@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.
- package/CHANGELOG.md +10 -0
- package/dist/adapters/http.d.ts +10 -4
- package/dist/adapters/http.d.ts.map +1 -1
- package/dist/adapters/http.js +43 -38
- package/dist/adapters/http.js.map +1 -1
- package/dist/{agent-profile-DDfIjdyS.d.ts → agent-profile-EpO0HURy.d.ts} +4 -4
- package/dist/{agent-profile-DDfIjdyS.d.ts.map → agent-profile-EpO0HURy.d.ts.map} +1 -1
- package/dist/analyst/index.d.ts +12 -12
- package/dist/analyst/index.d.ts.map +1 -1
- package/dist/analyst/index.js +6 -6
- package/dist/{analyze-runs-CPiKLJFs.d.ts → analyze-runs-Q74o4Hfj.d.ts} +2 -2
- package/dist/{analyze-runs-CPiKLJFs.d.ts.map → analyze-runs-Q74o4Hfj.d.ts.map} +1 -1
- package/dist/{benchmark-command-DFs_i62N.js → benchmark-command-6OBcB0sG.js} +7 -6
- package/dist/benchmark-command-6OBcB0sG.js.map +1 -0
- package/dist/benchmarks/index.d.ts +4 -4
- package/dist/benchmarks/index.js +3 -3
- package/dist/campaign/index.d.ts +6 -6
- package/dist/campaign/index.js +4 -4
- package/dist/{campaign-Btk1yFfw.js → campaign-DYhg4SjN.js} +6 -6
- package/dist/{campaign-Btk1yFfw.js.map → campaign-DYhg4SjN.js.map} +1 -1
- package/dist/{chat-client-DUhWNQQS.js → chat-client-5LLl4Ctl.js} +2 -2
- package/dist/{chat-client-DUhWNQQS.js.map → chat-client-5LLl4Ctl.js.map} +1 -1
- package/dist/{chat-json-call-Dx_sHzqm.js → chat-json-call-Digk-pfF.js} +2 -2
- package/dist/{chat-json-call-Dx_sHzqm.js.map → chat-json-call-Digk-pfF.js.map} +1 -1
- package/dist/cli.js +3 -3
- package/dist/contract/index.d.ts +10 -10
- package/dist/contract/index.js +7 -7
- package/dist/{default-registry-B-utjsmX.d.ts → default-registry-BHEfZHQd.d.ts} +6 -6
- package/dist/{default-registry-B-utjsmX.d.ts.map → default-registry-BHEfZHQd.d.ts.map} +1 -1
- package/dist/{default-registry-CpswV_S2.js → default-registry-wInsQWlQ.js} +2 -2
- package/dist/{default-registry-CpswV_S2.js.map → default-registry-wInsQWlQ.js.map} +1 -1
- package/dist/{define-agent-eval-CJN9dTQw.js → define-agent-eval-B25Cc3ki.js} +3 -3
- package/dist/{define-agent-eval-CJN9dTQw.js.map → define-agent-eval-B25Cc3ki.js.map} +1 -1
- package/dist/{define-agent-eval-MAc6BQrq.d.ts → define-agent-eval-D8bPoBM5.d.ts} +7 -7
- package/dist/{define-agent-eval-MAc6BQrq.d.ts.map → define-agent-eval-D8bPoBM5.d.ts.map} +1 -1
- package/dist/diagnosis/index.d.ts +2 -2
- package/dist/{dspy-rlm-engine-US8YsYGq.js → dspy-rlm-engine-D0h-P2Mz.js} +2 -2
- package/dist/{dspy-rlm-engine-US8YsYGq.js.map → dspy-rlm-engine-D0h-P2Mz.js.map} +1 -1
- package/dist/{engine-4_Q4oujc.d.ts → engine-F_PvwesR.d.ts} +4 -4
- package/dist/{engine-4_Q4oujc.d.ts.map → engine-F_PvwesR.d.ts.map} +1 -1
- package/dist/evaluation.d.ts +2 -2
- package/dist/{exact-types-KGxBWH4b.d.ts → exact-types-BBMMYk3g.d.ts} +2 -2
- package/dist/{exact-types-KGxBWH4b.d.ts.map → exact-types-BBMMYk3g.d.ts.map} +1 -1
- package/dist/experiment/index.d.ts +1 -1
- package/dist/{external-optimizer-process-KLcw0Vnc.js → external-optimizer-process-CIV4EqiT.js} +2 -2
- package/dist/{external-optimizer-process-KLcw0Vnc.js.map → external-optimizer-process-CIV4EqiT.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-CkyUUV7h.js → external-optimizer-subprocess-BmKtH3T6.js} +2 -2
- package/dist/{external-optimizer-subprocess-CkyUUV7h.js.map → external-optimizer-subprocess-BmKtH3T6.js.map} +1 -1
- package/dist/{extract-usage-DFATEuWW.js → extract-usage-BrnzRJ6H.js} +4 -3
- package/dist/extract-usage-BrnzRJ6H.js.map +1 -0
- package/dist/{feedback-trajectory-BXCGTRkz.d.ts → feedback-trajectory-C6NJxXEa.d.ts} +3 -3
- package/dist/{feedback-trajectory-BXCGTRkz.d.ts.map → feedback-trajectory-C6NJxXEa.d.ts.map} +1 -1
- package/dist/{index-CAtNOBQY.d.ts → index-B8n12TR2.d.ts} +12 -14
- package/dist/index-B8n12TR2.d.ts.map +1 -0
- package/dist/{index-DBk_4UPs.d.ts → index-C9l_-23M.d.ts} +7 -7
- package/dist/{index-DBk_4UPs.d.ts.map → index-C9l_-23M.d.ts.map} +1 -1
- package/dist/{index-Cuhxk05B.d.ts → index-CG7mtu_G.d.ts} +8 -8
- package/dist/{index-Cuhxk05B.d.ts.map → index-CG7mtu_G.d.ts.map} +1 -1
- package/dist/index.d.ts +19 -19
- package/dist/index.js +11 -11
- package/dist/{integrity-CaeNeusS.d.ts → integrity-BWO6RPxv.d.ts} +2 -2
- package/dist/{integrity-CaeNeusS.d.ts.map → integrity-BWO6RPxv.d.ts.map} +1 -1
- package/dist/{jev-review-DX6MI_jq.d.ts → jev-review-Cw06AKuF.d.ts} +2 -2
- package/dist/{jev-review-DX6MI_jq.d.ts.map → jev-review-Cw06AKuF.d.ts.map} +1 -1
- package/dist/jev.d.ts +3 -3
- package/dist/{llm-client-Bhqoglar.js → llm-client-DmCNzrc9.js} +77 -60
- package/dist/llm-client-DmCNzrc9.js.map +1 -0
- package/dist/{llm-judge-C-0P1WiT.js → llm-judge-DyZdrp6I.js} +57 -15
- package/dist/llm-judge-DyZdrp6I.js.map +1 -0
- package/dist/{matrix-BucOjLF1.d.ts → matrix-W1lbSx72.d.ts} +2 -2
- package/dist/{matrix-BucOjLF1.d.ts.map → matrix-W1lbSx72.d.ts.map} +1 -1
- package/dist/meta-eval/index.d.ts +1 -1
- package/dist/multishot/golden/index.d.ts +1 -1
- package/dist/multishot/index.d.ts +2 -2
- package/dist/openapi.json +1 -1
- package/dist/{prime-protocol-T0i6xcAs.d.ts → prime-protocol-IoN4Qb6c.d.ts} +2 -2
- package/dist/{prime-protocol-T0i6xcAs.d.ts.map → prime-protocol-IoN4Qb6c.d.ts.map} +1 -1
- package/dist/{produced-state-CR2tAPLf.js → produced-state-DLToUXrD.js} +3 -3
- package/dist/{produced-state-CR2tAPLf.js.map → produced-state-DLToUXrD.js.map} +1 -1
- package/dist/{abort-signal-CtzAM_sJ.js → proposal-findings-CXblsskH.js} +2 -11
- package/dist/proposal-findings-CXblsskH.js.map +1 -0
- package/dist/{registry-Bu3lNole.d.ts → registry-CT-hMSkR.d.ts} +4 -4
- package/dist/{registry-Bu3lNole.d.ts.map → registry-CT-hMSkR.d.ts.map} +1 -1
- package/dist/{researcher-DojEMb7E.d.ts → researcher-zTpxNJBy.d.ts} +3 -3
- package/dist/{researcher-DojEMb7E.d.ts.map → researcher-zTpxNJBy.d.ts.map} +1 -1
- package/dist/rl.d.ts +4 -4
- package/dist/search/index.d.ts +2 -2
- package/dist/{search-kernel-B8RC6BlK.d.ts → search-kernel-DNpsHb8P.d.ts} +2 -2
- package/dist/{search-kernel-B8RC6BlK.d.ts.map → search-kernel-DNpsHb8P.d.ts.map} +1 -1
- package/dist/{semantic-concept-judge-BcjYg37z.js → semantic-concept-judge-BgWxa4L_.js} +4 -4
- package/dist/{semantic-concept-judge-BcjYg37z.js.map → semantic-concept-judge-BgWxa4L_.js.map} +1 -1
- package/dist/{series-convergence-DY7ha0kx.d.ts → series-convergence-DLy0bQMV.d.ts} +2 -2
- package/dist/{series-convergence-DY7ha0kx.d.ts.map → series-convergence-DLy0bQMV.d.ts.map} +1 -1
- package/dist/{server-B4jMhqQn.js → server-CoBZO1fj.js} +2 -2
- package/dist/{server-B4jMhqQn.js.map → server-CoBZO1fj.js.map} +1 -1
- package/dist/{skillopt-optimization-method-Bls9t-G_.js → skillopt-optimization-method-Dz5CGZb8.js} +5 -5
- package/dist/{skillopt-optimization-method-Bls9t-G_.js.map → skillopt-optimization-method-Dz5CGZb8.js.map} +1 -1
- package/dist/{statistical-heldout-DJtQAUZB.d.ts → statistical-heldout-Dq3pnmxd.d.ts} +2 -2
- package/dist/{statistical-heldout-DJtQAUZB.d.ts.map → statistical-heldout-Dq3pnmxd.d.ts.map} +1 -1
- package/dist/{store-tool-spans-fkNF7Q9Y.d.ts → store-tool-spans-CAF3T-BS.d.ts} +5 -5
- package/dist/{store-tool-spans-fkNF7Q9Y.d.ts.map → store-tool-spans-CAF3T-BS.d.ts.map} +1 -1
- package/dist/{store-tool-spans-Biio4_Nw.js → store-tool-spans-D5kOU0mr.js} +2 -2
- package/dist/{store-tool-spans-Biio4_Nw.js.map → store-tool-spans-D5kOU0mr.js.map} +1 -1
- package/dist/{tool-groups-Zm84Pqpk.d.ts → tool-groups-B7_jUPHp.d.ts} +3 -3
- package/dist/tool-groups-B7_jUPHp.d.ts.map +1 -0
- package/dist/{trace-questions--lnJTaou.d.ts → trace-questions-Dd3AMf-U.d.ts} +3 -3
- package/dist/{trace-questions--lnJTaou.d.ts.map → trace-questions-Dd3AMf-U.d.ts.map} +1 -1
- package/dist/trace-repair/index.d.ts +1 -1
- package/dist/traces.d.ts +5 -5
- package/dist/traces.js +2 -2
- package/dist/{types-CeDfCRXp.d.ts → types-6VEehmEg.d.ts} +2 -2
- package/dist/{types-CeDfCRXp.d.ts.map → types-6VEehmEg.d.ts.map} +1 -1
- package/dist/{types-C0bWoSKC.d.ts → types-DLxe96rp.d.ts} +2 -2
- package/dist/{types-C0bWoSKC.d.ts.map → types-DLxe96rp.d.ts.map} +1 -1
- package/dist/{types-B6JadxWL.d.ts → types-DZQpUJse.d.ts} +2 -2
- package/dist/{types-B6JadxWL.d.ts.map → types-DZQpUJse.d.ts.map} +1 -1
- package/dist/{types-khIUQ0O5.d.ts → types-ZUy5HGY5.d.ts} +11 -5
- package/dist/{types-khIUQ0O5.d.ts.map → types-ZUy5HGY5.d.ts.map} +1 -1
- package/dist/wire/index.d.ts +2 -2
- package/dist/wire/index.js +1 -1
- package/docs/changes/llm-physical-attempts.md +30 -0
- package/docs/changes/shared-model-request-lifetime.md +30 -0
- package/docs/concepts.md +8 -0
- package/docs/eval-surface-map.md +12 -2
- package/docs/http-dispatch.md +33 -0
- package/docs/proofs/campaign-attempt-retention-20261001/README.md +70 -0
- 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
- 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
- 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
- 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
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-judge-attribution/attribution_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-judge-attribution/attribution_0/latest-attempt.json +1 -0
- 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
- 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
- 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
- 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
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-legacy/capture-retention_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-legacy/capture-retention_0/captured-response.json +6 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-legacy/capture-retention_0/failure-receipt.json +58 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-legacy/capture-retention_0/latest-attempt.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-legacy/capture-retention_0/spans.jsonl +1 -0
- 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
- 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
- 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
- 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
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/baseline/capture-retention_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/baseline/capture-retention_0/latest-attempt.json +1 -0
- 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
- 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
- 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
- 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
- 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
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/candidates/42b3fec6b7e55008/capture-retention_0/latest-attempt.json +1 -0
- 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
- 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
- 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
- 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
