@tangle-network/agent-eval 0.203.1 → 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 +4 -0
- package/dist/adapters/http.d.ts +2 -2
- package/dist/analyst/index.d.ts +5 -5
- package/dist/benchmarks/index.d.ts +4 -4
- package/dist/benchmarks/index.js +2 -2
- package/dist/campaign/index.d.ts +4 -4
- package/dist/campaign/index.js +3 -3
- package/dist/{campaign-BbObY2c6.js → campaign-DYhg4SjN.js} +4 -4
- package/dist/{campaign-BbObY2c6.js.map → campaign-DYhg4SjN.js.map} +1 -1
- package/dist/contract/index.d.ts +7 -7
- package/dist/contract/index.js +3 -3
- package/dist/{default-registry-CxEBVA6W.d.ts → default-registry-BHEfZHQd.d.ts} +3 -3
- package/dist/{default-registry-CxEBVA6W.d.ts.map → default-registry-BHEfZHQd.d.ts.map} +1 -1
- package/dist/{define-agent-eval-Da8mHyc7.js → define-agent-eval-B25Cc3ki.js} +2 -2
- package/dist/{define-agent-eval-Da8mHyc7.js.map → define-agent-eval-B25Cc3ki.js.map} +1 -1
- package/dist/{define-agent-eval-BbEXdM3H.d.ts → define-agent-eval-D8bPoBM5.d.ts} +5 -5
- package/dist/{define-agent-eval-BbEXdM3H.d.ts.map → define-agent-eval-D8bPoBM5.d.ts.map} +1 -1
- package/dist/{engine-CaSqEVlL.d.ts → engine-F_PvwesR.d.ts} +2 -2
- package/dist/{engine-CaSqEVlL.d.ts.map → engine-F_PvwesR.d.ts.map} +1 -1
- package/dist/evaluation.d.ts +1 -1
- package/dist/experiment/index.d.ts +1 -1
- package/dist/{index-Ce2HACcP.d.ts → index-B8n12TR2.d.ts} +9 -11
- package/dist/index-B8n12TR2.d.ts.map +1 -0
- package/dist/{index-CIkxRsLS.d.ts → index-C9l_-23M.d.ts} +3 -3
- package/dist/{index-CIkxRsLS.d.ts.map → index-C9l_-23M.d.ts.map} +1 -1
- package/dist/{index-CGAwP9l3.d.ts → index-CG7mtu_G.d.ts} +6 -6
- package/dist/{index-CGAwP9l3.d.ts.map → index-CG7mtu_G.d.ts.map} +1 -1
- package/dist/index.d.ts +9 -9
- package/dist/index.js +3 -3
- package/dist/jev.d.ts +1 -1
- package/dist/{llm-judge-EUMHOZOQ.js → llm-judge-DyZdrp6I.js} +54 -12
- package/dist/llm-judge-DyZdrp6I.js.map +1 -0
- package/dist/{matrix-BqHiMKIj.d.ts → matrix-W1lbSx72.d.ts} +2 -2
- package/dist/{matrix-BqHiMKIj.d.ts.map → matrix-W1lbSx72.d.ts.map} +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/{produced-state-g9BPgu4D.js → produced-state-DLToUXrD.js} +2 -2
- package/dist/{produced-state-g9BPgu4D.js.map → produced-state-DLToUXrD.js.map} +1 -1
- package/dist/rl.d.ts +2 -2
- package/dist/search/index.d.ts +2 -2
- package/dist/{search-kernel-C5jua_UB.d.ts → search-kernel-DNpsHb8P.d.ts} +2 -2
- package/dist/{search-kernel-C5jua_UB.d.ts.map → search-kernel-DNpsHb8P.d.ts.map} +1 -1
- package/dist/{skillopt-optimization-method-DZrdQGZl.js → skillopt-optimization-method-Dz5CGZb8.js} +2 -2
- package/dist/{skillopt-optimization-method-DZrdQGZl.js.map → skillopt-optimization-method-Dz5CGZb8.js.map} +1 -1
- package/dist/{statistical-heldout-CJYAO-4a.d.ts → statistical-heldout-Dq3pnmxd.d.ts} +2 -2
- package/dist/{statistical-heldout-CJYAO-4a.d.ts.map → statistical-heldout-Dq3pnmxd.d.ts.map} +1 -1
- package/dist/{store-tool-spans-fIE31bU8.d.ts → store-tool-spans-CAF3T-BS.d.ts} +3 -3
