@tangle-network/agent-eval 0.216.1 → 0.217.1
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 +102 -57
- package/README.md +2 -2
- package/dist/adapters/http.d.ts +2 -2
- package/dist/{agent-profile-VFcQncbW.d.ts → agent-profile-DTt8uGP9.d.ts} +4 -17
- package/dist/agent-profile-DTt8uGP9.d.ts.map +1 -0
- package/dist/{agent-profile-cell-BK8mlqlq.js → agent-profile-cell-B1V1gUZs.js} +4 -53
- package/dist/agent-profile-cell-B1V1gUZs.js.map +1 -0
- package/dist/{agent-profile-cell-BNp-Js9F.d.ts → agent-profile-cell-BcpFbNd5.d.ts} +2 -22
- package/dist/agent-profile-cell-BcpFbNd5.d.ts.map +1 -0
- package/dist/{agreement-irr-BEWLIF3_.js → agreement-irr-BhAqjluA.js} +5 -5
- package/dist/agreement-irr-BhAqjluA.js.map +1 -0
- package/dist/analyst/index.d.ts +36 -169
- package/dist/analyst/index.d.ts.map +1 -1
- package/dist/analyst/index.js +17 -192
- package/dist/analyst/index.js.map +1 -1
- package/dist/{analyze-runs-C84irm4i.js → analyze-runs-DqPJaJ4X.js} +225 -21
- package/dist/analyze-runs-DqPJaJ4X.js.map +1 -0
- package/dist/{analyze-runs-D68fd-su.d.ts → analyze-runs-DyeNcenW.d.ts} +5 -5
- package/dist/{analyze-runs-D68fd-su.d.ts.map → analyze-runs-DyeNcenW.d.ts.map} +1 -1
- package/dist/authenticity/index.d.ts +0 -23
- package/dist/authenticity/index.d.ts.map +1 -1
- package/dist/authenticity/index.js +1 -34
- package/dist/authenticity/index.js.map +1 -1
- package/dist/{benchmark-command-By0vh8q7.js → benchmark-command-xQK8nOI6.js} +554 -89
- package/dist/benchmark-command-xQK8nOI6.js.map +1 -0
- package/dist/benchmarks/index.d.ts +5 -6
- package/dist/benchmarks/index.d.ts.map +1 -1
- package/dist/benchmarks/index.js +4 -4
- package/dist/benchmarks/index.js.map +1 -1
- package/dist/builder-eval/index.d.ts +3 -3
- package/dist/builder-eval/index.js +2 -2
- package/dist/builder-eval/index.js.map +1 -1
- package/dist/campaign/index.d.ts +7 -7
- package/dist/campaign/index.js +15 -15
- package/dist/{campaign-BudZ70zo.js → campaign-CJGkxAQ6.js} +23 -530
- package/dist/campaign-CJGkxAQ6.js.map +1 -0
- package/dist/{campaign-evidence-Clrn4hBV.js → campaign-evidence-CQYw8_0H.js} +10 -18
- package/dist/{campaign-evidence-Clrn4hBV.js.map → campaign-evidence-CQYw8_0H.js.map} +1 -1
- package/dist/{canonical-B2RqLofv.js → canonical-Dq2hOQ8L.js} +2 -2
- package/dist/{canonical-B2RqLofv.js.map → canonical-Dq2hOQ8L.js.map} +1 -1
- package/dist/cell-spend-DEWR3nj-.js +90 -0
- package/dist/cell-spend-DEWR3nj-.js.map +1 -0
- package/dist/{chat-client-l-MVDlDk.js → chat-client-ldqxYYTN.js} +2 -2
- package/dist/{chat-client-l-MVDlDk.js.map → chat-client-ldqxYYTN.js.map} +1 -1
- package/dist/{chat-json-call-puMArmtW.js → chat-json-call-DavzexSp.js} +3 -2
- package/dist/chat-json-call-DavzexSp.js.map +1 -0
- package/dist/{claim-q4HQhK1G.js → claim-BPVuyHuy.js} +3 -3
- package/dist/{claim-q4HQhK1G.js.map → claim-BPVuyHuy.js.map} +1 -1
- package/dist/cli.js +113 -12
- package/dist/cli.js.map +1 -1
- package/dist/{client-BzL9niqi.d.ts → client-hwo3cJjR.d.ts} +3 -16
- package/dist/client-hwo3cJjR.d.ts.map +1 -0
- package/dist/{cluster-bootstrap-qcUWYFl-.js → cluster-bootstrap-DEwwhl0C.js} +3 -3
- package/dist/{cluster-bootstrap-qcUWYFl-.js.map → cluster-bootstrap-DEwwhl0C.js.map} +1 -1
- package/dist/{produced-state-BsDmnwxc.js → completion-verifier-B_jw6C9p.js} +9 -64
- package/dist/completion-verifier-B_jw6C9p.js.map +1 -0
- package/dist/contract/index.d.ts +12 -12
- package/dist/contract/index.js +25 -401
- package/dist/contract/index.js.map +1 -1
- package/dist/{cost-ledger-BP8V4kUQ.js → cost-ledger-D0352Z4d.js} +4 -8
- package/dist/cost-ledger-D0352Z4d.js.map +1 -0
- package/dist/dataset-DQqhOCPt.d.ts.map +1 -1
- package/dist/{default-registry-COhk9g36.js → default-registry-ClYjq3QE.js} +8 -11
- package/dist/default-registry-ClYjq3QE.js.map +1 -0
- package/dist/{default-registry-DzHiGruW.d.ts → default-registry-DXgLmoDI.d.ts} +7 -50
- package/dist/default-registry-DXgLmoDI.d.ts.map +1 -0
- package/dist/{define-agent-eval-CNTPzIkp.js → define-agent-eval-BDxNlP3D.js} +10 -10
- package/dist/{define-agent-eval-CNTPzIkp.js.map → define-agent-eval-BDxNlP3D.js.map} +1 -1
- package/dist/{define-agent-eval-WSck0CZJ.d.ts → define-agent-eval-BIkH_gAR.d.ts} +8 -8
- package/dist/define-agent-eval-BIkH_gAR.d.ts.map +1 -0
- package/dist/{descriptive-DboxPZrG.js → descriptive-CVNy48rR.js} +2 -2
- package/dist/{descriptive-DboxPZrG.js.map → descriptive-CVNy48rR.js.map} +1 -1
- package/dist/diagnosis/index.d.ts +4 -14
- package/dist/diagnosis/index.d.ts.map +1 -1
- package/dist/diagnosis/index.js +67 -28
- package/dist/diagnosis/index.js.map +1 -1
- package/dist/{dspy-rlm-engine-LRCu61m5.js → dspy-rlm-engine-DhdCr1Dc.js} +5 -10
- package/dist/dspy-rlm-engine-DhdCr1Dc.js.map +1 -0
- package/dist/{emitter-HcCmFQl_.d.ts → emitter-DABYYTcH.d.ts} +5 -5
- package/dist/emitter-DABYYTcH.d.ts.map +1 -0
- package/dist/emitter-DbMXUIVP.js.map +1 -1
- package/dist/{engine-3A_4jOlI.d.ts → engine-BvZ0R-36.d.ts} +17 -21
- package/dist/engine-BvZ0R-36.d.ts.map +1 -0
- package/dist/{equal-terms-BtdXayCJ.js → equal-terms-DbPGcVtI.js} +2 -2
- package/dist/equal-terms-DbPGcVtI.js.map +1 -0
- package/dist/{errors-DVZiqS7K.js → errors-BcOs5MKW.js} +2 -8
- package/dist/{errors-DVZiqS7K.js.map → errors-BcOs5MKW.js.map} +1 -1
- package/dist/{eval-campaign-90jeaCAI.js → eval-campaign-BLEfBGbA.js} +6 -6
- package/dist/{eval-campaign-90jeaCAI.js.map → eval-campaign-BLEfBGbA.js.map} +1 -1
- package/dist/{evaluation-6Yoi74gM.d.ts → evaluation-jTjC_QQD.d.ts} +7 -7
- package/dist/evaluation-jTjC_QQD.d.ts.map +1 -0
- package/dist/evaluation.d.ts +2 -2
- package/dist/evaluation.js +1 -1
- package/dist/evaluation.js.map +1 -1
- package/dist/{exact-types-ClugMJfI.d.ts → exact-types-CflbJ9f3.d.ts} +2 -2
- package/dist/{exact-types-ClugMJfI.d.ts.map → exact-types-CflbJ9f3.d.ts.map} +1 -1
- package/dist/{exec-ChJ6d5if.js → exec-CpbtOXVt.js} +2 -24
- package/dist/{exec-ChJ6d5if.js.map → exec-CpbtOXVt.js.map} +1 -1
- package/dist/execution-tracks-CpgFPpS5.js.map +1 -1
- package/dist/experiment/index.d.ts +9 -112
- package/dist/experiment/index.d.ts.map +1 -1
- package/dist/experiment/index.js +149 -241
- package/dist/experiment/index.js.map +1 -1
- package/dist/{exporters-X4xwla8A.js → exporters-Cl6IeNAu.js} +3 -3
- package/dist/exporters-Cl6IeNAu.js.map +1 -0
- package/dist/{exporters-ChTnR90l.d.ts → exporters-Dk3oEbK_.d.ts} +2 -4
- package/dist/{exporters-ChTnR90l.d.ts.map → exporters-Dk3oEbK_.d.ts.map} +1 -1
- package/dist/{external-optimizer-process-BjMGUuds.js → external-optimizer-process-KVFGvYX1.js} +4 -3
- package/dist/external-optimizer-process-KVFGvYX1.js.map +1 -0
- package/dist/{external-optimizer-subprocess-BqW0hHcI.js → external-optimizer-subprocess-XIHBOSKt.js} +9 -11
- package/dist/external-optimizer-subprocess-XIHBOSKt.js.map +1 -0
- package/dist/{extract-usage-BrnzRJ6H.js → extract-usage-BZLU01kR.js} +2 -23
- package/dist/{extract-usage-BrnzRJ6H.js.map → extract-usage-BZLU01kR.js.map} +1 -1
- package/dist/{failure-cluster-CbgEDcFJ.d.ts → failure-cluster-DN0t63Z0.d.ts} +4 -43
- package/dist/failure-cluster-DN0t63Z0.d.ts.map +1 -0
- package/dist/{failure-taxonomy-CjabTATj.js → failure-taxonomy-Bx5oWqaW.js} +2 -19
- package/dist/{failure-taxonomy-CjabTATj.js.map → failure-taxonomy-Bx5oWqaW.js.map} +1 -1
- package/dist/{feedback-trajectory-C4EG0nNH.d.ts → feedback-trajectory-BWmhRBGu.d.ts} +7 -9
- package/dist/feedback-trajectory-BWmhRBGu.d.ts.map +1 -0
- package/dist/{semantic-concept-judge-BQF8te0T.js → findings-store-BMVQtEsm.js} +9 -249
- package/dist/findings-store-BMVQtEsm.js.map +1 -0
- package/dist/fuzz.d.ts +19 -62
- package/dist/fuzz.d.ts.map +1 -1
- package/dist/fuzz.js +62 -108
- package/dist/fuzz.js.map +1 -1
- package/dist/{hf-dataset-BR2yhuNn.js → hf-dataset-D9f7-oX7.js} +4 -24
- package/dist/hf-dataset-D9f7-oX7.js.map +1 -0
- package/dist/hosted/index.d.ts +2 -2
- package/dist/hosted/index.js +2 -2
- package/dist/{index-DHKEAv3s.d.ts → index-9jJ2fKSX.d.ts} +2 -64
- package/dist/index-9jJ2fKSX.d.ts.map +1 -0
