnGB2T.d.ts.map → integrity-zeWMeTQt.d.ts.map} +1 -1
  87. package/dist/{judge-gate-CrN0PaAL.js → judge-gate-DsoI5tx7.js} +2 -2
  88. package/dist/{judge-gate-CrN0PaAL.js.map → judge-gate-DsoI5tx7.js.map} +1 -1
  89. package/dist/{kind-factory-CV0FNCMM.js → kind-factory-CJXaE1oF.js} +2 -2
  90. package/dist/{kind-factory-CV0FNCMM.js.map → kind-factory-CJXaE1oF.js.map} +1 -1
  91. package/dist/{llm-client-DMMSg-Ro.js → llm-client-Bj-TooRS.js} +11 -11
  92. package/dist/{llm-client-DMMSg-Ro.js.map → llm-client-Bj-TooRS.js.map} +1 -1
  93. package/dist/{llm-judge-Cbh7OHaO.js → llm-judge-DzAT7Wka.js} +5 -5
  94. package/dist/{llm-judge-Cbh7OHaO.js.map → llm-judge-DzAT7Wka.js.map} +1 -1
  95. package/dist/{matrix-8NhGDcmi.d.ts → matrix-Diw3yJCG.d.ts} +3 -4
  96. package/dist/matrix-Diw3yJCG.d.ts.map +1 -0
  97. package/dist/meta-eval/index.d.ts +2 -2
  98. package/dist/meta-eval/index.js +2 -2
  99. package/dist/multishot/golden/index.d.ts +1 -1
  100. package/dist/multishot/index.d.ts +3 -2
  101. package/dist/multishot/index.d.ts.map +1 -1
  102. package/dist/multishot/index.js +3 -3
  103. package/dist/multishot/index.js.map +1 -1
  104. package/dist/openapi.json +1 -1
  105. package/dist/{paired-promotion-decision-DJU3fZxH.d.ts → paired-promotion-decision-COeOgNHG.d.ts} +2 -2
  106. package/dist/{paired-promotion-decision-DJU3fZxH.d.ts.map → paired-promotion-decision-COeOgNHG.d.ts.map} +1 -1
  107. package/dist/{prime-protocol-CKxq_-cx.d.ts → prime-protocol-bRbh6dqd.d.ts} +2 -2
  108. package/dist/{prime-protocol-CKxq_-cx.d.ts.map → prime-protocol-bRbh6dqd.d.ts.map} +1 -1
  109. package/dist/{promotion-gate-HDtliGsh.d.ts → promotion-gate-XglbDCNo.d.ts} +2 -2
  110. package/dist/{promotion-gate-HDtliGsh.d.ts.map → promotion-gate-XglbDCNo.d.ts.map} +1 -1
  111. package/dist/{proposal-findings-DWs94yT8.js → proposal-findings-CS6LoGGp.js} +2 -2
  112. package/dist/{proposal-findings-DWs94yT8.js.map → proposal-findings-CS6LoGGp.js.map} +1 -1
  113. package/dist/{registry-BnenKvxM.d.ts → registry-Pfeqsr5k.d.ts} +4 -4
  114. package/dist/{registry-BnenKvxM.d.ts.map → registry-Pfeqsr5k.d.ts.map} +1 -1
  115. package/dist/reporting.d.ts +4 -4
  116. package/dist/{researcher-DEpedkOZ.d.ts → researcher-DlyjP5uq.d.ts} +4 -4
  117. package/dist/{researcher-DEpedkOZ.d.ts.map → researcher-DlyjP5uq.d.ts.map} +1 -1
  118. package/dist/rl.d.ts +8 -8
  119. package/dist/rl.js.map +1 -1
  120. package/dist/rollout/index.d.ts +2 -2
  121. package/dist/rollout/index.js +1 -1
  122. package/dist/{rollout-DWl3Gg6A.js → rollout-Dh5q6CPA.js} +2 -2
  123. package/dist/{rollout-DWl3Gg6A.js.map → rollout-Dh5q6CPA.js.map} +1 -1
  124. package/dist/search/index.d.ts +2 -2
  125. package/dist/{search-kernel-CoDJ21q9.d.ts → search-kernel-6P1dRpJ4.d.ts} +3 -3
  126. package/dist/{search-kernel-CoDJ21q9.d.ts.map → search-kernel-6P1dRpJ4.d.ts.map} +1 -1
  127. package/dist/{sequential-BCzXKQab.d.ts → sequential-OeSwJ2uG.d.ts} +2 -2
  128. package/dist/{sequential-BCzXKQab.d.ts.map → sequential-OeSwJ2uG.d.ts.map} +1 -1
  129. package/dist/{server-DFVwfoZq.js → server-BorE0Y3O.js} +16 -5
  130. package/dist/server-BorE0Y3O.js.map +1 -0
  131. package/dist/{skillopt-optimization-method-B1Hkfexf.js → skillopt-optimization-method-BtX7xXDd.js} +5 -5