- package/dist/{store-tool-spans-fIE31bU8.d.ts.map → store-tool-spans-CAF3T-BS.d.ts.map} +1 -1
- package/dist/{tool-groups-DP8b5kjF.d.ts → tool-groups-B7_jUPHp.d.ts} +2 -2
- package/dist/tool-groups-B7_jUPHp.d.ts.map +1 -0
- package/dist/{trace-questions-Cd0Fyhbv.d.ts → trace-questions-Dd3AMf-U.d.ts} +2 -2
- package/dist/{trace-questions-Cd0Fyhbv.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 +2 -2
- package/dist/{types-hH9APahd.d.ts → types-DZQpUJse.d.ts} +2 -2
- package/dist/{types-hH9APahd.d.ts.map → types-DZQpUJse.d.ts.map} +1 -1
- package/dist/{types-Duc6xOoW.d.ts → types-ZUy5HGY5.d.ts} +9 -3
- package/dist/{types-Duc6xOoW.d.ts.map → types-ZUy5HGY5.d.ts.map} +1 -1
- package/docs/eval-surface-map.md +12 -2
- 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/package.json +1 -1
- package/dist/index-Ce2HACcP.d.ts.map +0 -1
- package/dist/llm-judge-EUMHOZOQ.js.map +0 -1
- package/dist/tool-groups-DP8b5kjF.d.ts.map +0 -1
|
@@ -0,0 +1,608 @@
|
|
|
1
|
+
{
|
|
2
|
+
"invocations": 2,
|
|
3
|
+
"observed": [
|
|
4
|
+
{
|
|
5
|
+
"invocation": 1,
|
|
6
|
+
"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",
|
|
7
|
+
"sha256": "d088b238a483609d6aaf10a5be899164b69073b92c9cdb48b593170558642d57",
|
|
8
|
+
"value": {
|
|
9
|
+
"cellId": "capture-retention:0",
|
|
10
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
11
|
+
"invocation": 1,
|
|
12
|
+
"content": "Actual filesystem capture from invocation 1"
|
|
13
|
+
}
|
|
14
|
+
},
|
|
15
|
+
{
|
|
16
|
+
"invocation": 2,
|
|
17
|
+
"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",
|
|
18
|
+
"sha256": "d2faa147c54b57960d8b10015d64276ea9dd7613f657f98f8b22fed4850929f7",
|
|
19
|
+
"value": {
|
|
20
|
+
"cellId": "capture-retention:0",
|
|
21
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
22
|
+
"invocation": 2,
|
|
23
|
+
"content": "Actual filesystem capture from invocation 2"
|
|
24
|
+
}
|
|
25
|
+
}
|
|
26
|
+
],
|
|
27
|
+
"attempts": [
|
|
28
|
+
{
|
|
29
|
+
"identity": {
|
|
30
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
31
|
+
"number": 1,
|
|
32
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
33
|
+
"cellId": "capture-retention:0",
|
|
34
|
+
"scenarioId": "capture-retention",
|
|
35
|
+
"rep": 0,
|
|
36
|
+
"seed": 42,
|
|
37
|
+
"startedAt": "2026-10-01T17:12:40.129Z"
|
|
38
|
+
},
|
|
39
|
+
"result": {
|
|
40
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
41
|
+
"attempt": {
|
|
42
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
43
|
+
"number": 1
|
|
44
|
+
},
|
|
45
|
+
"cellId": "capture-retention:0",
|
|
46
|
+
"scenarioId": "capture-retention",
|
|
47
|
+
"rep": 0,
|
|
48
|
+
"artifact": null,
|
|
49
|
+
"judgeScores": {},
|
|
50
|
+
"costUsd": 0,
|
|
51
|
+
"costProvenance": {
|
|
52
|
+
"kind": "estimated",
|
|
53
|
+
"usd": 0
|
|
54
|
+
},
|
|
55
|
+
"costCallIds": [
|
|
56
|
+
"976568ff-db16-46e4-bd3f-dcd3c7664706"
|
|
57
|
+
],
|
|
58
|
+
"tokenUsage": {
|
|
59
