@tangle-network/agent-eval 0.221.7 → 0.222.0
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 +146 -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-CoD1Dotf.d.ts} +4 -10
- package/dist/agent-profile-CoD1Dotf.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-C-pP1sK9.d.ts} +7 -13
- package/dist/analyze-runs-C-pP1sK9.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-jYWR2gya.js} +61 -94
- package/dist/benchmark-command-jYWR2gya.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-B1DstPYg.d.ts} +53 -61
- package/dist/default-registry-B1DstPYg.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-CAffiFbN.d.ts} +19 -27
- package/dist/define-agent-eval-CAffiFbN.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-CaHaCc3W.d.ts} +10 -18
- package/dist/engine-CaHaCc3W.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-BPqUvZ66.d.ts} +4 -6
- package/dist/evaluation-BPqUvZ66.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-D98BKpxo.d.ts} +8 -10
- package/dist/exact-types-D98BKpxo.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-SBGncrCP.d.ts} +7 -13
- package/dist/feedback-trajectory-SBGncrCP.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-CdEVr8-O.d.ts → index-B7v2QHlD.d.ts} +24 -299
- package/dist/index-B7v2QHlD.d.ts.map +1 -0
- package/dist/{index-CTIRz-AB.d.ts → index-CUjeTDPv.d.ts} +14 -36
- package/dist/index-CUjeTDPv.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-D7fmC-1K.d.ts → index-DUn0GNiZ.d.ts} +183 -36
- package/dist/index-DUn0GNiZ.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-jnuFLxOs.d.ts → index-aDPVYBIM.d.ts} +11 -39
- package/dist/index-aDPVYBIM.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-CBUzIZs4.d.ts} +4 -12
- package/dist/matrix-CBUzIZs4.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-CdN909k1.js → otel-spans-BsGzTWIK.js} +5 -6
- package/dist/{otel-spans-CdN909k1.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-COFkxas6.d.ts} +4 -8
- package/dist/prime-protocol-COFkxas6.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-dhdDoAV8.d.ts} +5 -11
- package/dist/registry-dhdDoAV8.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-CBdwFZ05.d.ts} +10 -36
- package/dist/search-kernel-CBdwFZ05.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-Cei036nE.d.ts} +9 -37
- package/dist/statistical-heldout-Cei036nE.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-DxsFwtdr.d.ts → store-tool-spans-B3i3fl0A.d.ts} +11 -29
- package/dist/store-tool-spans-B3i3fl0A.d.ts.map +1 -0
- 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/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-Cw2txK5m.d.ts} +12 -24
- package/dist/systemone-Cw2txK5m.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-BLgxt-4Q.d.ts} +8 -10
- package/dist/systemone-review-BLgxt-4Q.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-BMsblaDh.js → task-failure-attributes-CC6VuIpk.js} +12 -11
- package/dist/task-failure-attributes-CC6VuIpk.js.map +1 -0
- package/dist/{tool-groups-b3oU_D88.d.ts → tool-groups-C3K-d3gd.d.ts} +3 -5
- package/dist/tool-groups-C3K-d3gd.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-ej5AEJ5X.d.ts} +4 -6
- package/dist/trace-questions-ej5AEJ5X.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-DZ0mYVw9.d.ts → types-0oMg7zee.d.ts} +3 -5
- package/dist/types-0oMg7zee.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-M6eF9m-r.d.ts → types-C54Uk_a1.d.ts} +84 -24
- package/dist/types-C54Uk_a1.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-6Lfdlshp.d.ts → types-DzCO-S1m.d.ts} +50 -62
- package/dist/types-DzCO-S1m.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/concepts.md +1 -0
- 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/task-failure-attributes-BMsblaDh.js.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
package/dist/campaign/index.d.ts
CHANGED
|
@@ -1,10 +1,10 @@
|
|
|
1
|
-
import { o as CostLedgerHandle, r as CostBreakdown, v as PendingCostCallView } from "../cost-ledger-
|
|
2
|
-
import { S as makeProposalFinding, v as ProposalFinding, y as ProposalFindingOrigin } from "../types-
|
|
3
|
-
import {
|
|
4
|
-
import { $ as paretoPolicy, G as AxisVerdict, H as PowerPreflightOptions, J as ObjectiveSource, K as BuildEvidenceVectorOptions, Q as buildEvidenceVector, U as powerPreflight, V as PowerPreflight, W as AxisEvidence, X as PromotionObjective, Y as ParetoSignificanceGateOptions, Z as PromotionPolicy, a as dimensionRegressions, c as SequentialDecideFn, d as SequentialObservation, et as paretoSignificanceGate, f as SequentialPairedGate, g as sequentialPairedGate, h as sequentialDecide, i as PairedHoldout, l as SequentialDecideOptions, m as SequentialStreamState, n as HeldoutSignificance, o as heldoutSignificance, p as SequentialPairedGateOptions, q as EvidenceVector, r as HeldoutSignificanceOptions, s as pairHoldout, t as DimensionRegression, u as SequentialGateDecision } from "../statistical-heldout-
|
|
5
|
-
import { $ as SearchTokenAccounting, A as SearchLedgerHash, B as SearchOperationStartedEvent, C as SearchEdgeOperator, D as SearchFailureReason, E as SearchExecutionIdentity, F as SearchNodeRegisteredEvent, G as SearchSplit, H as SearchProposerKind, I as SearchNodeStatus, J as SearchSurfaceEffect, K as SearchSplitTasks, L as SearchOpenedEvent, M as SearchNodeDecidedEvent, N as SearchNodeDecision, O as SearchLedgerEntry, P as SearchNodeRef, Q as SearchTaskOutcome, R as SearchOperationKind, S as SearchEdgeAttribution, T as SearchEstimateMethod, U as SearchReservation, V as SearchProposer, W as SearchSourceRef, X as SearchSurfaceKind, Y as SearchSurfaceEvidence, Z as SearchTask, _ as SearchClaimPower, a as SearchAttemptAccounting, b as SearchClosedEvent, c as SearchCancelReason, d as SearchCellCancelledEvent, et as SearchTraceRef, f as SearchCellSettledEvent, g as SearchClaimFinalist, h as SearchClaimEstimator, i as SearchArtifactRef, j as SearchModelIdentity, k as SearchLedgerEvent, l as SearchCandidateSurface, m as SearchClaim, n as SEARCH_LEDGER_SCHEMA, o as SearchAudit, p as SearchCellStage, q as SearchSplits, r as SearchArtifactKind, s as SearchBudget, t as NodeEstimate, tt as SearchUnknown, u as SearchCellAllocatedEvent, v as SearchClaimTest, w as SearchEdgeRecordedEvent, x as SearchCostAccounting, y as SearchCloseReason, z as SearchOperationRecordedEvent } from "../search-ledger-types-
|
|
6
|
-
import { $
|
|
7
|
-
import {
|
|
8
|
-