  132. package/dist/{skillopt-optimization-method-B1Hkfexf.js.map → skillopt-optimization-method-BtX7xXDd.js.map} +1 -1
  133. package/dist/{statistical-heldout-e2hbqDui.d.ts → statistical-heldout-vjrs8iyO.d.ts} +5 -5
  134. package/dist/{statistical-heldout-e2hbqDui.d.ts.map → statistical-heldout-vjrs8iyO.d.ts.map} +1 -1
  135. package/dist/statistics.d.ts +2 -2
  136. package/dist/{store-otlp-C-jyj3WX.js → store-otlp-CmgKDr0O.js} +2 -2
  137. package/dist/{store-otlp-C-jyj3WX.js.map → store-otlp-CmgKDr0O.js.map} +1 -1
  138. package/dist/{store-tool-spans-mlmGj6n3.d.ts → store-tool-spans-BemsQEhd.d.ts} +5 -5
  139. package/dist/{store-tool-spans-mlmGj6n3.d.ts.map → store-tool-spans-BemsQEhd.d.ts.map} +1 -1
  140. package/dist/{store-tool-spans-DalWlKY8.js → store-tool-spans-CW24axM0.js} +3 -3
  141. package/dist/{store-tool-spans-DalWlKY8.js.map → store-tool-spans-CW24axM0.js.map} +1 -1
  142. package/dist/{summary-report-DsiORzPc.d.ts → summary-report-vdhE20Zf.d.ts} +2 -2
  143. package/dist/{summary-report-DsiORzPc.d.ts.map → summary-report-vdhE20Zf.d.ts.map} +1 -1
  144. package/dist/supervisor-run/index.d.ts +1 -1
  145. package/dist/{systemone-B93onE_a.d.ts → systemone-DFUN9DL1.d.ts} +7 -7
  146. package/dist/{systemone-B93onE_a.d.ts.map → systemone-DFUN9DL1.d.ts.map} +1 -1
  147. package/dist/{systemone-review-BIggseLt.d.ts → systemone-review-4CKmHvr5.d.ts} +2 -2
  148. package/dist/{systemone-review-BIggseLt.d.ts.map → systemone-review-4CKmHvr5.d.ts.map} +1 -1
  149. package/dist/systemone.d.ts +3 -3
  150. package/dist/systemone.js +2 -2
  151. package/dist/{tool-groups-CCiATVZw.d.ts → tool-groups-Dpl3picW.d.ts} +3 -3
  152. package/dist/tool-groups-Dpl3picW.d.ts.map +1 -0
  153. package/dist/{trace-questions-B7lkyOlo.d.ts → trace-questions-oT2fZWJ8.d.ts} +3 -3
  154. package/dist/{trace-questions-B7lkyOlo.d.ts.map → trace-questions-oT2fZWJ8.d.ts.map} +1 -1
  155. package/dist/trace-repair/index.d.ts +1 -1
  156. package/dist/traces.d.ts +5 -5
  157. package/dist/traces.js +3 -3
  158. package/dist/{types-Ob1blgI2.d.ts → types-6GuPL-Vt.d.ts} +2 -2
  159. package/dist/{types-Ob1blgI2.d.ts.map → types-6GuPL-Vt.d.ts.map} +1 -1
  160. package/dist/{types-efIQ03Uo.d.ts → types-C9yEqDjQ.d.ts} +4 -4
  161. package/dist/{types-efIQ03Uo.d.ts.map → types-C9yEqDjQ.d.ts.map} +1 -1
  162. package/dist/{types-3uX8odDK.d.ts → types-CP3rkpir.d.ts} +2 -2
  163. package/dist/{types-3uX8odDK.d.ts.map → types-CP3rkpir.d.ts.map} +1 -1
  164. package/dist/{types-DUVhW5V1.d.ts → types-CwHU7U9J.d.ts} +2 -2
  165. package/dist/{types-DUVhW5V1.d.ts.map → types-CwHU7U9J.d.ts.map} +1 -1
  166. package/dist/wire/index.d.ts +4 -4
  167. package/dist/wire/index.d.ts.map +1 -1
  168. package/dist/wire/index.js +1 -1
  169. package/docs/wire-protocol.md +1 -1
  170. package/package.json +1 -1
  171. package/dist/chat-client-Zz97O3NB.js.map +0 -1
  172. package/dist/completion-verifier-DFyYFezn.js.map +0 -1
  173. package/dist/cost-ledger-DfhStGcZ.js.map +0 -1
  174. package/dist/matrix-8NhGDcmi.d.ts.map +0 -1
  175. package/dist/metrics-zALlR17s.js +0 -146
  176. package/dist/metrics-zALlR17s.js.map +0 -1
  177. package/dist/server-DFVwfoZq.js.map +0 -1