- package/dist/{index-DV-jSzxF.d.ts → index-BLGqZ81N.d.ts} +10 -15
- package/dist/{index-DV-jSzxF.d.ts.map → index-BLGqZ81N.d.ts.map} +1 -1
- package/dist/{index-DPcngtwt.d.ts → index-BiT6RtNH.d.ts} +11 -36
- package/dist/index-BiT6RtNH.d.ts.map +1 -0
- package/dist/{index-BO41eGj5.d.ts → index-CX-GAUpw.d.ts} +3 -7
- package/dist/index-CX-GAUpw.d.ts.map +1 -0
- package/dist/{index-BFRO6chs.d.ts → index-CiRwLv2R.d.ts} +5 -69
- package/dist/index-CiRwLv2R.d.ts.map +1 -0
- package/dist/{index-CyszSg4O.d.ts → index-ZuWx52WG.d.ts} +115 -347
- package/dist/index-ZuWx52WG.d.ts.map +1 -0
- package/dist/index.d.ts +211 -218
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +402 -1040
- package/dist/index.js.map +1 -1
- package/dist/{integrity-dLiDe2kD.js → integrity-DOd9Q86O.js} +7 -9
- package/dist/integrity-DOd9Q86O.js.map +1 -0
- package/dist/{integrity-BXOsAb3i.js → integrity-W4CADcFc.js} +3 -7
- package/dist/{integrity-BXOsAb3i.js.map → integrity-W4CADcFc.js.map} +1 -1
- package/dist/{integrity-DAMEDFaC.d.ts → integrity-_qgZLA5p.d.ts} +4 -6
- package/dist/{integrity-DAMEDFaC.d.ts.map → integrity-_qgZLA5p.d.ts.map} +1 -1
- package/dist/{internal-Epl-YBZW.js → internal-BTK91Bqt.js} +3 -3
- package/dist/{internal-Epl-YBZW.js.map → internal-BTK91Bqt.js.map} +1 -1
- package/dist/{interval-CELpye1n.js → interval-Dq9gt7jx.js} +3 -3
- package/dist/{interval-CELpye1n.js.map → interval-Dq9gt7jx.js.map} +1 -1
- package/dist/{journal-DfRz9kps.js → journal-CR4QikYp.js} +110 -110
- package/dist/journal-CR4QikYp.js.map +1 -0
- package/dist/json-BhdsJaOQ.js +111 -0
- package/dist/json-BhdsJaOQ.js.map +1 -0
- package/dist/{judge-calibration-BpCy5XQl.js → judge-calibration-BkXMIWA4.js} +3 -3
- package/dist/{judge-calibration-BpCy5XQl.js.map → judge-calibration-BkXMIWA4.js.map} +1 -1
- package/dist/{judge-gate-iY9s4L44.js → judge-gate-Co5tuh57.js} +4 -27
- package/dist/{judge-gate-iY9s4L44.js.map → judge-gate-Co5tuh57.js.map} +1 -1
- package/dist/{judge-gate-CE7CQAgt.d.ts → judge-gate-DsL0yFcb.d.ts} +8 -9
- package/dist/judge-gate-DsL0yFcb.d.ts.map +1 -0
- package/dist/{kind-factory-BobI5Jy3.js → kind-factory-Gy4fXixd.js} +88 -17
- package/dist/kind-factory-Gy4fXixd.js.map +1 -0
- package/dist/ledger-core/index.d.ts +2 -2
- package/dist/ledger-core/index.js +4 -4
- package/dist/{ledger-core-Gq7yp8lP.js → ledger-core-DrJhbms9.js} +5 -84
- package/dist/ledger-core-DrJhbms9.js.map +1 -0
- package/dist/{llm-client-CyBAoYiX.js → llm-client-BshCREn6.js} +6 -108
- package/dist/llm-client-BshCREn6.js.map +1 -0
- package/dist/{llm-judge-CG12SQJq.js → llm-judge-C8jMTNNG.js} +24 -32
- package/dist/llm-judge-C8jMTNNG.js.map +1 -0
- package/dist/matrix/index.d.ts +3 -3
- package/dist/matrix/index.js +3 -2
- package/dist/{matrix-Dm0BQW7U.js → matrix-B8l-Nsxd.js} +3 -88
- package/dist/matrix-B8l-Nsxd.js.map +1 -0
- package/dist/{matrix-DSrSxjjy.d.ts → matrix-DHe-dwVJ.d.ts} +9 -18
- package/dist/matrix-DHe-dwVJ.d.ts.map +1 -0
- package/dist/{mean-DHoMiudD.js → mean-BbA_JCzW.js} +39 -5
- package/dist/mean-BbA_JCzW.js.map +1 -0
- package/dist/meta-eval/index.d.ts +5 -5
- package/dist/meta-eval/index.js +10 -10
- package/dist/metrics-BKaPBtIG.js.map +1 -1
- package/dist/{mint-DNU8QPG4.d.ts → mint-BRZXuwRi.d.ts} +4 -4
- package/dist/{mint-DNU8QPG4.d.ts.map → mint-BRZXuwRi.d.ts.map} +1 -1
- package/dist/{mint-QaD5cA0u.js → mint-zszyYlpA.js} +4 -4
- package/dist/{mint-QaD5cA0u.js.map → mint-zszyYlpA.js.map} +1 -1
- package/dist/{multi-layer-verifier-BUaQ4C17.d.ts → multi-layer-verifier-DEgMUk6v.d.ts} +2 -2
- package/dist/{multi-layer-verifier-BUaQ4C17.d.ts.map → multi-layer-verifier-DEgMUk6v.d.ts.map} +1 -1
- package/dist/{multiplicity-DQUFcbW4.js → multiplicity-Sg0U7h8j.js} +2 -2
- package/dist/{multiplicity-DQUFcbW4.js.map → multiplicity-Sg0U7h8j.js.map} +1 -1
- package/dist/multishot/golden/index.d.ts +1 -23
- package/dist/multishot/golden/index.d.ts.map +1 -1
- package/dist/multishot/golden/index.js +3 -3
- package/dist/multishot/golden/index.js.map +1 -1
- package/dist/multishot/index.d.ts +3 -12
- package/dist/multishot/index.d.ts.map +1 -1
- package/dist/multishot/index.js +4 -3
- package/dist/multishot/index.js.map +1 -1
- package/dist/openapi.json +114 -10
- package/dist/{otel-spans-B9tUXRHF.js → otel-spans-CdN909k1.js} +3 -3
- package/dist/{otel-spans-B9tUXRHF.js.map → otel-spans-CdN909k1.js.map} +1 -1
- package/dist/{paired-arms-CwMGJsrS.js → paired-arms-Ct4LTKc1.js} +32 -9
- package/dist/paired-arms-Ct4LTKc1.js.map +1 -0
- package/dist/{paired-binary-CvpFHfKZ.js → paired-binary-B_nejA44.js} +3 -36
- package/dist/paired-binary-B_nejA44.js.map +1 -0
- package/dist/{paired-promotion-decision-DPKz8ZnF.d.ts → paired-promotion-decision-CQMJ5xGN.d.ts} +2 -34
- package/dist/paired-promotion-decision-CQMJ5xGN.d.ts.map +1 -0
- package/dist/{paired-promotion-decision-DsAPYGCm.js → paired-promotion-decision-D0SHb-ZL.js} +3 -3
- package/dist/{paired-promotion-decision-DsAPYGCm.js.map → paired-promotion-decision-D0SHb-ZL.js.map} +1 -1
- package/dist/{pareto-Cm6ndrWN.js → pareto-BzsxidFD.js} +11 -11
- package/dist/pareto-BzsxidFD.js.map +1 -0
- package/dist/pipelines/index.d.ts +16 -19
- package/dist/pipelines/index.d.ts.map +1 -1
- package/dist/pipelines/index.js +4 -68
- package/dist/pipelines/index.js.map +1 -1
- package/dist/{power-and-mde-Hj2ubb4B.js → power-and-mde-N5nU4_Ly.js} +3 -3
- package/dist/{power-and-mde-Hj2ubb4B.js.map → power-and-mde-N5nU4_Ly.js.map} +1 -1
- package/dist/{prime-protocol-CHKgA3a6.js → prime-protocol-BFgGrEMP.js} +3 -3
- package/dist/prime-protocol-BFgGrEMP.js.map +1 -0
- package/dist/{prime-protocol-HiB9HX7E.d.ts → prime-protocol-BWG2DQIk.d.ts} +3 -21
- package/dist/prime-protocol-BWG2DQIk.d.ts.map +1 -0
- package/dist/profile-cell.d.ts +2 -2
- package/dist/profile-cell.js +2 -2
- package/dist/{promotion-gate-BawrdPYr.d.ts → promotion-gate-DaY95yU6.d.ts} +7 -8
- package/dist/{promotion-gate-BawrdPYr.d.ts.map → promotion-gate-DaY95yU6.d.ts.map} +1 -1
- package/dist/{promotion-policy-KO57gbyV.js → promotion-policy-6A8osGr_.js} +4 -4
- package/dist/{promotion-policy-KO57gbyV.js.map → promotion-policy-6A8osGr_.js.map} +1 -1
- package/dist/{proposal-findings-CY1xKn_a.js → proposal-findings-BbWidvUk.js} +3 -3
- package/dist/proposal-findings-BbWidvUk.js.map +1 -0
- package/dist/{query-BUCZCK7F.js → query-B_RZ36an.js} +5 -5
- package/dist/query-B_RZ36an.js.map +1 -0
- package/dist/{query-DK1nWcmp.d.ts → query-Cmv_U8vf.d.ts} +4 -24
- package/dist/{query-DK1nWcmp.d.ts.map → query-Cmv_U8vf.d.ts.map} +1 -1
- package/dist/{random-ByUSnV9w.js → random-DDDcC7YU.js} +2 -2
- package/dist/{random-ByUSnV9w.js.map → random-DDDcC7YU.js.map} +1 -1
- package/dist/{raw-provider-sink-DoHz4JNG.js → raw-provider-sink-75w4FWBj.js} +2 -2
- package/dist/{raw-provider-sink-DoHz4JNG.js.map → raw-provider-sink-75w4FWBj.js.map} +1 -1
- package/dist/{redact-mtCCcT8z.d.ts → redact-DXdlFb_U.d.ts} +2 -3
- package/dist/redact-DXdlFb_U.d.ts.map +1 -0
- package/dist/{redact-B5xXKqQ3.js → redact-mtkTiJVk.js} +2 -7
- package/dist/{redact-B5xXKqQ3.js.map → redact-mtkTiJVk.js.map} +1 -1
- package/dist/{registry-D7_hm2fK.d.ts → registry-ssNoHzwD.d.ts} +4 -4
- package/dist/{registry-D7_hm2fK.d.ts.map → registry-ssNoHzwD.d.ts.map} +1 -1
- package/dist/{release-confidence-CpuP65S5.js → release-confidence-CYIFH-fp.js} +6 -11
- package/dist/release-confidence-CYIFH-fp.js.map +1 -0
- package/dist/{report-command-SC9nqb4M.js → render-C-w2EZ6p.js} +205 -363
- package/dist/render-C-w2EZ6p.js.map +1 -0
- package/dist/reporting.d.ts +6 -6
- package/dist/reporting.js +8 -8
- package/dist/{researcher-CEcfku-T.d.ts → researcher-0ABVIdGa.d.ts} +9 -9
- package/dist/{researcher-CEcfku-T.d.ts.map → researcher-0ABVIdGa.d.ts.map} +1 -1
- package/dist/{reward-CpCmAM8L.d.ts → reward-DrZ71vnf.d.ts} +3 -26
- package/dist/reward-DrZ71vnf.d.ts.map +1 -0
- package/dist/{reward-hacking-E0d-WVcP.js → reward-hacking-ByuETaIb.js} +3 -3
- package/dist/{reward-hacking-E0d-WVcP.js.map → reward-hacking-ByuETaIb.js.map} +1 -1
- package/dist/{reward-hacking-RVdZ6Kkt.d.ts → reward-hacking-E52wq86n.d.ts} +5 -18
