@tangle-network/agent-eval 0.221.8 → 0.222.1
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 +168 -0
- package/dist/adapters/http.d.ts +3 -5
- package/dist/adapters/http.d.ts.map +1 -1
- package/dist/adapters/http.js +1 -2
- package/dist/{adversarial-smnADNFS.d.ts → adversarial-DnQQFqTR.d.ts} +1 -3
- package/dist/adversarial-DnQQFqTR.d.ts.map +1 -0
- package/dist/{agent-profile-CNmwfaqq.d.ts → agent-profile-CSEQRC9l.d.ts} +4 -10
- package/dist/agent-profile-CSEQRC9l.d.ts.map +1 -0
- package/dist/agent-profile-Dngk3LZ0.js +153 -0
- package/dist/agent-profile-Dngk3LZ0.js.map +1 -0
- package/dist/{agent-profile-cell-BcpFbNd5.d.ts → agent-profile-cell-3kYxUsK2.d.ts} +3 -5
- package/dist/agent-profile-cell-3kYxUsK2.d.ts.map +1 -0
- package/dist/{agent-profile-cell-B1V1gUZs.js → agent-profile-cell-CT5JYueM.js} +5 -7
- package/dist/{agent-profile-cell-B1V1gUZs.js.map → agent-profile-cell-CT5JYueM.js.map} +1 -1
- package/dist/{agreement-irr-BhAqjluA.js → agreement-irr-BTjDN0i6.js} +4 -5
- package/dist/{agreement-irr-BhAqjluA.js.map → agreement-irr-BTjDN0i6.js.map} +1 -1
- package/dist/analyst/index.d.ts +95 -159
- package/dist/analyst/index.d.ts.map +1 -1
- package/dist/analyst/index.js +28 -39
- package/dist/analyst/index.js.map +1 -1
- package/dist/{analyze-runs-DfX6dHP0.d.ts → analyze-runs-CGLRgZGd.d.ts} +7 -13
- package/dist/analyze-runs-CGLRgZGd.d.ts.map +1 -0
- package/dist/{analyze-runs-CHmsykNJ.js → analyze-runs-CTheASYV.js} +23 -32
- package/dist/{analyze-runs-CHmsykNJ.js.map → analyze-runs-CTheASYV.js.map} +1 -1
- package/dist/{attribute-vocabulary-CcXqHODT.d.ts → attribute-vocabulary-5oo8ykIx.d.ts} +2 -4
- package/dist/attribute-vocabulary-5oo8ykIx.d.ts.map +1 -0
- package/dist/authenticity/index.d.ts +0 -2
- package/dist/authenticity/index.d.ts.map +1 -1
- package/dist/authenticity/index.js +1 -2
- package/dist/{benchmark-command-Cnjfo0Zh.js → benchmark-command-Cr7i7YSX.js} +61 -94
- package/dist/benchmark-command-Cr7i7YSX.js.map +1 -0
- package/dist/benchmarks/index.d.ts +5 -15
- package/dist/benchmarks/index.d.ts.map +1 -1
- package/dist/benchmarks/index.js +27 -38
- package/dist/benchmarks/index.js.map +1 -1
- package/dist/{bounded-process-Cj9YjXbE.js → bounded-process-BZrWXIKW.js} +2 -3
- package/dist/{bounded-process-Cj9YjXbE.js.map → bounded-process-BZrWXIKW.js.map} +1 -1
- package/dist/builder-eval/index.d.ts +5 -14
- package/dist/builder-eval/index.d.ts.map +1 -1
- package/dist/builder-eval/index.js +11 -17
- package/dist/builder-eval/index.js.map +1 -1
- package/dist/{calibration-CTwxpJ-t.js → calibration-6usfVYYo.js} +4 -5
- package/dist/{calibration-CTwxpJ-t.js.map → calibration-6usfVYYo.js.map} +1 -1
- package/dist/campaign/index.d.ts +9 -9
- package/dist/campaign/index.js +15 -14
- package/dist/{campaign-uddTlFg5.js → campaign-ddr1N3yD.js} +61 -58
- package/dist/campaign-ddr1N3yD.js.map +1 -0
- package/dist/{campaign-evidence-DunzcJGv.js → campaign-evidence-DN8rWuBj.js} +20 -322
- package/dist/campaign-evidence-DN8rWuBj.js.map +1 -0
- package/dist/{canonical-Dq2hOQ8L.js → canonical-CicJlINA.js} +3 -4
- package/dist/{canonical-Dq2hOQ8L.js.map → canonical-CicJlINA.js.map} +1 -1
- package/dist/{canonical-CFpojCN5.d.ts → canonical-gLeHiqzd.d.ts} +2 -4
- package/dist/canonical-gLeHiqzd.d.ts.map +1 -0
- package/dist/{cell-spend-DEWR3nj-.js → cell-spend-zxmzg4GH.js} +3 -4
- package/dist/{cell-spend-DEWR3nj-.js.map → cell-spend-zxmzg4GH.js.map} +1 -1
- package/dist/{chat-client-BNATB3rJ.js → chat-client-CO3uB33i.js} +3 -4
- package/dist/{chat-client-BNATB3rJ.js.map → chat-client-CO3uB33i.js.map} +1 -1
- package/dist/{chat-json-call-BzWYp18J.js → chat-json-call-8yUv4LFO.js} +4 -5
- package/dist/{chat-json-call-BzWYp18J.js.map → chat-json-call-8yUv4LFO.js.map} +1 -1
- package/dist/{claim-BPVuyHuy.js → claim-Dod_mjk3.js} +5 -6
- package/dist/{claim-BPVuyHuy.js.map → claim-Dod_mjk3.js.map} +1 -1
- package/dist/cli.js +179 -23
- package/dist/cli.js.map +1 -1
- package/dist/{client-hwo3cJjR.d.ts → client-BIFwUg7I.d.ts} +6 -12
- package/dist/client-BIFwUg7I.d.ts.map +1 -0
- package/dist/{cluster-bootstrap-DEwwhl0C.js → cluster-bootstrap-SbaeM44s.js} +4 -5
- package/dist/{cluster-bootstrap-DEwwhl0C.js.map → cluster-bootstrap-SbaeM44s.js.map} +1 -1
- package/dist/completion-verifier-7KEfroqx.d.ts +168 -0
- package/dist/completion-verifier-7KEfroqx.d.ts.map +1 -0
- package/dist/{completion-verifier-asaV1nSj.js → completion-verifier-rqbMh5-4.js} +9 -160
- package/dist/completion-verifier-rqbMh5-4.js.map +1 -0
- package/dist/contract/index.d.ts +13 -13
- package/dist/contract/index.js +41 -60
- package/dist/contract/index.js.map +1 -1
- package/dist/{cost-ledger-BGZB6yY2.js → cost-ledger-CS8R3IZj.js} +10 -17
- package/dist/{cost-ledger-BGZB6yY2.js.map → cost-ledger-CS8R3IZj.js.map} +1 -1
- package/dist/{cost-ledger-C1W3ICmP.d.ts → cost-ledger-Czro9wmo.d.ts} +1 -5
- package/dist/cost-ledger-Czro9wmo.d.ts.map +1 -0
- package/dist/{dataset-DQqhOCPt.d.ts → dataset-DThC1S6G.d.ts} +1 -3
- package/dist/dataset-DThC1S6G.d.ts.map +1 -0
- package/dist/{decision-point-0nJubuoM.js → decision-point-lhMgODOR.js} +14 -18
- package/dist/{decision-point-0nJubuoM.js.map → decision-point-lhMgODOR.js.map} +1 -1
- package/dist/{deep-freeze-CurcOQg3.js → deep-freeze-CK8-2W7p.js} +2 -3
- package/dist/{deep-freeze-CurcOQg3.js.map → deep-freeze-CK8-2W7p.js.map} +1 -1
- package/dist/{default-registry-CLG5YqGS.d.ts → default-registry-1f4ukAIX.d.ts} +53 -61
- package/dist/default-registry-1f4ukAIX.d.ts.map +1 -0
- package/dist/{default-registry-BBTDgkDM.js → default-registry-C1tPEjZ4.js} +29 -45
- package/dist/{default-registry-BBTDgkDM.js.map → default-registry-C1tPEjZ4.js.map} +1 -1
- package/dist/{define-agent-eval-CCbT7Kfr.js → define-agent-eval-BLoP4Lf9.js} +15 -18
- package/dist/{define-agent-eval-CCbT7Kfr.js.map → define-agent-eval-BLoP4Lf9.js.map} +1 -1
- package/dist/{define-agent-eval-JKt3gMQr.d.ts → define-agent-eval-sUYh9BFA.d.ts} +19 -27
- package/dist/define-agent-eval-sUYh9BFA.d.ts.map +1 -0
- package/dist/{descriptive-CVNy48rR.js → descriptive-cPwITVfy.js} +30 -8
- package/dist/descriptive-cPwITVfy.js.map +1 -0
- package/dist/diagnosis/index.d.ts +5 -15
- package/dist/diagnosis/index.d.ts.map +1 -1
- package/dist/diagnosis/index.js +13 -19
- package/dist/diagnosis/index.js.map +1 -1
- package/dist/{dspy-rlm-engine-DN1p_5jx.js → dspy-rlm-engine-jDiWkul6.js} +7 -10
- package/dist/{dspy-rlm-engine-DN1p_5jx.js.map → dspy-rlm-engine-jDiWkul6.js.map} +1 -1
- package/dist/{emitter-DbMXUIVP.js → emitter-DhQ65mza.js} +3 -4
- package/dist/{emitter-DbMXUIVP.js.map → emitter-DhQ65mza.js.map} +1 -1
- package/dist/{emitter-DABYYTcH.d.ts → emitter-mAXC5PSN.d.ts} +3 -5
- package/dist/emitter-mAXC5PSN.d.ts.map +1 -0