  178. package/dist/tool-groups-CCiATVZw.d.ts.map +0 -1
@@ -1 +1 @@
1
- {"version":3,"file":"proposal-findings-DWs94yT8.js","names":[],"sources":["../src/analyst/proposal-findings.ts"],"sourcesContent":["import { ProposalFindingSchema } from './finding-scope'\nimport type { ProposalFinding } from './types'\n\n/** True when a finding names a source candidate generation may learn from. */\nfunction isProposalFinding(finding: unknown): finding is ProposalFinding {\n return ProposalFindingSchema.safeParse(finding).success\n}\n\n/**\n * Reject findings whose source has not been explicitly admitted for candidate\n * generation. Search feedback and observed production behavior are allowed;\n * final evaluation data has no allowed origin.\n */\nexport function assertProposalFindings(\n findings: unknown,\n context = 'proposal findings',\n): ReadonlyArray<ProposalFinding> {\n if (!Array.isArray(findings)) {\n throw new TypeError(`${context}: expected an array`)\n }\n const rejected = findings.flatMap((finding, index) =>\n isProposalFinding(finding) ? [] : [findingLabel(finding, index)],\n )\n if (rejected.length > 0) {\n throw new Error(\n `${context}: every finding must match AnalystFinding and declare ` +\n `proposal_origin as search or production. ` +\n `Rejected findings: [${rejected.join(', ')}].`,\n )\n }\n return findings as ReadonlyArray<ProposalFinding>\n}\n\nfunction findingLabel(finding: unknown, index: number): string {\n if (typeof finding !== 'object' || finding === null) return `index ${index}`\n const id = (finding as { finding_id?: unknown }).finding_id\n return typeof id === 'string' && id.length > 0 ? id : `index ${index}`\n}\n"],"mappings":";;;AAIA,SAAS,kBAAkB,SAA8C;CACvE,OAAO,sBAAsB,UAAU,OAAO,CAAC,CAAC;AAClD;;;;;;AAOA,SAAgB,uBACd,UACA,UAAU,qBACsB;CAChC,IAAI,CAAC,MAAM,QAAQ,QAAQ,GACzB,MAAM,IAAI,UAAU,GAAG,QAAQ,oBAAoB;CAErD,MAAM,WAAW,SAAS,SAAS,SAAS,UAC1C,kBAAkB,OAAO,IAAI,CAAC,IAAI,CAAC,aAAa,SAAS,KAAK,CAAC,CACjE;CACA,IAAI,SAAS,SAAS,GACpB,MAAM,IAAI,MACR,GAAG,QAAQ,qHAEc,SAAS,KAAK,IAAI,EAAE,GAC/C;CAEF,OAAO;AACT;AAEA,SAAS,aAAa,SAAkB,OAAuB;CAC7D,IAAI,OAAO,YAAY,YAAY,YAAY,MAAM,OAAO,SAAS;CACrE,MAAM,KAAM,QAAqC;CACjD,OAAO,OAAO,OAAO,YAAY,GAAG,SAAS,IAAI,KAAK,SAAS;AACjE"}
1
+ {"version":3,"file":"proposal-findings-CS6LoGGp.js","names":[],"sources":["../src/analyst/proposal-findings.ts"],"sourcesContent":["import { ProposalFindingSchema } from './finding-scope'\nimport type { ProposalFinding } from './types'\n\n/** True when a finding names a source candidate generation may learn from. */\nfunction isProposalFinding(finding: unknown): finding is ProposalFinding {\n return ProposalFindingSchema.safeParse(finding).success\n}\n\n/**\n * Reject findings whose source has not been explicitly admitted for candidate\n * generation. Search feedback and observed production behavior are allowed;\n * final evaluation data has no allowed origin.\n */\nexport function assertProposalFindings(\n findings: unknown,\n context = 'proposal findings',\n): ReadonlyArray<ProposalFinding> {\n if (!Array.isArray(findings)) {\n throw new TypeError(`${context}: expected an array`)\n }\n const rejected = findings.flatMap((finding, index) =>\n isProposalFinding(finding) ? [] : [findingLabel(finding, index)],\n )\n if (rejected.length > 0) {\n throw new Error(\n `${context}: every finding must match AnalystFinding and declare ` +\n `proposal_origin as search or production. ` +\n `Rejected findings: [${rejected.join(', ')}].