- package/dist/reward-hacking-E52wq86n.d.ts.map +1 -0
- package/dist/reward-nw2xZGZG.js.map +1 -1
- package/dist/rl.d.ts +117 -295
- package/dist/rl.d.ts.map +1 -1
- package/dist/rl.js +27 -747
- package/dist/rl.js.map +1 -1
- package/dist/rollout/index.d.ts +5 -5
- package/dist/rollout/index.js +7 -8
- package/dist/{rollout-DZV1t2Wg.js → rollout-D6C50mfA.js} +88 -13
- package/dist/rollout-D6C50mfA.js.map +1 -0
- package/dist/{rubric-predictive-validity-DaFW70y9.js → rubric-predictive-validity-D3_oI72N.js} +3 -3
- package/dist/{rubric-predictive-validity-DaFW70y9.js.map → rubric-predictive-validity-D3_oI72N.js.map} +1 -1
- package/dist/{rubric-predictive-validity-C03EC7ZE.d.ts → rubric-predictive-validity-DiwT8PbL.d.ts} +2 -2
- package/dist/{rubric-predictive-validity-C03EC7ZE.d.ts.map → rubric-predictive-validity-DiwT8PbL.d.ts.map} +1 -1
- package/dist/{run-record-C1E24G2A.d.ts → run-record-CaxbLq6Y.d.ts} +4 -10
- package/dist/run-record-CaxbLq6Y.d.ts.map +1 -0
- package/dist/{run-record-C5e7FjhJ.js → run-record-CysKLB93.js} +3 -3
- package/dist/run-record-CysKLB93.js.map +1 -0
- package/dist/{run-record-Boc5IUTp.js → run-record-DhEJgoPz.js} +4 -4
- package/dist/{run-record-Boc5IUTp.js.map → run-record-DhEJgoPz.js.map} +1 -1
- package/dist/sandbox-harness-BlSOu4LX.d.ts.map +1 -1
- package/dist/{schema-CSf6qWgZ.js → schema-Bt2vLFm3.js} +2 -14
- package/dist/schema-Bt2vLFm3.js.map +1 -0
- package/dist/schema-C-ziyJD7.d.ts.map +1 -1
- package/dist/{schema-2g7trUwH.d.ts → schema-D6EVKzTB.d.ts} +4 -5
- package/dist/schema-D6EVKzTB.d.ts.map +1 -0
- package/dist/{schema-BuiujZ-g.js → schema-DxcKTzbi.js} +2 -2
- package/dist/schema-DxcKTzbi.js.map +1 -0
- package/dist/search/index.d.ts +3 -31
- package/dist/search/index.d.ts.map +1 -1
- package/dist/search/index.js +4 -4
- package/dist/{search-kernel-DyNEa6IU.d.ts → search-kernel-DxqjBxA-.d.ts} +10 -27
- package/dist/search-kernel-DxqjBxA-.d.ts.map +1 -0
- package/dist/{search-ledger-CVUGLDZp.js → search-ledger-TUvB5gWH.js} +7 -7
- package/dist/{search-ledger-CVUGLDZp.js.map → search-ledger-TUvB5gWH.js.map} +1 -1
- package/dist/{search-ledger-file-DDQUgvCg.js → search-ledger-file-UKCO2r8a.js} +4 -4
- package/dist/search-ledger-file-UKCO2r8a.js.map +1 -0
- package/dist/{search-shipper-BOSD46qv.js → search-shipper-Cw2z3-td.js} +4 -4
- package/dist/search-shipper-Cw2z3-td.js.map +1 -0
- package/dist/{search-summary-BQCSWf1g.js → search-summary-DQIvJsSB.js} +5 -5
- package/dist/{search-summary-BQCSWf1g.js.map → search-summary-DQIvJsSB.js.map} +1 -1
- package/dist/{sequential-CzK5DarL.js → sequential-BVv1sjsw.js} +2 -32
- package/dist/{sequential-CzK5DarL.js.map → sequential-BVv1sjsw.js.map} +1 -1
- package/dist/{sequential-Bf_pbE50.js → sequential-CimQrbjv.js} +5 -5
- package/dist/sequential-CimQrbjv.js.map +1 -0
- package/dist/{sequential-BhsrMupG.d.ts → sequential-KwGspfeR.d.ts} +2 -9
- package/dist/{sequential-BhsrMupG.d.ts.map → sequential-KwGspfeR.d.ts.map} +1 -1
- package/dist/{sequential-eprocess-DM_qENCl.js → sequential-eprocess-DodI6RoF.js} +2 -2
- package/dist/{sequential-eprocess-DM_qENCl.js.map → sequential-eprocess-DodI6RoF.js.map} +1 -1
- package/dist/{server-e9oWO_bN.js → server-DoVbqqtJ.js} +101 -101
- package/dist/server-DoVbqqtJ.js.map +1 -0
- package/dist/{single-run-lock-CV9Eyd_q.js → single-run-lock-U58yf8N6.js} +2 -2
- package/dist/{single-run-lock-CV9Eyd_q.js.map → single-run-lock-U58yf8N6.js.map} +1 -1
- package/dist/{skillopt-optimization-method-BiwA8nfX.js → skillopt-optimization-method-kyL94-ri.js} +17 -95
- package/dist/skillopt-optimization-method-kyL94-ri.js.map +1 -0
- package/dist/{spans-BHAUXMGN.d.ts → spans-CVngGij3.d.ts} +2 -2
- package/dist/{spans-BHAUXMGN.d.ts.map → spans-CVngGij3.d.ts.map} +1 -1
- package/dist/{statistical-heldout-42NlCNDe.js → statistical-heldout-Bksv92Th.js} +3 -3
- package/dist/statistical-heldout-Bksv92Th.js.map +1 -0
- package/dist/{statistical-heldout-CYpGaoDH.d.ts → statistical-heldout-C9AbgREN.d.ts} +8 -67
- package/dist/statistical-heldout-C9AbgREN.d.ts.map +1 -0
- package/dist/{steps-Bg5-KVx1.d.ts → steps-DlepJPwD.d.ts} +2 -56
- package/dist/steps-DlepJPwD.d.ts.map +1 -0
- package/dist/{store-C4avFjIi.d.ts → store-B3nWRy3B.d.ts} +2 -2
- package/dist/{store-C4avFjIi.d.ts.map → store-B3nWRy3B.d.ts.map} +1 -1
- package/dist/{store-otlp-DJ9dtA5x.js → store-otlp-BteDAbHK.js} +3 -3
- package/dist/{store-otlp-DJ9dtA5x.js.map → store-otlp-BteDAbHK.js.map} +1 -1
- package/dist/{store-tool-spans-CIX5arAG.js → store-tool-spans-D5l_OfTJ.js} +8 -8
- package/dist/store-tool-spans-D5l_OfTJ.js.map +1 -0
- package/dist/{store-tool-spans-DdCSPa7C.d.ts → store-tool-spans-DZS0HWVf.d.ts} +9 -18
- package/dist/store-tool-spans-DZS0HWVf.d.ts.map +1 -0
- package/dist/storyboard/index.d.ts +4 -11
- package/dist/storyboard/index.d.ts.map +1 -1
- package/dist/storyboard/index.js +1 -16
- package/dist/storyboard/index.js.map +1 -1
- package/dist/{summary-report-V_JACR4Y.js → summary-report-BsXJoYX5.js} +88 -10
- package/dist/summary-report-BsXJoYX5.js.map +1 -0
- package/dist/{summary-report-bdFDXq0b.d.ts → summary-report-DM7tlml1.d.ts} +10 -19
- package/dist/summary-report-DM7tlml1.d.ts.map +1 -0
- package/dist/supervisor-run/index.d.ts +60 -28
- package/dist/supervisor-run/index.d.ts.map +1 -1
- package/dist/supervisor-run/index.js +4 -458
- package/dist/supervisor-run/index.js.map +1 -1
- package/dist/{systemone-BomhHr2c.d.ts → systemone-Bp_Ac0p1.d.ts} +5 -5
- package/dist/{systemone-BomhHr2c.d.ts.map → systemone-Bp_Ac0p1.d.ts.map} +1 -1
- package/dist/systemone-protocol.js +1 -1
- package/dist/{systemone-review-CUVcQJu6.d.ts → systemone-review-CwE6JGIp.d.ts} +2 -2
- package/dist/{systemone-review-CUVcQJu6.d.ts.map → systemone-review-CwE6JGIp.d.ts.map} +1 -1
- package/dist/{systemone-review-D7sXwsqq.js → systemone-review-DzdOJ8I6.js} +3 -3
- package/dist/{systemone-review-D7sXwsqq.js.map → systemone-review-DzdOJ8I6.js.map} +1 -1
- package/dist/systemone.d.ts +5 -5
- package/dist/systemone.js +5 -5
- package/dist/systemone.js.map +1 -1
- package/dist/{task-failure-attributes-hEmASjix.js → task-failure-attributes-BMsblaDh.js} +3 -3
- package/dist/task-failure-attributes-BMsblaDh.js.map +1 -0
- package/dist/{terminal-record-DZyQZ3pz.js → terminal-record-78g6E4vh.js} +5 -5
- package/dist/terminal-record-78g6E4vh.js.map +1 -0
- package/dist/{tool-groups-C3NEcdCM.d.ts → tool-groups-DyNz2JZ_.d.ts} +3 -3
- package/dist/tool-groups-DyNz2JZ_.d.ts.map +1 -0
- package/dist/{tool-waste-BKgvd-TP.d.ts → tool-waste-CkjzWmeY.d.ts} +25 -28
- package/dist/tool-waste-CkjzWmeY.d.ts.map +1 -0
- package/dist/{tool-waste-BlesqTqW.js → tool-waste-DLsjS_p-.js} +5 -5
- package/dist/{tool-waste-BlesqTqW.js.map → tool-waste-DLsjS_p-.js.map} +1 -1
- package/dist/{trace-questions-CM2CAocU.d.ts → trace-questions-xxwgsuMi.d.ts} +4 -9
- package/dist/{trace-questions-CM2CAocU.d.ts.map → trace-questions-xxwgsuMi.d.ts.map} +1 -1
- package/dist/trace-repair/index.d.ts +14 -211
- package/dist/trace-repair/index.d.ts.map +1 -1
- package/dist/trace-repair/index.js +865 -981
- package/dist/trace-repair/index.js.map +1 -1
- package/dist/traces.d.ts +18 -46
- package/dist/traces.d.ts.map +1 -1
- package/dist/traces.js +13 -13
- package/dist/traces.js.map +1 -1
- package/dist/trajectory-replay/index.d.ts +3 -117
- package/dist/trajectory-replay/index.d.ts.map +1 -1
- package/dist/trajectory-replay/index.js +98 -56
- package/dist/trajectory-replay/index.js.map +1 -1
- package/dist/{tree-C1XYeXvy.js → tree-BGRglkKd.js} +13 -10
- package/dist/tree-BGRglkKd.js.map +1 -0
- package/dist/{types-B3jzCp0p.js → types-BXgASBCL.js} +2 -2
- package/dist/{types-B3jzCp0p.js.map → types-BXgASBCL.js.map} +1 -1
- package/dist/{types-C1tnkEgM.d.ts → types-CW8ewATZ.d.ts} +4 -113
- package/dist/types-CW8ewATZ.d.ts.map +1 -0
- package/dist/{types-BqkVmHfv.d.ts → types-CmXY0PQf.d.ts} +2 -4
- package/dist/types-CmXY0PQf.d.ts.map +1 -0
- package/dist/{types-C02qtnG8.d.ts → types-D-mgRzlh.d.ts} +7 -28
- package/dist/types-D-mgRzlh.d.ts.map +1 -0
