@tangle-network/agent-eval 0.215.1 → 0.216.1
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 +51 -0
- package/dist/adapters/http.d.ts +3 -3
- package/dist/{agent-profile-DmzL8KaJ.d.ts → agent-profile-VFcQncbW.d.ts} +5 -5
- package/dist/{agent-profile-DmzL8KaJ.d.ts.map → agent-profile-VFcQncbW.d.ts.map} +1 -1
- package/dist/{agent-profile-cell-BYc0bxsL.js → agent-profile-cell-BK8mlqlq.js} +3 -3
- package/dist/{agent-profile-cell-BYc0bxsL.js.map → agent-profile-cell-BK8mlqlq.js.map} +1 -1
- package/dist/agreement-irr-BEWLIF3_.js +186 -0
- package/dist/agreement-irr-BEWLIF3_.js.map +1 -0
- package/dist/analyst/index.d.ts +487 -14
- package/dist/analyst/index.d.ts.map +1 -1
- package/dist/analyst/index.js +1547 -12
- package/dist/analyst/index.js.map +1 -1
- package/dist/{analyze-runs-3xpby3p2.js → analyze-runs-C84irm4i.js} +608 -604
- package/dist/analyze-runs-C84irm4i.js.map +1 -0
- package/dist/{analyze-runs-BhFLOztS.d.ts → analyze-runs-D68fd-su.d.ts} +17 -15
- package/dist/analyze-runs-D68fd-su.d.ts.map +1 -0
- package/dist/{baseline-Dm8966du.js → baseline-Udi_PLuT.js} +5 -5
- package/dist/{baseline-Dm8966du.js.map → baseline-Udi_PLuT.js.map} +1 -1
- package/dist/{benchmark-BywPeAUo.js → benchmark-CXmVVHH2.js} +2 -2
- package/dist/{benchmark-BywPeAUo.js.map → benchmark-CXmVVHH2.js.map} +1 -1
- package/dist/{benchmark-command-CgpGi7vY.js → benchmark-command-By0vh8q7.js} +71 -70
- package/dist/benchmark-command-By0vh8q7.js.map +1 -0
- package/dist/benchmarks/index.d.ts +5 -5
- package/dist/benchmarks/index.js +3 -3
- package/dist/builder-eval/index.js +2 -2
- package/dist/campaign/index.d.ts +7 -7
- package/dist/campaign/index.js +14 -14
- package/dist/{campaign-CpnsCgfz.js → campaign-BudZ70zo.js} +245 -247
- package/dist/campaign-BudZ70zo.js.map +1 -0
- package/dist/{campaign-evidence-OcUciD-N.js → campaign-evidence-Clrn4hBV.js} +8 -8
- package/dist/campaign-evidence-Clrn4hBV.js.map +1 -0
- package/dist/{canonical-BsZVgBhW.js → canonical-B2RqLofv.js} +2 -2
- package/dist/{canonical-BsZVgBhW.js.map → canonical-B2RqLofv.js.map} +1 -1
- package/dist/{chat-client-DthPOYGd.js → chat-client-l-MVDlDk.js} +2 -2
- package/dist/{chat-client-DthPOYGd.js.map → chat-client-l-MVDlDk.js.map} +1 -1
- package/dist/{chat-json-call-BrKsyspg.js → chat-json-call-puMArmtW.js} +2 -2
- package/dist/{chat-json-call-BrKsyspg.js.map → chat-json-call-puMArmtW.js.map} +1 -1
- package/dist/{claim-BXcZw-jo.js → claim-q4HQhK1G.js} +4 -4
- package/dist/claim-q4HQhK1G.js.map +1 -0
- package/dist/cli.js +9 -9
- package/dist/cli.js.map +1 -1
- package/dist/cluster-bootstrap-qcUWYFl-.js +66 -0
- package/dist/cluster-bootstrap-qcUWYFl-.js.map +1 -0
- package/dist/contract/index.d.ts +11 -11
- package/dist/contract/index.js +1023 -1079
- package/dist/contract/index.js.map +1 -1
- package/dist/{cost-ledger-D2FnJPBF.js → cost-ledger-BP8V4kUQ.js} +835 -816
- package/dist/cost-ledger-BP8V4kUQ.js.map +1 -0
- package/dist/{cost-ledger-CO7UyEcQ.d.ts → cost-ledger-Jk62OBfF.d.ts} +12 -47
- package/dist/cost-ledger-Jk62OBfF.d.ts.map +1 -0
- package/dist/{counterfactual-DtMWvn7D.js → counterfactual-Dn3veueY.js} +2 -2
- package/dist/{counterfactual-DtMWvn7D.js.map → counterfactual-Dn3veueY.js.map} +1 -1
- package/dist/{default-registry-KP4-Tolz.js → default-registry-COhk9g36.js} +744 -759
- package/dist/default-registry-COhk9g36.js.map +1 -0
- package/dist/{default-registry-B4B7GMvI.d.ts → default-registry-DzHiGruW.d.ts} +6 -6
- package/dist/{default-registry-B4B7GMvI.d.ts.map → default-registry-DzHiGruW.d.ts.map} +1 -1
- package/dist/{define-agent-eval-BdfnqXeP.js → define-agent-eval-CNTPzIkp.js} +10 -10
- package/dist/define-agent-eval-CNTPzIkp.js.map +1 -0
- package/dist/{define-agent-eval-DfXvNho5.d.ts → define-agent-eval-WSck0CZJ.d.ts} +8 -8
- package/dist/{define-agent-eval-DfXvNho5.d.ts.map → define-agent-eval-WSck0CZJ.d.ts.map} +1 -1
- package/dist/{descriptive-CnU8Yq29.js → descriptive-DboxPZrG.js} +2 -2
- package/dist/{descriptive-CnU8Yq29.js.map → descriptive-DboxPZrG.js.map} +1 -1
- package/dist/diagnosis/index.d.ts +3 -3
- package/dist/diagnosis/index.js +4 -4
- package/dist/{dspy-rlm-engine-DZfHM7kT.js → dspy-rlm-engine-LRCu61m5.js} +3 -3
- package/dist/{dspy-rlm-engine-DZfHM7kT.js.map → dspy-rlm-engine-LRCu61m5.js.map} +1 -1
- package/dist/{effect-sizes-DiH8MGOH.js → effect-sizes-BhzEpXMc.js} +2 -2
- package/dist/{effect-sizes-DiH8MGOH.js.map → effect-sizes-BhzEpXMc.js.map} +1 -1
- package/dist/{engine-C37HS2kr.d.ts → engine-3A_4jOlI.d.ts} +5 -5
- package/dist/{engine-C37HS2kr.d.ts.map → engine-3A_4jOlI.d.ts.map} +1 -1
- package/dist/{equal-terms-Blx9Zk5I.js → equal-terms-BtdXayCJ.js} +2 -2
- package/dist/{equal-terms-Blx9Zk5I.js.map → equal-terms-BtdXayCJ.js.map} +1 -1
- package/dist/{errors-Dngq5h35.js → errors-DVZiqS7K.js} +6 -2
- package/dist/{errors-Dngq5h35.js.map → errors-DVZiqS7K.js.map} +1 -1
- package/dist/{eval-campaign-4byTtqrf.js → eval-campaign-90jeaCAI.js} +6 -6
- package/dist/{eval-campaign-4byTtqrf.js.map → eval-campaign-90jeaCAI.js.map} +1 -1
- package/dist/evaluation-6Yoi74gM.d.ts +75 -0
- package/dist/evaluation-6Yoi74gM.d.ts.map +1 -0
- package/dist/evaluation.d.ts +2 -74
- package/dist/evaluation.js +1 -1
- package/dist/{exact-types-Dlb73wN4.d.ts → exact-types-ClugMJfI.d.ts} +2 -2
- package/dist/{exact-types-Dlb73wN4.d.ts.map → exact-types-ClugMJfI.d.ts.map} +1 -1
- package/dist/experiment/index.d.ts +404 -3
- package/dist/experiment/index.d.ts.map +1 -1
- package/dist/experiment/index.js +313 -19
- package/dist/experiment/index.js.map +1 -1
- package/dist/{external-optimizer-process-GngofQML.js → external-optimizer-process-BjMGUuds.js} +3 -3
- package/dist/{external-optimizer-process-GngofQML.js.map → external-optimizer-process-BjMGUuds.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-BUxifb8g.js → external-optimizer-subprocess-BqW0hHcI.js} +791 -532
- package/dist/external-optimizer-subprocess-BqW0hHcI.js.map +1 -0
- package/dist/{feedback-trajectory-BLN_6fEs.d.ts → feedback-trajectory-C4EG0nNH.d.ts} +5 -8
- package/dist/feedback-trajectory-C4EG0nNH.d.ts.map +1 -0
- package/dist/fuzz.d.ts +1 -1
- package/dist/fuzz.js +3 -3
- package/dist/hosted/index.js +1 -1
- package/dist/{index-CPJJr98o.d.ts → index-BFRO6chs.d.ts} +3 -2
- package/dist/index-BFRO6chs.d.ts.map +1 -0
- package/dist/{index-Ch9VxCpK.d.ts → index-BO41eGj5.d.ts} +2 -2
- package/dist/{index-Ch9VxCpK.d.ts.map → index-BO41eGj5.d.ts.map} +1 -1
- package/dist/{index-B3WBm953.d.ts → index-CyszSg4O.d.ts} +165 -162
- package/dist/index-CyszSg4O.d.ts.map +1 -0
- package/dist/{index-B7m2uzJL.d.ts → index-DPcngtwt.d.ts} +30 -17
- package/dist/index-DPcngtwt.d.ts.map +1 -0
- package/dist/{index-tYLUvk_w.d.ts → index-DV-jSzxF.d.ts} +9 -9
- package/dist/{index-tYLUvk_w.d.ts.map → index-DV-jSzxF.d.ts.map} +1 -1
- package/dist/index.d.ts +1969 -1871
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +162 -201
- package/dist/index.js.map +1 -1
- package/dist/{integrity-BvnVARVl.js → integrity-BXOsAb3i.js} +2 -2
- package/dist/{integrity-BvnVARVl.js.map → integrity-BXOsAb3i.js.map} +1 -1
- package/dist/{integrity-CCAxGVK6.d.ts → integrity-DAMEDFaC.d.ts} +2 -2
- package/dist/{integrity-CCAxGVK6.d.ts.map → integrity-DAMEDFaC.d.ts.map} +1 -1
- package/dist/{integrity-BV1ZDf42.js → integrity-dLiDe2kD.js} +2 -5
- package/dist/integrity-dLiDe2kD.js.map +1 -0
- package/dist/{internal-CRHwS9iz.js → internal-Epl-YBZW.js} +3 -3
- package/dist/{internal-CRHwS9iz.js.map → internal-Epl-YBZW.js.map} +1 -1
- package/dist/interval-CELpye1n.js +219 -0
- package/dist/interval-CELpye1n.js.map +1 -0
- package/dist/interval-ohuOJurX.d.ts +47 -0
- package/dist/interval-ohuOJurX.d.ts.map +1 -0
- package/dist/{journal-DvKCvjty.js → journal-DfRz9kps.js} +2 -2
- package/dist/{journal-DvKCvjty.js.map → journal-DfRz9kps.js.map} +1 -1
- package/dist/{judge-calibration-N44V04Se.js → judge-calibration-BpCy5XQl.js} +3 -3
- package/dist/{judge-calibration-N44V04Se.js.map → judge-calibration-BpCy5XQl.js.map} +1 -1
- package/dist/{judge-gate-DNcUrRFA.d.ts → judge-gate-CE7CQAgt.d.ts} +3 -3
- package/dist/{judge-gate-DNcUrRFA.d.ts.map → judge-gate-CE7CQAgt.d.ts.map} +1 -1
- package/dist/{judge-gate-B1qYnkpi.js → judge-gate-iY9s4L44.js} +3 -3
- package/dist/judge-gate-iY9s4L44.js.map +1 -0
- package/dist/{kind-factory-AaEhUADH.js → kind-factory-BobI5Jy3.js} +5 -5
- package/dist/{kind-factory-AaEhUADH.js.map → kind-factory-BobI5Jy3.js.map} +1 -1
- package/dist/ledger-core/index.js +3 -3
- package/dist/{ledger-core-B2hahzrt.js → ledger-core-Gq7yp8lP.js} +4 -4