import { A as SearchRungDecision, At as SearchHistoryReceipt, B as beam, Bt as openSearchLedger, C as searchClaimDecision, Ct as surfaceNode, D as SearchAllocationView, Dt as SearchHistoryCoverage, E as AshaOptions, Et as SearchHistoryAuditSummary, F as SearchExpansion, Ft as FileSearchLedger, G as NodePosterior, Gt as createRunCostLedger, H as incumbent, Ht as replaySearchLedgerText, I as SearchPolicy, It as OpenSearchLedgerOptions, J as estimateNode, K as SEARCH_ESTIMATOR, Kt as fsCampaignStorage, L as SearchPolicyNode, Lt as SearchLedger, M as asha, Mt as createSearchHistoryReceipt, N as uniform, Nt as verifySearchHistoryArtifact, O as SearchAllocator, Ot as SearchHistoryCoverageRow, P as AideOptions, Pt as verifySearchHistoryReceipt, Q as searchPosterior, R as SearchPolicyView, Rt as SearchLedgerAppendResult, S as planSearchClaim, St as surfaceDiff, T as verifySearchClaim, Tt as SearchHistoryAdmissionOptions, U as incumbentWithOperatorBandit, Ut as validateSearchLedgerEvent, V as crowdedFrontierParent, Vt as parseSearchLedgerLine, W as EstimateNodeCellsInput, Wt as CampaignStorage, X as searchCellSetDigest, Y as estimateNodeFromCells, Z as searchEstimateMethod, _ as SEARCH_CLAIM_RULE, _t as SearchRecorderOptions, a as SearchArtifactCodec, b as SearchClaimVerification, bt as developmentClaim, c as SearchExecutor, d as SearchProposalRequest, dt as AllocateSearchCellInput, f as SearchProposalResult, ft as RecordSearchEdgeInput, g as PlanSearchClaimInput, gt as SearchRecorder, h as SearchRunResult, ht as SearchOpening, i as RunSearchOptions, j as SearchUnitExtension, jt as assertSearchHistoryMatchesState, k as SearchCellPlan, kt as SearchHistoryPolicy, l as SearchLane, m as SearchProposerPort, mt as SearchLedgerBinding, n as searchDivergence, o as SearchCellResult, p as SearchProposedChild, pt as RegisterSearchNodeInput, q as SearchPosterior, qt as inMemoryCampaignStorage, r as searchPolicyView, s as SearchCellWork, t as runSearch, u as SearchProposalBlob, v as SearchClaimDecision, vt as SearchRunIdentity, w as searchClaimReserveUsd, wt as CreateSearchHistoryReceiptInput, x as decideSearchClaim, xt as searchEdgeId, y as SearchClaimPlan, yt as SettleSearchCellInput, z as aide, zt as SearchLedgerTrustedHeadMode } from "../search-kernel-
|
|
9
|
-
import { B as ExternalOptimizerModelCallRequest, F as ExternalOptimizerEndpointFormat, G as ExternalOptimizerRunnerCommand, H as ExternalOptimizerModelExecutionObservation, I as ExternalOptimizerEvaluationObservation, L as ExternalOptimizerEvaluationRefusalReason, M as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, N as ExternalOptimizerCallbackLimits, P as ExternalOptimizerChatRequest, R as ExternalOptimizerModelBudget, U as ExternalOptimizerProcessLimits, V as ExternalOptimizerModelCallResult, j as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, q as ExternalTextCandidate, z as ExternalOptimizerModelCall } from "../agent-profile-
|
|
1
|
+
import { o as CostLedgerHandle, r as CostBreakdown, v as PendingCostCallView } from "../cost-ledger-Czro9wmo.js";
|
|
2
|
+
import { S as makeProposalFinding, v as ProposalFinding, y as ProposalFindingOrigin } from "../types-DzCO-S1m.js";
|
|
3
|
+
import { L as SeriesDistribution } from "../schema-DkFJr3Y8.js";
|
|
4
|
+
import { $ as paretoPolicy, G as AxisVerdict, H as PowerPreflightOptions, J as ObjectiveSource, K as BuildEvidenceVectorOptions, Q as buildEvidenceVector, U as powerPreflight, V as PowerPreflight, W as AxisEvidence, X as PromotionObjective, Y as ParetoSignificanceGateOptions, Z as PromotionPolicy, a as dimensionRegressions, c as SequentialDecideFn, d as SequentialObservation, et as paretoSignificanceGate, f as SequentialPairedGate, g as sequentialPairedGate, h as sequentialDecide, i as PairedHoldout, l as SequentialDecideOptions, m as SequentialStreamState, n as HeldoutSignificance, o as heldoutSignificance, p as SequentialPairedGateOptions, q as EvidenceVector, r as HeldoutSignificanceOptions, s as pairHoldout, t as DimensionRegression, u as SequentialGateDecision } from "../statistical-heldout-Cei036nE.js";
|
|
5
|
+
import { $ as SearchTokenAccounting, A as SearchLedgerHash, B as SearchOperationStartedEvent, C as SearchEdgeOperator, D as SearchFailureReason, E as SearchExecutionIdentity, F as SearchNodeRegisteredEvent, G as SearchSplit, H as SearchProposerKind, I as SearchNodeStatus, J as SearchSurfaceEffect, K as SearchSplitTasks, L as SearchOpenedEvent, M as SearchNodeDecidedEvent, N as SearchNodeDecision, O as SearchLedgerEntry, P as SearchNodeRef, Q as SearchTaskOutcome, R as SearchOperationKind, S as SearchEdgeAttribution, T as SearchEstimateMethod, U as SearchReservation, V as SearchProposer, W as SearchSourceRef, X as SearchSurfaceKind, Y as SearchSurfaceEvidence, Z as SearchTask, _ as SearchClaimPower, a as SearchAttemptAccounting, b as SearchClosedEvent, c as SearchCancelReason, d as SearchCellCancelledEvent, et as SearchTraceRef, f as SearchCellSettledEvent, g as SearchClaimFinalist, h as SearchClaimEstimator, i as SearchArtifactRef, j as SearchModelIdentity, k as SearchLedgerEvent, l as SearchCandidateSurface, m as SearchClaim, n as SEARCH_LEDGER_SCHEMA, o as SearchAudit, p as SearchCellStage, q as SearchSplits, r as SearchArtifactKind, s as SearchBudget, t as NodeEstimate, tt as SearchUnknown, u as SearchCellAllocatedEvent, v as SearchClaimTest, w as SearchEdgeRecordedEvent, x as SearchCostAccounting, y as SearchCloseReason, z as SearchOperationRecordedEvent } from "../search-ledger-types-qiYRUo_N.js";
|
|
6
|
+
import { $r as ExternalOptimizerObservationArtifact, $t as CrossSurfaceCandidate, A as RetainedDispatchOptions, Ai as CampaignRunPlanCell, An as BuildLoopProvenanceArgs, Ar as externalTextOptimizationMethod, At as importExternalEvaluations, B as RunProfileMatrixSegmentOptions, Bi as CampaignCoverage, Bn as emitLoopProvenance, Br as OptimizationMethodProvenance, Bt as LoadEvalFixtureScenariosOptions, C as searchModelIdentity, Ci as RunEvalOptions, Cn as CrossSurfaceSelectionPolicy, Cr as GepaEngineRun, Ct as RolloutCall, D as ScenarioSignal, Di as RunCampaignOptions, Dn as BuildTraceAnalystSurfaceDispatchOptions, Dr as gepaOptimizationMethod, Dt as GepaEvaluationImport, E as DiscriminationScore, Ei as CampaignCellRetryPolicy, En as measureCampaignCost, Er as GepaRunnerCommand, Et as ExternalSearchOptimizer, F as FinalizedProfileMatrixResult, Fi as CellScheduleSlot, Fn as LoopProvenanceCandidate, Fr as ExternalTextEvaluationResponse, Ft as EvalFixtureFile, G as ProfileSummary, Gi as campaignSplitDigestFromIdentities, Gn as HeldOutGateOptions, Gr as OptimizationTokenUsage, Gt as planEvalFixtureRun, H as finalizeProfileMatrix, Hi as campaignCoverage, Hr as OptimizationMethodRunOptions, Ht as discoverEvalFixtures, I as ProfileMatrixCoverage, Ii as buildCellSchedule, In as LoopProvenanceEvidence, Ir as decodeExternalTextCandidate, It as EvalFixtureLoadOptions, J as ScenarioRollup, Jn as DefaultProductionGateOptions, Jr as GepaCandidatePopulationCandidate, Jt as CrossSurfaceAdditionDecision, K as RunProfileMatrixOptions, Ki as SearchLedgerError, Kn as heldOutGate, Kr as costFromLedgerSummary, Kt as analyzeCrossSurfaceInteractions, L as ProfileMatrixPlan, Li as cellCachePath, Ln as LoopProvenanceOptimizationMethod, Lr as OptimizationMethod, Lt as EvalFixtureRunPlan, M as createRetainedDispatch, Mi as planCampaignRun, Mn as EmitLoopProvenanceResult, Mr as ExternalTextOptimizerContext, Mt as NeutralizationGateOptions, N as CreateProfileMatrixPlanOptions, Ni as CacheIssueReason, Nn as LoopProvenanceArgsFromResult, Nr as ExternalTextOptimizerResult, Nt as neutralizationGate, O as scoreDiscrimination, Oi as runCampaign, On as TraceAnalystArtifact, Or as OpenAICompatibleOptimizerModel, Ot as GepaPopulationImport, P as FinalizeProfileMatrixOptions, Pi as CacheRead, Pn as LoopProvenanceBackend, Pr as ExternalOptimizationExample, Pt as EvalFixture, Qr as ExternalOptimizerExecutionSummary, Qt as CrossSurfaceBootstrapPolicy, R as ProfileMatrixRow, Ri as FinalEvidencePolicy, Rn as LoopProvenanceRecord, Rr as OptimizationMethodComposition, Rt as EvalFixtureScenario, S