`,\n )\n }\n return findings as ReadonlyArray<ProposalFinding>\n}\n\nfunction findingLabel(finding: unknown, index: number): string {\n if (typeof finding !== 'object' || finding === null) return `index ${index}`\n const id = (finding as { finding_id?: unknown }).finding_id\n return typeof id === 'string' && id.length > 0 ? id : `index ${index}`\n}\n"],"mappings":";;;AAIA,SAAS,kBAAkB,SAA8C;CACvE,OAAO,sBAAsB,UAAU,OAAO,CAAC,CAAC;AAClD;;;;;;AAOA,SAAgB,uBACd,UACA,UAAU,qBACsB;CAChC,IAAI,CAAC,MAAM,QAAQ,QAAQ,GACzB,MAAM,IAAI,UAAU,GAAG,QAAQ,oBAAoB;CAErD,MAAM,WAAW,SAAS,SAAS,SAAS,UAC1C,kBAAkB,OAAO,IAAI,CAAC,IAAI,CAAC,aAAa,SAAS,KAAK,CAAC,CACjE;CACA,IAAI,SAAS,SAAS,GACpB,MAAM,IAAI,MACR,GAAG,QAAQ,qHAEc,SAAS,KAAK,IAAI,EAAE,GAC/C;CAEF,OAAO;AACT;AAEA,SAAS,aAAa,SAAkB,OAAuB;CAC7D,IAAI,OAAO,YAAY,YAAY,YAAY,MAAM,OAAO,SAAS;CACrE,MAAM,KAAM,QAAqC;CACjD,OAAO,OAAO,OAAO,YAAY,GAAG,SAAS,IAAI,KAAK,SAAS;AACjE"}
@@ -1,7 +1,7 @@
1
1
  import { o as CostLedgerHandle } from "./cost-ledger-Czro9wmo.js";
2
- import { c as AnalystRunEvent, d as AnalystRunSummary, i as AnalystFinding, l as AnalystRunInputs, n as AnalystContext, t as Analyst, u as AnalystRunResult } from "./types-Ob1blgI2.js";
3
- import { O as ChatClient } from "./cluster-Cd0MRbr9.js";
4
- import { i as ExactAnalystRunEvent, o as ExactAnalystRunResult, u as ExactExecutionComponentIdentity } from "./exact-types-DTlZSygs.js";
2
+ import { c as AnalystRunEvent, d as AnalystRunSummary, i as AnalystFinding, l as AnalystRunInputs, n as AnalystContext, t as Analyst, u as AnalystRunResult } from "./types-6GuPL-Vt.js";
3
+ import { O as ChatClient } from "./cluster-CkzP4h3T.js";
4
+ import { i as ExactAnalystRunEvent, o as ExactAnalystRunResult, u as ExactExecutionComponentIdentity } from "./exact-types-Blq7YvsK.js";
5
5
  interface AnalystHooks {
6
6
  /** Legacy runs may mutate ctx; exact runs provide a frozen observational context. */
7
7
  onBeforeAnalyze?(args: {
@@ -165,4 +165,4 @@ declare class AnalystRegistry {
165
165
  runStream(runId: string, inputs: AnalystRunInputs, runOpts?: RegistryRunOpts): AsyncGenerator<AnalystRunEvent, void, void>;
166
166
  }
167
167
  export { BudgetPolicy as a, RegistryRunOpts as c, AnalystRegistryOptions as i, ExactAnalystRunExecutionError as n, ExactAnalystBudgetPolicy as o, AnalystHooks as r, ExactRegistryRunOpts as s, AnalystRegistry as t };
168
- //# sourceMappingURL=registry-BnenKvxM.d.ts.map
168
+ //# sourceMappingURL=registry-Pfeqsr5k.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"registry-BnenKvxM.d.ts","names":[],"sources":["../src/analyst/registry-plan.ts","../src/analyst/registry-execution.ts","../src/analyst/registry.ts"],"mappings":";;;;UAiBiB;;EAEf,iBAAiB;IACf,SAAS;IACT,KAAK;IACL;aACS;;EAEX,gBAAgB;IACd,SAAS;IACT,SAAS;IACT,UAAU;IACV;aACS;;;;;;EAMX,SAAS;IACP,SAAS;IACT,OAAO;IACP;MACE,+BAA+B,QAAQ;;EAE3C,YAAY;IAAQ,QAAQ;aAA4B;;UAGzC;;EAEf;;EAEA,UAAU;;;;;;;EAOV,YAAY;IACV,SAAS;IACT;IACA;IACA;;;UAIa;;EAEf,OAAO;;EAEP,OAAO,aAAa,SAAS;;EAE7B,QAAQ;;EAER,gBAAgB;;EAEhB,gBAAgB;;EAEhB,eAAe;;UAGA;;EAEf