|
+
"input": 0,
|
|
60
|
+
"output": 0
|
|
61
|
+
},
|
|
62
|
+
"resolvedModels": [
|
|
63
|
+
"not-an-inference"
|
|
64
|
+
],
|
|
65
|
+
"resolvedModel": "not-an-inference",
|
|
66
|
+
"durationMs": 0,
|
|
67
|
+
"seed": 42,
|
|
68
|
+
"cached": false,
|
|
69
|
+
"errorStage": "dispatch",
|
|
70
|
+
"error": "Operational interruption after native artifact capture"
|
|
71
|
+
},
|
|
72
|
+
"trace": [
|
|
73
|
+
{
|
|
74
|
+
"name": "native.callback.capture",
|
|
75
|
+
"cellId": "capture-retention:0",
|
|
76
|
+
"startMs": 1790874760129,
|
|
77
|
+
"invocation": 1,
|
|
78
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
79
|
+
"durationMs": 0
|
|
80
|
+
},
|
|
81
|
+
{
|
|
82
|
+
"name": "cost.operational-file-read",
|
|
83
|
+
"cellId": "capture-retention:0",
|
|
84
|
+
"startMs": 1790874760129,
|
|
85
|
+
"amountUsd": 0,
|
|
86
|
+
"durationMs": 0
|
|
87
|
+
}
|
|
88
|
+
],
|
|
89
|
+
"path": "after-resume/capture-retention_0/attempts/2b022ccb-adb3-4468-8c21-48ee82d0634a/attempt-1"
|
|
90
|
+
},
|
|
91
|
+
{
|
|
92
|
+
"identity": {
|
|
93
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
94
|
+
"number": 1,
|
|
95
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
96
|
+
"cellId": "capture-retention:0",
|
|
97
|
+
"scenarioId": "capture-retention",
|
|
98
|
+
"rep": 0,
|
|
99
|
+
"seed": 42,
|
|
100
|
+
"startedAt": "2026-10-01T17:12:40.130Z"
|
|
101
|
+
},
|
|
102
|
+
"result": {
|
|
103
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
104
|
+
"attempt": {
|
|
105
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
106
|
+
"number": 1
|
|
107
|
+
},
|
|
108
|
+
"cellId": "capture-retention:0",
|
|
109
|
+
"scenarioId": "capture-retention",
|
|
110
|
+
"rep": 0,
|
|
111
|
+
"artifact": {
|
|
112
|
+
"cellId": "capture-retention:0",
|
|
113
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
114
|
+
"invocation": 2,
|
|
115
|
+
"content": "Actual filesystem capture from invocation 2"
|
|
116
|
+
},
|
|
117
|
+
"judgeScores": {},
|
|
118
|
+
"costUsd": 0,
|
|
119
|
+
"costProvenance": {
|
|
120
|
+
"kind": "estimated",
|
|
121
|
+
"usd": 0
|
|
122
|
+
},
|
|
123
|
+
"costCallIds": [
|
|
124
|
+
"4de937ba-5b1b-4736-acdf-4e7ee9530d32"
|
|
125
|
+
],
|
|
126
|
+
"tokenUsage": {
|
|
127
|
+
"input": 0,
|
|
128
|
+
"output": 0
|
|
129
|
+
},
|
|
130
|
+
"resolvedModels": [
|
|
131
|
+
"not-an-inference"
|
|
132
|
+
],
|
|
133
|
+
"resolvedModel": "not-an-inference",
|
|
134
|
+
"durationMs": 0,
|
|
135
|
+
"seed": 42,
|
|
136
|
+
"cached": false
|
|
137
|
+
},
|
|
138
|
+
"trace": [
|
|
139
|
+
{
|
|
140
|
+
"name": "native.callback.capture",
|
|
141
|
+
"cellId": "capture-retention:0",
|
|
142
|
+
"startMs": 1790874760130,
|
|
143
|
+
"invocation": 2,
|
|
144
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
145
|
+
"durationMs": 0
|
|
146
|
+
},
|
|
147
|
+
{
|
|
148
|
+
"name": "cost.operational-file-read",
|
|
149
|
+
"cellId": "capture-retention:0",
|
|
150
|
+
"startMs": 1790874760130,
|
|
151