- package/dist/{types-Cq8mSbq8.d.ts → types-D8GhJoHk.d.ts} +7 -9
- package/dist/{types-Cq8mSbq8.d.ts.map → types-D8GhJoHk.d.ts.map} +1 -1
- package/dist/{types-IXD2HJzu.js → types-DqFMNTgK.js} +2 -4
- package/dist/types-DqFMNTgK.js.map +1 -0
- package/dist/{types-Cbe2fWBa.d.ts → types-DtMbFw6U.d.ts} +3 -3
- package/dist/{types-Cbe2fWBa.d.ts.map → types-DtMbFw6U.d.ts.map} +1 -1
- package/dist/{verdict-Dl0tiT1X.js → verdict-BGQmxQ8A.js} +3 -3
- package/dist/verdict-BGQmxQ8A.js.map +1 -0
- package/dist/{verdict-E4eRNf7-.d.ts → verdict-C24NELtN.d.ts} +2 -15
- package/dist/{verdict-E4eRNf7-.d.ts.map → verdict-C24NELtN.d.ts.map} +1 -1
- package/dist/{verdict-cache-Bm8EKL2s.js → verdict-cache-CL_7tePD.js} +2 -2
- package/dist/verdict-cache-CL_7tePD.js.map +1 -0
- package/dist/wire/index.d.ts +31 -410
- package/dist/wire/index.d.ts.map +1 -1
- package/dist/wire/index.js +2 -2
- package/docs/changes/jev-evidence-review.md +1 -1
- package/docs/changes/jev-product-adoption.md +1 -1
- package/docs/changes/jev-recorded-evidence.md +1 -1
- package/docs/design/statistics-decisions.md +17 -17
- package/docs/design.md +1 -1
- package/docs/evaluation-integrity.md +1 -1
- package/docs/proofs/campaign-attempt-retention-20261001/README.md +2 -2
- package/docs/public-api.md +138 -529
- package/docs/rollout.md +1 -1
- package/docs/trace-repair-admission.md +1 -1
- package/docs/wire-protocol.md +1 -1
- package/package.json +1 -1
- package/dist/active-curriculum-B_Lc9kfO.js +0 -61
- package/dist/active-curriculum-B_Lc9kfO.js.map +0 -1
- package/dist/agent-profile-VFcQncbW.d.ts.map +0 -1
- package/dist/agent-profile-cell-BK8mlqlq.js.map +0 -1
- package/dist/agent-profile-cell-BNp-Js9F.d.ts.map +0 -1
- package/dist/agreement-irr-BEWLIF3_.js.map +0 -1
- package/dist/analyze-runs-C84irm4i.js.map +0 -1
- package/dist/baseline-Udi_PLuT.js +0 -259
- package/dist/baseline-Udi_PLuT.js.map +0 -1
- package/dist/benchmark-CXmVVHH2.js +0 -562
- package/dist/benchmark-CXmVVHH2.js.map +0 -1
- package/dist/benchmark-command-By0vh8q7.js.map +0 -1
- package/dist/campaign-BudZ70zo.js.map +0 -1
- package/dist/chat-json-call-puMArmtW.js.map +0 -1
- package/dist/claude-jsonl-dZir12m6.js +0 -86
- package/dist/claude-jsonl-dZir12m6.js.map +0 -1
- package/dist/client-BzL9niqi.d.ts.map +0 -1
- package/dist/cost-ledger-BP8V4kUQ.js.map +0 -1
- package/dist/counterfactual-CPaWN99-.d.ts +0 -60
- package/dist/counterfactual-CPaWN99-.d.ts.map +0 -1
- package/dist/counterfactual-Dn3veueY.js +0 -98
- package/dist/counterfactual-Dn3veueY.js.map +0 -1
- package/dist/default-registry-COhk9g36.js.map +0 -1
- package/dist/default-registry-DzHiGruW.d.ts.map +0 -1
- package/dist/define-agent-eval-WSck0CZJ.d.ts.map +0 -1
- package/dist/dspy-rlm-engine-LRCu61m5.js.map +0 -1
- package/dist/effect-sizes-BhzEpXMc.js +0 -82
- package/dist/effect-sizes-BhzEpXMc.js.map +0 -1
- package/dist/emitter-HcCmFQl_.d.ts.map +0 -1
- package/dist/engine-3A_4jOlI.d.ts.map +0 -1
- package/dist/equal-terms-BtdXayCJ.js.map +0 -1
- package/dist/evaluation-6Yoi74gM.d.ts.map +0 -1
- package/dist/exporters-X4xwla8A.js.map +0 -1
- package/dist/external-optimizer-process-BjMGUuds.js.map +0 -1
- package/dist/external-optimizer-subprocess-BqW0hHcI.js.map +0 -1
- package/dist/failure-cluster-CbgEDcFJ.d.ts.map +0 -1
- package/dist/feedback-trajectory-C4EG0nNH.d.ts.map +0 -1
- package/dist/hf-dataset-BR2yhuNn.js.map +0 -1
- package/dist/index-BFRO6chs.d.ts.map +0 -1
- package/dist/index-BO41eGj5.d.ts.map +0 -1
- package/dist/index-CyszSg4O.d.ts.map +0 -1
- package/dist/index-DHKEAv3s.d.ts.map +0 -1
- package/dist/index-DPcngtwt.d.ts.map +0 -1
- package/dist/integrity-dLiDe2kD.js.map +0 -1
- package/dist/journal-DfRz9kps.js.map +0 -1
- package/dist/judge-gate-CE7CQAgt.d.ts.map +0 -1
- package/dist/kind-factory-BobI5Jy3.js.map +0 -1
- package/dist/ledger-core-Gq7yp8lP.js.map +0 -1
- package/dist/llm-client-CyBAoYiX.js.map +0 -1
- package/dist/llm-judge-CG12SQJq.js.map +0 -1
- package/dist/matrix-DSrSxjjy.d.ts.map +0 -1
- package/dist/matrix-Dm0BQW7U.js.map +0 -1
- package/dist/mean-DHoMiudD.js.map +0 -1
- package/dist/normal-DsJZtrJc.js +0 -25
- package/dist/normal-DsJZtrJc.js.map +0 -1
- package/dist/paired-arms-CwMGJsrS.js.map +0 -1
- package/dist/paired-binary-CvpFHfKZ.js.map +0 -1
- package/dist/paired-promotion-decision-DPKz8ZnF.d.ts.map +0 -1
- package/dist/pareto-Cm6ndrWN.js.map +0 -1
- package/dist/prime-protocol-CHKgA3a6.js.map +0 -1
- package/dist/prime-protocol-HiB9HX7E.d.ts.map +0 -1
- package/dist/produced-state-BsDmnwxc.js.map +0 -1
- package/dist/proposal-findings-CY1xKn_a.js.map +0 -1
- package/dist/query-BUCZCK7F.js.map +0 -1
- package/dist/redact-mtCCcT8z.d.ts.map +0 -1
- package/dist/release-confidence-CpuP65S5.js.map +0 -1
- package/dist/report-command-SC9nqb4M.js.map +0 -1
- package/dist/reward-CpCmAM8L.d.ts.map +0 -1
- package/dist/reward-hacking-RVdZ6Kkt.d.ts.map +0 -1
- package/dist/rollout-DZV1t2Wg.js.map +0 -1
- package/dist/run-record-C1E24G2A.d.ts.map +0 -1
- package/dist/run-record-C5e7FjhJ.js.map +0 -1
- package/dist/schema-2g7trUwH.d.ts.map +0 -1
- package/dist/schema-BuiujZ-g.js.map +0 -1
- package/dist/schema-CSf6qWgZ.js.map +0 -1
- package/dist/search-kernel-DyNEa6IU.d.ts.map +0 -1
- package/dist/search-ledger-file-DDQUgvCg.js.map +0 -1
- package/dist/search-shipper-BOSD46qv.js.map +0 -1
- package/dist/semantic-concept-judge-BQF8te0T.js.map +0 -1
- package/dist/sequential-Bf_pbE50.js.map +0 -1
- package/dist/server-e9oWO_bN.js.map +0 -1
- package/dist/skillopt-optimization-method-BiwA8nfX.js.map +0 -1
- package/dist/statistical-heldout-42NlCNDe.js.map +0 -1
- package/dist/statistical-heldout-CYpGaoDH.d.ts.map +0 -1
- package/dist/steps-Bg5-KVx1.d.ts.map +0 -1
- package/dist/store-tool-spans-CIX5arAG.js.map +0 -1
- package/dist/store-tool-spans-DdCSPa7C.d.ts.map +0 -1
- package/dist/student-t-DVPW7CdL.js +0 -39
- package/dist/student-t-DVPW7CdL.js.map +0 -1
- package/dist/summary-report-V_JACR4Y.js.map +0 -1
- package/dist/summary-report-bdFDXq0b.d.ts.map +0 -1
- package/dist/task-failure-attributes-hEmASjix.js.map +0 -1
- package/dist/terminal-record-DZyQZ3pz.js.map +0 -1
- package/dist/tool-groups-C3NEcdCM.d.ts.map +0 -1
- package/dist/tool-waste-BKgvd-TP.d.ts.map +0 -1
- package/dist/trajectory-Y1fmdlw7.d.ts +0 -24
- package/dist/trajectory-Y1fmdlw7.d.ts.map +0 -1
- package/dist/tree-C1XYeXvy.js.map +0 -1
- package/dist/types-BqkVmHfv.d.ts.map +0 -1
- package/dist/types-C02qtnG8.d.ts.map +0 -1
- package/dist/types-C1tnkEgM.d.ts.map +0 -1
- package/dist/types-IXD2HJzu.js.map +0 -1
- package/dist/usage-receipt-Cx3xNeyB.js +0 -76
- package/dist/usage-receipt-Cx3xNeyB.js.map +0 -1
- package/dist/verdict-Dl0tiT1X.js.map +0 -1
- package/dist/verdict-cache-Bm8EKL2s.js.map +0 -1
- package/docs/design/self-improvement-evidence-audit.md +0 -263
|
@@ -279,19 +279,6 @@ declare class EquivalenceProtocolError extends Error {
|
|
|
279
279
|
* escape hatch: `arms: 3` or `blind: false` throws, never downgrades.
|
|
280
280
|
*/
|
|
281
281
|
declare function defineEquivalenceCheck(spec: EquivalenceCheckSpec): EquivalenceCheckDefinition;
|
|
282
|
-
/**
|
|
283
|
-
* Assemble the equivalence record, refusing every asymmetry.
|
|
284
|
-
*
|
|
285
|
-
* Refusals (each throws `EquivalenceProtocolError`):
|
|
286
|
-
* - an arm whose `blindness.toOtherArms` is false — it saw the other
|
|
287
|
-
* statement, so nothing was independently derived;
|
|
288
|
-
* - an arm whose `blindness.toOutcome` is false — it could steer its
|
|
289
|
-
* statement toward (or away from) the known result;
|
|
290
|
-
* - duplicate arm ids, empty statements, empty derivations;
|
|
291
|
-
* - an obligation whose fields contradict its status (see
|
|
292
|
-
* `EquivalenceObligation`).
|
|
293
|
-
*/
|
|
294
|
-
declare function buildEquivalenceRecord(definition: EquivalenceCheckDefinition, arms: readonly [EquivalenceArm, EquivalenceArm], obligation: EquivalenceObligation): EquivalenceRecord;
|
|
295
282
|
/**
|
|
296
283
|
* Discharge the obligation through an injected checker and return the
|
|
297
284
|
* record.