as campaignCellSearchResult, Sn as CrossSurfaceRelativeCost, Sr as GepaEngineOptions, St as RolloutArgumentDiffOptions, T as searchReceiptAccounting, Ti as CampaignCellFailureReceipt, Tn as CrossSurfaceTaskRow, Tr as GepaOptimizationRecipe, Tt as UngroundedLiteralReport, U as runProfileMatrixSegment, Ui as campaignScenarioIdentity, Un as campaignBreakdown, Ur as OptimizationPackageSource, Ut as loadEvalFixture, V as createProfileMatrixPlan, Vi as assertCampaignDesign, Vn as CampaignBreakdown, Vr as OptimizationMethodResult, Vt as PlanEvalFixtureRunOptions, W as ProfileDispatchFn, Wi as campaignSplitDigest, Wn as campaignMeanComposite, Wr as ComparisonCost, Wt as loadEvalFixtureScenarios, Xn as defaultProductionGate, Xr as GepaCandidateSelectionScore, Xt as CrossSurfaceAttemptCompleteness, Y as runProfileMatrix, Yn as DefaultProductionRewardHackingOptions, Yr as GepaCandidatePopulationSummary, Yt as CrossSurfaceAdditionRejectionReason, Zr as readGepaCandidatePopulationArtifact, Zt as CrossSurfaceBestSingleSelection, _ as surfaceDispatchRef, _i as RunOptimizationOptions, _n as CrossSurfacePairCompatibility, _r as OptimizationMethodComparison, _t as scopedOptimizationMethod, a as gitWorktreeAdapter, an as CrossSurfaceComponentEvidence, at as ScoreboardSummary, b as SingleRunLockOptions, bi as OpenAutoPrOptions, bn as CrossSurfacePairwiseEntry, br as compareOptimizationMethods, bt as FsLabeledScenarioStoreOptions, c as AutoevalsScoreLike, ci as LlmJudgeOptions, cn as CrossSurfaceEligibility, ct as renderScoreboardMarkdown, d as PhoenixEvaluatorLike, di as isTransientTransportFailure, dn as CrossSurfaceInteractionAwareSelection, dr as composeGate, dt as userStoryScoreboard, ei as ExternalOptimizerObservationSummary, en as CrossSurfaceCandidateComparison, et as PlaybackContext, f as assertCodeSurfaceIdentity, fi as transientDispatchFailure, fn as CrossSurfaceInteractionEffect, fr as SkillOptOptimizationMethodConfig, g as surfaceContentHash, gi as PremeasuredOptimizationBaseline, gn as CrossSurfaceNaiveStackSelection, gr as CompareOptimizationMethodsOptions, h as renderSurfaceDiff, hi as runImprovementLoop, hn as CrossSurfaceInteractionTask, hr as skillOptOptimizationMethod, i as WorktreeAdapter, ii as ReferenceEquivalenceJudgeOptions, in as CrossSurfaceComponent, it as ScoreboardRow, j as RetainedDispatchOutcome, ji as PlanCampaignRunOptions, jn as EmitLoopProvenanceArgs, jr as ExternalTextOptimizationMethodConfig, jt as importGepaPopulation, k as resolveRunDir, ki as CampaignRunPlan, kn as TraceAnalystScenario, kr as OptimizerModelBudget, kt as externalSurface, l as AutoevalsScorerLike, li as llmJudge, ln as CrossSurfaceEvidenceBreakdown, lt as scoreUserStory, m as componentSurfaceIdentityMaterial, mi as RunImprovementLoopResult, mn as CrossSurfaceInteractionReport, mr as SkillOptTrainerConfig, n as GitWorktreeAdapterOptions, ni as readExternalOptimizerObservationArtifact, nn as CrossSurfaceCandidateOutcome, nt as PlaybackStep, o as resolveWorktreePath, oi as ReferenceEquivalenceScenario, on as CrossSurfaceCompositionStep, ot as UserStory, p as codeSurfaceIdentityMaterial, pi as RunImprovementLoopOptions, pn as CrossSurfaceInteractionPath, pr as SkillOptRunnerCommand, q as RunProfileMatrixResult, qi as SearchLedgerIntegrityError, qn as DefaultProductionGateCheck, qr as GepaCandidatePopulationArtifact, qt as AnalyzeCrossSurfaceInteractionsInput, r as Worktree, rn as CrossSurfaceCandidateSummary, rt as ScoreboardRenderOptions, s as verifyCodeSurface, si as LlmJudgeDimension, sn as CrossSurfaceDistribution, st as UserStoryVerdict, t as CodeSurfaceVerification, ti as ExternalOptimizerSubmittedCandidate, tn as CrossSurfaceCandidateEvidence, tt as PlaybackDriver, u as PhoenixEvaluationResultLike, ui as TransientFailureOptions, un as CrossSurfaceIneligibilityReason, ut as scoreboardSummary, v as surfaceHash, vi as RunOptimizationResult, vn as CrossSurfacePairEvidence, vr as OptimizationMethodPairwise, vt as sequentialOptimizationMethod, w as searchProposalExecution, wi as runEval, wn as CrossSurfaceSelections, wr as GepaOptimizationMethodConfig, wt as ScoredRollout, x as acquireSingleRunLock, xi as OpenAutoPrResult, xn as CrossSurfaceRankedSingle, xr as GepaAdaptiveEngineRun, xt as RolloutArgumentDiff, y as SingleRunLock, yi as runOptimization, yn as CrossSurfacePairIncompatibilityReason, yr as OptimizationMethodScore, yt as neutralizeText, z as ProfileMatrixSegmentResult, zi as FinalEvidenceUse, zn as canonicalDigest, zr as OptimizationMethodInput, zt as EvalFixtureValidationMode } from "../index-B7v2QHlD.js";
|
|
7
|
+
import { A as LabeledScenarioWrite, B as ScoredSurfaceOutcome, C as JudgeDimension, D as LabeledScenarioSampleArgs, E as LabeledScenarioRecord, F as ProposeContext, G as SearchProposerView, H as SurfaceProposer, I as ProposedCandidate, J as searchProposerView, K as SearchSummaryOptions, L as RedactionStatus, M as OptimizerConfig, N as ParetoParent, O as LabeledScenarioSource, P as ProposalTrackContext, R as Scenario, S as JudgeConfig, T as LabelTrust, U as TraceSpan, V as SessionScript, W as isProposedCandidate, X as SearchStateView, Y as SearchState, Z as searchUnitScores, _ as GateDecision, a as CampaignResult, at as SearchCompletion, b as GenerationRecord, c as CampaignTraceWriter, ct as SearchOperation, d as DispatchContext, dt as SearchUnitScore, f as DispatchFn, ft as searchCellId, g as GateContribution, h as GateContext, i as CampaignCostMeter, it as SearchCell, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, lt as SearchScoredCell, m as GateCheckStatus, mt as searchTaskSetDigest, n as CampaignArtifactWriter, o as CampaignScenarioIdentity, ot as SearchDecisionRecord, p as Gate, pt as searchNodeId, q as renderSearchSummary, r as CampaignCellResult, s as CampaignTokenUsage, st as SearchNode, t as CampaignAggregates, u as ComponentSurface, ut as SearchSpend, v as GateResult, w as JudgeScore, x as JudgeAggregate, y as GenerationCandidate, z as ScenarioAggregate } from "../types-C54Uk_a1.js";
|
|
8
|
+
import { A as SearchRungDecision, At as SearchHistoryReceipt, B as beam, Bt as openSearchLedger, C as searchClaimDecision, Ct as surfaceNode, D as SearchAllocationView, Dt as SearchHistoryCoverage, E as AshaOptions, Et as SearchHistoryAuditSummary, F as SearchExpansion, Ft as FileSearchLedger, G as NodePosterior, Gt as createRunCostLedger, H as incumbent, Ht as replaySearchLedgerText, I as SearchPolicy, It as OpenSearchLedgerOptions, J as estimateNode, K as SEARCH_ESTIMATOR, Kt as fsCampaignStorage, L as SearchPolicyNode, Lt as SearchLedger, M as asha, Mt as createSearchHistoryReceipt, N as uniform, Nt as verifySearchHistoryArtifact, O as SearchAllocator, Ot as SearchHistoryCoverageRow, P as AideOptions, Pt as verifySearchHistoryReceipt, Q as searchPosterior, R as SearchPolicyView, Rt as SearchLedgerAppendResult, S as planSearchClaim, St as surfaceDiff, T as verifySearchClaim, Tt as SearchHistoryAdmissionOptions, U as incumbentWithOperatorBandit, Ut as validateSearchLedgerEvent, V as crowdedFrontierParent, Vt as parseSearchLedgerLine, W as EstimateNodeCellsInput, Wt as CampaignStorage, X as searchCellSetDigest, Y as estimateNodeFromCells, Z as searchEstimateMethod, _ as SEARCH_CLAIM_RULE, _t as SearchRecorderOptions, a as SearchArtifactCodec, b as SearchClaimVerification, bt as developmentClaim, c as SearchExecutor, d as SearchProposalRequest, dt as AllocateSearchCellInput, f as SearchProposalResult, ft as RecordSearchEdgeInput, g as PlanSearchClaimInput, gt as SearchRecorder, h as SearchRunResult, ht as SearchOpening, i as RunSearchOptions, j as SearchUnitExtension, jt as assertSearchHistoryMatchesState, k as SearchCellPlan, kt as SearchHistoryPolicy, l as SearchLane, m as SearchProposerPort, mt as SearchLedgerBinding, n as searchDivergence, o as SearchCellResult, p as SearchProposedChild, pt as RegisterSearchNodeInput, q as SearchPosterior, qt as inMemoryCampaignStorage, r as searchPolicyView, s as SearchCellWork, t as runSearch, u as SearchProposalBlob, v as SearchClaimDecision, vt as SearchRunIdentity, w as searchClaimReserveUsd, wt as CreateSearchHistoryReceiptInput, x as decideSearchClaim, xt as searchEdgeId, y as SearchClaimPlan, yt as SettleSearchCellInput, z as aide, zt as SearchLedgerTrustedHeadMode } from "../search-kernel-CBdwFZ05.js";