;;EAEA;;EAEA,SAAS;;EAET;;EAEA,SAAS;;EAET,aAAa;;EAEb;;EAEA,OAAO;;;;;;;;;EASP,gBAAgB,cAAc,kBAAkB,eAAe,cAAc;;;;;;EAM7E;;;KAIU;WAEG;WACA;;WAGA;WACA;;WAEA,SAAS,SAAS;;;;;;;;;;UAWhB;WACN;WACA,QAAQ;WACR;WACA,QAAQ;WACR,YAAY;WACZ,oBAAoB;WACpB;WACA,MAAM,SAAS;WACf,eACL,cAAc,kBACd,SAAS,eAAe,cAAc;WAEjC;WACA;WACA;WACA;;;cChHE,sCAAsC;WACxC;WACA,QAAQ;EAEL,YAAA,iBAAiB,QAAQ,uBAAuB,UAAU;;cCN3D;mBACM;mBACA;EAEL,YAAA,UAAS;EAIrB,SAAS,SAAS;EAsBlB,QAAQ;IACN;IACA;IACA;IACA,MAAM;;EAUF,IACJ,eACA,QAAQ,kBACR,UAAS,kBACR,QAAQ;;EAUL,SACJ,eACA,QAAQ,kBACR,SAAS,uBACR,QAAQ;;EAQJ,eACL,eACA,QAAQ,kBACR,SAAS,uBACR,eAAe;;;;;;;;;;;;EAmBX,UACL,eACA,QAAQ,kBACR,UAAS,kBACR,eAAe"}
1
+ {"version":3,"file":"registry-Pfeqsr5k.d.ts","names":[],"sources":["../src/analyst/registry-plan.ts","../src/analyst/registry-execution.ts","../src/analyst/registry.ts"],"mappings":";;;;UAiBiB;;EAEf,iBAAiB;IACf,SAAS;IACT,KAAK;IACL;aACS;;EAEX,gBAAgB;IACd,SAAS;IACT,SAAS;IACT,UAAU;IACV;aACS;;;;;;EAMX,SAAS;IACP,SAAS;IACT,OAAO;IACP;MACE,+BAA+B,QAAQ;;EAE3C,YAAY;IAAQ,QAAQ;aAA4B;;UAGzC;;EAEf;;EAEA,UAAU;;;;;;;EAOV,YAAY;IACV,SAAS;IACT;IACA;IACA;;;UAIa;;EAEf,OAAO;;EAEP,OAAO,aAAa,SAAS;;EAE7B,QAAQ;;EAER,gBAAgB;;EAEhB,gBAAgB;;EAEhB,eAAe;;UAGA;;EAEf;;EAEA;;EAEA,SAAS;;EAET;;EAEA,SAAS;;EAET,aAAa;;EAEb;;EAEA,OAAO;;;;;;;;;EASP,gBAAgB,cAAc,kBAAkB,eAAe,cAAc;;;;;;EAM7E;;;KAIU;WAEG;WACA;;WAGA;WACA;;WAEA,SAAS,SAAS;;;;;;;;;;UAWhB;WACN;WACA,QAAQ;WACR;WACA,QAAQ;WACR,YAAY;WACZ,oBAAoB;WACpB;WACA,MAAM,SAAS;WACf,eACL,cAAc,kBACd,SAAS,eAAe,cAAc;WAEjC;WACA;WACA;WACA;;;cChHE,sCAAsC;WACxC;WACA,QAAQ;EAEL,YAAA,iBAAiB,QAAQ,uBAAuB,UAAU;;cCN3D;mBACM;mBACA;EAEL,YAAA,UAAS;EAIrB,SAAS,SAAS;EAsBlB,QAAQ;IACN;IACA;IACA;IACA,MAAM;;EAUF,IACJ,eACA,QAAQ,kBACR,UAAS,kBACR,QAAQ;;EAUL,SACJ,eACA,QAAQ,kBACR,SAAS,uBACR,QAAQ;;EAQJ,eACL,eACA,QAAQ,kBACR,SAAS,uBACR,eAAe;;;;;;;;;;;;EAmBX,UACL,eACA,QAAQ,kBACR,UAAS,kBACR,eAAe"}
@@ -1,8 +1,8 @@
1
- import { A as pairedBootstrap, E as PairedBootstrapResult, J as benjaminiHochberg, T as PairedBootstrapOptions, g as wilcoxonSignedRank } from "./index-DNXFzSpc.js";
1
+ import { A as pairedBootstrap, E as PairedBootstrapResult, J as benjaminiHochberg, T as PairedBootstrapOptions, g as wilcoxonSignedRank } from "./index-BN6htPVi.js";
2
2
  import { a as RubricPredictiveValidityReport, i as RubricPredictiveValidityInput, o as RubricRanking, r as RubricOutcomePair, s as rubricPredictiveValidity } from "./rubric-predictive-validity-Bbi88Aut.js";
3
- import { _ as gainHistogram, a as ResearchReportOptions, c as RESEARCH_REPORT_HARD_PAIR_FLOOR, d as GainDistributionOptions, f as ParetoFigureSpec, g as SummaryTableRow, h as SummaryTableOptions, i as ResearchReportMethodology, l as GainDistributionBin, m as SummaryTable, n as ResearchReportCandidate, o as ResearchReportRecommendation, p as ParetoPoint, r as ResearchReportDecision, s as researchReport, t as ResearchReport, u as GainDistributionFigureSpec, v as summaryTable } from "./summary-report-DsiORzPc.js";
4
- import { a as PairedEvalueStep, c as SequentialDecision, i as PairedEvalueSequence, l as evaluateInterimReleaseConfidence, n as InterimReleaseConfidenceInput, r as PairedEvalueOptions, t as InterimReleaseConfidence, u as pairedEvalueSequence } from "./sequential-BCzXKQab.js";
5
