@tangle-network/agent-eval 0.203.0 → 0.203.2

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 (301) hide show
  1. package/CHANGELOG.md +10 -0
  2. package/dist/adapters/http.d.ts +10 -4
  3. package/dist/adapters/http.d.ts.map +1 -1
  4. package/dist/adapters/http.js +43 -38
  5. package/dist/adapters/http.js.map +1 -1
  6. package/dist/{agent-profile-DDfIjdyS.d.ts → agent-profile-EpO0HURy.d.ts} +4 -4
  7. package/dist/{agent-profile-DDfIjdyS.d.ts.map → agent-profile-EpO0HURy.d.ts.map} +1 -1
  8. package/dist/analyst/index.d.ts +12 -12
  9. package/dist/analyst/index.d.ts.map +1 -1
  10. package/dist/analyst/index.js +6 -6
  11. package/dist/{analyze-runs-CPiKLJFs.d.ts → analyze-runs-Q74o4Hfj.d.ts} +2 -2
  12. package/dist/{analyze-runs-CPiKLJFs.d.ts.map → analyze-runs-Q74o4Hfj.d.ts.map} +1 -1
  13. package/dist/{benchmark-command-DFs_i62N.js → benchmark-command-6OBcB0sG.js} +7 -6
  14. package/dist/benchmark-command-6OBcB0sG.js.map +1 -0
  15. package/dist/benchmarks/index.d.ts +4 -4
  16. package/dist/benchmarks/index.js +3 -3
  17. package/dist/campaign/index.d.ts +6 -6
  18. package/dist/campaign/index.js +4 -4
  19. package/dist/{campaign-Btk1yFfw.js → campaign-DYhg4SjN.js} +6 -6
  20. package/dist/{campaign-Btk1yFfw.js.map → campaign-DYhg4SjN.js.map} +1 -1
  21. package/dist/{chat-client-DUhWNQQS.js → chat-client-5LLl4Ctl.js} +2 -2
  22. package/dist/{chat-client-DUhWNQQS.js.map → chat-client-5LLl4Ctl.js.map} +1 -1
  23. package/dist/{chat-json-call-Dx_sHzqm.js → chat-json-call-Digk-pfF.js} +2 -2
  24. package/dist/{chat-json-call-Dx_sHzqm.js.map → chat-json-call-Digk-pfF.js.map} +1 -1
  25. package/dist/cli.js +3 -3
  26. package/dist/contract/index.d.ts +10 -10
  27. package/dist/contract/index.js +7 -7
  28. package/dist/{default-registry-B-utjsmX.d.ts → default-registry-BHEfZHQd.d.ts} +6 -6
  29. package/dist/{default-registry-B-utjsmX.d.ts.map → default-registry-BHEfZHQd.d.ts.map} +1 -1
  30. package/dist/{default-registry-CpswV_S2.js → default-registry-wInsQWlQ.js} +2 -2
  31. package/dist/{default-registry-CpswV_S2.js.map → default-registry-wInsQWlQ.js.map} +1 -1
  32. package/dist/{define-agent-eval-CJN9dTQw.js → define-agent-eval-B25Cc3ki.js} +3 -3
  33. package/dist/{define-agent-eval-CJN9dTQw.js.map → define-agent-eval-B25Cc3ki.js.map} +1 -1
  34. package/dist/{define-agent-eval-MAc6BQrq.d.ts → define-agent-eval-D8bPoBM5.d.ts} +7 -7
  35. package/dist/{define-agent-eval-MAc6BQrq.d.ts.map → define-agent-eval-D8bPoBM5.d.ts.map} +1 -1
  36. package/dist/diagnosis/index.d.ts +2 -2
  37. package/dist/{dspy-rlm-engine-US8YsYGq.js → dspy-rlm-engine-D0h-P2Mz.js} +2 -2
  38. package/dist/{dspy-rlm-engine-US8YsYGq.js.map → dspy-rlm-engine-D0h-P2Mz.js.map} +1 -1
  39. package/dist/{engine-4_Q4oujc.d.ts → engine-F_PvwesR.d.ts} +4 -4
  40. package/dist/{engine-4_Q4oujc.d.ts.map → engine-F_PvwesR.d.ts.map} +1 -1
  41. package/dist/evaluation.d.ts +2 -2
  42. package/dist/{exact-types-KGxBWH4b.d.ts → exact-types-BBMMYk3g.d.ts} +2 -2
  43. package/dist/{exact-types-KGxBWH4b.d.ts.map → exact-types-BBMMYk3g.d.ts.map} +1 -1
  44. package/dist/experiment/index.d.ts +1 -1
  45. package/dist/{external-optimizer-process-KLcw0Vnc.js → external-optimizer-process-CIV4EqiT.js} +2 -2
  46. package/dist/{external-optimizer-process-KLcw0Vnc.js.map → external-optimizer-process-CIV4EqiT.js.map} +1 -1
  47. package/dist/{external-optimizer-subprocess-CkyUUV7h.js → external-optimizer-subprocess-BmKtH3T6.js} +2 -2
  48. package/dist/{external-optimizer-subprocess-CkyUUV7h.js.map → external-optimizer-subprocess-BmKtH3T6.js.map} +1 -1
  49. package/dist/{extract-usage-DFATEuWW.js → extract-usage-BrnzRJ6H.js} +4 -3
  50. package/dist/extract-usage-BrnzRJ6H.js.map +1 -0
  51. package/dist/{feedback-trajectory-BXCGTRkz.d.ts → feedback-trajectory-C6NJxXEa.d.ts} +3 -3
  52. package/dist/{feedback-trajectory-BXCGTRkz.d.ts.map → feedback-trajectory-C6NJxXEa.d.ts.map} +1 -1
  53. package/dist/{index-CAtNOBQY.d.ts → index-B8n12TR2.d.ts} +12 -14
  54. package/dist/index-B8n12TR2.d.ts.map +1 -0
  55. package/dist/{index-DBk_4UPs.d.ts → index-C9l_-23M.d.ts} +7 -7
  56. package/dist/{index-DBk_4UPs.d.ts.map → index-C9l_-23M.d.ts.map} +1 -1
  57. package/dist/{index-Cuhxk05B.d.ts → index-CG7mtu_G.d.ts} +8 -8
  58. package/dist/{index-Cuhxk05B.d.ts.map → index-CG7mtu_G.d.ts.map} +1 -1
  59. package/dist/index.d.ts +19 -19
  60. package/dist/index.js +11 -11
  61. package/dist/{integrity-CaeNeusS.d.ts → integrity-BWO6RPxv.d.ts} +2 -2
  62. package/dist/{integrity-CaeNeusS.d.ts.map → integrity-BWO6RPxv.d.ts.map} +1 -1
  63. package/dist/{jev-review-DX6MI_jq.d.ts → jev-review-Cw06AKuF.d.ts} +2 -2
  64. package/dist/{jev-review-DX6MI_jq.d.ts.map → jev-review-Cw06AKuF.d.ts.map} +1 -1
  65. package/dist/jev.d.ts +3 -3
  66. package/dist/{llm-client-Bhqoglar.js → llm-client-DmCNzrc9.js} +77 -60
  67. package/dist/llm-client-DmCNzrc9.js.map +1 -0
  68. package/dist/{llm-judge-C-0P1WiT.js → llm-judge-DyZdrp6I.js} +57 -15
  69. package/dist/llm-judge-DyZdrp6I.js.map +1 -0
  70. package/dist/{matrix-BucOjLF1.d.ts → matrix-W1lbSx72.d.ts} +2 -2
  71. package/dist/{matrix-BucOjLF1.d.ts.map → matrix-W1lbSx72.d.ts.map} +1 -1
  72. package/dist/meta-eval/index.d.ts +1 -1
  73. package/dist/multishot/golden/index.d.ts +1 -1
  74. package/dist/multishot/index.d.ts +2 -2
  75. package/dist/openapi.json +1 -1
  76. package/dist/{prime-protocol-T0i6xcAs.d.ts → prime-protocol-IoN4Qb6c.d.ts} +2 -2
  77. package/dist/{prime-protocol-T0i6xcAs.d.ts.map → prime-protocol-IoN4Qb6c.d.ts.map} +1 -1
  78. package/dist/{produced-state-CR2tAPLf.js → produced-state-DLToUXrD.js} +3 -3
  79. package/dist/{produced-state-CR2tAPLf.js.map → produced-state-DLToUXrD.js.map} +1 -1
  80. package/dist/{abort-signal-CtzAM_sJ.js → proposal-findings-CXblsskH.js} +2 -11
  81. package/dist/proposal-findings-CXblsskH.js.map +1 -0
  82. package/dist/{registry-Bu3lNole.d.ts → registry-CT-hMSkR.d.ts} +4 -4
  83. package/dist/{registry-Bu3lNole.d.ts.map → registry-CT-hMSkR.d.ts.map} +1 -1
  84. package/dist/{researcher-DojEMb7E.d.ts → researcher-zTpxNJBy.d.ts} +3 -3
  85. package/dist/{researcher-DojEMb7E.d.ts.map → researcher-zTpxNJBy.d.ts.map} +1 -1
  86. package/dist/rl.d.ts +4 -4
  87. package/dist/search/index.d.ts +2 -2
  88. package/dist/{search-kernel-B8RC6BlK.d.ts → search-kernel-DNpsHb8P.d.ts} +2 -2
  89. package/dist/{search-kernel-B8RC6BlK.d.ts.map → search-kernel-DNpsHb8P.d.ts.map} +1 -1
  90. package/dist/{semantic-concept-judge-BcjYg37z.js → semantic-concept-judge-BgWxa4L_.js} +4 -4
  91. package/dist/{semantic-concept-judge-BcjYg37z.js.map → semantic-concept-judge-BgWxa4L_.js.map} +1 -1
  92. package/dist/{series-convergence-DY7ha0kx.d.ts → series-convergence-DLy0bQMV.d.ts} +2 -2
  93. package/dist/{series-convergence-DY7ha0kx.d.ts.map → series-convergence-DLy0bQMV.d.ts.map} +1 -1
  94. package/dist/{server-B4jMhqQn.js → server-CoBZO1fj.js} +2 -2
  95. package/dist/{server-B4jMhqQn.js.map → server-CoBZO1fj.js.map} +1 -1
  96. package/dist/{skillopt-optimization-method-Bls9t-G_.js → skillopt-optimization-method-Dz5CGZb8.js} +5 -5
  97. package/dist/{skillopt-optimization-method-Bls9t-G_.js.map → skillopt-optimization-method-Dz5CGZb8.js.map} +1 -1
  98. package/dist/{statistical-heldout-DJtQAUZB.d.ts → statistical-heldout-Dq3pnmxd.d.ts} +2 -2
  99. package/dist/{statistical-heldout-DJtQAUZB.d.ts.map → statistical-heldout-Dq3pnmxd.d.ts.map} +1 -1
  100. package/dist/{store-tool-spans-fkNF7Q9Y.d.ts → store-tool-spans-CAF3T-BS.d.ts} +5 -5
  101. package/dist/{store-tool-spans-fkNF7Q9Y.d.ts.map → store-tool-spans-CAF3T-BS.d.ts.map} +1 -1
  102. package/dist/{store-tool-spans-Biio4_Nw.js → store-tool-spans-D5kOU0mr.js} +2 -2
  103. package/dist/{store-tool-spans-Biio4_Nw.js.map → store-tool-spans-D5kOU0mr.js.map} +1 -1
  104. package/dist/{tool-groups-Zm84Pqpk.d.ts → tool-groups-B7_jUPHp.d.ts} +3 -3
  105. package/dist/tool-groups-B7_jUPHp.d.ts.map +1 -0
  106. package/dist/{trace-questions--lnJTaou.d.ts → trace-questions-Dd3AMf-U.d.ts} +3 -3
  107. package/dist/{trace-questions--lnJTaou.d.ts.map → trace-questions-Dd3AMf-U.d.ts.map} +1 -1
  108. package/dist/trace-repair/index.d.ts +1 -1
  109. package/dist/traces.d.ts +5 -5
  110. package/dist/traces.js +2 -2
  111. package/dist/{types-CeDfCRXp.d.ts → types-6VEehmEg.d.ts} +2 -2
  112. package/dist/{types-CeDfCRXp.d.ts.map → types-6VEehmEg.d.ts.map} +1 -1
  113. package/dist/{types-C0bWoSKC.d.ts → types-DLxe96rp.d.ts} +2 -2
  114. package/dist/{types-C0bWoSKC.d.ts.map → types-DLxe96rp.d.ts.map} +1 -1
  115. package/dist/{types-B6JadxWL.d.ts → types-DZQpUJse.d.ts} +2 -2
  116. package/dist/{types-B6JadxWL.d.ts.map → types-DZQpUJse.d.ts.map} +1 -1
  117. package/dist/{types-khIUQ0O5.d.ts → types-ZUy5HGY5.d.ts} +11 -5
  118. package/dist/{types-khIUQ0O5.d.ts.map → types-ZUy5HGY5.d.ts.map} +1 -1
  119. package/dist/wire/index.d.ts +2 -2
  120. package/dist/wire/index.js +1 -1
  121. package/docs/changes/llm-physical-attempts.md +30 -0
  122. package/docs/changes/shared-model-request-lifetime.md +30 -0
  123. package/docs/concepts.md +8 -0
  124. package/docs/eval-surface-map.md +12 -2
  125. package/docs/http-dispatch.md +33 -0
  126. package/docs/proofs/campaign-attempt-retention-20261001/README.md +70 -0
  127. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-judge-attribution/attribution_0/attempts/e7949644-3d60-4ab9-817e-f80d0a48f5d8/attempt-1/artifacts/capture.json +3 -0
  128. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-judge-attribution/attribution_0/attempts/e7949644-3d60-4ab9-817e-f80d0a48f5d8/attempt-1/identity.json +10 -0
  129. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-judge-attribution/attribution_0/attempts/e7949644-3d60-4ab9-817e-f80d0a48f5d8/attempt-1/result.json +1 -0
  130. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-judge-attribution/attribution_0/attempts/e7949644-3d60-4ab9-817e-f80d0a48f5d8/attempt-1/trace/spans.jsonl +0 -0
  131. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-judge-attribution/attribution_0/cached-result.json +1 -0
  132. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-judge-attribution/attribution_0/latest-attempt.json +1 -0
  133. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-legacy/capture-retention_0/attempts/7c7718f6-c85e-4867-9e3a-13ef9513de80/attempt-1/artifacts/captured-response.json +4 -0
  134. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-legacy/capture-retention_0/attempts/7c7718f6-c85e-4867-9e3a-13ef9513de80/attempt-1/identity.json +10 -0
  135. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-legacy/capture-retention_0/attempts/7c7718f6-c85e-4867-9e3a-13ef9513de80/attempt-1/result.json +1 -0
  136. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-legacy/capture-retention_0/attempts/7c7718f6-c85e-4867-9e3a-13ef9513de80/attempt-1/trace/spans.jsonl +1 -0
  137. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-legacy/capture-retention_0/cached-result.json +1 -0
  138. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-legacy/capture-retention_0/captured-response.json +6 -0
  139. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-legacy/capture-retention_0/failure-receipt.json +58 -0
  140. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-legacy/capture-retention_0/latest-attempt.json +1 -0
  141. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-legacy/capture-retention_0/spans.jsonl +1 -0
  142. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/baseline/capture-retention_0/attempts/3652118b-e1dc-4b17-bc15-6b802dc71eb1/attempt-1/artifacts/captured-response.json +4 -0
  143. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/baseline/capture-retention_0/attempts/3652118b-e1dc-4b17-bc15-6b802dc71eb1/attempt-1/identity.json +10 -0
  144. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/baseline/capture-retention_0/attempts/3652118b-e1dc-4b17-bc15-6b802dc71eb1/attempt-1/result.json +1 -0
  145. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/baseline/capture-retention_0/attempts/3652118b-e1dc-4b17-bc15-6b802dc71eb1/attempt-1/trace/spans.jsonl +1 -0
  146. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/baseline/capture-retention_0/cached-result.json +1 -0
  147. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/baseline/capture-retention_0/latest-attempt.json +1 -0
  148. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/candidates/42b3fec6b7e55008/capture-retention_0/attempts/40e411fe-ad12-4e6b-a279-2cf07c356a2f/attempt-1/artifacts/captured-response.json +4 -0
  149. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/candidates/42b3fec6b7e55008/capture-retention_0/attempts/40e411fe-ad12-4e6b-a279-2cf07c356a2f/attempt-1/failure-receipt.json +62 -0
  150. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/candidates/42b3fec6b7e55008/capture-retention_0/attempts/40e411fe-ad12-4e6b-a279-2cf07c356a2f/attempt-1/identity.json +10 -0
  151. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/candidates/42b3fec6b7e55008/capture-retention_0/attempts/40e411fe-ad12-4e6b-a279-2cf07c356a2f/attempt-1/result.json +1 -0
  152. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/candidates/42b3fec6b7e55008/capture-retention_0/attempts/40e411fe-ad12-4e6b-a279-2cf07c356a2f/attempt-1/trace/spans.jsonl +1 -0
  153. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/candidates/42b3fec6b7e55008/capture-retention_0/latest-attempt.json +1 -0
  154. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/candidates/87740cc101602a08/capture-retention_0/attempts/9e7da63e-2d19-4ec1-a5ec-ec804e2b0f40/attempt-1/artifacts/captured-response.json +4 -0
  155. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/candidates/87740cc101602a08/capture-retention_0/attempts/9e7da63e-2d19-4ec1-a5ec-ec804e2b0f40/attempt-1/identity.json +10 -0
  156. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/candidates/87740cc101602a08/capture-retention_0/attempts/9e7da63e-2d19-4ec1-a5ec-ec804e2b0f40/attempt-1/result.json +1 -0
  157. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/candidates/87740cc101602a08/capture-retention_0/attempts/9e7da63e-2d19-4ec1-a5ec-ec804e2b0f40/attempt-1/trace/spans.jsonl +1 -0
  158. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/candidates/87740cc101602a08/capture-retention_0/cached-result.json +1 -0
  159. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/candidates/87740cc101602a08/capture-retention_0/latest-attempt.json +1 -0
  160. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/first-interruption.json +38 -0
  161. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/blobs/19147243d263e55b1eaf1c940691947382a7cce17c5f49e4a4da765a473d2a55.json +1 -0
  162. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/blobs/5dd0d871d559d1c55b2ac04b3fbd26d62f7c9e25d7632d4542502cd292818fc0.json +1 -0
  163. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/blobs/5de45d632937e75ff90fec9f1f2ec425904d3632da71f6b26f83b1e1eb8cde5a.json +1 -0
  164. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/blobs/6caba627806e499aa3d9e156aec9d5a731a3ef30e29756d343b85b7217c77c11.json +1 -0
  165. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/blobs/ce2770851b3b7651f27cedcbad0372b9191e9a9b016665987e3ddd2ad53012b5.json +1 -0
  166. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/blobs/d8ac4ba8d9844afb4defa5cd277fe3cb944d99e58cf41c07d3ae199b6f467966.json +1 -0
  167. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/blobs/f59a02035fe96ea2f5582dd715d2e3ad8c184ae959310978962f1ac0780753bd.json +1 -0
  168. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/ledger.jsonl +23 -0
  169. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/ledger.jsonl.head +1 -0
  170. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-path-guard/capture-retention_0/attempts/3c8272d6-d35b-477e-9d08-085261b8dc64/attempt-1/artifacts/capture.json +3 -0
  171. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-path-guard/capture-retention_0/attempts/3c8272d6-d35b-477e-9d08-085261b8dc64/attempt-1/identity.json +10 -0
  172. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-path-guard/capture-retention_0/attempts/3c8272d6-d35b-477e-9d08-085261b8dc64/attempt-1/result.json +1 -0
  173. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-path-guard/capture-retention_0/attempts/3c8272d6-d35b-477e-9d08-085261b8dc64/attempt-1/trace/spans.jsonl +0 -0
  174. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-path-guard/capture-retention_0/cached-result.json +1 -0
  175. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-path-guard/capture-retention_0/latest-attempt.json +1 -0
  176. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-resume/capture-retention_0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1/artifacts/captured-response.json +6 -0
  177. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-resume/capture-retention_0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1/failure-receipt.json +81 -0
  178. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-resume/capture-retention_0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1/identity.json +10 -0