- package/dist/{engine-dtRVeuQs.d.ts → engine-BITiErg-.d.ts} +10 -18
- package/dist/engine-BITiErg-.d.ts.map +1 -0
- package/dist/{equal-terms-DbPGcVtI.js → equal-terms-K11VHmR3.js} +3 -4
- package/dist/{equal-terms-DbPGcVtI.js.map → equal-terms-K11VHmR3.js.map} +1 -1
- package/dist/{errors-BcOs5MKW.js → errors-DeEEV9Vx.js} +2 -3
- package/dist/{errors-BcOs5MKW.js.map → errors-DeEEV9Vx.js.map} +1 -1
- package/dist/{errors-DEE6u6ot.d.ts → errors-E9I2hldP.d.ts} +1 -3
- package/dist/errors-E9I2hldP.d.ts.map +1 -0
- package/dist/{eval-campaign-DoX5vz3E.js → eval-campaign-Cgixrvf7.js} +8 -9
- package/dist/{eval-campaign-DoX5vz3E.js.map → eval-campaign-Cgixrvf7.js.map} +1 -1
- package/dist/{evaluation-wdGyZ7jV.js → evaluation-B8H3GYej.js} +4 -6
- package/dist/{evaluation-wdGyZ7jV.js.map → evaluation-B8H3GYej.js.map} +1 -1
- package/dist/{evaluation-DblGquCB.d.ts → evaluation-Djq7-Ybj.d.ts} +4 -6
- package/dist/evaluation-Djq7-Ybj.d.ts.map +1 -0
- package/dist/evaluation.d.ts +1 -1
- package/dist/evaluation.js +1 -2
- package/dist/{exact-types-Do63KoKn.d.ts → exact-types-CndDT_h7.d.ts} +8 -10
- package/dist/exact-types-CndDT_h7.d.ts.map +1 -0
- package/dist/{exec-CpbtOXVt.js → exec-C0_AkLUv.js} +3 -5
- package/dist/{exec-CpbtOXVt.js.map → exec-C0_AkLUv.js.map} +1 -1
- package/dist/{execution-tracks-CpgFPpS5.js → execution-tracks-DaDXYwAq.js} +2 -3
- package/dist/{execution-tracks-CpgFPpS5.js.map → execution-tracks-DaDXYwAq.js.map} +1 -1
- package/dist/experiment/index.d.ts +9 -37
- package/dist/experiment/index.d.ts.map +1 -1
- package/dist/experiment/index.js +29 -40
- package/dist/experiment/index.js.map +1 -1
- package/dist/{exporters-Dk3oEbK_.d.ts → exporters-B5Ze_RKh.d.ts} +4 -4
- package/dist/exporters-B5Ze_RKh.d.ts.map +1 -0
- package/dist/{exporters-DWvM16LE.js → exporters-ByLcILYU.js} +23 -5
- package/dist/exporters-ByLcILYU.js.map +1 -0
- package/dist/{external-optimizer-process-DEzJ2CGX.js → external-optimizer-process-BRVmaDFe.js} +5 -6
- package/dist/{external-optimizer-process-DEzJ2CGX.js.map → external-optimizer-process-BRVmaDFe.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-B_ukPtRD.js → external-optimizer-subprocess-U616hV_l.js} +24 -37
- package/dist/{external-optimizer-subprocess-B_ukPtRD.js.map → external-optimizer-subprocess-U616hV_l.js.map} +1 -1
- package/dist/{extract-usage-BZLU01kR.js → extract-usage-ISFqhfDz.js} +2 -3
- package/dist/{extract-usage-BZLU01kR.js.map → extract-usage-ISFqhfDz.js.map} +1 -1
- package/dist/{failure-cluster-DN0t63Z0.d.ts → failure-cluster-DOoa7CBg.d.ts} +3 -7
- package/dist/failure-cluster-DOoa7CBg.d.ts.map +1 -0
- package/dist/{failure-taxonomy-Bx5oWqaW.js → failure-taxonomy-9eQfXKoW.js} +2 -3
- package/dist/{failure-taxonomy-Bx5oWqaW.js.map → failure-taxonomy-9eQfXKoW.js.map} +1 -1
- package/dist/{feedback-trajectory-CYVJXolG.d.ts → feedback-trajectory-DsHR9krG.d.ts} +7 -13
- package/dist/feedback-trajectory-DsHR9krG.d.ts.map +1 -0
- package/dist/{finding-composition-8n0fFgVI.js → finding-composition-DlvE1Qul.js} +4 -6
- package/dist/{finding-composition-8n0fFgVI.js.map → finding-composition-DlvE1Qul.js.map} +1 -1
- package/dist/{findings-store-Bg4sdS-e.js → findings-store-B75Dspaq.js} +11 -14
- package/dist/{findings-store-Bg4sdS-e.js.map → findings-store-B75Dspaq.js.map} +1 -1
- package/dist/fuzz.d.ts +3 -17
- package/dist/fuzz.d.ts.map +1 -1
- package/dist/fuzz.js +30 -29
- package/dist/fuzz.js.map +1 -1
- package/dist/{hf-dataset-CKxIJQba.js → hf-dataset-DPVQXW5n.js} +362 -21
- package/dist/hf-dataset-DPVQXW5n.js.map +1 -0
- package/dist/hosted/index.d.ts +2 -6
- package/dist/hosted/index.d.ts.map +1 -1
- package/dist/hosted/index.js +1 -2
- package/dist/{index-CalAsXMR.d.ts → index-B1Bm9oJA.d.ts} +3 -15
- package/dist/index-B1Bm9oJA.d.ts.map +1 -0
- package/dist/{index-D7fmC-1K.d.ts → index-BdKzYYkC.d.ts} +183 -36
- package/dist/index-BdKzYYkC.d.ts.map +1 -0
- package/dist/{index-CdEVr8-O.d.ts → index-Bdcpx2s8.d.ts} +24 -299
- package/dist/index-Bdcpx2s8.d.ts.map +1 -0
- package/dist/{index-jnuFLxOs.d.ts → index-Cu12qxJb.d.ts} +11 -39
- package/dist/index-Cu12qxJb.d.ts.map +1 -0
- package/dist/{index-CfJULcW5.d.ts → index-DAk2quYu.d.ts} +18 -38
- package/dist/index-DAk2quYu.d.ts.map +1 -0
- package/dist/{index-CHxwfW5g.d.ts → index-DwUGehqX.d.ts} +3 -9
- package/dist/index-DwUGehqX.d.ts.map +1 -0
- package/dist/{index-CTIRz-AB.d.ts → index-IzWTz5EP.d.ts} +14 -36
- package/dist/index-IzWTz5EP.d.ts.map +1 -0
- package/dist/index.d.ts +55 -188
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +122 -195
- package/dist/index.js.map +1 -1
- package/dist/{integrity-D-s2iHiT.js → integrity-B1V6klyO.js} +12 -363
- package/dist/integrity-B1V6klyO.js.map +1 -0
- package/dist/{integrity-D9lBnv_K.d.ts → integrity-BAdVzT8u.d.ts} +4 -6
- package/dist/integrity-BAdVzT8u.d.ts.map +1 -0
- package/dist/{integrity-W4CADcFc.js → integrity-DOHt1mwU.js} +3 -4
- package/dist/{integrity-W4CADcFc.js.map → integrity-DOHt1mwU.js.map} +1 -1
- package/dist/{internal-BTK91Bqt.js → internal-v6XfOeJW.js} +6 -8
- package/dist/{internal-BTK91Bqt.js.map → internal-v6XfOeJW.js.map} +1 -1
- package/dist/{interval-Dq9gt7jx.js → interval-Bd2f4CRT.js} +6 -8
- package/dist/{interval-Dq9gt7jx.js.map → interval-Bd2f4CRT.js.map} +1 -1
- package/dist/{interval-ohuOJurX.d.ts → interval-CWyxqDpE.d.ts} +1 -5
- package/dist/interval-CWyxqDpE.d.ts.map +1 -0
- package/dist/{journal-C7om8Gum.js → journal-C1uLJi0w.js} +115 -121
- package/dist/{journal-C7om8Gum.js.map → journal-C1uLJi0w.js.map} +1 -1
- package/dist/{json-BhdsJaOQ.js → json-DtvZNAU9.js} +4 -6
- package/dist/{json-BhdsJaOQ.js.map → json-DtvZNAU9.js.map} +1 -1
- package/dist/json-recovery-DK4MDm5r.js +296 -0
- package/dist/json-recovery-DK4MDm5r.js.map +1 -0
- package/dist/{judge-calibration-BkXMIWA4.js → judge-calibration-DGqlfWW8.js} +5 -6
- package/dist/{judge-calibration-BkXMIWA4.js.map → judge-calibration-DGqlfWW8.js.map} +1 -1
- package/dist/{judge-calibration-DFtEMlde.d.ts → judge-calibration-DnrbQaXI.d.ts} +1 -3
- package/dist/judge-calibration-DnrbQaXI.d.ts.map +1 -0
- package/dist/{judge-gate-BVvmP00P.js → judge-gate-C2BCOt-M.js} +8 -12
- package/dist/{judge-gate-BVvmP00P.js.map → judge-gate-C2BCOt-M.js.map} +1 -1
- package/dist/judges-BLxg5p8E.js +20 -0
- package/dist/judges-BLxg5p8E.js.map +1 -0
- package/dist/{kind-factory-Dl0PmNQL.js → kind-factory-VKysp1ds.js} +23 -39
- package/dist/{kind-factory-Dl0PmNQL.js.map → kind-factory-VKysp1ds.js.map} +1 -1
- package/dist/ledger-core/index.d.ts +2 -2
- package/dist/ledger-core/index.js +3 -3
- package/dist/ledger-core/lock-holder.d.ts +1 -1
- package/dist/ledger-core/lock-holder.js +1 -2
- package/dist/{ledger-core-CAxUdDk0.js → ledger-core-CfT-FzPf.js} +5 -6