|
|
9
|
+
import { B as ExternalOptimizerModelCallRequest, F as ExternalOptimizerEndpointFormat, G as ExternalOptimizerRunnerCommand, H as ExternalOptimizerModelExecutionObservation, I as ExternalOptimizerEvaluationObservation, L as ExternalOptimizerEvaluationRefusalReason, M as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, N as ExternalOptimizerCallbackLimits, P as ExternalOptimizerChatRequest, R as ExternalOptimizerModelBudget, U as ExternalOptimizerProcessLimits, V as ExternalOptimizerModelCallResult, j as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, q as ExternalTextCandidate, z as ExternalOptimizerModelCall } from "../agent-profile-CoD1Dotf.js";
|
|
10
10
|
export { type AideOptions, type AllocateSearchCellInput, type AnalyzeCrossSurfaceInteractionsInput, type AshaOptions, type AutoevalsScoreLike, type AutoevalsScorerLike, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type BuildLoopProvenanceArgs, type BuildTraceAnalystSurfaceDispatchOptions, type CacheIssueReason, type CacheRead, type CampaignAggregates, type CampaignArtifactWriter, type CampaignBreakdown, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCellRetryPolicy, type CampaignCostMeter, type CampaignCoverage, type CampaignResult, type CampaignRunPlan, type CampaignRunPlanCell, type CampaignScenarioIdentity, type CampaignStorage, type CampaignTokenUsage, type CampaignTraceWriter, type CellScheduleSlot, type CodeSurface, type CodeSurfaceVerification, type CompareOptimizationMethodsOptions, type ComparisonCost, type ComponentSurface, type CostBreakdown, type CostLedgerHandle, type CreateProfileMatrixPlanOptions, type CreateSearchHistoryReceiptInput, type CrossSurfaceAdditionDecision, type CrossSurfaceAdditionRejectionReason, type CrossSurfaceAttemptCompleteness, type CrossSurfaceBestSingleSelection, type CrossSurfaceBootstrapPolicy, type CrossSurfaceCandidate, type CrossSurfaceCandidateComparison, type CrossSurfaceCandidateEvidence, type CrossSurfaceCandidateOutcome, type CrossSurfaceCandidateSummary, type CrossSurfaceComponent, type CrossSurfaceComponentEvidence, type CrossSurfaceCompositionStep, type CrossSurfaceDistribution, type CrossSurfaceEligibility, type CrossSurfaceEvidenceBreakdown, type CrossSurfaceIneligibilityReason, type CrossSurfaceInteractionAwareSelection, type CrossSurfaceInteractionEffect, type CrossSurfaceInteractionPath, type CrossSurfaceInteractionReport, type CrossSurfaceInteractionTask, type CrossSurfaceNaiveStackSelection, type CrossSurfacePairCompatibility, type CrossSurfacePairEvidence, type CrossSurfacePairIncompatibilityReason, type CrossSurfacePairwiseEntry, type CrossSurfaceRankedSingle, type CrossSurfaceRelativeCost, type CrossSurfaceSelectionPolicy, type CrossSurfaceSelections, type CrossSurfaceTaskRow, DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DimensionRegression, type DiscriminationScore, type DispatchContext, type DispatchFn, type EmitLoopProvenanceArgs, type EmitLoopProvenanceResult, type EstimateNodeCellsInput, type EvalFixture, type EvalFixtureFile, type EvalFixtureLoadOptions, type EvalFixtureRunPlan, type EvalFixtureScenario, type EvalFixtureValidationMode, type EvidenceVector, type ExternalOptimizationExample, type ExternalOptimizerCallbackLimits, type ExternalOptimizerChatRequest, type ExternalOptimizerEndpointFormat, type ExternalOptimizerEvaluationObservation, type ExternalOptimizerEvaluationRefusalReason, type ExternalOptimizerExecutionSummary, type ExternalOptimizerModelBudget, type ExternalOptimizerModelCall, type ExternalOptimizerModelCallRequest, type ExternalOptimizerModelCallResult, type ExternalOptimizerModelExecutionObservation, type ExternalOptimizerObservationArtifact, type ExternalOptimizerObservationSummary, type ExternalOptimizerProcessLimits, type ExternalOptimizerRunnerCommand, type ExternalOptimizerSubmittedCandidate, type ExternalSearchOptimizer, type ExternalTextCandidate, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, FileSearchLedger, type FinalEvidencePolicy, type FinalEvidenceUse, type FinalizeProfileMatrixOptions, type FinalizedProfileMatrixResult, type FsLabeledScenarioStoreOptions, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaCandidatePopulationArtifact, type GepaCandidatePopulationCandidate, type GepaCandidatePopulationSummary, type GepaCandidateSelectionScore, type GepaEngineOptions, type GepaEngineRun, type GepaEvaluationImport, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaPopulationImport, type GepaRunnerCommand, type GitWorktreeAdapterOptions, type HeldOutGateOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type JudgeAggregate, type JudgeConfig, type JudgeDimension, type JudgeScore, type LabelTrust, type LabeledScenarioRecord, type LabeledScenarioSampleArgs, type LabeledScenarioSource, type LabeledScenarioStore, type LabeledScenarioWrite, type LlmJudgeDimension, type LlmJudgeOptions, type LoadEvalFixtureScenariosOptions, type LoopProvenanceArgsFromResult, type LoopProvenanceBackend, type LoopProvenanceCandidate, type LoopProvenanceEvidence, type LoopProvenanceOptimizationMethod, type LoopProvenanceRecord, type MutableSurface, type NeutralizationGateOptions, type NodeEstimate, type NodePosterior, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OpenAutoPrOptions, type OpenAutoPrResult, type OpenSearchLedgerOptions, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodComposition, type OptimizationMethodInput, type OptimizationMethodPairwise, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationMethodRunOptions, type OptimizationMethodScore, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type PairedHoldout, type ParetoParent, type ParetoSignificanceGateOptions, type PendingCostCallView, type PhoenixEvaluationResultLike, type PhoenixEvaluatorLike, type PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlanSearchClaimInput, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PowerPreflight, type PowerPreflightOptions, type PremeasuredOptimizationBaseline, type ProfileDispatchFn, type ProfileMatrixCoverage, type ProfileMatrixPlan, type ProfileMatrixRow, type ProfileMatrixSegmentResult, type ProfileSummary, type PromotionObjective, type PromotionPolicy, type