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/candidates/87740cc101602a08/capture-retention_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/candidates/87740cc101602a08/capture-retention_0/latest-attempt.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/first-interruption.json +38 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/blobs/19147243d263e55b1eaf1c940691947382a7cce17c5f49e4a4da765a473d2a55.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/blobs/5dd0d871d559d1c55b2ac04b3fbd26d62f7c9e25d7632d4542502cd292818fc0.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/blobs/5de45d632937e75ff90fec9f1f2ec425904d3632da71f6b26f83b1e1eb8cde5a.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/blobs/6caba627806e499aa3d9e156aec9d5a731a3ef30e29756d343b85b7217c77c11.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/blobs/ce2770851b3b7651f27cedcbad0372b9191e9a9b016665987e3ddd2ad53012b5.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/blobs/d8ac4ba8d9844afb4defa5cd277fe3cb944d99e58cf41c07d3ae199b6f467966.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/blobs/f59a02035fe96ea2f5582dd715d2e3ad8c184ae959310978962f1ac0780753bd.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/ledger.jsonl +23 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-optimization/search/ledger.jsonl.head +1 -0
- 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
- 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
- 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
- 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
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-path-guard/capture-retention_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-path-guard/capture-retention_0/latest-attempt.json +1 -0
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-resume/capture-retention_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-resume/capture-retention_0/latest-attempt.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-resume/observation.json +608 -0
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-retry/capture-retention_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-retry/capture-retention_0/latest-attempt.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-retry/observation.json +514 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/after-summary.json +1853 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/judge-attribution.json +140 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/qualification.log +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/qualify.mjs +165 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduce.mjs +39 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-resume/capture-retention_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-resume/capture-retention_0/captured-response.json +5 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-resume/capture-retention_0/failure-receipt.json +58 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-resume/capture-retention_0/spans.jsonl +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-resume/observer-retained-captures.json +20 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-retry/capture-retention_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-retry/capture-retention_0/captured-response.json +5 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-retry/capture-retention_0/failure-receipt.attempt-1.json +58 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-retry/capture-retention_0/spans.jsonl +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-retry/observer-retained-captures.json +20 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction-summary.json +315 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/reproduction.log +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/source-and-build.json +2409 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/typecheck.log +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/current/verify-package.log +340 -0
- 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
- 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
- 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
- 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
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-legacy/capture-retention_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-legacy/capture-retention_0/captured-response.json +6 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-legacy/capture-retention_0/failure-receipt.json +58 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-legacy/capture-retention_0/latest-attempt.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-legacy/capture-retention_0/spans.jsonl +1 -0
- 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
- 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
- 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
- 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
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/baseline/capture-retention_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/baseline/capture-retention_0/latest-attempt.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/first-interruption.json +5 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/search/blobs/5de45d632937e75ff90fec9f1f2ec425904d3632da71f6b26f83b1e1eb8cde5a.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/search/ledger.jsonl +7 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-optimization/search/ledger.jsonl.head +1 -0
- 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
- 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
- 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
- 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
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-path-guard/capture-retention_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-path-guard/capture-retention_0/latest-attempt.json +1 -0
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-resume/capture-retention_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-resume/capture-retention_0/latest-attempt.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-resume/observation.json +458 -0
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-retry/capture-retention_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-retry/capture-retention_0/latest-attempt.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/after-retry/observation.json +380 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-resume/capture-retention_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-resume/capture-retention_0/captured-response.json +6 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-resume/capture-retention_0/failure-receipt.json +58 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-resume/capture-retention_0/spans.jsonl +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-resume/observation.json +198 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-retry/capture-retention_0/cached-result.json +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-retry/capture-retention_0/captured-response.json +6 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-retry/capture-retention_0/failure-receipt.attempt-1.json +58 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-retry/capture-retention_0/spans.jsonl +1 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-retry/observation.json +121 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/before-summary.json +321 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/build.attribution.log +491 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/build.final.log +491 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/build.log +495 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/qualification.log +18 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/qualification.wrong-root.log +9 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/qualify.mjs +148 -0
- package/docs/proofs/campaign-attempt-retention-20261001/artifacts/predecessor/qualify.wrong-root.mjs +148 -0
- package/docs/proofs/campaign-attempt-retention-20261001/manifest.json +808 -0
- package/docs/proofs/router-cache-usage-20261001/after.json +2058 -0
- package/docs/proofs/router-cache-usage-20261001/component-projections.json +584 -0
- package/docs/proofs/router-cache-usage-20261001/component-projections.nested-only.json +306 -0
- package/docs/proofs/router-cache-usage-20261001/manifest.json +32 -0
- package/docs/proofs/router-cache-usage-20261001/summary.json +49 -0
- package/docs/proofs/router-cache-usage-20261001/summary.nested-only.json +29 -0
- package/package.json +1 -1
- package/dist/abort-signal-CtzAM_sJ.js.map +0 -1
- package/dist/benchmark-command-DFs_i62N.js.map +0 -1
- package/dist/extract-usage-DFATEuWW.js.map +0 -1
- package/dist/index-CAtNOBQY.d.ts.map +0 -1
- package/dist/llm-client-Bhqoglar.js.map +0 -1
- package/dist/llm-judge-C-0P1WiT.js.map +0 -1
- package/dist/tool-groups-Zm84Pqpk.d.ts.map +0 -1
|
@@ -0,0 +1,1853 @@
|
|
|
1
|
+
{
|
|
2
|
+
"module": {
|
|
3
|
+
"path": "/home/drew/code/_wt/eval-router-cache-usage-20261001/dist/campaign/index.js",
|
|
4
|
+
"sha256": "2464ee9209cc4d5aa72fb0f64dd8950e0f5d7a3a59bdab5ad3a3148b4db52061"
|
|
5
|
+
},
|
|
6
|
+
"retry": {
|
|
7
|
+
"invocations": 2,
|
|
8
|
+
"observed": [
|
|
9
|
+
{
|
|
10
|
+
"invocation": 1,
|
|
11
|
+
"path": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-1/artifacts/captured-response.json",
|
|
12
|
+
"sha256": "0b87b7e969e1233f4927f091369327480e998e28ac98bbe97a69423bd282fdac",
|
|
13
|
+
"value": {
|
|
14
|
+
"cellId": "capture-retention:0",
|
|
15
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
16
|
+
"invocation": 1,
|
|
17
|
+
"content": "Actual filesystem capture from invocation 1"
|
|
18
|
+
}
|
|
19
|
+
},
|
|
20
|
+
{
|
|
21
|
+
"invocation": 2,
|
|
22
|
+
"path": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-2/artifacts/captured-response.json",
|
|
23
|
+
"sha256": "74e46c137ea0c465395cd01a36465b52cc5658161344233e62d8bfbbc7b6ebd0",
|
|
24
|
+
"value": {
|
|
25
|
+
"cellId": "capture-retention:0",
|
|
26
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
27
|
+
"invocation": 2,
|
|
28
|
+
"content": "Actual filesystem capture from invocation 2"
|
|
29
|
+
}
|
|
30
|
+
}
|
|
31
|
+
],
|
|
32
|
+
"attempts": [
|
|
33
|
+
{
|
|
34
|
+
"identity": {
|
|
35
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
36
|
+
"number": 1,
|
|
37
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
38
|
+
"cellId": "capture-retention:0",
|
|
39
|
+
"scenarioId": "capture-retention",
|
|
40
|
+
"rep": 0,
|
|
41
|
+
"seed": 42,
|
|
42
|
+
"startedAt": "2026-10-01T17:12:40.120Z"
|
|
43
|
+
},
|
|
44
|
+
"result": {
|
|
45
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
46
|
+
"attempt": {
|
|
47
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
48
|
+
"number": 1
|
|
49
|
+
},
|
|
50
|
+
"cellId": "capture-retention:0",
|
|
51
|
+
"scenarioId": "capture-retention",
