@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
package/CHANGELOG.md
CHANGED
|
@@ -4,6 +4,174 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
|
|
|
4
4
|
|
|
5
5
|
---
|
|
6
6
|
|
|
7
|
+
## [0.222.1] — 2026-10-26
|
|
8
|
+
|
|
9
|
+
- fix(deps): depend on zod ^4.6.6 so consumers resolve one copy (#1133)
|
|
10
|
+
- chore(test): retire incomplete Runtime adoption snapshot
|
|
11
|
+
- fix(release): build public exports before integration tests
|
|
12
|
+
- chore(release): 0.222.0
|
|
13
|
+
- feat(campaign): share concurrency for measured scenario adapters
|
|
14
|
+
- fix(supervisor-run): staged play-check reward is stages passed, read from the SCORE line only (#1126)
|
|
15
|
+
- refactor(statistics): delete duplicate numeric summary helpers
|
|
16
|
+
- fix(rollout): import survives a damaged deliverables archive; named ids are not keys (#1125)
|
|
17
|
+
- feat(supervisor-run): rollouts importer for Discovery runs, harness sessions and gate receipts (#1124)
|
|
18
|
+
- feat(statistics): extend series summaries with nearest-rank evidence
|
|
19
|
+
- test(rollout): isolate bundle permission probe from runner identity
|
|
20
|
+
- fix(rollout): license class is provenance and an opt-in filter, not a default exclusion (#1121)
|
|
21
|
+
- fix(rollout): allow nonroot images to read the base bundle
|
|
22
|
+
- feat(rollout): license class on rollout lines and whole-row privacy drop in rollout-release (#1119)
|
|
23
|
+
- test(rollout): inherit configured identity for Git fixtures
|
|
24
|
+
- test(rollout): capture empty tool definitions at the ledger boundary
|
|
25
|
+
- feat(rollout): write verified episodes as Harbor tasks
|
|
26
|
+
- feat(rollout): project captured policy steps into Tinker data
|
|
27
|
+
- adopt-stack: delete the adoption branch when its squash merge lands (#1115)
|
|
28
|
+
- chore(deps)!: require Python 3.11 for patched locks and refresh dependencies
|
|
29
|
+
- feat(rollout): read retained Codex sessions through the shared reader
|
|
30
|
+
- chore(release): 0.221.8
|
|
31
|
+
- test(adopt): retain failed browser child outcomes
|
|
32
|
+
- docs: leave changelog preparation to the release workflow
|
|
33
|
+
- chore(release): 0.221.7
|
|
34
|
+
- feat(multishot): conversation legs report the USD they spent
|
|
35
|
+
- chore(release): 0.221.6
|
|
36
|
+
- docs(traces): bound the incomplete usage coverage claim
|
|
37
|
+
- feat: own judge-panel calibration, multishot conversation legs and served-panel identity
|
|
38
|
+
- test(traces): typecheck OTLP usage integration proof
|
|
39
|
+
- fix(traces): retain incomplete OTLP token usage
|
|
40
|
+
- chore(release): 0.221.5
|
|
41
|
+
- ci: retain private adopter failure evidence (#1106)
|
|
42
|
+
- test(process): isolate process-tree fixture environment
|
|
43
|
+
- fix(adopt): invoke each resolved Playwright CLI (#1096)
|
|
44
|
+
- test(systemone): close disposable provider connections
|
|
45
|
+
- chore(release): 0.221.4
|
|
46
|
+
- test(adoption): inherit the maintained integration deadline
|
|
47
|
+
- fix(adoption): scope guard allowance to disposable Git fixtures
|
|
48
|
+
- fix(analyst): carry root-cause target into failure-mode rubric
|
|
49
|
+
- ci: check locally and in publication paths
|
|
50
|
+