ProposalFinding, type ProposalFindingOrigin, type ProposalTrackContext, type ProposeContext, type ProposedCandidate, type RecordSearchEdgeInput, type RedactionStatus, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceScenario, type RegisterSearchNodeInput, type RetainedDispatchOptions, type RetainedDispatchOutcome, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, type RunCampaignOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunOptimizationOptions, type RunOptimizationResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, type RunProfileMatrixSegmentOptions, type RunSearchOptions, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_LEDGER_SCHEMA, type Scenario, type ScenarioAggregate, type ScenarioRollup, type ScenarioSignal, type ScoreboardRenderOptions, type ScoreboardRow, type ScoreboardSummary, type ScoredRollout, type ScoredSurfaceOutcome, type SearchAllocationView, type SearchAllocator, type SearchArtifactCodec, type SearchArtifactKind, type SearchArtifactRef, type SearchAttemptAccounting, type SearchAudit, type SearchBudget, type SearchCancelReason, type SearchCandidateSurface, type SearchCell, type SearchCellAllocatedEvent, type SearchCellCancelledEvent, type SearchCellPlan, type SearchCellResult, type SearchCellSettledEvent, type SearchCellStage, type SearchCellWork, type SearchClaim, type SearchClaimDecision, type SearchClaimEstimator, type SearchClaimFinalist, type SearchClaimPlan, type SearchClaimPower, type SearchClaimTest, type SearchClaimVerification, type SearchCloseReason, type SearchClosedEvent, type SearchCompletion, type SearchCostAccounting, type SearchDecisionRecord, type SearchEdgeAttribution, type SearchEdgeOperator, type SearchEdgeRecordedEvent, type SearchEstimateMethod, type SearchExecutionIdentity, type SearchExecutor, type SearchExpansion, type SearchFailureReason, type SearchHistoryAdmissionOptions, type SearchHistoryAuditSummary, type SearchHistoryCoverage, type SearchHistoryCoverageRow, type SearchHistoryPolicy, type SearchHistoryReceipt, type SearchLane, type SearchLedger, type SearchLedgerAppendResult, type SearchLedgerBinding, type SearchLedgerEntry, SearchLedgerError, type SearchLedgerEvent, type SearchLedgerHash, SearchLedgerIntegrityError, type SearchLedgerTrustedHeadMode, type SearchModelIdentity, type SearchNode, type SearchNodeDecidedEvent, type SearchNodeDecision, type SearchNodeRef, type SearchNodeRegisteredEvent, type SearchNodeStatus, type SearchOpenedEvent, type SearchOpening, type SearchOperation, type SearchOperationKind, type SearchOperationRecordedEvent, type SearchOperationStartedEvent, type SearchPolicy, type SearchPolicyNode, type SearchPolicyView, type SearchPosterior, type SearchProposalBlob, type SearchProposalRequest, type SearchProposalResult, type SearchProposedChild, type SearchProposer, type SearchProposerKind, type SearchProposerPort, type SearchProposerView, SearchRecorder, type SearchRecorderOptions, type SearchReservation, type SearchRunIdentity, type SearchRunResult, type SearchRungDecision, type SearchScoredCell, type SearchSourceRef, type SearchSpend, type SearchSplit, type SearchSplitTasks, type SearchSplits, SearchState, SearchStateView, type SearchSummaryOptions, type SearchSurfaceEffect, type SearchSurfaceEvidence, type SearchSurfaceKind, type SearchTask, type SearchTaskOutcome, type SearchTokenAccounting, type SearchTraceRef, type SearchUnitExtension, type SearchUnitScore, type SearchUnknown, type SequentialDecideFn, type SequentialDecideOptions, type SequentialGateDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SequentialStreamState, type SeriesDistribution, type SessionScript, type SettleSearchCellInput, type SingleRunLock, type SingleRunLockOptions, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SurfaceProposer, type TraceAnalystArtifact, type TraceAnalystScenario, type TraceSpan, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCodeSurfaceIdentity, assertSearchHistoryMatchesState, beam, buildCellSchedule, buildEvidenceVector, campaignBreakdown, campaignCellSearchResult, campaignCoverage, campaignMeanComposite, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, codeSurfaceIdentityMaterial, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, makeProposalFinding, measureCampaignCost, neutralizationGate, neutralizeText, openSearchLedger, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveRunDir, resolveWorktreePath, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimDecision, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchEstimateMethod, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };
|
package/dist/campaign/index.js
CHANGED
|
@@ -1,15 +1,16 @@
|
|
|
1
|
-
import { a as heldoutSignificance, i as dimensionRegressions, o as pairHoldout } from "../statistical-heldout-
|
|
2
|
-
import { $ as
|
|
3
|
-
import {
|
|
4
|
-
import { a as SearchLedgerError, o as SearchLedgerIntegrityError } from "../search-ledger-file-
|
|
5
|
-
import { M as fsCampaignStorage, N as inMemoryCampaignStorage, j as createRunCostLedger, l as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, u as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS } from "../external-optimizer-subprocess-
|
|
6
|
-
import { a as
|
|
7
|
-
import {
|
|
8
|
-
import {
|
|
9
|
-
import { t as
|
|
10
|
-
import {
|
|
11
|
-
import {
|
|
12
|
-
import { a as
|
|
13
|
-
import { n as
|
|
14
|
-
import { n as
|
|
1
|
+
import { a as heldoutSignificance, i as dimensionRegressions, o as pairHoldout } from "../statistical-heldout-TP6ndL0B.js";
|
|
2
|
+
import { $ as resolveRunDir, A as campaignCellSearchResult, C as surfaceDiff, D as createSearchHistoryReceipt, E as assertSearchHistoryMatchesState, F as uniform, M as searchProposalExecution, N as searchReceiptAccounting, O as verifySearchHistoryArtifact, P as asha, Q as planCampaignRun, S as searchEdgeId, X as runEval, Z as runCampaign, _ as crowdedFrontierParent, b as SearchRecorder, c as readGepaCandidatePopulationArtifact, et as buildCellSchedule, f as runImprovementLoop, g as beam, h as aide, j as searchModelIdentity, k as verifySearchHistoryReceipt, m as isProposedCandidate, n as isTransientTransportFailure, p as runOptimization, r as transientDispatchFailure, t as llmJudge, tt as cellCachePath, u as costFromLedgerSummary, v as incumbent, w as surfaceNode, x as developmentClaim, y as incumbentWithOperatorBandit, z as defaultProductionGate } from "../llm-judge-cgbhTKQp.js";
|
|
3
|
+
import { A as campaignSplitDigest, O as campaignCoverage, T as assertCampaignDesign, f as campaignBreakdown, j as campaignSplitDigestFromIdentities, k as campaignScenarioIdentity, l as canonicalDigest, p as campaignMeanComposite, u as emitLoopProvenance, v as powerPreflight } from "../campaign-evidence-DN8rWuBj.js";
|
|
4
|
+
import { a as SearchLedgerError, o as SearchLedgerIntegrityError } from "../search-ledger-file-DTBD6X-y.js";
|
|
5
|
+
import { M as fsCampaignStorage, N as inMemoryCampaignStorage, j as createRunCostLedger, l as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, u as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS } from "../external-optimizer-subprocess-U616hV_l.js";
|
|
6
|
+