|
|
52
|
+
"rep": 0,
|
|
53
|
+
"artifact": null,
|
|
54
|
+
"judgeScores": {},
|
|
55
|
+
"costUsd": 0,
|
|
56
|
+
"costProvenance": {
|
|
57
|
+
"kind": "estimated",
|
|
58
|
+
"usd": 0
|
|
59
|
+
},
|
|
60
|
+
"costCallIds": [
|
|
61
|
+
"c26674f7-275c-4c6c-aceb-8516c1eff8e9"
|
|
62
|
+
],
|
|
63
|
+
"tokenUsage": {
|
|
64
|
+
"input": 0,
|
|
65
|
+
"output": 0
|
|
66
|
+
},
|
|
67
|
+
"resolvedModels": [
|
|
68
|
+
"not-an-inference"
|
|
69
|
+
],
|
|
70
|
+
"resolvedModel": "not-an-inference",
|
|
71
|
+
"durationMs": 6,
|
|
72
|
+
"seed": 42,
|
|
73
|
+
"cached": false,
|
|
74
|
+
"errorStage": "dispatch",
|
|
75
|
+
"error": "Operational interruption after native artifact capture"
|
|
76
|
+
},
|
|
77
|
+
"trace": [
|
|
78
|
+
{
|
|
79
|
+
"name": "native.callback.capture",
|
|
80
|
+
"cellId": "capture-retention:0",
|
|
81
|
+
"startMs": 1790874760121,
|
|
82
|
+
"invocation": 1,
|
|
83
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
84
|
+
"durationMs": 0
|
|
85
|
+
},
|
|
86
|
+
{
|
|
87
|
+
"name": "cost.operational-file-read",
|
|
88
|
+
"cellId": "capture-retention:0",
|
|
89
|
+
"startMs": 1790874760125,
|
|
90
|
+
"amountUsd": 0,
|
|
91
|
+
"durationMs": 0
|
|
92
|
+
}
|
|
93
|
+
],
|
|
94
|
+
"path": "after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-1"
|
|
95
|
+
},
|
|
96
|
+
{
|
|
97
|
+
"identity": {
|
|
98
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
99
|
+
"number": 2,
|
|
100
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
101
|
+
"cellId": "capture-retention:0",
|
|
102
|
+
"scenarioId": "capture-retention",
|
|
103
|
+
"rep": 0,
|
|
104
|
+
"seed": 42,
|
|
105
|
+
"startedAt": "2026-10-01T17:12:40.126Z"
|
|
106
|
+
},
|
|
107
|
+
"result": {
|
|
108
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
109
|
+
"attempt": {
|
|
110
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
111
|
+
"number": 2
|
|
112
|
+
},
|
|
113
|
+
"cellId": "capture-retention:0",
|
|
114
|
+
"scenarioId": "capture-retention",
|
|
115
|
+
"rep": 0,
|
|
116
|
+
"artifact": {
|
|
117
|
+
"cellId": "capture-retention:0",
|
|
118
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
119
|
+
"invocation": 2,
|
|
120
|
+
"content": "Actual filesystem capture from invocation 2"
|
|
121
|
+
},
|
|
122
|
+
"judgeScores": {},
|
|
123
|
+
"costUsd": 0,
|
|
124
|
+
"costProvenance": {
|
|
125
|
+
"kind": "estimated",
|
|
126
|
+
"usd": 0
|
|
127
|
+
},
|
|
128
|
+
"costCallIds": [
|
|
129
|
+
"a409cf65-dba3-4dbd-b9b4-826f8ddd0c67",
|
|
130
|
+
"c26674f7-275c-4c6c-aceb-8516c1eff8e9"
|
|
131
|
+
],
|
|
132
|
+
"tokenUsage": {
|
|
133
|
+
"input": 0,
|
|
134
|
+
"output": 0
|
|
135
|
+
},
|
|
136
|
+
"resolvedModels": [
|
|
137
|
+
"not-an-inference"
|
|
138
|
+
],
|
|
139
|
+
"resolvedModel": "not-an-inference",
|
|
140
|
+
"durationMs": 1,
|
|
141
|
+
"seed": 42,
|
|
142
|
+
"cached": false,
|
|
143
|
+
"retryAttempts": 1
|
|
144
|
+
},
|
|
145
|
+
"trace": [
|
|
146
|
+
{
|
|
147
|
+
"name": "native.callback.capture",
|
|
148
|
+
"cellId": "capture-retention:0",
|
|
149
|
+
"startMs": 1790874760126,
|
|
150
|
+
"invocation": 2,
|
|
151
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
152
|
+
"durationMs": 0
|
|
153
|
+
},
|
|
154
|
+
{
|
|
155
|
+
"name": "cost.operational-file-read",
|
|
156
|
+
"cellId": "capture-retention:0",
|
|
157
|
+
"startMs": 1790874760127,
|
|
158
|
+
"amountUsd": 0,
|
|
159
|
+
"durationMs": 0
|
|
160
|
+
}
|
|
161
|
+
],
|
|
162
|
+
"path": "after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-2"
|
|
163
|
+
}
|
|
164
|
+
],
|
|
165
|
+
"receipt": {
|
|
166
|
+
"schemaVersion": 1,
|
|
167
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
168
|
+
"recordedAt": "2026-10-01T17:12:40.126Z",
|
|
169
|
+
"failure": {
|
|
170
|
+
"stage": "dispatch",
|
|
171
|
+
"error": {
|
|
172
|
+
"name": "Error",
|
|
173
|
+
"message": "Operational interruption after native artifact capture",
|
|
174
|
+
"stack": "Error: Operational interruption after native artifact capture\n at Object.dispatch (file:///mnt/traces/eval-attempt-artifacts-20261001-r2/qualify.mjs:36:33)\n at async executeCell (file:///home/drew/code/_wt/eval-router-cache-usage-20261001/dist/llm-judge-DyZdrp6I.js:491:19)\n at async file:///home/drew/code/_wt/eval-router-cache-usage-20261001/dist/llm-judge-DyZdrp6I.js:923:15\n at async Promise.allSettled (index 0)\n at async runCampaign (file:///home/drew/code/_wt/eval-router-cache-usage-20261001/dist/llm-judge-DyZdrp6I.js:970:22)\n at async qualify (file:///mnt/traces/eval-attempt-artifacts-20261001-r2/qualify.mjs:40:38)\n at async file:///mnt/traces/eval-attempt-artifacts-20261001-r2/qualify.mjs:85:96"
|
|
175
|
+
}
|
|
176
|
+
},
|
|
177
|
+
"cell": {
|
|
178
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
179
|
+
"attempt": {
|
|
180
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
181
|
+
"number": 1
|
|
182
|
+
},
|
|
183
|
+
"cellId": "capture-retention:0",
|
|
184
|
+
"scenarioId": "capture-retention",
|
|
185
|
+
"rep": 0,
|
|
186
|
+
"artifact": null,
|
|
187
|
+
"judgeScores": {},
|
|
188
|
+
"costUsd": 0,
|
|
189
|
+
"costProvenance": {
|
|
190
|
+
"kind": "estimated",
|
|
191
|
+
"usd": 0
|
|
192
|
+
},
|
|
193
|
+
"costCallIds": [
|
|
194
|
+
"c26674f7-275c-4c6c-aceb-8516c1eff8e9"
|
|
195
|
+
],
|
|
196
|
+
"tokenUsage": {
|
|
197
|
+
"input": 0,
|
|
198
|
+
"output": 0
|
|
199
|
+
},
|
|
200
|
+
"resolvedModels": [
|
|
201
|
+
"not-an-inference"
|
|
202
|
+
],
|
|
203
|
+
"resolvedModel": "not-an-inference",
|
|
204
|
+
"durationMs": 6,
|
|
205
|
+
"seed": 42,
|
|
206
|
+
"cached": false,
|
|
207
|
+
"errorStage": "dispatch",
|
|
208
|
+
"error": "Operational interruption after native artifact capture"
|
|
209
|
+
},
|
|
210
|
+
"cost": {
|
|
211
|
+
"totalCalls": 1,
|
|
212
|
+
"pendingCalls": 0,
|
|
213
|
+
"unresolvedCalls": 0,
|
|
214
|
+
"reservedCostUsd": 0,
|
|
215
|
+
"inputTokens": 0,
|
|
216
|
+
"outputTokens": 0,
|
|
217
|
+
"reasoningTokens": 0,
|
|
218
|
+
"cachedTokens": 0,
|
|
219
|
+
"cacheWriteTokens": 0,
|
|
220
|
+
"totalCostUsd": 0,
|
|
221
|
+
"costProvenance": {
|
|
222
|
+
"kind": "estimated",
|
|
223
|
+
"usd": 0
|
|
224
|
+
},
|
|
225
|
+
"byChannel": [
|
|
226
|
+
{
|
|
227
|
+
"channel": "agent",
|
|
228
|
+
"calls": 1,
|
|
229
|
+
"inputTokens": 0,
|
|
230
|
+
"outputTokens": 0,
|
|
231
|
+
"reasoningTokens": 0,
|
|
232
|
+
"cachedTokens": 0,
|
|
233
|
+
"cacheWriteTokens": 0,
|
|
234
|
+
"costUsd": 0,
|
|
235
|
+
"unpricedCalls": 0,
|
|
236
|
+
"unknownUsageCalls": 0
|
|
237
|
+
}
|
|
238
|
+
],
|
|
239
|
+
"unpricedModels": [],
|
|
240
|
+
"fullyPriced": true,
|
|
241
|
+
"usageComplete": true,
|
|
242
|
+
"accountingComplete": true,
|
|
243
|
+
"incompleteReasons": []
|
|
244
|
+
}
|
|
245
|
+
},
|
|
246
|
+
"pointer": {
|
|
247
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
248
|
+
"number": 2
|
|
249
|
+
},
|
|
250
|
+
"reports": [
|
|
251
|
+
{
|
|
252
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
253
|
+
"splitDigest": "sha256:23ea149e81be5435966bd1d185776c7d2e119739830f98a47d4738bdbd46ade6",
|
|
254
|
+
"seed": 42,
|
|
255
|
+
"reps": 1,
|
|
256
|
+
"startedAt": "2026-10-01T17:12:40.119Z",
|
|
257
|
+
"endedAt": "2026-10-01T17:12:40.127Z",
|
|
258
|
+
"durationMs": 8,
|
|
259
|
+
"cells": [
|
|
260
|
+
{
|
|
261
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
262
|
+
"attempt": {
|
|
263
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
264
|
+
"number": 2
|
|
265
|
+
},
|
|
266
|
+
"cellId": "capture-retention:0",
|
|
267
|
+
"scenarioId": "capture-retention",
|
|
268
|
+
"rep": 0,
|
|
269
|
+
"artifact": {
|
|
270
|
+
"cellId": "capture-retention:0",
|
|
271
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
272
|
+
"invocation": 2,
|
|
273
|
+
"content": "Actual filesystem capture from invocation 2"
|
|
274
|
+
},
|
|
275
|
+
"judgeScores": {},
|
|
276
|
+
"costUsd": 0,
|
|
277
|
+
"costProvenance": {
|
|
278
|
+
"kind": "estimated",
|
|
279
|
+
"usd": 0
|
|
280
|
+
},
|
|
281
|
+
"costCallIds": [
|
|
282
|
+
"a409cf65-dba3-4dbd-b9b4-826f8ddd0c67",
|
|
283
|
+
"c26674f7-275c-4c6c-aceb-8516c1eff8e9"
|
|
284
|
+
],
|
|
285
|
+
"tokenUsage": {
|
|
286
|
+
"input": 0,
|
|
287
|
+
"output": 0
|
|
288
|
+
},
|
|
289
|
+
"resolvedModels": [
|
|
290
|
+
"not-an-inference"
|
|
291
|
+
],
|
|
292
|
+
"resolvedModel": "not-an-inference",
|
|
293
|
+
"durationMs": 1,
|
|
294
|
+
"seed": 42,
|
|
295
|
+
"cached": false,
|
|
296
|
+
"retryAttempts": 1
|
|
297
|
+
}
|
|
298
|
+
],
|
|
299
|
+
"aggregates": {
|
|
300
|
+
"byJudge": {},
|
|
301
|
+
"byScenario": {},
|
|
302
|
+
"cost": {
|
|
303
|
+
"totalCalls": 2,
|
|
304
|
+
"pendingCalls": 0,
|
|
305
|
+
"unresolvedCalls": 0,
|
|
306
|
+
"reservedCostUsd": 0,
|
|
307
|
+
"inputTokens": 0,
|
|
308
|
+
"outputTokens": 0,
|
|
309
|
+
"reasoningTokens": 0,
|
|
310
|
+
"cachedTokens": 0,
|
|
311
|
+
"cacheWriteTokens": 0,
|
|
312
|
+
"totalCostUsd": 0,
|
|
313
|
+
"costProvenance": {
|
|
314
|
+
"kind": "estimated",
|
|
315
|
+
"usd": 0
|
|
316
|
+
},
|
|
317
|
+
"byChannel": [
|
|
318
|
+
{
|
|
319
|
+
"channel": "agent",
|
|
320
|
+
"calls": 2,
|
|
321
|
+
"inputTokens": 0,
|
|
322
|
+
"outputTokens": 0,
|
|
323
|
+
"reasoningTokens": 0,
|
|
324
|
+
"cachedTokens": 0,
|
|
325
|
+
"cacheWriteTokens": 0,
|
|
326
|
+
"costUsd": 0,
|
|
327
|
+
"unpricedCalls": 0,
|
|
328
|
+
"unknownUsageCalls": 0
|
|
329
|
+
}
|
|
330
|
+
],
|
|
331
|
+
"unpricedModels": [],
|
|
332
|
+
"fullyPriced": true,
|
|
333
|
+
"usageComplete": true,
|
|
334
|
+
"accountingComplete": true,
|
|
335
|
+
"incompleteReasons": []
|
|
336
|
+
},
|
|
337
|
+
"cellsExecuted": 1,
|
|
338
|
+
"cellsSkipped": 0,
|
|
339
|
+
"cellsCached": 0,
|
|
340
|
+
"cellsFailed": 0,
|
|
341
|
+
"cellsDispatchFailed": 0,
|
|
342
|
+
"cellsJudgeFailed": 0,
|
|
343
|
+
"cellsUnclassifiedFailed": 0
|
|
344
|
+
},
|
|
345
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-retry",
|
|
346
|
+
"artifactsByPath": {
|
|
347
|
+
"capture-retention:0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-1/identity.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-1/identity.json",
|
|
348
|
+
"capture-retention:0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-1/artifacts/captured-response.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-1/artifacts/captured-response.json",
|
|
349
|
+
"capture-retention:0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-1/failure-receipt.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-1/failure-receipt.json",
|
|
350
|
+
"capture-retention:0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-1/result.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-1/result.json",
|
|
351
|
+
"capture-retention:0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-2/identity.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-2/identity.json",
|
|
352
|
+