- test(adopt): give real-git fixtures their own commit identity (#1101)
|
|
51
|
+
- chore(release): 0.221.3 (#1099)
|
|
52
|
+
- fix(analyst): repin the benchmark implementation digest after #1097 (#1098)
|
|
53
|
+
- fix(trace-analyst): build OTLP stores from bytes, not one joined string
|
|
54
|
+
- fix(adopt): prepare caller-locked Chromium before gates
|
|
55
|
+
- test(adopt): retain admitted identity in real Git fixtures
|
|
56
|
+
- fix(adopt): retain the sparse caller entrypoint
|
|
57
|
+
- fix(adopt): settle pushed proposal reads against remote refs
|
|
58
|
+
- chore(release): 0.221.2
|
|
59
|
+
- Return plain analyst findings from decision replay
|
|
60
|
+
- chore(release): 0.221.1 (#1091)
|
|
61
|
+
- fix(deps): pin zod 4.6.6 (GHSA-mrhx-c7xc-m3w5) (#1088)
|
|
62
|
+
- Give the disposable adoption fixture its own Git author
|
|
63
|
+
- Fix real adoption drift fixture under the declared package manager
|
|
64
|
+
- Capture supporting trace events in the original native decision
|
|
65
|
+
- test: inherit the integration default for behavioral questions
|
|
66
|
+
- test: set the integration default from the observed loaded tail
|
|
67
|
+
- chore(release): 0.221.0
|
|
68
|
+
- test: own the measured integration deadline in Vitest config
|
|
69
|
+
- test: await observed integration completion across tool paths
|
|
70
|
+
- feat: preserve finding composition and assigned arm populations
|
|
71
|
+
- test: await decision-point HTTP and ledger completion
|
|
72
|
+
- test: await durable readiness in filesystem integration cases
|
|
73
|
+
- test(process): wait for argv child readiness before abort
|
|
74
|
+
- feat(analyst): add bounded calibrated trace question observations
|
|
75
|
+
- fix(analyst): reject surrounding whitespace in caller gold ids
|
|
76
|
+
- test(analyst): reproduce whitespace source-id bypass
|
|
77
|
+
- test(optimizer): retire fake CLI metering harness
|
|
78
|
+
- test(analyst): isolate recorded gold admission scenarios
|
|
79
|
+
- fix(systemone): validate decision boundaries before routing
|
|
80
|
+
- ci: pin the eval publication dispatcher body
|
|
81
|
+
- fix(analyst): reserve excluded source ids across caller gold
|
|
82
|
+
- fix(analyst): validate root provenance before gold admission
|
|
83
|
+
- fix(analyst): preserve actor boundaries in AgentRx calibration gold
|
|
84
|
+
- Preserve manual drafts observed while recording failed adoption (#1076)
|
|
85
|
+
- Recover updater-owned failure drafts while preserving manual holds (#1075)
|
|
86
|
+
- feat(systemone): persist calibrated decision points and outcomes
|
|
87
|
+
- ci: move eval verification and Python publishing to owned runners
|
|
88
|
+
- chore(deps): bump opentelemetry-api in /clients/python (#1068)
|
|
89
|
+
- fix(adopt): quote optional workflow flags
|
|
90
|
+
- fix(adopt): preserve shallow history and caller-owned dependency references
|
|
91
|
+
- fix(adopt): reconcile effective workspace pins and preserve proposal history
|
|
92
|
+
|
|
93
|
+
## [0.222.0] — 2026-10-26
|
|
94
|
+
|
|
95
|
+
- feat(campaign): share concurrency for measured scenario adapters
|
|
96
|
+
- fix(supervisor-run): staged play-check reward is stages passed, read from the SCORE line only (#1126)
|
|
97
|
+