- package/dist/{ledger-core-CAxUdDk0.js.map → ledger-core-CfT-FzPf.js.map} +1 -1
- package/dist/{llm-client-b1ysZVIU.js → llm-client-kJ9EOqjt.js} +22 -30
- package/dist/{llm-client-b1ysZVIU.js.map → llm-client-kJ9EOqjt.js.map} +1 -1
- package/dist/{llm-judge-Cybq29bL.js → llm-judge-cgbhTKQp.js} +108 -164
- package/dist/llm-judge-cgbhTKQp.js.map +1 -0
- package/dist/{lock-holder-DaJbfpci.d.ts → lock-holder-DoxlZFkQ.d.ts} +1 -3
- package/dist/lock-holder-DoxlZFkQ.d.ts.map +1 -0
- package/dist/matrix/index.d.ts +3 -3
- package/dist/matrix/index.js +2 -2
- package/dist/{matrix-DtkZLD_3.d.ts → matrix-8X6i5kJ6.d.ts} +4 -12
- package/dist/matrix-8X6i5kJ6.d.ts.map +1 -0
- package/dist/{matrix-B8l-Nsxd.js → matrix-CgpX8wdm.js} +8 -13
- package/dist/matrix-CgpX8wdm.js.map +1 -0
- package/dist/{mean-BbA_JCzW.js → mean-CdDYG-us.js} +6 -9
- package/dist/{mean-BbA_JCzW.js.map → mean-CdDYG-us.js.map} +1 -1
- package/dist/meta-eval/index.d.ts +3 -3
- package/dist/meta-eval/index.js +17 -21
- package/dist/meta-eval/index.js.map +1 -1
- package/dist/{metrics-BKaPBtIG.js → metrics-DNCAUtuB.js} +2 -3
- package/dist/{metrics-BKaPBtIG.js.map → metrics-DNCAUtuB.js.map} +1 -1
- package/dist/{mint-CbdwjTPW.js → mint-D9ZysYtY.js} +7 -8
- package/dist/{mint-CbdwjTPW.js.map → mint-D9ZysYtY.js.map} +1 -1
- package/dist/{multi-layer-verifier-DEgMUk6v.d.ts → multi-layer-verifier-lM4aVN78.d.ts} +2 -4
- package/dist/multi-layer-verifier-lM4aVN78.d.ts.map +1 -0
- package/dist/{multiplicity-Sg0U7h8j.js → multiplicity-DAd4xaDC.js} +3 -4
- package/dist/{multiplicity-Sg0U7h8j.js.map → multiplicity-DAd4xaDC.js.map} +1 -1
- package/dist/{multiplicity-DIWHvysC.d.ts → multiplicity-D_SEsc5j.d.ts} +1 -3
- package/dist/multiplicity-D_SEsc5j.d.ts.map +1 -0
- package/dist/multishot/golden/index.d.ts +1 -17
- package/dist/multishot/golden/index.d.ts.map +1 -1
- package/dist/multishot/golden/index.js +10 -17
- package/dist/multishot/golden/index.js.map +1 -1
- package/dist/multishot/index.d.ts +2 -10
- package/dist/multishot/index.d.ts.map +1 -1
- package/dist/multishot/index.js +11 -18
- package/dist/multishot/index.js.map +1 -1
- package/dist/{nonnegative-number-DvdMxg_f.js → nonnegative-number-lEYbOsuh.js} +2 -3
- package/dist/{nonnegative-number-DvdMxg_f.js.map → nonnegative-number-lEYbOsuh.js.map} +1 -1
- package/dist/openapi.json +1 -1
- package/dist/{otel-spans-64hmQqON.js → otel-spans-BsGzTWIK.js} +5 -6
- package/dist/{otel-spans-64hmQqON.js.map → otel-spans-BsGzTWIK.js.map} +1 -1
- package/dist/{otlp-attributes-D9ptWaIJ.js → otlp-attributes-Boeb7gtH.js} +2 -3
- package/dist/{otlp-attributes-D9ptWaIJ.js.map → otlp-attributes-Boeb7gtH.js.map} +1 -1
- package/dist/{outcome-observations-FLmRpgYa.js → outcome-observations-D4snOtUb.js} +4 -5
- package/dist/{outcome-observations-FLmRpgYa.js.map → outcome-observations-D4snOtUb.js.map} +1 -1
- package/dist/{outcome-store-DUcMj-3D.js → outcome-store-TarrcfHG.js} +2 -3
- package/dist/{outcome-store-DUcMj-3D.js.map → outcome-store-TarrcfHG.js.map} +1 -1
- package/dist/{package-version-D7lQHt_-.js → package-version-CJl_pY9W.js} +2 -3
- package/dist/{package-version-D7lQHt_-.js.map → package-version-CJl_pY9W.js.map} +1 -1
- package/dist/{paired-arms-B2ttLc0_.js → paired-arms-D-54hO-k.js} +10 -14
- package/dist/{paired-arms-B2ttLc0_.js.map → paired-arms-D-54hO-k.js.map} +1 -1
- package/dist/{paired-binary-B_nejA44.js → paired-binary-BR0nwHOA.js} +3 -4
- package/dist/{paired-binary-B_nejA44.js.map → paired-binary-BR0nwHOA.js.map} +1 -1
- package/dist/{paired-promotion-decision-CQMJ5xGN.d.ts → paired-promotion-decision-BDnqepZ1.d.ts} +1 -5
- package/dist/paired-promotion-decision-BDnqepZ1.d.ts.map +1 -0
- package/dist/{paired-promotion-decision-D0SHb-ZL.js → paired-promotion-decision-FGvHA4p_.js} +5 -7
- package/dist/{paired-promotion-decision-D0SHb-ZL.js.map → paired-promotion-decision-FGvHA4p_.js.map} +1 -1
- package/dist/{pareto-BiBnQXAE.js → pareto-B9udAt2i.js} +28 -50
- package/dist/pareto-B9udAt2i.js.map +1 -0
- package/dist/pipelines/index.d.ts +5 -11
- package/dist/pipelines/index.d.ts.map +1 -1
- package/dist/pipelines/index.js +6 -8
- package/dist/pipelines/index.js.map +1 -1
- package/dist/{power-and-mde-CQX-0weG.js → power-and-mde-ClI6RWw4.js} +4 -5
- package/dist/{power-and-mde-CQX-0weG.js.map → power-and-mde-ClI6RWw4.js.map} +1 -1
- package/dist/{predicate-FdKmElA_.js → predicate-BYHA4MBn.js} +2 -3
- package/dist/{predicate-FdKmElA_.js.map → predicate-BYHA4MBn.js.map} +1 -1
- package/dist/{prime-bridge-transport-6feEglLf.d.ts → prime-bridge-transport-Cfqir_Im.d.ts} +1 -3
- package/dist/prime-bridge-transport-Cfqir_Im.d.ts.map +1 -0
- package/dist/{prime-bridge-transport-Z3cpKGoV.js → prime-bridge-transport-Csd5SQg7.js} +2 -3
- package/dist/{prime-bridge-transport-Z3cpKGoV.js.map → prime-bridge-transport-Csd5SQg7.js.map} +1 -1
- package/dist/{prime-protocol-BGxjfhM4.d.ts → prime-protocol-BpQlTowq.d.ts} +4 -8
- package/dist/prime-protocol-BpQlTowq.d.ts.map +1 -0
- package/dist/{prime-protocol-BFgGrEMP.js → prime-protocol-CrhpXRz9.js} +4 -6
- package/dist/{prime-protocol-BFgGrEMP.js.map → prime-protocol-CrhpXRz9.js.map} +1 -1
- package/dist/profile-cell.d.ts +1 -1
- package/dist/profile-cell.js +1 -2
- package/dist/{promotion-gate-CkVZHagC.d.ts → promotion-gate-CixshGI8.d.ts} +5 -11
- package/dist/promotion-gate-CixshGI8.d.ts.map +1 -0
- package/dist/{promotion-policy-6A8osGr_.js → promotion-policy-CBYeD4zi.js} +5 -6
- package/dist/{promotion-policy-6A8osGr_.js.map → promotion-policy-CBYeD4zi.js.map} +1 -1
- package/dist/{proposal-findings-DqLUWsAG.js → proposal-findings-WFQrxnfe.js} +3 -4
- package/dist/{proposal-findings-DqLUWsAG.js.map → proposal-findings-WFQrxnfe.js.map} +1 -1
- package/dist/{quantile--0yb07pD.js → quantile-CncXP-Ss.js} +2 -3
- package/dist/{quantile--0yb07pD.js.map → quantile-CncXP-Ss.js.map} +1 -1
- package/dist/{query-B_RZ36an.js → query-CGXzO9d4.js} +5 -6
- package/dist/{query-B_RZ36an.js.map → query-CGXzO9d4.js.map} +1 -1
- package/dist/{query-Cmv_U8vf.d.ts → query-UD0B08Sb.d.ts} +4 -6
- package/dist/query-UD0B08Sb.d.ts.map +1 -0
- package/dist/{random-DGEyr3zG.d.ts → random-BN150Q_X.d.ts} +1 -5
- package/dist/random-BN150Q_X.d.ts.map +1 -0
- package/dist/{random-DDDcC7YU.js → random-DrzcgC7F.js} +3 -4
- package/dist/{random-DDDcC7YU.js.map → random-DrzcgC7F.js.map} +1 -1
- package/dist/{raw-provider-sink-75w4FWBj.js → raw-provider-sink-DDLkP_eF.js} +3 -4
- package/dist/{raw-provider-sink-75w4FWBj.js.map → raw-provider-sink-DDLkP_eF.js.map} +1 -1
- package/dist/{record-id-DUgsK5qp.js → record-id-CWZXFh58.js} +2 -3
- package/dist/{record-id-DUgsK5qp.js.map → record-id-CWZXFh58.js.map} +1 -1
- package/dist/{redact-DXdlFb_U.d.ts → redact-C_wzC4c_.d.ts} +1 -3
- package/dist/redact-C_wzC4c_.d.ts.map +1 -0