"capture-retention:0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-2/artifacts/captured-response.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-2/artifacts/captured-response.json",
|
|
353
|
+
"capture-retention:0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-2/result.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-retry/capture-retention_0/attempts/74c13884-d47e-4b6d-a823-5bd44ee0e6cb/attempt-2/result.json"
|
|
354
|
+
},
|
|
355
|
+
"scenarios": [
|
|
356
|
+
{
|
|
357
|
+
"id": "capture-retention",
|
|
358
|
+
"kind": "operational",
|
|
359
|
+
"scenarioDigest": "sha256:70a52f79806918ce9e4833c4ed81360245988fced43a4605ab703d7afacb20a3"
|
|
360
|
+
}
|
|
361
|
+
]
|
|
362
|
+
}
|
|
363
|
+
],
|
|
364
|
+
"ledgerReceipts": [
|
|
365
|
+
{
|
|
366
|
+
"status": "settled",
|
|
367
|
+
"callId": "c26674f7-275c-4c6c-aceb-8516c1eff8e9",
|
|
368
|
+
"channel": "agent",
|
|
369
|
+
"phase": "campaign",
|
|
370
|
+
"actor": "operational-file-read",
|
|
371
|
+
"model": "not-an-inference",
|
|
372
|
+
"timestamp": 1790874760122,
|
|
373
|
+
"inputTokens": 0,
|
|
374
|
+
"outputTokens": 0,
|
|
375
|
+
"costUsd": 0,
|
|
376
|
+
"costUnknown": false,
|
|
377
|
+
"usageUnknown": false,
|
|
378
|
+
"estimatedCostUsd": 0,
|
|
379
|
+
"tags": {
|
|
380
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-retry",
|
|
381
|
+
"cellId": "capture-retention:0",
|
|
382
|
+
"scenarioId": "capture-retention",
|
|
383
|
+
"rep": "0",
|
|
384
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
385
|
+
"attemptNumber": "1"
|
|
386
|
+
}
|
|
387
|
+
},
|
|
388
|
+
{
|
|
389
|
+
"status": "settled",
|
|
390
|
+
"callId": "a409cf65-dba3-4dbd-b9b4-826f8ddd0c67",
|
|
391
|
+
"channel": "agent",
|
|
392
|
+
"phase": "campaign",
|
|
393
|
+
"actor": "operational-file-read",
|
|
394
|
+
"model": "not-an-inference",
|
|
395
|
+
"timestamp": 1790874760126,
|
|
396
|
+
"inputTokens": 0,
|
|
397
|
+
"outputTokens": 0,
|
|
398
|
+
"costUsd": 0,
|
|
399
|
+
"costUnknown": false,
|
|
400
|
+
"usageUnknown": false,
|
|
401
|
+
"estimatedCostUsd": 0,
|
|
402
|
+
"tags": {
|
|
403
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-retry",
|
|
404
|
+
"cellId": "capture-retention:0",
|
|
405
|
+
"scenarioId": "capture-retention",
|
|
406
|
+
"rep": "0",
|
|
407
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
408
|
+
"attemptNumber": "2"
|
|
409
|
+
}
|
|
410
|
+
}
|
|
411
|
+
],
|
|
412
|
+
"cacheReuse": {
|
|
413
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
414
|
+
"splitDigest": "sha256:23ea149e81be5435966bd1d185776c7d2e119739830f98a47d4738bdbd46ade6",
|
|
415
|
+
"seed": 42,
|
|
416
|
+
"reps": 1,
|
|
417
|
+
"startedAt": "2026-10-01T17:12:40.128Z",
|
|
418
|
+
"endedAt": "2026-10-01T17:12:40.128Z",
|
|
419
|
+
"durationMs": 0,
|
|
420
|
+
"cells": [
|
|
421
|
+
{
|
|
422
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
423
|
+
"attempt": {
|
|
424
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
425
|
+
"number": 2
|
|
426
|
+
},
|
|
427
|
+
"cellId": "capture-retention:0",
|
|
428
|
+
"scenarioId": "capture-retention",
|
|
429
|
+
"rep": 0,
|
|
430
|
+
"artifact": {
|
|
431
|
+
"cellId": "capture-retention:0",
|
|
432
|
+
"runAttemptId": "74c13884-d47e-4b6d-a823-5bd44ee0e6cb",
|
|
433
|
+
"invocation": 2,
|
|
434
|
+
"content": "Actual filesystem capture from invocation 2"
|
|
435
|
+
},
|
|
436
|
+
"judgeScores": {},
|
|
437
|
+
"costUsd": 0,
|
|
438
|
+
"costProvenance": {
|
|
439
|
+
"kind": "estimated",
|
|
440
|
+
"usd": 0
|
|
441
|
+
},
|
|
442
|
+
"costCallIds": [
|
|
443
|
+
"a409cf65-dba3-4dbd-b9b4-826f8ddd0c67",
|
|
444
|
+
"c26674f7-275c-4c6c-aceb-8516c1eff8e9"
|
|
445
|
+
],
|
|
446
|
+
"tokenUsage": {
|
|
447
|
+
"input": 0,
|
|
448
|
+
"output": 0
|
|
449
|
+
},
|
|
450
|
+
"durationMs": 1,
|
|
451
|
+
"seed": 42,
|
|
452
|
+
"cached": true,
|
|
453
|
+
"retryAttempts": 1,
|
|
454
|
+
"resolvedModels": [
|
|
455
|
+
"not-an-inference"
|
|
456
|
+
],
|
|
457
|
+
"resolvedModel": "not-an-inference"
|
|
458
|
+
}
|
|
459
|
+
],
|
|
460
|
+
"aggregates": {
|
|
461
|
+
"byJudge": {},
|
|
462
|
+
"byScenario": {},
|
|
463
|
+
"cost": {
|
|
464
|
+
"totalCalls": 2,
|
|
465
|
+
"pendingCalls": 0,
|
|
466
|
+
"unresolvedCalls": 0,
|
|
467
|
+
"reservedCostUsd": 0,
|
|
468
|
+
"inputTokens": 0,
|
|
469
|
+
"outputTokens": 0,
|
|
470
|
+
"reasoningTokens": 0,
|
|
471
|
+
"cachedTokens": 0,
|
|
472
|
+
"cacheWriteTokens": 0,
|
|
473
|
+
"totalCostUsd": 0,
|
|
474
|
+
"costProvenance": {
|
|
475
|
+
"kind": "estimated",
|
|
476
|
+
"usd": 0
|
|
477
|
+
},
|
|
478
|
+
"byChannel": [
|
|
479
|
+
{
|
|
480
|
+
"channel": "agent",
|
|
481
|
+
"calls": 2,
|
|
482
|
+
"inputTokens": 0,
|
|
483
|
+
"outputTokens": 0,
|
|
484
|
+
"reasoningTokens": 0,
|
|
485
|
+
"cachedTokens": 0,
|
|
486
|
+
"cacheWriteTokens": 0,
|
|
487
|
+
"costUsd": 0,
|
|
488
|
+
"unpricedCalls": 0,
|
|
489
|
+
"unknownUsageCalls": 0
|
|
490
|
+
}
|
|
491
|
+
],
|
|
492
|
+
"unpricedModels": [],
|
|
493
|
+
"fullyPriced": true,
|
|
494
|
+
"usageComplete": true,
|
|
495
|
+
"accountingComplete": true,
|
|
496
|
+
"incompleteReasons": []
|
|
497
|
+
},
|
|
498
|
+
"cellsExecuted": 1,
|
|
499
|
+
"cellsSkipped": 0,
|
|
500
|
+
"cellsCached": 1,
|
|
501
|
+
"cellsFailed": 0,
|
|
502
|
+
"cellsDispatchFailed": 0,
|
|
503
|
+
"cellsJudgeFailed": 0,
|
|
504
|
+
"cellsUnclassifiedFailed": 0
|
|
505
|
+
},
|
|
506
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-retry",
|
|
507
|
+
"artifactsByPath": {},
|
|
508
|
+
"scenarios": [
|
|
509
|
+
{
|
|
510
|
+
"id": "capture-retention",
|
|
511
|
+
"kind": "operational",
|
|
512
|
+
"scenarioDigest": "sha256:70a52f79806918ce9e4833c4ed81360245988fced43a4605ab703d7afacb20a3"
|
|
513
|
+
}
|
|
514
|
+
]
|
|
515
|
+
},
|
|
516
|
+
"firstContentStillPresent": true,
|
|
517
|
+
"modelInference": false,
|
|
518
|
+
"researchUnits": 0
|
|
519
|
+
},
|
|
520
|
+
"resume": {
|
|
521
|
+
"invocations": 2,
|
|
522
|
+
"observed": [
|
|
523
|
+
{
|
|
524
|
+
"invocation": 1,
|
|
525
|
+
"path": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume/capture-retention_0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1/artifacts/captured-response.json",
|
|
526
|
+
"sha256": "d088b238a483609d6aaf10a5be899164b69073b92c9cdb48b593170558642d57",
|
|
527
|
+
"value": {
|
|
528
|
+
"cellId": "capture-retention:0",
|
|
529
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
530
|
+
"invocation": 1,
|
|
531
|
+
"content": "Actual filesystem capture from invocation 1"
|
|
532
|
+
}
|
|
533
|
+
},
|
|
534
|
+
{
|
|
535
|
+
"invocation": 2,
|
|
536
|
+
"path": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume/capture-retention_0/attempts/c4859951-1e7c-4d23-b89a-0ca8fa79e2cf/attempt-1/artifacts/captured-response.json",
|
|
537
|
+
"sha256": "d2faa147c54b57960d8b10015d64276ea9dd7613f657f98f8b22fed4850929f7",
|
|
538
|
+
"value": {
|
|
539
|
+
"cellId": "capture-retention:0",
|
|
540
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
541
|
+
"invocation": 2,
|
|
542
|
+
"content": "Actual filesystem capture from invocation 2"
|
|
543
|
+
}
|
|
544
|
+
}
|
|
545
|
+
],
|
|
546
|
+
"attempts": [
|
|
547
|
+
{
|
|
548
|
+
"identity": {
|
|
549
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
550
|
+
"number": 1,
|
|
551
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
552
|
+
"cellId": "capture-retention:0",
|
|
553
|
+
"scenarioId": "capture-retention",
|
|
554
|
+
"rep": 0,
|
|
555
|
+
"seed": 42,
|
|
556
|
+
"startedAt": "2026-10-01T17:12:40.129Z"
|
|
557
|
+
},
|
|
558
|
+
"result": {
|
|
559
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
560
|
+
"attempt": {
|
|
561
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
562
|
+
"number": 1
|
|
563
|
+
},
|
|
564
|
+
"cellId": "capture-retention:0",
|
|
565
|
+
"scenarioId": "capture-retention",
|
|
566
|
+
"rep": 0,
|
|
567
|
+
"artifact": null,
|
|
568
|
+
"judgeScores": {},
|
|
569
|
+
"costUsd": 0,
|
|
570
|
+
"costProvenance": {
|
|
571
|
+
"kind": "estimated",
|
|
572
|
+
"usd": 0
|
|
573
|
+
},
|
|
574
|
+
"costCallIds": [
|
|
575
|
+
"976568ff-db16-46e4-bd3f-dcd3c7664706"
|
|
576
|
+
],
|
|
577
|
+
"tokenUsage": {
|
|
578
|
+
"input": 0,
|
|
579
|
+
"output": 0
|
|
580
|
+
},
|
|
581
|
+
"resolvedModels": [
|
|
582
|
+
"not-an-inference"
|
|
583
|
+
],
|
|
584
|
+
"resolvedModel": "not-an-inference",
|
|
585
|
+
"durationMs": 0,
|
|
586
|
+
"seed": 42,
|
|
587
|
+
"cached": false,
|
|
588
|
+
"errorStage": "dispatch",
|
|
589
|
+
"error": "Operational interruption after native artifact capture"
|
|
590
|
+
},
|
|
591
|
+
"trace": [
|
|
592
|
+
{
|
|
593
|
+
"name": "native.callback.capture",
|
|
594
|
+
"cellId": "capture-retention:0",
|
|
595
|
+
"startMs": 1790874760129,
|
|
596
|
+
"invocation": 1,
|
|
597
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
598
|
+
"durationMs": 0
|
|
599
|
+
},
|
|
600
|
+
{
|
|
601
|
+
"name": "cost.operational-file-read",
|
|
602
|
+
"cellId": "capture-retention:0",
|
|
603
|
+
"startMs": 1790874760129,
|
|
604
|
+
"amountUsd": 0,
|
|
605
|
+
"durationMs": 0
|
|
606
|
+
}
|
|
607
|
+
],
|
|
608
|
+
"path": "after-resume/capture-retention_0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1"
|
|
609
|
+
},
|
|
610
|
+
{
|
|
611
|
+
"identity": {
|
|
612
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
613
|
+
"number": 1,
|
|
614
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
615
|
+
"cellId": "capture-retention:0",
|
|
616
|
+
"scenarioId": "capture-retention",
|
|
617
|
+
"rep": 0,
|
|
618
|
+
"seed": 42,
|
|
619
|
+
"startedAt": "2026-10-01T17:12:40.130Z"
|
|
620
|
+
},
|
|
621
|
+
"result": {
|
|
622
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
623
|
+
"attempt": {
|
|
624
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
625
|
+
"number": 1
|
|
626
|
+
},
|
|
627
|
+
"cellId": "capture-retention:0",
|
|
628
|
+
"scenarioId": "capture-retention",
|
|
629
|
+
"rep": 0,
|
|
630
|
+
"artifact": {
|
|
631
|
+
"cellId": "capture-retention:0",
|
|
632
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
633
|
+
"invocation": 2,
|
|
634
|
+
"content": "Actual filesystem capture from invocation 2"
|
|
635
|
+
},
|
|
636
|
+
"judgeScores": {},
|
|
637
|
+
"costUsd": 0,
|
|
638
|
+
"costProvenance": {
|
|
639
|
+
"kind": "estimated",
|
|
640
|
+
"usd": 0
|
|
641
|
+
},
|
|
642
|
+
"costCallIds": [
|
|
643
|
+
"4de937ba-5b1b-4736-acdf-4e7ee9530d32"
|
|
644
|
+
],
|
|
645
|
+
"tokenUsage": {
|
|
646
|
+
"input": 0,
|
|
647
|
+
"output": 0
|
|
648
|
+
},
|
|
649
|
+
"resolvedModels": [
|
|
650
|
+
"not-an-inference"
|
|
651
|
+
],
|
|
652
|
+
"resolvedModel": "not-an-inference",
|
|
653
|
+
"durationMs": 0,
|
|
654
|
+
"seed": 42,
|
|
655
|
+
"cached": false
|
|
656
|
+
},
|
|
657
|
+
"trace": [
|
|
658
|
+
{
|
|
659
|
+
"name": "native.callback.capture",
|
|
660
|
+
"cellId": "capture-retention:0",
|
|
661
|
+
"startMs": 1790874760130,
|
|
662
|
+
"invocation": 2,
|
|
663
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