  179. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-resume/capture-retention_0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1/result.json +1 -0
  180. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-resume/capture-retention_0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1/trace/spans.jsonl +2 -0
  181. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-resume/capture-retention_0/attempts/c4859951-1e7c-4d23-b89a-0ca8fa79e2cf/attempt-1/artifacts/captured-response.json +6 -0
  182. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-resume/capture-retention_0/attempts/c4859951-1e7c-4d23-b89a-0ca8fa79e2cf/attempt-1/identity.json +10 -0
  183. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-resume/capture-retention_0/attempts/c4859951-1e7c-4d23-b89a-0ca8fa79e2cf/attempt-1/result.json +1 -0
  184. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-resume/capture-retention_0/attempts/c4859951-1e7c-4d23-b89a-0ca8fa79e2cf/attempt-1/trace/spans.jsonl +2 -0
  185. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-resume/capture-retention_0/cached-result.json +1 -0
  186. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-resume/capture-retention_0/latest-attempt.json +1 -0
  187. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-resume/observation.json +608 -0
  188. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-1/artifacts/captured-response.json +6 -0
  189. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-1/failure-receipt.json +81 -0
  190. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-1/identity.json +10 -0
  191. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-1/result.json +1 -0
  192. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-1/trace/spans.jsonl +2 -0
  193. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-2/artifacts/captured-response.json +6 -0
  194. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-2/identity.json +10 -0
  195. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-2/result.json +1 -0
  196. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-2/trace/spans.jsonl +2 -0
  197. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-retry/capture-retention_0/cached-result.json +1 -0
  198. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-retry/capture-retention_0/latest-attempt.json +1 -0
  199. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-retry/observation.json +514 -0
  200. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-summary.json +1853 -0
  201. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/judge-attribution.json +140 -0
  202. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/qualification.log +1 -0
  203. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/qualify.mjs +165 -0
  204. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduce.mjs +39 -0
  205. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-resume/capture-retention_0/cached-result.json +1 -0
  206. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-resume/capture-retention_0/captured-response.json +5 -0
  207. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-resume/capture-retention_0/failure-receipt.json +58 -0
  208. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-resume/capture-retention_0/spans.jsonl +1 -0
  209. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-resume/observer-retained-captures.json +20 -0
  210. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-retry/capture-retention_0/cached-result.json +1 -0
  211. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-retry/capture-retention_0/captured-response.json +5 -0
  212. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-retry/capture-retention_0/failure-receipt.attempt-1.json +58 -0
  213. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-retry/capture-retention_0/spans.jsonl +1 -0
  214. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-retry/observer-retained-captures.json +20 -0
  215. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-summary.json +315 -0
  216. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction.log +1 -0
  217. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/source-and-build.json +2409 -0
  218. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/typecheck.log +1 -0
  219. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/verify-package.log +340 -0
  220. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-legacy/capture-retention_0/attempts/33d57ccb-d4d8-4c23-92e3-c1ddf7db6058/attempt-1/artifacts/captured-response.json +4 -0
  221. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-legacy/capture-retention_0/attempts/33d57ccb-d4d8-4c23-92e3-c1ddf7db6058/attempt-1/identity.json +10 -0
  222. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-legacy/capture-retention_0/attempts/33d57ccb-d4d8-4c23-92e3-c1ddf7db6058/attempt-1/result.json +1 -0
  223. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-legacy/capture-retention_0/attempts/33d57ccb-d4d8-4c23-92e3-c1ddf7db6058/attempt-1/trace/spans.jsonl +1 -0
  224. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-legacy/capture-retention_0/cached-result.json +1 -0
  225. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-legacy/capture-retention_0/captured-response.json +6 -0
  226. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-legacy/capture-retention_0/failure-receipt.json +58 -0
  227. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-legacy/capture-retention_0/latest-attempt.json +1 -0
  228. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-legacy/capture-retention_0/spans.jsonl +1 -0
  229. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/baseline/capture-retention_0/attempts/6b25a1d2-5324-445b-b59d-bdc7381cd4b4/attempt-1/artifacts/captured-response.json +4 -0
  230. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/baseline/capture-retention_0/attempts/6b25a1d2-5324-445b-b59d-bdc7381cd4b4/attempt-1/identity.json +10 -0
  231. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/baseline/capture-retention_0/attempts/6b25a1d2-5324-445b-b59d-bdc7381cd4b4/attempt-1/result.json +1 -0
  232. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/baseline/capture-retention_0/attempts/6b25a1d2-5324-445b-b59d-bdc7381cd4b4/attempt-1/trace/spans.jsonl +1 -0
  233. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/baseline/capture-retention_0/cached-result.json +1 -0
  234. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/baseline/capture-retention_0/latest-attempt.json +1 -0
  235. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/first-interruption.json +5 -0
  236. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/search/blobs/5de45d632937e75ff90fec9f1f2ec425904d3632da71f6b26f83b1e1eb8cde5a.json +1 -0
  237. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/search/ledger.jsonl +7 -0
  238. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/search/ledger.jsonl.head +1 -0
  239. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-path-guard/capture-retention_0/attempts/723fa9c6-099e-437c-bddf-a4cbd28a8508/attempt-1/artifacts/capture.json +3 -0
  240. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-path-guard/capture-retention_0/attempts/723fa9c6-099e-437c-bddf-a4cbd28a8508/attempt-1/identity.json +10 -0
  241. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-path-guard/capture-retention_0/attempts/723fa9c6-099e-437c-bddf-a4cbd28a8508/attempt-1/result.json +1 -0
  242. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-path-guard/capture-retention_0/attempts/723fa9c6-099e-437c-bddf-a4cbd28a8508/attempt-1/trace/spans.jsonl +0 -0
  243. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-path-guard/capture-retention_0/cached-result.json +1 -0
  244. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-path-guard/capture-retention_0/latest-attempt.json +1 -0
  245. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-resume/capture-retention_0/attempts/06db3429-c5c6-42c1-8623-08e2da1f563e/attempt-1/artifacts/captured-response.json +6 -0
  246. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-resume/capture-retention_0/attempts/06db3429-c5c6-42c1-8623-08e2da1f563e/attempt-1/failure-receipt.json +62 -0
  247. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-resume/capture-retention_0/attempts/06db3429-c5c6-42c1-8623-08e2da1f563e/attempt-1/identity.json +10 -0
  248. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-resume/capture-retention_0/attempts/06db3429-c5c6-42c1-8623-08e2da1f563e/attempt-1/result.json +1 -0
  249. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-resume/capture-retention_0/attempts/06db3429-c5c6-42c1-8623-08e2da1f563e/attempt-1/trace/spans.jsonl +1 -0
  250. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-resume/capture-retention_0/attempts/d49f7e40-c4d2-4cb6-98d1-5e49f45c6e2c/attempt-1/artifacts/captured-response.json +6 -0
  251. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-resume/capture-retention_0/attempts/d49f7e40-c4d2-4cb6-98d1-5e49f45c6e2c/attempt-1/identity.json +10 -0
  252. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-resume/capture-retention_0/attempts/d49f7e40-c4d2-4cb6-98d1-5e49f45c6e2c/attempt-1/result.json +1 -0
  253. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-resume/capture-retention_0/attempts/d49f7e40-c4d2-4cb6-98d1-5e49f45c6e2c/attempt-1/trace/spans.jsonl +1 -0
  254. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-resume/capture-retention_0/cached-result.json +1 -0
  255. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-resume/capture-retention_0/latest-attempt.json +1 -0
  256. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-resume/observation.json +458 -0
  257. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-retry/capture-retention_0/attempts/cf34f352-67e5-4e8b-a510-0e896bcc96fc/attempt-1/artifacts/captured-response.json +6 -0
  258. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-retry/capture-retention_0/attempts/cf34f352-67e5-4e8b-a510-0e896bcc96fc/attempt-1/failure-receipt.json +62 -0
  259. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-retry/capture-retention_0/attempts/cf34f352-67e5-4e8b-a510-0e896bcc96fc/attempt-1/identity.json +10 -0
  260. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-retry/capture-retention_0/attempts/cf34f352-67e5-4e8b-a510-0e896bcc96fc/attempt-1/result.json +1 -0
  261. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-retry/capture-retention_0/attempts/cf34f352-67e5-4e8b-a510-0e896bcc96fc/attempt-1/trace/spans.jsonl +1 -0
  262. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-retry/capture-retention_0/attempts/cf34f352-67e5-4e8b-a510-0e896bcc96fc/attempt-2/artifacts/captured-response.json +6 -0
  263. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-retry/capture-retention_0/attempts/cf34f352-67e5-4e8b-a510-0e896bcc96fc/attempt-2/identity.json +10 -0
  264. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-retry/capture-retention_0/attempts/cf34f352-67e5-4e8b-a510-0e896bcc96fc/attempt-2/result.json +1 -0
  265. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-retry/capture-retention_0/attempts/cf34f352-67e5-4e8b-a510-0e896bcc96fc/attempt-2/trace/spans.jsonl +1 -0
  266. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-retry/capture-retention_0/cached-result.json +1 -0
  267. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-retry/capture-retention_0/latest-attempt.json +1 -0
  268. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-retry/observation.json +380 -0
  269. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-resume/capture-retention_0/cached-result.json +1 -0
  270. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-resume/capture-retention_0/captured-response.json +6 -0
  271. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-resume/capture-retention_0/failure-receipt.json +58 -0
  272. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-resume/capture-retention_0/spans.jsonl +1 -0
  273. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-resume/observation.json +198 -0
  274. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-retry/capture-retention_0/cached-result.json +1 -0
  275. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-retry/capture-retention_0/captured-response.json +6 -0
  276. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-retry/capture-retention_0/failure-receipt.attempt-1.json +58 -0
  277. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-retry/capture-retention_0/spans.jsonl +1 -0
  278. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-retry/observation.json +121 -0
  279. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-summary.json +321 -0
  280. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/build.attribution.log +491 -0
  281. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/build.final.log +491 -0
  282. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/build.log +495 -0
  283. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/qualification.log +18 -0
  284. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/qualification.wrong-root.log +9 -0
  285. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/qualify.mjs +148 -0
  286. package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/qualify.wrong-root.mjs +148 -0
  287. package/docs/proofs/campaign-attempt-retention-20261001/manifest.json +808 -0
  288. package/docs/proofs/router-cache-usage-20261001/after.json +2058 -0
  289. package/docs/proofs/router-cache-usage-20261001/component-projections.json +584 -0
  290. package/docs/proofs/router-cache-usage-20261001/component-projections.nested-only.json +306 -0
  291. package/docs/proofs/router-cache-usage-20261001/manifest.json +32 -0
  292. package/docs/proofs/router-cache-usage-20261001/summary.json +49 -0
  293. package/docs/proofs/router-cache-usage-20261001/summary.nested-only.json +29 -0
  294. package/package.json +1 -1
  295. package/dist/abort-signal-CtzAM_sJ.js.map +0 -1
  296. package/dist/benchmark-command-DFs_i62N.js.map +0 -1
  297. package/dist/extract-usage-DFATEuWW.js.map +0 -1
  298. package/dist/index-CAtNOBQY.d.ts.map +0 -1
  299. package/dist/llm-client-Bhqoglar.js.map +0 -1
  300. package/dist/llm-judge-C-0P1WiT.js.map +0 -1
  301. package/dist/tool-groups-Zm84Pqpk.d.ts.map +0 -1
@@ -2,6 +2,15 @@ import { t as AgentEvalError } from "./errors-Dngq5h35.js";
2
2
  import { c as costForTokenPricing } from "./cost-ledger-_IbkejCF.js";
3
3
  import { t as newRecordId } from "./record-id-DUgsK5qp.js";
4
4
  import { a as providerFromBaseUrl, i as defaultProviderRedactor } from "./raw-provider-sink-rB2DYxr6.js";
5
+ //#region src/abort-signal.ts
6
+ /** Combine active cancellation sources without wrapping a single source. */
7
+ function combineAbortSignals(...signals) {
8
+ const active = [...new Set(signals.filter((signal) => signal !== void 0))];
9
+ if (active.length === 0) return void 0;
10
+ if (active.length === 1) return active[0];
11
+ return AbortSignal.any(active);
12
+ }
13
+ //#endregion
5
14
  //#region src/judge-families.ts
6
15
  /** Explicit `provider/...` prefix → family (models.dev / OpenRouter style). */
7
16
  const PROVIDER_PREFIX = {
@@ -322,11 +331,9 @@ function maximumChargeForLlmRequest(request, options = {}) {
322
331
  if (request.messages.some((message) => Array.isArray(message.content) && message.content.some((part) => part.type === "image_url"))) return;
323
332
  const attempts = resolveMaximumAttempts(options.maximumAttempts);
324
333
  const forceJsonObject = options.jsonSchemaTransport === "json-object";
325
- const requestBytes = new TextEncoder().encode(JSON.stringify(buildBody(request, forceJsonObject, options.thinking))).byteLength;
326
- const batches = request.jsonSchema && !forceJsonObject ? 2 : 1;
327
334
  const usage = {
328
- inputTokens: requestBytes * attempts * batches,
329
- outputTokens: request.maxTokens * attempts * batches
335
+ inputTokens: new TextEncoder().encode(JSON.stringify(buildBody(request, forceJsonObject, options.thinking))).byteLength * attempts,
336
+ outputTokens: request.maxTokens * attempts
330
337
  };
331
338
  return options.customTokenPricing ? {
332
339
  customTokenPricing: options.customTokenPricing,
@@ -585,22 +592,21 @@ function parseWireToolCalls(value, model) {
585
592
  };
586
593
  });
587
594
  }
588
- async function sleep(ms) {
589
- return new Promise((resolve) => setTimeout(resolve, ms));
590
- }
591
- /**
592
- * Combine the per-attempt timeout signal with an optional caller signal into
593
- * one signal the fetch listens on. Prefers the native `AbortSignal.any`; falls
594
- * back to manual wiring on runtimes that predate it. The caller signal is also
595
- * propagated to the timeout controller so aborting it cancels the in-flight
596
- * fetch immediately.
597
- */
598
- function linkSignals(timeoutController, caller) {
599
- if (!caller) return timeoutController.signal;
600
- if (typeof AbortSignal.any === "function") return AbortSignal.any([timeoutController.signal, caller]);
601
- if (caller.aborted) timeoutController.abort();
602
- else caller.addEventListener("abort", () => timeoutController.abort(), { once: true });