- refactor(statistics): delete duplicate numeric summary helpers
|
|
98
|
+
- fix(rollout): import survives a damaged deliverables archive; named ids are not keys (#1125)
|
|
99
|
+
- feat(supervisor-run): rollouts importer for Discovery runs, harness sessions and gate receipts (#1124)
|
|
100
|
+
- feat(statistics): extend series summaries with nearest-rank evidence
|
|
101
|
+
- test(rollout): isolate bundle permission probe from runner identity
|
|
102
|
+
- fix(rollout): license class is provenance and an opt-in filter, not a default exclusion (#1121)
|
|
103
|
+
- fix(rollout): allow nonroot images to read the base bundle
|
|
104
|
+
- feat(rollout): license class on rollout lines and whole-row privacy drop in rollout-release (#1119)
|
|
105
|
+
- test(rollout): inherit configured identity for Git fixtures
|
|
106
|
+
- test(rollout): capture empty tool definitions at the ledger boundary
|
|
107
|
+
- feat(rollout): write verified episodes as Harbor tasks
|
|
108
|
+
- feat(rollout): project captured policy steps into Tinker data
|
|
109
|
+
- adopt-stack: delete the adoption branch when its squash merge lands (#1115)
|
|
110
|
+
- chore(deps)!: require Python 3.11 for patched locks and refresh dependencies
|
|
111
|
+
- feat(rollout): read retained Codex sessions through the shared reader
|
|
112
|
+
- chore(release): 0.221.8
|
|
113
|
+
- test(adopt): retain failed browser child outcomes
|
|
114
|
+
- docs: leave changelog preparation to the release workflow
|
|
115
|
+
- chore(release): 0.221.7
|
|
116
|
+
- feat(multishot): conversation legs report the USD they spent
|
|
117
|
+
- chore(release): 0.221.6
|
|
118
|
+
- docs(traces): bound the incomplete usage coverage claim
|
|
119
|
+
- feat: own judge-panel calibration, multishot conversation legs and served-panel identity
|
|
120
|
+
- test(traces): typecheck OTLP usage integration proof
|
|
121
|
+
- fix(traces): retain incomplete OTLP token usage
|
|
122
|
+
- chore(release): 0.221.5
|
|
123
|
+
- ci: retain private adopter failure evidence (#1106)
|
|
124
|
+
- test(process): isolate process-tree fixture environment
|
|
125
|
+
- fix(adopt): invoke each resolved Playwright CLI (#1096)
|
|
126
|
+
- test(systemone): close disposable provider connections
|
|
127
|
+
- chore(release): 0.221.4
|
|
128
|
+
- test(adoption): inherit the maintained integration deadline
|
|
129
|
+
- fix(adoption): scope guard allowance to disposable Git fixtures
|
|
130
|
+
- fix(analyst): carry root-cause target into failure-mode rubric
|
|
131
|
+
- ci: check locally and in publication paths
|
|
132
|
+
- test(adopt): give real-git fixtures their own commit identity (#1101)
|
|
133
|
+
- chore(release): 0.221.3 (#1099)
|
|
134
|
+
- fix(analyst): repin the benchmark implementation digest after #1097 (#1098)
|
|
135
|
+
- fix(trace-analyst): build OTLP stores from bytes, not one joined string
|
|
136
|
+
- fix(adopt): prepare caller-locked Chromium before gates
|
|
137
|
+
- test(adopt): retain admitted identity in real Git fixtures
|
|
138
|
+
- fix(adopt): retain the sparse caller entrypoint
|
|
139
|
+
- fix(adopt): settle pushed proposal reads against remote refs
|
|
140
|
+
- chore(release): 0.221.2
|
|
141
|
+
- Return plain analyst findings from decision replay
|
|
142
|
+
- chore(release): 0.221.1 (#1091)