|
+
"amountUsd": 0,
|
|
152
|
+
"durationMs": 0
|
|
153
|
+
}
|
|
154
|
+
],
|
|
155
|
+
"path": "after-resume/capture-retention_0/attempts/c4859951-1e7c-4d23-b89a-0ca8fa79e2cf/attempt-1"
|
|
156
|
+
}
|
|
157
|
+
],
|
|
158
|
+
"receipt": {
|
|
159
|
+
"schemaVersion": 1,
|
|
160
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
161
|
+
"recordedAt": "2026-10-01T17:12:40.129Z",
|
|
162
|
+
"failure": {
|
|
163
|
+
"stage": "dispatch",
|
|
164
|
+
"error": {
|
|
165
|
+
"name": "Error",
|
|
166
|
+
"message": "Operational interruption after native artifact capture",
|
|
167
|
+
"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"
|
|
168
|
+
}
|
|
169
|
+
},
|
|
170
|
+
"cell": {
|
|
171
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
172
|
+
"attempt": {
|
|
173
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
174
|
+
"number": 1
|
|
175
|
+
},
|
|
176
|
+
"cellId": "capture-retention:0",
|
|
177
|
+
"scenarioId": "capture-retention",
|
|
178
|
+
"rep": 0,
|
|
179
|
+
"artifact": null,
|
|
180
|
+
"judgeScores": {},
|
|
181
|
+
"costUsd": 0,
|
|
182
|
+
"costProvenance": {
|
|
183
|
+
"kind": "estimated",
|
|
184
|
+
"usd": 0
|
|
185
|
+
},
|
|
186
|
+
"costCallIds": [
|
|
187
|
+
"976568ff-db16-46e4-bd3f-dcd3c7664706"
|
|
188
|
+
],
|
|
189
|
+
"tokenUsage": {
|
|
190
|
+
"input": 0,
|
|
191
|
+
"output": 0
|
|
192
|
+
},
|
|
193
|
+
"resolvedModels": [
|
|
194
|
+
"not-an-inference"
|
|
195
|
+
],
|
|
196
|
+
"resolvedModel": "not-an-inference",
|
|
197
|
+
"durationMs": 0,
|
|
198
|
+
"seed": 42,
|
|
199
|
+
"cached": false,
|
|
200
|
+
"errorStage": "dispatch",
|
|
201
|
+
"error": "Operational interruption after native artifact capture"
|
|
202
|
+
},
|
|
203
|
+
"cost": {
|
|
204
|
+
"totalCalls": 1,
|
|
205
|
+
"pendingCalls": 0,
|
|
206
|
+
"unresolvedCalls": 0,
|
|
207
|
+
"reservedCostUsd": 0,
|
|
208
|
+
"inputTokens": 0,
|
|
209
|
+
"outputTokens": 0,
|
|
210
|
+
"reasoningTokens": 0,
|
|
211
|
+
"cachedTokens": 0,
|
|
212
|
+
"cacheWriteTokens": 0,
|
|
213
|
+
"totalCostUsd": 0,
|
|
214
|
+
"costProvenance": {
|
|
215
|
+
"kind": "estimated",
|
|
216
|
+
"usd": 0
|
|
217
|
+
},
|
|
218
|
+
"byChannel": [
|
|
219
|
+
{
|
|
220
|
+
"channel": "agent",
|
|
221
|
+
"calls": 1,
|
|
222
|
+
"inputTokens": 0,
|
|
223
|
+
"outputTokens": 0,
|
|
224
|
+
"reasoningTokens": 0,
|
|
225
|
+
"cachedTokens": 0,
|
|
226
|
+
"cacheWriteTokens": 0,
|
|
227
|
+
"costUsd": 0,
|
|
228
|
+
"unpricedCalls": 0,
|
|
229
|
+
"unknownUsageCalls": 0
|
|
230
|
+
}
|
|
231
|
+
],
|
|
232
|
+
"unpricedModels": [],
|
|
233
|
+
"fullyPriced": true,
|
|
234
|
+
"usageComplete": true,
|
|
235
|
+
"accountingComplete": true,
|
|
236
|
+
"incompleteReasons": []
|
|
237
|
+
}
|
|
238
|
+
},
|
|
239
|
+
"pointer": {
|
|
240
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
241
|
+
"number": 1
|
|
242
|
+
},
|
|
243
|
+
"reports": [
|
|
244
|
+
{
|
|
245
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
246
|
+