import { a as componentSurfaceIdentityMaterial, c as surfaceDispatchRef, i as codeSurfaceIdentityMaterial, l as surfaceHash, o as renderSurfaceDiff, r as assertCodeSurfaceIdentity, s as surfaceContentHash } from "../json-recovery-DK4MDm5r.js";
|
|
7
|
+
import { a as makeProposalFinding } from "../types-BK0blWhe.js";
|
|
8
|
+
import { C as SEARCH_ESTIMATOR, D as searchEstimateMethod, E as searchCellSetDigest, O as searchPosterior, T as estimateNodeFromCells, _ as planSearchClaim, a as searchDivergence, b as verifySearchClaim, g as decideSearchClaim, h as SEARCH_CLAIM_RULE, i as runSearch, o as searchPolicyView, v as searchClaimDecision, w as estimateNode, y as searchClaimReserveUsd } from "../pareto-B9udAt2i.js";
|
|
9
|
+
import { a as validateSearchLedgerEvent, c as SearchStateView, d as searchNodeId, f as searchTaskSetDigest, i as replaySearchLedgerText, l as searchUnitScores, n as openSearchLedger, o as SEARCH_LEDGER_SCHEMA, r as parseSearchLedgerLine, s as SearchState, t as FileSearchLedger, u as searchCellId } from "../search-ledger-BsVi-Ngv.js";
|
|
10
|
+
import { t as acquireSingleRunLock } from "../single-run-lock-BGlW3OSk.js";
|
|
11
|
+
import { n as searchProposerView, t as renderSearchSummary } from "../search-summary-Ct7SrmVc.js";
|
|
12
|
+
import { C as measureCampaignCost, S as analyzeCrossSurfaceInteractions, _ as neutralizationGate, a as createRetainedDispatch, b as loadEvalFixtureScenarios, c as runProfileMatrixSegment, d as scoreUserStory, f as scoreboardSummary, g as neutralizeText, h as sequentialOptimizationMethod, i as scoreDiscrimination, l as runProfileMatrix, m as scopedOptimizationMethod, n as resolveWorktreePath, o as createProfileMatrixPlan, p as userStoryScoreboard, r as verifyCodeSurface, s as finalizeProfileMatrix, t as gitWorktreeAdapter, u as renderScoreboardMarkdown, v as discoverEvalFixtures, x as planEvalFixtureRun, y as loadEvalFixture } from "../campaign-ddr1N3yD.js";
|
|
13
|
+
import { a as importExternalEvaluations, c as composeGate, d as readExternalOptimizerObservationArtifact, i as externalSurface, l as externalTextOptimizationMethod, n as compareOptimizationMethods, o as importGepaPopulation, r as gepaOptimizationMethod, s as heldOutGate, t as skillOptOptimizationMethod, u as decodeExternalTextCandidate } from "../skillopt-optimization-method-qB1M554K.js";
|
|
14
|
+
import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-CBYeD4zi.js";
|
|
15
|
+
import { n as sequentialPairedGate, t as sequentialDecide } from "../sequential-B1I0tRfL.js";
|
|
15
16
|
export { DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, FileSearchLedger, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_LEDGER_SCHEMA, SearchLedgerError, SearchLedgerIntegrityError, SearchRecorder, SearchState, SearchStateView, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCodeSurfaceIdentity, assertSearchHistoryMatchesState, beam, buildCellSchedule, buildEvidenceVector, campaignBreakdown, campaignCellSearchResult, campaignCoverage, campaignMeanComposite, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, codeSurfaceIdentityMaterial, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, makeProposalFinding, measureCampaignCost, neutralizationGate, neutralizeText, openSearchLedger, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveRunDir, resolveWorktreePath, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimDecision, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchEstimateMethod, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };
|
|
@@ -1,31 +1,33 @@
|
|
|
1
|
-
import { s as ValidationError, t as AgentEvalError } from "./errors-
|
|
2
|
-
import { a as hashCanonical, i as compareCodeUnits, r as canonicalString } from "./canonical-
|
|
3
|
-
import {
|
|
4
|
-
import { i as buildAgentProfileCell } from "./agent-profile-cell-
|
|
5
|
-
import { c as validateRunRecord, i as modelHasSnapshot, n as UNKNOWN_MODEL, o as runTaskScore } from "./run-record-
|
|
6
|
-
import { t as comparePairedArms } from "./paired-arms-
|
|
7
|
-
import { i as pairedBootstrap } from "./mean-
|
|
8
|
-
import { o as pairHoldout } from "./statistical-heldout-
|
|
9
|
-
import {
|
|
10
|
-
import {
|
|
11
|
-
import {
|
|
12
|
-
import {
|
|
13
|
-
import {
|
|
14
|
-
import { M as fsCampaignStorage } from "./external-optimizer-subprocess-
|
|
15
|
-
import "./
|
|
16
|
-
import "./
|
|
17
|
-
import "./
|
|
18
|
-
import "./search-
|
|
19
|
-
import "./
|
|
20
|
-
import "./
|
|
21
|
-
import "./
|
|
1
|
+
import { s as ValidationError, t as AgentEvalError } from "./errors-DeEEV9Vx.js";
|
|
2
|
+
import { a as hashCanonical, i as compareCodeUnits, r as canonicalString } from "./canonical-CicJlINA.js";
|
|
3
|
+
import { a as agentProfileModelId, i as agentProfileId, r as agentProfileHash, s as harnessAxisOf } from "./agent-profile-Dngk3LZ0.js";
|
|
4
|
+
import { i as buildAgentProfileCell } from "./agent-profile-cell-CT5JYueM.js";
|
|
5
|
+
import { c as validateRunRecord, i as modelHasSnapshot, n as UNKNOWN_MODEL, o as runTaskScore } from "./run-record-DEjx43cI.js";
|
|
6
|
+
import { t as comparePairedArms } from "./paired-arms-D-54hO-k.js";
|
|
7
|
+
import { i as pairedBootstrap, t as mean } from "./mean-CdDYG-us.js";
|
|
8
|
+
import { o as pairHoldout } from "./statistical-heldout-TP6ndL0B.js";
|
|
9
|
+
import { d as mapConcurrent } from "./journal-C1uLJi0w.js";
|
|
10
|
+
import { n as contentHash, t as canonicalJson } from "./verdict-cache-BVH-fl18.js";
|
|
11
|
+
import { $ as resolveRunDir, Q as planCampaignRun, Z as runCampaign, l as combineComparisonCosts, nt as computeManifestHash, o as executeOptimizationMethod, tt as cellCachePath } from "./llm-judge-cgbhTKQp.js";
|
|
12
|
+
import { n as campaignCellToRunRecord, r as projectCampaignCellQuality, t as campaignCellCostProvenance } from "./run-record-CZy4-TnZ.js";
|
|
13
|
+
import { C as assertRealBackend, T as assertCampaignDesign, k as campaignScenarioIdentity, w as summarizeBackendIntegrity } from "./campaign-evidence-DN8rWuBj.js";
|
|
14
|
+
import { M as fsCampaignStorage } from "./external-optimizer-subprocess-U616hV_l.js";
|
|
15
|
+
import { r as assertCodeSurfaceIdentity, s as surfaceContentHash } from "./json-recovery-DK4MDm5r.js";
|
|
16
|
+
import "./types-BK0blWhe.js";
|
|
17
|
+
import "./pareto-B9udAt2i.js";
|
|
18
|
+
import "./search-ledger-BsVi-Ngv.js";
|
|
19
|
+
import "./search-summary-Ct7SrmVc.js";
|
|
20
|
+
import { r as verifyCompletion } from "./completion-verifier-rqbMh5-4.js";
|
|
21
|
+
import "./external-optimizer-process-BRVmaDFe.js";
|
|
22
|
+
import "./skillopt-optimization-method-qB1M554K.js";
|
|
23
|
+
import "./sequential-B1I0tRfL.js";
|
|
22
24
|
import { createHash } from "node:crypto";
|
|
23
25
|
import { z } from "zod";
|
|
24
26
|
import { closeSync, constants, existsSync, fstatSync, lstatSync, mkdirSync, mkdtempSync, openSync, readFileSync, readSync, readdirSync, readlinkSync, realpathSync, rmSync, statSync } from "node:fs";
|
|
25
27
|
import { basename, dirname, isAbsolute, join, relative, resolve, sep } from "node:path";
|
|
26
28
|
import { devNull, tmpdir } from "node:os";
|
|
27
29
|
import { execFileSync } from "node:child_process";
|
|
28
|
-
|
|
30
|
+
// src/campaign/cost-breakdown.ts
|
|
29
31
|
/** Read the source rates for this execution, never a copied model price table.