603
- return timeoutController.signal;
595
+ /** Backoff belongs to the caller's operation and must stop when it is cancelled. */
596
+ async function sleep(ms, signal) {
597
+ signal?.throwIfAborted();
598
+ return new Promise((resolve, reject) => {
599
+ const onAbort = () => {
600
+ clearTimeout(timer);
601
+ signal?.removeEventListener("abort", onAbort);
602
+ reject(signal?.reason);
603
+ };
604
+ const timer = setTimeout(() => {
605
+ signal?.removeEventListener("abort", onAbort);
606
+ resolve();
607
+ }, ms);
608
+ signal?.addEventListener("abort", onAbort, { once: true });
609
+ });
604
610
  }
605
611
  /** True once the cross-attempt wall-clock budget (if any) is exhausted. */
606
612
  function deadlineExceeded(start, deadlineMs) {
@@ -672,6 +678,11 @@ function extractBalancedJson(input, start) {
672
678
  * degrade use `callLlmJson`.
673
679
  */
674
680
  async function callLlm(req, opts = {}) {
681
+ return callLlmAttempts(req, opts, false);
682
+ }
683
+ /** One retry owner for raw and structured calls. Fallback changes the next
684
+ * request, not the attempt allowance, deadline origin, or raw-event sequence. */
685
+ async function callLlmAttempts(req, opts, allowSchemaFallback) {
675
686
  if (!opts.baseUrl) throw new Error("callLlm: opts.baseUrl is required");
676
687
  const baseUrl = opts.baseUrl.replace(/\/+$/, "");
677
688
  const url = `${baseUrl}/chat/completions`;
@@ -693,41 +704,46 @@ async function callLlm(req, opts = {}) {
693
704
  });
694
705
  let lastErr;
695
706
  let effectiveRequest = req;
707
+ const remainingMs = () => deadlineMs === void 0 ? Infinity : Math.max(0, deadlineMs - (Date.now() - deadlineStart));
708
+ const retryDelay = (ms) => sleep(Math.min(ms, remainingMs()), callerSignal);
696
709
  for (let attempt = 0; attempt < maximumAttempts; attempt++) {
697
710
  if (callerSignal?.aborted) throw new DOMException("callLlm aborted by caller signal", "AbortError");
698
- if (attempt > 0 && deadlineExceeded(deadlineStart, deadlineMs)) throw lastErr instanceof Error ? lastErr : new Error(String(lastErr));
711
+ if (deadlineExceeded(deadlineStart, deadlineMs)) throw lastErr instanceof Error ? lastErr : new DOMException("callLlm deadline exceeded", "TimeoutError");
712
+ const requestBody = buildBody(effectiveRequest, opts.jsonSchemaTransport === "json-object", opts.thinking);
699
713
  const controller = new AbortController();
700
- const attemptSignal = linkSignals(controller, callerSignal);
701
- const timeoutHandle = setTimeout(() => controller.abort(), timeoutMs);
714
+ const attemptSignal = combineAbortSignals(controller.signal, callerSignal);
715
+ const timeoutHandle = setTimeout(() => controller.abort(), Math.min(timeoutMs, remainingMs()));
702
716
  const started = Date.now();
703
- const requestBody = buildBody(effectiveRequest, opts.jsonSchemaTransport === "json-object", opts.thinking);
704
717
  let attemptErrorRecorded = false;
705
- if (sink) await recordRaw(sink, redactor, {
706
- eventId: newRecordId(),
707
- runId: traceContext?.runId,
708
- spanId: traceContext?.spanId,
709
- provider,
710
- model: req.model,
711
- endpoint,
712
- baseUrl,
713
- attemptIndex: attempt,
714
- direction: "request",
715
- timestamp: started,
716
- requestHeaders: headers,
717
- requestBody,
718
- redactedFields: []
719
- });
720
718
  try {
719
+ if (sink) await recordRaw(sink, redactor, {
720
+ eventId: newRecordId(),
721
+ runId: traceContext?.runId,
722
+ spanId: traceContext?.spanId,
723
+ provider,
724
+ model: req.model,
725
+ endpoint,
726
+ baseUrl,
727
+ attemptIndex: attempt,
728
+ direction: "request",
729
+ timestamp: started,
730
+ requestHeaders: headers,
731
+ requestBody,
732
+ redactedFields: []
733
+ });
734
+ attemptSignal.throwIfAborted();
721
735
  const res = await fetchFn(url, {
722
736
  method: "POST",
723
737
  headers,
724
738
  body: JSON.stringify(requestBody),
725
739
  signal: attemptSignal
726
740
  });
727
- clearTimeout(timeoutHandle);
728
741
  const responseHeaders = sink ? headersToObject(res.headers) : void 0;
742
+ const text = await res.text();
743
+ attemptSignal.throwIfAborted();
744
+ clearTimeout(timeoutHandle);
729
745
  if (!res.ok) {
730
- const body = await res.text();
746
+ const body = text;
731
747
  if (sink) {
732
748
  await recordRaw(sink, redactor, {
733
749
  eventId: newRecordId(),
@@ -758,14 +774,22 @@ async function callLlm(req, opts = {}) {
758
774
  };
759
775
  continue;
760
776
  }
777
+ if (allowSchemaFallback && opts.jsonSchemaTransport !== "json-object" && effectiveRequest.jsonSchema && isSchemaRejection(res.status, body) && attempt < maximumAttempts - 1 && !deadlineExceeded(deadlineStart, deadlineMs)) {
778
+ lastErr = err;
779
+ effectiveRequest = {
780
+ ...effectiveRequest,
781
+ jsonMode: true,
782
+ jsonSchema: void 0
783
+ };
784
+ continue;
785
+ }
761
786
  if (RETRYABLE_STATUS.has(res.status) && attempt < maximumAttempts - 1 && !deadlineExceeded(deadlineStart, deadlineMs)) {
762
787
  lastErr = err;
763
- await sleep(parseRetryAfter(res.headers) ?? backoffMs(attempt));
788
+ await retryDelay(parseRetryAfter(res.headers) ?? backoffMs(attempt));
764
789
  continue;
765
790
  }
766
791
  throw err;
767
792
  }
768
- const text = await res.text();
769
793
  let json;
770
794
  try {
771
795
  json = JSON.parse(text);
@@ -821,7 +845,7 @@ async function callLlm(req, opts = {}) {
821
845
  const cachedRaw = usageRaw?.prompt_tokens_details && typeof usageRaw.prompt_tokens_details === "object" && !Array.isArray(usageRaw.prompt_tokens_details) ? usageRaw.prompt_tokens_details.cached_tokens : void 0;
822
846
  const cachedPromptTokens = cachedRaw === void 0 ? void 0 : providerTokenCount(cachedRaw);
823
847
  const usageCaptured = promptTokens !== void 0 && completionTokens !== void 0 && (reasoningRaw === void 0 || reasoningTokens !== void 0 && reasoningTokens <= completionTokens) && (cachedRaw === void 0 || cachedPromptTokens !== void 0 && cachedPromptTokens <= promptTokens) && (totalTokens === void 0 || totalTokens === promptTokens + completionTokens);
824
- const costFromProxy = json._response_cost ?? json.cost_usd;
848
+ const costFromProxy = providerReportedCost(json);
825
849
  const content = choice?.message?.content ?? "";
826
850
  const configuredCost = typeof costFromProxy !== "number" && usageCaptured && opts.customTokenPricing ? costForTokenPricing(opts.customTokenPricing, {
827
851
  inputTokens: promptTokens - (cachedPromptTokens ?? 0),
@@ -887,10 +911,12 @@ async function callLlm(req, opts = {}) {
887
911
  redactedFields: []
888
912
  });
889
913
  if (attempt < maximumAttempts - 1 && isTransientLlmError(err) && !deadlineExceeded(deadlineStart, deadlineMs)) {
890
- await sleep(backoffMs(attempt));
914
+ await retryDelay(backoffMs(attempt));
891
915
  continue;
892
916
  }
893
917
  throw err;
918
+ } finally {
919
+ clearTimeout(timeoutHandle);
894
920
  }
895
921
  }
896
922
  throw lastErr instanceof Error ? lastErr : new Error(String(lastErr));
@@ -922,19 +948,10 @@ async function callLlmJson(req, opts = {}) {
922
948
  }
923
949
  /** Shared schema-to-JSON-mode fallback that preserves the raw result. */
924
950
  async function callLlmStructured(req, opts = {}) {
925
- try {
926
- return await callLlm({
927
- ...req,
928
- jsonMode: req.jsonMode ?? !req.jsonSchema
929
- }, opts);
930
- } catch (err) {
931
- if (opts.jsonSchemaTransport !== "json-object" && err instanceof LlmCallError && isSchemaRejection(err.status, err.body) && req.jsonSchema) return await callLlm({
932
- ...req,
933
- jsonMode: true,
934
- jsonSchema: void 0
935
- }, opts);
936
- throw err;
937
- }
951
+ return callLlmAttempts({
952
+ ...req,
953
+ jsonMode: req.jsonMode ?? !req.jsonSchema
954
+ }, opts, true);
938
955
  }
939
956
  function parseJsonResult(result, jsonPayloadMode) {
940
957
  try {
@@ -981,6 +998,6 @@ var LlmClient = class {
981
998
  }
982
999
  };
983
1000
  //#endregion
984
- export { judgeFamily as S, assertServedModels as _, callLlmJson as a, CrossFamilyError as b, extractJsonPayload as c, stripFencedJson as d, ModelSubstitutionError as f, assertServedModelPolicy as g, assertServedModel as h, backoffMs as i, isTransientLlmError as l, assertCrossFamilyServed as m, LlmClient as n, costReceiptFromLlm as o, ServedCrossFamilyError as p, LlmResponseError as r, costReceiptFromLlmError as s, LlmCallError as t, maximumChargeForLlmRequest as u, checkServedModel as v, assertCrossFamily as x, servedModelAcceptable as y };
1001
+ export { combineAbortSignals as C, judgeFamily as S, assertServedModels as _, callLlmJson as a, CrossFamilyError as b, extractJsonPayload as c, stripFencedJson as d, ModelSubstitutionError as f, assertServedModelPolicy as g, assertServedModel as h, backoffMs as i, isTransientLlmError as l, assertCrossFamilyServed as m, LlmClient as n, costReceiptFromLlm as o, ServedCrossFamilyError as p, LlmResponseError as r, costReceiptFromLlmError as s, LlmCallError as t, maximumChargeForLlmRequest as u, checkServedModel as v, assertCrossFamily as x, servedModelAcceptable as y };
985
1002
 
986
- //# sourceMappingURL=llm-client-Bhqoglar.js.map
1003
+ //# sourceMappingURL=llm-client-DmCNzrc9.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"llm-client-DmCNzrc9.js","names":[],"sources":["../src/abort-signal.ts","../src/judge-families.ts","../src/integrity/served-model.ts","../src/llm-client.ts"],"sourcesContent":["/** Combine active cancellation sources without wrapping a single source. */\nexport function combineAbortSignals(\n ...signals: Array<AbortSignal | undefined>\n): AbortSignal | undefined {\n const active = [\n ...new Set(signals.filter((signal): signal is AbortSignal => signal !== undefined)),\n ]\n if (active.length === 0) return undefined\n if (active.length === 1) return active[0]\n return AbortSignal.any(active)\n}\n","/**\n * Judge model-family classification + cross-family enforcement.\n *\n * A judge ensemble built entirely from one provider family shares that\n * family's blind spots and self-preference — its \"agreement\" is correlated\n * bias, not independent signal. `assertCrossFamily` makes the consumer prove\n * the ensemble spans ≥2 families; `judgeFamily` is the single regex map that\n * replaces the per-consumer copies (tax/legal/creative/gtm each ship one).\n */\n\n/** Provider family a model belongs to. `unknown` when no rule matches. */\nexport type JudgeFamily =\n | 'anthropic'\n | 'openai'\n | 'google'\n | 'meta'\n | 'mistral'\n | 'deepseek'\n | 'xai'\n | 'qwen'\n | 'cohere'\n | 'amazon'\n | 'moonshot'\n | 'zhipu'\n | 'unknown'\n\n/** Explicit `provider/...` prefix → family (models.dev / OpenRouter style). */\nconst PROVIDER_PREFIX: Record<string, JudgeFamily> = {\n anthropic: 'anthropic',\n openai: 'openai',\n 'azure-openai': 'openai',\n google: 'google',\n 'google-vertex': 'google',\n meta: 'meta',\n 'meta-llama': 'meta',\n mistral: 'mistral',\n mistralai: 'mistral',\n deepseek: 'deepseek',\n xai: 'xai',\n qwen: 'qwen',\n alibaba: 'qwen',\n cohere: 'cohere',\n amazon: 'amazon',\n bedrock: 'amazon',\n moonshot: 'moonshot',\n moonshotai: 'moonshot',\n kimi: 'moonshot',\n 'kimi-code': 'moonshot',\n zhipu: 'zhipu',\n zhipuai: 'zhipu',\n zai: 'zhipu',\n 'z-ai': 'zhipu',\n glm: 'zhipu',\n}\n\n/** Fallback model-name patterns when there's no recognised provider prefix. */\nconst NAME_PATTERNS: Array<[RegExp, JudgeFamily]> = [\n [/claude/i, 'anthropic'],\n // Anthropic's own bare aliases (`claude --model sonnet|opus|haiku`). A run that requested `sonnet`\n // and was served claude-sonnet-5 is an alias resolution, not a cross-family substitution: without\n // this row the alias classified as `unknown` and assertServedModel refused the pair.\n [/^(?:sonnet|opus|haiku)(?:[-@\\s]|$)/i, 'anthropic'],\n [/\\b(gpt|davinci|babbage)\\b|^o[134]\\b|[-/]o[134]\\b|gpt-/i, 'openai'],\n [/gemini|palm|gemma|bison/i, 'google'],\n [/llama/i, 'meta'],\n [/mi(s|x)tral|codestral|magistral/i, 'mistral'],\n [/deepseek/i, 'deepseek'],\n [/grok/i, 'xai'],\n [/qwen/i, 'qwen'],\n [/command-?(r|a)?/i, 'cohere'],\n [/\\b(nova|titan)\\b/i, 'amazon'],\n [/\\bkimi\\b|moonshot/i, 'moonshot'],\n [/\\bglm\\b|zhipu|\\bz-?ai\\b/i, 'zhipu'],\n]\n\n/**\n * Classify a model id into its provider family. Strips a `@snapshot` suffix\n * and prefers an explicit `provider/...` prefix; otherwise matches the model\n * name. Returns `unknown` when nothing matches (callers decide whether that's\n * acceptable — `assertCrossFamily` counts it as its own family).\n */\nexport function judgeFamily(modelId: string): JudgeFamily {\n const id = modelId.trim().split('@')[0]!.toLowerCase()\n const slash = id.indexOf('/')\n if (slash > 0) {\n const prefix = id.slice(0, slash)\n const mapped = PROVIDER_PREFIX[prefix]\n if (mapped) return mapped\n }\n for (const [pattern, family] of NAME_PATTERNS) {\n if (pattern.test(id)) return family\n }\n return 'unknown'\n}\n\nexport interface AssertCrossFamilyOptions {\n /** Minimum number of distinct families the ensemble must span. Default 2. */\n minFamilies?: number\n /** When false (default), `unknown`-family models do NOT count toward the\n * family total — an ensemble of all-unclassifiable models is not provably\n * cross-family. Set true to count `unknown` as one shared family. */\n allowUnknown?: boolean\n}\n\nexport class CrossFamilyError extends Error {\n constructor(\n message: string,\n public readonly families: JudgeFamily[],\n public readonly models: string[],\n ) {\n super(message)\n this.name = 'CrossFamilyError'\n }\n}\n\n/**\n * Throw unless the judge models span at least `minFamilies` distinct provider\n * families. Pass the model ids backing your judge ensemble. Fail-loud by\n * design — a correlated single-family ensemble silently inflates agreement.\n *\n * Scope: this reads the ids you REQUEST. It proves the panel was configured\n * across families; it cannot prove the panel RAN across families, because a\n * routing gateway may answer several different ids from one provider. Where\n * the diversity claim is load-bearing (a published leaderboard, a\n * certification, a non-self-judging exclusion), assert on the ids the\n * provider echoed instead: `assertCrossFamilyServed` in\n * ./integrity/served-model.\n */\nexport function assertCrossFamily(\n models: string[],\n opts: AssertCrossFamilyOptions = {},\n): JudgeFamily[] {\n const minFamilies = opts.minFamilies ?? 2\n const families = new Set<JudgeFamily>()\n for (const m of models) {\n const f = judgeFamily(m)\n if (f === 'unknown' && !opts.allowUnknown) continue\n families.add(f)\n }\n const list = [...families].sort()\n if (list.length < minFamilies) {\n throw new CrossFamilyError(\n `judge ensemble spans ${list.length} provider famil${list.length === 1 ? 'y' : 'ies'} ` +\n `(${list.join(', ') || 'none'}) but ${minFamilies} required — a single-family ensemble ` +\n 'is correlated bias, not independent signal',\n list,\n models,\n )\n }\n return list\n}\n","/**\n * Served-model identity: prove the model that ANSWERED is the model that was\n * REQUESTED.\n *\n * A routing gateway can accept `model: \"gpt-4.1-mini\"` and answer from a\n * different model entirely. Every guard that reasons about the requested id —\n * cross-family judge diversity, per-model leaderboard rows, non-self-judging\n * exclusions, cost attribution — then passes while measuring something else.\n * The requested id is an intent; only the id echoed on the response is\n * evidence.\n *\n * `checkServedModel` classifies one requested/served pair; `assertServedModel`\n * and `assertServedModels` are the fail-loud gates; `assertCrossFamilyServed`\n * is the family-diversity rule computed over SERVED ids (the requested-id\n * version lives in ../judge-families and cannot see substitution).\n *\n * Aliasing is tolerated, substitution is not: `openai/gpt-4.1-mini` and\n * `gpt-4.1-mini` are the same request expressed two ways, so a response\n * echoing either satisfies the other. A response echoing `gemini-2.5-flash-lite`\n * does not.\n */\n\nimport { AgentEvalError } from '../errors'\nimport { type JudgeFamily, judgeFamily } from '../judge-families'\n\n/**\n * Output-token budget a liveness probe must grant a model.\n *\n * Identity is only readable off a response the provider actually produced. A\n * reasoning model spends budget on hidden reasoning tokens before it emits a\n * single visible one, so a cap of a few tokens makes a HEALTHY deepseek/glm\n * model fail with `reasoning_budget_exhausted` — it names no model, its\n * identity reads as `unreported`, and a preflight scores it DEAD. 64 clears\n * that floor. Every probe in this package reads this constant, so two probes\n * cannot reach two different answers about the same router.\n *\n * Cost: a probe spends at most `PROBE_MAX_TOKENS` output tokens per model,\n * plus whatever reasoning tokens a reasoning model bills — roughly 400 output\n * tokens for a six-model preflight. That is fractions of a cent, and far\n * cheaper than a campaign that runs on a model nobody proved was alive.