664
|
+
"durationMs": 0
|
|
665
|
+
},
|
|
666
|
+
{
|
|
667
|
+
"name": "cost.operational-file-read",
|
|
668
|
+
"cellId": "capture-retention:0",
|
|
669
|
+
"startMs": 1790874760130,
|
|
670
|
+
"amountUsd": 0,
|
|
671
|
+
"durationMs": 0
|
|
672
|
+
}
|
|
673
|
+
],
|
|
674
|
+
"path": "after-resume/capture-retention_0/attempts/c4859951-1e7c-4d23-b89a-0ca8fa79e2cf/attempt-1"
|
|
675
|
+
}
|
|
676
|
+
],
|
|
677
|
+
"receipt": {
|
|
678
|
+
"schemaVersion": 1,
|
|
679
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
680
|
+
"recordedAt": "2026-10-01T17:12:40.129Z",
|
|
681
|
+
"failure": {
|
|
682
|
+
"stage": "dispatch",
|
|
683
|
+
"error": {
|
|
684
|
+
"name": "Error",
|
|
685
|
+
"message": "Operational interruption after native artifact capture",
|
|
686
|
+
"stack": "Error: Operational interruption after native artifact capture\n at Object.dispatch (file:///mnt/traces/eval-attempt-artifacts-20261001-r2/qualify.mjs:36:33)\n at async executeCell (file:///home/drew/code/_wt/eval-router-cache-usage-20261001/dist/llm-judge-DyZdrp6I.js:491:19)\n at async file:///home/drew/code/_wt/eval-router-cache-usage-20261001/dist/llm-judge-DyZdrp6I.js:923:15\n at async Promise.allSettled (index 0)\n at async runCampaign (file:///home/drew/code/_wt/eval-router-cache-usage-20261001/dist/llm-judge-DyZdrp6I.js:970:22)\n at async qualify (file:///mnt/traces/eval-attempt-artifacts-20261001-r2/qualify.mjs:42:18)\n at async file:///mnt/traces/eval-attempt-artifacts-20261001-r2/qualify.mjs:85:128"
|
|
687
|
+
}
|
|
688
|
+
},
|
|
689
|
+
"cell": {
|
|
690
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
691
|
+
"attempt": {
|
|
692
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
693
|
+
"number": 1
|
|
694
|
+
},
|
|
695
|
+
"cellId": "capture-retention:0",
|
|
696
|
+
"scenarioId": "capture-retention",
|
|
697
|
+
"rep": 0,
|
|
698
|
+
"artifact": null,
|
|
699
|
+
"judgeScores": {},
|
|
700
|
+
"costUsd": 0,
|
|
701
|
+
"costProvenance": {
|
|
702
|
+
"kind": "estimated",
|
|
703
|
+
"usd": 0
|
|
704
|
+
},
|
|
705
|
+
"costCallIds": [
|
|
706
|
+
"976568ff-db16-46e4-bd3f-dcd3c7664706"
|
|
707
|
+
],
|
|
708
|
+
"tokenUsage": {
|
|
709
|
+
"input": 0,
|
|
710
|
+
"output": 0
|
|
711
|
+
},
|
|
712
|
+
"resolvedModels": [
|
|
713
|
+
"not-an-inference"
|
|
714
|
+
],
|
|
715
|
+
"resolvedModel": "not-an-inference",
|
|
716
|
+
"durationMs": 0,
|
|
717
|
+
"seed": 42,
|
|
718
|
+
"cached": false,
|
|
719
|
+
"errorStage": "dispatch",
|
|
720
|
+
"error": "Operational interruption after native artifact capture"
|
|
721
|
+
},
|
|
722
|
+
"cost": {
|
|
723
|
+
"totalCalls": 1,
|
|
724
|
+
"pendingCalls": 0,
|
|
725
|
+
"unresolvedCalls": 0,
|
|
726
|
+
"reservedCostUsd": 0,
|
|
727
|
+
"inputTokens": 0,
|
|
728
|
+
"outputTokens": 0,
|
|
729
|
+
"reasoningTokens": 0,
|
|
730
|
+
"cachedTokens": 0,
|
|
731
|
+
"cacheWriteTokens": 0,
|
|
732
|
+
"totalCostUsd": 0,
|
|
733
|
+
"costProvenance": {
|
|
734
|
+
"kind": "estimated",
|
|
735
|
+
"usd": 0
|
|
736
|
+
},
|
|
737
|
+
"byChannel": [
|
|
738
|
+
{
|
|
739
|
+
"channel": "agent",
|
|
740
|
+
"calls": 1,
|
|
741
|
+
"inputTokens": 0,
|
|
742
|
+
"outputTokens": 0,
|
|
743
|
+
"reasoningTokens": 0,
|
|
744
|
+
"cachedTokens": 0,
|
|
745
|
+
"cacheWriteTokens": 0,
|
|
746
|
+
"costUsd": 0,
|
|
747
|
+
"unpricedCalls": 0,
|
|
748
|
+
"unknownUsageCalls": 0
|
|
749
|
+
}
|
|
750
|
+
],
|
|
751
|
+
"unpricedModels": [],
|
|
752
|
+
"fullyPriced": true,
|
|
753
|
+
"usageComplete": true,
|
|
754
|
+
"accountingComplete": true,
|
|
755
|
+
"incompleteReasons": []
|
|
756
|
+
}
|
|
757
|
+
},
|
|
758
|
+
"pointer": {
|
|
759
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
760
|
+
"number": 1
|
|
761
|
+
},
|
|
762
|
+
"reports": [
|
|
763
|
+
{
|
|
764
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
765
|
+
"splitDigest": "sha256:23ea149e81be5435966bd1d185776c7d2e119739830f98a47d4738bdbd46ade6",
|
|
766
|
+
"seed": 42,
|
|
767
|
+
"reps": 1,
|
|
768
|
+
"startedAt": "2026-10-01T17:12:40.129Z",
|
|
769
|
+
"endedAt": "2026-10-01T17:12:40.129Z",
|
|
770
|
+
"durationMs": 0,
|
|
771
|
+
"cells": [
|
|
772
|
+
{
|
|
773
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
774
|
+
"attempt": {
|
|
775
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
776
|
+
"number": 1
|
|
777
|
+
},
|
|
778
|
+
"cellId": "capture-retention:0",
|
|
779
|
+
"scenarioId": "capture-retention",
|
|
780
|
+
"rep": 0,
|
|
781
|
+
"artifact": null,
|
|
782
|
+
"judgeScores": {},
|
|
783
|
+
"costUsd": 0,
|
|
784
|
+
"costProvenance": {
|
|
785
|
+
"kind": "estimated",
|
|
786
|
+
"usd": 0
|
|
787
|
+
},
|
|
788
|
+
"costCallIds": [
|
|
789
|
+
"976568ff-db16-46e4-bd3f-dcd3c7664706"
|
|
790
|
+
],
|
|
791
|
+
"tokenUsage": {
|
|
792
|
+
"input": 0,
|
|
793
|
+
"output": 0
|
|
794
|
+
},
|
|
795
|
+
"resolvedModels": [
|
|
796
|
+
"not-an-inference"
|
|
797
|
+
],
|
|
798
|
+
"resolvedModel": "not-an-inference",
|
|
799
|
+
"durationMs": 0,
|
|
800
|
+
"seed": 42,
|
|
801
|
+
"cached": false,
|
|
802
|
+
"errorStage": "dispatch",
|
|
803
|
+
"error": "Operational interruption after native artifact capture"
|
|
804
|
+
}
|
|
805
|
+
],
|
|
806
|
+
"aggregates": {
|
|
807
|
+
"byJudge": {},
|
|
808
|
+
"byScenario": {},
|
|
809
|
+
"cost": {
|
|
810
|
+
"totalCalls": 1,
|
|
811
|
+
"pendingCalls": 0,
|
|
812
|
+
"unresolvedCalls": 0,
|
|
813
|
+
"reservedCostUsd": 0,
|
|
814
|
+
"inputTokens": 0,
|
|
815
|
+
"outputTokens": 0,
|
|
816
|
+
"reasoningTokens": 0,
|
|
817
|
+
"cachedTokens": 0,
|
|
818
|
+
"cacheWriteTokens": 0,
|
|
819
|
+
"totalCostUsd": 0,
|
|
820
|
+
"costProvenance": {
|
|
821
|
+
"kind": "estimated",
|
|
822
|
+
"usd": 0
|
|
823
|
+
},
|
|
824
|
+
"byChannel": [
|
|
825
|
+
{
|
|
826
|
+
"channel": "agent",
|
|
827
|
+
"calls": 1,
|
|
828
|
+
"inputTokens": 0,
|
|
829
|
+
"outputTokens": 0,
|
|
830
|
+
"reasoningTokens": 0,
|
|
831
|
+
"cachedTokens": 0,
|
|
832
|
+
"cacheWriteTokens": 0,
|
|
833
|
+
"costUsd": 0,
|
|
834
|
+
"unpricedCalls": 0,
|
|
835
|
+
"unknownUsageCalls": 0
|
|
836
|
+
}
|
|
837
|
+
],
|
|
838
|
+
"unpricedModels": [],
|
|
839
|
+
"fullyPriced": true,
|
|
840
|
+
"usageComplete": true,
|
|
841
|
+
"accountingComplete": true,
|
|
842
|
+
"incompleteReasons": []
|
|
843
|
+
},
|
|
844
|
+
"cellsExecuted": 0,
|
|
845
|
+
"cellsSkipped": 0,
|
|
846
|
+
"cellsCached": 0,
|
|
847
|
+
"cellsFailed": 1,
|
|
848
|
+
"cellsDispatchFailed": 1,
|
|
849
|
+
"cellsJudgeFailed": 0,
|
|
850
|
+
"cellsUnclassifiedFailed": 0
|
|
851
|
+
},
|
|
852
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume",
|
|
853
|
+
"artifactsByPath": {
|
|
854
|
+
"capture-retention:0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1/identity.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume/capture-retention_0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1/identity.json",
|
|
855
|
+
"capture-retention:0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1/artifacts/captured-response.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume/capture-retention_0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1/artifacts/captured-response.json",
|
|
856
|
+
"capture-retention:0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1/failure-receipt.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume/capture-retention_0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1/failure-receipt.json",
|
|
857
|
+
"capture-retention:0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1/result.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume/capture-retention_0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1/result.json"
|
|
858
|
+
},
|
|
859
|
+
"scenarios": [
|
|
860
|
+
{
|
|
861
|
+
"id": "capture-retention",
|
|
862
|
+
"kind": "operational",
|
|
863
|
+
"scenarioDigest": "sha256:70a52f79806918ce9e4833c4ed81360245988fced43a4605ab703d7afacb20a3"
|
|
864
|
+
}
|
|
865
|
+
]
|
|
866
|
+
},
|
|
867
|
+
{
|
|
868
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
869
|
+
"splitDigest": "sha256:23ea149e81be5435966bd1d185776c7d2e119739830f98a47d4738bdbd46ade6",
|
|
870
|
+
"seed": 42,
|
|
871
|
+
"reps": 1,
|
|
872
|
+
"startedAt": "2026-10-01T17:12:40.130Z",
|
|
873
|
+
"endedAt": "2026-10-01T17:12:40.130Z",
|
|
874
|
+
"durationMs": 0,
|
|
875
|
+
"cells": [
|
|
876
|
+
{
|
|
877
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
878
|
+
"attempt": {
|
|
879
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
880
|
+
"number": 1
|
|
881
|
+
},
|
|
882
|
+
"cellId": "capture-retention:0",
|
|
883
|
+
"scenarioId": "capture-retention",
|
|
884
|
+
"rep": 0,
|
|
885
|
+
"artifact": {
|
|
886
|
+
"cellId": "capture-retention:0",
|
|
887
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
888
|
+
"invocation": 2,
|
|
889
|
+
"content": "Actual filesystem capture from invocation 2"
|
|
890
|
+
},
|
|
891
|
+
"judgeScores": {},
|
|
892
|
+
"costUsd": 0,
|
|
893
|
+
"costProvenance": {
|
|
894
|
+
"kind": "estimated",
|
|
895
|
+
"usd": 0
|
|
896
|
+
},
|
|
897
|
+
"costCallIds": [
|
|
898
|
+
"4de937ba-5b1b-4736-acdf-4e7ee9530d32"
|
|
899
|
+
],
|
|
900
|
+
"tokenUsage": {
|
|
901
|
+
"input": 0,
|
|
902
|
+
"output": 0
|
|
903
|
+
},
|
|
904
|
+
"resolvedModels": [
|
|
905
|
+
"not-an-inference"
|
|
906
|
+
],
|
|
907
|
+
"resolvedModel": "not-an-inference",
|
|
908
|
+
"durationMs": 0,
|
|
909
|
+
"seed": 42,
|
|
910
|
+
"cached": false
|
|
911
|
+
}
|
|
912
|
+
],
|
|
913
|
+
"aggregates": {
|
|
914
|
+
"byJudge": {},
|
|
915
|
+
"byScenario": {},
|
|
916
|
+
"cost": {
|
|
917
|
+
"totalCalls": 2,
|
|
918
|
+
"pendingCalls": 0,
|
|
919
|
+
"unresolvedCalls": 0,
|
|
920
|
+
"reservedCostUsd": 0,
|
|
921
|
+
"inputTokens": 0,
|
|
922
|
+
"outputTokens": 0,
|
|
923
|
+
"reasoningTokens": 0,
|
|
924
|
+
"cachedTokens": 0,
|
|
925
|
+
"cacheWriteTokens": 0,
|
|
926
|
+
"totalCostUsd": 0,
|
|
927
|
+
"costProvenance": {
|
|
928
|
+
"kind": "estimated",
|
|
929
|
+
"usd": 0
|
|
930
|
+
},
|
|
931
|
+
"byChannel": [
|
|
932
|
+
{
|
|
933
|
+
"channel": "agent",
|
|
934
|
+
"calls": 2,
|
|
935
|
+
"inputTokens": 0,
|
|
936
|
+
"outputTokens": 0,
|
|
937
|
+
"reasoningTokens": 0,
|
|
938
|
+
"cachedTokens": 0,
|
|
939
|
+
"cacheWriteTokens": 0,
|
|
940
|
+
"costUsd": 0,
|
|
941
|
+
"unpricedCalls": 0,
|
|
942
|
+
"unknownUsageCalls": 0
|
|
943
|
+
}
|
|
944
|
+
],
|
|
945
|
+
"unpricedModels": [],
|
|
946
|
+
"fullyPriced": true,
|
|
947
|
+
"usageComplete": true,
|
|
948
|
+
"accountingComplete": true,
|
|
949
|
+
"incompleteReasons": []
|
|
950
|
+
},
|
|
951
|
+
"cellsExecuted": 1,
|
|
952
|
+
"cellsSkipped": 0,
|
|
953
|
+
"cellsCached": 0,
|
|
954
|
+
"cellsFailed": 0,
|
|
955
|
+
"cellsDispatchFailed": 0,
|
|
956
|
+
"cellsJudgeFailed": 0,
|
|
957
|
+
"cellsUnclassifiedFailed": 0
|
|
958
|
+
},
|
|
959
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume",
|
|
960
|
+
"artifactsByPath": {
|
|
961
|
+