|
|
@@ -388,5 +375,5 @@ declare function certificationEvidenceDigest(evidence: unknown): string;
|
|
|
388
375
|
*/
|
|
389
376
|
declare function equivalenceVerdict(record: EquivalenceRecord): DefaultVerdict;
|
|
390
377
|
//#endregion
|
|
391
|
-
export {
|
|
392
|
-
//# sourceMappingURL=verdict-
|
|
378
|
+
export { runEquivalenceCheck as C, defineEquivalenceCheck as S, StrategyChecker as _, CheckerIdentity as a, VerificationStrategyProfile as b, EquivalenceCheckDefinition as c, EquivalenceCheckerInput as d, EquivalenceCheckerResult as f, EquivalenceRecord as g, EquivalenceProtocolError as h, equivalenceVerdict as i, EquivalenceCheckSpec as l, EquivalenceObligationStatus as m, VerdictCertification as n, CheckerOutcome as o, EquivalenceObligation as p, certificationEvidenceDigest as r, EquivalenceArm as s, DefaultVerdict as t, EquivalenceChecker as u, VERIFICATION_STRATEGIES as v, VerificationStrategySource as x, VERIFICATION_STRATEGY_SOURCES as y };
|
|
379
|
+
//# sourceMappingURL=verdict-C24NELtN.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"verdict-
|
|
1
|
+
{"version":3,"file":"verdict-C24NELtN.d.ts","names":[],"sources":["../src/verification-strategy.ts","../src/verdict.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KAqEY;;;;;;;;;UAoBK;;EAEf;;EAEA;;;;;;;;cASW,yBAAyB,OACpC,4BACA;;cAkDW,wCAEC;;;;;;UAOG;;EAEf;;EAEA;;;;;EAKA,OAAO;;;;;;;KAQG,eAAe;EACrB;EAAiB,OAAO;;EACxB;EAAkB;;;;;;;;;;;;;UAaP,gBAAgB,OAAO;;EAEtC,UAAU;EACV,UAAU;;EAEV;EACA,MAAM,OAAO,QAAQ,QAAQ,eAAe;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;UAmC7B;;;;;;EAMf,QAAQ;;;;;;EAMR;;EAEA;;;;;;EAMA;;;UAIe;WACN,MAAM,SAAS;;;UAIT;EACf;;EAEA;;;;;;;EAOA;;;;;;EAMA;;IAEE;;IAEA;;;KAIQ;;;;;;;;;;;;;UAiBK;EACf,QAAQ;;EAER;;EAEA;;EAEA,SAAS;;EAET;;;UAIe;EACf,MAAM,SAAS;EACf,gBAAgB,gBAAgB;EAChC,YAAY;;;UAIG;EACf;EACA;;;;;;;;UASe;EACf;;EAEA;EACA;;;KAIU,qBAAqB,gBAAgB,yBAAyB;;cAG7D,iCAAiC;WACnC;EAeT,YAAY,MAAM,kCAAkC;;;;;;;;;iBActC,uBAAuB,MAAM,uBAAuB;;;;;;;;;;;;;;iBAmE9C,oBACpB,YAAY,4BACZ,gBAAgB,gBAAgB,iBAChC,SAAS,qBACR,QAAQ;;;;;;;;;;;;UCjaM;;;;;;EAMf,UAAU;;;;;;EAMV,SAAS;;;;;;;EAOT;;EAEA;;;;;;;;;;UAWe;;EAEf;;EAEA;;EAEA,SAAS;;EAET;;;;;;;;EAQA,gBAAgB;;;;;;;;;;iBAWF,4BAA4B;;;;;;;;;;;;iBAqB5B,mBAAmB,QAAQ,oBAAoB"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { a as hashCanonical, r as canonicalString } from "./canonical-
|
|
1
|
+
import { a as hashCanonical, r as canonicalString } from "./canonical-Dq2hOQ8L.js";
|
|
2
2
|
import { appendFileSync, existsSync, readFileSync } from "node:fs";
|
|
3
3
|
//#region src/verdict-cache.ts
|
|
4
4
|
/**
|
|
@@ -85,4 +85,4 @@ function fileVerdictCache(path) {
|
|
|
85
85
|
//#endregion
|
|
86
86
|
export { contentHash as n, fileVerdictCache as r, canonicalJson as t };
|
|
87
87
|
|
|
88
|
-
//# sourceMappingURL=verdict-cache-
|
|
88
|
+
//# sourceMappingURL=verdict-cache-CL_7tePD.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"verdict-cache-CL_7tePD.js","names":[],"sources":["../src/verdict-cache.ts"],"sourcesContent":["/**\n * Content-addressed judge-verdict caching.\n *\n * LAW: cache JUDGE VERDICTS only — judging the same artifact with the same\n * judge+rubric is pure. NEVER cache agent rollouts. (A router that cached\n * identical fanout prompts silently destroyed best-of-N diversity; rollout\n * caching reintroduces that failure class. Judging has no diversity to\n * destroy — same artifact + same rubric ⇒ same verdict is the desired\n * property, not a bug.)\n *\n * The cache key is `v2:` + a sha-256 over the canonical JSON of everything\n * that can change a verdict: the artifact content, the scenario id, the judge\n * name, the full dimension list (key + description — the description IS the\n * rubric text shown to the judge), and a caller-supplied `judgeVersion`.\n * `judgeVersion` is REQUIRED: a judge whose prompt/model/ensemble changes\n * without a version bump would otherwise silently serve stale verdicts.\n * The `v2:` prefix names the key scheme; a store written under another scheme\n * misses once and repopulates.\n *\n * Canonical JSON is RFC 8785 from `ledger-core/canonical` — the package's one\n * identity scheme. It refuses `undefined`, functions, symbols, non-finite\n * numbers, class instances (including `Date`), `Map`, `Set`, cycles, and\n * unpaired surrogates: an artifact that cannot be unambiguously serialized\n * cannot be content-addressed, and coercing it would let two different\n * artifacts collide on one key. Pass timestamps as ISO strings.\n */\n\nimport { appendFileSync, existsSync, readFileSync } from 'node:fs'\nimport type { JudgeScore } from './campaign/types'\nimport { canonicalString, hashCanonical } from './ledger-core/canonical'\n\n// ── canonical JSON + content hash ─────────────────────────────────────────\n\n\n\n/**\n * Canonical JSON (RFC 8785): object keys sorted recursively, so two\n * semantically equal values produce byte-identical output regardless of key\n * insertion order. Throws `LedgerCanonicalizationError` on any value with no\n * faithful JSON form — see `ledger-core/canonical`.\n */\nexport function canonicalJson(value: unknown): string {\n return canonicalString(value)\n}\n\n/** Hex sha-256 over `canonicalJson(value)`. The content address used by the\n * verdict cache and report attestation. Equals `hashCanonical(value)` with\n * the `sha256:` prefix removed. */\nexport function contentHash(value: unknown): string {\n return hashCanonical(value).slice('sha256:'.length)\n}\n\n// ── store contract ─────────────────────────────────────────────────────────\n\n/** Pluggable verdict store. Sync or async on both legs — `cachedJudge`\n * awaits the results either way. */\nexport interface VerdictCacheStore {\n get(key: string): Promise<JudgeScore | undefined> | JudgeScore | undefined\n set(key: string, score: JudgeScore): Promise<void> | void\n}\n\n\n\ninterface VerdictCacheLine {\n key: string\n score: JudgeScore\n}\n\nfunction parseCacheLine(line: string, path: string, lineNo: number): VerdictCacheLine {\n let parsed: unknown\n try {\n parsed = JSON.parse(line)\n } catch (err) {\n throw new Error(\n `fileVerdictCache: corrupt JSONL at ${path}:${lineNo} — ${err instanceof Error ? err.message : String(err)}`,\n )\n }\n const rec = parsed as Partial<VerdictCacheLine>\n if (\n typeof rec !== 'object' ||\n rec === null ||\n typeof rec.key !== 'string' ||\n typeof rec.score !== 'object' ||\n rec.score === null ||\n typeof rec.score.composite !== 'number' ||\n typeof rec.score.dimensions !== 'object'\n ) {\n throw new Error(\n `fileVerdictCache: invalid record shape at ${path}:${lineNo} — expected {key, score:{dimensions, composite, notes}}`,\n )\n }\n return rec as VerdictCacheLine\n}\n\n/**\n * JSONL-file-backed store: the full file is loaded into an in-memory index at\n * construction; every `set` appends one line synchronously (durable before\n * the verdict is returned). A corrupt or malformed line throws at load with\n * file:line — a skipped line would silently re-judge (cost) or, worse, mask\n * a half-written file that needs operator attention.\n */\nexport function fileVerdictCache(path: string): VerdictCacheStore {\n const entries = new Map<string, JudgeScore>()\n if (existsSync(path)) {\n const lines = readFileSync(path, 'utf8').split('\\n')\n for (let i = 0; i < lines.length; i++) {\n const line = lines[i]\n if (line === undefined || line.trim() === '') continue\n const rec = parseCacheLine(line, path, i + 1)\n entries.set(rec.key, rec.score)\n }\n }\n return {\n get: (key) => entries.get(key),\n set: (key, score) => {\n appendFileSync(path, `${JSON.stringify({ key, score })}\\n`, 'utf8')\n entries.set(key, score)\n },\n }\n}\n\n// ── cached judge wrapper ───────────────────────────────────────────────────\n\n\n\n\n\n\n\n\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAyCA,SAAgB,cAAc,OAAwB;CACpD,OAAO,gBAAgB,KAAK;AAC9B;;;;AAKA,SAAgB,YAAY,OAAwB;CAClD,OAAO,cAAc,KAAK,CAAC,CAAC,MAAM,CAAgB;AACpD;AAkBA,SAAS,eAAe,MAAc,MAAc,QAAkC;CACpF,IAAI;CACJ,IAAI;EACF,SAAS,KAAK,MAAM,IAAI;CAC1B,SAAS,KAAK;EACZ,MAAM,IAAI,MACR,sCAAsC,KAAK,GAAG,OAAO,KAAK,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,GAC3G;CACF;CACA,MAAM,MAAM;CACZ,IACE,OAAO,QAAQ,YACf,QAAQ,QACR,OAAO,IAAI,QAAQ,YACnB,OAAO,IAAI,UAAU,YACrB,IAAI,UAAU,QACd,OAAO,IAAI,MAAM,cAAc,YAC/B,OAAO,IAAI,MAAM,eAAe,UAEhC,MAAM,IAAI,MACR,6CAA6C,KAAK,GAAG,OAAO,wDAC9D;CAEF,OAAO;AACT;;;;;;;;AASA,SAAgB,iBAAiB,MAAiC;CAChE,MAAM,0BAAU,IAAI,IAAwB;CAC5C,IAAI,WAAW,IAAI,GAAG;EACpB,MAAM,QAAQ,aAAa,MAAM,MAAM,CAAC,CAAC,MAAM,IAAI;EACnD,KAAK,IAAI,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;GACrC,MAAM,OAAO,MAAM;GACnB,IAAI,SAAS,KAAA,KAAa,KAAK,KAAK,MAAM,IAAI;GAC9C,MAAM,MAAM,eAAe,MAAM,MAAM,IAAI,CAAC;GAC5C,QAAQ,IAAI,IAAI,KAAK,IAAI,KAAK;EAChC;CACF;CACA,OAAO;EACL,MAAM,QAAQ,QAAQ,IAAI,GAAG;EAC7B,MAAM,KAAK,UAAU;GACnB,eAAe,MAAM,GAAG,KAAK,UAAU;IAAE;IAAK;GAAM,CAAC,EAAE,KAAK,MAAM;GAClE,QAAQ,IAAI,KAAK,KAAK;EACxB;CACF;AACF"}
|
package/dist/wire/index.d.ts
CHANGED
|
@@ -1,12 +1,12 @@
|
|
|
1
1
|
import { a as CostLedgerHandle, p as CustomTokenPricing } from "../cost-ledger-Jk62OBfF.js";
|
|
2
|
-
import { x as ChatClient } from "../types-
|
|
3
|
-
import { s as TraceStore } from "../store-
|
|
4
|
-
import { m as FeedbackTrajectoryStore } from "../feedback-trajectory-
|
|
2
|
+
import { x as ChatClient } from "../types-D-mgRzlh.js";
|
|
3
|
+
import { s as TraceStore } from "../store-B3nWRy3B.js";
|
|
4
|
+
import { m as FeedbackTrajectoryStore } from "../feedback-trajectory-BWmhRBGu.js";
|
|
5
5
|
import { z } from "zod";
|
|
6
6
|
import { ServerType } from "@hono/node-server";
|
|
7
7
|
import { Hono } from "hono";
|
|
8
8
|
//#region src/wire/schemas.d.ts
|
|
9
|
-
|
|
9
|
+
declare const RubricDimensionSchema: z.ZodObject<{
|
|
10
10
|
id: z.ZodString;
|
|
11
11
|
description: z.ZodString;
|
|
12
12
|
weight: z.ZodDefault<z.ZodNumber>;
|
|
@@ -17,7 +17,7 @@ declare const FailureModeSchema: z.ZodObject<{
|
|
|
17
17
|
id: z.ZodString;
|
|
18
18
|
description: z.ZodString;
|
|
19
19
|
}, z.core.$strip>;
|
|
20
|
-
|
|
20
|
+
declare const RubricSchema: z.ZodObject<{
|
|
21
21
|
name: z.ZodString;
|
|
22
22
|
description: z.ZodString;
|
|
23
23
|
systemPrompt: z.ZodString;
|
|
@@ -37,7 +37,7 @@ export declare const RubricSchema: z.ZodObject<{
|
|
|
37
37
|
description: z.ZodString;
|
|
38
38
|
}, z.core.$strip>>>;
|
|
39
39
|
}, z.core.$strip>;
|
|
40
|
-
|
|
40
|
+
declare const JudgeRequestSchema: z.ZodObject<{
|
|
41
41
|
rubricName: z.ZodOptional<z.ZodString>;
|
|
42
42
|
rubric: z.ZodOptional<z.ZodObject<{
|
|
43
43
|
name: z.ZodString;
|
|
@@ -63,7 +63,7 @@ export declare const JudgeRequestSchema: z.ZodObject<{
|
|
|
63
63
|
context: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
64
64
|
model: z.ZodOptional<z.ZodString>;
|
|
65
65
|
}, z.core.$strip>;
|
|
66
|
-
|
|
66
|
+
declare const JudgeResultSchema: z.ZodObject<{
|
|
67
67
|
composite: z.ZodNumber;
|
|
68
68
|
dimensions: z.ZodRecord<z.ZodString, z.ZodNumber>;
|
|
69
69
|
failureModes: z.ZodDefault<z.ZodArray<z.ZodString>>;
|
|
@@ -84,7 +84,7 @@ declare const RubricInfoSchema: z.ZodObject<{
|
|
|
84
84
|
failureModes: z.ZodDefault<z.ZodArray<z.ZodString>>;
|
|
85
85
|
rubricVersion: z.ZodString;
|
|
86
86
|
}, z.core.$strip>;
|
|
87
|
-
|
|
87
|
+
declare const ListRubricsResponseSchema: z.ZodObject<{
|
|
88
88
|
rubrics: z.ZodArray<z.ZodObject<{
|
|
89
89
|
name: z.ZodString;
|
|
90
90
|
description: z.ZodString;
|
|
@@ -97,19 +97,15 @@ export declare const ListRubricsResponseSchema: z.ZodObject<{
|
|
|
97
97
|
rubricVersion: z.ZodString;
|
|
98
98
|
}, z.core.$strip>>;
|
|
99
99
|
}, z.core.$strip>;
|
|
100
|
-
|
|
100
|
+
declare const VersionResponseSchema: z.ZodObject<{
|
|
101
101
|
package: z.ZodString;
|
|
102
102
|
version: z.ZodString;
|
|
103
103
|
wireVersion: z.ZodString;
|
|
104
104
|
apiSurface: z.ZodArray<z.ZodString>;
|
|
105
105
|
}, z.core.$strip>;
|
|
106
|
-
export declare const HealthResponseSchema: z.ZodObject<{
|
|
107
|
-
status: z.ZodLiteral<"ok">;
|
|
108
|
-
uptimeSec: z.ZodNumber;
|
|
109
|
-
}, z.core.$strip>;
|
|
110
106
|
/**
|
|
111
|
-
* Minimal `
|
|
112
|
-
* Matches `trace/schema.ts` `
|
|
107
|
+
* Minimal `WireTraceEvent` shape that the production runtime emits.