- import { _ as evaluateReleaseConfidence, a as bootstrapCi, c as ReleaseConfidenceAxis, d as ReleaseConfidenceIssue, f as ReleaseConfidenceMetrics, g as ReleaseTraceEvidence, h as ReleaseConfidenceThresholds, i as Verdict, l as ReleaseConfidenceAxisName, m as ReleaseConfidenceStatus, n as BootstrapResult, o as judgeReplayGate, p as ReleaseConfidenceScorecard, r as JudgeReplayGateArgs, t as BootstrapOptions, u as ReleaseConfidenceInput } from "./promotion-gate-HDtliGsh.js";
3
+ import { _ as gainHistogram, a as ResearchReportOptions, c as RESEARCH_REPORT_HARD_PAIR_FLOOR, d as GainDistributionOptions, f as ParetoFigureSpec, g as SummaryTableRow, h as SummaryTableOptions, i as ResearchReportMethodology, l as GainDistributionBin, m as SummaryTable, n as ResearchReportCandidate, o as ResearchReportRecommendation, p as ParetoPoint, r as ResearchReportDecision, s as researchReport, t as ResearchReport, u as GainDistributionFigureSpec, v as summaryTable } from "./summary-report-vdhE20Zf.js";
4
+ import { a as PairedEvalueStep, c as SequentialDecision, i as PairedEvalueSequence, l as evaluateInterimReleaseConfidence, n as InterimReleaseConfidenceInput, r as PairedEvalueOptions, t as InterimReleaseConfidence, u as pairedEvalueSequence } from "./sequential-OeSwJ2uG.js";
5
+ import { _ as evaluateReleaseConfidence, a as bootstrapCi, c as ReleaseConfidenceAxis, d as ReleaseConfidenceIssue, f as ReleaseConfidenceMetrics, g as ReleaseTraceEvidence, h as ReleaseConfidenceThresholds, i as Verdict, l as ReleaseConfidenceAxisName, m as ReleaseConfidenceStatus, n as BootstrapResult, o as judgeReplayGate, p as ReleaseConfidenceScorecard, r as JudgeReplayGateArgs, t as BootstrapOptions, u as ReleaseConfidenceInput } from "./promotion-gate-XglbDCNo.js";
6
6
  /** A text report column; `none` leaves the cell unpadded. */
7
7
  export interface ReportTableColumn {
8
8
  header: string;
@@ -1,10 +1,10 @@
1
1
  import { i as AgentProfileCellInput, r as AgentProfileCell } from "./agent-profile-cell-3kYxUsK2.js";
2
2
  import { a as RunRecord, c as RunSplitTag, d as RunTokenUsage, l as RunTaskFailure, n as RunCostProvenance, r as RunJudgeMetadata, t as JudgeScoresRecord } from "./run-record-pBlhZizT.js";
3
- import { O as ChatClient, it as RawProviderSink } from "./cluster-Cd0MRbr9.js";
3
+ import { O as ChatClient, at as RawProviderSink } from "./cluster-CkzP4h3T.js";
4
4
  import { s as TraceStore } from "./store-CMs5wGXP.js";
5
- import { S as HeldOutGateDecision, a as ResearchReportOptions, t as ResearchReport } from "./summary-report-DsiORzPc.js";
5
+ import { S as HeldOutGateDecision, a as ResearchReportOptions, t as ResearchReport } from "./summary-report-vdhE20Zf.js";
6
6
  import { o as TraceEmitter, r as RunCompleteHook } from "./emitter-mAXC5PSN.js";
7
- import { a as RunIntegrityReport, n as RunIntegrityExpectations } from "./integrity-CrmnGB2T.js";
7
+ import { a as RunIntegrityReport, n as RunIntegrityExpectations } from "./integrity-zeWMeTQt.js";
8
8
  interface CampaignVariant<V> {
9
9
  id: string;
10
10
  payload: V;
@@ -286,4 +286,4 @@ interface Researcher {
286
286
  evaluateChange(plan: ExperimentPlan): Promise<ExperimentResult>;
287
287
  }
288
288