|
|
143
|
+
- fix(deps): pin zod 4.6.6 (GHSA-mrhx-c7xc-m3w5) (#1088)
|
|
144
|
+
- Give the disposable adoption fixture its own Git author
|
|
145
|
+
- Fix real adoption drift fixture under the declared package manager
|
|
146
|
+
- Capture supporting trace events in the original native decision
|
|
147
|
+
- test: inherit the integration default for behavioral questions
|
|
148
|
+
- test: set the integration default from the observed loaded tail
|
|
149
|
+
- chore(release): 0.221.0
|
|
150
|
+
- test: own the measured integration deadline in Vitest config
|
|
151
|
+
- test: await observed integration completion across tool paths
|
|
152
|
+
- feat: preserve finding composition and assigned arm populations
|
|
153
|
+
- test: await decision-point HTTP and ledger completion
|
|
154
|
+
- test: await durable readiness in filesystem integration cases
|
|
155
|
+
- test(process): wait for argv child readiness before abort
|
|
156
|
+
- feat(analyst): add bounded calibrated trace question observations
|
|
157
|
+
- fix(analyst): reject surrounding whitespace in caller gold ids
|
|
158
|
+
- test(analyst): reproduce whitespace source-id bypass
|
|
159
|
+
- test(optimizer): retire fake CLI metering harness
|
|
160
|
+
- test(analyst): isolate recorded gold admission scenarios
|
|
161
|
+
- fix(systemone): validate decision boundaries before routing
|
|
162
|
+
- ci: pin the eval publication dispatcher body
|
|
163
|
+
- fix(analyst): reserve excluded source ids across caller gold
|
|
164
|
+
- fix(analyst): validate root provenance before gold admission
|
|
165
|
+
- fix(analyst): preserve actor boundaries in AgentRx calibration gold
|
|
166
|
+
- Preserve manual drafts observed while recording failed adoption (#1076)
|
|
167
|
+
- Recover updater-owned failure drafts while preserving manual holds (#1075)
|
|
168
|
+
- feat(systemone): persist calibrated decision points and outcomes
|
|
169
|
+
- ci: move eval verification and Python publishing to owned runners
|
|
170
|
+
- chore(deps): bump opentelemetry-api in /clients/python (#1068)
|
|
171
|
+
- fix(adopt): quote optional workflow flags
|
|
172
|
+
- fix(adopt): preserve shallow history and caller-owned dependency references
|
|
173
|
+
- fix(adopt): reconcile effective workspace pins and preserve proposal history
|
|
174
|
+
|
|
7
175
|
## [0.221.8] — 2026-10-25
|
|
8
176
|
|
|
9
177
|
- test(adopt): retain failed browser child outcomes
|
package/dist/adapters/http.d.ts
CHANGED
|
@@ -1,7 +1,6 @@
|
|
|
1
|
-
import { d as CostReceipt } from "../cost-ledger-
|
|
2
|
-
import { R as Scenario, d as DispatchContext, f as DispatchFn } from "../types-
|
|
3
|
-
import "../index-
|
|
4
|
-
//#region src/adapters/http.d.ts
|
|
1
|
+
import { d as CostReceipt } from "../cost-ledger-Czro9wmo.js";
|
|
2
|
+
import { R as Scenario, d as DispatchContext, f as DispatchFn } from "../types-e3HGaI7P.js";
|
|
3
|
+
import "../index-Cu12qxJb.js";
|
|
5
4
|
export interface HttpDispatchOptions<TScenario extends Scenario, _TArtifact> {
|
|
6
5
|
/** Static endpoint URL. Mutually exclusive with `resolveUrl`. */
|
|
7
6
|
url?: string;
|
|
@@ -118,5 +117,4 @@ export interface DispatchServerHandle {
|
|
|
118
117
|
* minimal — works in plain Node, sandbox, or any container.