"splitDigest": "sha256:23ea149e81be5435966bd1d185776c7d2e119739830f98a47d4738bdbd46ade6",
|
|
247
|
+
"seed": 42,
|
|
248
|
+
"reps": 1,
|
|
249
|
+
"startedAt": "2026-10-01T17:12:40.129Z",
|
|
250
|
+
"endedAt": "2026-10-01T17:12:40.129Z",
|
|
251
|
+
"durationMs": 0,
|
|
252
|
+
"cells": [
|
|
253
|
+
{
|
|
254
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
255
|
+
"attempt": {
|
|
256
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
257
|
+
"number": 1
|
|
258
|
+
},
|
|
259
|
+
"cellId": "capture-retention:0",
|
|
260
|
+
"scenarioId": "capture-retention",
|
|
261
|
+
"rep": 0,
|
|
262
|
+
"artifact": null,
|
|
263
|
+
"judgeScores": {},
|
|
264
|
+
"costUsd": 0,
|
|
265
|
+
"costProvenance": {
|
|
266
|
+
"kind": "estimated",
|
|
267
|
+
"usd": 0
|
|
268
|
+
},
|
|
269
|
+
"costCallIds": [
|
|
270
|
+
"976568ff-db16-46e4-bd3f-dcd3c7664706"
|
|
271
|
+
],
|
|
272
|
+
"tokenUsage": {
|
|
273
|
+
"input": 0,
|
|
274
|
+
"output": 0
|
|
275
|
+
},
|
|
276
|
+
"resolvedModels": [
|
|
277
|
+
"not-an-inference"
|
|
278
|
+
],
|
|
279
|
+
"resolvedModel": "not-an-inference",
|
|
280
|
+
"durationMs": 0,
|
|
281
|
+
"seed": 42,
|
|
282
|
+
"cached": false,
|
|
283
|
+
"errorStage": "dispatch",
|
|
284
|
+
"error": "Operational interruption after native artifact capture"
|
|
285
|
+
}
|
|
286
|
+
],
|
|
287
|
+
"aggregates": {
|
|
288
|
+
"byJudge": {},
|
|
289
|
+
"byScenario": {},
|
|
290
|
+
"cost": {
|
|
291
|
+
"totalCalls": 1,
|
|
292
|
+
"pendingCalls": 0,
|
|
293
|
+
"unresolvedCalls": 0,
|
|
294
|
+
"reservedCostUsd": 0,
|
|
295
|
+
"inputTokens": 0,
|
|
296
|
+
"outputTokens": 0,
|
|
297
|
+
"reasoningTokens": 0,
|
|
298
|
+
"cachedTokens": 0,
|
|
299
|
+
"cacheWriteTokens": 0,
|
|
300
|
+
"totalCostUsd": 0,
|
|
301
|
+
"costProvenance": {
|
|
302
|
+
"kind": "estimated",
|
|
303
|
+
"usd": 0
|
|
304
|
+
},
|
|
305
|
+
"byChannel": [
|
|
306
|
+
{
|
|
307
|
+
"channel": "agent",
|
|
308
|
+
"calls": 1,
|
|
309
|
+
"inputTokens": 0,
|
|
310
|
+
"outputTokens": 0,
|
|
311
|
+
"reasoningTokens": 0,
|
|
312
|
+
"cachedTokens": 0,
|
|
313
|
+
"cacheWriteTokens": 0,
|
|
314
|
+
"costUsd": 0,
|
|
315
|
+
"unpricedCalls": 0,
|
|
316
|
+
"unknownUsageCalls": 0
|
|
317
|
+
}
|
|
318
|
+
],
|
|
319
|
+
"unpricedModels": [],
|
|
320
|
+
"fullyPriced": true,
|
|
321
|
+
"usageComplete": true,
|
|
322
|
+
"accountingComplete": true,
|
|
323
|
+
"incompleteReasons": []
|
|
324
|
+
},
|
|
325
|
+
"cellsExecuted": 0,
|
|
326
|
+
"cellsSkipped": 0,
|
|
327
|
+
"cellsCached": 0,
|
|
328
|
+
"cellsFailed": 1,
|
|
329
|
+
"cellsDispatchFailed": 1,
|
|
330
|
+
"cellsJudgeFailed": 0,
|
|
331
|
+
"cellsUnclassifiedFailed": 0
|
|
332
|
+
},
|
|
333
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume",
|
|
334
|
+
"artifactsByPath": {
|
|
335
|
+
"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",
|
|
336
|
+
"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",
|
|
337
|
+
"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",