|
|
108
|
+
* Matches `trace/schema.ts` `WireTraceEvent` but is duplicated here as a
|
|
113
109
|
* wire schema so non-TypeScript clients can validate without depending
|
|
114
110
|
* on internal types.
|
|
115
111
|
*/
|
|
@@ -130,7 +126,7 @@ export declare const TraceEventSchema: z.ZodObject<{
|
|
|
130
126
|
timestamp: z.ZodNumber;
|
|
131
127
|
payload: z.ZodRecord<z.ZodString, z.ZodUnknown>;
|
|
132
128
|
}, z.core.$strip>;
|
|
133
|
-
|
|
129
|
+
declare const TracesIngestRequestSchema: z.ZodObject<{
|
|
134
130
|
events: z.ZodArray<z.ZodObject<{
|
|
135
131
|
eventId: z.ZodString;
|
|
136
132
|
runId: z.ZodString;
|
|
@@ -149,7 +145,7 @@ export declare const TracesIngestRequestSchema: z.ZodObject<{
|
|
|
149
145
|
payload: z.ZodRecord<z.ZodString, z.ZodUnknown>;
|
|
150
146
|
}, z.core.$strip>>;
|
|
151
147
|
}, z.core.$strip>;
|
|
152
|
-
|
|
148
|
+
declare const TracesIngestResponseSchema: z.ZodObject<{
|
|
153
149
|
accepted: z.ZodNumber;
|
|
154
150
|
rejected: z.ZodNumber;
|
|
155
151
|
errors: z.ZodDefault<z.ZodArray<z.ZodObject<{
|
|
@@ -282,15 +278,15 @@ export declare const FeedbackTrajectorySchema: z.ZodObject<{
|
|
|
282
278
|
updatedAt: z.ZodOptional<z.ZodString>;
|
|
283
279
|
metadata: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
284
280
|
}, z.core.$strip>;
|
|
285
|
-
|
|
281
|
+
declare const FeedbackIngestResponseSchema: z.ZodObject<{
|
|
286
282
|
id: z.ZodString;
|
|
287
283
|
persisted: z.ZodBoolean;
|
|
288
284
|
}, z.core.$strip>;
|
|
289
|
-
|
|
290
|
-
|
|
291
|
-
|
|
292
|
-
|
|
293
|
-
|
|
285
|
+
type WireTraceEvent = z.infer<typeof TraceEventSchema>;
|
|
286
|
+
type TracesIngestRequest = z.infer<typeof TracesIngestRequestSchema>;
|
|
287
|
+
type TracesIngestResponse = z.infer<typeof TracesIngestResponseSchema>;
|
|
288
|
+
type WireFeedbackTrajectory = z.infer<typeof FeedbackTrajectorySchema>;
|
|
289
|
+
type FeedbackIngestResponse = z.infer<typeof FeedbackIngestResponseSchema>;
|
|
294
290
|
export declare const ErrorResponseSchema: z.ZodObject<{
|
|
295
291
|
error: z.ZodObject<{
|
|
296
292
|
code: z.ZodString;
|
|
@@ -298,15 +294,15 @@ export declare const ErrorResponseSchema: z.ZodObject<{
|
|
|
298
294
|
details: z.ZodOptional<z.ZodUnknown>;
|
|
299
295
|
}, z.core.$strip>;
|
|
300
296
|
}, z.core.$strip>;
|
|
301
|
-
|
|
302
|
-
|
|
303
|
-
|
|
304
|
-
|
|
305
|
-
|
|
306
|
-
|
|
307
|
-
|
|
308
|
-
|
|
309
|
-
|
|
297
|
+
type WireRubricDimension = z.infer<typeof RubricDimensionSchema>;
|
|
298
|
+
type WireFailureMode = z.infer<typeof FailureModeSchema>;
|
|
299
|
+
type Rubric = z.infer<typeof RubricSchema>;
|
|
300
|
+
type JudgeRequest = z.infer<typeof JudgeRequestSchema>;
|
|
301
|
+
type JudgeResult = z.infer<typeof JudgeResultSchema>;
|
|
302
|
+
type RubricInfo = z.infer<typeof RubricInfoSchema>;
|
|
303
|
+
type ListRubricsResponse = z.infer<typeof ListRubricsResponseSchema>;
|
|
304
|
+
type VersionResponse = z.infer<typeof VersionResponseSchema>;
|
|
305
|
+
type ErrorResponse = z.infer<typeof ErrorResponseSchema>;
|
|
310
306
|
/**
|
|
311
307
|
* Bump the major on any breaking change to a request/response schema — the
|
|
312
308
|
* Python client compares majors and refuses a mismatch. Bump the minor when a
|
|
@@ -314,24 +310,6 @@ export type ErrorResponse = z.infer<typeof ErrorResponseSchema>;
|
|
|
314
310
|
* can detect but does not have to act on.
|
|
315
311
|
*/
|
|
316
312
|
export declare const WIRE_VERSION = "1.1.0";
|
|
317
|
-
/**
|
|
318
|
-
* The digest scheme `hashRubric` mints, named inside the value it produces.
|
|
319
|
-
* A consumer holding a tag from an earlier release reads a different scheme
|
|
320
|
-
* here and can tell "this package changed how it hashes" from "the rubric
|
|
321
|
-
* changed" — the two are indistinguishable when the value is an untagged hex
|
|
322
|
-
* string.
|
|
323
|
-
*/
|
|
324
|
-
export declare const RUBRIC_VERSION_SCHEME = "sha256-rfc8785";
|
|
325
|
-
/**
|
|
326
|
-
* Stable identity of a rubric, as `<name>@sha256-rfc8785:<hex>`. Two scores
|
|
327
|
-
* are comparable when this matches exactly.
|
|
328
|
-
*
|
|
329
|
-
* The digest comes from `hashCanonical`, this package's only canonical-JSON
|
|
330
|
-
* encoder (RFC 8785). RFC 8785 orders keys by UTF-16 code unit, which is a
|
|
331
|
-
* property of the value alone — so the same rubric produces the same tag on
|
|
332
|
-
* every machine, which a locale-collated key order does not guarantee.
|
|
333
|
-
*/
|
|
334
|
-
export declare function hashRubric(rubric: Rubric): string;
|
|
335
313
|
//#endregion
|
|
336
314
|
//#region src/wire/handlers.d.ts
|
|
337
315
|
/** Caller-fixable error. The transport renders this to 4xx + ErrorResponse. */
|
|
@@ -341,7 +319,7 @@ export declare class WireError extends Error {
|
|
|
341
319
|
readonly details?: unknown;
|
|
342
320
|
constructor(code: string, message: string, status?: number, details?: unknown);
|
|
343
321
|
}
|
|
344
|
-
|
|
322
|
+
interface HandleJudgeOptions {
|
|
345
323
|
costLedger?: CostLedgerHandle;
|
|
346
324
|
costPhase?: string;
|
|
347
325
|
/**
|
|
@@ -355,9 +333,6 @@ export interface HandleJudgeOptions {
|
|
|
355
333
|
pricing?: CustomTokenPricing;
|
|
356
334
|
signal?: AbortSignal;
|
|
357
335
|
}
|
|
358
|
-
export declare function handleJudge(req: JudgeRequest, options?: HandleJudgeOptions): Promise<JudgeResult>;
|
|
359
|
-
export declare function handleListRubrics(): ListRubricsResponse;
|
|
360
|
-
export declare function handleVersion(): VersionResponse;
|
|
361
336
|
/**
|
|
362
337
|
* Pluggable stores the wire layer routes ingestion writes into. Both
|
|
363
338
|
* are optional — when omitted, the corresponding endpoint returns 503.
|
|
@@ -366,344 +341,10 @@ export declare function handleVersion(): VersionResponse;
|
|
|
366
341
|
* `FileSystemFeedbackTrajectoryStore` here. Tests substitute in-memory
|
|
367
342
|
* stores.
|
|
368
343
|
*/
|
|
369
|
-
|
|
344
|
+
interface IngestionStores {
|
|
370
345
|
traceStore?: TraceStore;
|
|
371
346
|
feedbackStore?: FeedbackTrajectoryStore;
|
|
372
347
|
}
|
|
373
|
-
/**
|
|
374
|
-
* `POST /v1/traces/ingest` — accept a batch of `TraceEvent`s from the
|
|
375
|
-
* production runtime. Best-effort: each event is appended independently;
|
|
376
|
-
* one bad event does not poison the batch.
|
|
377
|
-
*
|
|
378
|
-
* Idempotency: the underlying store is append-only; consumers retrying
|
|
379
|
-
* the same payload will get duplicate events. Consumers should
|
|
380
|
-
* de-duplicate by `eventId` downstream — production traces frequently
|
|
381
|
-
* land via at-least-once buses (Kafka, SQS) where dedup is unavoidable.
|
|
382
|
-
*/
|
|
383
|
-
export declare function handleTracesIngest(req: TracesIngestRequest, stores: IngestionStores): Promise<TracesIngestResponse>;
|
|
384
|
-
/**
|
|
385
|
-
* `POST /v1/feedback` — accept a single `FeedbackTrajectory` from the
|
|
386
|
-
* production runtime. Idempotent on `id`: re-posting the same trajectory
|
|
387
|
-
* replaces the prior record.