- package/dist/{redact-mtkTiJVk.js → redact-D6APJS1d.js} +3 -5
- package/dist/{redact-mtkTiJVk.js.map → redact-D6APJS1d.js.map} +1 -1
- package/dist/{registry-BcdJmyoU.d.ts → registry-D9A7Tybn.d.ts} +5 -11
- package/dist/registry-D9A7Tybn.d.ts.map +1 -0
- package/dist/{release-confidence-CYIFH-fp.js → release-confidence-CTUJz5mM.js} +10 -16
- package/dist/release-confidence-CTUJz5mM.js.map +1 -0
- package/dist/{render-BhJqXoBH.js → render-pfAZ7IR4.js} +266 -33
- package/dist/render-pfAZ7IR4.js.map +1 -0
- package/dist/{reporting-html-Kvlc8oMt.js → reporting-html-DMXIjATy.js} +2 -3
- package/dist/{reporting-html-Kvlc8oMt.js.map → reporting-html-DMXIjATy.js.map} +1 -1
- package/dist/reporting.d.ts +6 -6
- package/dist/reporting.js +7 -7
- package/dist/{researcher-DKWtoMWk.d.ts → researcher-CLnNJ7Or.d.ts} +8 -12
- package/dist/researcher-CLnNJ7Or.d.ts.map +1 -0
- package/dist/{reward-nw2xZGZG.js → reward-P0VD_Aft.js} +2 -3
- package/dist/{reward-nw2xZGZG.js.map → reward-P0VD_Aft.js.map} +1 -1
- package/dist/{reward-BeUgVVC_.d.ts → reward-ST4PXNXF.d.ts} +2 -4
- package/dist/reward-ST4PXNXF.d.ts.map +1 -0
- package/dist/{reward-hacking-WTxUbtbb.d.ts → reward-hacking-C9hPVT0Q.d.ts} +4 -8
- package/dist/reward-hacking-C9hPVT0Q.d.ts.map +1 -0
- package/dist/{reward-hacking-ByuETaIb.js → reward-hacking-DZzHZlS1.js} +5 -7
- package/dist/{reward-hacking-ByuETaIb.js.map → reward-hacking-DZzHZlS1.js.map} +1 -1
- package/dist/rl.d.ts +26 -63
- package/dist/rl.d.ts.map +1 -1
- package/dist/rl.js +35 -52
- package/dist/rl.js.map +1 -1
- package/dist/{rolldown-runtime-8H4AJuhK.js → rolldown-runtime-BJohB4LM.js} +1 -2
- package/dist/rollout/index.d.ts +6 -6
- package/dist/rollout/index.js +7 -7
- package/dist/{rollout-Dd1ucT_M.js → rollout-BhkMpAJT.js} +279 -16
- package/dist/rollout-BhkMpAJT.js.map +1 -0
- package/dist/{terminal-record-CB9BFtjS.js → rollout-nodes-BHR9Q5jP.js} +351 -11
- package/dist/rollout-nodes-BHR9Q5jP.js.map +1 -0
- package/dist/{round-decimal-COrKglMh.js → round-decimal-ZmjAQPo1.js} +2 -3
- package/dist/{round-decimal-COrKglMh.js.map → round-decimal-ZmjAQPo1.js.map} +1 -1
- package/dist/{rubric-predictive-validity-Cd5SCtAf.d.ts → rubric-predictive-validity-B3RzbwOx.d.ts} +2 -8
- package/dist/rubric-predictive-validity-B3RzbwOx.d.ts.map +1 -0
- package/dist/{rubric-predictive-validity-CeKW2GQN.js → rubric-predictive-validity-CQp_M0vy.js} +3 -4
- package/dist/{rubric-predictive-validity-CeKW2GQN.js.map → rubric-predictive-validity-CQp_M0vy.js.map} +1 -1
- package/dist/{run-record-CysKLB93.js → run-record-CZy4-TnZ.js} +4 -5
- package/dist/{run-record-CysKLB93.js.map → run-record-CZy4-TnZ.js.map} +1 -1
- package/dist/{run-record-C3awiXPf.d.ts → run-record-D2BFMPGl.d.ts} +5 -7
- package/dist/run-record-D2BFMPGl.d.ts.map +1 -0
- package/dist/{run-record-DhEJgoPz.js → run-record-DEjx43cI.js} +6 -7
- package/dist/{run-record-DhEJgoPz.js.map → run-record-DEjx43cI.js.map} +1 -1
- package/dist/{sandbox-harness-BlSOu4LX.d.ts → sandbox-harness-B3Kx1fOA.d.ts} +1 -3
- package/dist/sandbox-harness-B3Kx1fOA.d.ts.map +1 -0
- package/dist/{schema-D6EVKzTB.d.ts → schema-CkIvnmLU.d.ts} +2 -43
- package/dist/schema-CkIvnmLU.d.ts.map +1 -0
- package/dist/{schema-Bt2vLFm3.js → schema-D-xu3MU_.js} +2 -3
- package/dist/{schema-Bt2vLFm3.js.map → schema-D-xu3MU_.js.map} +1 -1
- package/dist/{schema-C-ziyJD7.d.ts → schema-DkFJr3Y8.d.ts} +111 -48
- package/dist/schema-DkFJr3Y8.d.ts.map +1 -0
- package/dist/{schema-CvhfLjtZ.js → schema-YUIXT8iW.js} +37 -7
- package/dist/schema-YUIXT8iW.js.map +1 -0
- package/dist/search/index.d.ts +3 -35
- package/dist/search/index.d.ts.map +1 -1
- package/dist/search/index.js +4 -5
- package/dist/{search-kernel-CctpiDd2.d.ts → search-kernel-B77eE7Ac.d.ts} +10 -36
- package/dist/search-kernel-B77eE7Ac.d.ts.map +1 -0
- package/dist/{search-ledger-CZRqtFbW.js → search-ledger-BsVi-Ngv.js} +17 -26
- package/dist/{search-ledger-CZRqtFbW.js.map → search-ledger-BsVi-Ngv.js.map} +1 -1
- package/dist/{search-ledger-file-Bd5Tioq6.js → search-ledger-file-DTBD6X-y.js} +6 -8
- package/dist/{search-ledger-file-Bd5Tioq6.js.map → search-ledger-file-DTBD6X-y.js.map} +1 -1
- package/dist/{search-ledger-types-p2xKfiDJ.d.ts → search-ledger-types-qiYRUo_N.d.ts} +5 -9
- package/dist/search-ledger-types-qiYRUo_N.d.ts.map +1 -0
- package/dist/{search-shipper-Dn3NLsuH.js → search-shipper-pdg4viK7.js} +8 -13
- package/dist/{search-shipper-Dn3NLsuH.js.map → search-shipper-pdg4viK7.js.map} +1 -1
- package/dist/{search-summary-D4bwYnAE.js → search-summary-Ct7SrmVc.js} +6 -7
- package/dist/{search-summary-D4bwYnAE.js.map → search-summary-Ct7SrmVc.js.map} +1 -1
- package/dist/{sequential-CimQrbjv.js → sequential-B1I0tRfL.js} +6 -7
- package/dist/{sequential-CimQrbjv.js.map → sequential-B1I0tRfL.js.map} +1 -1
- package/dist/{sequential-KwGspfeR.d.ts → sequential-Bb_pdTqG.d.ts} +1 -3
- package/dist/sequential-Bb_pdTqG.d.ts.map +1 -0
- package/dist/{sequential-BVv1sjsw.js → sequential-Bjs5roiK.js} +2 -3
- package/dist/{sequential-BVv1sjsw.js.map → sequential-Bjs5roiK.js.map} +1 -1
- package/dist/{sequential-eprocess-DodI6RoF.js → sequential-eprocess-C1YLwFaz.js} +3 -4
- package/dist/{sequential-eprocess-DodI6RoF.js.map → sequential-eprocess-C1YLwFaz.js.map} +1 -1
- package/dist/{server-ChhXPCgX.js → server-DA6h3co5.js} +11 -17
- package/dist/{server-ChhXPCgX.js.map → server-DA6h3co5.js.map} +1 -1
- package/dist/{single-run-lock-Dekd8_Wv.js → single-run-lock-BGlW3OSk.js} +3 -4
- package/dist/{single-run-lock-Dekd8_Wv.js.map → single-run-lock-BGlW3OSk.js.map} +1 -1
- package/dist/{skillopt-optimization-method-sB_-yi1G.js → skillopt-optimization-method-qB1M554K.js} +38 -57
- package/dist/{skillopt-optimization-method-sB_-yi1G.js.map → skillopt-optimization-method-qB1M554K.js.map} +1 -1
- package/dist/{spans-CVngGij3.d.ts → spans-vZoW__3q.d.ts} +2 -4
- package/dist/spans-vZoW__3q.d.ts.map +1 -0
- package/dist/{statistical-heldout-BjzSOmLH.d.ts → statistical-heldout-BYxHe3KO.d.ts} +9 -37
- package/dist/statistical-heldout-BYxHe3KO.d.ts.map +1 -0
- package/dist/{statistical-heldout-Bksv92Th.js → statistical-heldout-TP6ndL0B.js} +4 -5
- package/dist/{statistical-heldout-Bksv92Th.js.map → statistical-heldout-TP6ndL0B.js.map} +1 -1
- package/dist/{steps-DlepJPwD.d.ts → steps-B8fYY_Vo.d.ts} +1 -3
- package/dist/steps-B8fYY_Vo.d.ts.map +1 -0
- package/dist/{store-B3nWRy3B.d.ts → store-CMs5wGXP.d.ts} +2 -4
- package/dist/store-CMs5wGXP.d.ts.map +1 -0
- package/dist/{store-BhNuLdJw.js → store-ZykIyQmy.js} +2 -3
- package/dist/{store-BhNuLdJw.js.map → store-ZykIyQmy.js.map} +1 -1
- package/dist/{store-otlp-CFf27E8_.js → store-otlp-DmLm65XW.js} +8 -13
- package/dist/{store-otlp-CFf27E8_.js.map → store-otlp-DmLm65XW.js.map} +1 -1