|
|
338
|
+
"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"
|
|
339
|
+
},
|
|
340
|
+
"scenarios": [
|
|
341
|
+
{
|
|
342
|
+
"id": "capture-retention",
|
|
343
|
+
"kind": "operational",
|
|
344
|
+
"scenarioDigest": "sha256:70a52f79806918ce9e4833c4ed81360245988fced43a4605ab703d7afacb20a3"
|
|
345
|
+
}
|
|
346
|
+
]
|
|
347
|
+
},
|
|
348
|
+
{
|
|
349
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
350
|
+
"splitDigest": "sha256:23ea149e81be5435966bd1d185776c7d2e119739830f98a47d4738bdbd46ade6",
|
|
351
|
+
"seed": 42,
|
|
352
|
+
"reps": 1,
|
|
353
|
+
"startedAt": "2026-10-01T17:12:40.130Z",
|
|
354
|
+
"endedAt": "2026-10-01T17:12:40.130Z",
|
|
355
|
+
"durationMs": 0,
|
|
356
|
+
"cells": [
|
|
357
|
+
{
|
|
358
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
359
|
+
"attempt": {
|
|
360
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
361
|
+
"number": 1
|
|
362
|
+
},
|
|
363
|
+
"cellId": "capture-retention:0",
|
|
364
|
+
"scenarioId": "capture-retention",
|
|
365
|
+
"rep": 0,
|
|
366
|
+
"artifact": {
|
|
367
|
+
"cellId": "capture-retention:0",
|
|
368
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
369
|
+
"invocation": 2,
|
|
370
|
+
"content": "Actual filesystem capture from invocation 2"
|
|
371
|
+
},
|
|
372
|
+
"judgeScores": {},
|
|
373
|
+
"costUsd": 0,
|
|
374
|
+
"costProvenance": {
|
|
375
|
+
"kind": "estimated",
|
|
376
|
+
"usd": 0
|
|
377
|
+
},
|
|
378
|
+
"costCallIds": [
|
|
379
|
+
"4de937ba-5b1b-4736-acdf-4e7ee9530d32"
|
|
380
|
+
],
|
|
381
|
+
"tokenUsage": {
|
|
382
|
+
"input": 0,
|
|
383
|
+
"output": 0
|
|
384
|
+
},
|
|
385
|
+
"resolvedModels": [
|
|
386
|
+
"not-an-inference"
|
|
387
|
+
],
|
|
388
|
+
"resolvedModel": "not-an-inference",
|
|
389
|
+
"durationMs": 0,
|
|
390
|
+
"seed": 42,
|
|
391
|
+
"cached": false
|
|
392
|
+
}
|
|
393
|
+
],
|
|
394
|
+
"aggregates": {
|
|
395
|
+
"byJudge": {},
|
|
396
|
+
"byScenario": {},
|
|
397
|
+
"cost": {
|
|
398
|
+
"totalCalls": 2,
|
|
399
|
+
"pendingCalls": 0,
|
|
400
|
+
"unresolvedCalls": 0,
|
|
401
|
+
"reservedCostUsd": 0,
|
|
402
|
+
"inputTokens": 0,
|
|
403
|
+
"outputTokens": 0,
|
|
404
|
+
"reasoningTokens": 0,
|
|
405
|
+
"cachedTokens": 0,
|
|
406
|
+
"cacheWriteTokens": 0,
|
|
407
|
+
"totalCostUsd": 0,
|
|
408
|
+
"costProvenance": {
|
|
409
|
+
"kind": "estimated",
|
|
410
|
+
"usd": 0
|
|
411
|
+
},
|
|
412
|
+
"byChannel": [
|
|
413
|
+
{
|
|
414
|
+
"channel": "agent",
|
|
415
|
+
"calls": 2,
|
|
416
|
+
"inputTokens": 0,
|
|
417
|
+
"outputTokens": 0,
|
|
418
|
+
"reasoningTokens": 0,
|
|
419
|
+
"cachedTokens": 0,
|
|
420
|
+
"cacheWriteTokens": 0,
|
|
421
|
+
"costUsd": 0,
|
|
422
|
+
"unpricedCalls": 0,
|
|
423
|
+
"unknownUsageCalls": 0
|
|
424
|
+
}
|
|
425
|
+
],
|
|
426
|
+
"unpricedModels": [],
|
|
427
|
+
"fullyPriced": true,
|
|
428
|
+
"usageComplete": true,
|
|
429
|
+
"accountingComplete": true,
|
|
430
|
+
"incompleteReasons": []
|
|
431
|
+
},
|
|
432
|
+