- package/dist/{ledger-core-B2hahzrt.js.map → ledger-core-Gq7yp8lP.js.map} +1 -1
- package/dist/{llm-client-DQveoxSM.js → llm-client-CyBAoYiX.js} +192 -219
- package/dist/llm-client-CyBAoYiX.js.map +1 -0
- package/dist/{llm-judge-Cd5s5Ei6.js → llm-judge-CG12SQJq.js} +499 -921
- package/dist/llm-judge-CG12SQJq.js.map +1 -0
- package/dist/matrix/index.d.ts +2 -2
- package/dist/matrix/index.js +1 -1
- package/dist/{matrix-DXVAJMLD.d.ts → matrix-DSrSxjjy.d.ts} +4 -4
- package/dist/{matrix-DXVAJMLD.d.ts.map → matrix-DSrSxjjy.d.ts.map} +1 -1
- package/dist/{matrix-_BV9-Lk5.js → matrix-Dm0BQW7U.js} +7 -11
- package/dist/matrix-Dm0BQW7U.js.map +1 -0
- package/dist/{paired-tests-jbwPdSz5.js → mean-DHoMiudD.js} +11 -5
- package/dist/mean-DHoMiudD.js.map +1 -0
- package/dist/meta-eval/index.d.ts +3 -3
- package/dist/meta-eval/index.js +9 -70
- package/dist/meta-eval/index.js.map +1 -1
- package/dist/mint-DNU8QPG4.d.ts +81 -0
- package/dist/mint-DNU8QPG4.d.ts.map +1 -0
- package/dist/{mint-Csor1BLJ.js → mint-QaD5cA0u.js} +3 -3
- package/dist/{mint-Csor1BLJ.js.map → mint-QaD5cA0u.js.map} +1 -1
- package/dist/{multiplicity-D9cjmStq.js → multiplicity-DQUFcbW4.js} +2 -2
- package/dist/{multiplicity-D9cjmStq.js.map → multiplicity-DQUFcbW4.js.map} +1 -1
- package/dist/multishot/golden/index.d.ts +1 -1
- package/dist/multishot/golden/index.d.ts.map +1 -1
- package/dist/multishot/golden/index.js +3 -6
- package/dist/multishot/golden/index.js.map +1 -1
- package/dist/multishot/index.d.ts +2 -2
- package/dist/multishot/index.js +1 -1
- package/dist/nonnegative-number-DvdMxg_f.js +9 -0
- package/dist/nonnegative-number-DvdMxg_f.js.map +1 -0
- package/dist/openapi.json +1 -1
- package/dist/{otel-spans-C8YIJWy1.js → otel-spans-B9tUXRHF.js} +3 -3
- package/dist/{otel-spans-C8YIJWy1.js.map → otel-spans-B9tUXRHF.js.map} +1 -1
- package/dist/{paired-arms-DzKn0TfY.js → paired-arms-CwMGJsrS.js} +5 -5
- package/dist/{paired-arms-DzKn0TfY.js.map → paired-arms-CwMGJsrS.js.map} +1 -1
- package/dist/{paired-binary-Q36bhn2R.js → paired-binary-CvpFHfKZ.js} +2 -2
- package/dist/{paired-binary-Q36bhn2R.js.map → paired-binary-CvpFHfKZ.js.map} +1 -1
- package/dist/{paired-promotion-decision-BN6IQTiu.js → paired-promotion-decision-DsAPYGCm.js} +3 -3
- package/dist/{paired-promotion-decision-BN6IQTiu.js.map → paired-promotion-decision-DsAPYGCm.js.map} +1 -1
- package/dist/{search-kernel-BztlERLv.js → pareto-Cm6ndrWN.js} +883 -861
- package/dist/pareto-Cm6ndrWN.js.map +1 -0
- package/dist/pipelines/index.d.ts +2 -2
- package/dist/pipelines/index.js +3 -3
- package/dist/{power-and-mde-fK5Qocwo.js → power-and-mde-Hj2ubb4B.js} +2 -2
- package/dist/{power-and-mde-fK5Qocwo.js.map → power-and-mde-Hj2ubb4B.js.map} +1 -1
- package/dist/predicate-FdKmElA_.js +38 -0
- package/dist/predicate-FdKmElA_.js.map +1 -0
- package/dist/{prime-protocol-DcAK9pC8.js → prime-protocol-CHKgA3a6.js} +2 -2
- package/dist/{prime-protocol-DcAK9pC8.js.map → prime-protocol-CHKgA3a6.js.map} +1 -1
- package/dist/{prime-protocol-BHia9vDz.d.ts → prime-protocol-HiB9HX7E.d.ts} +3 -3
- package/dist/{prime-protocol-BHia9vDz.d.ts.map → prime-protocol-HiB9HX7E.d.ts.map} +1 -1
- package/dist/{produced-state-BGp4BX8t.js → produced-state-BsDmnwxc.js} +8 -8
- package/dist/{produced-state-BGp4BX8t.js.map → produced-state-BsDmnwxc.js.map} +1 -1
- package/dist/profile-cell.js +1 -1
- package/dist/{release-confidence-BK2QCB5M.d.ts → promotion-gate-BawrdPYr.d.ts} +102 -102
- package/dist/promotion-gate-BawrdPYr.d.ts.map +1 -0
- package/dist/{promotion-policy-CWW35UWq.js → promotion-policy-KO57gbyV.js} +4 -4
- package/dist/{promotion-policy-CWW35UWq.js.map → promotion-policy-KO57gbyV.js.map} +1 -1
- package/dist/{proposal-findings-CXblsskH.js → proposal-findings-CY1xKn_a.js} +2 -36
- package/dist/proposal-findings-CY1xKn_a.js.map +1 -0
- package/dist/{query-B1D1qp2Z.js → query-BUCZCK7F.js} +3 -3
- package/dist/{query-B1D1qp2Z.js.map → query-BUCZCK7F.js.map} +1 -1
- package/dist/{random-Dn5fPWkt.js → random-ByUSnV9w.js} +2 -2
- package/dist/{random-Dn5fPWkt.js.map → random-ByUSnV9w.js.map} +1 -1
- package/dist/{raw-provider-sink-rB2DYxr6.js → raw-provider-sink-DoHz4JNG.js} +2 -2
- package/dist/{raw-provider-sink-rB2DYxr6.js.map → raw-provider-sink-DoHz4JNG.js.map} +1 -1
- package/dist/{redact-Dk1ZdJEm.js → redact-B5xXKqQ3.js} +97 -98
- package/dist/redact-B5xXKqQ3.js.map +1 -0
- package/dist/{redact-Cveoq6fJ.d.ts → redact-mtCCcT8z.d.ts} +4 -20
- package/dist/redact-mtCCcT8z.d.ts.map +1 -0
- package/dist/{registry-CMYIaDgS.d.ts → registry-D7_hm2fK.d.ts} +11 -13
- package/dist/registry-D7_hm2fK.d.ts.map +1 -0
- package/dist/{release-confidence-Dbe9AS_X.js → release-confidence-CpuP65S5.js} +5 -5
- package/dist/{release-confidence-Dbe9AS_X.js.map → release-confidence-CpuP65S5.js.map} +1 -1
- package/dist/{report-command-CtIIsQKP.js → report-command-SC9nqb4M.js} +128 -177
- package/dist/report-command-SC9nqb4M.js.map +1 -0
- package/dist/reporting-html-Kvlc8oMt.js +100 -0
- package/dist/reporting-html-Kvlc8oMt.js.map +1 -0
- package/dist/reporting.d.ts +3 -3
- package/dist/reporting.js +6 -6
- package/dist/{researcher-f6mZhh58.d.ts → researcher-CEcfku-T.d.ts} +5 -5
- package/dist/{researcher-f6mZhh58.d.ts.map → researcher-CEcfku-T.d.ts.map} +1 -1
- package/dist/reward-CpCmAM8L.d.ts +78 -0
- package/dist/reward-CpCmAM8L.d.ts.map +1 -0
- package/dist/{reward-hacking-BFcQgkux.js → reward-hacking-E0d-WVcP.js} +2 -2
- package/dist/{reward-hacking-BFcQgkux.js.map → reward-hacking-E0d-WVcP.js.map} +1 -1
- package/dist/{reward-hacking-DHA7c1l8.d.ts → reward-hacking-RVdZ6Kkt.d.ts} +2 -2
- package/dist/{reward-hacking-DHA7c1l8.d.ts.map → reward-hacking-RVdZ6Kkt.d.ts.map} +1 -1
- package/dist/rl.d.ts +8 -8
- package/dist/rl.js +14 -14
- package/dist/rollout/index.d.ts +3 -2
- package/dist/rollout/index.js +2 -2
- package/dist/{rollout-D9bKKBnZ.js → rollout-DZV1t2Wg.js} +2 -2
- package/dist/{rollout-D9bKKBnZ.js.map → rollout-DZV1t2Wg.js.map} +1 -1
- package/dist/round-decimal-COrKglMh.js +10 -0
- package/dist/round-decimal-COrKglMh.js.map +1 -0
- package/dist/{rubric-predictive-validity-2SgHxoa5.d.ts → rubric-predictive-validity-C03EC7ZE.d.ts} +2 -2
- package/dist/{rubric-predictive-validity-2SgHxoa5.d.ts.map → rubric-predictive-validity-C03EC7ZE.d.ts.map} +1 -1
- package/dist/{rubric-predictive-validity-BsgkBfT0.js → rubric-predictive-validity-DaFW70y9.js} +3 -3
- package/dist/{rubric-predictive-validity-BsgkBfT0.js.map → rubric-predictive-validity-DaFW70y9.js.map} +1 -1
- package/dist/{run-record-DlTZ_lqe.js → run-record-Boc5IUTp.js} +4 -4
- package/dist/{run-record-DlTZ_lqe.js.map → run-record-Boc5IUTp.js.map} +1 -1
- package/dist/{run-record-D1Hw-Jml.d.ts → run-record-C1E24G2A.d.ts} +2 -2
- package/dist/{run-record-D1Hw-Jml.d.ts.map → run-record-C1E24G2A.d.ts.map} +1 -1
- package/dist/{run-record-ES07wfz9.js → run-record-C5e7FjhJ.js} +3 -5
- package/dist/run-record-C5e7FjhJ.js.map +1 -0
- package/dist/search/index.d.ts +46 -32
- package/dist/search/index.d.ts.map +1 -1
- package/dist/search/index.js +3 -3
- package/dist/{search-kernel-BqyBZ9Wm.d.ts → search-kernel-DyNEa6IU.d.ts} +14 -10
- package/dist/search-kernel-DyNEa6IU.d.ts.map +1 -0
- package/dist/{search-ledger-d7V8o9Lk.js → search-ledger-CVUGLDZp.js} +189 -199
- package/dist/search-ledger-CVUGLDZp.js.map +1 -0
- package/dist/{search-ledger-file-BqOo-bVo.js → search-ledger-file-DDQUgvCg.js} +4 -4
- package/dist/{search-ledger-file-BqOo-bVo.js.map → search-ledger-file-DDQUgvCg.js.map} +1 -1
- package/dist/search-ledger-types-p2xKfiDJ.d.ts.map +1 -1
- package/dist/{search-shipper-er4pK5_Y.js → search-shipper-BOSD46qv.js} +3 -3
- package/dist/{search-shipper-er4pK5_Y.js.map → search-shipper-BOSD46qv.js.map} +1 -1
- package/dist/{search-summary-CBvz3u_S.js → search-summary-BQCSWf1g.js} +5 -5
- package/dist/{search-summary-CBvz3u_S.js.map → search-summary-BQCSWf1g.js.map} +1 -1
- package/dist/{semantic-concept-judge-BPqbwIVs.js → semantic-concept-judge-BQF8te0T.js} +12 -14
- package/dist/semantic-concept-judge-BQF8te0T.js.map +1 -0
- package/dist/{sequential-D-1PPCmM.js → sequential-Bf_pbE50.js} +5 -5
- package/dist/{sequential-D-1PPCmM.js.map → sequential-Bf_pbE50.js.map} +1 -1
- package/dist/{sequential-eprocess-D1jKoihe.js → sequential-eprocess-DM_qENCl.js} +2 -2
- package/dist/{sequential-eprocess-D1jKoihe.js.map → sequential-eprocess-DM_qENCl.js.map} +1 -1
- package/dist/{server-DbcC7xMl.js → server-e9oWO_bN.js} +4 -4
- package/dist/{server-DbcC7xMl.js.map → server-e9oWO_bN.js.map} +1 -1