"capture-retention:0/attempts/c4859951-1e7c-4d23-b89a-0ca8fa79e2cf/attempt-1/identity.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume/capture-retention_0/attempts/c4859951-1e7c-4d23-b89a-0ca8fa79e2cf/attempt-1/identity.json",
|
|
962
|
+
"capture-retention:0/attempts/c4859951-1e7c-4d23-b89a-0ca8fa79e2cf/attempt-1/artifacts/captured-response.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume/capture-retention_0/attempts/c4859951-1e7c-4d23-b89a-0ca8fa79e2cf/attempt-1/artifacts/captured-response.json",
|
|
963
|
+
"capture-retention:0/attempts/c4859951-1e7c-4d23-b89a-0ca8fa79e2cf/attempt-1/result.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume/capture-retention_0/attempts/c4859951-1e7c-4d23-b89a-0ca8fa79e2cf/attempt-1/result.json"
|
|
964
|
+
},
|
|
965
|
+
"scenarios": [
|
|
966
|
+
{
|
|
967
|
+
"id": "capture-retention",
|
|
968
|
+
"kind": "operational",
|
|
969
|
+
"scenarioDigest": "sha256:70a52f79806918ce9e4833c4ed81360245988fced43a4605ab703d7afacb20a3"
|
|
970
|
+
}
|
|
971
|
+
]
|
|
972
|
+
}
|
|
973
|
+
],
|
|
974
|
+
"ledgerReceipts": [
|
|
975
|
+
{
|
|
976
|
+
"status": "settled",
|
|
977
|
+
"callId": "976568ff-db16-46e4-bd3f-dcd3c7664706",
|
|
978
|
+
"channel": "agent",
|
|
979
|
+
"phase": "campaign",
|
|
980
|
+
"actor": "operational-file-read",
|
|
981
|
+
"model": "not-an-inference",
|
|
982
|
+
"timestamp": 1790874760129,
|
|
983
|
+
"inputTokens": 0,
|
|
984
|
+
"outputTokens": 0,
|
|
985
|
+
"costUsd": 0,
|
|
986
|
+
"costUnknown": false,
|
|
987
|
+
"usageUnknown": false,
|
|
988
|
+
"estimatedCostUsd": 0,
|
|
989
|
+
"tags": {
|
|
990
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume",
|
|
991
|
+
"cellId": "capture-retention:0",
|
|
992
|
+
"scenarioId": "capture-retention",
|
|
993
|
+
"rep": "0",
|
|
994
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
995
|
+
"attemptNumber": "1"
|
|
996
|
+
}
|
|
997
|
+
},
|
|
998
|
+
{
|
|
999
|
+
"status": "settled",
|
|
1000
|
+
"callId": "4de937ba-5b1b-4736-acdf-4e7ee9530d32",
|
|
1001
|
+
"channel": "agent",
|
|
1002
|
+
"phase": "campaign",
|
|
1003
|
+
"actor": "operational-file-read",
|
|
1004
|
+
"model": "not-an-inference",
|
|
1005
|
+
"timestamp": 1790874760130,
|
|
1006
|
+
"inputTokens": 0,
|
|
1007
|
+
"outputTokens": 0,
|
|
1008
|
+
"costUsd": 0,
|
|
1009
|
+
"costUnknown": false,
|
|
1010
|
+
"usageUnknown": false,
|
|
1011
|
+
"estimatedCostUsd": 0,
|
|
1012
|
+
"tags": {
|
|
1013
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume",
|
|
1014
|
+
"cellId": "capture-retention:0",
|
|
1015
|
+
"scenarioId": "capture-retention",
|
|
1016
|
+
"rep": "0",
|
|
1017
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
1018
|
+
"attemptNumber": "1"
|
|
1019
|
+
}
|
|
1020
|
+
}
|
|
1021
|
+
],
|
|
1022
|
+
"cacheReuse": {
|
|
1023
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
1024
|
+
"splitDigest": "sha256:23ea149e81be5435966bd1d185776c7d2e119739830f98a47d4738bdbd46ade6",
|
|
1025
|
+
"seed": 42,
|
|
1026
|
+
"reps": 1,
|
|
1027
|
+
"startedAt": "2026-10-01T17:12:40.131Z",
|
|
1028
|
+
"endedAt": "2026-10-01T17:12:40.131Z",
|
|
1029
|
+
"durationMs": 0,
|
|
1030
|
+
"cells": [
|
|
1031
|
+
{
|
|
1032
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
1033
|
+
"attempt": {
|
|
1034
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
1035
|
+
"number": 1
|
|
1036
|
+
},
|
|
1037
|
+
"cellId": "capture-retention:0",
|
|
1038
|
+
"scenarioId": "capture-retention",
|
|
1039
|
+
"rep": 0,
|
|
1040
|
+
"artifact": {
|
|
1041
|
+
"cellId": "capture-retention:0",
|
|
1042
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
1043
|
+
"invocation": 2,
|
|
1044
|
+
"content": "Actual filesystem capture from invocation 2"
|
|
1045
|
+
},
|
|
1046
|
+
"judgeScores": {},
|
|
1047
|
+
"costUsd": 0,
|
|
1048
|
+
"costProvenance": {
|
|
1049
|
+
"kind": "estimated",
|
|
1050
|
+
"usd": 0
|
|
1051
|
+
},
|
|
1052
|
+
"costCallIds": [
|
|
1053
|
+
"4de937ba-5b1b-4736-acdf-4e7ee9530d32"
|
|
1054
|
+
],
|
|
1055
|
+
"tokenUsage": {
|
|
1056
|
+
"input": 0,
|
|
1057
|
+
"output": 0
|
|
1058
|
+
},
|
|
1059
|
+
"durationMs": 0,
|
|
1060
|
+
"seed": 42,
|
|
1061
|
+
"cached": true,
|
|
1062
|
+
"resolvedModels": [
|
|
1063
|
+
"not-an-inference"
|
|
1064
|
+
],
|
|
1065
|
+
"resolvedModel": "not-an-inference"
|
|
1066
|
+
}
|
|
1067
|
+
],
|
|
1068
|
+
"aggregates": {
|
|
1069
|
+
"byJudge": {},
|
|
1070
|
+
"byScenario": {},
|
|
1071
|
+
"cost": {
|
|
1072
|
+
"totalCalls": 2,
|
|
1073
|
+
"pendingCalls": 0,
|
|
1074
|
+
"unresolvedCalls": 0,
|
|
1075
|
+
"reservedCostUsd": 0,
|
|
1076
|
+
"inputTokens": 0,
|
|
1077
|
+
"outputTokens": 0,
|
|
1078
|
+
"reasoningTokens": 0,
|
|
1079
|
+
"cachedTokens": 0,
|
|
1080
|
+
"cacheWriteTokens": 0,
|
|
1081
|
+
"totalCostUsd": 0,
|
|
1082
|
+
"costProvenance": {
|
|
1083
|
+
"kind": "estimated",
|
|
1084
|
+
"usd": 0
|
|
1085
|
+
},
|
|
1086
|
+
"byChannel": [
|
|
1087
|
+
{
|
|
1088
|
+
"channel": "agent",
|
|
1089
|
+
"calls": 2,
|
|
1090
|
+
"inputTokens": 0,
|
|
1091
|
+
"outputTokens": 0,
|
|
1092
|
+
"reasoningTokens": 0,
|
|
1093
|
+
"cachedTokens": 0,
|
|
1094
|
+
"cacheWriteTokens": 0,
|
|
1095
|
+
"costUsd": 0,
|
|
1096
|
+
"unpricedCalls": 0,
|
|
1097
|
+
"unknownUsageCalls": 0
|
|
1098
|
+
}
|
|
1099
|
+
],
|
|
1100
|
+
"unpricedModels": [],
|
|
1101
|
+
"fullyPriced": true,
|
|
1102
|
+
"usageComplete": true,
|
|
1103
|
+
"accountingComplete": true,
|
|
1104
|
+
"incompleteReasons": []
|
|
1105
|
+
},
|
|
1106
|
+
"cellsExecuted": 1,
|
|
1107
|
+
"cellsSkipped": 0,
|
|
1108
|
+
"cellsCached": 1,
|
|
1109
|
+
"cellsFailed": 0,
|
|
1110
|
+
"cellsDispatchFailed": 0,
|
|
1111
|
+
"cellsJudgeFailed": 0,
|
|
1112
|
+
"cellsUnclassifiedFailed": 0
|
|
1113
|
+
},
|
|
1114
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume",
|
|
1115
|
+
"artifactsByPath": {},
|
|
1116
|
+
"scenarios": [
|
|
1117
|
+
{
|
|
1118
|
+
"id": "capture-retention",
|
|
1119
|
+
"kind": "operational",
|
|
1120
|
+
"scenarioDigest": "sha256:70a52f79806918ce9e4833c4ed81360245988fced43a4605ab703d7afacb20a3"
|
|
1121
|
+
}
|
|
1122
|
+
]
|
|
1123
|
+
},
|
|
1124
|
+
"firstContentStillPresent": true,
|
|
1125
|
+
"modelInference": false,
|
|
1126
|
+
"researchUnits": 0
|
|
1127
|
+
},
|
|
1128
|
+
"legacy": {
|
|
1129
|
+
"originalCell": "/mnt/traces/eval-attempt-artifacts-20261001/before-resume/capture-retention_0",
|
|
1130
|
+
"legacySnapshot": {
|
|
1131
|
+
"cached-result.json": "173d6f94429b9bc4017a61ea0d23e3cd004676661cdbc8b6ec9971f2acd3cc34",
|
|
1132
|
+
"captured-response.json": "bd867bf7291384d653ab799dc2eebf768878ca3a8681fa869684d6da0808cfd7",
|
|
1133
|
+
"failure-receipt.json": "109bd5739ab17e84ead2309caec31e8127399b6b008845d05c1fdf518d4fefa1",
|
|
1134
|
+
"spans.jsonl": "186c6229ad5a7a846d9d92c66c1a3291d5974cbc39eb73eccab68f0659b6ff30"
|
|
1135
|
+
},
|
|
1136
|
+
"report": {
|
|
1137
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
1138
|
+
"splitDigest": "sha256:23ea149e81be5435966bd1d185776c7d2e119739830f98a47d4738bdbd46ade6",
|
|
1139
|
+
"seed": 42,
|
|
1140
|
+
"reps": 1,
|
|
1141
|
+
"startedAt": "2026-10-01T17:12:40.489Z",
|
|
1142
|
+
"endedAt": "2026-10-01T17:12:40.489Z",
|
|
1143
|
+
"durationMs": 0,
|
|
1144
|
+
"cells": [
|
|
1145
|
+
{
|
|
1146
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
1147
|
+
"attempt": {
|
|
1148
|
+
"runAttemptId": "7c7718f6-c85e-4867-9e3a-13ef9513de80",
|
|
1149
|
+
"number": 1
|
|
1150
|
+
},
|
|
1151
|
+
"cellId": "capture-retention:0",
|
|
1152
|
+
"scenarioId": "capture-retention",
|
|
1153
|
+
"rep": 0,
|
|
1154
|
+
"artifact": {
|
|
1155
|
+
"continuation": true
|
|
1156
|
+
},
|
|
1157
|
+
"judgeScores": {},
|
|
1158
|
+
"costUsd": 0,
|
|
1159
|
+
"costProvenance": {
|
|
1160
|
+
"kind": "observed",
|
|
1161
|
+
"usd": 0
|
|
1162
|
+
},
|
|
1163
|
+
"costCallIds": [],
|
|
1164
|
+
"tokenUsage": {
|
|
1165
|
+
"input": 0,
|
|
1166
|
+
"output": 0
|
|
1167
|
+
},
|
|
1168
|
+
"durationMs": 0,
|
|
1169
|
+
"seed": 42,
|
|
1170
|
+
"cached": false
|
|
1171
|
+
}
|
|
1172
|
+
],
|
|
1173
|
+
"aggregates": {
|
|
1174
|
+
"byJudge": {},
|
|
1175
|
+
"byScenario": {},
|
|
1176
|
+
"cost": {
|
|
1177
|
+
"totalCalls": 0,
|
|
1178
|
+
"pendingCalls": 0,
|
|
1179
|
+
"unresolvedCalls": 0,
|
|
1180
|
+
"reservedCostUsd": 0,
|
|
1181
|
+
"inputTokens": 0,
|
|
1182
|
+
"outputTokens": 0,
|
|
1183
|
+
"reasoningTokens": 0,
|
|
1184
|
+
"cachedTokens": 0,
|
|
1185
|
+
"cacheWriteTokens": 0,
|
|
1186
|
+
"totalCostUsd": 0,
|
|
1187
|
+
"costProvenance": {
|
|
1188
|
+
"kind": "observed",
|
|
1189
|
+
"usd": 0
|
|
1190
|
+
},
|
|
1191
|
+
"byChannel": [],
|
|
1192
|
+
"unpricedModels": [],
|
|
1193
|
+
"fullyPriced": true,
|
|
1194
|
+
"usageComplete": true,
|
|
1195
|
+
"accountingComplete": true,
|
|
1196
|
+
"incompleteReasons": []
|
|
1197
|
+
},
|
|
1198
|
+
"cellsExecuted": 1,
|
|
1199
|
+
"cellsSkipped": 0,
|
|
1200
|
+
"cellsCached": 0,
|
|
1201
|
+
"cellsFailed": 0,
|
|
1202
|
+
"cellsDispatchFailed": 0,
|
|
1203
|
+
"cellsJudgeFailed": 0,
|
|
1204
|
+
"cellsUnclassifiedFailed": 0
|
|
1205
|
+
},
|
|
1206
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-legacy",
|
|
1207
|
+
"artifactsByPath": {
|
|
1208
|
+
"capture-retention:0/attempts/7c7718f6-c85e-4867-9e3a-13ef9513de80/attempt-1/identity.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-legacy/capture-retention_0/attempts/7c7718f6-c85e-4867-9e3a-13ef9513de80/attempt-1/identity.json",
|
|
1209
|
+
"capture-retention:0/attempts/7c7718f6-c85e-4867-9e3a-13ef9513de80/attempt-1/artifacts/captured-response.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-legacy/capture-retention_0/attempts/7c7718f6-c85e-4867-9e3a-13ef9513de80/attempt-1/artifacts/captured-response.json",
|
|
1210
|
+
"capture-retention:0/attempts/7c7718f6-c85e-4867-9e3a-13ef9513de80/attempt-1/result.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-legacy/capture-retention_0/attempts/7c7718f6-c85e-4867-9e3a-13ef9513de80/attempt-1/result.json"
|
|
1211
|
+
},
|
|
1212
|
+
"scenarios": [
|
|
1213
|
+
{
|
|
1214
|
+
"id": "capture-retention",
|
|
1215
|
+
"kind": "operational",
|
|
1216
|
+
"scenarioDigest": "sha256:70a52f79806918ce9e4833c4ed81360245988fced43a4605ab703d7afacb20a3"
|
|
1217
|
+
}
|
|
1218
|
+
]
|
|
1219
|
+
},
|
|
1220
|
+
"allHistoricalCellFilesUnchanged": true
|
|
1221
|
+
},
|
|
1222
|
+
"pathGuard": {
|
|
1223
|
+
"rejection": "Artifact path must remain inside this execution attempt",
|
|
1224
|
+
"identityBefore": "68f295075a12c564c0480445de4192b0379cd53fe04de7891dfe3ad934840abf",
|
|
1225
|
+
"identityAfter": "68f295075a12c564c0480445de4192b0379cd53fe04de7891dfe3ad934840abf",
|
|
1226
|
+
"report": {
|
|
1227
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
1228
|
+
"splitDigest": "sha256:23ea149e81be5435966bd1d185776c7d2e119739830f98a47d4738bdbd46ade6",
|
|
1229
|
+
"seed": 42,
|
|
1230
|
+
"reps": 1,
|
|
1231
|
+
"startedAt": "2026-10-01T17:12:40.490Z",
|
|
1232
|
+
"endedAt": "2026-10-01T17:12:40.490Z",