"cellsExecuted": 1,
|
|
433
|
+
"cellsSkipped": 0,
|
|
434
|
+
"cellsCached": 0,
|
|
435
|
+
"cellsFailed": 0,
|
|
436
|
+
"cellsDispatchFailed": 0,
|
|
437
|
+
"cellsJudgeFailed": 0,
|
|
438
|
+
"cellsUnclassifiedFailed": 0
|
|
439
|
+
},
|
|
440
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume",
|
|
441
|
+
"artifactsByPath": {
|
|
442
|
+
"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",
|
|
443
|
+
"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",
|
|
444
|
+
"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"
|
|
445
|
+
},
|
|
446
|
+
"scenarios": [
|
|
447
|
+
{
|
|
448
|
+
"id": "capture-retention",
|
|
449
|
+
"kind": "operational",
|
|
450
|
+
"scenarioDigest": "sha256:70a52f79806918ce9e4833c4ed81360245988fced43a4605ab703d7afacb20a3"
|
|
451
|
+
}
|
|
452
|
+
]
|
|
453
|
+
}
|
|
454
|
+
],
|
|
455
|
+
"ledgerReceipts": [
|
|
456
|
+
{
|
|
457
|
+
"status": "settled",
|
|
458
|
+
"callId": "976568ff-db16-46e4-bd3f-dcd3c7664706",
|
|
459
|
+
"channel": "agent",
|
|
460
|
+
"phase": "campaign",
|
|
461
|
+
"actor": "operational-file-read",
|
|
462
|
+
"model": "not-an-inference",
|
|
463
|
+
"timestamp": 1790874760129,
|
|
464
|
+
"inputTokens": 0,
|
|
465
|
+
"outputTokens": 0,
|
|
466
|
+
"costUsd": 0,
|
|
467
|
+
"costUnknown": false,
|
|
468
|
+
"usageUnknown": false,
|
|
469
|
+
"estimatedCostUsd": 0,
|
|
470
|
+
"tags": {
|
|
471
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume",
|
|
472
|
+
"cellId": "capture-retention:0",
|
|
473
|
+
"scenarioId": "capture-retention",
|
|
474
|
+
"rep": "0",
|
|
475
|
+
"runAttemptId": "2b022ccb-adb3-4468-8c21-48ee82d0634a",
|
|
476
|
+
"attemptNumber": "1"
|
|
477
|
+
}
|
|
478
|
+
},
|
|
479
|
+
{
|
|
480
|
+
"status": "settled",
|
|
481
|
+
"callId": "4de937ba-5b1b-4736-acdf-4e7ee9530d32",
|
|
482
|
+
"channel": "agent",
|
|
483
|
+
"phase": "campaign",
|
|
484
|
+
"actor": "operational-file-read",
|
|
485
|
+
"model": "not-an-inference",
|
|
486
|
+
"timestamp": 1790874760130,
|
|
487
|
+
"inputTokens": 0,
|
|
488
|
+
"outputTokens": 0,
|
|
489
|
+
"costUsd": 0,
|
|
490
|
+
"costUnknown": false,
|
|
491
|
+
"usageUnknown": false,
|
|
492
|
+
"estimatedCostUsd": 0,
|
|
493
|
+
"tags": {
|
|
494
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume",
|
|
495
|
+
"cellId": "capture-retention:0",
|
|
496
|
+
"scenarioId": "capture-retention",
|
|
497
|
+
"rep": "0",
|
|
498
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
499
|
+
"attemptNumber": "1"
|
|
500
|
+
}
|
|
501
|
+
}
|
|
502
|
+
],
|
|
503
|
+
"cacheReuse": {
|
|
504
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
505
|
+
"splitDigest": "sha256:23ea149e81be5435966bd1d185776c7d2e119739830f98a47d4738bdbd46ade6",
|
|
506
|
+
"seed": 42,
|
|
507
|
+
"reps": 1,
|
|
508
|
+
"startedAt": "2026-10-01T17:12:40.131Z",
|
|
509
|
+
"endedAt": "2026-10-01T17:12:40.131Z",
|
|
510
|
+
"durationMs": 0,
|
|
511
|
+
"cells": [
|
|
512
|
+
{
|
|
513
|
+