- package/dist/{single-run-lock-fVu3MZWH.js → single-run-lock-CV9Eyd_q.js} +2 -2
- package/dist/{single-run-lock-fVu3MZWH.js.map → single-run-lock-CV9Eyd_q.js.map} +1 -1
- package/dist/{skillopt-optimization-method-Dcepr6QL.js → skillopt-optimization-method-BiwA8nfX.js} +427 -12
- package/dist/skillopt-optimization-method-BiwA8nfX.js.map +1 -0
- package/dist/{spans-D9_ehFuk.d.ts → spans-BHAUXMGN.d.ts} +2 -2
- package/dist/{spans-D9_ehFuk.d.ts.map → spans-BHAUXMGN.d.ts.map} +1 -1
- package/dist/{statistical-heldout-1Gbmex7v.js → statistical-heldout-42NlCNDe.js} +3 -3
- package/dist/{statistical-heldout-1Gbmex7v.js.map → statistical-heldout-42NlCNDe.js.map} +1 -1
- package/dist/{statistical-heldout-CJTj2EoS.d.ts → statistical-heldout-CYpGaoDH.d.ts} +3 -3
- package/dist/{statistical-heldout-CJTj2EoS.d.ts.map → statistical-heldout-CYpGaoDH.d.ts.map} +1 -1
- package/dist/{store-otlp-DWVhUgEj.js → store-otlp-DJ9dtA5x.js} +135 -132
- package/dist/store-otlp-DJ9dtA5x.js.map +1 -0
- package/dist/{store-tool-spans-BNICrhsb.js → store-tool-spans-CIX5arAG.js} +7 -7
- package/dist/{store-tool-spans-BNICrhsb.js.map → store-tool-spans-CIX5arAG.js.map} +1 -1
- package/dist/{store-tool-spans-BNzls3OD.d.ts → store-tool-spans-DdCSPa7C.d.ts} +162 -161
- package/dist/store-tool-spans-DdCSPa7C.d.ts.map +1 -0
- package/dist/storyboard/index.d.ts.map +1 -1
- package/dist/storyboard/index.js +20 -3
- package/dist/storyboard/index.js.map +1 -1
- package/dist/{student-t-BQeKJx2N.js → student-t-DVPW7CdL.js} +2 -2
- package/dist/{student-t-BQeKJx2N.js.map → student-t-DVPW7CdL.js.map} +1 -1
- package/dist/{summary-report-DtdRucM0.js → summary-report-V_JACR4Y.js} +526 -614
- package/dist/summary-report-V_JACR4Y.js.map +1 -0
- package/dist/{summary-report-C-CtItCN.d.ts → summary-report-bdFDXq0b.d.ts} +10 -5
- package/dist/summary-report-bdFDXq0b.d.ts.map +1 -0
- package/dist/supervisor-run/index.d.ts +154 -150
- package/dist/supervisor-run/index.d.ts.map +1 -1
- package/dist/supervisor-run/index.js +65 -3
- package/dist/supervisor-run/index.js.map +1 -1
- package/dist/systemone-BomhHr2c.d.ts +135 -0
- package/dist/systemone-BomhHr2c.d.ts.map +1 -0
- package/dist/systemone-protocol.js +1 -1
- package/dist/{systemone-review-C4dZgrgN.d.ts → systemone-review-CUVcQJu6.d.ts} +2 -2
- package/dist/{systemone-review-C4dZgrgN.d.ts.map → systemone-review-CUVcQJu6.d.ts.map} +1 -1
- package/dist/{systemone-review-DTClwgmE.js → systemone-review-D7sXwsqq.js} +3 -3
- package/dist/{systemone-review-DTClwgmE.js.map → systemone-review-D7sXwsqq.js.map} +1 -1
- package/dist/systemone.d.ts +4 -134
- package/dist/systemone.js +4 -4
- package/dist/{task-failure-attributes-DngaLdE4.js → task-failure-attributes-hEmASjix.js} +2 -2
- package/dist/{task-failure-attributes-DngaLdE4.js.map → task-failure-attributes-hEmASjix.js.map} +1 -1
- package/dist/{terminal-record-BHocwFw1.js → terminal-record-DZyQZ3pz.js} +14 -10
- package/dist/terminal-record-DZyQZ3pz.js.map +1 -0
- package/dist/{tool-groups-BKRrtqy4.d.ts → tool-groups-C3NEcdCM.d.ts} +3 -3
- package/dist/tool-groups-C3NEcdCM.d.ts.map +1 -0
- package/dist/{tool-waste-Wi4g6Ekb.d.ts → tool-waste-BKgvd-TP.d.ts} +30 -30
- package/dist/tool-waste-BKgvd-TP.d.ts.map +1 -0
- package/dist/{tool-waste-BYt5rdt8.js → tool-waste-BlesqTqW.js} +5 -158
- package/dist/tool-waste-BlesqTqW.js.map +1 -0
- package/dist/{trace-questions-CLx8vt6t.d.ts → trace-questions-CM2CAocU.d.ts} +4 -4
- package/dist/{trace-questions-CLx8vt6t.d.ts.map → trace-questions-CM2CAocU.d.ts.map} +1 -1
- package/dist/trace-repair/index.d.ts +3 -3
- package/dist/trace-repair/index.d.ts.map +1 -1
- package/dist/trace-repair/index.js +7 -13
- package/dist/trace-repair/index.js.map +1 -1
- package/dist/traces.d.ts +76 -58
- package/dist/traces.d.ts.map +1 -1
- package/dist/traces.js +9 -9
- package/dist/trajectory-replay/index.js +3 -3
- package/dist/{tree-rXLW3M_4.js → tree-C1XYeXvy.js} +2277 -2339
- package/dist/tree-C1XYeXvy.js.map +1 -0
- package/dist/{types-DjOMwGY_.d.ts → types-BqkVmHfv.d.ts} +7 -1
- package/dist/types-BqkVmHfv.d.ts.map +1 -0
- package/dist/{types-DBZ7keha.d.ts → types-C02qtnG8.d.ts} +4 -46
- package/dist/types-C02qtnG8.d.ts.map +1 -0
- package/dist/{types-DHb5TI36.d.ts → types-C1tnkEgM.d.ts} +347 -6
- package/dist/types-C1tnkEgM.d.ts.map +1 -0
- package/dist/{types-DinHtRJp.d.ts → types-Cbe2fWBa.d.ts} +3 -3
- package/dist/{types-DinHtRJp.d.ts.map → types-Cbe2fWBa.d.ts.map} +1 -1
- package/dist/{types-CGEmVYOk.d.ts → types-Cq8mSbq8.d.ts} +78 -72
- package/dist/types-Cq8mSbq8.d.ts.map +1 -0
- package/dist/{types-CW5b90JH.js → types-IXD2HJzu.js} +16 -3
- package/dist/types-IXD2HJzu.js.map +1 -0
- package/dist/{verdict-DhPoXApk.js → verdict-Dl0tiT1X.js} +2 -2
- package/dist/{verdict-DhPoXApk.js.map → verdict-Dl0tiT1X.js.map} +1 -1
- package/dist/{verdict-cache-BQEsVy9n.js → verdict-cache-Bm8EKL2s.js} +2 -2
- package/dist/{verdict-cache-BQEsVy9n.js.map → verdict-cache-Bm8EKL2s.js.map} +1 -1
- package/dist/wire/index.d.ts +3 -3
- package/dist/wire/index.js +1 -1
- package/docs/concepts.md +1 -0
- package/docs/design/mlbenchmarks-book-review.md +1 -1
- package/docs/design/mlbenchmarks-review/probes.mts +23 -31
- package/docs/experiment.md +1 -1
- package/docs/finding-layers.md +69 -0
- package/package.json +1 -1
- package/dist/analyze-runs-3xpby3p2.js.map +0 -1
- package/dist/analyze-runs-BhFLOztS.d.ts.map +0 -1
- package/dist/ast-DJ6Bc7ab.js +0 -558
- package/dist/ast-DJ6Bc7ab.js.map +0 -1
- package/dist/ast-hI-vjW6J.d.ts +0 -457
- package/dist/ast-hI-vjW6J.d.ts.map +0 -1
- package/dist/benchmark-command-CgpGi7vY.js.map +0 -1
- package/dist/campaign-CpnsCgfz.js.map +0 -1
- package/dist/campaign-evidence-OcUciD-N.js.map +0 -1
- package/dist/claim-BXcZw-jo.js.map +0 -1
- package/dist/cost-ledger-CO7UyEcQ.d.ts.map +0 -1
- package/dist/cost-ledger-D2FnJPBF.js.map +0 -1
- package/dist/default-registry-KP4-Tolz.js.map +0 -1
- package/dist/define-agent-eval-BdfnqXeP.js.map +0 -1
- package/dist/evaluation.d.ts.map +0 -1
- package/dist/external-optimizer-subprocess-BUxifb8g.js.map +0 -1
- package/dist/feedback-trajectory-BLN_6fEs.d.ts.map +0 -1
- package/dist/index-B3WBm953.d.ts.map +0 -1
- package/dist/index-B7m2uzJL.d.ts.map +0 -1
- package/dist/index-CPJJr98o.d.ts.map +0 -1
- package/dist/integrity-BV1ZDf42.js.map +0 -1
- package/dist/judge-gate-B1qYnkpi.js.map +0 -1
- package/dist/llm-client-DQveoxSM.js.map +0 -1
- package/dist/llm-judge-Cd5s5Ei6.js.map +0 -1
- package/dist/matrix-_BV9-Lk5.js.map +0 -1
- package/dist/mint-DCGoBYhB.d.ts +0 -155
- package/dist/mint-DCGoBYhB.d.ts.map +0 -1
- package/dist/paired-tests-jbwPdSz5.js.map +0 -1
- package/dist/proposal-findings-CXblsskH.js.map +0 -1
- package/dist/redact-Cveoq6fJ.d.ts.map +0 -1
- package/dist/redact-Dk1ZdJEm.js.map +0 -1
- package/dist/registry-CMYIaDgS.d.ts.map +0 -1
- package/dist/release-confidence-BK2QCB5M.d.ts.map +0 -1
- package/dist/report-command-CtIIsQKP.js.map +0 -1
- package/dist/run-record-ES07wfz9.js.map +0 -1
- package/dist/search-kernel-BqyBZ9Wm.d.ts.map +0 -1
- package/dist/search-kernel-BztlERLv.js.map +0 -1
- package/dist/search-ledger-d7V8o9Lk.js.map +0 -1
- package/dist/semantic-concept-judge-BPqbwIVs.js.map +0 -1
- package/dist/skillopt-optimization-method-Dcepr6QL.js.map +0 -1
- package/dist/store-otlp-DWVhUgEj.js.map +0 -1
- package/dist/store-tool-spans-BNzls3OD.d.ts.map +0 -1
- package/dist/summary-report-C-CtItCN.d.ts.map +0 -1
- package/dist/summary-report-DtdRucM0.js.map +0 -1
- package/dist/systemone.d.ts.map +0 -1
- package/dist/terminal-record-BHocwFw1.js.map +0 -1
- package/dist/tool-groups-BKRrtqy4.d.ts.map +0 -1
- package/dist/tool-waste-BYt5rdt8.js.map +0 -1
- package/dist/tool-waste-Wi4g6Ekb.d.ts.map +0 -1
- package/dist/tree-rXLW3M_4.js.map +0 -1
- package/dist/types-CGEmVYOk.d.ts.map +0 -1
- package/dist/types-CW5b90JH.js.map +0 -1
- package/dist/types-DBZ7keha.d.ts.map +0 -1
- package/dist/types-DHb5TI36.d.ts.map +0 -1
- package/dist/types-DjOMwGY_.d.ts.map +0 -1
|
@@ -164,6 +164,12 @@ interface SupervisorRunSources {
|
|
|
164
164
|
readonly patch: string | null;
|
|
165
165
|
/** Outer-driver log (used for the driver's steer verbs + deadline evidence). */
|
|
166
166
|
readonly driverLog: string | null;
|
|
167
|
+
/**
|
|
168
|
+
* Runtime `coordination-log.jsonl` (JSONL of `{ runId, seq, at, event }`):
|
|
169
|
+
* questions, steers, answers and peer mail between nodes. Absent for stores
|
|
170
|
+
* that keep no coordination log; null when the Runtime run dir has none.