|
|
119
118
|
*/
|
|
120
119
|
export declare function runDispatchServer<TScenario extends Scenario, TArtifact>(opts: RunDispatchServerOptions<TScenario, TArtifact>): Promise<DispatchServerHandle>;
|
|
121
|
-
//#endregion
|
|
122
120
|
//# sourceMappingURL=http.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"http.d.ts","names":["Dispatch"],"sources":["../../src/adapters/http.ts"],"mappings":"
|
|
1
|
+
{"version":3,"file":"http.d.ts","names":["Dispatch"],"sources":["../../src/adapters/http.ts"],"mappings":";;;iBA2DiB,oBAAoB,kBAAkB,UAAU;;EAE/D;;;;;;EAMA,cAAc;IAAS,UAAU;IAAW;IAAoB;;;EAEhE,gCAAgC;;EAEhC,UAAU;;EAEV;;;;;;;EAOA;;EAEA,mBAAmB;;iBAGJ,wBAAwB,kBAAkB;EACzD,UAAU;EACV;EACA;EACA;EACA;EACA;EACA;EACA;;iBAGe,yBAAyB;EACxC,UAAU;;;;;;;;EAQV,WAAW;;;;;;;;;;;wBAkBG,aAAa,kBAAkB,UAAU,WACvD,MAAM,oBAAoB,WAAW,aACpCA,WAAS,WAAW;iBAiKN,yBAAyB,kBAAkB,UAAU;;EAEpE,UAAUA,WAAS,WAAW;;EAE9B;;EAEA;;;;EAIA;;EAEA;;;;;;;;EAQA,kBACE,KAAK,wBAAwB,YAC7B,QAAQ,gBACL,QAAQ;;EAEb;;EAEA,aAAa;IACX;IACA;IACA;IACA;;;iBAIa;;EAEf;;EAEA,aAAa;;;;;;;;;;;;;;;;;;;;wBAqBO,kBAAkB,kBAAkB,UAAU,WAClE,MAAM,yBAAyB,WAAW,aACzC,QAAQ"}
|
package/dist/adapters/http.js
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
|
|
1
|
+
// src/adapters/http.ts
|
|
2
2
|
function resolveAuth(auth) {
|
|
3
3
|
if (!auth) return Promise.resolve(null);
|
|
4
4
|
if (typeof auth === "string") return Promise.resolve(auth);
|
|
@@ -276,7 +276,6 @@ const NOOP_ARTIFACTS = {
|
|
|
276
276
|
const UNCONFIGURED_COST = { async runPaidCall() {
|
|
277
277
|
throw new Error("runDispatchServer: paid work requires a contextFactory cost meter.");
|
|
278
278
|
} };
|
|
279
|
-
//#endregion
|
|
280
279
|
export { httpDispatch, runDispatchServer };
|
|
281
280
|
|
|
282
281
|
//# sourceMappingURL=http.js.map
|
|
@@ -1,4 +1,3 @@
|
|
|
1
|
-
//#region src/rl/adversarial.d.ts
|
|
2
1
|
/**
|
|
3
2
|
* Adversarial mutation contract.
|
|
4
3
|
*
|
|
@@ -16,6 +15,5 @@ interface AdversarialMutation<S> {
|
|
|
16
15
|
*/
|
|
17
16
|
mutate(parent: S, rng: () => number): Promise<S[]> | S[];
|
|
18
17
|
}
|
|
19
|
-
//#endregion
|
|
20
18
|
export { AdversarialMutation as t };
|
|
21
|
-
//# sourceMappingURL=adversarial-
|
|
19
|
+
//# sourceMappingURL=adversarial-DnQQFqTR.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"adversarial-DnQQFqTR.d.ts","names":[],"sources":["../src/rl/adversarial.ts"],"mappings":";;;;;;;;;UAUiB,oBAAoB;EACnC;;;;;EAKA,OAAO,QAAQ,GAAG,oBAAoB,QAAQ,OAAO"}
|
|
@@ -1,8 +1,7 @@
|
|
|
1
|
-
import { f as CostReceiptInput, m as CustomTokenPricing } from "./cost-ledger-
|
|
2
|
-
import { J as TraceAnalysisStore, i as AnalystFinding, m as EvidenceRef, p as AnalystUsageReceipt } from "./types-
|
|
3
|
-
import { C as ChatResponse, S as ChatRequest } from "./types-
|
|
1
|
+
import { f as CostReceiptInput, m as CustomTokenPricing } from "./cost-ledger-Czro9wmo.js";
|
|
2
|
+
import { J as TraceAnalysisStore, i as AnalystFinding, m as EvidenceRef, p as AnalystUsageReceipt } from "./types-BEvjDziM.js";
|
|
3
|
+
import { C as ChatResponse, S as ChatRequest } from "./types-0KglnQ9o.js";
|
|
4
4
|
import { AgentProfile, AgentProfile as AgentProfile$1, HarnessType, HarnessType as HarnessType$1 } from "@tangle-network/agent-interface";
|
|
5
|
-
//#region src/campaign/external-optimizer-contracts.d.ts
|
|
6
5
|
interface ExternalOptimizerProcessLimits {