- package/dist/{store-tool-spans-BK_8FjGG.js → store-tool-spans-C2g8UYm-.js} +17 -41
- package/dist/{store-tool-spans-BK_8FjGG.js.map → store-tool-spans-C2g8UYm-.js.map} +1 -1
- package/dist/{store-tool-spans-DxsFwtdr.d.ts → store-tool-spans-DMM6wMwB.d.ts} +11 -29
- package/dist/store-tool-spans-DMM6wMwB.d.ts.map +1 -0
- package/dist/storyboard/index.d.ts +1 -5
- package/dist/storyboard/index.d.ts.map +1 -1
- package/dist/storyboard/index.js +3 -5
- package/dist/storyboard/index.js.map +1 -1
- package/dist/{summary-report-C8V49Jt3.d.ts → summary-report-0La5vejY.d.ts} +5 -13
- package/dist/summary-report-0La5vejY.d.ts.map +1 -0
- package/dist/{summary-report-DHDWxe6p.js → summary-report-CPfsrLac.js} +21 -30
- package/dist/summary-report-CPfsrLac.js.map +1 -0
- package/dist/supervisor-run/index.d.ts +30 -29
- package/dist/supervisor-run/index.d.ts.map +1 -1
- package/dist/supervisor-run/index.js +6 -7
- package/dist/{systemone-CHnqwmYg.d.ts → systemone-UHbKD6LH.d.ts} +12 -24
- package/dist/systemone-UHbKD6LH.d.ts.map +1 -0
- package/dist/{systemone-protocol-Bl2EI2KQ.d.ts → systemone-protocol-C4t7JH4Y.d.ts} +1 -3
- package/dist/systemone-protocol-C4t7JH4Y.d.ts.map +1 -0
- package/dist/systemone-protocol.d.ts +1 -1
- package/dist/systemone-protocol.js +2 -3
- package/dist/{systemone-review-cQ9acAsb.d.ts → systemone-review-Cn9Uj0xG.d.ts} +8 -10
- package/dist/systemone-review-Cn9Uj0xG.d.ts.map +1 -0
- package/dist/{systemone-review-DFGtIUcf.js → systemone-review-S49gsy_y.js} +5 -6
- package/dist/{systemone-review-DFGtIUcf.js.map → systemone-review-S49gsy_y.js.map} +1 -1
- package/dist/systemone.d.ts +4 -4
- package/dist/systemone.js +3 -3
- package/dist/{task-failure-attributes-oIK3BhCR.js → task-failure-attributes-CC6VuIpk.js} +7 -10
- package/dist/{task-failure-attributes-oIK3BhCR.js.map → task-failure-attributes-CC6VuIpk.js.map} +1 -1
- package/dist/{tool-groups-b3oU_D88.d.ts → tool-groups-CFWmAGCV.d.ts} +3 -5
- package/dist/tool-groups-CFWmAGCV.d.ts.map +1 -0
- package/dist/{tool-waste-CkjzWmeY.d.ts → tool-waste-BFG1vwTV.d.ts} +4 -16
- package/dist/tool-waste-BFG1vwTV.d.ts.map +1 -0
- package/dist/{tool-waste-DLsjS_p-.js → tool-waste-CLJ46AbT.js} +10 -26
- package/dist/{tool-waste-DLsjS_p-.js.map → tool-waste-CLJ46AbT.js.map} +1 -1
- package/dist/trace-attributes.d.ts +1 -1
- package/dist/trace-attributes.js +1 -2
- package/dist/{trace-questions-BmT8M6UQ.d.ts → trace-questions-BFCQAzoT.d.ts} +4 -6
- package/dist/trace-questions-BFCQAzoT.d.ts.map +1 -0
- package/dist/trace-repair/index.d.ts +46 -92
- package/dist/trace-repair/index.d.ts.map +1 -1
- package/dist/trace-repair/index.js +32 -54
- package/dist/trace-repair/index.js.map +1 -1
- package/dist/traces.d.ts +13 -29
- package/dist/traces.d.ts.map +1 -1
- package/dist/traces.js +18 -22
- package/dist/traces.js.map +1 -1
- package/dist/{trajectory-D_7rLrvE.js → trajectory-O3GDHmev.js} +2 -3
- package/dist/{trajectory-D_7rLrvE.js.map → trajectory-O3GDHmev.js.map} +1 -1
- package/dist/trajectory-replay/index.d.ts +2 -22
- package/dist/trajectory-replay/index.d.ts.map +1 -1
- package/dist/trajectory-replay/index.js +19 -30
- package/dist/trajectory-replay/index.js.map +1 -1
- package/dist/{tree-DSqwJpL9.js → tree-DrIChF4a.js} +39 -68
- package/dist/{tree-DSqwJpL9.js.map → tree-DrIChF4a.js.map} +1 -1
- package/dist/{types-B_IFcKp5.d.ts → types-0KglnQ9o.d.ts} +3 -15
- package/dist/types-0KglnQ9o.d.ts.map +1 -0
- package/dist/{types-6Lfdlshp.d.ts → types-BEvjDziM.d.ts} +50 -62
- package/dist/types-BEvjDziM.d.ts.map +1 -0
- package/dist/{types-CnWmj8NW.js → types-BK0blWhe.js} +3 -4
- package/dist/{types-CnWmj8NW.js.map → types-BK0blWhe.js.map} +1 -1
- package/dist/{types-CmXY0PQf.d.ts → types-C-aOqgt2.d.ts} +2 -5
- package/dist/types-C-aOqgt2.d.ts.map +1 -0
- package/dist/{types-BXgASBCL.js → types-DK4vNv5q.js} +2 -3
- package/dist/{types-BXgASBCL.js.map → types-DK4vNv5q.js.map} +1 -1
- package/dist/{types-DZ0mYVw9.d.ts → types-DUR0dqET.d.ts} +3 -5
- package/dist/types-DUR0dqET.d.ts.map +1 -0
- package/dist/{types-M6eF9m-r.d.ts → types-e3HGaI7P.d.ts} +84 -24
- package/dist/types-e3HGaI7P.d.ts.map +1 -0
- package/dist/{verdict-BGQmxQ8A.js → verdict-C8TcdwfD.js} +3 -4
- package/dist/{verdict-BGQmxQ8A.js.map → verdict-C8TcdwfD.js.map} +1 -1
- package/dist/{verdict-C24NELtN.d.ts → verdict-D_qiBQ6_.d.ts} +1 -5
- package/dist/verdict-D_qiBQ6_.d.ts.map +1 -0
- package/dist/{verdict-cache-CL_7tePD.js → verdict-cache-BVH-fl18.js} +3 -4
- package/dist/{verdict-cache-CL_7tePD.js.map → verdict-cache-BVH-fl18.js.map} +1 -1
- package/dist/wire/index.d.ts +43 -51
- package/dist/wire/index.d.ts.map +1 -1
- package/dist/wire/index.js +1 -2
- package/docs/eval-surface-map.md +15 -0
- package/docs/rollout.md +132 -6
- package/package.json +18 -18
- package/dist/adversarial-smnADNFS.d.ts.map +0 -1
- package/dist/agent-profile-CNmwfaqq.d.ts.map +0 -1
- package/dist/agent-profile-cell-BcpFbNd5.d.ts.map +0 -1
- package/dist/analyze-runs-DfX6dHP0.d.ts.map +0 -1
- package/dist/attribute-vocabulary-CcXqHODT.d.ts.map +0 -1
- package/dist/benchmark-command-Cnjfo0Zh.js.map +0 -1
- package/dist/campaign-evidence-DunzcJGv.js.map +0 -1
- package/dist/campaign-uddTlFg5.js.map +0 -1
- package/dist/canonical-CFpojCN5.d.ts.map +0 -1
- package/dist/client-hwo3cJjR.d.ts.map +0 -1
- package/dist/completion-verifier-asaV1nSj.js.map +0 -1
- package/dist/cost-ledger-C1W3ICmP.d.ts.map +0 -1
- package/dist/dataset-DQqhOCPt.d.ts.map +0 -1
- package/dist/default-registry-CLG5YqGS.d.ts.map +0 -1
- package/dist/define-agent-eval-JKt3gMQr.d.ts.map +0 -1
- package/dist/descriptive-B2iPaT9J.d.ts +0 -89
- package/dist/descriptive-B2iPaT9J.d.ts.map +0 -1
- package/dist/descriptive-CVNy48rR.js.map +0 -1
- package/dist/emitter-DABYYTcH.d.ts.map +0 -1
- package/dist/engine-dtRVeuQs.d.ts.map +0 -1
- package/dist/errors-DEE6u6ot.d.ts.map +0 -1
- package/dist/evaluation-DblGquCB.d.ts.map +0 -1
- package/dist/exact-types-Do63KoKn.d.ts.map +0 -1
- package/dist/exporters-DWvM16LE.js.map +0 -1
- package/dist/exporters-Dk3oEbK_.d.ts.map +0 -1
- package/dist/failure-cluster-DN0t63Z0.d.ts.map +0 -1
- package/dist/feedback-trajectory-CYVJXolG.d.ts.map +0 -1
- package/dist/hf-dataset-CKxIJQba.js.map +0 -1
- package/dist/index-CHxwfW5g.d.ts.map +0 -1
- package/dist/index-CTIRz-AB.d.ts.map +0 -1
- package/dist/index-CalAsXMR.d.ts.map +0 -1
- package/dist/index-CdEVr8-O.d.ts.map +0 -1
- package/dist/index-CfJULcW5.d.ts.map +0 -1
- package/dist/index-D7fmC-1K.d.ts.map +0 -1
- package/dist/index-jnuFLxOs.d.ts.map +0 -1
- package/dist/integrity-D-s2iHiT.js.map +0 -1
- package/dist/integrity-D9lBnv_K.d.ts.map +0 -1
- package/dist/interval-ohuOJurX.d.ts.map +0 -1
- package/dist/judge-calibration-DFtEMlde.d.ts.map +0 -1