|
|
30
32
|
* inputTokens includes cachedTokens. Seat credits are standard model credits;
|
|
31
33
|
* discounts and MCP charges require provider observations, not token guesses.
|
|
@@ -163,8 +165,7 @@ async function measureCampaignCost(input) {
|
|
|
163
165
|
function finite(value, label) {
|
|
164
166
|
if (typeof value !== "number" || !Number.isFinite(value) || value < 0) throw new Error(`${label} must be finite and nonnegative`);
|
|
165
167
|
}
|
|
166
|
-
|
|
167
|
-
//#region src/campaign/cross-surface-context.ts
|
|
168
|
+
// src/campaign/cross-surface-context.ts
|
|
168
169
|
function validateCrossSurfaceInput(input) {
|
|
169
170
|
assertNonEmptyUnique(input.taskOrder, "taskOrder");
|
|
170
171
|
assertNonEmptyUnique(input.componentOrder, "componentOrder");
|
|
@@ -336,8 +337,7 @@ function assertExactSet(expected, actualIterable, actualLabel, expectedLabel) {
|
|
|
336
337
|
function arraysEqual(left, right) {
|
|
337
338
|
return left.length === right.length && left.every((value, index) => value === right[index]);
|
|
338
339
|
}
|
|
339
|
-
|
|
340
|
-
//#region src/campaign/cross-surface-evidence.ts
|
|
340
|
+
// src/campaign/cross-surface-evidence.ts
|
|
341
341
|
function summarizeCandidateEvidence(rows, componentIds, field) {
|
|
342
342
|
if (componentIds.length === 0) return {
|
|
343
343
|
byComponent: [],
|
|
@@ -412,8 +412,7 @@ function appendEvidenceReasons(reasons, evidence, minimum, requireObserved, belo
|
|
|
412
412
|
if (evidence.byComponent.some((component) => component.observedTaskIds.length < minimum)) reasons.push(belowReason);
|
|
413
413
|
if (requireObserved && evidence.byComponent.some((component) => component.unobservedTaskIds.length > 0)) reasons.push(unobservedReason);
|
|
414
414
|
}
|
|
415
|
-
|
|
416
|
-
//#region src/campaign/cross-surface-interaction.ts
|
|
415
|
+
// src/campaign/cross-surface-interaction.ts
|
|
417
416
|
/**
|
|
418
417
|
* Task-paired comparison and deterministic selection for independently
|
|
419
418
|
* proposed agent surfaces. Unlike factorial cell-mean attribution, every
|
|
@@ -889,8 +888,7 @@ function compareSingleSummaries(context, left, right) {
|
|
|
889
888
|
if (byBytes !== 0) return byBytes;
|
|
890
889
|
return context.candidateIndex.get(left.candidate.candidateId) - context.candidateIndex.get(right.candidate.candidateId);
|
|
891
890
|
}
|
|
892
|
-
|
|
893
|
-
//#region src/campaign/fixtures.ts
|
|
891
|
+
// src/campaign/fixtures.ts
|
|
894
892
|
/** Walk `evalsDir` and return the relative name of every fixture directory (one containing an exact-case `PROMPT.md`). */
|
|
895
893
|
function discoverEvalFixtures(evalsDir) {
|
|
896
894
|
const root = resolve(evalsDir);
|
|
@@ -1033,18 +1031,35 @@ function assertInside(root, target, label) {
|
|
|
1033
1031
|
if (rel === "" || !rel.startsWith("..") && !isAbsolute(rel)) return;
|
|
1034
1032
|
throw new Error(`loadEvalFixture: fixture path escapes evalsDir: ${label}`);
|
|
1035
1033
|
}
|
|
1036
|
-
|
|
1037
|
-
|
|
1038
|
-
|
|
1039
|
-
|
|
1040
|
-
|
|
1041
|
-
|
|
1034
|
+
// src/campaign/gates/neutralization-gate.ts
|
|
1035
|
+
/**
|
|
1036
|
+
* @module
|
|
1037
|
+
* Composable placebo / neutralization promotion gate.
|
|
1038
|
+
*
|
|
1039
|
+
* A held-out gate proves a candidate beat baseline. It CANNOT prove the lift came
|
|
1040
|
+
* from the candidate's CONTENT rather than from the prompt/mount FOOTPRINT the
|
|
1041
|
+
* content happened to add (more bytes, a longer prompt). This gate closes that
|
|
1042
|
+
* hole: it compares the candidate's held-out lift against the lift of a
|
|
1043
|
+
* FOOTPRINT-MATCHED neutralized variant (same layout + length, zero content, via
|
|
1044
|
+
* `neutralizeText`). If the neutralized variant reproduces more than
|
|
1045
|
+
* `maxDecorativeFraction` of the candidate's lift, the lift is decorative — it
|
|
1046
|
+
* survives blanking the content — and the candidate is HELD regardless of how
|
|
1047
|
+
* large or significant its raw lift is.
|
|
1048
|
+
*
|
|
1049
|
+
* Compose it AFTER the significance gate — significance says the lift is real,
|
|
1050
|
+
* this says the lift is CAUSED BY THE CONTENT:
|
|
1051
|
+
* composeGate(heldOutGate({ ... }), neutralizationGate({ ... }))
|
|
1052
|
+
*
|
|
1053
|
+
* Requires `ctx.neutralizedJudgeScores`, populated by `runImprovementLoop` when it
|
|
1054
|
+
* is given a `neutralize` function. A gate composed without that wiring fails
|
|
1055
|
+
* loud rather than silently passing an unproven candidate.