|
|
171
|
+
*/
|
|
172
|
+
readonly coordinationLog?: string | null;
|
|
167
173
|
/**
|
|
168
174
|
* Worker tokens recovered from a harness session store; null = store unavailable.
|
|
169
175
|
* `store` names the store in the report's provenance line (e.g. `opencode`).
|
|
@@ -561,4 +567,4 @@ interface SupervisorRunTreeGap {
|
|
|
561
567
|
}
|
|
562
568
|
//#endregion
|
|
563
569
|
export { WallDistribution as A, SupervisorRunSources as C, SupervisorStatusSource as D, SupervisorRunTreeGapCode as E, showMeasured as F, unavailable as I, WorkerNativeSession as M, WorkerTurnCoverage as N, TerminalFailure as O, isUnavailable as P, SupervisorRunRollup as S, SupervisorRunTreeGap as T, SpendMeasurements as _, NamedResourceReceipt as a, SupervisorRunReader as b, PatchStats as c, RoleSpend as d, RollupCellRow as f, SpendMeasurement as g, SourceLimits as h, NO_SOURCE_LIMITS as i, WorkerLogSource as j, Unavailable as k, PerWorkerRow as l, SUPERVISOR_RUN_SCHEMA as m, EconomicsMetrics as n, OrchestrationMetrics as o, SUPERVISOR_RUN_ROLLUP_SCHEMA as p, Measured as r, OutcomeMetrics as s, DecisionMetrics as t, ResourceSpendRecord as u, SteerBreakdown as v, SupervisorRunTree as w, SupervisorRunReport as x, SupervisorRunNodeRole as y };
|
|
564
|
-
//# sourceMappingURL=types-
|
|
570
|
+
//# sourceMappingURL=types-BqkVmHfv.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"types-BqkVmHfv.d.ts","names":[],"sources":["../src/supervisor-run/types.ts"],"mappings":";;;;UAkCiB;WACN;;;KAIC,SAAS,KAAK,IAAI;iBAEd,YAAY,iBAAiB;iBAI7B,cAAc,aAAa,KAAK;;iBAKhC,aAAa,GAAG;;KAWpB;;UAGK;WACN;WACA;;;KAIC;WAEG;WACA;;;WAGA;;;;WAIA;;WAEA;WAAgC;;;;;;UAM9B;;;;;WAKN;;WAEA;;WAEA;;WAEA;;WAEA;;WAEA;;;;;;WAMA,QAAQ;;;;;;WAMR,gBAAgB;;WAEhB,6BAA6B;;WAE7B;;WAEA;WACA;WACA;WACA;;;;;;;;;;;;;UAcM;;WAEN;;WAEA;;WAEA;;WAEA;;WAEA;;;cAIE,kBAAkB;;;;;;;;;;UAiBd;;WAEN;WACA;;WAEA;;WAEA;;;;;;WAMA;;;;;;WAMA;;WAEA;;WAEA;;WAEA;;WAEA;;WAEA,kBAAkB;;WAElB;;;;WAIA;;;;;;WAMA;;;;;;;WAOA;;WAEA;;WAEA;;WAEA;;;;;;WAMA;;;;;WAKA;IACP;IACA;IACA;IACA;;IAEA;IACA;;WAEO;;WAEA,QAAQ;;;;;WAKR;;WAEA,iCAAiC;;;;;WAKjC;;;;;;UAOM;;WAEN;EACT,QAAQ,QAAQ;;cAOL;cACA;UAEI;;WAEN;WACA;;WAEA;;WAEA;;UAGM;WACN,gBAAgB;WAChB,gBAAgB;WAChB,kBAAkB;;WAElB,QAAQ;WACR,iBAAiB;WACjB,gBAAgB,kBAAkB;;WAElC,kBAAkB;;;;;;WAMlB,OAAO;WACP,WAAW;WACX,gBAAgB;;WAEhB,UAAU;;WAEV,gBAAgB;;WAEhB,iBAAiB;WACjB,oBAAoB;WACpB,kBAAkB;;;;;;;;;WASlB,sBAAsB;;WAEtB,QAAQ;WACR,SAAS;;WAET,mBAAmB;;UAGb;WACN,iBAAiB,SAAS;WAC1B,iBAAiB,SAAS;;;;;;;WAO1B,UAAU;;WAEV,UAAU;;WAEV,WAAW;;WAEX,oBAAoB;;WAEpB,wBAAwB;;WAExB,eAAe;WACf,qBAAqB;;UAGf;WACN,UAAU;WACV,WAAW;;;;;;WAMX,WAAW;WACX,YAAY;WACZ,KAAK;WACL;;UAGM;;WAEN;WACA;;WAEA,MAAM;;WAEN;;WAEA;;WAEA;;WAEA;;WAEA;WACA;;WAEA;WACA;WACA;WACA;WACA;;WAEA;;WAEA,4BAA4B;;;;;;;KAQ3B,mBAAmB;;UAGd;WACN,KAAK;;WAEL;;;;;;;WAOA;;WAEA;;WAEA;;;;;;;;;;;;;UAcM;WACN,gBAAgB;WAChB,aAAa;;;UAIP;WACN,MAAM;WACN,MAAM;WACN,QAAQ;WACR,OAAO;;;UAID;WACN;WACA;WACA;WACA,WAAW,kBAAkB;;UAGvB;;WAEN,kBAAkB,kBAAkB;;WAEpC,OAAO;;;;;;;;WAQP,kBAAkB;;WAElB,SAAS;;WAET,OAAO;;;;;;;;WAQP,UAAU;;;;;;WAMV;WACA,yBAAyB;WACzB,0BAA0B,SAAS;WACnC,WAAW,kBAAkB;;UAGvB;WACN;WACA;WACA;WACA;;;KAIC;;;;;;UAOK;;;;;WAKN;;WAEA;;WAEA;;WAEA;;;;;;;;WAQA;;UAGM;;;;;;WAMN,WAAW;WACX,iBAAiB,SAAS;;;;;;WAM1B,WAAW;;;;;;;WAOX,SAAS,SAAS;WAClB,YAAY;WACZ,WAAW;WACX,eAAe;WACf,YAAY;WACZ,aAAa;WACb,YAAY;WACZ,YAAY;WACZ,UAAU;WACV,OAAO,SAAS;;WAEhB;;UAGM;WACN,eAAe;;WAEf;WACA;WACA;WACA,cAAc;WACd,yBAAyB;WACzB;WACA,eAAe;WACf,UAAU;WACV,WAAW;WACX,SAAS;;WAET,gCAAgC;;WAEhC;;WAEA;;UAGM;;WAEN,kBAAkB,kBAAkB;WACpC;WACA;WACA,QAAQ;WACR,OAAO;WACP,aAAa;WACb,SAAS;WACT,UAAU;WACV,KAAK;;UAGC;WACN,eAAe;WACf;WACA,aAAa;WACb,iBAAiB;WACjB;WACA,WAAW;WACX,mBAAmB;WACnB,iBAAiB;WACjB,aAAa;WACb,qBAAqB;WACrB,eAAe;;;;;WAKf,UAAU;;;;;;;WAOV;aACE;eAA2B,OAAO;eAA2B;;aAC7D;eAAwB,OAAO;eAA2B;;;WAE5D,eAAe;WACf,kBAAkB;;;;;;;;UASZ;WACN;WACA,gBAAgB;;WAEhB,eAAe;;;KAId;UAWK;WACN,MAAM;WACN;WACA;WACA"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
+
import { a as CostLedgerHandle, c as CostLedgerSummary, p as CustomTokenPricing } from "./cost-ledger-Jk62OBfF.js";
|
|
1
2
|
import { t as AgentEvalError } from "./errors-DEE6u6ot.js";
|
|
2
|
-
import { b as CustomTokenPricing, c as CostLedgerHandle, f as CostLedgerSummary, g as CostReceiptInput, x as MaximumCharge } from "./cost-ledger-CO7UyEcQ.js";
|
|
3
3
|
//#region src/judge-families.d.ts
|
|
4
4
|
/**
|
|
5
5
|
* Judge model-family classification + cross-family enforcement.
|
|
@@ -286,7 +286,7 @@ declare class FileSystemRawProviderSink implements RawProviderSink {
|
|
|
286
286
|
*/
|
|
287
287
|
declare function providerFromBaseUrl(baseUrl: string): string;
|
|
288
288
|
//#endregion
|
|
289
|
-
//#region src/llm-client.d.ts
|
|
289
|
+
//#region src/llm-client/contract.d.ts
|
|
290
290
|
interface LlmMessage {
|
|
291
291
|
role: 'system' | 'user' | 'assistant' | 'tool';
|
|
292
292
|
/**
|
|
@@ -374,7 +374,6 @@ interface LlmChargeBounds {
|
|
|
374
374
|
/** Token rates used when the provider omits cost or package pricing does not cover the model. */
|
|
375
375
|
customTokenPricing?: CustomTokenPricing;
|
|
376
376
|
}
|
|
377
|
-
declare function maximumChargeForLlmRequest(request: Pick<LlmCallRequest, 'model' | 'messages' | 'jsonSchema' | 'tools' | 'toolChoice' | 'maxTokens' | 'thinking'>, options?: LlmChargeBounds): MaximumCharge | undefined;
|
|
378
377
|
interface LlmUsage {
|
|
379
378
|
promptTokens: number;
|
|
380
379
|
completionTokens: number;
|
|
@@ -458,29 +457,6 @@ interface LlmCallResult {
|
|
|
458
457
|
raw: Record<string, unknown>;
|
|
459
458
|
}
|
|
460
459
|
type LlmCallMetadata = Pick<LlmCallResult, 'usage' | 'costUsd' | 'model' | 'durationMs'>;
|
|
461
|
-
/** Convert a provider result into the canonical paid-call receipt input.
|
|
462
|
-
* The receipt is JSON-clean: absent optional fields are omitted, never carried
|
|
463
|
-
* as explicit-undefined keys. Receipts cross JSON boundaries (the external
|
|
464
|
-
* optimizer's loopback proxy validates them with `assertJsonValue`), where an
|
|
465
|
-
* explicit-undefined value is rejected as non-serializable. */
|
|
466
|
-
declare function costReceiptFromLlm(result: LlmCallResult, customTokenPricing?: CustomTokenPricing): CostReceiptInput;
|
|
467
|
-
/** Structured-response failures retain their completed provider receipt. */
|
|
468
|
-
declare function costReceiptFromLlmError(error: Error, customTokenPricing?: CustomTokenPricing): CostReceiptInput | undefined;
|
|
469
|
-
declare class LlmCallError extends AgentEvalError {
|
|
470
|
-
readonly status: number;
|
|
471
|
-
readonly body: string;
|
|
472
|
-
readonly model: string;
|
|
473
|
-
constructor(message: string, status: number, body: string, model: string);
|
|
474
|
-
}
|
|
475
|
-
/** A provider response completed and incurred measurable usage, but its content
|
|
476
|
-
* could not satisfy the caller's response contract. The response envelope is
|
|
477
|
-
* retained so accounting can commit the receipt before the error propagates. */
|
|
478
|
-
declare class LlmResponseError extends AgentEvalError {
|
|
479
|
-
readonly result: LlmCallResult;
|
|
480
|
-
constructor(message: string, result: LlmCallResult, options?: {
|
|
481
|
-
cause?: unknown;
|
|
482
|
-
});
|
|
483
|
-
}
|
|
484
460
|
interface LlmClientOptions extends LlmChargeBounds {
|
|
485
461
|
/** Base URL (without trailing slash), ending at the `/v1` prefix. Required: there is no default endpoint. */
|
|
486
462
|
baseUrl?: string;
|
|
@@ -557,24 +533,6 @@ interface LlmClientOptions extends LlmChargeBounds {
|
|
|
557
533
|
*/
|
|
558
534
|
assertServedModel?: boolean | AssertServedModelOptions;
|
|
559
535
|
}
|
|
560
|
-
/**
|
|
561
|
-
* True when an error is a transient transport/network fault worth retrying,
|
|
562
|
-
* as opposed to a deterministic failure (4xx schema reject, JSON parse) that
|
|
563
|
-
* a retry cannot fix. Inspects `LlmCallError.status`, then the error's
|
|
564
|
-
* name/message/code, then recurses into `error.cause` — undici nests the
|
|
565
|
-
* real socket fault one or more levels under `.cause`.