- package/dist/llm-judge-Cybq29bL.js.map +0 -1
- package/dist/lock-holder-DaJbfpci.d.ts.map +0 -1
- package/dist/matrix-B8l-Nsxd.js.map +0 -1
- package/dist/matrix-DtkZLD_3.d.ts.map +0 -1
- package/dist/mint-CkzpVPdz.d.ts +0 -81
- package/dist/mint-CkzpVPdz.d.ts.map +0 -1
- package/dist/multi-layer-verifier-DEgMUk6v.d.ts.map +0 -1
- package/dist/multiplicity-DIWHvysC.d.ts.map +0 -1
- package/dist/paired-promotion-decision-CQMJ5xGN.d.ts.map +0 -1
- package/dist/pareto-BiBnQXAE.js.map +0 -1
- package/dist/prime-bridge-transport-6feEglLf.d.ts.map +0 -1
- package/dist/prime-protocol-BGxjfhM4.d.ts.map +0 -1
- package/dist/promotion-gate-CkVZHagC.d.ts.map +0 -1
- package/dist/query-Cmv_U8vf.d.ts.map +0 -1
- package/dist/random-DGEyr3zG.d.ts.map +0 -1
- package/dist/redact-DXdlFb_U.d.ts.map +0 -1
- package/dist/registry-BcdJmyoU.d.ts.map +0 -1
- package/dist/release-confidence-CYIFH-fp.js.map +0 -1
- package/dist/render-BhJqXoBH.js.map +0 -1
- package/dist/researcher-DKWtoMWk.d.ts.map +0 -1
- package/dist/reward-BeUgVVC_.d.ts.map +0 -1
- package/dist/reward-hacking-WTxUbtbb.d.ts.map +0 -1
- package/dist/rollout-Dd1ucT_M.js.map +0 -1
- package/dist/rubric-predictive-validity-Cd5SCtAf.d.ts.map +0 -1
- package/dist/run-record-C3awiXPf.d.ts.map +0 -1
- package/dist/sandbox-harness-BlSOu4LX.d.ts.map +0 -1
- package/dist/schema-C-ziyJD7.d.ts.map +0 -1
- package/dist/schema-CvhfLjtZ.js.map +0 -1
- package/dist/schema-D6EVKzTB.d.ts.map +0 -1
- package/dist/search-kernel-CctpiDd2.d.ts.map +0 -1
- package/dist/search-ledger-types-p2xKfiDJ.d.ts.map +0 -1
- package/dist/sequential-KwGspfeR.d.ts.map +0 -1
- package/dist/spans-CVngGij3.d.ts.map +0 -1
- package/dist/statistical-heldout-BjzSOmLH.d.ts.map +0 -1
- package/dist/steps-DlepJPwD.d.ts.map +0 -1
- package/dist/store-B3nWRy3B.d.ts.map +0 -1
- package/dist/store-tool-spans-DxsFwtdr.d.ts.map +0 -1
- package/dist/summary-report-C8V49Jt3.d.ts.map +0 -1
- package/dist/summary-report-DHDWxe6p.js.map +0 -1
- package/dist/systemone-CHnqwmYg.d.ts.map +0 -1
- package/dist/systemone-protocol-Bl2EI2KQ.d.ts.map +0 -1
- package/dist/systemone-review-cQ9acAsb.d.ts.map +0 -1
- package/dist/terminal-record-CB9BFtjS.js.map +0 -1
- package/dist/tool-groups-b3oU_D88.d.ts.map +0 -1
- package/dist/tool-waste-CkjzWmeY.d.ts.map +0 -1
- package/dist/trace-questions-BmT8M6UQ.d.ts.map +0 -1
- package/dist/types-6Lfdlshp.d.ts.map +0 -1
- package/dist/types-B_IFcKp5.d.ts.map +0 -1
- package/dist/types-CmXY0PQf.d.ts.map +0 -1
- package/dist/types-DZ0mYVw9.d.ts.map +0 -1
- package/dist/types-M6eF9m-r.d.ts.map +0 -1
- package/dist/verdict-C24NELtN.d.ts.map +0 -1
|
@@ -1,6 +1,5 @@
|
|
|
1
|
-
import { t as DefaultVerdict } from "../verdict-
|
|
2
|
-
import { a as RecordedTrajectoryTurn, c as finalRecordedOutcome, i as RecordedTrajectoryStep, n as RecordedFinalOutcome, o as classifyObservation, r as RecordedObservationKind, s as decodeRecordedTurns, t as DecodedTrajectory } from "../steps-
|
|
3
|
-
//#region src/trajectory-replay/corpus.d.ts
|
|
1
|
+
import { t as DefaultVerdict } from "../verdict-D_qiBQ6_.js";
|
|
2
|
+
import { a as RecordedTrajectoryTurn, c as finalRecordedOutcome, i as RecordedTrajectoryStep, n as RecordedFinalOutcome, o as classifyObservation, r as RecordedObservationKind, s as decodeRecordedTurns, t as DecodedTrajectory } from "../steps-B8fYY_Vo.js";
|
|
4
3
|
interface CorpusSpec {
|
|
5
4
|
readonly name: string;
|
|
6
5
|
readonly labelsPath: string;
|
|
@@ -52,8 +51,6 @@ interface EnumerationResult {
|
|
|
52
51
|
readonly excluded: ExcludedCase[];
|
|
53
52
|
readonly labelEntryCount: number;
|
|
54
53
|
}
|
|
55
|
-
//#endregion
|
|
56
|
-
//#region src/trajectory-replay/exec.d.ts
|
|
57
54
|
/**
|
|
58
55
|
* The execution boundary replay runs across.
|
|
59
56
|
*
|
|
@@ -80,8 +77,6 @@ interface ReplayExecBackend {
|
|
|
80
77
|
* internally (batch, corpus wire, finding verification) take this instead of
|
|
81
78
|
* a backend, so every case runs against its own image. */
|
|
82
79
|
type ReplayExecBackendFactory = (image: string) => ReplayExecBackend;
|
|
83
|
-
//#endregion
|
|
84
|
-
//#region src/trajectory-replay/fix.d.ts
|
|
85
80
|
interface ChatUsage {
|
|
86
81
|
readonly promptTokens: number;
|
|
87
82
|
readonly completionTokens: number;
|
|
@@ -127,8 +122,6 @@ type FixGenerationOutcome = {
|
|
|
127
122
|
readonly succeeded: false;
|
|
128
123
|
readonly error: string;
|
|
129
124
|
};
|
|
130
|
-
//#endregion
|
|
131
|
-
//#region src/trajectory-replay/fix-loop.d.ts
|
|
132
125
|
/** Result of executing one corrected command as a full arm. */
|
|
133
126
|
interface FixArmExecution {
|
|
134
127
|
readonly exitCode: number;
|
|
@@ -187,8 +180,6 @@ interface FixLoopResult {
|
|
|
187
180
|
* bound rather than a measurement. */
|
|
188
181
|
readonly callsWithoutUsage: number;
|
|
189
182
|
}
|
|
190
|
-
//#endregion
|
|
191
|
-
//#region src/trajectory-replay/image-preparer.d.ts
|
|
192
183
|
/**
|
|
193
184
|
* Replay-ready image derivation.
|
|
194
185
|
*
|
|
@@ -223,8 +214,6 @@ interface DockerImagePreparerOptions {
|
|
|
223
214
|
* replays on the base image directly.
|
|
224
215
|
*/
|
|
225
216
|
export declare function dockerImagePreparer(options?: DockerImagePreparerOptions): ImagePreparer;
|
|
226
|
-
//#endregion
|
|
227
|
-
//#region src/trajectory-replay/batch.d.ts
|
|
228
217
|
interface ReplayBatchOptions {
|
|
229
218
|
readonly corpora: readonly CorpusSpec[];
|
|
230
219
|
readonly out: string;
|
|
@@ -400,8 +389,6 @@ interface ReplayBatchReport {
|
|
|
400
389
|
readonly cases: readonly ReplayBatchCaseRow[];
|
|
401
390
|
}
|
|
402
391
|
export declare function runReplayBatch(options: ReplayBatchOptions): Promise<ReplayBatchReport>;
|
|
403
|
-
//#endregion
|
|
404
|
-
//#region src/trajectory-replay/verify.d.ts
|
|
405
392
|
interface IngestedTrajectory {
|
|
406
393
|
runId: string;
|
|
407
394
|
stepCount: number;
|
|
@@ -519,12 +506,8 @@ interface ReplayVerdict extends DefaultVerdict {
|
|
|
519
506
|
};
|
|
520
507
|
}
|
|
521
508
|
export declare function replayVerify(options: ReplayVerifyOptions): Promise<ReplayVerdict>;
|
|
522
|
-
//#endregion
|
|
523
|
-
//#region src/trajectory-replay/batch-fix.d.ts
|
|
524
509
|
/** Arm B and the counterfactual fix phase of a replay batch. */
|
|
525
510
|
export declare function seededSample<T>(items: readonly T[], size: number, seed: number): Set<T>;
|
|
526
|
-
//#endregion
|
|
527
|
-
//#region src/trajectory-replay/findings.d.ts
|
|
528
511
|
interface VerifiableFinding {
|
|
529
512
|
readonly finding_id?: string;
|
|
530
513
|
readonly analyst_id?: string;
|
|
@@ -634,8 +617,6 @@ export declare function classifyVerdict(verdict: ReplayVerdict): FindingVerifica
|
|
|
634
617
|
* run-level verifications.json.