|
|
1233
|
+
"durationMs": 0,
|
|
1234
|
+
"cells": [
|
|
1235
|
+
{
|
|
1236
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
1237
|
+
"attempt": {
|
|
1238
|
+
"runAttemptId": "3c8272d6-d35b-477e-9d08-085261b8dc64",
|
|
1239
|
+
"number": 1
|
|
1240
|
+
},
|
|
1241
|
+
"cellId": "capture-retention:0",
|
|
1242
|
+
"scenarioId": "capture-retention",
|
|
1243
|
+
"rep": 0,
|
|
1244
|
+
"artifact": {
|
|
1245
|
+
"rejection": "Artifact path must remain inside this execution attempt",
|
|
1246
|
+
"identityBefore": "68f295075a12c564c0480445de4192b0379cd53fe04de7891dfe3ad934840abf",
|
|
1247
|
+
"identityAfter": "68f295075a12c564c0480445de4192b0379cd53fe04de7891dfe3ad934840abf"
|
|
1248
|
+
},
|
|
1249
|
+
"judgeScores": {},
|
|
1250
|
+
"costUsd": 0,
|
|
1251
|
+
"costProvenance": {
|
|
1252
|
+
"kind": "observed",
|
|
1253
|
+
"usd": 0
|
|
1254
|
+
},
|
|
1255
|
+
"costCallIds": [],
|
|
1256
|
+
"tokenUsage": {
|
|
1257
|
+
"input": 0,
|
|
1258
|
+
"output": 0
|
|
1259
|
+
},
|
|
1260
|
+
"durationMs": 0,
|
|
1261
|
+
"seed": 42,
|
|
1262
|
+
"cached": false
|
|
1263
|
+
}
|
|
1264
|
+
],
|
|
1265
|
+
"aggregates": {
|
|
1266
|
+
"byJudge": {},
|
|
1267
|
+
"byScenario": {},
|
|
1268
|
+
"cost": {
|
|
1269
|
+
"totalCalls": 0,
|
|
1270
|
+
"pendingCalls": 0,
|
|
1271
|
+
"unresolvedCalls": 0,
|
|
1272
|
+
"reservedCostUsd": 0,
|
|
1273
|
+
"inputTokens": 0,
|
|
1274
|
+
"outputTokens": 0,
|
|
1275
|
+
"reasoningTokens": 0,
|
|
1276
|
+
"cachedTokens": 0,
|
|
1277
|
+
"cacheWriteTokens": 0,
|
|
1278
|
+
"totalCostUsd": 0,
|
|
1279
|
+
"costProvenance": {
|
|
1280
|
+
"kind": "observed",
|
|
1281
|
+
"usd": 0
|
|
1282
|
+
},
|
|
1283
|
+
"byChannel": [],
|
|
1284
|
+
"unpricedModels": [],
|
|
1285
|
+
"fullyPriced": true,
|
|
1286
|
+
"usageComplete": true,
|
|
1287
|
+
"accountingComplete": true,
|
|
1288
|
+
"incompleteReasons": []
|
|
1289
|
+
},
|
|
1290
|
+
"cellsExecuted": 1,
|
|
1291
|
+
"cellsSkipped": 0,
|
|
1292
|
+
"cellsCached": 0,
|
|
1293
|
+
"cellsFailed": 0,
|
|
1294
|
+
"cellsDispatchFailed": 0,
|
|
1295
|
+
"cellsJudgeFailed": 0,
|
|
1296
|
+
"cellsUnclassifiedFailed": 0
|
|
1297
|
+
},
|
|
1298
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-path-guard",
|
|
1299
|
+
"artifactsByPath": {
|
|
1300
|
+
"capture-retention:0/attempts/3c8272d6-d35b-477e-9d08-085261b8dc64/attempt-1/identity.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-path-guard/capture-retention_0/attempts/3c8272d6-d35b-477e-9d08-085261b8dc64/attempt-1/identity.json",
|
|
1301
|
+
"capture-retention:0/attempts/3c8272d6-d35b-477e-9d08-085261b8dc64/attempt-1/artifacts/capture.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-path-guard/capture-retention_0/attempts/3c8272d6-d35b-477e-9d08-085261b8dc64/attempt-1/artifacts/capture.json",
|
|
1302
|
+
"capture-retention:0/attempts/3c8272d6-d35b-477e-9d08-085261b8dc64/attempt-1/result.json": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-path-guard/capture-retention_0/attempts/3c8272d6-d35b-477e-9d08-085261b8dc64/attempt-1/result.json"
|
|
1303
|
+
},
|
|
1304
|
+
"scenarios": [
|
|
1305
|
+
{
|
|
1306
|
+
"id": "capture-retention",
|
|
1307
|
+
"kind": "operational",
|
|
1308
|
+
"scenarioDigest": "sha256:70a52f79806918ce9e4833c4ed81360245988fced43a4605ab703d7afacb20a3"
|
|
1309
|
+
}
|
|
1310
|
+
]
|
|
1311
|
+
}
|
|
1312
|
+
},
|
|
1313
|
+
"optimization": {
|
|
1314
|
+
"firstError": "Operational stop after candidate settlement",
|
|
1315
|
+
"proposals": 3,
|
|
1316
|
+
"histories": [
|
|
1317
|
+
{
|
|
1318
|
+
"proposal": 1,
|
|
1319
|
+
"history": []
|
|
1320
|
+
},
|
|
1321
|
+
{
|
|
1322
|
+
"proposal": 2,
|
|
1323
|
+
"history": [
|
|
1324
|
+
{
|
|
1325
|
+
"generationIndex": 0,
|
|
1326
|
+
"candidates": [
|
|
1327
|
+
{
|
|
1328
|
+
"surfaceHash": "42b3fec6b7e55008",
|
|
1329
|
+
"composite": null,
|
|
1330
|
+
"eligibleForPromotion": false,
|
|
1331
|
+
"coverage": {
|
|
1332
|
+
"expectedCells": 1,
|
|
1333
|
+
"scorableCells": 0,
|
|
1334
|
+
"unscorableCells": [
|
|
1335
|
+
{
|
|
1336
|
+
"cellId": "capture-retention:0",
|
|
1337
|
+
"reason": "Operational candidate interruption; missing artifact"
|
|
1338
|
+
}
|
|
1339
|
+
]
|
|
1340
|
+
},
|
|
1341
|
+
"dimensions": {},
|
|
1342
|
+
"scenarios": []
|
|
1343
|
+
}
|
|
1344
|
+
],
|
|
1345
|
+
"promoted": []
|
|
1346
|
+
}
|
|
1347
|
+
]
|
|
1348
|
+
},
|
|
1349
|
+
{
|
|
1350
|
+
"proposal": 3,
|
|
1351
|
+
"history": [
|
|
1352
|
+
{
|
|
1353
|
+
"generationIndex": 0,
|
|
1354
|
+
"candidates": [
|
|
1355
|
+
{
|
|
1356
|
+
"surfaceHash": "42b3fec6b7e55008",
|
|
1357
|
+
"composite": null,
|
|
1358
|
+
"eligibleForPromotion": false,
|
|
1359
|
+
"coverage": {
|
|
1360
|
+
"expectedCells": 1,
|
|
1361
|
+
"scorableCells": 0,
|
|
1362
|
+
"unscorableCells": [
|
|
1363
|
+
{
|
|
1364
|
+
"cellId": "capture-retention:0",
|
|
1365
|
+
"reason": "Operational candidate interruption; missing artifact"
|
|
1366
|
+
}
|
|
1367
|
+
]
|
|
1368
|
+
},
|
|
1369
|
+
"dimensions": {},
|
|
1370
|
+
"scenarios": []
|
|
1371
|
+
}
|
|
1372
|
+
],
|
|
1373
|
+
"promoted": []
|
|
1374
|
+
}
|
|
1375
|
+
]
|
|
1376
|
+
}
|
|
1377
|
+
],
|
|
1378
|
+
"failedCandidateRecovered": true,
|
|
1379
|
+
"optimized": {
|
|
1380
|
+
"generations": [
|
|
1381
|
+
{
|
|
1382
|
+
"record": {
|
|
1383
|
+
"generationIndex": 0,
|
|
1384
|
+
"candidates": [
|
|
1385
|
+
{
|
|
1386
|
+
"surfaceHash": "42b3fec6b7e55008",
|
|
1387
|
+
"composite": null,
|
|
1388
|
+
"eligibleForPromotion": false,
|
|
1389
|
+
"coverage": {
|
|
1390
|
+
"expectedCells": 1,
|
|
1391
|
+
"scorableCells": 0,
|
|
1392
|
+
"unscorableCells": [
|
|
1393
|
+
{
|
|
1394
|
+
"cellId": "capture-retention:0",
|
|
1395
|
+
"reason": "Operational candidate interruption; missing artifact"
|
|
1396
|
+
}
|
|
1397
|
+
]
|
|
1398
|
+
},
|
|
1399
|
+
"dimensions": {},
|
|
1400
|
+
"scenarios": []
|
|
1401
|
+
}
|
|
1402
|
+
],
|
|
1403
|
+
"promoted": []
|
|
1404
|
+
},
|
|
1405
|
+
"surfaces": [
|
|
1406
|
+
{
|
|
1407
|
+
"surfaceHash": "42b3fec6b7e55008",
|
|
1408
|
+
"surface": "failed-candidate",
|
|
1409
|
+
"campaign": {
|
|
1410
|
+
"manifestHash": "21cc03334665253d4cea184a67430a9c5937ea12a25f1ae823d44dfb5ee168f6",
|
|
1411
|
+
"splitDigest": "sha256:23ea149e81be5435966bd1d185776c7d2e119739830f98a47d4738bdbd46ade6",
|
|
1412
|
+
"seed": 42,
|
|
1413
|
+
"reps": 1,
|
|
1414
|
+
"startedAt": "2026-10-01T17:12:55.571Z",
|
|
1415
|
+
"endedAt": "2026-10-01T17:12:55.571Z",
|
|
1416
|
+
"durationMs": 0,
|
|
1417
|
+
"cells": [
|
|
1418
|
+
{
|
|
1419
|
+
"manifestHash": "21cc03334665253d4cea184a67430a9c5937ea12a25f1ae823d44dfb5ee168f6",
|
|
1420
|
+
"attempt": {
|
|
1421
|
+
"runAttemptId": "40e411fe-ad12-4e6b-a279-2cf07c356a2f",
|
|
1422
|
+
"number": 1
|
|
1423
|
+
},
|
|
1424
|
+
"cellId": "capture-retention:0",
|
|
1425
|
+
"scenarioId": "capture-retention",
|
|
1426
|
+
"rep": 0,
|
|
1427
|
+
"artifact": null,
|
|
1428
|
+
"judgeScores": {},
|
|
1429
|
+
"costUsd": 0,
|
|
1430
|
+
"costProvenance": {
|
|
1431
|
+
"kind": "observed",
|
|
1432
|
+
"usd": 0
|
|
1433
|
+
},
|
|
1434
|
+
"costCallIds": [],
|
|
1435
|
+
"tokenUsage": {
|
|
1436
|
+
"input": 0,
|
|
1437
|
+
"output": 0
|
|
1438
|
+
},
|
|
1439
|
+
"durationMs": 0,
|
|
1440
|
+
"seed": 42,
|
|
1441
|
+
"cached": false,
|
|
1442
|
+
"errorStage": "dispatch",
|
|
1443
|
+
"error": "Operational candidate interruption"
|
|
1444
|
+
}
|
|
1445
|
+
],
|
|
1446
|
+
"aggregates": {
|
|
1447
|
+
"byJudge": {},
|
|
1448
|
+
"byScenario": {},
|
|
1449
|
+
"cost": {
|
|
1450
|
+
"totalCalls": 0,
|
|
1451
|
+
"pendingCalls": 0,
|
|
1452
|
+
"unresolvedCalls": 0,
|
|
1453
|
+
"reservedCostUsd": 0,
|
|
1454
|
+
"inputTokens": 0,
|
|
1455
|
+
"outputTokens": 0,
|
|
1456
|
+
"reasoningTokens": 0,
|
|
1457
|
+
"cachedTokens": 0,
|
|
1458
|
+
"cacheWriteTokens": 0,
|
|
1459
|
+
"totalCostUsd": 0,
|
|
1460
|
+
"costProvenance": {
|
|
1461
|
+
"kind": "observed",
|
|
1462
|
+
"usd": 0
|
|
1463
|
+
},
|
|
1464
|
+
"byChannel": [],
|
|
1465
|
+
"unpricedModels": [],
|
|
1466
|
+
"fullyPriced": true,
|
|
1467
|
+
"usageComplete": true,
|
|
1468
|
+
"accountingComplete": true,
|
|
1469
|
+
"incompleteReasons": []
|
|
1470
|
+
},
|
|
1471
|
+
"cellsExecuted": 0,
|
|
1472
|
+
"cellsSkipped": 0,
|
|
1473
|
+
"cellsCached": 0,
|
|
1474
|
+
"cellsFailed": 1,
|
|
1475
|
+
"cellsDispatchFailed": 1,
|
|
1476
|
+
"cellsJudgeFailed": 0,
|
|
1477
|
+
"cellsUnclassifiedFailed": 0
|
|
1478
|
+
},
|
|
1479
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-optimization/candidates/42b3fec6b7e55008",
|
|
1480
|
+
"artifactsByPath": {},
|
|
1481
|
+
"scenarios": [
|
|
1482
|
+
{
|
|
1483
|
+
"id": "capture-retention",
|
|
1484
|
+
"kind": "operational",
|
|
1485
|
+
"scenarioDigest": "sha256:70a52f79806918ce9e4833c4ed81360245988fced43a4605ab703d7afacb20a3"
|
|
1486
|
+
}
|
|
1487
|
+
]
|
|
1488
|
+
}
|
|
1489
|
+
}
|
|
1490
|
+
]
|
|
1491
|
+
},
|
|
1492
|
+
{
|
|
1493
|
+
"record": {
|
|
1494
|
+
"generationIndex": 1,
|
|
1495
|
+
"candidates": [
|
|
1496
|
+
{
|
|
1497
|
+
"surfaceHash": "87740cc101602a08",
|
|
1498
|
+
"composite": 1,
|
|
1499
|
+
"eligibleForPromotion": true,
|
|
1500
|
+
"coverage": {
|
|
1501
|
+
"expectedCells": 1,
|
|
1502
|
+
"scorableCells": 1,
|
|
1503
|
+
"unscorableCells": []
|
|
1504
|
+
},
|
|
1505
|
+
"dimensions": {
|
|
1506
|
+
"captured": 1
|
|
1507
|
+
},
|
|
1508
|
+
"scenarios": [
|
|
1509
|
+
{
|
|
1510
|
+
"scenarioId": "capture-retention",
|
|
1511
|
+
"composite": 1,
|
|
1512
|
+
"notes": "Operational artifact only; no research outcome."
|
|
1513
|
+
}
|
|
1514
|
+
]
|
|
1515
|
+
}
|
|
1516
|
+
],
|
|
1517
|
+
"promoted": []
|
|
1518
|
+
},
|
|
1519
|
+
"surfaces": [
|
|
1520
|
+
{
|
|
1521
|
+
"surfaceHash": "87740cc101602a08",
|
|
1522
|
+
"surface": "resumed-candidate",
|
|
1523
|
+
"campaign": {
|
|
1524
|
+
"manifestHash": "64c7f454b64ca41db2e7bbc6ffba46d77b541fd334eaaf40401e335b83ff8fd9",
|
|
1525
|
+
"splitDigest": "sha256:23ea149e81be5435966bd1d185776c7d2e119739830f98a47d4738bdbd46ade6",
|
|
1526
|
+
"seed": 42,
|
|
1527
|
+
"reps": 1,
|
|
1528
|
+
"startedAt": "2026-10-01T17:13:03.153Z",
|
|
1529
|
+
"endedAt": "2026-10-01T17:13:03.153Z",
|
|
1530
|
+
"durationMs": 1,
|
|
1531
|
+
"cells": [
|
|
1532
|
+
{
|
|
1533
|
+
"manifestHash": "64c7f454b64ca41db2e7bbc6ffba46d77b541fd334eaaf40401e335b83ff8fd9",
|
|
1534
|
+
"attempt": {
|
|
1535
|
+
"runAttemptId": "9e7da63e-2d19-4ec1-a5ec-ec804e2b0f40",
|
|
1536
|
+
"number": 1
|
|
1537
|
+
},
|
|
1538
|
+
"cellId": "capture-retention:0",
|
|
1539
|
+
"scenarioId": "capture-retention",
|
|
1540
|
+
"rep": 0,
|
|
1541
|
+
"artifact": {
|
|
1542
|
+
"surface": "resumed-candidate"
|
|
1543
|
+
},
|
|
1544
|
+
"judgeScores": {
|
|
1545
|
+
"operational-quality": {
|
|
1546
|
+
"dimensions": {
|
|
1547
|
+
"captured": 1
|
|
1548
|
+
},
|
|
1549
|
+
"composite": 1,
|
|
1550
|
+
"notes": "Operational artifact only; no research outcome."