\n */\nexport const PROBE_MAX_TOKENS = 64\n\n/** How a served id relates to the id that was requested. */\nexport type ServedModelVerdict =\n /** Byte-identical after normalisation — the requested model answered. */\n | 'exact'\n /** Same model, different spelling (provider prefix, snapshot, tier suffix). */\n | 'alias'\n /** A different model of the SAME provider family answered. */\n | 'substituted-within-family'\n /** A different provider's model answered. */\n | 'substituted-cross-family'\n /** The response carried no model id — identity is unproven either way. */\n | 'unreported'\n\nexport interface ServedModelCheck {\n /** The id the caller asked for. */\n requested: string\n /** The id echoed on the response; `null` when the response omitted it. */\n served: string | null\n requestedFamily: JudgeFamily\n /** `null` when `served` is null. */\n servedFamily: JudgeFamily | null\n verdict: ServedModelVerdict\n /** True for every verdict except `exact` and `alias`. */\n substituted: boolean\n}\n\n/**\n * Reduce a model id to its comparable core: lowercase, no surrounding space,\n * no `provider/` prefix, no `@snapshot` / `:batch` / `:free` tier suffix, no\n * trailing build date, and `.`/`_` folded to `-` so one version is spelled one\n * way.\n *\n * Dropping the build date is what makes snapshot resolution legible as the\n * non-event it is: a router answering `gpt-4o-mini` with\n * `gpt-4o-mini-2024-07-18` pinned a floating alias to a reproducible build —\n * the same model, which is the behaviour we want. Only routing decoration is\n * stripped; version digits are load-bearing, so `deepseek-v3.2` and\n * `deepseek-v4-flash` stay distinct, and comparison is EXACT equality rather\n * than a prefix test (a prefix rule would accept `gpt-5` → `gpt-5-mini`, a\n * silent downgrade wearing the right vendor name).\n */\nexport function normalizeModelId(modelId: string): string {\n let id = modelId.trim().toLowerCase()\n const at = id.indexOf('@')\n if (at > 0) id = id.slice(0, at)\n const colon = id.indexOf(':')\n if (colon > 0) id = id.slice(0, colon)\n const slash = id.lastIndexOf('/')\n if (slash >= 0) id = id.slice(slash + 1)\n return id\n .replace(/-\\d{4}-\\d{2}-\\d{2}$/, '')\n .replace(/-\\d{8}$/, '')\n .replace(/[._]/g, '-')\n .replace(/-+$/, '')\n .trim()\n}\n\n/**\n * Classify one requested/served pair. Pure — no I/O — so it is safe inside\n * response handlers, reducers, and CI gates.\n *\n * `served` is the id echoed by the provider (OpenAI-compatible bodies put it\n * at `model`). `null`/`undefined` means the body omitted it; that is\n * `unreported`, NOT a pass — a provider that does not name what answered has\n * not proven identity, and a transport that drops the field must not read as\n * agreement.\n */\nexport function checkServedModel(\n requested: string,\n served: string | null | undefined,\n): ServedModelCheck {\n const requestedFamily = judgeFamily(requested)\n if (served === null || served === undefined || served.trim() === '') {\n return {\n requested,\n served: null,\n requestedFamily,\n servedFamily: null,\n verdict: 'unreported',\n substituted: true,\n }\n }\n const servedFamily = judgeFamily(served)\n if (requested.trim().toLowerCase() === served.trim().toLowerCase()) {\n return {\n requested,\n served,\n requestedFamily,\n servedFamily,\n verdict: 'exact',\n substituted: false,\n }\n }\n if (normalizeModelId(requested) === normalizeModelId(served)) {\n return {\n requested,\n served,\n requestedFamily,\n servedFamily,\n verdict: 'alias',\n substituted: false,\n }\n }\n return {\n requested,\n served,\n requestedFamily,\n servedFamily,\n verdict:\n requestedFamily === servedFamily ? 'substituted-within-family' : 'substituted-cross-family',\n substituted: true,\n }\n}\n\nexport class ModelSubstitutionError extends AgentEvalError {\n constructor(\n message: string,\n public readonly checks: ReadonlyArray<ServedModelCheck>,\n ) {\n super('model_substitution', message)\n this.name = 'ModelSubstitutionError'\n }\n}\n\n/**\n * Consumer-facing name for the substitution policy a metered surface applies.\n * `'exact'` rejects every substitution. `'allow-within-family'` accepts a\n * different model of the same provider family; it keeps family-level claims\n * valid and forfeits per-model claims. Maps to\n * `AssertServedModelOptions.allowWithinFamily`.\n */\nexport type ServedModelPolicy = 'exact' | 'allow-within-family'\n\nexport function assertServedModelPolicy(\n value: unknown,\n label: string,\n): asserts value is ServedModelPolicy | undefined {\n if (value !== undefined && value !== 'exact' && value !== 'allow-within-family') {\n throw new Error(`${label} must be 'exact' or 'allow-within-family'`)\n }\n}\n\nexport interface AssertServedModelOptions {\n /**\n * Accept a different model of the same provider family (e.g. requested\n * `deepseek-v3.2`, served `deepseek-v4-flash`). Default false. Setting this\n * keeps family-level claims valid and forfeits per-model claims.\n */\n allowWithinFamily?: boolean\n /**\n * Accept a response that carried no model id. Default false — an\n * unidentified response cannot support a per-model or per-family claim.\n */\n allowUnreported?: boolean\n /** Prefixed to the thrown message, e.g. the judge or campaign cell name. */\n context?: string\n}\n\n/**\n * The one place the accept/reject policy lives, so a caller that reports\n * substitution (a preflight table, a run record) and a caller that throws on it\n * can never drift apart. A cross-family substitution is never acceptable.\n */\nexport function servedModelAcceptable(\n check: ServedModelCheck,\n opts: AssertServedModelOptions = {},\n): boolean {\n switch (check.verdict) {\n case 'exact':\n case 'alias':\n return true\n case 'unreported':\n return opts.allowUnreported === true\n case 'substituted-within-family':\n return opts.allowWithinFamily === true\n default:\n return false\n }\n}\n\nfunction describe(check: ServedModelCheck): string {\n if (check.verdict === 'unreported')\n return `${check.requested}: response carried no model id (identity unproven)`\n return (\n `${check.requested} (${check.requestedFamily}) → served ${check.served} ` +\n `(${check.servedFamily}) [${check.verdict}]`\n )\n}\n\n/**\n * Throw `ModelSubstitutionError` unless the served id is the requested model.\n * Returns the check on success so callers can record the served id alongside\n * the result.\n */\nexport function assertServedModel(\n requested: string,\n served: string | null | undefined,\n opts: AssertServedModelOptions = {},\n): ServedModelCheck {\n const check = checkServedModel(requested, served)\n if (servedModelAcceptable(check, opts)) return check\n const prefix = opts.context ? `${opts.context}: ` : ''\n throw new ModelSubstitutionError(\n `${prefix}model substitution — ${describe(check)}. The measurement is of the SERVED model, ` +\n 'not the requested one; any per-model or per-family claim from this call is invalid.',\n [check],\n )\n}\n\n/**\n * Batch form: check every pair and throw naming EVERY substitution, so one\n * failure does not hide the rest. Returns all checks on success.\n */\nexport function assertServedModels(\n pairs: ReadonlyArray<{ requested: string; served: string | null | undefined }>,\n opts: AssertServedModelOptions = {},\n): ServedModelCheck[] {\n const checks = pairs.map((p) => checkServedModel(p.requested, p.served))\n const bad = checks.filter((c) => !servedModelAcceptable(c, opts))\n if (bad.length > 0) {\n const prefix = opts.context ? `${opts.context}: ` : ''\n throw new ModelSubstitutionError(\n `${prefix}${bad.length}/${checks.length} call(s) were answered by a different model than ` +\n `requested — ${bad.map(describe).join('; ')}. Per-model and per-family claims from this ` +\n 'run are invalid until the ids are re-measured.',\n checks,\n )\n }\n return checks\n}\n\nexport interface AssertCrossFamilyServedOptions extends AssertServedModelOptions {\n /** Minimum distinct SERVED families required. Default 2. */\n minFamilies?: number\n /** Count `unknown`-family served ids toward the total. Default false. */\n allowUnknown?: boolean\n}\n\nexport class ServedCrossFamilyError extends AgentEvalError {\n constructor(\n message: string,\n public readonly families: JudgeFamily[],\n public readonly checks: ReadonlyArray<ServedModelCheck>,\n ) {\n super('model_substitution', message)\n this.name = 'ServedCrossFamilyError'\n }\n}\n\n/**\n * Family-diversity rule over the models that actually ANSWERED.\n *\n * `assertCrossFamily` (../judge-families) reads the requested ids and so\n * cannot see a gateway that answers three \"different\" requests from one\n * provider. This one asserts no substitution first, then counts families from\n * the served ids — a panel that collapsed to one family under the hood fails\n * here even though its request list looked diverse.\n */\nexport function assertCrossFamilyServed(\n pairs: ReadonlyArray<{ requested: string; served: string | null | undefined }>,\n opts: AssertCrossFamilyServedOptions = {},\n): JudgeFamily[] {\n const checks = assertServedModels(pairs, opts)\n const families = new Set<JudgeFamily>()\n for (const check of checks) {\n const family = check.servedFamily\n if (family === null) continue\n if (family === 'unknown' && !opts.allowUnknown) continue\n families.add(family)\n }\n const list = [...families].sort()\n const minFamilies = opts.minFamilies ?? 2\n if (list.length < minFamilies) {\n const prefix = opts.context ? `${opts.context}: ` : ''\n throw new ServedCrossFamilyError(\n `${prefix}the models that ANSWERED span ${list.length} provider famil` +\n `${list.length === 1 ? 'y' : 'ies'} (${list.join(', ') || 'none'}) but ${minFamilies} ` +\n `required — served ids: ${checks.map((c) => c.served ?? 'unreported').join(', ')}`,\n list,\n checks,\n )\n }\n return list\n}\n","/**\n * OpenAI-compatible client. The CLASS is internal; one transport now drives it\n * for public callers.\n *\n * agent-eval never goes looking for a credential — it reads no environment\n * variable to find one — and this module is still not reachable through any\n * export subpath. What reaches a consumer is `createChatClient({ transport:\n * 'openai-compatible', baseUrl, apiKey })`, the sanctioned public route, which\n * constructs this client from an endpoint and a credential the caller passed as\n * values. Anything richer than that transport's options stays in-repo.\n *\n * Three in-repo callers hold the class directly:\n * - `src/analyst/chat-client.ts`, the public transport above;\n * - the `agent-eval` binary (`src/cli-config.ts`), a deployed server whose\n * caller is a JSON-RPC client in another language and which therefore\n * configures its own endpoint from its own environment;\n * - the loopback optimizer proxy path (`src/analyst/benchmark-public-model.ts`,\n * `src/analyst/dspy-rlm-engine.ts`), which targets `http://127.0.0.1:<port>/v1`\n * with an ephemeral token and executes nothing itself.\n *\n * The canonical request/result TYPES it defines ARE public — they are the\n * contract every caller-owned transport speaks.\n *\n * OpenAI-compatible `/v1/chat/completions` client with:\n * - Exponential-backoff retry on 429 + 5xx gateway errors (502/503/504).\n * - Retry on transient network errors (fetch failed, AbortError, ECONNRESET).\n * - One retry at temperature 1 when a model explicitly requires it.\n * - Graceful json_schema → json_object degrade on 400 with schema-reject body.\n * - Fenced-JSON stripping (```json ... ```) for models that wrap structured output.\n * - Configurable base URL + api key / bearer, works with LiteLLM proxies, OpenAI\n * directly, cli-bridge subscriptions, and any router that speaks the spec.\n *\n * Usage:\n * const { value, result } = await callLlmJson<MyType>(\n * { model: 'gpt-4o', messages: [...], jsonSchema: { name: 'x', schema: {...} } },\n * { baseUrl: process.env.AGENT_EVAL_LLM_BASE_URL, apiKey: process.env.AGENT_EVAL_LLM_API_KEY },\n * )\n *\n */\n\nimport { combineAbortSignals } from './abort-signal'\nimport {\n type CostReceiptInput,\n type CustomTokenPricing,\n costForTokenPricing,\n type MaximumCharge,\n} from './cost-ledger'\nimport { AgentEvalError } from './errors'\nimport {\n type AssertServedModelOptions,\n assertServedModel as assertServedModelIdentity,\n} from './integrity/served-model'\nimport { newRecordId } from './record-id'\nimport {\n defaultProviderRedactor,\n type ProviderRedactor,\n providerFromBaseUrl,\n type RawProviderEvent,\n type RawProviderSink,\n} from './trace/raw-provider-sink'\n\n// ─── Types ──────────────────────────────────────────────────────────────\n\nexport interface LlmMessage {\n role: 'system' | 'user' | 'assistant' | 'tool'\n /**\n * Either a plain text content string OR a multimodal content array\n * (text + image_url parts) for vision-capable models.\n */\n content:\n | string\n | Array<\n | { type: 'text'; text: string }\n | { type: 'image_url'; image_url: { url: string; detail?: 'auto' | 'low' | 'high' } }\n >\n /** Tool invocations made by an `assistant` message earlier in the turn. */\n toolCalls?: LlmToolCall[]\n /** The invocation a `tool` message answers. Required when role is `tool`. */\n toolCallId?: string\n}\n\nexport type LlmThinkingMode = 'enabled' | 'disabled'\n\n/** Canonical function-tool definition offered to the model. */\nexport interface LlmToolDefinition {\n type: 'function'\n function: {\n name: string\n description?: string\n parameters: Record<string, unknown>\n }\n}\n\n/** Canonical tool-choice policy; meaningful only when `tools` is present. */\nexport type LlmToolChoice =\n | 'auto'\n | 'none'\n | 'required'\n | { type: 'function'; function: { name: string } }\n\n/** One tool invocation on a response or an assistant history message. */\nexport interface LlmToolCall {\n id: string\n name: string\n /** JSON-encoded arguments, exactly as the provider produced them. */\n argumentsJson: string\n}\n\nexport interface LlmCallRequest {\n model: string\n messages: LlmMessage[]\n /** Optional JSON-mode response format (response_format: json_object). */\n jsonMode?: boolean\n /** Optional structured output via JSON Schema. Falls back to json_object on 400. */\n jsonSchema?: { name: string; schema: Record<string, unknown> }\n /** Function tools offered to the model for this call. */\n tools?: LlmToolDefinition[]\n /** Tool-choice policy for `tools`. */\n toolChoice?: LlmToolChoice\n temperature?: number\n maxTokens?: number\n /** Ask the provider for the log-probability of each sampled token and its\n * `topLogprobs` most likely alternatives. Sends `logprobs: true` with\n * `top_logprobs`. A provider that ignores the field returns\n * `LlmCallResult.logprobs === null`; nothing is inferred from its absence. */\n logprobs?: { topLogprobs: number }\n /** OpenAI-compatible reasoning mode. Omitted when the provider default should apply. */\n thinking?: LlmThinkingMode\n /** Per-call timeout, default 300s. */\n timeoutMs?: number\n}\n\n/** Conservative priced bound for the exact text request sent to a provider.\n * Returns undefined when output or multimodal input is not bounded, causing a\n * capped CostLedger to reject the call before execution. Pass\n * `customTokenPricing` when package pricing does not cover the model or endpoint. */\nexport interface LlmChargeBounds {\n /** Total physical requests, including retries and schema/temperature fallback. Default 3. */\n maximumAttempts?: number\n /** The transport sends JSON mode instead of a response schema. */\n jsonSchemaTransport?: 'native' | 'json-object'\n /** Default provider reasoning mode the transport applies. */\n thinking?: LlmThinkingMode\n /** Token rates used when the provider omits cost or package pricing does not cover the model. */\n customTokenPricing?: CustomTokenPricing\n}\n\nexport function maximumChargeForLlmRequest(\n request: Pick<\n LlmCallRequest,\n 'model' | 'messages' | 'jsonSchema' | 'tools' | 'toolChoice' | 'maxTokens' | 'thinking'\n >,\n options: LlmChargeBounds = {},\n): MaximumCharge | undefined {\n if (request.maxTokens === undefined) return undefined\n if (!Number.isInteger(request.maxTokens) || request.maxTokens <= 0) {\n throw new RangeError(`maximumChargeForLlmRequest: maxTokens must be a positive integer`)\n }\n if (\n request.messages.some(\n (message) =>\n Array.isArray(message.content) && message.content.some((part) => part.type === 'image_url'),\n )\n ) {\n return undefined\n }\n\n const attempts = resolveMaximumAttempts(options.maximumAttempts)\n const forceJsonObject = options.jsonSchemaTransport === 'json-object'\n // A byte-level tokenizer cannot emit more input tokens than request bytes.\n // Pricing the complete body also covers role/schema framing omitted from content-only estimates.\n const requestBytes = new TextEncoder().encode(\n JSON.stringify(buildBody(request, forceJsonObject, options.thinking)),\n ).byteLength\n // Schema fallback shares the same physical-attempt budget; it cannot buy a second batch.