|
|
635
618
|
*/
|
|
636
619
|
export declare function verifyFindings(findings: readonly VerifiableFinding[], options: VerifyFindingsOptions): Promise<VerifyFindingsRun>;
|
|
637
|
-
//#endregion
|
|
638
|
-
//#region src/trajectory-replay/wire.d.ts
|
|
639
620
|
interface IncorrectStepsSubject {
|
|
640
621
|
readonly firstStep: number;
|
|
641
622
|
readonly lastStep: number;
|
|
@@ -676,6 +657,5 @@ interface ReplayFindingResult {
|
|
|
676
657
|
* that needs a derived image must run an `ImagePreparer` first.
|
|
677
658
|
*/
|
|
678
659
|
export declare function replayVerifyFinding(finding: AnalystReplayFinding, options: ReplayFindingOptions): Promise<ReplayFindingResult>;
|
|
679
|
-
//#endregion
|
|
680
660
|
export { type AnalystReplayFinding, type ArmExecutionResult, type CaseResources, type ChatCompletionCaller, type ChatOutcome, type ChatUsage, type CorpusSpec, type CwdSource, type DecodedTrajectory, type DockerImagePreparerOptions, type EnumerationResult, type ExcludedCase, type FailedFixAttempt, type FindingReplaySource, type FindingReplayability, type FindingVerification, type FindingVerificationStatus, type FixArmExecution, type FixArmExecutor, type FixGenerationOutcome, type FixLoopAttemptRecord, type FixLoopOptions, type FixLoopResult, type FixPromptInput, type ImagePreparation, type ImagePreparer, type IncorrectStepsSubject, type IngestedTrajectory, type PrefixDivergence, type PrefixDivergenceKind, type PrefixReplayResult, type RecordedFinalOutcome, type RecordedObservationKind, type RecordedTrajectoryStep, type RecordedTrajectoryTurn, type ReplayArmVerdict, type ReplayBatchCaseRow, type ReplayBatchFixResult, type ReplayBatchOptions, type ReplayBatchReport, type ReplayExclusionReason, type ReplayExecBackend, type ReplayExecBackendFactory, type ReplayExecResult, type ReplayExecSession, type ReplayFindingOptions, type ReplayFindingResult, type ReplayVerdict, type ReplayVerifyOptions, type ReplayableCase, type ResolvedFindingReplay, type ResolvedReplayInvocation, type ResourceResolution, type SandboxCounterfactualRunnerOptions, type VerifiableFinding, type VerifyFindingsOptions, type VerifyFindingsRun, classifyObservation, decodeRecordedTurns, finalRecordedOutcome };
|
|
681
661
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/trajectory-replay/corpus.ts","../../src/trajectory-replay/exec.ts","../../src/trajectory-replay/fix.ts","../../src/trajectory-replay/fix-loop.ts","../../src/trajectory-replay/image-preparer.ts","../../src/trajectory-replay/batch.ts","../../src/trajectory-replay/verify.ts","../../src/trajectory-replay/batch-fix.ts","../../src/trajectory-replay/findings.ts","../../src/trajectory-replay/wire.ts"],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/trajectory-replay/corpus.ts","../../src/trajectory-replay/exec.ts","../../src/trajectory-replay/fix.ts","../../src/trajectory-replay/fix-loop.ts","../../src/trajectory-replay/image-preparer.ts","../../src/trajectory-replay/batch.ts","../../src/trajectory-replay/verify.ts","../../src/trajectory-replay/batch-fix.ts","../../src/trajectory-replay/findings.ts","../../src/trajectory-replay/wire.ts"],"mappings":";;UAyBiB;WACN;WACA;WACA;;KAgBC;UAYK;WACN;WACA;WACA,QAAQ;WACR;;KAGC;;UAGK;WACN;WACA;WACA;WACA,gBAAgB;WAChB;WACA;WACA;WACA,WAAW;;WAEX;;UAGM,uBAAuB;;WAE7B;;;;WAIA;;WAEA;;WAEA;;KA0EC;WACG;WAAyB,WAAW;;WACpC;WAA0B,QAAQ;WAAgC;;UAuEhE;WACN,YAAY;WACZ,UAAU;WACV;;;;;;;;;;;UCrOM;EACf;EACA;EACA;;UAGe;EACf,KAAK,iBAAiB,oBAAoB,QAAQ;EAClD,SAAS;;UAGM;;EAEf,QAAQ,QAAQ;;;;;KAMN,4BAA4B,kBAAkB;UCbzC;WACN;WACA;;KAGC;WACG;WAA0B;IAAS;IAAiB,OAAO;;;WAC3D;WAA2B;;UAEzB;EACf,SAAS,gBAAgB,eAAe,QAAQ;;UAGjC;WACN;WACA,gBAAgB;;WAEhB;;WAEA;;;UA4DM;WACN;;WAEA;WACA;WACA;WACA;WACA;;KAsEC;WACG;WAA0B;IAAS;IAAiB,OAAO;;;WAC3D;WAA2B;;;UClJzB;WACN;WACA;WACA;;WAEA;WACA;WACA;WACA;;;KAIC,kBAAkB,iBAAiB,oBAAoB,QAAQ;UAE1D;;WAEN;;WAEA;;WAEA;WACA,cAAc;;UAGR;WACN;WACA;WACA;WACA;aAAkB;aAA+B;;WACjD;WACA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;UAGM;WACN;WACA;WACA,mBAAmB;;WAEnB;WACA;;;WAGA;WACA;WACA;;;;WAIA;;;;;;;;;;;KCjEC;WAEG;WACA;IAAS;IAAsB;IAAiB;;;WAEhD;WAA2B;;UAEzB;EACf,OAAO,eAAe,cAAc,QAAQ;;UAG7B;WACN;WACA;;;;;;;;wBAcK,oBAAoB,UAAS,6BAAkC;UCD9D;WACN,kBAAkB;WAClB;;;;WAIA;;WAEA;WACA,YAAY;WACZ;;WAEA;WACA;;WAEA;WACA;;WAEA;;WAEA;;WAEA,WAAW;;WAEX,gBAAgB;WAChB,cAAc;;UAGR;;WAEN;WACA;WACA;WACA;;;WAGA,OAAO;WACP;WACA;WACA;WACA;WACA;WACA;;WAEA,mBAAmB;;;WAGnB;;UAGM;WACN;WACA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;WAEA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;WAEA;WACA;;;WAGA;WACA;WACA;WACA;;;WAGA;WACA,KAAK;WACL;;UAGM;WACN;WACA;IAAoB;IAAc;IAAoB;;WACtD;aACE;aACA;aACA;aACA,kBAAkB;;;aAGlB,qBAAqB;MAE1B;MAAyB;;;WAGtB;aACE;MAAqB;MAAmB;MAAqB;;aAC7D;MAAuB;MAAmB;MAAqB;;;;;aAI/D;eACE;eACA;eACA;eACA;;eAEA;eACA;eACA;eACA;;aAEF;MAAe;MAAmB;MAAqB;;;aAEvD;MACP;MACA;MACA;;;;aAIO;MACP;MACA;MACA;;;aAGO,gBAAgB;;WAElB;aACE;aACA;aACA;aACA;aACA;;;;aAIA;;WAEF;IAAqB;IAAgB;IAAgB;IAAgB;;WACrE;IAAyB;IAAgB;IAAgB;IAAe;;WACxE,gBAAgB;;wBAmUL,eAAe,SAAS,qBAAqB,QAAQ;UC/c1D;EACf;EACA;;;KAyCU;;;;;UAMK;EACf;EACA,MAAM;;EAEN;EACA;;UAGe;EACf;;EAEA,mBAAmB;;EAEnB;EACA;EACA;;;EAGA;;EAEA;EACA;;;;qBAKW;UAiDI;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;;;EAGA;EACA,cAAc;;UAwGC;EACf;EACA;;EAEA;;EAEA;EACA;EACA;EACA;;EAEA;EACA;EACA;;EAEA;;EAEA,SAAS;EACT,cAAc;;UAGC;EACf;EACA;EACA;EACA,QAAQ;;;;;;;;UASO,sBAAsB;EACrC;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,mBAAmB;;EAEnB;EACA;EACA;EACA;EACA;EACA,MAAM;IAAqB;;EAC3B,OAAO;IAAqB;;EAC5B;IAAW;IAAgB;IAAuB;;EAClD;IAAU;IAAkB;IAAc;;;wBAUtB,aAAa,SAAS,sBAAsB,QAAQ;;wBCpV1D,aAAa,GAAG,gBAAgB,KAAK,cAAc,eAAe,IAAI;UCwBrE;WACN;WACA;WACA;WACA;WACA;WACA;aACE;aACA;aACA;;WAEF,WAAW,SAAS;;;;;;;;;;;KA+CnB;WAEG;WACA;WACA;WACA;WACA;;WAGA;WACA,kBAAkB;WAClB,WAAW;;UAGT;WACN;WACA;;WAEA;WACA;;WAEA;WACA;WACA;;KAGC;WACG;WAA2B,UAAU;;WACrC;WAA4B;;KAuH/B;UAMK;WACN;WACA;WACA;;WAEA;WACA,UAAU;;WAEV;;WAEA;WACA;;WAEA;;;;;;;;UASM,0BAA0B;WAChC;WACA,wBAAwB;WACxB,QAAQ,SAAS,OAAO;;WAExB;;UAGM;WACN,QAAQ;;WAER;;WAEA;WACA;WACA;;WAEA,gBAAgB;;;WAGhB,kBAAkB;WAClB,cAAc;;;;;;;;wBAST,gBAAgB,SAAS,gBAAgB;;;;;;;wBAyGnC,eACpB,mBAAmB,qBACnB,SAAS,wBACR,QAAQ;UC5YM;WACN;WACA;WACA;WACA;;UAiBM;WACN;WACA;;UAGM;WACN,WAAW;WACX,SAAS;;WAET;;UAmDM;WACN,kBAAkB;WAClB;;WAEA,YAAY;;WAEZ;WACA;WACA;;WAEA,gBAAgB;WAChB,cAAc;;UAGR;WACN,YAAY;WACZ;WACA,SAAS;;;;;;;wBAQE,oBACpB,SAAS,sBACT,SAAS,uBACR,QAAQ"}
|
|
@@ -1,18 +1,18 @@
|
|
|
1
|
-
import { o as NotFoundError, s as ValidationError } from "../errors-
|
|
2
|
-
import { t as mulberry32 } from "../random-
|
|
3
|
-