  export { SteeringChange as a, CampaignRunContext as c, CampaignVariant as d, EvalCampaignOptions as f, runEvalCampaign as h, Researcher as i, CampaignRunOutcome as l, FailedRun as m, ExperimentResult as n, CampaignFactoryParams as o, EvalCampaignResult as p, FailureMode as r, CampaignIntegrityPolicy as s, ExperimentPlan as t, CampaignScenario as u };
289
- //# sourceMappingURL=researcher-DEpedkOZ.d.ts.map
289
+ //# sourceMappingURL=researcher-DlyjP5uq.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"researcher-DEpedkOZ.d.ts","names":[],"sources":["../src/eval-campaign.ts","../src/researcher.ts"],"mappings":";;;;;;;UAuEiB,gBAAgB;EAC/B;EACA,SAAS;;UAGM;EACf;;EAEA,OAAO;;UAGQ,mBAAmB;;EAElC;;EAEA;EACA,SAAS;EACT;EACA;EACA,cAAc;EACd;EACA,UAAU;;;;;;;EAOV,SAAS;EACT,OAAO;EACP,SAAS;;;;;EAKT,MAAM;;;UAIS;;;;;;EAMf,SAAS;EACT;;UAGQ;;EAER;;EAEA;;EAEA;;EAEA,gBAAgB;EAChB,YAAY;;EAEZ;;EAEA;;EAEA;;EAEA,MAAM;;EAEN,gBAAgB;;;;;;EAMhB,cAAc;;;;;;EAMd,eAAe,mBAAmB;;;KAIxB,qBAAqB,2BAA2B;KAEhD,eAAe,MAAM,KAAK,mBAAmB,OAAO,QAAQ;KAE5D;UAEK,oBAAoB;;;;;EAKnC;EACA,UAAU,gBAAgB;EAC1B,WAAW;;EAEX;;EAEA,WAAW;;EAEX;;;;;;;;EAQA,cAAc,QAAQ,uBAAuB;;;;;;;EAO7C;;;;;;EAMA,eAAe,QAAQ,0BAA0B;;;;;;;EAOjD,kBAAkB,QAAQ,0BAA0B;;;;;EAKpD;;;;;EAKA,gBAAgB;;;;;;EAMhB,YAAY;;EAEZ,qBAAqB;;;;;EAKrB,QAAQ,eAAe;;;;;EAKvB;IAAW;MAAwB,KACjC;;;;;EAOF;;EAEA;;;;EAIA;;EAEA,SAAS,QAAQ;;;;;;;EAOjB,eACI,mBACA,0BAEE,QAAQ;IACN,SAAS;IACT,cAAc;QAGd,mBACA,wBACA,QAAQ,mBAAmB;;UAGpB;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;;EAEA;EACA;;EAEA,MAAM;;EAEN,kBAAkB;EAClB,YAAY;;EAEZ,SAAS;EACT;EACA;;iBAyDoB,gBAAgB,GACpC,MAAM,oBAAoB,KACzB,QAAQ;;UChUM;;;EAGf;;EAEA;EACA;;;IAGE;;IAEA;;;;UAKa;EACf;;;;EAIA;;;EAGA;;EAEA;;;UAIe;EACf;EACA;EACA,SAAS;;;EAGT;;EAEA;IAAU;IAAkB;;;;UAIb;EACf,MAAM;EACN,MAAM;EACN,cAAc;;;;;;;;;;;;;;;;;;;UAoBC;EACf,gBAAgB,MAAM,cAAc,QAAQ;EAC5C,cAAc,UAAU,gBAAgB,QAAQ;EAChD,YAAY,SAAS,kBAAkB,UAAU,iBAAiB,QAAQ;EAC1E,eAAe,MAAM,iBAAiB,QAAQ"}
1
+ {"version":3,"file":"researcher-DlyjP5uq.d.ts","names":[],"sources":["../src/eval-campaign.ts","../src/researcher.ts"],"mappings":";;;;;;;UAuEiB,gBAAgB;EAC/B;EACA,SAAS;;UAGM;EACf;;EAEA,OAAO;;UAGQ,mBAAmB;;EAElC;;EAEA;EACA,SAAS;EACT;EACA;EACA,cAAc;EACd;EACA,UAAU;;;;;;;EAOV,SAAS;EACT,OAAO;EACP,SAAS;;;;;EAKT,MAAM;;;UAIS;;;;;;EAMf,SAAS;EACT;;UAGQ;;EAER;;EAEA;;EAEA;;EAEA,gBAAgB;EAChB,YAAY;;EAEZ;;EAEA;;EAEA;;EAEA,MAAM;;EAEN,gBAAgB;;;;;;EAMhB,cAAc;;;;;;EAMd,eAAe,mBAAmB;;;KAIxB,qBAAqB,2BAA2B;KAEhD,eAAe,MAAM,KAAK,mBAAmB,OAAO,QAAQ;KAE5D;UAEK,oBAAoB;;;;;EAKnC;EACA,UAAU,gBAAgB;EAC1B,WAAW;;EAEX;;EAEA,WAAW;;EAEX;;;;;;;;EAQA,cAAc,QAAQ,uBAAuB;;;;;;;EAO7C;;;;;;EAMA,eAAe,QAAQ,0BAA0B;;;;;;;EAOjD,kBAAkB,QAAQ,0BAA0B;;;;;EAKpD;;;;;EAKA,gBAAgB;;;;;;EAMhB,YAAY;;EAEZ,qBAAqB;;;;;EAKrB,QAAQ,eAAe;;;;;EAKvB;IAAW;MAAwB,KACjC;;;;;EAOF;;EAEA;;;;EAIA;;EAEA,SAAS,QAAQ;;;;;;;EAOjB,eACI,mBACA,0BAEE,QAAQ;IACN,SAAS;IACT,cAAc;QAGd,mBACA,wBACA,QAAQ,mBAAmB;;UAGpB;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;;EAEA;EACA;;EAEA,MAAM;;EAEN,kBAAkB;EAClB,YAAY;;EAEZ,SAAS;EACT;EACA;;iBAyDoB,gBAAgB,GACpC,MAAM,oBAAoB,KACzB,QAAQ;;UChUM;;;EAGf;;EAEA;EACA;;;IAGE;;IAEA;;;;UAKa;EACf;;;;EAIA;;;EAGA;;EAEA;;;UAIe;EACf;EACA;EACA,SAAS;;;EAGT;;EAEA;IAAU;IAAkB;;;;UAIb;EACf,MAAM;EACN,MAAM;EACN,cAAc;;;;;;;;;;;;;;;;;;;UAoBC;EACf,gBAAgB,MAAM,cAAc,QAAQ;EAC5C,cAAc,UAAU,gBAAgB,QAAQ;EAChD,YAAY,SAAS,kBAAkB,UAAU,iBAAiB,QAAQ;EAC1E,eAAe,MAAM,iBAAiB,QAAQ"}