\n const usage = {\n inputTokens: requestBytes * attempts,\n outputTokens: request.maxTokens * attempts,\n }\n return options.customTokenPricing\n ? { customTokenPricing: options.customTokenPricing, ...usage }\n : { model: request.model, ...usage }\n}\n\nexport interface LlmUsage {\n promptTokens: number\n completionTokens: number\n totalTokens: number\n /** False when the provider omitted or malformed prompt/completion usage. */\n captured?: boolean\n /** Reasoning-token subset of completionTokens, when reported. */\n reasoningTokens?: number\n /** Proxies populate this when prompt caching is on. */\n cachedPromptTokens?: number\n}\n\n/** One sampled token with its own log probability and the alternatives the\n * provider ranked at that position. */\nexport interface LlmTokenLogprob {\n token: string\n logprob: number\n top: ReadonlyArray<{ token: string; logprob: number }>\n}\n\nexport interface LlmCallResult {\n /** The text content of the first choice. Empty string if none. */\n content: string\n /** Tool invocations from the first choice, when the model called tools. */\n toolCalls?: LlmToolCall[]\n usage: LlmUsage\n /**\n * Cost in USD. Uses the provider's reported cost when present, otherwise\n * caller-supplied token pricing. `null` when neither is available.\n */\n costUsd: number | null\n /**\n * Model id used for attribution (cost, pricing, log lines). The response's\n * echoed id when the provider sent one, else the requested id.\n *\n * NOT evidence of which model answered — read `servedModel` for that. A\n * provider that omits `model` makes this equal to the request, which is\n * exactly the case an identity check must be able to distinguish.\n */\n model: string\n /**\n * The model id the provider echoed on the response, verbatim; `null` when\n * the body carried none. This is the only field that can witness a gateway\n * substituting a different model for the one requested — compare it with\n * `assertServedModel` / `checkServedModel` (src/integrity/served-model.ts).\n *\n * Optional so hand-built results (mock/custom transports) still typecheck,\n * but omitting it is not a pass: the identity checks read `undefined` as\n * `unreported` and reject it by default. A transport that knows which model\n * answered should say so.\n */\n servedModel?: string | null\n /** Wall-clock duration of the HTTP call (last attempt, if retried). */\n durationMs: number\n /**\n * `finish_reason` echoed from the first choice (`stop`, `length`,\n * `content_filter`, `tool_calls`, ...). `null` when the provider omits it.\n * Exposed so a free-form `callLlm` caller CAN detect a truncated answer\n * (`length`) instead of treating a cut-off completion as complete. Note:\n * `callLlm` does not itself reject on it — acting on this signal is the\n * caller's responsibility (in-repo free-form drivers do not yet enforce it).\n */\n finishReason?: string | null\n /**\n * True when `content.trim()` is empty. An empty completion is a silent zero\n * for free-form `callLlm` callers; this flag is the signal a caller can\n * inspect to fail loud rather than proceed on an empty string. `callLlm`\n * surfaces it but does not throw on it.\n */\n contentEmpty?: boolean\n /**\n * Per-token log probabilities for the first choice, in emission order, when\n * the request asked for them and the provider returned\n * `choices[0].logprobs.content`. `null` when the provider returned none, so a\n * caller can tell \"not requested or not supported\" from \"requested and\n * empty\". Absent when the request did not ask for logprobs.\n */\n logprobs?: ReadonlyArray<LlmTokenLogprob> | null\n /** Raw response body. */\n raw: Record<string, unknown>\n}\n\nexport type LlmCallMetadata = Pick<LlmCallResult, 'usage' | 'costUsd' | 'model' | 'durationMs'>\n\n/** Convert a provider result into the canonical paid-call receipt input.\n * The receipt is JSON-clean: absent optional fields are omitted, never carried\n * as explicit-undefined keys. Receipts cross JSON boundaries (the external\n * optimizer's loopback proxy validates them with `assertJsonValue`), where an\n * explicit-undefined value is rejected as non-serializable. */\nexport function costReceiptFromLlm(\n result: LlmCallResult,\n customTokenPricing?: CustomTokenPricing,\n): CostReceiptInput {\n const cachedTokens = result.usage.cachedPromptTokens ?? 0\n const inputTokens = Math.max(0, result.usage.promptTokens - cachedTokens)\n const providerCostUsd = providerReportedCost(result.raw)\n return {\n model: result.model,\n inputTokens,\n outputTokens: result.usage.completionTokens,\n ...(result.usage.reasoningTokens === undefined\n ? {}\n : { reasoningTokens: result.usage.reasoningTokens }),\n ...(cachedTokens > 0 ? { cachedTokens } : {}),\n ...(providerCostUsd === undefined\n ? customTokenPricing && result.usage.captured !== false\n ? { customTokenPricing }\n : result.costUsd === null\n ? {}\n : { estimatedCostUsd: result.costUsd }\n : { actualCostUsd: providerCostUsd }),\n usageUnknown: result.usage.captured === false,\n }\n}\n\nfunction providerReportedCost(raw: Record<string, unknown>): number | undefined {\n const value = raw._response_cost ?? raw.cost_usd\n return typeof value === 'number' && Number.isFinite(value) && value >= 0 ? value : undefined\n}\n\n/** Structured-response failures retain their completed provider receipt. */\nexport function costReceiptFromLlmError(\n error: Error,\n customTokenPricing?: CustomTokenPricing,\n): CostReceiptInput | undefined {\n return error instanceof LlmResponseError\n ? costReceiptFromLlm(error.result, customTokenPricing)\n : undefined\n}\n\nexport class LlmCallError extends AgentEvalError {\n constructor(\n message: string,\n public readonly status: number,\n public readonly body: string,\n public readonly model: string,\n ) {\n super('judge', message)\n }\n}\n\n/** A provider response completed and incurred measurable usage, but its content\n * could not satisfy the caller's response contract. The response envelope is\n * retained so accounting can commit the receipt before the error propagates. */\nexport class LlmResponseError extends AgentEvalError {\n constructor(\n message: string,\n public readonly result: LlmCallResult,\n options?: { cause?: unknown },\n ) {\n super('judge', message, options)\n }\n}\n\nexport interface LlmClientOptions extends LlmChargeBounds {\n /** Base URL (without trailing slash), ending at the `/v1` prefix. Required: there is no default endpoint. */\n baseUrl?: string\n /** Bearer token — either `apiKey` or `bearer` populates `Authorization: Bearer ...`. */\n apiKey?: string\n bearer?: string\n /** Override for the `Authorization` header (e.g. `X-Auth: ...`). Takes precedence over apiKey/bearer. */\n authHeader?: { name: string; value: string }\n /** Stable provider idempotency key, reused across retries of this logical call. */\n idempotencyKey?: string\n /** Default timeout in ms. Per-call can override. */\n defaultTimeoutMs?: number\n /**\n * Caller-supplied abort signal — e.g. a campaign-wide cancel. Linked to\n * each attempt's per-attempt timeout controller, so aborting it cancels\n * the in-flight fetch. A caller abort is FATAL: it is not retried even\n * though an AbortError otherwise matches the transient patterns.\n */\n signal?: AbortSignal\n /**\n * Cross-attempt wall-clock budget in ms, measured from the first attempt.\n * Before launching each attempt the loop checks the remaining budget and\n * stops retrying once it is exhausted, rather than waiting the full\n * per-attempt timeout on every retry. Bounds total time independent of\n * total attempts × `timeoutMs`.\n */\n deadlineMs?: number\n /**\n * JSON payload parsing policy. `extract` accepts fenced or prose-prefixed JSON.\n * `exact` requires the complete response content to be one JSON value.\n * Default: `extract`.\n */\n jsonPayloadMode?: 'extract' | 'exact'\n /** Fetch implementation — defaults to global `fetch`. Override for custom transport (e.g. tests). */\n fetch?: typeof fetch\n /**\n * Optional raw HTTP capture sink. When provided, every request, response,\n * and error (across all retry attempts) is recorded to the sink, with auth\n * headers and credential-shaped body fields redacted by default. This is\n * the layer-1 forensics primitive: structured `LlmSpan`s record intent,\n * raw events record what actually crossed the wire.\n */\n rawSink?: RawProviderSink\n /**\n * Logical provider id attached to raw events. When omitted, derived from\n * `baseUrl` via `providerFromBaseUrl`.\n */\n provider?: string\n /** Trace context attached to raw events; populated by emitter-aware callers. */\n traceContext?: { runId?: string; spanId?: string }\n /** Override the redaction strategy for this call. Defaults to `defaultProviderRedactor`. */\n redactor?: ProviderRedactor\n /**\n * Reject a response whose echoed model is not the model that was requested.\n * A routing gateway can accept one id and answer from another, which\n * silently invalidates every per-model and per-family claim downstream.\n * `true` uses the strict default (aliases pass, substitutions and\n * unidentified responses throw `ModelSubstitutionError`); pass an options\n * object to relax a specific case. Off by default — turning it on for a\n * measurement run is the point.\n */\n assertServedModel?: boolean | AssertServedModelOptions\n}\n\n// ─── Internals ──────────────────────────────────────────────────────────\n\n// Flagship / reasoning models routinely take several minutes on large prompts (a\n// reflection over many failures, a long tool transcript). A tight cap aborts a\n// legitimately-slow but healthy call — and because every retry attempt re-uses\n// the same window, such a model aborts on ALL attempts and the loop throws. The\n// default is generous enough to let those complete, bounded enough that a truly\n// hung call still fails over after retries, and tunable per deployment via\n// TANGLE_LLM_TIMEOUT_MS. Per-call `req.timeoutMs` / `opts.defaultTimeoutMs`\n// still win for callers that know their model's latency.\nconst DEFAULT_TIMEOUT_MS = Number(process.env.TANGLE_LLM_TIMEOUT_MS) || 300_000\nconst DEFAULT_MAXIMUM_ATTEMPTS =\n process.env.TANGLE_LLM_MAXIMUM_ATTEMPTS === undefined\n ? 3\n : Number(process.env.TANGLE_LLM_MAXIMUM_ATTEMPTS)\n\nfunction resolveMaximumAttempts(configured: number | undefined): number {\n const attempts = configured ?? DEFAULT_MAXIMUM_ATTEMPTS\n if (!Number.isInteger(attempts) || attempts <= 0) {\n throw new RangeError('LLM maximum attempts must be a positive integer')\n }\n return attempts\n}\n\nfunction providerTokenCount(value: unknown): number | undefined {\n return typeof value === 'number' && Number.isSafeInteger(value) && value >= 0 ? value : undefined\n}\n\n// 522/524: Cloudflare edge timeouts (origin connect/respond) — transient\n// exactly like 504. Router deployments serve through Cloudflare, so a long\n// non-streaming call can draw these from the edge, not the origin.\nconst RETRYABLE_STATUS = new Set([429, 502, 503, 504, 522, 524])\n\n/**\n * Transient transport/network error signatures, matched against an error's\n * name, message, and `code`. Covers fetch/undici network failures, aborts\n * and timeouts, and — critically — HTTP/2 transport faults a keep-alive\n * connection raises mid-response: `terminated`, `NGHTTP2_INTERNAL_ERROR`,\n * `UND_ERR_*`, `other side closed`. Those last ones carry no clean HTTP\n * status; unrecognised, they escape the retry loop and surface as an\n * uncaught rejection.\n */\nconst TRANSIENT_ERROR_PATTERNS: readonly RegExp[] = [\n /AbortError/i,\n /TimeoutError/i,\n /this operation was aborted/i,\n /fetch failed/i,\n /ECONNRESET/i,\n /ETIMEDOUT/i,\n /EAI_AGAIN/i,\n /socket hang up/i,\n /stream.*ended.*unexpectedly/i,\n /terminated/i,\n /other side closed/i,\n /NGHTTP2/i,\n /UND_ERR/i,\n]\n\n/**\n * True when an error is a transient transport/network fault worth retrying,\n * as opposed to a deterministic failure (4xx schema reject, JSON parse) that\n * a retry cannot fix. Inspects `LlmCallError.status`, then the error's\n * name/message/code, then recurses into `error.cause` — undici nests the\n * real socket fault one or more levels under `.cause`.\n *\n * This is the retry classifier for the package: `callLlm` and\n * `withJudgeRetry` both route through it, so connection failures are treated\n * consistently across transports.\n */\nexport function isTransientLlmError(err: unknown): boolean {\n return classifyTransient(err, 0)\n}\n\nfunction classifyTransient(err: unknown, depth: number): boolean {\n if (err instanceof LlmCallError) return RETRYABLE_STATUS.has(err.status)\n if (!(err instanceof Error)) return false\n // Foreign transport errors can carry a numeric HTTP status without being an\n // LlmCallError. A retryable status is decisive.\n const status = (err as { status?: unknown }).status\n if (typeof status === 'number' && RETRYABLE_STATUS.has(status)) return true\n const code = (err as { code?: unknown }).code\n const haystack = `${err.name}\\n${err.message}\\n${typeof code === 'string' ? code : ''}`\n if (TRANSIENT_ERROR_PATTERNS.some((p) => p.test(haystack))) return true\n const cause = (err as { cause?: unknown }).cause\n if (depth < 4 && cause instanceof Error && cause !== err) {\n return classifyTransient(cause, depth + 1)\n }\n return false\n}\n\nfunction parseRetryAfter(headers: Headers): number | null {\n const h = headers.get('retry-after')\n if (!h) return null\n const asNumber = Number(h)\n if (Number.isFinite(asNumber) && asNumber > 0) return asNumber * 1000\n const asDate = Date.parse(h)\n if (Number.isFinite(asDate)) return Math.max(0, asDate - Date.now())\n return null\n}\n\n/** Exponential backoff: 500ms, 1s, 2s, 4s, ... capped at 16s. Attempt is 0-indexed. */\nexport function backoffMs(attempt: number): number {\n return Math.min(500 * 2 ** attempt, 16_000)\n}\n\nfunction buildHeaders(opts: LlmClientOptions): Record<string, string> {\n const headers: Record<string, string> = {\n 'Content-Type': 'application/json',\n Accept: 'application/json',\n }\n if (opts.authHeader) {\n headers[opts.authHeader.name] = opts.authHeader.value\n } else if (opts.bearer || opts.apiKey) {\n headers.Authorization = `Bearer ${opts.bearer ?? opts.apiKey}`\n }\n if (opts.idempotencyKey) headers['Idempotency-Key'] = opts.idempotencyKey\n return headers\n}\n\nfunction isSchemaRejection(status: number, body: string): boolean {\n if (status !== 400) return false\n const lower = body.toLowerCase()\n return (\n lower.includes('response_format') ||\n lower.includes('json_schema') ||\n lower.includes('is unavailable') ||\n lower.includes('not supported')\n )\n}\n\nfunction isTemperatureOneRejection(status: number, body: string): boolean {\n if (status !== 400 || !/temperature/i.test(body)) return false\n return (\n /temperature[^.\\n]{0,120}\\b(?:only|must|should|required|requires?)\\b[^.\\n]{0,40}\\b1(?:\\.0+)?\\b/i.test(\n body,\n ) || /\\bonly\\s+1(?:\\.0+)?\\s+is\\s+allowed\\b[^.\\n]{0,120}\\btemperature\\b/i.test(body)\n )\n}\n\nfunction buildBody(\n req: LlmCallRequest,\n forceJsonObject: boolean,\n defaultThinking?: LlmThinkingMode,\n): Record<string, unknown> {\n const body: Record<string, unknown> = {\n model: req.model,\n messages: req.messages.map(encodeWireMessage),\n temperature: req.temperature ?? 0,\n }\n if (req.maxTokens != null) {\n if (usesMaxCompletionTokens(req.model)) body.max_completion_tokens = req.maxTokens\n else body.max_tokens = req.maxTokens\n }\n if (req.logprobs !== undefined) {\n body.logprobs = true\n body.top_logprobs = req.logprobs.topLogprobs\n }\n if (req.tools !== undefined) body.tools = req.tools\n if (req.toolChoice !== undefined) body.tool_choice = req.toolChoice\n const thinking = req.thinking ?? defaultThinking\n if (thinking !== undefined) {\n body.thinking = { type: thinking }\n }\n\n if (req.jsonSchema && !forceJsonObject) {\n body.response_format = {\n type: 'json_schema',\n json_schema: { name: req.jsonSchema.name, schema: req.jsonSchema.schema, strict: true },\n }\n } else if (req.jsonMode || req.jsonSchema) {\n body.response_format = { type: 'json_object' }\n }\n\n return body\n}\n\nfunction usesMaxCompletionTokens(model: string): boolean {\n return /^gpt-5(?:[.-]|$)/i.test(model)\n}\n\n/** Encode one canonical message as its OpenAI chat-completions wire shape. */\nfunction encodeWireMessage(message: LlmMessage): Record<string, unknown> {\n if (message.role === 'tool') {\n return { role: 'tool', tool_call_id: message.toolCallId, content: message.content }\n }\n if (message.toolCalls === undefined) {\n return { role: message.role, content: message.content }\n }\n return {\n role: message.role,\n content: message.content,\n tool_calls: message.toolCalls.map((call) => ({\n id: call.id,\n type: 'function',\n function: { name: call.name, arguments: call.argumentsJson },\n })),\n }\n}\n\n/**\n * Parse `choices[0].logprobs.content` into the canonical per-token shape.\n * `null` when the provider returned nothing for a request that asked: a\n * provider that ignores `logprobs` is a fact the caller must be able to read,\n * not an error. A malformed entry is a contract violation and throws.