|
|
1551
|
+
}
|
|
1552
|
+
},
|
|
1553
|
+
"costUsd": 0,
|
|
1554
|
+
"costProvenance": {
|
|
1555
|
+
"kind": "observed",
|
|
1556
|
+
"usd": 0
|
|
1557
|
+
},
|
|
1558
|
+
"costCallIds": [],
|
|
1559
|
+
"tokenUsage": {
|
|
1560
|
+
"input": 0,
|
|
1561
|
+
"output": 0
|
|
1562
|
+
},
|
|
1563
|
+
"durationMs": 1,
|
|
1564
|
+
"seed": 42,
|
|
1565
|
+
"cached": false
|
|
1566
|
+
}
|
|
1567
|
+
],
|
|
1568
|
+
"aggregates": {
|
|
1569
|
+
"byJudge": {
|
|
1570
|
+
"operational-quality": {
|
|
1571
|
+
"mean": 1,
|
|
1572
|
+
"stdev": 0,
|
|
1573
|
+
"ci95": [
|
|
1574
|
+
1,
|
|
1575
|
+
1
|
|
1576
|
+
],
|
|
1577
|
+
"n": 1,
|
|
1578
|
+
"distribution": {
|
|
1579
|
+
"n": 1,
|
|
1580
|
+
"min": 1,
|
|
1581
|
+
"p50": 1,
|
|
1582
|
+
"p90": 1,
|
|
1583
|
+
"max": 1,
|
|
1584
|
+
"sum": 1
|
|
1585
|
+
}
|
|
1586
|
+
}
|
|
1587
|
+
},
|
|
1588
|
+
"byScenario": {
|
|
1589
|
+
"capture-retention": {
|
|
1590
|
+
"meanComposite": 1,
|
|
1591
|
+
"ci95": [
|
|
1592
|
+
1,
|
|
1593
|
+
1
|
|
1594
|
+
],
|
|
1595
|
+
"n": 1,
|
|
1596
|
+
"distribution": {
|
|
1597
|
+
"n": 1,
|
|
1598
|
+
"min": 1,
|
|
1599
|
+
"p50": 1,
|
|
1600
|
+
"p90": 1,
|
|
1601
|
+
"max": 1,
|
|
1602
|
+
"sum": 1
|
|
1603
|
+
}
|
|
1604
|
+
}
|
|
1605
|
+
},
|
|
1606
|
+
"cost": {
|
|
1607
|
+
"totalCalls": 0,
|
|
1608
|
+
"pendingCalls": 0,
|
|
1609
|
+
"unresolvedCalls": 0,
|
|
1610
|
+
"reservedCostUsd": 0,
|
|
1611
|
+
"inputTokens": 0,
|
|
1612
|
+
"outputTokens": 0,
|
|
1613
|
+
"reasoningTokens": 0,
|
|
1614
|
+
"cachedTokens": 0,
|
|
1615
|
+
"cacheWriteTokens": 0,
|
|
1616
|
+
"totalCostUsd": 0,
|
|
1617
|
+
"costProvenance": {
|
|
1618
|
+
"kind": "observed",
|
|
1619
|
+
"usd": 0
|
|
1620
|
+
},
|
|
1621
|
+
"byChannel": [],
|
|
1622
|
+
"unpricedModels": [],
|
|
1623
|
+
"fullyPriced": true,
|
|
1624
|
+
"usageComplete": true,
|
|
1625
|
+
"accountingComplete": true,
|
|
1626
|
+
"incompleteReasons": []
|
|
1627
|
+
},
|
|
1628
|
+
"cellsExecuted": 1,
|
|
1629
|
+
"cellsSkipped": 0,
|
|
1630
|
+
"cellsCached": 0,
|
|
1631
|
+
"cellsFailed": 0,
|
|
1632
|
+
"cellsDispatchFailed": 0,
|
|
1633
|
+
"cellsJudgeFailed": 0,
|
|
1634
|
+
"cellsUnclassifiedFailed": 0
|
|
1635
|
+
},
|
|
1636
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-optimization/candidates/87740cc101602a08",
|
|
1637
|
+
"artifactsByPath": {},
|
|
1638
|
+
"scenarios": [
|
|
1639
|
+
{
|
|
1640
|
+
"id": "capture-retention",
|
|
1641
|
+
"kind": "operational",
|
|
1642
|
+
"scenarioDigest": "sha256:70a52f79806918ce9e4833c4ed81360245988fced43a4605ab703d7afacb20a3"
|
|
1643
|
+
}
|
|
1644
|
+
]
|
|
1645
|
+
}
|
|
1646
|
+
}
|
|
1647
|
+
]
|
|
1648
|
+
}
|
|
1649
|
+
],
|
|
1650
|
+
"baselineSurface": "baseline",
|
|
1651
|
+
"winnerSurface": "baseline",
|
|
1652
|
+
"winnerSurfaceHash": "8ba8496a2525ae17",
|
|
1653
|
+
"baselineCampaign": {
|
|
1654
|
+
"manifestHash": "481089fe6a54224e4b72ba7c72436736d97ffb1cb3599430ef99c426b76f8fc9",
|
|
1655
|
+
"splitDigest": "sha256:23ea149e81be5435966bd1d185776c7d2e119739830f98a47d4738bdbd46ade6",
|
|
1656
|
+
"seed": 42,
|
|
1657
|
+
"reps": 1,
|
|
1658
|
+
"startedAt": "2026-10-01T17:12:55.015Z",
|
|
1659
|
+
"endedAt": "2026-10-01T17:12:55.015Z",
|
|
1660
|
+
"durationMs": 0,
|
|
1661
|
+
"cells": [
|
|
1662
|
+
{
|
|
1663
|
+
"manifestHash": "481089fe6a54224e4b72ba7c72436736d97ffb1cb3599430ef99c426b76f8fc9",
|
|
1664
|
+
"attempt": {
|
|
1665
|
+
"runAttemptId": "3652118b-e1dc-4b17-bc15-6b802dc71eb1",
|
|
1666
|
+
"number": 1
|
|
1667
|
+
},
|
|
1668
|
+
"cellId": "capture-retention:0",
|
|
1669
|
+
"scenarioId": "capture-retention",
|
|
1670
|
+
"rep": 0,
|
|
1671
|
+
"artifact": {
|
|
1672
|
+
"surface": "baseline"
|
|
1673
|
+
},
|
|
1674
|
+
"judgeScores": {
|
|
1675
|
+
"operational-quality": {
|
|
1676
|
+
"dimensions": {
|
|
1677
|
+
"captured": 1
|
|
1678
|
+
},
|
|
1679
|
+
"composite": 1,
|
|
1680
|
+
"notes": "Operational artifact only; no research outcome."
|
|
1681
|
+
}
|
|
1682
|
+
},
|
|
1683
|
+
"costUsd": 0,
|
|
1684
|
+
"costProvenance": {
|
|
1685
|
+
"kind": "observed",
|
|
1686
|
+
"usd": 0
|
|
1687
|
+
},
|
|
1688
|
+
"costCallIds": [],
|
|
1689
|
+
"tokenUsage": {
|
|
1690
|
+
"input": 0,
|
|
1691
|
+
"output": 0
|
|
1692
|
+
},
|
|
1693
|
+
"durationMs": 0,
|
|
1694
|
+
"seed": 42,
|
|
1695
|
+
"cached": false
|
|
1696
|
+
}
|
|
1697
|
+
],
|
|
1698
|
+
"aggregates": {
|
|
1699
|
+
"byJudge": {
|
|
1700
|
+
"operational-quality": {
|
|
1701
|
+
"mean": 1,
|
|
1702
|
+
"stdev": 0,
|
|
1703
|
+
"ci95": [
|
|
1704
|
+
1,
|
|
1705
|
+
1
|
|
1706
|
+
],
|
|
1707
|
+
"n": 1,
|
|
1708
|
+
"distribution": {
|
|
1709
|
+
"n": 1,
|
|
1710
|
+
"min": 1,
|
|
1711
|
+
"p50": 1,
|
|
1712
|
+
"p90": 1,
|
|
1713
|
+
"max": 1,
|
|
1714
|
+
"sum": 1
|
|
1715
|
+
}
|
|
1716
|
+
}
|
|
1717
|
+
},
|
|
1718
|
+
"byScenario": {
|
|
1719
|
+
"capture-retention": {
|
|
1720
|
+
"meanComposite": 1,
|
|
1721
|
+
"ci95": [
|
|
1722
|
+
1,
|
|
1723
|
+
1
|
|
1724
|
+
],
|
|
1725
|
+
"n": 1,
|
|
1726
|
+
"distribution": {
|
|
1727
|
+
"n": 1,
|
|
1728
|
+
"min": 1,
|
|
1729
|
+
"p50": 1,
|
|
1730
|
+
"p90": 1,
|
|
1731
|
+
"max": 1,
|
|
1732
|
+
"sum": 1
|
|
1733
|
+
}
|
|
1734
|
+
}
|
|
1735
|
+
},
|
|
1736
|
+
"cost": {
|
|
1737
|
+
"totalCalls": 0,
|
|
1738
|
+
"pendingCalls": 0,
|
|
1739
|
+
"unresolvedCalls": 0,
|
|
1740
|
+
"reservedCostUsd": 0,
|
|
1741
|
+
"inputTokens": 0,
|
|
1742
|
+
"outputTokens": 0,
|
|
1743
|
+
"reasoningTokens": 0,
|
|
1744
|
+
"cachedTokens": 0,
|
|
1745
|
+
"cacheWriteTokens": 0,
|
|
1746
|
+
"totalCostUsd": 0,
|
|
1747
|
+
"costProvenance": {
|
|
1748
|
+
"kind": "observed",
|
|
1749
|
+
"usd": 0
|
|
1750
|
+
},
|
|
1751
|
+
"byChannel": [],
|
|
1752
|
+
"unpricedModels": [],
|
|
1753
|
+
"fullyPriced": true,
|
|
1754
|
+
"usageComplete": true,
|
|
1755
|
+
"accountingComplete": true,
|
|
1756
|
+
"incompleteReasons": []
|
|
1757
|
+
},
|
|
1758
|
+
"cellsExecuted": 1,
|
|
1759
|
+
"cellsSkipped": 0,
|
|
1760
|
+
"cellsCached": 0,
|
|
1761
|
+
"cellsFailed": 0,
|
|
1762
|
+
"cellsDispatchFailed": 0,
|
|
1763
|
+
"cellsJudgeFailed": 0,
|
|
1764
|
+
"cellsUnclassifiedFailed": 0
|
|
1765
|
+
},
|
|
1766
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-optimization/baseline",
|
|
1767
|
+
"artifactsByPath": {},
|
|
1768
|
+
"scenarios": [
|
|
1769
|
+
{
|
|
1770
|
+
"id": "capture-retention",
|
|
1771
|
+
"kind": "operational",
|
|
1772
|
+
"scenarioDigest": "sha256:70a52f79806918ce9e4833c4ed81360245988fced43a4605ab703d7afacb20a3"
|
|
1773
|
+
}
|
|
1774
|
+
]
|
|
1775
|
+
},
|
|
1776
|
+
"cost": {
|
|
1777
|
+
"totalCalls": 0,
|
|
1778
|
+
"pendingCalls": 0,
|
|
1779
|
+
"unresolvedCalls": 0,
|
|
1780
|
+
"reservedCostUsd": 0,
|
|
1781
|
+
"inputTokens": 0,
|
|
1782
|
+
"outputTokens": 0,
|
|
1783
|
+
"reasoningTokens": 0,
|
|
1784
|
+
"cachedTokens": 0,
|
|
1785
|
+
"cacheWriteTokens": 0,
|
|
1786
|
+
"totalCostUsd": 0,
|
|
1787
|
+
"costProvenance": {
|
|
1788
|
+
"kind": "observed",
|
|
1789
|
+
"usd": 0
|
|
1790
|
+
},
|
|
1791
|
+
"byChannel": [],
|
|
1792
|
+
"unpricedModels": [],
|
|
1793
|
+
"fullyPriced": true,
|
|
1794
|
+
"usageComplete": true,
|
|
1795
|
+
"accountingComplete": true,
|
|
1796
|
+
"incompleteReasons": []
|
|
1797
|
+
},
|
|
1798
|
+
"searchHistory": {
|
|
1799
|
+
"schemaVersion": "2026-09",
|
|
1800
|
+
"kind": "search-history-receipt",
|
|
1801
|
+
"digestAlgorithm": "rfc8785-sha256",
|
|
1802
|
+
"producerId": "operational-retention-qualification",
|
|
1803
|
+
"runId": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-optimization",
|
|
1804
|
+
"ledger": {
|
|
1805
|
+
"role": "search-ledger",
|
|
1806
|
+
"uri": "file:///mnt/traces/eval-attempt-artifacts-20261001-r2/after-optimization/search/ledger.jsonl",
|
|
1807
|
+
"sha256": "sha256:61ab0dd415f9c7a0f028d46a6c21050e368fa65138c990d1c86b6436743a87a2",
|
|
1808
|
+
"byteLength": 22813
|
|
1809
|
+
},
|
|
1810
|
+
"auditDigest": "sha256:c896811e9236800a3604b361d99c88d9738241651e60fca895c2922568d24841",
|
|
1811
|
+
"summary": {
|
|
1812
|
+
"searchId": "optimization-e3a93285eef1a96e",
|
|
1813
|
+
"headHash": "sha256:561a0bcc9cfa56c63e60bd289f13a1e24f9cb69af8d1236fc4ccbdd7d1d3ee58",
|
|
1814
|
+
"status": "closed",
|
|
1815
|
+
"closeReason": "max-nodes",
|
|
1816
|
+
"claimDecision": null,
|
|
1817
|
+
"selectedNodeId": "node_9a513e45f8fd7702584f5d14b4735aaa",
|
|
1818
|
+
"eventCount": 23,
|
|
1819
|
+
"nodes": 3,
|
|
1820
|
+
"edges": 3,
|
|
1821
|
+
"cells": 3,
|
|
1822
|
+
"openCells": 0,
|
|
1823
|
+
"openOperations": 0,
|
|
1824
|
+
"nodesWithoutEdge": 0,
|
|
1825
|
+
"undecidedNodes": 0
|
|
1826
|
+
},
|
|
1827
|
+
"complete": true,
|
|
1828
|
+
"incompleteReasons": [],
|
|
1829
|
+
"receiptDigest": "sha256:f0c989dee5e062bf7fcdab35b1bddd7bce96b3bc284a7399e74ebe10145009fe"
|
|
1830
|
+
},
|
|
1831
|
+
"paretoFrontier": [
|
|
1832
|
+
{
|
|
1833
|
+
"surface": "baseline",
|
|
1834
|
+
"surfaceHash": "8ba8496a2525ae17",
|
|
1835
|
+
"objectives": {
|
|
1836
|
+
"capture-retention": 1
|
|
1837
|
+
},
|
|
1838
|
+
"composite": 1,
|
|
1839
|
+
"generation": -1
|
|
1840
|
+
},
|
|
1841
|
+
{
|
|
1842
|
+
"surface": "resumed-candidate",
|
|
1843
|
+
"surfaceHash": "87740cc101602a08",
|
|
1844
|
+
"objectives": {
|
|
1845
|
+
"capture-retention": 1
|
|
1846
|
+
},
|
|
1847
|
+
"composite": 1,
|
|
1848
|
+
"generation": 1
|
|
1849
|
+
}
|
|
1850
|
+
]
|
|
1851
|
+
}
|
|
1852
|
+
}
|
|
1853
|
+
}
|