"manifestHash": "6d4d16fdd6cbaa1119fc301e329b128e6485461e9bc013bdbeaea9ec0270b19c",
|
|
514
|
+
"attempt": {
|
|
515
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
516
|
+
"number": 1
|
|
517
|
+
},
|
|
518
|
+
"cellId": "capture-retention:0",
|
|
519
|
+
"scenarioId": "capture-retention",
|
|
520
|
+
"rep": 0,
|
|
521
|
+
"artifact": {
|
|
522
|
+
"cellId": "capture-retention:0",
|
|
523
|
+
"runAttemptId": "c4859951-1e7c-4d23-b89a-0ca8fa79e2cf",
|
|
524
|
+
"invocation": 2,
|
|
525
|
+
"content": "Actual filesystem capture from invocation 2"
|
|
526
|
+
},
|
|
527
|
+
"judgeScores": {},
|
|
528
|
+
"costUsd": 0,
|
|
529
|
+
"costProvenance": {
|
|
530
|
+
"kind": "estimated",
|
|
531
|
+
"usd": 0
|
|
532
|
+
},
|
|
533
|
+
"costCallIds": [
|
|
534
|
+
"4de937ba-5b1b-4736-acdf-4e7ee9530d32"
|
|
535
|
+
],
|
|
536
|
+
"tokenUsage": {
|
|
537
|
+
"input": 0,
|
|
538
|
+
"output": 0
|
|
539
|
+
},
|
|
540
|
+
"durationMs": 0,
|
|
541
|
+
"seed": 42,
|
|
542
|
+
"cached": true,
|
|
543
|
+
"resolvedModels": [
|
|
544
|
+
"not-an-inference"
|
|
545
|
+
],
|
|
546
|
+
"resolvedModel": "not-an-inference"
|
|
547
|
+
}
|
|
548
|
+
],
|
|
549
|
+
"aggregates": {
|
|
550
|
+
"byJudge": {},
|
|
551
|
+
"byScenario": {},
|
|
552
|
+
"cost": {
|
|
553
|
+
"totalCalls": 2,
|
|
554
|
+
"pendingCalls": 0,
|
|
555
|
+
"unresolvedCalls": 0,
|
|
556
|
+
"reservedCostUsd": 0,
|
|
557
|
+
"inputTokens": 0,
|
|
558
|
+
"outputTokens": 0,
|
|
559
|
+
"reasoningTokens": 0,
|
|
560
|
+
"cachedTokens": 0,
|
|
561
|
+
"cacheWriteTokens": 0,
|
|
562
|
+
"totalCostUsd": 0,
|
|
563
|
+
"costProvenance": {
|
|
564
|
+
"kind": "estimated",
|
|
565
|
+
"usd": 0
|
|
566
|
+
},
|
|
567
|
+
"byChannel": [
|
|
568
|
+
{
|
|
569
|
+
"channel": "agent",
|
|
570
|
+
"calls": 2,
|
|
571
|
+
"inputTokens": 0,
|
|
572
|
+
"outputTokens": 0,
|
|
573
|
+
"reasoningTokens": 0,
|
|
574
|
+
"cachedTokens": 0,
|
|
575
|
+
"cacheWriteTokens": 0,
|
|
576
|
+
"costUsd": 0,
|
|
577
|
+
"unpricedCalls": 0,
|
|
578
|
+
"unknownUsageCalls": 0
|
|
579
|
+
}
|
|
580
|
+
],
|
|
581
|
+
"unpricedModels": [],
|
|
582
|
+
"fullyPriced": true,
|
|
583
|
+
"usageComplete": true,
|
|
584
|
+
"accountingComplete": true,
|
|
585
|
+
"incompleteReasons": []
|
|
586
|
+
},
|
|
587
|
+
"cellsExecuted": 1,
|
|
588
|
+
"cellsSkipped": 0,
|
|
589
|
+
"cellsCached": 1,
|
|
590
|
+
"cellsFailed": 0,
|
|
591
|
+
"cellsDispatchFailed": 0,
|
|
592
|
+
"cellsJudgeFailed": 0,
|
|
593
|
+
"cellsUnclassifiedFailed": 0
|
|
594
|
+
},
|
|
595
|
+
"runDir": "/mnt/traces/eval-attempt-artifacts-20261001-r2/after-resume",
|
|
596
|
+
"artifactsByPath": {},
|
|
597
|
+
"scenarios": [
|
|
598
|
+
{
|
|
599
|
+
"id": "capture-retention",
|
|
600
|
+
"kind": "operational",
|
|
601
|
+
"scenarioDigest": "sha256:70a52f79806918ce9e4833c4ed81360245988fced43a4605ab703d7afacb20a3"
|
|
602
|
+
}
|
|
603
|
+
]
|
|
604
|
+
},
|
|
605
|
+
"firstContentStillPresent": true,
|
|
606
|
+
"modelInference": false,
|
|
607
|
+
"researchUnits": 0
|
|
608
|
+
}
|