\n */\nfunction parseWireLogprobs(value: unknown, model: string): ReadonlyArray<LlmTokenLogprob> | null {\n if (value === undefined || value === null) return null\n if (!Array.isArray(value)) {\n throw new Error(`LLM response logprobs.content must be an array (model=${model})`)\n }\n return value.map((entry, index) => {\n const record = entry as { token?: unknown; logprob?: unknown; top_logprobs?: unknown } | null\n if (\n !record ||\n typeof record !== 'object' ||\n typeof record.token !== 'string' ||\n typeof record.logprob !== 'number' ||\n !Number.isFinite(record.logprob)\n ) {\n throw new Error(\n `LLM response logprobs.content[${index}] is not a token with a finite logprob (model=${model})`,\n )\n }\n const top = record.top_logprobs\n if (top !== undefined && top !== null && !Array.isArray(top)) {\n throw new Error(\n `LLM response logprobs.content[${index}].top_logprobs must be an array (model=${model})`,\n )\n }\n return {\n token: record.token,\n logprob: record.logprob,\n top: (Array.isArray(top) ? top : []).map((alternative, position) => {\n const candidate = alternative as { token?: unknown; logprob?: unknown } | null\n if (\n !candidate ||\n typeof candidate !== 'object' ||\n typeof candidate.token !== 'string' ||\n typeof candidate.logprob !== 'number' ||\n !Number.isFinite(candidate.logprob)\n ) {\n throw new Error(\n `LLM response logprobs.content[${index}].top_logprobs[${position}] is not a token with a finite logprob (model=${model})`,\n )\n }\n return { token: candidate.token, logprob: candidate.logprob }\n }),\n }\n })\n}\n\n/** Parse provider tool_calls into the canonical shape. A malformed entry is a\n * provider-contract violation and throws rather than degrading silently. */\nfunction parseWireToolCalls(value: unknown, model: string): LlmToolCall[] | undefined {\n if (value === undefined || value === null) return undefined\n if (!Array.isArray(value)) {\n throw new Error(`LLM response tool_calls must be an array (model=${model})`)\n }\n if (value.length === 0) return undefined\n return value.map((entry, index) => {\n const record = entry as {\n id?: unknown\n type?: unknown\n function?: { name?: unknown; arguments?: unknown } | null\n } | null\n const fn = record && typeof record === 'object' ? record.function : undefined\n if (\n !record ||\n typeof record !== 'object' ||\n typeof record.id !== 'string' ||\n record.id.length === 0 ||\n (record.type !== undefined && record.type !== 'function') ||\n !fn ||\n typeof fn !== 'object' ||\n typeof fn.name !== 'string' ||\n fn.name.length === 0 ||\n typeof fn.arguments !== 'string'\n ) {\n throw new Error(\n `LLM response tool_calls[${index}] is not a function call with string arguments (model=${model})`,\n )\n }\n return { id: record.id, name: fn.name, argumentsJson: fn.arguments }\n })\n}\n\n/** Backoff belongs to the caller's operation and must stop when it is cancelled. */\nasync function sleep(ms: number, signal?: AbortSignal): Promise<void> {\n signal?.throwIfAborted()\n return new Promise((resolve, reject) => {\n const onAbort = () => {\n clearTimeout(timer)\n signal?.removeEventListener('abort', onAbort)\n reject(signal?.reason)\n }\n const timer = setTimeout(() => {\n signal?.removeEventListener('abort', onAbort)\n resolve()\n }, ms)\n signal?.addEventListener('abort', onAbort, { once: true })\n })\n}\n\n/** True once the cross-attempt wall-clock budget (if any) is exhausted. */\nfunction deadlineExceeded(start: number, deadlineMs: number | undefined): boolean {\n return deadlineMs != null && Date.now() - start >= deadlineMs\n}\n\n// ─── Public API ─────────────────────────────────────────────────────────\n\n/**\n * Strip a ```json / ``` code fence if the model emitted one.\n * Idempotent for naked JSON. Some models (claude-code via router, certain\n * deepseek models) wrap output even under json_object.\n */\nexport function stripFencedJson(raw: string): string {\n const trimmed = raw.trim()\n const m = trimmed.match(/^```(?:json)?\\s*\\n?([\\s\\S]*?)\\n?```\\s*$/)\n return m ? m[1]!.trim() : trimmed\n}\n\nexport function extractJsonPayload(raw: string): string {\n const stripped = stripFencedJson(raw)\n try {\n JSON.parse(stripped)\n return stripped\n } catch {\n // A response that declares a JSON root must parse as that complete root.\n // Scanning onward could turn a truncated object into one of its valid nested\n // arrays or objects and silently change the response schema.\n if (stripped.startsWith('{') || stripped.startsWith('[')) return stripped\n }\n\n // Only prose-leading responses may contain a recoverable JSON payload.\n const starts = [...stripped.matchAll(/[[{]/g)]\n .map((match) => match.index)\n .filter((index) => index != null)\n for (const start of starts) {\n const candidate = extractBalancedJson(stripped, start)\n if (!candidate) continue\n try {\n JSON.parse(candidate)\n return candidate\n } catch {\n // Keep scanning; earlier braces may belong to prose.\n }\n }\n\n return stripped\n}\n\nfunction extractBalancedJson(input: string, start: number): string | null {\n const opener = input[start]\n const closer = opener === '{' ? '}' : opener === '[' ? ']' : null\n if (!closer) return null\n\n const stack: string[] = [closer]\n let isInString = false\n let isEscaped = false\n\n for (let i = start + 1; i < input.length; i++) {\n const char = input[i]!\n if (isEscaped) {\n isEscaped = false\n continue\n }\n if (char === '\\\\') {\n isEscaped = isInString\n continue\n }\n if (char === '\"') {\n isInString = !isInString\n continue\n }\n if (isInString) continue\n\n if (char === '{') stack.push('}')\n else if (char === '[') stack.push(']')\n else if (char === stack[stack.length - 1]) {\n stack.pop()\n if (stack.length === 0) return input.slice(start, i + 1)\n }\n }\n\n return null\n}\n\n/**\n * Low-level call. Returns raw content + usage + cost. Retries on transient\n * failures; does NOT degrade schema here — callers that want graceful\n * degrade use `callLlmJson`.\n */\nexport async function callLlm(\n req: LlmCallRequest,\n opts: LlmClientOptions = {},\n): Promise<LlmCallResult> {\n return callLlmAttempts(req, opts, false)\n}\n\n/** One retry owner for raw and structured calls. Fallback changes the next\n * request, not the attempt allowance, deadline origin, or raw-event sequence. */\nasync function callLlmAttempts(\n req: LlmCallRequest,\n opts: LlmClientOptions,\n allowSchemaFallback: boolean,\n): Promise<LlmCallResult> {\n // No default endpoint. This client is internal to the `agent-eval` binary\n // and the loopback optimizer proxy; both name their endpoint explicitly, and\n // a fallback would let a misconfigured caller bill an unintended provider.\n if (!opts.baseUrl) throw new Error('callLlm: opts.baseUrl is required')\n const baseUrl = opts.baseUrl.replace(/\\/+$/, '')\n const url = `${baseUrl}/chat/completions`\n const endpoint = '/chat/completions'\n const timeoutMs = req.timeoutMs ?? opts.defaultTimeoutMs ?? DEFAULT_TIMEOUT_MS\n const maximumAttempts = resolveMaximumAttempts(opts.maximumAttempts)\n const fetchFn = opts.fetch ?? globalThis.fetch\n const headers = buildHeaders(opts)\n const provider = opts.provider ?? providerFromBaseUrl(baseUrl)\n const sink = opts.rawSink\n const redactor = opts.redactor ?? defaultProviderRedactor\n const traceContext = opts.traceContext\n const callerSignal = opts.signal\n const deadlineMs = opts.deadlineMs\n const deadlineStart = Date.now()\n if (opts.customTokenPricing) {\n costForTokenPricing(opts.customTokenPricing, { inputTokens: 0, outputTokens: 0 })\n }\n\n let lastErr: unknown\n let effectiveRequest = req\n const remainingMs = () =>\n deadlineMs === undefined ? Infinity : Math.max(0, deadlineMs - (Date.now() - deadlineStart))\n const retryDelay = (ms: number) => sleep(Math.min(ms, remainingMs()), callerSignal)\n for (let attempt = 0; attempt < maximumAttempts; attempt++) {\n // A caller cancel is fatal — never retried. Checking before each attempt\n // means an already-aborted signal short-circuits without firing fetch.\n if (callerSignal?.aborted) {\n throw new DOMException('callLlm aborted by caller signal', 'AbortError')\n }\n // Stop retrying once the cross-attempt budget is spent rather than burning\n // a full per-attempt timeout on each remaining retry.\n if (deadlineExceeded(deadlineStart, deadlineMs)) {\n throw lastErr instanceof Error\n ? lastErr\n : new DOMException('callLlm deadline exceeded', 'TimeoutError')\n }\n const requestBody = buildBody(\n effectiveRequest,\n opts.jsonSchemaTransport === 'json-object',\n opts.thinking,\n )\n const controller = new AbortController()\n const attemptSignal = combineAbortSignals(controller.signal, callerSignal)!\n const timeoutHandle = setTimeout(() => controller.abort(), Math.min(timeoutMs, remainingMs()))\n const started = Date.now()\n let attemptErrorRecorded = false\n try {\n if (sink) {\n await recordRaw(sink, redactor, {\n eventId: newRecordId(),\n runId: traceContext?.runId,\n spanId: traceContext?.spanId,\n provider,\n model: req.model,\n endpoint,\n baseUrl,\n attemptIndex: attempt,\n direction: 'request',\n timestamp: started,\n requestHeaders: headers,\n requestBody,\n redactedFields: [],\n })\n }\n attemptSignal.throwIfAborted()\n const res = await fetchFn(url, {\n method: 'POST',\n headers,\n body: JSON.stringify(requestBody),\n signal: attemptSignal,\n })\n const responseHeaders = sink ? headersToObject(res.headers) : undefined\n // Headers are not completion: the same timeout/cancellation covers either response body.\n const text = await res.text()\n attemptSignal.throwIfAborted()\n clearTimeout(timeoutHandle)\n\n if (!res.ok) {\n const body = text\n if (sink) {\n await recordRaw(sink, redactor, {\n eventId: newRecordId(),\n runId: traceContext?.runId,\n spanId: traceContext?.spanId,\n provider,\n model: req.model,\n endpoint,\n baseUrl,\n attemptIndex: attempt,\n direction: 'error',\n timestamp: Date.now(),\n durationMs: Date.now() - started,\n statusCode: res.status,\n responseHeaders,\n responseBody: body,\n errorMessage: `HTTP ${res.status}`,\n redactedFields: [],\n })\n attemptErrorRecorded = true\n }\n const err = new LlmCallError(\n `LLM call failed with HTTP ${res.status}`,\n res.status,\n body,\n req.model,\n )\n if (\n isTemperatureOneRejection(res.status, body) &&\n effectiveRequest.temperature !== 1 &&\n attempt < maximumAttempts - 1 &&\n !deadlineExceeded(deadlineStart, deadlineMs)\n ) {\n lastErr = err\n effectiveRequest = { ...effectiveRequest, temperature: 1 }\n continue\n }\n if (\n allowSchemaFallback &&\n opts.jsonSchemaTransport !== 'json-object' &&\n effectiveRequest.jsonSchema &&\n isSchemaRejection(res.status, body) &&\n attempt < maximumAttempts - 1 &&\n !deadlineExceeded(deadlineStart, deadlineMs)\n ) {\n lastErr = err\n effectiveRequest = { ...effectiveRequest, jsonMode: true, jsonSchema: undefined }\n continue\n }\n if (\n RETRYABLE_STATUS.has(res.status) &&\n attempt < maximumAttempts - 1 &&\n !deadlineExceeded(deadlineStart, deadlineMs)\n ) {\n lastErr = err\n const retryAfter = parseRetryAfter(res.headers)\n await retryDelay(retryAfter ?? backoffMs(attempt))\n continue\n }\n throw err\n }\n\n let json: Record<string, unknown>\n try {\n json = JSON.parse(text) as Record<string, unknown>\n } catch (parseErr) {\n if (sink) {\n await recordRaw(sink, redactor, {\n eventId: newRecordId(),\n runId: traceContext?.runId,\n spanId: traceContext?.spanId,\n provider,\n model: req.model,\n endpoint,\n baseUrl,\n attemptIndex: attempt,\n direction: 'error',\n timestamp: Date.now(),\n durationMs: Date.now() - started,\n statusCode: res.status,\n responseHeaders,\n responseBody: text,\n errorMessage: `non-JSON response: ${parseErr instanceof Error ? parseErr.message : String(parseErr)}`,\n redactedFields: [],\n })\n attemptErrorRecorded = true\n }\n throw parseErr\n }\n if (sink) {\n await recordRaw(sink, redactor, {\n eventId: newRecordId(),\n runId: traceContext?.runId,\n spanId: traceContext?.spanId,\n provider,\n model: req.model,\n endpoint,\n baseUrl,\n attemptIndex: attempt,\n direction: 'response',\n timestamp: Date.now(),\n durationMs: Date.now() - started,\n statusCode: res.status,\n responseHeaders,\n responseBody: json,\n redactedFields: [],\n })\n }\n const choice = (\n json.choices as\n | Array<{\n message?: { content?: string | null; tool_calls?: unknown }\n finish_reason?: string | null\n logprobs?: { content?: unknown } | null\n }>\n | undefined\n )?.[0]\n const toolCalls = parseWireToolCalls(choice?.message?.tool_calls, req.model)\n const usageRaw =\n json.usage && typeof json.usage === 'object' && !Array.isArray(json.usage)\n ? (json.usage as Record<string, unknown>)\n : undefined\n const promptTokens = providerTokenCount(usageRaw?.prompt_tokens)\n const completionTokens = providerTokenCount(usageRaw?.completion_tokens)\n const totalTokens = providerTokenCount(usageRaw?.total_tokens)\n const completionDetails =\n usageRaw?.completion_tokens_details &&\n typeof usageRaw.completion_tokens_details === 'object' &&\n !Array.isArray(usageRaw.completion_tokens_details)\n ? (usageRaw.completion_tokens_details as Record<string, unknown>)\n : undefined\n const reasoningRaw = completionDetails?.reasoning_tokens\n const reasoningTokens =\n reasoningRaw === undefined ? undefined : providerTokenCount(reasoningRaw)\n const cachedRaw =\n usageRaw?.prompt_tokens_details &&\n typeof usageRaw.prompt_tokens_details === 'object' &&\n !Array.isArray(usageRaw.prompt_tokens_details)\n ? (usageRaw.prompt_tokens_details as Record<string, unknown>).cached_tokens\n : undefined\n const cachedPromptTokens = cachedRaw === undefined ? undefined : providerTokenCount(cachedRaw)\n const usageCaptured =\n promptTokens !== undefined &&\n completionTokens !== undefined &&\n (reasoningRaw === undefined ||\n (reasoningTokens !== undefined && reasoningTokens <= completionTokens)) &&\n (cachedRaw === undefined ||\n (cachedPromptTokens !== undefined && cachedPromptTokens <= promptTokens)) &&\n (totalTokens === undefined || totalTokens === promptTokens + completionTokens)\n const costFromProxy = providerReportedCost(json)\n const content = choice?.message?.content ?? ''\n\n const configuredCost =\n typeof costFromProxy !== 'number' && usageCaptured && opts.customTokenPricing\n ? costForTokenPricing(opts.customTokenPricing, {\n inputTokens: promptTokens! - (cachedPromptTokens ?? 0),\n ...(cachedPromptTokens ? { cachedTokens: cachedPromptTokens } : {}),\n outputTokens: completionTokens!,\n })\n : undefined\n\n // The echoed id, kept separate from the attribution id: a provider that\n // omits it must read as \"unproven\", never as \"the model I asked for\".\n const servedModel =\n typeof json.model === 'string' && json.model.trim() !== '' ? json.model : null\n if (opts.assertServedModel) {\n assertServedModelIdentity(\n req.model,\n servedModel,\n opts.assertServedModel === true ? {} : opts.assertServedModel,\n )\n }\n\n return {\n content,\n ...(toolCalls === undefined ? {} : { toolCalls }),\n ...(req.logprobs === undefined\n ? {}\n : { logprobs: parseWireLogprobs(choice?.logprobs?.content, req.model) }),\n finishReason: choice?.finish_reason ?? null,\n contentEmpty: content.trim().length === 0,\n usage: {\n promptTokens: promptTokens ?? 0,\n completionTokens: completionTokens ?? 0,\n totalTokens: totalTokens ?? (promptTokens ?? 0) + (completionTokens ?? 0),\n captured: usageCaptured,\n reasoningTokens,\n cachedPromptTokens,\n },\n costUsd: typeof costFromProxy === 'number' ? costFromProxy : (configuredCost ?? null),\n model: servedModel ?? req.model,\n servedModel,\n durationMs: Date.now() - started,\n raw: json,\n }\n } catch (err) {\n clearTimeout(timeoutHandle)\n lastErr = err\n // A caller cancel is fatal even though an AbortError matches the\n // transient patterns — a cancelled call must surface immediately, not\n // be retried against the same dead intent.\n if (callerSignal?.aborted) {\n if (sink && !attemptErrorRecorded) {\n await recordRaw(sink, redactor, {\n eventId: newRecordId(),\n runId: traceContext?.runId,\n spanId: traceContext?.spanId,\n provider,\n model: req.model,\n endpoint,\n baseUrl,\n attemptIndex: attempt,\n direction: 'error',\n timestamp: Date.now(),\n durationMs: Date.now() - started,\n errorMessage: err instanceof Error ? err.message : String(err),\n redactedFields: [],\n })\n }\n throw err\n }\n if (sink && !attemptErrorRecorded) {\n // Record only if neither the !res.ok branch nor the JSON.parse catch\n // already produced an error event for this attempt. Covers network\n // failures, timeouts, and aborts.\n await recordRaw(sink, redactor, {\n eventId: newRecordId(),\n runId: traceContext?.runId,\n spanId: traceContext?.spanId,\n provider,\n model: req.model,\n endpoint,\n baseUrl,\n attemptIndex: attempt,\n direction: 'error',\n timestamp: Date.now(),\n durationMs: Date.now() - started,\n errorMessage: err instanceof Error ? err.message : String(err),\n redactedFields: [],\n })\n }\n if (\n attempt < maximumAttempts - 1 &&\n isTransientLlmError(err) &&\n !deadlineExceeded(deadlineStart, deadlineMs)\n ) {\n await retryDelay(backoffMs(attempt))\n continue\n }\n throw err\n } finally {\n clearTimeout(timeoutHandle)\n }\n }\n throw lastErr instanceof Error ? lastErr : new Error(String(lastErr))\n}\n\nasync function recordRaw(\n sink: RawProviderSink,\n redactor: ProviderRedactor,\n event: RawProviderEvent,\n): Promise<void> {\n // Errors from sinks must not crash the LLM call. Forensic capture is\n // best-effort; the structured trace is the system of record.