|
|
566
|
-
*
|
|
567
|
-
* This is the retry classifier for the package: `callLlm` and
|
|
568
|
-
* `withJudgeRetry` both route through it, so connection failures are treated
|
|
569
|
-
* consistently across transports.
|
|
570
|
-
*/
|
|
571
|
-
declare function isTransientLlmError(err: unknown): boolean;
|
|
572
|
-
/**
|
|
573
|
-
* Strip a ```json / ``` code fence if the model emitted one.
|
|
574
|
-
* Idempotent for naked JSON. Some models (claude-code via router, certain
|
|
575
|
-
* deepseek models) wrap output even under json_object.
|
|
576
|
-
*/
|
|
577
|
-
declare function stripFencedJson(raw: string): string;
|
|
578
536
|
//#endregion
|
|
579
537
|
//#region src/analyst/chat-client.d.ts
|
|
580
538
|
/**
|
|
@@ -868,5 +826,5 @@ interface CheckResult {
|
|
|
868
826
|
actual: string;
|
|
869
827
|
}
|
|
870
828
|
//#endregion
|
|
871
|
-
export {
|
|
872
|
-
//# sourceMappingURL=types-
|
|
829
|
+
export { providerFromBaseUrl as $, createChatClient as A, LlmToolDefinition as B, ChatResponse as C, MockTransportOpts as D, CustomTransportOpts as E, LlmMessage as F, InMemoryRawProviderSinkOptions as G, FileSystemRawProviderSink as H, LlmThinkingMode as I, RawProviderDirection as J, NoopRawProviderSink as K, LlmTokenLogprob as L, LlmCallRequest as M, LlmCallResult as N, OpenAiCompatibleTransportOpts as O, LlmChargeBounds as P, defaultProviderRedactor as Q, LlmToolCall as R, ChatRequest as S, CreateChatClientOpts as T, FileSystemRawProviderSinkOptions as U, LlmUsage as V, InMemoryRawProviderSink as W, RawProviderSink as X, RawProviderEvent as Y, RawProviderSinkFilter as Z, ScenarioResult as _, CompletionCriterion as a, ServedModelPolicy as at, ChatCallOpts as b, JudgeInput as c, assertServedModel as ct, PersonaConfig as d, servedModelAcceptable as dt, AssertCrossFamilyServedOptions as et, ProductClientConfig as f, AssertCrossFamilyOptions as ft, ScenarioFile as g, judgeFamily as gt, Scenario as h, assertCrossFamily as ht, CollectedArtifacts as i, ServedModelCheck as it, LlmCallMetadata as j, SandboxSdkTransportOpts as k, JudgeRubric as l, assertServedModels as lt, RubricDimension as m, JudgeFamily as mt, ArtifactResult as n, ModelSubstitutionError as nt, DriverState as o, ServedModelVerdict as ot, RouteMap as p, CrossFamilyError as pt, ProviderRedactor as q, CheckResult as r, ServedCrossFamilyError as rt, JudgeFn as s, assertCrossFamilyServed as st, ArtifactCheck as t, AssertServedModelOptions as tt, JudgeScore as u, checkServedModel as ut, Turn as v, ChatTransport as w, ChatClient as x, TurnResult as y, LlmToolChoice as z };
|
|
830
|
+
//# sourceMappingURL=types-C02qtnG8.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"types-C02qtnG8.d.ts","names":[],"sources":["../src/judge-families.ts","../src/integrity/served-model.ts","../src/trace/raw-provider-sink.ts","../src/llm-client/contract.ts","../src/analyst/chat-client.ts","../src/types.ts"],"mappings":";;;;;;;;;;;;;KAWY;;;;;;;iBAsEI,YAAY,kBAAkB;UAc7B;;EAEf;;;;EAIA;;cAGW,yBAAyB;WAGlB,UAAU;WACV;EAHlB,YACE,iBACgB,UAAU,eACV;;;;;;;;;;;;;;;iBAoBJ,kBACd,kBACA,OAAM,2BACL;;;;KCvFS;;;;;;;;;;;UAYK;;EAEf;;EAEA;EACA,iBAAiB;;EAEjB,cAAc;EACd,SAAS;;EAET;;;;;;;;;;;;iBA4Cc,iBACd,mBACA,oCACC;cA4CU,+BAA+B;WAGxB,QAAQ,cAAc;EAFxC,YACE,iBACgB,QAAQ,cAAc;;;;;;;;;KAc9B;UAWK;;;;;;EAMf;;;;;EAKA;;EAEA;;;;;;;iBAQc,sBACd,OAAO,kBACP,OAAM;;;;;;iBA6BQ,kBACd,mBACA,mCACA,OAAM,2BACL;;;;;iBAea,mBACd,OAAO;EAAgB;EAAmB;IAC1C,OAAM,2BACL;UAec,uCAAuC;;EAEtD;;EAEA;;cAGW,+BAA+B;WAGxB,UAAU;WACV,QAAQ,cAAc;EAHxC,YACE,iBACgB,UAAU,eACV,QAAQ,cAAc;;;;;;;;;;;iBAgB1B,wBACd,OAAO;EAAgB;EAAmB;IAC1C,OAAM,iCACL;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KCjRS;UAEK;;EAEf;;EAEA;EACA;;;;;;EAMA;EACA;;EAEA;;EAEA;;EAEA;EACA,WAAW;;EAEX;;EAEA;EACA;EACA,iBAAiB;EACjB;EACA,kBAAkB;EAClB;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA,YAAY;EACZ;;UAGe;EACf,OAAO,OAAO,mBAAmB;;EAEjC,MAAM,SAAS,wBAAwB,QAAQ;;EAE/C,UAAU;;KAGA,oBAAoB,OAAO,qBAAqB;;;;;iBAM5C,wBAAwB,OAAO,mBAAmB;UAqBjD;EACf,WAAW;;cAGA,mCAAmC;UACtC;UACA;EAER,YAAY,OAAM;EAIZ,OAAO,OAAO,mBAAmB;EAIjC,KAAK,SAAQ,wBAA6B,QAAQ;EAUxD;;cAKW,+BAA+B;EACpC,UAAU;;;;;;;EASV,QAAQ,QAAQ;;UAOP;;EAEf;;EAEA;;EAEA;EACA,WAAW;;cAGA,qCAAqC;UACxC;UACA;UACA;UACA;UACA;UACA;UACA;EAER,YAAY,MAAM;UAOJ;UAON;EAKF,OAAO,OAAO,mBAAmB;EAYjC,KAAK,SAAQ,wBAA6B,QAAQ;;;;;;iBAkC1C,oBAAoB;;;UC5OnB;EACf;;;;;EAKA,kBAEI;IACM;IAAc;;IACd;IAAmB;MAAa;MAAa;;;;EAGvD,YAAY;;EAEZ;;KAGU;;UAGK;EACf;EACA;IACE;IACA;IACA,YAAY;;;;KAKJ;EAIN;EAAkB;IAAY;;;;UAGnB;EACf;EACA;;EAEA;;UAGe;EACf;EACA,UAAU;;EAEV;;EAEA;IAAe;IAAc,QAAQ;;;EAErC,QAAQ;;EAER,aAAa;EACb;EACA;;;;;EAKA;IAAa;;;EAEb,WAAW;;EAEX;;;;;;UAOe;;EAEf;;EAEA;;EAEA,WAAW;;EAEX,qBAAqB;;UAGN;EACf;EACA;EACA;;EAEA;;EAEA;;EAEA;;;;UAKe;EACf;EACA;EACA,KAAK;IAAgB;IAAe;;;UAGrB;;EAEf;;EAEA,YAAY;EACZ,OAAO;;;;;EAKP;;;;;;;;;EASA;;;;;;;;;;;;EAYA;;EAEA;;;;;;;;;EASA;;;;;;;EAOA;;;;;;;;EAQA,WAAW,cAAc;;EAEzB,KAAK;;KAGK,kBAAkB,KAAK;UAElB,yBAAyB;;EAExC;;EAEA;EACA;;EAEA;IAAe;IAAc;;;EAE7B;;EAEA;;;;;;;EAOA;;;;;;;EAOA,SAAS;;;;;;;;EAQT;;;;;;EAMA;;EAEA,eAAe;;;;;;;;EAQf,UAAU;;;;;EAKV;;EAEA;IAAiB;IAAgB;;;EAEjC,WAAW;;;;;;;;;;EAUX,8BAA8B;;;;;;;UCzMf;;WAEN,WAAW;;WAEX;;WAEA;;EAGT,KAAK,KAAK,aAAa,OAAO,eAAe,QAAQ;;KAG3C;UAMK,oBAAoB,KAAK;;EAExC;;KAGU,eAAe;UAEV;;EAEf,SAAS;;EAET;;EAEA;;EAEA;;KAKU,uBACR,0BACA,sBACA,gCACA;UAEM;EACR;;EAEA;;;;;;UAOe,gCAAgC;EAC/C;EACA,OAAO,KAAK,aAAa,OAAO,iBAAiB,QAAQ;;;UAI1C,4BAA4B;EAC3C;EACA,OAAO,KAAK,aAAa,OAAO,iBAAiB,QAAQ;;;;;;;;;;;;;;;;;;;;UAqB1C,sCACP,mBACN,KACE;EAcJ;;EAEA;;EAEA;;EAEA;;EAEA;IAAe;IAAc;;;;;;;UAOd,0BAA0B;EACzC;EACA,UAAU,KAAK,aAAa,OAAO,iBAAiB,QAAQ;;;;;;iBAO9C,iBAAiB,MAAM,uBAAuB;;;UChK7C;EACf;EACA;EACA;EACA;EACA;EACA,OAAO;EACP,gBAAgB;EAChB;;UAGe;EACf;EACA;EACA;EACA;;UAKe;EACf;EAQA;EACA;EACA;EACA;;UAKe;EACf;EACA;EACA,YAAY;;UAGG;EACf;EACA;EACA;EACA;EACA;;UAKe;EACf;EACA;EACA,OAAO;EACP,iBAAiB;EACjB,aAAa;EACb;EACA;EACA;EACA,WAAW;;EAEX,OAAO;;UAGQ;EACf;EACA;EACA;EACA;EACA;IAAmB;IAAc;;EACjC;EACA;;UAGe;EACf,OAAO;EACP;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;IAAc;IAAc;;EAC5B;IAAmB;IAAc,QAAQ;;EACzC;IAAc;IAAkB;;EAChC;;UAKe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;GACC;;UAGc;EACf;EACA,QAAQ;;EAER;;UAKe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;IACE;MACE;MACA;MACA;;;EAGJ,OAAO;EACP,gBAAgB;;UAKD;EACf;EACA,QAAQ,OAAO;EACf,YAAY,OAAO;;;;;;;;;;KAWT;UAEK;EACf;EACA;EACA;EACA,oBAAoB;EACpB;;EAEA,QAAQ;;;;;;;EAOR;;;;;;;EAOA;;;;;;EAMA;;UAGe;EACf;EACA;EACA;IAAa;IAAiB;IAAkB;;EAChD;EACA;EACA;;UAGe;EACf,UAAU;EACV,OAAO;EACP,WAAW;;EAEX,aAAa;EACb;EACA,WAAW;EACX,SAAS;;KAGC,WAAW,MAAM,YAAY,OAAO,eAAe,QAAQ;UAYtD;EACf;EACA;EACA;EACA"}
|
|
@@ -1,6 +1,7 @@
|
|
|
1
|
-
import {
|
|
2
|
-
import { a as RunRecord, d as RunTokenUsage } from "./run-record-
|
|
3
|
-
import { c as JudgeInput, x as ChatClient } from "./types-
|
|
1
|
+
import { a as CostLedgerHandle, l as CostProvenance } from "./cost-ledger-Jk62OBfF.js";
|
|
2
|
+
import { a as RunRecord, d as RunTokenUsage } from "./run-record-C1E24G2A.js";
|
|
3
|
+
import { c as JudgeInput, x as ChatClient } from "./types-C02qtnG8.js";
|
|
4
|
+
import { z } from "zod";
|
|
4
5
|
import { SpanKind } from "@tangle-network/agent-trace-contract";
|
|
5
6
|
import { RE2JS } from "re2js";
|
|
6
7
|
//#region src/trace-analyst/types.d.ts
|
|
@@ -289,13 +290,326 @@ interface BoundedTraceAnalysisStoreOptions {
|
|
|
289
290
|
/** Apply the public validation, cancellation, not-found, and size rules to any adapter. */
|
|
290
291
|
declare function createBoundedTraceAnalysisStore(source: TraceAnalysisStore, options?: BoundedTraceAnalysisStoreOptions): TraceAnalysisStore;
|
|
291
292
|
//#endregion
|
|
293
|
+
//#region src/analyst/finding-scope.d.ts
|
|
294
|
+
/**
|
|
295
|
+
* The layer ladder, lowest first. `problem` is what discovery-lab calls a
|
|
296
|
+
* theatre. `meta` sits outside the ladder: it is about the instrument, not the
|
|
297
|
+
* agents, so it never promotes and never bounds another layer.