package/dist/rl.d.ts CHANGED
@@ -1,16 +1,16 @@
1
1
  import { _ as Span } from "./schema-CkIvnmLU.js";
2
2
  import { a as RunRecord, c as RunSplitTag } from "./run-record-pBlhZizT.js";
3
- import { E as PairedBootstrapResult } from "./index-DNXFzSpc.js";
4
- import { Q as MintRolloutOptions, a as CampaignResult } from "./types-efIQ03Uo.js";
3
+ import { E as PairedBootstrapResult } from "./index-BN6htPVi.js";
4
+ import { Q as MintRolloutOptions, a as CampaignResult } from "./types-C9yEqDjQ.js";
5
5
  import { a as MintedRolloutLine, b as RolloutSplit, t as ChatMessage } from "./schema-PDKQ_iNE.js";
6
- import { a as PairedPromotionDecision } from "./paired-promotion-decision-DJU3fZxH.js";
6
+ import { a as PairedPromotionDecision } from "./paired-promotion-decision-COeOgNHG.js";
7
7
  import { a as detectRewardHacking, c as VerifiableRewardSource, i as RewardHackingSignal, l as extractVerifiableReward, n as RewardHackingFinding, o as VerifiableReward, r as RewardHackingReport, s as VerifiableRewardExtractionOptions, t as DetectRewardHackingInput, u as extractVerifiableRewardsFromRecords } from "./reward-hacking-CIdVwVe4.js";
8
8
  import { a as RubricPredictiveValidityReport, d as FileSystemOutcomeStore, f as FileSystemOutcomeStoreOptions, h as OutcomeStore, p as InMemoryOutcomeStore, t as OutcomeMetricSpec, u as DeploymentOutcome } from "./rubric-predictive-validity-Bbi88Aut.js";
9
- import { a as SystemOneReviewReport, t as PreparedSystemOneReview } from "./systemone-review-BIggseLt.js";
10
- import { n as TraceQuestionOutcome, r as TraceQuestionSpec } from "./trace-questions-B7lkyOlo.js";
11
- import { a as SteeringChange, f as EvalCampaignOptions, h as runEvalCampaign, i as Researcher, n as ExperimentResult, p as EvalCampaignResult, r as FailureMode, t as ExperimentPlan } from "./researcher-DEpedkOZ.js";
9
+ import { a as SystemOneReviewReport, t as PreparedSystemOneReview } from "./systemone-review-4CKmHvr5.js";
10
+ import { n as TraceQuestionOutcome, r as TraceQuestionSpec } from "./trace-questions-oT2fZWJ8.js";
11
+ import { a as SteeringChange, f as EvalCampaignOptions, h as runEvalCampaign, i as Researcher, n as ExperimentResult, p as EvalCampaignResult, r as FailureMode, t as ExperimentPlan } from "./researcher-DlyjP5uq.js";
12
12
  import { o as TrainingExportOptions } from "./exporters-dQKYD_Zo.js";
13
- import { t as InterimReleaseConfidence } from "./sequential-BCzXKQab.js";
13
+ import { t as InterimReleaseConfidence } from "./sequential-OeSwJ2uG.js";
14
14
  import { t as AdversarialMutation } from "./adversarial-DnQQFqTR.js";
15
15
  /**
16
16
  * Adaptive curriculum: active scenario selection by variance.
@@ -1372,7 +1372,7 @@ export declare function runRLCampaign<V>(opts: RunRLCampaignOptions<V>): Promise
1372
1372
  interface AdapterContext {
1373
1373
  /** Logical experiment id — typically the campaign or sweep identifier. */
1374
1374
  experimentId: string;
1375
- /** Snapshot model id (e.g. `claude-sonnet-4-6@2025-04-15`). */
1375
+ /** Snapshot model id (e.g. `claude-sonnet-5-5` or `gpt-4o-2024-11-20`). */
1376
1376
  model: string;
1377
1377
  /** Git SHA the harness was run from. */
1378
1378
  commitSha: string;