\n try {\n await sink.record(redactor(event))\n } catch {\n // Intentionally swallowed.\n }\n}\n\nfunction headersToObject(h: Headers): Record<string, string> {\n const out: Record<string, string> = {}\n h.forEach((value, key) => {\n out[key] = value\n })\n return out\n}\n\n/**\n * Structured-output call. Returns parsed JSON plus the raw result envelope.\n * Degrades `jsonSchema` → `jsonMode` on a 400 that names the schema param —\n * critical for deepseek-v3/v4, kimi-k2.6, and other models that don't accept\n * the `response_format.json_schema` shape but DO accept `json_object`.\n */\nexport async function callLlmJson<T = unknown>(\n req: LlmCallRequest,\n opts: LlmClientOptions = {},\n): Promise<{ value: T; result: LlmCallResult }> {\n const result = await callLlmStructured(req, opts)\n const value = parseJsonResult<T>(result, opts.jsonPayloadMode ?? 'extract')\n return { value, result }\n}\n\n/** Shared schema-to-JSON-mode fallback that preserves the raw result. */\nasync function callLlmStructured(\n req: LlmCallRequest,\n opts: LlmClientOptions = {},\n): Promise<LlmCallResult> {\n return callLlmAttempts({ ...req, jsonMode: req.jsonMode ?? !req.jsonSchema }, opts, true)\n}\n\nfunction parseJsonResult<T>(\n result: LlmCallResult,\n jsonPayloadMode: NonNullable<LlmClientOptions['jsonPayloadMode']>,\n): T {\n try {\n if (result.finishReason === 'length') {\n throw new Error(\n `LLM returned truncated JSON content (model=${result.model}, finishReason=length)`,\n )\n }\n return parseJsonSafely<T>(result.content, result.model, jsonPayloadMode)\n } catch (error) {\n if (error instanceof LlmResponseError) throw error\n const cause = error instanceof Error ? error : new Error(String(error))\n throw new LlmResponseError(cause.message, result, { cause })\n }\n}\n\nfunction parseJsonSafely<T>(\n content: string,\n model: string,\n jsonPayloadMode: NonNullable<LlmClientOptions['jsonPayloadMode']>,\n): T {\n const payload = jsonPayloadMode === 'exact' ? content : extractJsonPayload(content)\n try {\n return JSON.parse(payload) as T\n } catch {\n throw new Error(`LLM returned non-JSON content (model=${model})`)\n }\n}\n\n/**\n * Stateful client — construct once with defaults, call many times.\n * Thin wrapper around the free functions; exists for callers that want\n * to inject a single configured instance into multiple primitives.\n */\nexport class LlmClient {\n readonly maximumAttempts: number\n private readonly opts: LlmClientOptions\n\n constructor(opts: LlmClientOptions = {}) {\n this.opts = opts\n this.maximumAttempts = resolveMaximumAttempts(opts.maximumAttempts)\n }\n\n call(req: LlmCallRequest, per?: LlmClientOptions): Promise<LlmCallResult> {\n const options = { ...this.opts, ...per }\n return req.jsonSchema ? callLlmStructured(req, options) : callLlm(req, options)\n }\n\n callJson<T = unknown>(\n req: LlmCallRequest,\n per?: LlmClientOptions,\n ): Promise<{ value: T; result: LlmCallResult }> {\n return callLlmJson<T>(req, { ...this.opts, ...per })\n }\n}\n"],"mappings":";;;;;;AACA,SAAgB,oBACd,GAAG,SACsB;CACzB,MAAM,SAAS,CACb,GAAG,IAAI,IAAI,QAAQ,QAAQ,WAAkC,WAAW,KAAA,CAAS,CAAC,CACpF;CACA,IAAI,OAAO,WAAW,GAAG,OAAO,KAAA;CAChC,IAAI,OAAO,WAAW,GAAG,OAAO,OAAO;CACvC,OAAO,YAAY,IAAI,MAAM;AAC/B;;;;ACiBA,MAAM,kBAA+C;CACnD,WAAW;CACX,QAAQ;CACR,gBAAgB;CAChB,QAAQ;CACR,iBAAiB;CACjB,MAAM;CACN,cAAc;CACd,SAAS;CACT,WAAW;CACX,UAAU;CACV,KAAK;CACL,MAAM;CACN,SAAS;CACT,QAAQ;CACR,QAAQ;CACR,SAAS;CACT,UAAU;CACV,YAAY;CACZ,MAAM;CACN,aAAa;CACb,OAAO;CACP,SAAS;CACT,KAAK;CACL,QAAQ;CACR,KAAK;AACP;;AAGA,MAAM,gBAA8C;CAClD,CAAC,WAAW,WAAW;CAIvB,CAAC,uCAAuC,WAAW;CACnD,CAAC,0DAA0D,QAAQ;CACnE,CAAC,4BAA4B,QAAQ;CACrC,CAAC,UAAU,MAAM;CACjB,CAAC,oCAAoC,SAAS;CAC9C,CAAC,aAAa,UAAU;CACxB,CAAC,SAAS,KAAK;CACf,CAAC,SAAS,MAAM;CAChB,CAAC,oBAAoB,QAAQ;CAC7B,CAAC,qBAAqB,QAAQ;CAC9B,CAAC,sBAAsB,UAAU;CACjC,CAAC,4BAA4B,OAAO;AACtC;;;;;;;AAQA,SAAgB,YAAY,SAA8B;CACxD,MAAM,KAAK,QAAQ,KAAK,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,EAAE,CAAE,YAAY;CACrD,MAAM,QAAQ,GAAG,QAAQ,GAAG;CAC5B,IAAI,QAAQ,GAAG;EACb,MAAM,SAAS,GAAG,MAAM,GAAG,KAAK;EAChC,MAAM,SAAS,gBAAgB;EAC/B,IAAI,QAAQ,OAAO;CACrB;CACA,KAAK,MAAM,CAAC,SAAS,WAAW,eAC9B,IAAI,QAAQ,KAAK,EAAE,GAAG,OAAO;CAE/B,OAAO;AACT;AAWA,IAAa,mBAAb,cAAsC,MAAM;CAGxB;CACA;CAHlB,YACE,SACA,UACA,QACA;EACA,MAAM,OAAO;EAHG,KAAA,WAAA;EACA,KAAA,SAAA;EAGhB,KAAK,OAAO;CACd;AACF;;;;;;;;;;;;;;AAeA,SAAgB,kBACd,QACA,OAAiC,CAAC,GACnB;CACf,MAAM,cAAc,KAAK,eAAe;CACxC,MAAM,2BAAW,IAAI,IAAiB;CACtC,KAAK,MAAM,KAAK,QAAQ;EACtB,MAAM,IAAI,YAAY,CAAC;EACvB,IAAI,MAAM,aAAa,CAAC,KAAK,cAAc;EAC3C,SAAS,IAAI,CAAC;CAChB;CACA,MAAM,OAAO,CAAC,GAAG,QAAQ,CAAC,CAAC,KAAK;CAChC,IAAI,KAAK,SAAS,aAChB,MAAM,IAAI,iBACR,wBAAwB,KAAK,OAAO,iBAAiB,KAAK,WAAW,IAAI,MAAM,MAAM,IAC/E,KAAK,KAAK,IAAI,KAAK,OAAO,QAAQ,YAAY,kFAEpD,MACA,MACF;CAEF,OAAO;AACT;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AClEA,SAAgB,iBAAiB,SAAyB;CACxD,IAAI,KAAK,QAAQ,KAAK,CAAC,CAAC,YAAY;CACpC,MAAM,KAAK,GAAG,QAAQ,GAAG;CACzB,IAAI,KAAK,GAAG,KAAK,GAAG,MAAM,GAAG,EAAE;CAC/B,MAAM,QAAQ,GAAG,QAAQ,GAAG;CAC5B,IAAI,QAAQ,GAAG,KAAK,GAAG,MAAM,GAAG,KAAK;CACrC,MAAM,QAAQ,GAAG,YAAY,GAAG;CAChC,IAAI,SAAS,GAAG,KAAK,GAAG,MAAM,QAAQ,CAAC;CACvC,OAAO,GACJ,QAAQ,uBAAuB,EAAE,CAAC,CAClC,QAAQ,WAAW,EAAE,CAAC,CACtB,QAAQ,SAAS,GAAG,CAAC,CACrB,QAAQ,OAAO,EAAE,CAAC,CAClB,KAAK;AACV;;;;;;;;;;;AAYA,SAAgB,iBACd,WACA,QACkB;CAClB,MAAM,kBAAkB,YAAY,SAAS;CAC7C,IAAI,WAAW,QAAQ,WAAW,KAAA,KAAa,OAAO,KAAK,MAAM,IAC/D,OAAO;EACL;EACA,QAAQ;EACR;EACA,cAAc;EACd,SAAS;EACT,aAAa;CACf;CAEF,MAAM,eAAe,YAAY,MAAM;CACvC,IAAI,UAAU,KAAK,CAAC,CAAC,YAAY,MAAM,OAAO,KAAK,CAAC,CAAC,YAAY,GAC/D,OAAO;EACL;EACA;EACA;EACA;EACA,SAAS;EACT,aAAa;CACf;CAEF,IAAI,iBAAiB,SAAS,MAAM,iBAAiB,MAAM,GACzD,OAAO;EACL;EACA;EACA;EACA;EACA,SAAS;EACT,aAAa;CACf;CAEF,OAAO;EACL;EACA;EACA;EACA;EACA,SACE,oBAAoB,eAAe,8BAA8B;EACnE,aAAa;CACf;AACF;AAEA,IAAa,yBAAb,cAA4C,eAAe;CAGvC;CAFlB,YACE,SACA,QACA;EACA,MAAM,sBAAsB,OAAO;EAFnB,KAAA,SAAA;EAGhB,KAAK,OAAO;CACd;AACF;AAWA,SAAgB,wBACd,OACA,OACgD;CAChD,IAAI,UAAU,KAAA,KAAa,UAAU,WAAW,UAAU,uBACxD,MAAM,IAAI,MAAM,GAAG,MAAM,0CAA0C;AAEvE;;;;;;AAuBA,SAAgB,sBACd,OACA,OAAiC,CAAC,GACzB;CACT,QAAQ,MAAM,SAAd;EACE,KAAK;EACL,KAAK,SACH,OAAO;EACT,KAAK,cACH,OAAO,KAAK,oBAAoB;EAClC,KAAK,6BACH,OAAO,KAAK,sBAAsB;EACpC,SACE,OAAO;CACX;AACF;AAEA,SAAS,SAAS,OAAiC;CACjD,IAAI,MAAM,YAAY,cACpB,OAAO,GAAG,MAAM,UAAU;CAC5B,OACE,GAAG,MAAM,UAAU,IAAI,MAAM,gBAAgB,aAAa,MAAM,OAAO,IACnE,MAAM,aAAa,KAAK,MAAM,QAAQ;AAE9C;;;;;;AAOA,SAAgB,kBACd,WACA,QACA,OAAiC,CAAC,GAChB;CAClB,MAAM,QAAQ,iBAAiB,WAAW,MAAM;CAChD,IAAI,sBAAsB,OAAO,IAAI,GAAG,OAAO;CAE/C,MAAM,IAAI,uBACR,GAFa,KAAK,UAAU,GAAG,KAAK,QAAQ,MAAM,GAExC,uBAAuB,SAAS,KAAK,EAAE,gIAEjD,CAAC,KAAK,CACR;AACF;;;;;AAMA,SAAgB,mBACd,OACA,OAAiC,CAAC,GACd;CACpB,MAAM,SAAS,MAAM,KAAK,MAAM,iBAAiB,EAAE,WAAW,EAAE,MAAM,CAAC;CACvE,MAAM,MAAM,OAAO,QAAQ,MAAM,CAAC,sBAAsB,GAAG,IAAI,CAAC;CAChE,IAAI,IAAI,SAAS,GAEf,MAAM,IAAI,uBACR,GAFa,KAAK,UAAU,GAAG,KAAK,QAAQ,MAAM,KAEtC,IAAI,OAAO,GAAG,OAAO,OAAO,+DACvB,IAAI,IAAI,QAAQ,CAAC,CAAC,KAAK,IAAI,EAAE,6FAE9C,MACF;CAEF,OAAO;AACT;AASA,IAAa,yBAAb,cAA4C,eAAe;CAGvC;CACA;CAHlB,YACE,SACA,UACA,QACA;EACA,MAAM,sBAAsB,OAAO;EAHnB,KAAA,WAAA;EACA,KAAA,SAAA;EAGhB,KAAK,OAAO;CACd;AACF;;;;;;;;;;AAWA,SAAgB,wBACd,OACA,OAAuC,CAAC,GACzB;CACf,MAAM,SAAS,mBAAmB,OAAO,IAAI;CAC7C,MAAM,2BAAW,IAAI,IAAiB;CACtC,KAAK,MAAM,SAAS,QAAQ;EAC1B,MAAM,SAAS,MAAM;EACrB,IAAI,WAAW,MAAM;EACrB,IAAI,WAAW,aAAa,CAAC,KAAK,cAAc;EAChD,SAAS,IAAI,MAAM;CACrB;CACA,MAAM,OAAO,CAAC,GAAG,QAAQ,CAAC,CAAC,KAAK;CAChC,MAAM,cAAc,KAAK,eAAe;CACxC,IAAI,KAAK,SAAS,aAEhB,MAAM,IAAI,uBACR,GAFa,KAAK,UAAU,GAAG,KAAK,QAAQ,MAAM,GAExC,gCAAgC,KAAK,OAAO,iBACjD,KAAK,WAAW,IAAI,MAAM,MAAM,IAAI,KAAK,KAAK,IAAI,KAAK,OAAO,QAAQ,YAAY,0BAC3D,OAAO,KAAK,MAAM,EAAE,UAAU,YAAY,CAAC,CAAC,KAAK,IAAI,KACjF,MACA,MACF;CAEF,OAAO;AACT;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;ACnLA,SAAgB,2BACd,SAIA,UAA2B,CAAC,GACD;CAC3B,IAAI,QAAQ,cAAc,KAAA,GAAW,OAAO,KAAA;CAC5C,IAAI,CAAC,OAAO,UAAU,QAAQ,SAAS,KAAK,QAAQ,aAAa,GAC/D,MAAM,IAAI,WAAW,kEAAkE;CAEzF,IACE,QAAQ,SAAS,MACd,YACC,MAAM,QAAQ,QAAQ,OAAO,KAAK,QAAQ,QAAQ,MAAM,SAAS,KAAK,SAAS,WAAW,CAC9F,GAEA;CAGF,MAAM,WAAW,uBAAuB,QAAQ,eAAe;CAC/D,MAAM,kBAAkB,QAAQ,wBAAwB;CAOxD,MAAM,QAAQ;EACZ,aALmB,IAAI,YAAY,CAAC,CAAC,OACrC,KAAK,UAAU,UAAU,SAAS,iBAAiB,QAAQ,QAAQ,CAAC,CACtE,CAAC,CAAC,aAG4B;EAC5B,cAAc,QAAQ,YAAY;CACpC;CACA,OAAO,QAAQ,qBACX;EAAE,oBAAoB,QAAQ;EAAoB,GAAG;CAAM,IAC3D;EAAE,OAAO,QAAQ;EAAO,GAAG;CAAM;AACvC;;;;;;AA2FA,SAAgB,mBACd,QACA,oBACkB;CAClB,MAAM,eAAe,OAAO,MAAM,sBAAsB;CACxD,MAAM,cAAc,KAAK,IAAI,GAAG,OAAO,MAAM,eAAe,YAAY;CACxE,MAAM,kBAAkB,qBAAqB,OAAO,GAAG;CACvD,OAAO;EACL,OAAO,OAAO;EACd;EACA,cAAc,OAAO,MAAM;EAC3B,GAAI,OAAO,MAAM,oBAAoB,KAAA,IACjC,CAAC,IACD,EAAE,iBAAiB,OAAO,MAAM,gBAAgB;EACpD,GAAI,eAAe,IAAI,EAAE,aAAa,IAAI,CAAC;EAC3C,GAAI,oBAAoB,KAAA,IACpB,sBAAsB,OAAO,MAAM,aAAa,QAC9C,EAAE,mBAAmB,IACrB,OAAO,YAAY,OACjB,CAAC,IACD,EAAE,kBAAkB,OAAO,QAAQ,IACvC,EAAE,eAAe,gBAAgB;EACrC,cAAc,OAAO,MAAM,aAAa;CAC1C;AACF;AAEA,SAAS,qBAAqB,KAAkD;CAC9E,MAAM,QAAQ,IAAI,kBAAkB,IAAI;CACxC,OAAO,OAAO,UAAU,YAAY,OAAO,SAAS,KAAK,KAAK,SAAS,IAAI,QAAQ,KAAA;AACrF;;AAGA,SAAgB,wBACd,OACA,oBAC8B;CAC9B,OAAO,iBAAiB,mBACpB,mBAAmB,MAAM,QAAQ,kBAAkB,IACnD,KAAA;AACN;AAEA,IAAa,eAAb,cAAkC,eAAe;CAG7B;CACA;CACA;CAJlB,YACE,SACA,QACA,MACA,OACA;EACA,MAAM,SAAS,OAAO;EAJN,KAAA,SAAA;EACA,KAAA,OAAA;EACA,KAAA,QAAA;CAGlB;AACF;;;;AAKA,IAAa,mBAAb,cAAsC,eAAe;CAGjC;CAFlB,YACE,SACA,QACA,SACA;EACA,MAAM,SAAS,SAAS,OAAO;EAHf,KAAA,SAAA;CAIlB;AACF;AA4EA,MAAM,qBAAqB,OAAO,QAAQ,IAAI,qBAAqB,KAAK;AACxE,MAAM,2BACJ,QAAQ,IAAI,gCAAgC,KAAA,IACxC,IACA,OAAO,QAAQ,IAAI,2BAA2B;AAEpD,SAAS,uBAAuB,YAAwC;CACtE,MAAM,WAAW,cAAc;CAC/B,IAAI,CAAC,OAAO,UAAU,QAAQ,KAAK,YAAY,GAC7C,MAAM,IAAI,WAAW,iDAAiD;CAExE,OAAO;AACT;AAEA,SAAS,mBAAmB,OAAoC;CAC9D,OAAO,OAAO,UAAU,YAAY,OAAO,cAAc,KAAK,KAAK,SAAS,IAAI,QAAQ,KAAA;AAC1F;AAKA,MAAM,mCAAmB,IAAI,IAAI;CAAC;CAAK;CAAK;CAAK;CAAK;CAAK;AAAG,CAAC;;;;;;;;;;AAW/D,MAAM,2BAA8C;CAClD;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;CACA;AACF;;;;;;;;;;;;AAaA,SAAgB,oBAAoB,KAAuB;CACzD,OAAO,kBAAkB,KAAK,CAAC;AACjC;AAEA,SAAS,kBAAkB,KAAc,OAAwB;CAC/D,IAAI,eAAe,cAAc,OAAO,iBAAiB,IAAI,IAAI,MAAM;CACvE,IAAI,EAAE,eAAe,QAAQ,OAAO;CAGpC,MAAM,SAAU,IAA6B;CAC7C,IAAI,OAAO,WAAW,YAAY,iBAAiB,IAAI,MAAM,GAAG,OAAO;CACvE,MAAM,OAAQ,IAA2B;CACzC,MAAM,WAAW,GAAG,IAAI,KAAK,IAAI,IAAI,QAAQ,IAAI,OAAO,SAAS,WAAW,OAAO;CACnF,IAAI,yBAAyB,MAAM,MAAM,EAAE,KAAK,QAAQ,CAAC,GAAG,OAAO;CACnE,MAAM,QAAS,IAA4B;CAC3C,IAAI,QAAQ,KAAK,iBAAiB,SAAS,UAAU,KACnD,OAAO,kBAAkB,OAAO,QAAQ,CAAC;CAE3C,OAAO;AACT;AAEA,SAAS,gBAAgB,SAAiC;CACxD,MAAM,IAAI,QAAQ,IAAI,aAAa;CACnC,IAAI,CAAC,GAAG,OAAO;CACf,MAAM,WAAW,OAAO,CAAC;CACzB,IAAI,OAAO,SAAS,QAAQ,KAAK,WAAW,GAAG,OAAO,WAAW;CACjE,MAAM,SAAS,KAAK,MAAM,CAAC;CAC3B,IAAI,OAAO,SAAS,MAAM,GAAG,OAAO,KAAK,IAAI,GAAG,SAAS,KAAK,IAAI,CAAC;CACnE,OAAO;AACT;;AAGA,SAAgB,UAAU,SAAyB;CACjD,OAAO,KAAK,IAAI,MAAM,KAAK,SAAS,IAAM;AAC5C;AAEA,SAAS,aAAa,MAAgD;CACpE,MAAM,UAAkC;EACtC,gBAAgB;EAChB,QAAQ;CACV;CACA,IAAI,KAAK,YACP,QAAQ,KAAK,WAAW,QAAQ,KAAK,WAAW;MAC3C,IAAI,KAAK,UAAU,KAAK,QAC7B,QAAQ,gBAAgB,UAAU,KAAK,UAAU,KAAK;CAExD,IAAI,KAAK,gBAAgB,QAAQ,qBAAqB,KAAK;CAC3D,OAAO;AACT;AAEA,SAAS,kBAAkB,QAAgB,MAAuB;CAChE,IAAI,WAAW,KAAK,OAAO;CAC3B,MAAM,QAAQ,KAAK,YAAY;CAC/B,OACE,MAAM,SAAS,iBAAiB,KAChC,MAAM,SAAS,aAAa,KAC5B,MAAM,SAAS,gBAAgB,KAC/B,MAAM,SAAS,eAAe;AAElC;AAEA,SAAS,0BAA0B,QAAgB,MAAuB;CACxE,IAAI,WAAW,OAAO,CAAC,eAAe,KAAK,IAAI,GAAG,OAAO;CACzD,OACE,iGAAiG,KAC/F,IACF,KAAK,oEAAoE,KAAK,IAAI;AAEtF;AAEA,SAAS,UACP,KACA,iBACA,iBACyB;CACzB,MAAM,OAAgC;EACpC,OAAO,IAAI;EACX,UAAU,IAAI,SAAS,IAAI,iBAAiB;EAC5C,aAAa,IAAI,eAAe;CAClC;CACA,IAAI,IAAI,aAAa,MAAM;EACzB,IAAI,wBAAwB,IAAI,KAAK,GAAG,KAAK,wBAAwB,IAAI;OACpE,KAAK,aAAa,IAAI;CAC7B;CACA,IAAI,IAAI,aAAa,KAAA,GAAW;EAC9B,KAAK,WAAW;EAChB,KAAK,eAAe,IAAI,SAAS;CACnC;CACA,IAAI,IAAI,UAAU,KAAA,GAAW,KAAK,QAAQ,IAAI;CAC9C,IAAI,IAAI,eAAe,KAAA,GAAW,KAAK,cAAc,IAAI;CACzD,MAAM,WAAW,IAAI,YAAY;CACjC,IAAI,aAAa,KAAA,GACf,KAAK,WAAW,EAAE,MAAM,SAAS;CAGnC,IAAI,IAAI,cAAc,CAAC,iBACrB,KAAK,kBAAkB;EACrB,MAAM;EACN,aAAa;GAAE,MAAM,IAAI,WAAW;GAAM,QAAQ,IAAI,WAAW;GAAQ,QAAQ;EAAK;CACxF;MACK,IAAI,IAAI,YAAY,IAAI,YAC7B,KAAK,kBAAkB,EAAE,MAAM,cAAc;CAG/C,OAAO;AACT;AAEA,SAAS,wBAAwB,OAAwB;CACvD,OAAO,oBAAoB,KAAK,KAAK;AACvC;;AAGA,SAAS,kBAAkB,SAA8C;CACvE,IAAI,QAAQ,SAAS,QACnB,OAAO;EAAE,MAAM;EAAQ,cAAc,QAAQ;EAAY,SAAS,QAAQ;CAAQ;CAEpF,IAAI,QAAQ,cAAc,KAAA,GACxB,OAAO;EAAE,MAAM,QAAQ;EAAM,SAAS,QAAQ;CAAQ;CAExD,OAAO;EACL,MAAM,QAAQ;EACd,SAAS,QAAQ;EACjB,YAAY,QAAQ,UAAU,KAAK,UAAU;GAC3C,IAAI,KAAK;GACT,MAAM;GACN,UAAU;IAAE,MAAM,KAAK;IAAM,WAAW,KAAK;GAAc;EAC7D,EAAE;CACJ;AACF;;;;;;;AAQA,SAAS,kBAAkB,OAAgB,OAAsD;CAC/F,IAAI,UAAU,KAAA,KAAa,UAAU,MAAM,OAAO;CAClD,IAAI,CAAC,MAAM,QAAQ,KAAK,GACtB,MAAM,IAAI,MAAM,yDAAyD,MAAM,EAAE;CAEnF,OAAO,MAAM,KAAK,OAAO,UAAU;EACjC,MAAM,SAAS;EACf,IACE,CAAC,UACD,OAAO,WAAW,YAClB,OAAO,OAAO,UAAU,YACxB,OAAO,OAAO,YAAY,YAC1B,CAAC,OAAO,SAAS,OAAO,OAAO,GAE/B,MAAM,IAAI,MACR,iCAAiC,MAAM,gDAAgD,MAAM,EAC/F;EAEF,MAAM,MAAM,OAAO;EACnB,IAAI,QAAQ,KAAA,KAAa,QAAQ,QAAQ,CAAC,MAAM,QAAQ,GAAG,GACzD,MAAM,IAAI,MACR,iCAAiC,MAAM,yCAAyC,MAAM,EACxF;EAEF,OAAO;GACL,OAAO,OAAO;GACd,SAAS,OAAO;GAChB,MAAM,MAAM,QAAQ,GAAG,IAAI,MAAM,CAAC,EAAA,CAAG,KAAK,aAAa,aAAa;IAClE,MAAM,YAAY;IAClB,IACE,CAAC,aACD,OAAO,cAAc,YACrB,OAAO,UAAU,UAAU,YAC3B,OAAO,UAAU,YAAY,YAC7B,CAAC,OAAO,SAAS,UAAU,OAAO,GAElC,MAAM,IAAI,MACR,iCAAiC,MAAM,iBAAiB,SAAS,gDAAgD,MAAM,EACzH;IAEF,OAAO;KAAE,OAAO,UAAU;KAAO,SAAS,UAAU;IAAQ;GAC9D,CAAC;EACH;CACF,CAAC;AACH;;;AAIA,SAAS,mBAAmB,OAAgB,OAA0C;CACpF,IAAI,UAAU,KAAA,KAAa,UAAU,MAAM,OAAO,KAAA;CAClD,IAAI,CAAC,MAAM,QAAQ,KAAK,GACtB,MAAM,IAAI,MAAM,mDAAmD,MAAM,EAAE;CAE7E,IAAI,MAAM,WAAW,GAAG,OAAO,KAAA;CAC/B,OAAO,MAAM,KAAK,OAAO,UAAU;EACjC,MAAM,SAAS;EAKf,MAAM,KAAK,UAAU,OAAO,WAAW,WAAW,OAAO,WAAW,KAAA;EACpE,IACE,CAAC,UACD,OAAO,WAAW,YAClB,OAAO,OAAO,OAAO,YACrB,OAAO,GAAG,WAAW,KACpB,OAAO,SAAS,KAAA,KAAa,OAAO,SAAS,cAC9C,CAAC,MACD,OAAO,OAAO,YACd,OAAO,GAAG,SAAS,YACnB,GAAG,KAAK,WAAW,KACnB,OAAO,GAAG,cAAc,UAExB,MAAM,IAAI,MACR,2BAA2B,MAAM,wDAAwD,MAAM,EACjG;EAEF,OAAO;GAAE,IAAI,OAAO;GAAI,MAAM,GAAG;GAAM,eAAe,GAAG;EAAU;CACrE,CAAC;AACH;;AAGA,eAAe,MAAM,IAAY,QAAqC;CACpE,QAAQ,eAAe;CACvB,OAAO,IAAI,SAAS,SAAS,WAAW;EACtC,MAAM,gBAAgB;GACpB,aAAa,KAAK;GAClB,QAAQ,oBAAoB,SAAS,OAAO;GAC5C,OAAO,QAAQ,MAAM;EACvB;EACA,MAAM,QAAQ,iBAAiB;GAC7B,QAAQ,oBAAoB,SAAS,OAAO;GAC5C,QAAQ;EACV,GAAG,EAAE;EACL,QAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM,KAAK,CAAC;CAC3D,CAAC;AACH;;AAGA,SAAS,iBAAiB,OAAe,YAAyC;CAChF,OAAO,cAAc,QAAQ,KAAK,IAAI,IAAI,SAAS;AACrD;;;;;;AASA,SAAgB,gBAAgB,KAAqB;CACnD,MAAM,UAAU,IAAI,KAAK;CACzB,MAAM,IAAI,QAAQ,MAAM,yCAAyC;CACjE,OAAO,IAAI,EAAE,EAAE,CAAE,KAAK,IAAI;AAC5B;AAEA,SAAgB,mBAAmB,KAAqB;CACtD,MAAM,WAAW,gBAAgB,GAAG;CACpC,IAAI;EACF,KAAK,MAAM,QAAQ;EACnB,OAAO;CACT,QAAQ;EAIN,IAAI,SAAS,WAAW,GAAG,KAAK,SAAS,WAAW,GAAG,GAAG,OAAO;CACnE;CAGA,MAAM,SAAS,CAAC,GAAG,SAAS,SAAS,OAAO,CAAC,CAAC,CAC3C,KAAK,UAAU,MAAM,KAAK,CAAC,CAC3B,QAAQ,UAAU,SAAS,IAAI;CAClC,KAAK,MAAM,SAAS,QAAQ;EAC1B,MAAM,YAAY,oBAAoB,UAAU,KAAK;EACrD,IAAI,CAAC,WAAW;EAChB,IAAI;GACF,KAAK,MAAM,SAAS;GACpB,OAAO;EACT,QAAQ,CAER;CACF;CAEA,OAAO;AACT;AAEA,SAAS,oBAAoB,OAAe,OAA8B;CACxE,MAAM,SAAS,MAAM;CACrB,MAAM,SAAS,WAAW,MAAM,MAAM,WAAW,MAAM,MAAM;CAC7D,IAAI,CAAC,QAAQ,OAAO;CAEpB,MAAM,QAAkB,CAAC,MAAM;CAC/B,IAAI,aAAa;CACjB,IAAI,YAAY;CAEhB,KAAK,IAAI,IAAI,QAAQ,GAAG,IAAI,MAAM,QAAQ,KAAK;EAC7C,MAAM,OAAO,MAAM;EACnB,IAAI,WAAW;GACb,YAAY;GACZ;EACF;EACA,IAAI,SAAS,MAAM;GACjB,YAAY;GACZ;EACF;EACA,IAAI,SAAS,MAAK;GAChB,aAAa,CAAC;GACd;EACF;EACA,IAAI,YAAY;EAEhB,IAAI,SAAS,KAAK,MAAM,KAAK,GAAG;OAC3B,IAAI,SAAS,KAAK,MAAM,KAAK,GAAG;OAChC,IAAI,SAAS,MAAM,MAAM,SAAS,IAAI;GACzC,MAAM,IAAI;GACV,IAAI,MAAM,WAAW,GAAG,OAAO,MAAM,MAAM,OAAO,IAAI,CAAC;EACzD;CACF;CAEA,OAAO;AACT;;;;;;AAOA,eAAsB,QACpB,KACA,OAAyB,CAAC,GACF;CACxB,OAAO,gBAAgB,KAAK,MAAM,KAAK;AACzC;;;AAIA,eAAe,gBACb,KACA,MACA,qBACwB;CAIxB,IAAI,CAAC,KAAK,SAAS,MAAM,IAAI,MAAM,mCAAmC;CACtE,MAAM,UAAU,KAAK,QAAQ,QAAQ,QAAQ,EAAE;CAC/C,MAAM,MAAM,GAAG,QAAQ;CACvB,MAAM,WAAW;CACjB,MAAM,YAAY,IAAI,aAAa,KAAK,oBAAoB;CAC5D,MAAM,kBAAkB,uBAAuB,KAAK,eAAe;CACnE,MAAM,UAAU,KAAK,SAAS,WAAW;CACzC,MAAM,UAAU,aAAa,IAAI;CACjC,MAAM,WAAW,KAAK,YAAY,oBAAoB,OAAO;CAC7D,MAAM,OAAO,KAAK;CAClB,MAAM,WAAW,KAAK,YAAY;CAClC,MAAM,eAAe,KAAK;CAC1B,MAAM,eAAe,KAAK;CAC1B,MAAM,aAAa,KAAK;CACxB,MAAM,gBAAgB,KAAK,IAAI;CAC/B,IAAI,KAAK,oBACP,oBAAoB,KAAK,oBAAoB;EAAE,aAAa;EAAG,cAAc;CAAE,CAAC;CAGlF,IAAI;CACJ,IAAI,mBAAmB;CACvB,MAAM,oBACJ,eAAe,KAAA,IAAY,WAAW,KAAK,IAAI,GAAG,cAAc,KAAK,IAAI,IAAI,cAAc;CAC7F,MAAM,cAAc,OAAe,MAAM,KAAK,IAAI,IAAI,YAAY,CAAC,GAAG,YAAY;CAClF,KAAK,IAAI,UAAU,GAAG,UAAU,iBAAiB,WAAW;EAG1D,IAAI,cAAc,SAChB,MAAM,IAAI,aAAa,oCAAoC,YAAY;EAIzE,IAAI,iBAAiB,eAAe,UAAU,GAC5C,MAAM,mBAAmB,QACrB,UACA,IAAI,aAAa,6BAA6B,cAAc;EAElE,MAAM,cAAc,UAClB,kBACA,KAAK,wBAAwB,eAC7B,KAAK,QACP;EACA,MAAM,aAAa,IAAI,gBAAgB;EACvC,MAAM,gBAAgB,oBAAoB,WAAW,QAAQ,YAAY;EACzE,MAAM,gBAAgB,iBAAiB,WAAW,MAAM,GAAG,KAAK,IAAI,WAAW,YAAY,CAAC,CAAC;EAC7F,MAAM,UAAU,KAAK,IAAI;EACzB,IAAI,uBAAuB;EAC3B,IAAI;GACF,IAAI,MACF,MAAM,UAAU,MAAM,UAAU;IAC9B,SAAS,YAAY;IACrB,OAAO,cAAc;IACrB,QAAQ,cAAc;IACtB;IACA,OAAO,IAAI;IACX;IACA;IACA,cAAc;IACd,WAAW;IACX,WAAW;IACX,gBAAgB;IAChB;IACA,gBAAgB,CAAC;GACnB,CAAC;GAEH,cAAc,eAAe;GAC7B,MAAM,MAAM,MAAM,QAAQ,KAAK;IAC7B,QAAQ;IACR;IACA,MAAM,KAAK,UAAU,WAAW;IAChC,QAAQ;GACV,CAAC;GACD,MAAM,kBAAkB,OAAO,gBAAgB,IAAI,OAAO,IAAI,KAAA;GAE9D,MAAM,OAAO,MAAM,IAAI,KAAK;GAC5B,cAAc,eAAe;GAC7B,aAAa,aAAa;GAE1B,IAAI,CAAC,IAAI,IAAI;IACX,MAAM,OAAO;IACb,IAAI,MAAM;KACR,MAAM,UAAU,MAAM,UAAU;MAC9B,SAAS,YAAY;MACrB,OAAO,cAAc;MACrB,QAAQ,cAAc;MACtB;MACA,OAAO,IAAI;MACX;MACA;MACA,cAAc;MACd,WAAW;MACX,WAAW,KAAK,IAAI;MACpB,YAAY,KAAK,IAAI,IAAI;MACzB,YAAY,IAAI;MAChB;MACA,cAAc;MACd,cAAc,QAAQ,IAAI;MAC1B,gBAAgB,CAAC;KACnB,CAAC;KACD,uBAAuB;IACzB;IACA,MAAM,MAAM,IAAI,aACd,6BAA6B,IAAI,UACjC,IAAI,QACJ,MACA,IAAI,KACN;IACA,IACE,0BAA0B,IAAI,QAAQ,IAAI,KAC1C,iBAAiB,gBAAgB,KACjC,UAAU,kBAAkB,KAC5B,CAAC,iBAAiB,eAAe,UAAU,GAC3C;KACA,UAAU;KACV,mBAAmB;MAAE,GAAG;MAAkB,aAAa;KAAE;KACzD;IACF;IACA,IACE,uBACA,KAAK,wBAAwB,iBAC7B,iBAAiB,cACjB,kBAAkB,IAAI,QAAQ,IAAI,KAClC,UAAU,kBAAkB,KAC5B,CAAC,iBAAiB,eAAe,UAAU,GAC3C;KACA,UAAU;KACV,mBAAmB;MAAE,GAAG;MAAkB,UAAU;MAAM,YAAY,KAAA;KAAU;KAChF;IACF;IACA,IACE,iBAAiB,IAAI,IAAI,MAAM,KAC/B,UAAU,kBAAkB,KAC5B,CAAC,iBAAiB,eAAe,UAAU,GAC3C;KACA,UAAU;KAEV,MAAM,WADa,gBAAgB,IAAI,OACb,KAAK,UAAU,OAAO,CAAC;KACjD;IACF;IACA,MAAM;GACR;GAEA,IAAI;GACJ,IAAI;IACF,OAAO,KAAK,MAAM,IAAI;GACxB,SAAS,UAAU;IACjB,IAAI,MAAM;KACR,MAAM,UAAU,MAAM,UAAU;MAC9B,SAAS,YAAY;MACrB,OAAO,cAAc;MACrB,QAAQ,cAAc;MACtB;MACA,OAAO,IAAI;MACX;MACA;MACA,cAAc;MACd,WAAW;MACX,WAAW,KAAK,IAAI;MACpB,YAAY,KAAK,IAAI,IAAI;MACzB,YAAY,IAAI;MAChB;MACA,cAAc;MACd,cAAc,sBAAsB,oBAAoB,QAAQ,SAAS,UAAU,OAAO,QAAQ;MAClG,gBAAgB,CAAC;KACnB,CAAC;KACD,uBAAuB;IACzB;IACA,MAAM;GACR;GACA,IAAI,MACF,MAAM,UAAU,MAAM,UAAU;IAC9B,SAAS,YAAY;IACrB,OAAO,cAAc;IACrB,QAAQ,cAAc;IACtB;IACA,OAAO,IAAI;IACX;IACA;IACA,cAAc;IACd,WAAW;IACX,WAAW,KAAK,IAAI;IACpB,YAAY,KAAK,IAAI,IAAI;IACzB,YAAY,IAAI;IAChB;IACA,cAAc;IACd,gBAAgB,CAAC;GACnB,CAAC;GAEH,MAAM,SACJ,KAAK,UAOH;GACJ,MAAM,YAAY,mBAAmB,QAAQ,SAAS,YAAY,IAAI,KAAK;GAC3E,MAAM,WACJ,KAAK,SAAS,OAAO,KAAK,UAAU,YAAY,CAAC,MAAM,QAAQ,KAAK,KAAK,IACpE,KAAK,QACN,KAAA;GACN,MAAM,eAAe,mBAAmB,UAAU,aAAa;GAC/D,MAAM,mBAAmB,mBAAmB,UAAU,iBAAiB;GACvE,MAAM,cAAc,mBAAmB,UAAU,YAAY;GAO7D,MAAM,gBALJ,UAAU,6BACV,OAAO,SAAS,8BAA8B,YAC9C,CAAC,MAAM,QAAQ,SAAS,yBAAyB,IAC5C,SAAS,4BACV,KAAA,EAAA,EACkC;GACxC,MAAM,kBACJ,iBAAiB,KAAA,IAAY,KAAA,IAAY,mBAAmB,YAAY;GAC1E,MAAM,YACJ,UAAU,yBACV,OAAO,SAAS,0BAA0B,YAC1C,CAAC,MAAM,QAAQ,SAAS,qBAAqB,IACxC,SAAS,sBAAkD,gBAC5D,KAAA;GACN,MAAM,qBAAqB,cAAc,KAAA,IAAY,KAAA,IAAY,mBAAmB,SAAS;GAC7F,MAAM,gBACJ,iBAAiB,KAAA,KACjB,qBAAqB,KAAA,MACpB,iBAAiB,KAAA,KACf,oBAAoB,KAAA,KAAa,mBAAmB,sBACtD,cAAc,KAAA,KACZ,uBAAuB,KAAA,KAAa,sBAAsB,kBAC5D,gBAAgB,KAAA,KAAa,gBAAgB,eAAe;GAC/D,MAAM,gBAAgB,qBAAqB,IAAI;GAC/C,MAAM,UAAU,QAAQ,SAAS,WAAW;GAE5C,MAAM,iBACJ,OAAO,kBAAkB,YAAY,iBAAiB,KAAK,qBACvD,oBAAoB,KAAK,oBAAoB;IAC3C,aAAa,gBAAiB,sBAAsB;IACpD,GAAI,qBAAqB,EAAE,cAAc,mBAAmB,IAAI,CAAC;IACjE,cAAc;GAChB,CAAC,IACD,KAAA;GAIN,MAAM,cACJ,OAAO,KAAK,UAAU,YAAY,KAAK,MAAM,KAAK,MAAM,KAAK,KAAK,QAAQ;GAC5E,IAAI,KAAK,mBACP,kBACE,IAAI,OACJ,aACA,KAAK,sBAAsB,OAAO,CAAC,IAAI,KAAK,iBAC9C;GAGF,OAAO;IACL;IACA,GAAI,cAAc,KAAA,IAAY,CAAC,IAAI,EAAE,UAAU;IAC/C,GAAI,IAAI,aAAa,KAAA,IACjB,CAAC,IACD,EAAE,UAAU,kBAAkB,QAAQ,UAAU,SAAS,IAAI,KAAK,EAAE;IACxE,cAAc,QAAQ,iBAAiB;IACvC,cAAc,QAAQ,KAAK,CAAC,CAAC,WAAW;IACxC,OAAO;KACL,cAAc,gBAAgB;KAC9B,kBAAkB,oBAAoB;KACtC,aAAa,gBAAgB,gBAAgB,MAAM,oBAAoB;KACvE,UAAU;KACV;KACA;IACF;IACA,SAAS,OAAO,kBAAkB,WAAW,gBAAiB,kBAAkB;IAChF,OAAO,eAAe,IAAI;IAC1B;IACA,YAAY,KAAK,IAAI,IAAI;IACzB,KAAK;GACP;EACF,SAAS,KAAK;GACZ,aAAa,aAAa;GAC1B,UAAU;GAIV,IAAI,cAAc,SAAS;IACzB,IAAI,QAAQ,CAAC,sBACX,MAAM,UAAU,MAAM,UAAU;KAC9B,SAAS,YAAY;KACrB,OAAO,cAAc;KACrB,QAAQ,cAAc;KACtB;KACA,OAAO,IAAI;KACX;KACA;KACA,cAAc;KACd,WAAW;KACX,WAAW,KAAK,IAAI;KACpB,YAAY,KAAK,IAAI,IAAI;KACzB,cAAc,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;KAC7D,gBAAgB,CAAC;IACnB,CAAC;IAEH,MAAM;GACR;GACA,IAAI,QAAQ,CAAC,sBAIX,MAAM,UAAU,MAAM,UAAU;IAC9B,SAAS,YAAY;IACrB,OAAO,cAAc;IACrB,QAAQ,cAAc;IACtB;IACA,OAAO,IAAI;IACX;IACA;IACA,cAAc;IACd,WAAW;IACX,WAAW,KAAK,IAAI;IACpB,YAAY,KAAK,IAAI,IAAI;IACzB,cAAc,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;IAC7D,gBAAgB,CAAC;GACnB,CAAC;GAEH,IACE,UAAU,kBAAkB,KAC5B,oBAAoB,GAAG,KACvB,CAAC,iBAAiB,eAAe,UAAU,GAC3C;IACA,MAAM,WAAW,UAAU,OAAO,CAAC;IACnC;GACF;GACA,MAAM;EACR,UAAU;GACR,aAAa,aAAa;EAC5B;CACF;CACA,MAAM,mBAAmB,QAAQ,UAAU,IAAI,MAAM,OAAO,OAAO,CAAC;AACtE;AAEA,eAAe,UACb,MACA,UACA,OACe;CAGf,IAAI;EACF,MAAM,KAAK,OAAO,SAAS,KAAK,CAAC;CACnC,QAAQ,CAER;AACF;AAEA,SAAS,gBAAgB,GAAoC;CAC3D,MAAM,MAA8B,CAAC;CACrC,EAAE,SAAS,OAAO,QAAQ;EACxB,IAAI,OAAO;CACb,CAAC;CACD,OAAO;AACT;;;;;;;AAQA,eAAsB,YACpB,KACA,OAAyB,CAAC,GACoB;CAC9C,MAAM,SAAS,MAAM,kBAAkB,KAAK,IAAI;CAEhD,OAAO;EAAE,OADK,gBAAmB,QAAQ,KAAK,mBAAmB,SACpD;EAAG;CAAO;AACzB;;AAGA,eAAe,kBACb,KACA,OAAyB,CAAC,GACF;CACxB,OAAO,gBAAgB;EAAE,GAAG;EAAK,UAAU,IAAI,YAAY,CAAC,IAAI;CAAW,GAAG,MAAM,IAAI;AAC1F;AAEA,SAAS,gBACP,QACA,iBACG;CACH,IAAI;EACF,IAAI,OAAO,iBAAiB,UAC1B,MAAM,IAAI,MACR,8CAA8C,OAAO,MAAM,uBAC7D;EAEF,OAAO,gBAAmB,OAAO,SAAS,OAAO,OAAO,eAAe;CACzE,SAAS,OAAO;EACd,IAAI,iBAAiB,kBAAkB,MAAM;EAC7C,MAAM,QAAQ,iBAAiB,QAAQ,QAAQ,IAAI,MAAM,OAAO,KAAK,CAAC;EACtE,MAAM,IAAI,iBAAiB,MAAM,SAAS,QAAQ,EAAE,MAAM,CAAC;CAC7D;AACF;AAEA,SAAS,gBACP,SACA,OACA,iBACG;CACH,MAAM,UAAU,oBAAoB,UAAU,UAAU,mBAAmB,OAAO;CAClF,IAAI;EACF,OAAO,KAAK,MAAM,OAAO;CAC3B,QAAQ;EACN,MAAM,IAAI,MAAM,wCAAwC,MAAM,EAAE;CAClE;AACF;;;;;;AAOA,IAAa,YAAb,MAAuB;CACrB;CACA;CAEA,YAAY,OAAyB,CAAC,GAAG;EACvC,KAAK,OAAO;EACZ,KAAK,kBAAkB,uBAAuB,KAAK,eAAe;CACpE;CAEA,KAAK,KAAqB,KAAgD;EACxE,MAAM,UAAU;GAAE,GAAG,KAAK;GAAM,GAAG;EAAI;EACvC,OAAO,IAAI,aAAa,kBAAkB,KAAK,OAAO,IAAI,QAAQ,KAAK,OAAO;CAChF;CAEA,SACE,KACA,KAC8C;EAC9C,OAAO,YAAe,KAAK;GAAE,GAAG,KAAK;GAAM,GAAG;EAAI,CAAC;CACrD;AACF"}
@@ -11,19 +11,19 @@ import { i as CostLedger, t as CostAccountingIncompleteError } from "./cost-ledg
11
11
  import { F as surfaceContentHash, G as campaignCoverage, H as assertCampaignDesign, I as surfaceDispatchRef, K as campaignScenarioIdentity, L as surfaceHash, O as FinalEvidenceConflictError, P as renderSurfaceDiff, R as surfaceHashMatches, S as campaignMeanCompositeOrNull, U as assertCampaignSplitIdentity, W as assertCompleteCampaign, Y as formatCoverageFailures, b as campaignBreakdown, k as FinalEvidenceError, q as campaignSplitDigest, t as createCampaignEvidenceReceipt, x as campaignMeanComposite, z as BackendIntegrityError } from "./campaign-evidence-Dt_rWPk2.js";
12
12
  import { m as mapConcurrent } from "./journal-DvKCvjty.js";
13
13
  import { t as deepFreezeCanonicalJson } from "./deep-freeze-CurcOQg3.js";
14
- import { S as fsCampaignStorage, g as isRecord, h as isExternalTextCandidate, x as createRunCostLedger } from "./external-optimizer-subprocess-CkyUUV7h.js";
14
+ import { S as fsCampaignStorage, g as isRecord, h as isExternalTextCandidate, x as createRunCostLedger } from "./external-optimizer-subprocess-BmKtH3T6.js";
15
15
  import { r as summarizeEvaluationUnits, t as defineEvaluationClaim } from "./claim-BXcZw-jo.js";
16
16
  import { f as redactText, n as REDACTION_VERSION, o as detectCredential, s as detectPersonalData } from "./redact-Dk1ZdJEm.js";
17
17
  import { t as detectRewardHacking } from "./reward-hacking-C-a__ShZ.js";
18
- import { n as assertProposalFindings, t as combineAbortSignals } from "./abort-signal-CtzAM_sJ.js";
18
+ import { t as assertProposalFindings } from "./proposal-findings-CXblsskH.js";
19
19
  import { D as paretoFrontier, O as paretoFrontierWithCrowding, a as searchPolicyView, b as estimateNode, i as searchExpansionIndex, n as runSearch, p as EXPANSION_OPERATORS, t as SEARCH_KERNEL_SOURCE } from "./search-kernel-Dpniloks.js";
20
20
  import { a as validateSearchLedgerEvent, d as searchTaskSetDigest, i as replaySearchLedgerText, l as searchCellId, n as openSearchLedger, u as searchNodeId } from "./search-ledger-CXQPwdWz.js";
21
21
  import { n as searchProposerView, t as renderSearchSummary } from "./search-summary-C1nAHgkB.js";
22
- import { d as stripFencedJson, o as costReceiptFromLlm, s as costReceiptFromLlmError, u as maximumChargeForLlmRequest } from "./llm-client-Bhqoglar.js";
22
+ import { C as combineAbortSignals, d as stripFencedJson, o as costReceiptFromLlm, s as costReceiptFromLlmError, u as maximumChargeForLlmRequest } from "./llm-client-DmCNzrc9.js";
23
23
  import { createHash } from "node:crypto";
24
24
  import { writeFileSync } from "node:fs";
25
25
  import { z } from "zod";
26
- import { basename, dirname, isAbsolute, join } from "node:path";
26
+ import { basename, dirname, isAbsolute, join, relative, resolve, sep } from "node:path";
27
27
  import { homedir, tmpdir } from "node:os";
28
28
  import { execSync } from "node:child_process";
29
29
  import { fileURLToPath, pathToFileURL } from "node:url";
@@ -175,6 +175,10 @@ function buildCellSchedule(scenarios, seed, reps) {
175
175
  function cellDirectory(runDir, cellId) {
176
176
  return join(runDir, cellId.replace(/[^a-zA-Z0-9_-]/g, "_"));
177
177
  }
178
+ function cellAttemptDirectory(cellDir, attempt) {
179
+ if (!/^[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12}$/i.test(attempt.runAttemptId) || !Number.isSafeInteger(attempt.number) || attempt.number < 1) throw new Error("Invalid campaign attempt identity");
180
+ return join(cellDir, "attempts", attempt.runAttemptId, `attempt-${attempt.number}`);
181
+ }
178
182
  function cellCachePath(runDir, cellId) {
179
183
  return join(cellDirectory(runDir, cellId), "cached-result.json");
180
184
  }
@@ -342,6 +346,10 @@ async function executeCell(args) {
342
346
  ...stableCostTags,
343
347
  runAttemptId: args.runAttemptId
344
348
  };
349
+ const attemptCostTags = {
350
+ ...costTags,
351
+ attemptNumber: String(args.attempt)
352
+ };
345
353
  const cachePath = join(cellDir, "cached-result.json");
346
354
  if (args.resumable) {
347
355
  const cached = readCachedCell({
@@ -375,15 +383,39 @@ async function executeCell(args) {
375
383
  }
376
384
  }
377
385
  const startMs = Date.now();
378
- const trace = args.buildTraceWriter(args.slot.cellId, cellDir);
386
+ const attempt = {
387
+ runAttemptId: args.runAttemptId,
388
+ number: args.attempt
389
+ };
390
+ const attemptDir = cellAttemptDirectory(cellDir, attempt);
391
+ const artifactDir = join(attemptDir, "artifacts");
392
+ const traceDir = join(attemptDir, "trace");
393
+ storage.ensureDir(artifactDir);
394
+ storage.ensureDir(traceDir);
395
+ const attemptPrefix = relative(cellDir, attemptDir).split(sep).join("/");
379
396
  const artifactsByPath = {};
397
+ const identityPath = join(attemptDir, "identity.json");
398
+ storage.write(identityPath, JSON.stringify({
399
+ ...attempt,
400
+ manifestHash: args.manifestHash,
401
+ cellId: args.slot.cellId,
402
+ scenarioId: args.slot.scenario.id,
403
+ rep: args.slot.rep,
404
+ seed: args.slot.cellSeed,
405
+ startedAt: args.now().toISOString()
406
+ }, null, 2));
407
+ artifactsByPath[`${args.slot.cellId}/${attemptPrefix}/identity.json`] = identityPath;
408
+ storage.write(join(cellDir, "latest-attempt.json"), JSON.stringify(attempt));
409
+ const trace = args.buildTraceWriter(args.slot.cellId, traceDir);
380
410
  let paidCallStarted = false;
381
411
  const artifacts = {
382
412
  async write(path, content) {
383
- const fullPath = join(cellDir, path);
413
+ const fullPath = resolve(artifactDir, path);
414
+ const artifactPath = relative(artifactDir, fullPath);
415
+ if (!artifactPath || artifactPath === ".." || artifactPath.startsWith(".." + sep) || isAbsolute(artifactPath)) throw new Error("Artifact path must remain inside this execution attempt");
384
416
  storage.ensureDir(join(fullPath, ".."));
385
417
  storage.write(fullPath, content);
386
- artifactsByPath[`${args.slot.cellId}/${path}`] = fullPath;
418
+ artifactsByPath[`${args.slot.cellId}/${attemptPrefix}/artifacts/${artifactPath.split(sep).join("/")}`] = fullPath;
387
419
  return fullPath;
388
420
  },
389
421
  async writeJson(path, value) {
@@ -397,7 +429,7 @@ async function executeCell(args) {
397
429
  channel: input.channel ?? "agent",
398
430
  phase: args.costPhase,
399
431
  actor: input.actor,
400
- tags: costTags,
432
+ tags: attemptCostTags,
401
433
  signal: cellAbort.signal
402
434
  });
403
435
  if (result.receipt) trace.span(`cost.${result.receipt.actor}`, { amountUsd: result.receipt.costUsd }).end();
@@ -531,7 +563,7 @@ async function executeCell(args) {
531
563
  signal: args.signal,
532
564
  costLedger: args.costLedger,
533
565
  costPhase: args.costPhase,
534
- costTags
566
+ costTags: attemptCostTags
535
567
  });
536
568
  judgeScores[judge.name] = score;
537
569
  } catch (err) {
@@ -555,6 +587,7 @@ async function executeCell(args) {
555
587
  const costCallIds = args.costLedger.list({ tags: costTags }).map((receipt) => receipt.callId).sort();
556
588
  const cell = {
557
589
  manifestHash: args.manifestHash,
590
+ attempt,
558
591
  cellId: args.slot.cellId,
559
592
  scenarioId: args.slot.scenario.id,
560
593
  rep: args.slot.rep,
@@ -591,13 +624,15 @@ async function executeCell(args) {
591
624
  })
592
625
  };
593
626
  retry = args.cellRetry !== void 0 && args.attempt < args.cellRetry.attempts && !args.signal.aborted && fatalCellError === void 0 && args.cellRetry.retryable(receipt.failure);
594
- const receiptName = retry ? `failure-receipt.attempt-${args.attempt}.json` : "failure-receipt.json";
595
- const failurePath = join(cellDir, receiptName);
627
+ const failurePath = join(attemptDir, "failure-receipt.json");
596
628
  storage.write(failurePath, JSON.stringify(receipt, null, 2));
597
- artifactsByPath[`${args.slot.cellId}/${receiptName}`] = failurePath;
629
+ artifactsByPath[`${args.slot.cellId}/${attemptPrefix}/failure-receipt.json`] = failurePath;
598
630
  if (!retry) args.onFailure?.(failure);
599
631
  }
600
632
  await trace.flush();
633
+ const resultPath = join(attemptDir, "result.json");
634
+ storage.write(resultPath, JSON.stringify(cell));
635
+ artifactsByPath[`${args.slot.cellId}/${attemptPrefix}/result.json`] = resultPath;
601
636
  if (!errorMessage && args.resumable) storage.write(cachePath, JSON.stringify(cell));
602
637
  if (fatalCellError !== void 0) throw fatalCellError;
603
638
  return {
@@ -4556,14 +4591,21 @@ var SurfaceNodes = class {
4556
4591
  if (!cell) throw new Error(`runOptimization: premeasured baseline lacks ${work.taskId}:${work.rep}`);
4557
4592
  return this.cellResult(cell, work);
4558
4593
  }
4559
- /** A cell's result as its campaign stored it: the cache of a scored cell,
4560
- * the final failure receipt of a failed one. */
4561
4594
  /** A cell's result as its campaign stored it for this surface: the cache of
4562
4595
  * a scored cell, the final failure receipt of a failed one. A result another
4563
4596
  * surface left in the same directory does not count. */
4564
4597
  storedCell(dir, surface, taskId, rep) {
4565
4598
  const cellId = `${taskId}:${rep}`;
4566
4599
  const manifestHash = this.manifestFor(surface);
4600
+ const cellDir = cellDirectory(dir, cellId);
4601
+ const latest = this.storage.read(join(cellDir, "latest-attempt.json"));
4602
+ if (latest !== void 0) {
4603
+ const attemptDir = cellAttemptDirectory(cellDir, JSON.parse(latest));
4604
+ const result = this.storage.read(join(attemptDir, "result.json"));
4605
+ const receipt = this.storage.read(join(attemptDir, "failure-receipt.json"));
4606
+ const cell = result !== void 0 ? JSON.parse(result) : receipt !== void 0 ? JSON.parse(receipt).cell : void 0;
4607
+ return cell?.manifestHash === manifestHash ? cell : void 0;
4608
+ }
4567
4609
  const cached = this.storage.read(cellCachePath(dir, cellId));
4568
4610
  const scored = cached === void 0 ? void 0 : JSON.parse(cached);
4569
4611
  if (scored?.manifestHash === manifestHash) return scored;
@@ -6132,4 +6174,4 @@ function firstString(value) {
6132
6174
  //#endregion
6133
6175
  export { captureFinalEvidencePolicy as $, assertCompleteSearchHistory as A, searchReceiptAccounting as B, incumbentWithOperatorBandit as C, surfaceDiff as D, searchEdgeId as E, verifySearchHistoryArtifact as F, captureJudge as G, uniform as H, verifySearchHistoryReceipt as I, redTeamDataset as J, defaultProductionGate as K, campaignCellSearchResult as L, assertSearchHistoryMatchesState as M, createSearchHistoryReceipt as N, surfaceNode as O, searchHistoryCoverageRow as P, assertIndependentEvaluationSplit as Q, searchModelIdentity as R, incumbent as S, developmentClaim as T, runFinalComparison as U, asha as V, openAutoPr as W, scoreRedTeamOutput as X, redTeamReport as Y, runCanaries as Z, isProposedCandidate as _, transientDispatchFailure as a, resolveRunDir as at, beam as b, compareOptimizationMethods as c, buildCellSchedule as ct, combineComparisonCosts as d, evaluationUnitMap as et, costFromLedgerSummary as f, runOptimization as g, runImprovementLoop as h, quotaExhaustedUntil as i, planCampaignRun as it, assertSearchHistoryAdmissionOptions as j, SearchHistoryRequiredError as k, optimizationTokenUsageFromSummary as l, cellCachePath as lt, readGepaCandidatePopulationArtifact as m, JudgeParseError as n, runEval as nt, aggregateRunScore as o, tangleTracesRoot as ot, assertGepaCandidatePopulationSummary as p, DEFAULT_RED_TEAM_CORPUS as q, isTransientTransportFailure as r, runCampaign as rt, clamp01 as s, readCachedCell as st, llmJudge as t, reserveFinalEvidence as tt, executeOptimizationMethod as u, computeManifestHash as ut, labelTrustRank as v, SearchRecorder as w, crowdedFrontierParent as x, aide as y, searchProposalExecution as z };
6134
6176
 
6135
- //# sourceMappingURL=llm-judge-C-0P1WiT.js.map
6177
+ //# sourceMappingURL=llm-judge-DyZdrp6I.js.map