|
|
298
|
+
*/
|
|
299
|
+
declare const FINDING_LAYERS: readonly ['event', 'turn', 'agent', 'team', 'run', 'play', 'problem', 'domain', 'lab', 'meta'];
|
|
300
|
+
type FindingLayer = (typeof FINDING_LAYERS)[number];
|
|
301
|
+
/** Rank on the ladder; `meta` has none. */
|
|
302
|
+
declare function findingLayerRank(layer: FindingLayer): number | null;
|
|
303
|
+
interface FindingScope {
|
|
304
|
+
/**
|
|
305
|
+
* Ids of the measured units. A population is stored resolved, never as a
|
|
306
|
+
* query: the ids are what `n`, `evidence_units` and promotion's lineage
|
|
307
|
+
* check are compared against, and a query re-run later can select a
|
|
308
|
+
* different set. Select populations with `queryFindings` (layer, subject,
|
|
309
|
+
* domain path, analyst) and store the ids it returned.
|
|
310
|
+
*/
|
|
311
|
+
population: string[];
|
|
312
|
+
/** Units measured. Below 2, readers label the finding an anecdote. */
|
|
313
|
+
n: number;
|
|
314
|
+
/** ISO-8601 bounds of the measured activity. */
|
|
315
|
+
window: {
|
|
316
|
+
from: string;
|
|
317
|
+
to: string;
|
|
318
|
+
};
|
|
319
|
+
/** Domain taxonomy path of any depth, broadest first. */
|
|
320
|
+
domain_path?: string[];
|
|
321
|
+
}
|
|
322
|
+
interface FindingCalibration {
|
|
323
|
+
/** Gold items the judge was scored against. */
|
|
324
|
+
gold_n: number;
|
|
325
|
+
/** Share of gold items where the judge matched the gold label. */
|
|
326
|
+
agreement: number;
|
|
327
|
+
/** Chance-corrected agreement (Cohen's κ). */
|
|
328
|
+
kappa: number;
|
|
329
|
+
}
|
|
330
|
+
interface FindingMethod {
|
|
331
|
+
kind: 'computed' | 'measured' | 'judged';
|
|
332
|
+
name: string;
|
|
333
|
+
version: string;
|
|
334
|
+
calibration?: FindingCalibration;
|
|
335
|
+
}
|
|
336
|
+
interface FindingMeasure {
|
|
337
|
+
name: string;
|
|
338
|
+
value: number;
|
|
339
|
+
unit: string;
|
|
340
|
+
interval?: [number, number];
|
|
341
|
+
/** Confidence level of `interval`, in (0, 1). */
|
|
342
|
+
level?: number;
|
|
343
|
+
}
|
|
344
|
+
/**
|
|
345
|
+
* One measured unit. `derived_from` is the unit this one inherited state from
|
|
346
|
+
* (run 2 built on run 1's code): `null` records an independent start, and an
|
|
347
|
+
* absent field records lineage nobody captured. Unknown lineage counts as
|
|
348
|
+
* shared, so it can never manufacture independent support for a claim.
|
|
349
|
+
*/
|
|
350
|
+
interface FindingEvidenceUnit {
|
|
351
|
+
id: string;
|
|
352
|
+
layer: FindingLayer;
|
|
353
|
+
derived_from?: string | null;
|
|
354
|
+
}
|
|
355
|
+
/** The one validator for stored findings of either schema version. Unknown fields fail. */
|
|
356
|
+
declare const AnalystFindingSchema: z.ZodObject<{
|
|
357
|
+
schema_version: z.ZodEnum<{
|
|
358
|
+
"1.0.0": "1.0.0";
|
|
359
|
+
"1.1.0": "1.1.0";
|
|
360
|
+
}>;
|
|
361
|
+
finding_id: z.ZodString;
|
|
362
|
+
analyst_id: z.ZodString;
|
|
363
|
+
produced_at: z.ZodString;
|
|
364
|
+
severity: z.ZodEnum<{
|
|
365
|
+
critical: "critical";
|
|
366
|
+
high: "high";
|
|
367
|
+
info: "info";
|
|
368
|
+
low: "low";
|
|
369
|
+
medium: "medium";
|
|
370
|
+
}>;
|
|
371
|
+
area: z.ZodString;
|
|
372
|
+
claim: z.ZodString;
|
|
373
|
+
rationale: z.ZodOptional<z.ZodString>;
|
|
374
|
+
evidence_refs: z.ZodArray<z.ZodObject<{
|
|
375
|
+
kind: z.ZodEnum<{
|
|
376
|
+
artifact: "artifact";
|
|
377
|
+
event: "event";
|
|
378
|
+
finding: "finding";
|
|
379
|
+
metric: "metric";
|
|
380
|
+
span: "span";
|
|
381
|
+
}>;
|
|
382
|
+
uri: z.ZodString;
|
|
383
|
+
excerpt: z.ZodOptional<z.ZodString>;
|
|
384
|
+
}, z.core.$strict>>;
|
|
385
|
+
recommended_action: z.ZodOptional<z.ZodString>;
|
|
386
|
+
validation_plan: z.ZodOptional<z.ZodString>;
|
|
387
|
+
confidence: z.ZodNumber;
|
|
388
|
+
subject: z.ZodOptional<z.ZodString>;
|
|
389
|
+
derived_from_judge: z.ZodOptional<z.ZodBoolean>;
|
|
390
|
+
metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
391
|
+
layer: z.ZodOptional<z.ZodEnum<{
|
|
392
|
+
agent: "agent";
|
|
393
|
+
domain: "domain";
|
|
394
|
+
event: "event";
|
|
395
|
+
lab: "lab";
|
|
396
|
+
meta: "meta";
|
|
397
|
+
play: "play";
|
|
398
|
+
problem: "problem";
|
|
399
|
+
run: "run";
|
|
400
|
+
team: "team";
|
|
401
|
+
turn: "turn";
|
|
402
|
+
}>>;
|
|
403
|
+
scope: z.ZodOptional<z.ZodObject<{
|
|
404
|
+
population: z.ZodArray<z.ZodString>;
|
|
405
|
+
n: z.ZodNumber;
|
|
406
|
+
window: z.ZodObject<{
|
|
407
|
+
from: z.ZodString;
|
|
408
|
+
to: z.ZodString;
|
|
409
|
+
}, z.core.$strict>;
|
|
410
|
+
domain_path: z.ZodOptional<z.ZodArray<z.ZodString>>;
|
|
411
|
+
}, z.core.$strict>>;
|
|
412
|
+
method: z.ZodOptional<z.ZodObject<{
|
|
413
|
+
kind: z.ZodEnum<{
|
|
414
|
+
computed: "computed";
|
|
415
|
+
judged: "judged";
|
|
416
|
+
measured: "measured";
|
|
417
|
+
}>;
|
|
418
|
+
name: z.ZodString;
|
|
419
|
+
version: z.ZodString;
|
|
420
|
+
calibration: z.ZodOptional<z.ZodObject<{
|
|
421
|
+
gold_n: z.ZodNumber;
|
|
422
|
+
agreement: z.ZodNumber;
|
|
423
|
+
kappa: z.ZodNumber;
|
|
424
|
+
}, z.core.$strict>>;
|
|
425
|
+
}, z.core.$strict>>;
|
|
426
|
+
measure: z.ZodOptional<z.ZodObject<{
|
|
427
|
+
name: z.ZodString;
|
|
428
|
+
value: z.ZodNumber;
|
|
429
|
+
unit: z.ZodString;
|
|
430
|
+
interval: z.ZodOptional<z.ZodTuple<[z.ZodNumber, z.ZodNumber], null>>;
|
|
431
|
+
level: z.ZodOptional<z.ZodNumber>;
|
|
432
|
+
}, z.core.$strict>>;
|
|
433
|
+
confounders: z.ZodOptional<z.ZodArray<z.ZodString>>;
|
|
434
|
+
valid_for: z.ZodOptional<z.ZodString>;
|
|
435
|
+
claim_id: z.ZodOptional<z.ZodString>;
|
|
436
|
+
evidence_units: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
437
|
+
id: z.ZodString;
|
|
438
|
+
layer: z.ZodEnum<{
|
|
439
|
+
agent: "agent";
|
|
440
|
+
domain: "domain";
|
|
441
|
+
event: "event";
|
|
442
|
+
lab: "lab";
|
|
443
|
+
meta: "meta";
|
|
444
|
+
play: "play";
|
|
445
|
+
problem: "problem";
|
|
446
|
+
run: "run";
|
|
447
|
+
team: "team";
|
|
448
|
+
turn: "turn";
|
|
449
|
+
}>;
|
|
450
|
+
derived_from: z.ZodOptional<z.ZodNullable<z.ZodString>>;
|
|
451
|
+
}, z.core.$strict>>>;
|
|
452
|
+
}, z.core.$strict>;
|
|
453
|
+
/** A finding admitted as candidate-generation input; same rules plus its origin. */
|
|
454
|
+
declare const ProposalFindingSchema: z.ZodObject<{
|
|
455
|
+
schema_version: z.ZodEnum<{
|
|
456
|
+
"1.0.0": "1.0.0";
|
|
457
|
+
"1.1.0": "1.1.0";
|
|
458
|
+
}>;
|
|
459
|
+
finding_id: z.ZodString;
|
|
460
|
+
analyst_id: z.ZodString;
|
|
461
|
+
produced_at: z.ZodString;
|
|
462
|
+
severity: z.ZodEnum<{
|
|
463
|
+
critical: "critical";
|
|
464
|
+
high: "high";
|
|
465
|
+
info: "info";
|
|
466
|
+
low: "low";
|
|
467
|
+
medium: "medium";
|
|
468
|
+
}>;
|
|
469
|
+
area: z.ZodString;
|
|
470
|
+
claim: z.ZodString;
|
|
471
|
+
rationale: z.ZodOptional<z.ZodString>;
|
|
472
|
+
evidence_refs: z.ZodArray<z.ZodObject<{
|
|
473
|
+
kind: z.ZodEnum<{
|
|
474
|
+
artifact: "artifact";
|
|
475
|
+
event: "event";
|
|
476
|
+
finding: "finding";
|
|
477
|
+
metric: "metric";
|
|
478
|
+
span: "span";
|
|
479
|
+
}>;
|
|
480
|
+
uri: z.ZodString;
|
|
481
|
+
excerpt: z.ZodOptional<z.ZodString>;
|
|
482
|
+
}, z.core.$strict>>;
|
|
483
|
+
recommended_action: z.ZodOptional<z.ZodString>;
|
|
484
|
+
validation_plan: z.ZodOptional<z.ZodString>;
|
|
485
|
+
confidence: z.ZodNumber;
|
|
486
|
+
subject: z.ZodOptional<z.ZodString>;
|
|
487
|
+
derived_from_judge: z.ZodOptional<z.ZodBoolean>;
|
|
488
|
+
metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
489
|
+
layer: z.ZodOptional<z.ZodEnum<{
|
|
490
|
+
agent: "agent";
|
|
491
|
+
domain: "domain";
|
|
492
|
+
event: "event";
|
|
493
|
+
lab: "lab";
|
|
494
|
+
meta: "meta";
|
|
495
|
+
play: "play";
|
|
496
|
+
problem: "problem";
|
|
497
|
+
run: "run";
|
|
498
|
+
team: "team";
|
|
499
|
+
turn: "turn";
|
|
500
|
+
}>>;
|
|
501
|
+
scope: z.ZodOptional<z.ZodObject<{
|
|
502
|
+
population: z.ZodArray<z.ZodString>;
|
|
503
|