|
|
388
|
-
*/
|
|
389
|
-
export declare function handleFeedbackIngest(req: FeedbackTrajectory, stores: IngestionStores): Promise<FeedbackIngestResponse>;
|
|
390
|
-
//#endregion
|
|
391
|
-
//#region node_modules/.pnpm/openapi3-ts@4.6.1/node_modules/openapi3-ts/dist/model/specification-extension.d.ts
|
|
392
|
-
type IExtensionName = `x-${string}`;
|
|
393
|
-
type IExtensionType = any;
|
|
394
|
-
type ISpecificationExtension = {
|
|
395
|
-
[extensionName: IExtensionName]: IExtensionType;
|
|
396
|
-
};
|
|
397
|
-
//#endregion
|
|
398
|
-
//#region node_modules/.pnpm/openapi3-ts@4.6.1/node_modules/openapi3-ts/dist/model/oas-common.d.ts
|
|
399
|
-
interface ServerObject extends ISpecificationExtension {
|
|
400
|
-
url: string;
|
|
401
|
-
description?: string;
|
|
402
|
-
variables?: {
|
|
403
|
-
[v: string]: ServerVariableObject;
|
|
404
|
-
};
|
|
405
|
-
}
|
|
406
|
-
interface ServerVariableObject extends ISpecificationExtension {
|
|
407
|
-
enum?: string[] | boolean[] | number[];
|
|
408
|
-
default: string | boolean | number;
|
|
409
|
-
description?: string;
|
|
410
|
-
}
|
|
411
|
-
//#endregion
|
|
412
|
-
//#region node_modules/.pnpm/openapi3-ts@4.6.1/node_modules/openapi3-ts/dist/model/openapi31.d.ts
|
|
413
|
-
interface OpenAPIObject extends ISpecificationExtension {
|
|
414
|
-
openapi: string;
|
|
415
|
-
info: InfoObject;
|
|
416
|
-
servers?: ServerObject[];
|
|
417
|
-
paths?: PathsObject;
|
|
418
|
-
components?: ComponentsObject;
|
|
419
|
-
security?: SecurityRequirementObject[];
|
|
420
|
-
tags?: TagObject[];
|
|
421
|
-
externalDocs?: ExternalDocumentationObject;
|
|
422
|
-
webhooks?: PathsObject;
|
|
423
|
-
}
|
|
424
|
-
interface InfoObject extends ISpecificationExtension {
|
|
425
|
-
title: string;
|
|
426
|
-
description?: string;
|
|
427
|
-
termsOfService?: string;
|
|
428
|
-
contact?: ContactObject;
|
|
429
|
-
license?: LicenseObject;
|
|
430
|
-
version: string;
|
|
431
|
-
}
|
|
432
|
-
interface ContactObject extends ISpecificationExtension {
|
|
433
|
-
name?: string;
|
|
434
|
-
url?: string;
|
|
435
|
-
email?: string;
|
|
436
|
-
}
|
|
437
|
-
interface LicenseObject extends ISpecificationExtension {
|
|
438
|
-
name: string;
|
|
439
|
-
identifier?: string;
|
|
440
|
-
url?: string;
|
|
441
|
-
}
|
|
442
|
-
interface ComponentsObject extends ISpecificationExtension {
|
|
443
|
-
schemas?: {
|
|
444
|
-
[schema: string]: SchemaObject | ReferenceObject;
|
|
445
|
-
};
|
|
446
|
-
responses?: {
|
|
447
|
-
[response: string]: ResponseObject | ReferenceObject;
|
|
448
|
-
};
|
|
449
|
-
parameters?: {
|
|
450
|
-
[parameter: string]: ParameterObject | ReferenceObject;
|
|
451
|
-
};
|
|
452
|
-
examples?: {
|
|
453
|
-
[example: string]: ExampleObject | ReferenceObject;
|
|
454
|
-
};
|
|
455
|
-
requestBodies?: {
|
|
456
|
-
[request: string]: RequestBodyObject | ReferenceObject;
|
|
457
|
-
};
|
|
458
|
-
headers?: {
|
|
459
|
-
[header: string]: HeaderObject | ReferenceObject;
|
|
460
|
-
};
|
|
461
|
-
securitySchemes?: {
|
|
462
|
-
[securityScheme: string]: SecuritySchemeObject | ReferenceObject;
|
|
463
|
-
};
|
|
464
|
-
links?: {
|
|
465
|
-
[link: string]: LinkObject | ReferenceObject;
|
|
466
|
-
};
|
|
467
|
-
callbacks?: {
|
|
468
|
-
[callback: string]: CallbackObject | ReferenceObject;
|
|
469
|
-
};
|
|
470
|
-
pathItems?: {
|
|
471
|
-
[pathItem: string]: PathItemObject | ReferenceObject;
|
|
472
|
-
};
|
|
473
|
-
}
|
|
474
|
-
interface PathsObject extends ISpecificationExtension {
|
|
475
|
-
[path: string]: PathItemObject;
|
|
476
|
-
}
|
|
477
|
-
interface PathItemObject extends ISpecificationExtension {
|
|
478
|
-
$ref?: string;
|
|
479
|
-
summary?: string;
|
|
480
|
-
description?: string;
|
|
481
|
-
get?: OperationObject;
|
|
482
|
-
put?: OperationObject;
|
|
483
|
-
post?: OperationObject;
|
|
484
|
-
delete?: OperationObject;
|
|
485
|
-
options?: OperationObject;
|
|
486
|
-
head?: OperationObject;
|
|
487
|
-
patch?: OperationObject;
|
|
488
|
-
trace?: OperationObject;
|
|
489
|
-
servers?: ServerObject[];
|
|
490
|
-
parameters?: (ParameterObject | ReferenceObject)[];
|
|
491
|
-
}
|
|
492
|
-
interface OperationObject extends ISpecificationExtension {
|
|
493
|
-
tags?: string[];
|
|
494
|
-
summary?: string;
|
|
495
|
-
description?: string;
|
|
496
|
-
externalDocs?: ExternalDocumentationObject;
|
|
497
|
-
operationId?: string;
|
|
498
|
-
parameters?: (ParameterObject | ReferenceObject)[];
|
|
499
|
-
requestBody?: RequestBodyObject | ReferenceObject;
|
|
500
|
-
responses?: ResponsesObject;
|
|
501
|
-
callbacks?: CallbacksObject;
|
|
502
|
-
deprecated?: boolean;
|
|
503
|
-
security?: SecurityRequirementObject[];
|
|
504
|
-
servers?: ServerObject[];
|
|
505
|
-
}
|
|
506
|
-
interface ExternalDocumentationObject extends ISpecificationExtension {
|
|
507
|
-
description?: string;
|
|
508
|
-
url: string;
|
|
509
|
-
}
|
|
510
|
-
type ParameterLocation = 'query' | 'header' | 'path' | 'cookie';
|
|
511
|
-
type ParameterStyle = 'matrix' | 'label' | 'form' | 'simple' | 'spaceDelimited' | 'pipeDelimited' | 'deepObject';
|
|
512
|
-
interface BaseParameterObject extends ISpecificationExtension {
|
|
513
|
-
description?: string;
|
|
514
|
-
required?: boolean;
|
|
515
|
-
deprecated?: boolean;
|
|
516
|
-
allowEmptyValue?: boolean;
|
|
517
|
-
style?: ParameterStyle;
|
|
518
|
-
explode?: boolean;
|
|
519
|
-
allowReserved?: boolean;
|
|
520
|
-
schema?: SchemaObject | ReferenceObject;
|
|
521
|
-
examples?: {
|
|
522
|
-
[param: string]: ExampleObject | ReferenceObject;
|
|
523
|
-
};
|
|
524
|
-
example?: any;
|
|
525
|
-
content?: ContentObject;
|
|
526
|
-
}
|
|
527
|
-
interface ParameterObject extends BaseParameterObject {
|
|
528
|
-
name: string;
|
|
529
|
-
in: ParameterLocation;
|
|
530
|
-
}
|
|
531
|
-
interface RequestBodyObject extends ISpecificationExtension {
|
|
532
|
-
description?: string;
|
|
533
|
-
content: ContentObject;
|
|
534
|
-
required?: boolean;
|
|
535
|
-
}
|
|
536
|
-
interface ContentObject {
|
|
537
|
-
[mediatype: string]: MediaTypeObject;
|
|
538
|
-
}
|
|
539
|
-
interface MediaTypeObject extends ISpecificationExtension {
|
|
540
|
-
schema?: SchemaObject | ReferenceObject;
|
|
541
|
-
examples?: ExamplesObject;
|
|
542
|
-
example?: any;
|
|
543
|
-
encoding?: EncodingObject;
|
|
544
|
-
}
|
|
545
|
-
interface EncodingObject extends ISpecificationExtension {
|
|
546
|
-
[property: string]: EncodingPropertyObject | any;
|
|
547
|
-
}
|
|
548
|
-
interface EncodingPropertyObject {
|
|
549
|
-
contentType?: string;
|
|
550
|
-
headers?: {
|
|
551
|
-
[key: string]: HeaderObject | ReferenceObject;
|
|
552
|
-
};
|
|
553
|
-
style?: string;
|
|
554
|
-
explode?: boolean;
|
|
555
|
-
allowReserved?: boolean;
|
|
556
|
-
[key: string]: any;
|
|
557
|
-
}
|
|
558
|
-
interface ResponsesObject extends ISpecificationExtension {
|
|
559
|
-
default?: ResponseObject | ReferenceObject;
|
|
560
|
-
[statuscode: string]: ResponseObject | ReferenceObject | any;
|
|
561
|
-
}
|
|
562
|
-
interface ResponseObject extends ISpecificationExtension {
|
|
563
|
-
description: string;
|
|
564
|
-
headers?: HeadersObject;
|
|
565
|
-
content?: ContentObject;
|
|
566
|
-
links?: LinksObject;
|
|
567
|
-
}
|
|
568
|
-
interface CallbacksObject extends ISpecificationExtension {
|
|
569
|
-
[name: string]: CallbackObject | ReferenceObject | any;
|
|
570
|
-
}
|
|
571
|
-
interface CallbackObject extends ISpecificationExtension {
|
|
572
|
-
[name: string]: PathItemObject | any;
|
|
573
|
-
}
|
|
574
|
-
interface HeadersObject {
|
|
575
|
-
[name: string]: HeaderObject | ReferenceObject;
|
|
576
|
-
}
|
|
577
|
-
interface ExampleObject {
|
|
578
|
-
summary?: string;
|
|
579
|
-
description?: string;
|
|
580
|
-
value?: any;
|
|
581
|
-
externalValue?: string;
|
|
582
|
-
[property: string]: any;
|
|
583
|
-
}
|
|
584
|
-
interface LinksObject {
|
|
585
|
-
[name: string]: LinkObject | ReferenceObject;
|
|
586
|
-
}
|
|
587
|
-
interface LinkObject extends ISpecificationExtension {
|
|
588
|
-
operationRef?: string;
|
|
589
|
-
operationId?: string;
|
|
590
|
-
parameters?: LinkParametersObject;
|
|
591
|
-
requestBody?: any | string;
|
|
592
|
-
description?: string;
|
|
593
|
-
server?: ServerObject;
|
|
594
|
-
[property: string]: any;
|
|
595
|
-
}
|
|
596
|
-
interface LinkParametersObject {
|
|
597
|
-
[name: string]: any | string;
|
|
598
|
-
}
|
|
599
|
-
interface HeaderObject extends BaseParameterObject {
|
|
600
|
-
$ref?: string;
|
|
601
|
-
}
|
|
602
|
-
interface TagObject extends ISpecificationExtension {
|
|
603
|
-
name: string;
|
|
604
|
-
description?: string;
|
|
605
|
-
externalDocs?: ExternalDocumentationObject;
|
|
606
|
-
[extension: string]: any;
|
|
607
|
-
}
|
|
608
|
-
interface ExamplesObject {
|
|
609
|
-
[name: string]: ExampleObject | ReferenceObject;
|
|
610
|
-
}
|
|
611
|
-
interface ReferenceObject {
|