import { t as certificationEvidenceDigest } from "../verdict-
|
|
4
|
-
import { t as TraceEmitter } from "../emitter-
|
|
5
|
-
import { n as InMemoryTraceStore } from "../store-
|
|
6
|
-
import { t as packageVersion } from "../package-version-
|
|
7
|
-
import { t as buildTrajectory } from "../trajectory-
|
|
8
|
-
import { a as classifyObservation, c as finalRecordedOutcome, d as isSubmitOnlyAction, f as parseObservationOutput, o as decodeRecordedTurns, p as parseRecordedReturncode, s as deriveFailureSignature, t as wrapActionForExec, u as isSubmitAction } from "../exec-
|
|
1
|
+
import { o as NotFoundError, s as ValidationError } from "../errors-DeEEV9Vx.js";
|
|
2
|
+
import { t as mulberry32 } from "../random-DrzcgC7F.js";
|
|
3
|
+
import { t as certificationEvidenceDigest } from "../verdict-C8TcdwfD.js";
|
|
4
|
+
import { t as TraceEmitter } from "../emitter-DhQ65mza.js";
|
|
5
|
+
import { n as InMemoryTraceStore } from "../store-ZykIyQmy.js";
|
|
6
|
+
import { t as packageVersion } from "../package-version-CJl_pY9W.js";
|
|
7
|
+
import { t as buildTrajectory } from "../trajectory-O3GDHmev.js";
|
|
8
|
+
import { a as classifyObservation, c as finalRecordedOutcome, d as isSubmitOnlyAction, f as parseObservationOutput, o as decodeRecordedTurns, p as parseRecordedReturncode, s as deriveFailureSignature, t as wrapActionForExec, u as isSubmitAction } from "../exec-C0_AkLUv.js";
|
|
9
9
|
import { createHash } from "node:crypto";
|
|
10
10
|
import { appendFileSync, existsSync, mkdirSync, mkdtempSync, readFileSync, readdirSync, writeFileSync } from "node:fs";
|
|
11
11
|
import { join } from "node:path";
|
|
12
12
|
import { tmpdir } from "node:os";
|
|
13
13
|
import { execFile } from "node:child_process";
|
|
14
14
|
import { promisify } from "node:util";
|
|
15
|
-
|
|
15
|
+
// src/counterfactual.ts
|
|
16
16
|
/**
|
|
17
17
|
* Counterfactual replay — "what would have happened if we'd changed
|
|
18
18
|
* exactly one thing at turn N?"
|
|
@@ -103,8 +103,7 @@ function applyMutation(step, mutation) {
|
|
|
103
103
|
if (mutation.kind === "custom") return mutation.apply(step);
|
|
104
104
|
return step;
|
|
105
105
|
}
|
|
106
|
-
|
|
107
|
-
//#region src/trajectory-replay/fix.ts
|
|
106
|
+
// src/trajectory-replay/fix.ts
|
|
108
107
|
/**
|
|
109
108
|
* Counterfactual patch synthesis: turn a recorded incorrect step into a
|
|
110
109
|
* corrected shell command for replay arm B.
|
|
@@ -233,8 +232,7 @@ async function generateFixCommand(caller, input) {
|
|
|
233
232
|
}
|
|
234
233
|
};
|
|
235
234
|
}
|
|
236
|
-
|
|
237
|
-
//#region src/trajectory-replay/fix-loop.ts
|
|
235
|
+
// src/trajectory-replay/fix-loop.ts
|
|
238
236
|
/**
|
|
239
237
|
* Iterative counterfactual fix loop.
|
|
240
238
|
*
|
|
@@ -380,8 +378,7 @@ async function runFixLoop(caller, input, executor, options) {
|
|
|
380
378
|
callsWithoutUsage
|
|
381
379
|
};
|
|
382
380
|
}
|
|
383
|
-
|
|
384
|
-
//#region src/trajectory-replay/verify.ts
|
|
381
|
+
// src/trajectory-replay/verify.ts
|
|
385
382
|
/**
|
|
386
383
|
* Execution-replay verification for recorded shell trajectories.
|
|
387
384
|
*
|
|
@@ -772,8 +769,7 @@ function renderReport(verdict, target, armA, armB) {
|
|
|
772
769
|
lines.push("");
|
|
773
770
|
return lines.join("\n");
|
|
774
771
|
}
|
|
775
|
-
|
|
776
|
-
//#region src/trajectory-replay/batch-fix.ts
|
|
772
|
+
// src/trajectory-replay/batch-fix.ts
|
|
777
773
|
/** Arm B and the counterfactual fix phase of a replay batch. */
|
|
778
774
|
function seededSample(items, size, seed) {
|
|
779
775
|
const pool = [...items];
|
|
@@ -1020,8 +1016,7 @@ async function runFixPhase(run, selected, rows, verdicts) {
|
|
|
1020
1016
|
...tally
|
|
1021
1017
|
};
|
|
1022
1018
|
}
|
|
1023
|
-
|
|
1024
|
-
//#region src/trajectory-replay/batch-report.ts
|
|
1019
|
+
// src/trajectory-replay/batch-report.ts
|
|
1025
1020
|
function pct(value) {
|
|
1026
1021
|
return value === null ? "—" : `${(value * 100).toFixed(1)}%`;
|
|
1027
1022
|
}
|
|
@@ -1130,8 +1125,7 @@ function renderBatchReport(report) {
|
|
|
1130
1125
|
lines.push("");
|
|
1131
1126
|
return lines.join("\n");
|
|
1132
1127
|
}
|
|
1133
|
-
|
|
1134
|
-
//#region src/trajectory-replay/corpus.ts
|
|
1128
|
+
// src/trajectory-replay/corpus.ts
|
|
1135
1129
|
/**
|
|
1136
1130
|
* Corpus enumeration for batch replay verification.
|
|
1137
1131
|
*
|
|
@@ -1353,8 +1347,7 @@ function enumerateReplayableCases(corpora) {
|
|
|
1353
1347
|
labelEntryCount
|
|
1354
1348
|
};
|
|
1355
1349
|
}
|
|
1356
|
-
|
|
1357
|
-
//#region src/trajectory-replay/image-preparer.ts
|
|
1350
|
+
// src/trajectory-replay/image-preparer.ts
|
|
1358
1351
|
/**
|
|
1359
1352
|
* Replay-ready image derivation.
|
|
1360
1353
|
*
|
|
@@ -1473,8 +1466,7 @@ function dockerImagePreparer(options = {}) {
|
|
|
1473
1466
|
};
|
|
1474
1467
|
} };
|
|
1475
1468
|
}
|
|
1476
|
-
|
|
1477
|
-
//#region src/trajectory-replay/batch.ts
|
|
1469
|
+
// src/trajectory-replay/batch.ts
|
|
1478
1470
|
/**
|
|
1479
1471
|
* Batch replay verification across gold-labeled trajectory corpora.
|
|
1480
1472
|
*
|
|
@@ -1755,8 +1747,7 @@ async function runReplayBatch(options) {
|
|
|
1755
1747
|
writeFileSync(join(options.out, "batch-report.md"), renderBatchReport(report));
|
|
1756
1748
|
return report;
|
|
1757
1749
|
}
|
|
1758
|
-
|
|
1759
|
-
//#region src/trajectory-replay/wire.ts
|
|
1750
|
+
// src/trajectory-replay/wire.ts
|
|
1760
1751
|
/**
|
|
1761
1752
|
* Finding-to-replay wire: the entry point that turns a cited incorrect-steps
|
|
1762
1753
|
* finding into an executed replay proof.
|
|
@@ -1847,8 +1838,7 @@ async function replayVerifyFinding(finding, options) {
|
|
|
1847
1838
|
verdict
|
|
1848
1839
|
};
|
|
1849
1840
|
}
|
|
1850
|
-
|
|
1851
|
-
//#region src/trajectory-replay/findings.ts
|
|
1841
|
+
// src/trajectory-replay/findings.ts
|
|
1852
1842
|
/**
|
|
1853
1843
|
* Proof-carrying findings: execute analyst findings as replays.
|
|
1854
1844
|
*
|
|
@@ -2226,7 +2216,6 @@ async function verifyFindings(findings, options) {
|
|
|
2226
2216
|
}, null, 2)}\n`);
|
|
2227
2217
|
return run;
|
|
2228
2218
|
}
|
|
2229
|
-
//#endregion
|
|
2230
2219
|
export { PREFIX_DIVERGENCE_TOLERANCE_PCT, classifyObservation, classifyVerdict, decodeRecordedTurns, dockerImagePreparer, finalRecordedOutcome, replayVerify, replayVerifyFinding, runReplayBatch, seededSample, verifyFindings };
|
|
2231
2220
|
|
|
2232
2221
|
//# sourceMappingURL=index.js.map
|