+
n: z.ZodNumber;
|
|
504
|
+
window: z.ZodObject<{
|
|
505
|
+
from: z.ZodString;
|
|
506
|
+
to: z.ZodString;
|
|
507
|
+
}, z.core.$strict>;
|
|
508
|
+
domain_path: z.ZodOptional<z.ZodArray<z.ZodString>>;
|
|
509
|
+
}, z.core.$strict>>;
|
|
510
|
+
method: z.ZodOptional<z.ZodObject<{
|
|
511
|
+
kind: z.ZodEnum<{
|
|
512
|
+
computed: "computed";
|
|
513
|
+
judged: "judged";
|
|
514
|
+
measured: "measured";
|
|
515
|
+
}>;
|
|
516
|
+
name: z.ZodString;
|
|
517
|
+
version: z.ZodString;
|
|
518
|
+
calibration: z.ZodOptional<z.ZodObject<{
|
|
519
|
+
gold_n: z.ZodNumber;
|
|
520
|
+
agreement: z.ZodNumber;
|
|
521
|
+
kappa: z.ZodNumber;
|
|
522
|
+
}, z.core.$strict>>;
|
|
523
|
+
}, z.core.$strict>>;
|
|
524
|
+
measure: z.ZodOptional<z.ZodObject<{
|
|
525
|
+
name: z.ZodString;
|
|
526
|
+
value: z.ZodNumber;
|
|
527
|
+
unit: z.ZodString;
|
|
528
|
+
interval: z.ZodOptional<z.ZodTuple<[z.ZodNumber, z.ZodNumber], null>>;
|
|
529
|
+
level: z.ZodOptional<z.ZodNumber>;
|
|
530
|
+
}, z.core.$strict>>;
|
|
531
|
+
confounders: z.ZodOptional<z.ZodArray<z.ZodString>>;
|
|
532
|
+
valid_for: z.ZodOptional<z.ZodString>;
|
|
533
|
+
claim_id: z.ZodOptional<z.ZodString>;
|
|
534
|
+
evidence_units: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
535
|
+
id: z.ZodString;
|
|
536
|
+
layer: z.ZodEnum<{
|
|
537
|
+
agent: "agent";
|
|
538
|
+
domain: "domain";
|
|
539
|
+
event: "event";
|
|
540
|
+
lab: "lab";
|
|
541
|
+
meta: "meta";
|
|
542
|
+
play: "play";
|
|
543
|
+
problem: "problem";
|
|
544
|
+
run: "run";
|
|
545
|
+
team: "team";
|
|
546
|
+
turn: "turn";
|
|
547
|
+
}>;
|
|
548
|
+
derived_from: z.ZodOptional<z.ZodNullable<z.ZodString>>;
|
|
549
|
+
}, z.core.$strict>>>;
|
|
550
|
+
proposal_origin: z.ZodEnum<{
|
|
551
|
+
production: "production";
|
|
552
|
+
search: "search";
|
|
553
|
+
}>;
|
|
554
|
+
}, z.core.$strict>;
|
|
555
|
+
type FindingParseOutcome = {
|
|
556
|
+
succeeded: true;
|
|
557
|
+
value: AnalystFinding;
|
|
558
|
+
} | {
|
|
559
|
+
succeeded: false;
|
|
560
|
+
error: {
|
|
561
|
+
path: string;
|
|
562
|
+
message: string;
|
|
563
|
+
};
|
|
564
|
+
};
|
|
565
|
+
/** Validate one stored finding of schema 1.0.0 or 1.1.0. */
|
|
566
|
+
declare function parseAnalystFinding(value: unknown): FindingParseOutcome;
|
|
567
|
+
/**
|
|
568
|
+
* A finding's scope as a reader sees it. A 1.0.0 finding states neither layer
|
|
569
|
+
* nor population, so both read `unknown` — never a guessed value.
|
|
570
|
+
*/
|
|
571
|
+
interface FindingScopeReading {
|
|
572
|
+
layer: FindingLayer | 'unknown';
|
|
573
|
+
n: number | 'unknown';
|
|
574
|
+
/** Units with distinct lineage roots; unknown lineage counts as one shared root. */
|
|
575
|
+
independent_n: number | 'unknown';
|
|
576
|
+
domain_path: readonly string[] | null;
|
|
577
|
+
labels: FindingReaderLabel[];
|
|
578
|
+
}
|
|
579
|
+
/**
|
|
580
|
+
* - `unscoped`: a 1.0.0 finding with no layer or population.
|
|
581
|
+
* - `anecdote`: measured on fewer than two units.
|
|
582
|
+
* - `uncalibrated`: judged, with no calibration against gold labels.
|
|
583
|
+
* - `lineage-unknown`: at least one unit has no recorded lineage.
|
|
584
|
+
*/
|
|
585
|
+
type FindingReaderLabel = 'unscoped' | 'anecdote' | 'uncalibrated' | 'lineage-unknown';
|
|
586
|
+
declare function readFindingScope(finding: AnalystFinding): FindingScopeReading;
|
|
587
|
+
/** Root token shared by every unit whose lineage was not recorded. */
|
|
588
|
+
declare const UNKNOWN_LINEAGE = "?unknown-lineage";
|
|
589
|
+
/**
|
|
590
|
+
* Lineage root of each unit, keyed by unit id. A unit derived from another
|
|
591
|
+
* unit in the list shares that unit's root; a unit derived from an id outside
|
|
592
|
+
* the list takes that id as its root; unknown lineage maps to one shared root.
|
|
593
|
+
*/
|
|
594
|
+
declare function lineageRoots(units: ReadonlyArray<FindingEvidenceUnit>): Map<string, string>;
|
|
595
|
+
/** Units with distinct lineage roots. */
|
|
596
|
+
declare function independentUnitCount(units: ReadonlyArray<FindingEvidenceUnit>): number;
|
|
597
|
+
/** One entry per unit id, sorted; a later copy that records lineage fills an earlier unknown. */
|
|
598
|
+
declare function mergeEvidenceUnits(units: readonly FindingEvidenceUnit[]): FindingEvidenceUnit[];
|
|
599
|
+
//#endregion
|
|
292
600
|
//#region src/analyst/types.d.ts
|
|
293
601
|
/**
|
|
294
602
|
* Unified envelope every analyst emits. Schema-versioned so renderers
|
|
295
603
|
* and time-series diffs survive future field additions.
|
|
604
|
+
*
|
|
605
|
+
* 1.1.0 adds what a finding is about and what it was measured on: `layer`,
|
|
606
|
+
* `scope`, `method`, `measure`, `confounders`, `valid_for`, `claim_id` and
|
|
607
|
+
* `evidence_units`. They are all optional, and a 1.0.0 finding carries none of
|
|
608
|
+
* them; `readFindingScope` reads such a finding as layer and n unknown.
|
|
609
|
+
* `parseAnalystFinding` (finding-scope.ts) is the one validator for both.
|
|
296
610
|
*/
|
|
297
611
|
interface AnalystFinding {
|
|
298
|
-
schema_version:
|
|
612
|
+
schema_version: AnalystFindingSchemaVersion;
|
|
299
613
|
/**
|
|
300
614
|
* Stable hash over identity-defining fields (analyst_id + canonical
|
|
301
615
|
* claim + area + optional subject). Two findings from two runs that
|
|
@@ -332,7 +646,32 @@ interface AnalystFinding {
|
|
|
332
646
|
derived_from_judge?: boolean;
|
|
333
647
|
/** Analyst-private extras; renderers ignore unless they know the analyst. */
|
|
334
648
|
metadata?: Record<string, unknown>;
|
|
649
|
+
/** 1.1.0: the level the claim is about, from one event up to the whole lab. */
|
|
650
|
+
layer?: FindingLayer;
|
|
651
|
+
/** 1.1.0: the population the claim was measured on. */
|
|
652
|
+
scope?: FindingScope;
|
|
653
|
+
/** 1.1.0: how the claim was produced, and the judge's calibration when judged. */
|
|
654
|
+
method?: FindingMethod;
|
|
655
|
+
/** 1.1.0: the number behind the claim, when it has one. */
|
|
656
|
+
measure?: FindingMeasure;
|
|
657
|
+
/** 1.1.0: known differences between the compared units besides the treatment. */
|
|
658
|
+
confounders?: string[];
|
|
659
|
+
/** 1.1.0: the conditions the claim was measured under, in words. */
|
|
660
|
+
valid_for?: string;
|
|
661
|
+
/**
|
|
662
|
+
* 1.1.0: stable key for the claim across subjects and layers. `promote`
|
|
663
|
+
* groups children by it; a finding without one never lifts.
|
|
664
|
+
*/
|
|
665
|
+
claim_id?: string;
|
|
666
|
+
/**
|
|
667
|
+
* 1.1.0: the measured units with their lineage. When present, `scope.n` and
|
|
668
|
+
* `scope.population` must match them and no unit may sit above `layer`.
|
|
669
|
+
*/
|
|
670
|
+
evidence_units?: FindingEvidenceUnit[];
|
|
335
671
|
}
|
|
672
|
+
/** Every finding schema this package reads. Writers emit 1.1.0 only when a 1.1.0 field is set. */
|
|
673
|
+
declare const ANALYST_FINDING_SCHEMA_VERSIONS: readonly ['1.0.0', '1.1.0'];
|
|
674
|
+
type AnalystFindingSchemaVersion = (typeof ANALYST_FINDING_SCHEMA_VERSIONS)[number];
|
|
336
675
|
/** Finding severity. `AnalystSeverity` derives from this array, so the type
|
|
337
676
|
* and the schema that validates a finding cannot name different levels. */
|
|
338
677
|
declare const ANALYST_SEVERITIES: readonly ['critical', 'high', 'medium', 'low', 'info'];
|
|
@@ -549,6 +888,8 @@ declare function computeFindingId(input: {
|
|
|
549
888
|
/** Override the claim for hashing — use when the displayed claim has run-specific bits. */
|
|
550
889
|
id_basis?: string;
|
|
551
890
|
}): string;
|
|
891
|
+
/** Fields that exist only in schema 1.1.0. */
|
|
892
|
+
declare const FINDING_SCOPE_FIELDS: readonly ["layer", "scope", "method", "measure", "confounders", "valid_for", "claim_id", "evidence_units"];
|
|
552
893
|
/**
|
|
553
894
|
* Convenience factory: produce a fully-formed AnalystFinding with the
|
|
554
895
|
* id computed automatically. Analyst code stays terse.