|
612
|
-
$ref: string;
|
|
613
|
-
summary?: string;
|
|
614
|
-
description?: string;
|
|
615
|
-
}
|
|
616
|
-
type SchemaObjectType = 'integer' | 'number' | 'string' | 'boolean' | 'object' | 'null' | 'array';
|
|
617
|
-
interface SchemaObject extends ISpecificationExtension {
|
|
618
|
-
$ref?: string;
|
|
619
|
-
discriminator?: DiscriminatorObject;
|
|
620
|
-
readOnly?: boolean;
|
|
621
|
-
writeOnly?: boolean;
|
|
622
|
-
xml?: XmlObject;
|
|
623
|
-
externalDocs?: ExternalDocumentationObject;
|
|
624
|
-
example?: any;
|
|
625
|
-
examples?: any[];
|
|
626
|
-
deprecated?: boolean;
|
|
627
|
-
type?: SchemaObjectType | SchemaObjectType[];
|
|
628
|
-
format?: 'int32' | 'int64' | 'float' | 'double' | 'byte' | 'binary' | 'date' | 'date-time' | 'password' | string;
|
|
629
|
-
allOf?: (SchemaObject | ReferenceObject)[];
|
|
630
|
-
oneOf?: (SchemaObject | ReferenceObject)[];
|
|
631
|
-
anyOf?: (SchemaObject | ReferenceObject)[];
|
|
632
|
-
not?: SchemaObject | ReferenceObject;
|
|
633
|
-
items?: SchemaObject | ReferenceObject;
|
|
634
|
-
properties?: {
|
|
635
|
-
[propertyName: string]: SchemaObject | ReferenceObject;
|
|
636
|
-
};
|
|
637
|
-
additionalProperties?: SchemaObject | ReferenceObject | boolean;
|
|
638
|
-
propertyNames?: SchemaObject | ReferenceObject;
|
|
639
|
-
description?: string;
|
|
640
|
-
default?: any;
|
|
641
|
-
title?: string;
|
|
642
|
-
multipleOf?: number;
|
|
643
|
-
maximum?: number;
|
|
644
|
-
const?: any;
|
|
645
|
-
exclusiveMaximum?: number;
|
|
646
|
-
minimum?: number;
|
|
647
|
-
exclusiveMinimum?: number;
|
|
648
|
-
maxLength?: number;
|
|
649
|
-
minLength?: number;
|
|
650
|
-
pattern?: string;
|
|
651
|
-
maxItems?: number;
|
|
652
|
-
minItems?: number;
|
|
653
|
-
uniqueItems?: boolean;
|
|
654
|
-
maxProperties?: number;
|
|
655
|
-
minProperties?: number;
|
|
656
|
-
required?: string[];
|
|
657
|
-
enum?: any[];
|
|
658
|
-
prefixItems?: (SchemaObject | ReferenceObject)[];
|
|
659
|
-
contentMediaType?: string;
|
|
660
|
-
contentEncoding?: string;
|
|
661
|
-
}
|
|
662
|
-
interface DiscriminatorObject {
|
|
663
|
-
propertyName: string;
|
|
664
|
-
mapping?: {
|
|
665
|
-
[key: string]: string;
|
|
666
|
-
};
|
|
667
|
-
}
|
|
668
|
-
interface XmlObject extends ISpecificationExtension {
|
|
669
|
-
name?: string;
|
|
670
|
-
namespace?: string;
|
|
671
|
-
prefix?: string;
|
|
672
|
-
attribute?: boolean;
|
|
673
|
-
wrapped?: boolean;
|
|
674
|
-
}
|
|
675
|
-
type SecuritySchemeType = 'apiKey' | 'http' | 'oauth2' | 'openIdConnect';
|
|
676
|
-
interface SecuritySchemeObject extends ISpecificationExtension {
|
|
677
|
-
type: SecuritySchemeType;
|
|
678
|
-
description?: string;
|
|
679
|
-
name?: string;
|
|
680
|
-
in?: string;
|
|
681
|
-
scheme?: string;
|
|
682
|
-
bearerFormat?: string;
|
|
683
|
-
flows?: OAuthFlowsObject;
|
|
684
|
-
openIdConnectUrl?: string;
|
|
685
|
-
}
|
|
686
|
-
interface OAuthFlowsObject extends ISpecificationExtension {
|
|
687
|
-
implicit?: OAuthFlowObject;
|
|
688
|
-
password?: OAuthFlowObject;
|
|
689
|
-
clientCredentials?: OAuthFlowObject;
|
|
690
|
-
authorizationCode?: OAuthFlowObject;
|
|
691
|
-
}
|
|
692
|
-
interface OAuthFlowObject extends ISpecificationExtension {
|
|
693
|
-
authorizationUrl?: string;
|
|
694
|
-
tokenUrl?: string;
|
|
695
|
-
refreshUrl?: string;
|
|
696
|
-
scopes: ScopesObject;
|
|
697
|
-
}
|
|
698
|
-
interface ScopesObject extends ISpecificationExtension {
|
|
699
|
-
[scope: string]: any;
|
|
700
|
-
}
|
|
701
|
-
interface SecurityRequirementObject {
|
|
702
|
-
[name: string]: string[];
|
|
703
|
-
}
|
|
704
|
-
//#endregion
|
|
705
|
-
//#region src/wire/openapi.d.ts
|
|
706
|
-
export declare function buildOpenApi(packageVersion: string): OpenAPIObject;
|
|
707
348
|
//#endregion
|
|
708
349
|
//#region src/wire/rpc.d.ts
|
|
709
350
|
interface RpcRequest {
|
|
@@ -726,26 +367,6 @@ interface RpcOptions {
|
|
|
726
367
|
judgeModel?: string;
|
|
727
368
|
}
|
|
728
369
|
export declare function dispatchRpc(req: RpcRequest, options?: RpcOptions): Promise<RpcSuccess | RpcError>;
|
|
729
|
-
/** Read one JSON request from stdin, write one JSON response to stdout. */
|
|
730
|
-
export declare function runRpcOnce(method?: string, options?: RpcOptions): Promise<number>;
|
|
731
|
-
/** Read JSONL requests from stdin, write JSONL responses to stdout. */
|
|
732
|
-
export declare function runRpcBatch(method?: string, options?: RpcOptions): Promise<number>;
|
|
733
|
-
//#endregion
|
|
734
|
-
//#region src/wire/rubrics.d.ts
|
|
735
|
-
/** Get a built-in rubric by name, or undefined. */
|
|
736
|
-
export declare function getBuiltinRubric(name: string): Rubric | undefined;
|
|
737
|
-
/** List built-in rubrics with their stable versions. */
|
|
738
|
-
export declare function listBuiltinRubrics(): {
|
|
739
|
-
name: string;
|
|
740
|
-
description: string;
|
|
741
|
-
dimensions: {
|
|
742
|
-
id: string;
|
|
743
|
-
description: string;
|
|
744
|
-
weight: number;
|
|
745
|
-
}[];
|
|
746
|
-
failureModes: string[];
|
|
747
|
-
rubricVersion: string;
|
|
748
|
-
}[];
|
|
749
370
|
//#endregion
|
|
750
371
|
//#region src/wire/server.d.ts
|
|
751
372
|
interface CreateAppOptions {
|
|
@@ -798,5 +419,5 @@ interface StartedServer {
|
|
|
798
419
|
*/
|
|
799
420
|
export declare function startServerAsync(opts?: ServeOptions): Promise<StartedServer>;
|
|
800
421
|
//#endregion
|
|
801
|
-
export type { RpcOptions, ServeOptions, StartedServer };
|
|
422
|
+
export type { ErrorResponse, FeedbackIngestResponse, HandleJudgeOptions, IngestionStores, JudgeRequest, JudgeResult, ListRubricsResponse, RpcOptions, Rubric, RubricInfo, ServeOptions, StartedServer, TracesIngestRequest, TracesIngestResponse, VersionResponse, WireFailureMode, WireFeedbackTrajectory, WireRubricDimension, WireTraceEvent };
|
|
802
423
|
//# sourceMappingURL=index.d.ts.map
|
package/dist/wire/index.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","names":[
|
|
1
|
+
{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/wire/schemas.ts","../../src/wire/handlers.ts","../../src/wire/rpc.ts","../../src/wire/server.ts"],"mappings":";;;;;;;;cAoBa,uBAAqB,EAAA;;;;;;GAkBD,EAAA,KAAA;cAE3B,mBAAiB,EAAA;;;GAKM,EAAA,KAAA;cAIhB,cAAY,EAAA;;;;;;;;;;;;;;;;;;;GA6BL,EAAA,KAAA;cAIP,oBAAkB,EAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAwBL,EAAA,KAAA;cAEb,mBAAiB,EAAA;;;;;;;;;GA6BL,EAAA,KAAA;cAInB,kBAAgB,EAAA;;;;;;;;;;GAcE,EAAA,KAAA;cAEX,2BAAyB,EAAA;;;;;;;;;;;;GAIL,EAAA,KAAA;cAIpB,uBAAqB,EAAA;;;;;GAWL,EAAA,KAAA;;;;;;;qBAiBhB,kBAAgB,EAAA;;;;;;;;;;;;;;;;GA0BD,EAAA,KAAA;cAEf,2BAAyB,EAAA;;;;;;;;;;;;;;;;;;GAQL,EAAA,KAAA;cAEpB,4BAA0B,EAAA;;;;;;;GAiBL,EAAA,KAAA;qBA2DrB,0BAAwB,EAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BD,EAAA,KAAA;cAEvB,8BAA4B,EAAA;;;GAKL,EAAA,KAAA;KAExB,iBAAiB,EAAE,aAAa;KAChC,sBAAsB,EAAE,aAAa;KACrC,uBAAuB,EAAE,aAAa;KACtC,yBAAyB,EAAE,aAAa;KACxC,yBAAyB,EAAE,aAAa;qBAIvC,qBAAmB,EAAA;;;;;;GAcL,EAAA,KAAA;KAIf,sBAAsB,EAAE,aAAa;KACrC,kBAAkB,EAAE,aAAa;KACjC,SAAS,EAAE,aAAa;KACxB,eAAe,EAAE,aAAa;KAC9B,cAAc,EAAE,aAAa;KAC7B,aAAa,EAAE,aAAa;KAC5B,sBAAsB,EAAE,aAAa;KACrC,kBAAkB,EAAE,aAAa;KACjC,gBAAgB,EAAE,aAAa;;;;;;;qBAU9B;;;;qBChWA,kBAAkB;WAEX;WAEA;WACA;EAJlB,YACkB,cAChB,iBACgB,iBACA;;UAgJH;EACf,aAAa;EACb;;;;;;EAMA,OAAO;EACP;;EAEA,UAAU;EACV,SAAS;;;;;;;;;;UAsGM;EACf,aAAa;EACb,gBAAgB;;;;UC5RR;EACR;EACA;;UAGQ;EACR;;UAGQ;EACR;IAAS;IAAc;IAAiB;;;UAGzB;;EAEf,OAAO;EACP;;wBAGoB,YACpB,KAAK,YACL,UAAS,aACR,QAAQ,aAAa;;;UCNd;;EAER,SAAS;;;;;;EAMT,OAAO;;EAEP;;;;;;;;;;EAUA;IACE,mBAAmB,4BAA4B;;;wBAMnC,UAAU,OAAM,mBAAqB,0BAAA,+BAAA;UA0IpC,qBAAqB;;EAEpC;;EAEA;;wBAGc,YAAY,OAAM,eAAoB;UAUrC;EACf,QAAQ;;;;EAIR;;EAEA;;EAEA,SAAS;;;;;;;wBAQK,iBAAiB,OAAM,eAAoB,QAAQ"}
|
package/dist/wire/index.js
CHANGED
|
@@ -1,2 +1,2 @@
|
|
|
1
|
-
import {
|
|
2
|
-
export { ErrorResponseSchema,
|
|
1
|
+
import { a as dispatchRpc, c as WireError, d as FeedbackTrajectorySchema, f as TraceEventSchema, n as startServer, p as WIRE_VERSION, r as startServerAsync, t as createApp, u as ErrorResponseSchema } from "../server-DoVbqqtJ.js";
|
|
2
|
+
export { ErrorResponseSchema, FeedbackTrajectorySchema, TraceEventSchema, WIRE_VERSION, WireError, createApp, dispatchRpc, startServer, startServerAsync };
|
|
@@ -5,7 +5,7 @@
|
|
|
5
5
|
thresholds and result policy. Review identity includes authored option order; request JSON is
|
|
6
6
|
not sorted before dispatch. Missing or partial negative evidence cannot become a clean review.
|
|
7
7
|
|
|
8
|
-
The optional
|
|
8
|
+
The optional behavior-review recipe (its example file is not in this tree) covers reward gaming, evaluator tampering,
|
|
9
9
|
policy/authorization/data-boundary breaches, injection-following, contradicted completion,
|
|
10
10
|
oversight bypass, matched-baseline anomalies and cross-run behavior. It composes the existing
|
|
11
11
|
`jevAnalyst`; the taxonomy is not installed in the generic evaluator and is not a calibrated
|