|
|
1056
|
+
*/
|
|
1042
1057
|
/** Paired mean held-out lift of `arm` over baseline, on the in-scope cells. */
|
|
1043
1058
|
function pairedLift(arm, baseline, scenarioIds) {
|
|
1044
1059
|
const paired = pairHoldout(arm, baseline, scenarioIds, (s) => s.composite);
|
|
1045
1060
|
const deltas = paired.after.map((a, i) => a - (paired.before[i] ?? 0));
|
|
1046
1061
|
return {
|
|
1047
|
-
lift: mean
|
|
1062
|
+
lift: deltas.length ? mean(deltas) : 0,
|
|
1048
1063
|
n: deltas.length
|
|
1049
1064
|
};
|
|
1050
1065
|
}
|
|
@@ -1098,8 +1113,7 @@ function neutralizationGate(options) {
|
|
|
1098
1113
|
}
|
|
1099
1114
|
};
|
|
1100
1115
|
}
|
|
1101
|
-
|
|
1102
|
-
//#region src/campaign/neutralize.ts
|
|
1116
|
+
// src/campaign/neutralize.ts
|
|
1103
1117
|
/**
|
|
1104
1118
|
* @module
|
|
1105
1119
|
* Footprint-matched neutralization — the placebo control for content-vs-footprint
|
|
@@ -1133,8 +1147,7 @@ const FILLER = "#";
|
|
|
1133
1147
|
function neutralizeText(content) {
|
|
1134
1148
|
return content.replace(/\S/g, FILLER);
|
|
1135
1149
|
}
|
|
1136
|
-
|
|
1137
|
-
//#region src/campaign/optimization-method-composition.ts
|
|
1150
|
+
// src/campaign/optimization-method-composition.ts
|
|
1138
1151
|
/** Optimize a projection while evaluating every proposal in the complete candidate. */
|
|
1139
1152
|
function scopedOptimizationMethod(options) {
|
|
1140
1153
|
return {
|
|
@@ -1235,8 +1248,7 @@ function sequentialOptimizationMethod(options) {
|
|
|
1235
1248
|
}
|
|
1236
1249
|
};
|
|
1237
1250
|
}
|
|
1238
|
-
|
|
1239
|
-
//#region src/campaign/presets/playback.ts
|
|
1251
|
+
// src/campaign/presets/playback.ts
|
|
1240
1252
|
/**
|
|
1241
1253
|
* Score one story's produced state against its requirements. Thin wrapper over
|
|
1242
1254
|
* `verifyCompletion` that builds the gold from the story and returns a
|
|
@@ -1337,8 +1349,7 @@ function renderScoreboardMarkdown(rows, opts = {}) {
|
|
|
1337
1349
|
}
|
|
1338
1350
|
return out.join("\n");
|
|
1339
1351
|
}
|
|
1340
|
-
|
|
1341
|
-
//#region src/campaign/presets/run-profile-matrix.ts
|
|
1352
|
+
// src/campaign/presets/run-profile-matrix.ts
|
|
1342
1353
|
/**
|
|
1343
1354
|
* `runProfileMatrix` — the missing keystone between `runAgentMatrix` and the
|
|
1344
1355
|
* backend-integrity guard.
|
|
@@ -1425,9 +1436,6 @@ function sanitize(id) {
|
|
|
1425
1436
|
function sha(input) {
|
|
1426
1437
|
return createHash("sha256").update(JSON.stringify(input)).digest("hex");
|
|
1427
1438
|
}
|
|
1428
|
-
function mean(xs) {
|
|
1429
|
-
return xs.length === 0 ? 0 : xs.reduce((a, b) => a + b, 0) / xs.length;
|
|
1430
|
-
}
|
|
1431
1439
|
function receiptModels(cell) {
|
|
1432
1440
|
const reported = cell.resolvedModels ?? (cell.resolvedModel ? [cell.resolvedModel] : []);
|
|
1433
1441
|
return [...new Set(reported.map((model) => model.trim()).filter(Boolean))];
|
|
@@ -1707,8 +1715,7 @@ function rollupByPersona(records, scenarios, personaOf) {
|
|
|
1707
1715
|
for (const s of scenarios) personaByScenarioId.set(s.id, personaOf(s));
|
|
1708
1716
|
return rollup(records, (r) => r.scenarioId ? personaByScenarioId.get(r.scenarioId) : void 0);
|
|
1709
1717
|
}
|
|
1710
|
-
|
|
1711
|
-
//#region src/campaign/presets/segmented-profile-matrix.ts
|
|
1718
|
+
// src/campaign/presets/segmented-profile-matrix.ts
|
|
1712
1719
|
/**
|
|
1713
1720
|
* Eval-owned execution for profile matrices whose external authorization is
|
|
1714
1721
|
* shorter than the complete matrix.
|
|
@@ -2146,8 +2153,7 @@ function requireIdentifier(value, name) {
|
|
|
2146
2153
|
function byteLength(value) {
|
|
2147
2154
|
return new TextEncoder().encode(value).byteLength;
|
|
2148
2155
|
}
|
|
2149
|
-
|
|
2150
|
-
//#region src/campaign/retained-dispatch.ts
|
|
2156
|
+
// src/campaign/retained-dispatch.ts
|
|
2151
2157
|
const digest = z.string().regex(/^sha256:[a-f0-9]{64}$/);
|
|
2152
2158
|
const headerSchema = z.object({
|
|
2153
2159
|
kind: z.literal("retained-dispatch-v1"),
|
|
@@ -2519,8 +2525,7 @@ async function dispatchOnce(journal, id, request) {
|
|
|
2519
2525
|
}
|
|
2520
2526
|
return retainResult(journal, id, value);
|
|
2521
2527
|
}
|
|
2522
|
-
|
|
2523
|
-
//#region src/campaign/scenario-selection.ts
|
|
2528
|
+
// src/campaign/scenario-selection.ts
|
|
2524
2529
|
const DEFAULT_SATURATION_CEILING = .999;
|
|
2525
2530
|
/** Variance below this is treated as "every candidate scored the same". */
|
|
2526
2531
|
const VARIANCE_EPSILON = 1e-9;
|
|
@@ -2574,8 +2579,7 @@ function scoreDiscrimination(signals, opts) {
|
|
|
2574
2579
|
};
|
|
2575
2580
|
}).sort(compareDiscrimination);
|
|
2576
2581
|
}
|
|
2577
|
-
|
|
2578
|
-
//#region src/campaign/worktree/index.ts
|
|
2582
|
+
// src/campaign/worktree/index.ts
|
|
2579
2583
|
/**
|
|
2580
2584
|
* VCS-pluggable worktree adapter. One improvement = one worktree, PR-like
|
|
2581
2585
|
* (multiple commits allowed). A code-tier proposer's `propose()` creates a
|
|
@@ -3095,7 +3099,6 @@ function verifyCodeSurface(surface, worktreeDir) {
|
|
|
3095
3099
|
function resolveWorktreePath(surface, worktreeDir) {
|
|
3096
3100
|
return verifyCodeSurface(surface, worktreeDir).path;
|
|
3097
3101
|
}
|
|
3098
|
-
//#endregion
|
|
3099
3102
|
export { measureCampaignCost as C, analyzeCrossSurfaceInteractions as S, neutralizationGate as _, createRetainedDispatch as a, loadEvalFixtureScenarios as b, runProfileMatrixSegment as c, scoreUserStory as d, scoreboardSummary as f, neutralizeText as g, sequentialOptimizationMethod as h, scoreDiscrimination as i, runProfileMatrix as l, scopedOptimizationMethod as m, resolveWorktreePath as n, createProfileMatrixPlan as o, userStoryScoreboard as p, verifyCodeSurface as r, finalizeProfileMatrix as s, gitWorktreeAdapter as t, renderScoreboardMarkdown as u, discoverEvalFixtures as v, planEvalFixtureRun as x, loadEvalFixture as y };
|
|
3100
3103
|
|
|
3101
|
-
//# sourceMappingURL=campaign-
|
|
3104
|
+
//# sourceMappingURL=campaign-ddr1N3yD.js.map
|