|
|
@@ -629,5 +970,5 @@ type AnalystRunEvent = {
|
|
|
629
970
|
result: AnalystRunResult;
|
|
630
971
|
};
|
|
631
972
|
//#endregion
|
|
632
|
-
export {
|
|
633
|
-
//# sourceMappingURL=types-
|
|
973
|
+
export { DatasetOverview as $, FindingMeasure as A, lineageRoots as B, makeFinding as C, FindingCalibration as D, FINDING_LAYERS as E, FindingScopeReading as F, BoundedTraceAnalysisStoreOptions as G, parseAnalystFinding as H, ProposalFindingSchema as I, SpanSourceReader as J, ReadSpanSourceInput as K, UNKNOWN_LINEAGE as L, FindingParseOutcome as M, FindingReaderLabel as N, FindingEvidenceUnit as O, FindingScope as P, DEFAULT_TRACE_ANALYST_BUDGETS as Q, findingLayerRank as R, computeFindingId as S, AnalystFindingSchema as T, readFindingScope as U, mergeEvidenceUnits as V, createBoundedTraceAnalysisStore as W, TraceAnalysisStore as X, TRACE_ANALYSIS_LIMITS as Y, TraceAnalysisStoreContext as Z, ExecutionProbeOutcome as _, AnalystFinding as a, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX as at, ProposalFinding as b, AnalystRequirements as c, TraceAnalystSpan as ct, AnalystRunResult as d, TraceAnalystTraceSummary as dt, ErrorCluster as et, AnalystRunSummary as f, ViewSpansResult as ft, ExecutionProbe as g, EvidenceRef as h, AnalystCost as i, SpanMatchRecord as it, FindingMethod as j, FindingLayer as k, AnalystRunEvent as l, TraceAnalystSpanKind as lt, AnalystUsageReceipt as m, ViewTraceResult as mt, Analyst as n, SearchSpanResult as nt, AnalystFindingSchemaVersion as o, TraceAnalystByteBudgets as ot, AnalystSeverity as p, ViewTraceOversized as pt, ReadSpanSourceResult as q, AnalystContext as r, SearchTraceResult as rt, AnalystInputKind as s, TraceAnalystFilters as st, ANALYST_FINDING_SCHEMA_VERSIONS as t, QueryTracesPage as tt, AnalystRunInputs as u, TraceAnalystSpanStatus as ut, ExecutionProbeRequest as v, makeProposalFinding as w, ProposalFindingOrigin as x, FINDING_SCOPE_FIELDS as y, independentUnitCount as z };
|
|
974
|
+
//# sourceMappingURL=types-C1tnkEgM.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"types-C1tnkEgM.d.ts","names":[],"sources":["../src/trace-analyst/types.ts","../src/trace-analyst/store-contract.ts","../src/trace-analyst/store-boundary.ts","../src/analyst/finding-scope.ts","../src/analyst/types.ts"],"mappings":";;;;;;;;;;;;KAsBY,uBAAuB;KAEvB;;;;UAKK;EACf;EACA;EACA;EACA;EACA,MAAM;EACN;EACA;EACA;EACA,QAAQ;EACR;EACA;EACA;EACA;EACA;;;EAGA,YAAY;;UAGG;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA;;;;;;;UAQe;;EAEf;;EAEA;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA;IAAU;IAAqB;;;;;;EAK/B,gBAAgB;EAChB;IAAc;IAAkB;;;UAGjB;EACf,QAAQ;EACR;EACA;;;;;UAMe;EACf;EACA,QAAQ;EACR,YAAY;;UAGG;EACf;;EAEA,gBAAgB;;EAEhB;EACA;;UAGe;EACf;EACA,OAAO;;EAEP;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA,WAAW;;;EAGX;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA,MAAM;EACN;;UAGe;EACf;EACA;EACA,MAAM;EACN;;;UAIe;;;EAGf;;;EAGA;;;EAGA;;;EAGA;;cAGW,+BAA+B;;;cAS/B;;;cC3MA;WACX;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;UAGe;EACf,SAAS;;;UAIM;EACf;EACA;EACA;EACA;EACA;EACA;;KAGU;EAEN;EACA;EACA;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;IACE;IACA;IACA;IACA;IACA;;;;KAKI,oBACV,OAAO,qBACP,UAAU,8BACP,QAAQ;;;;;;;;UASI;;EAEf,iBAAiB;EAEjB,SAAS,kBAAkB,UAAU,4BAA4B;EAEjE,SACE;IAAS;IAAkB;KAC3B,UAAU,4BACT;EAEH,YACE,UAAU,qBACV,UAAU,4BACT,QAAQ;EAEX,YACE;IAAS,UAAU;IAAqB;IAAe;KACvD,UAAU,4BACT,QAAQ;EAEX,YAAY,UAAU,qBAAqB,UAAU,4BAA4B;EAEjF,UACE;IACE;IACA;KAEF,UAAU,4BACT,QAAQ;EAEX,UACE;IACE;IACA;IACA;KAEF,UAAU,4BACT,QAAQ;EAEX,YACE;IACE;IACA;IACA;KAEF,UAAU,4BACT,QAAQ;EAEX,WACE;IACE;IACA;IACA;IACA;KAEF,UAAU,4BACT,QAAQ;;UAGI;EACf,UAAU,QAAQ;;;;;iBC1GJ,gCACd,QAAQ,oBACR,UAAS,mCACR;;;;;;;;cCCU;KAaD,uBAAuB;;iBAGnB,iBAAiB,OAAO;UAIvB;;;;;;;;EAQf;;EAEA;;EAEA;IAAU;IAAc;;;EAExB;;UAGe;;EAEf;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA,cAAc;;UAGC;EACf;EACA;EACA;EACA;;EAEA;;;;;;;;UASe;EACf;EACA,OAAO;EACP;;;cAoKW,sBAAoB,EAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAA0C,EAAA,KAAA;;cAG9D,uBAAqB,EAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAGN,EAAA,KAAA;KAQhB;EACN;EAAiB,OAAO;;EACxB;EAAkB;IAAS;IAAc;;;;iBAG/B,oBAAoB,iBAAiB;;;;;UAiBpC;EACf,OAAO;EACP;;EAEA;EACA;EACA,QAAQ;;;;;;;;KASE;iBAEI,iBAAiB,SAAS,iBAAiB;;cA2B9C;;;;;;iBAOG,aAAa,OAAO,cAAc,uBAAuB;;iBA0BzD,qBAAqB,OAAO,cAAc;;iBAK1C,mBAAmB,gBAAgB,wBAAwB;;;;;;;;;;;;;UC5V1D;EACf,gBAAgB;;;;;;;EAOhB;EACA;EACA;EACA,UAAU;;;;;;;EAOV;EACA;EACA;EACA,eAAe;EACf;EACA;;EAEA;;;;;;EAMA;;;;EAIA;;EAEA,WAAW;;EAEX,QAAQ;;EAER,QAAQ;;EAER,SAAS;;EAET,UAAU;;EAEV;;EAEA;;;;;EAKA;;;;;EAKA,iBAAiB;;;cAIN;KAED,sCAAsC;;;cAIrC;KAED,0BAA0B;;KAG1B;;KAGA,kBAAkB;WACnB,iBAAiB;;UAGX;;;;;;;EAOf;EACA;EACA;;;;;;;;KAWU;UAOK;;EAEf;;EAEA;;EAEA;;EAEA;;UAGe;;EAEf;;EAEA;;;;;;UAOe;EACf,aAAa;EACb;EACA,YAAY;EACZ,aAAa;;EAEb,SAAS;;UAGM;EACf;;EAEA;;EAEA;;EAEA;;EAEA,aAAa;;EAEb;;;;;;;EAOA,OAAO;;;;;;;;;;EAUP,gBAAgB,cAAc;;;;;;;EAO9B,mBAAmB,cAAc;;;;;EAKjC,eAAe,SAAS;;EAExB,OAAO;;EAEP,OAAO,aAAa,SAAS;;EAE7B,SAAS;;;;;;;;EAQT,QAAQ;;;UAMO;EACf;;EAEA;;EAEA;;EAEA;EACA,SAAS;;;;;;;KAQC;EAEN;EACA;EACA;EACA;EACA;;EAEA;;EAEA;EAAkB;IAAS;IAAe;;;;;;;;UAO/B;;WAEN;EACT,QAAQ,SAAS,wBAAwB,QAAQ;;;;;;;;UASlC,QAAQ;;WAEd;;WAEA;WACA,WAAW;WACX,MAAM;WACN,WAAW;;WAEX;EACT,QAAQ,OAAO,QAAQ,KAAK,iBAAiB,QAAQ;;;UAItC;;EAEf;;EAEA,QAAQ;;EAER,MAAM;;EAEN;;;;;;;;EAQA;IAAkB;IAAsB;;;;;;;;EAOxC;;;;;;;;;;iBAac,iBAAiB;EAC/B;EACA;EACA;EACA;;EAEA;;;cAsBW;;;;;iBAeG,YACd,MAAM,KAAK;EACT;EACA;IAED;;iBAmBa,oBACd,MAAM,KAAK;EACT;EACA;IAED;UAOc;EACf;EACA;;EAEA;EACA;EACA;;EAEA,OAAO;;EAEP;IAAU;IAAe;;;UAGV;EACf;EACA;EACA;EACA;EACA,UAAU;EACV,aAAa;;EAEb;;;;;EAKA,wBAAwB;;;;;;;;;;;;;;;KAkBd;EAEN;EACA;EACA;EACA;;EAEA,aAAa;;EAGb;EACA,SAAS;;EAGT;EACA;EACA;;EAGA;;EAEA,SAAS;EACT,UAAU,cAAc;;EAGxB;EACA,QAAQ"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { c as RunSplitTag } from "./run-record-
|
|
2
|
-
import { R as Scenario, d as DispatchContext } from "./types-
|
|
1
|
+
import { c as RunSplitTag } from "./run-record-C1E24G2A.js";
|
|
2
|
+
import { R as Scenario, d as DispatchContext } from "./types-Cq8mSbq8.js";
|
|
3
3
|
//#region src/benchmarks/types.d.ts
|
|
4
4
|
type BenchmarkTaskKind = 'retrieval' | 'rag-answer' | 'hallucination' | 'kb-improvement' | 'routing' | 'custom';
|
|
5
5
|
type BenchmarkFamily = 'beir' | 'mteb-retrieval' | 'msmarco' | 'trec-dl' | 'miracl' | 'lotte' | 'bright' | 'crag' | 'hotpotqa' | 'kilt' | 'ragtruth' | 'faithbench' | 'first-party' | 'custom';
|
|
@@ -99,4 +99,4 @@ declare const BENCHMARK_SPLIT_SEED = "agent-eval-v1";
|
|
|
99
99
|
declare function deterministicSplit(itemId: string, seed?: string): RunSplitTag;
|
|
100
100
|
//#endregion
|
|
101
101
|
export { BenchmarkFamily as a, BenchmarkSource as c, BenchmarkEvaluation as i, BenchmarkTaskKind as l, BenchmarkAdapter as n, BenchmarkResponder as o, BenchmarkDatasetItem as r, BenchmarkScenario as s, BENCHMARK_SPLIT_SEED as t, deterministicSplit as u };
|
|
102
|
-
//# sourceMappingURL=types-
|
|
102
|
+
//# sourceMappingURL=types-Cbe2fWBa.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types-
|
|
1
|
+
{"version":3,"file":"types-Cbe2fWBa.d.ts","names":[],"sources":["../src/benchmarks/types.ts"],"mappings":";;;KASY;KAQA;UAgBK,qBAAqB;;;EAGpC;;EAEA,SAAS;;EAET,QAAQ;;EAER,SAAS;;EAET,WAAW;;EAEX;;EAEA,SAAS;EACT,WAAW;;UAGI;;;;EAIf;;EAEA;;EAEA,aAAa;;;EAGb,KAAK;EACL;;UAGe;EACf;EACA;EACA;;EAEA;EACA;EACA;;EAEA;;;;;EAKA;;;UAQe,iBAAiB,kBAAkB,oBAAoB;;EAEtE;EACA,SAAS;EACT,WAAW;EACX;EACA,SAAS;EACT;;;;;EAKA,YAAY,OAAO,cAAc,QAAQ,qBAAqB;;EAE9D,SAAS,MAAM,qBAAqB,WAAW,UAAU,YAAY,QAAQ;;;;EAI7E,YAAY,iBAAiB;;UAGd,kBAAkB,4BAA4B;EAC7D;EACA;EACA,QAAQ;EACR,UAAU;EACV,UAAU;EACV,MAAM,qBAAqB;;KAGjB,mBAAmB,oBAAoB,uBAAuB;EACxE,UAAU,kBAAkB;EAC5B,MAAM,qBAAqB;EAC3B,SAAS;MACL,QAAQ,aAAa;;;cAoBd;;;;;;;;;iBAUG,mBACd,gBACA,gBACC"}
|