|
|
7
6
|
/** Maximum serialized input written for the child process. */
|
|
8
7
|
maxInputBytes: number;
|
|
@@ -167,8 +166,6 @@ interface ExternalOptimizerWireCounts {
|
|
|
167
166
|
/** Successful 2xx responses recorded on this wire. */
|
|
168
167
|
successfulCompletions: number;
|
|
169
168
|
}
|
|
170
|
-
//#endregion
|
|
171
|
-
//#region src/analyst/benchmark.d.ts
|
|
172
169
|
interface AnalystEvidenceExpectation {
|
|
173
170
|
uri: string;
|
|
174
171
|
kind?: EvidenceRef['kind'];
|
|
@@ -389,8 +386,6 @@ interface RunAnalystBenchmarkOptions<TInput> {
|
|
|
389
386
|
signal?: AbortSignal;
|
|
390
387
|
}
|
|
391
388
|
declare function runAnalystBenchmark<TInput>(options: RunAnalystBenchmarkOptions<TInput>): Promise<AnalystBenchmarkResult>;
|
|
392
|
-
//#endregion
|
|
393
|
-
//#region src/agent-profile.d.ts
|
|
394
389
|
/**
|
|
395
390
|
* The agentic coding harnesses an eval sweeps by default — the ones we care about
|
|
396
391
|
* ranking. This is the SINGLE source of that list; consumers import it instead of
|
|
@@ -470,6 +465,5 @@ declare function agentProfileId(profile: AgentProfile): string;
|
|
|
470
465
|
* as absent; explicit `null` fields remain hash-bearing.
|
|
471
466
|
*/
|
|
472
467
|
declare function agentProfileHash(profile: AgentProfile): string;
|
|
473
|
-
//#endregion
|
|
474
468
|
export { traceStoreEvidenceResolver as A, ExternalOptimizerModelCallRequest as B, AnalystEvidenceResolutionError as C, AnalystLatencyDistribution as D, AnalystIssueExpectation as E, ExternalOptimizerEndpointFormat as F, ExternalOptimizerRunnerCommand as G, ExternalOptimizerModelExecutionObservation as H, ExternalOptimizerEvaluationObservation as I, ExternalTextEvaluationRequest as J, ExternalOptimizerWireCounts as K, ExternalOptimizerEvaluationRefusalReason as L, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS as M, ExternalOptimizerCallbackLimits as N, RunAnalystBenchmarkOptions as O, ExternalOptimizerChatRequest as P, ExternalOptimizerModelBudget as R, AnalystEvidenceResolution as S, AnalystFindingScore as T, ExternalOptimizerProcessLimits as U, ExternalOptimizerModelCallResult as V, ExternalOptimizerResumeMode as W, AnalystBenchmarkProvenance as _, ProfileAxisSpec as a, AnalystBenchmarkSummary as b, expandProfileAxes as c, AnalystBenchmarkDatasetRef as d, AnalystBenchmarkDescriptor as f, AnalystBenchmarkOutput as g, AnalystBenchmarkObservation as h, HarnessType$1 as i, DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS as j, runAnalystBenchmark as k, harnessAxisOf as l, AnalystBenchmarkLabelState as m, CODING_HARNESSES as n, agentProfileHash as o, AnalystBenchmarkError as p, ExternalTextCandidate as q, HARNESS_NATIVE_MODEL as r, agentProfileId as s, AgentProfile$1 as t, AnalystBenchmarkCase as u, AnalystBenchmarkResult as v, AnalystEvidenceResolver as w, AnalystEvidenceExpectation as x, AnalystBenchmarkRunner as y, ExternalOptimizerModelCall as z };
|
|
475
|
-
//# sourceMappingURL=agent-profile-
|
|
469
|
+
//# sourceMappingURL=agent-profile-CSEQRC9l.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agent-profile-CSEQRC9l.d.ts","names":["_$___cost_ledger0.CostReceiptInput"],"sources":["../src/campaign/external-optimizer-contracts.ts","../src/analyst/benchmark.ts","../src/agent-profile.ts"],"mappings":";;;;UAciB;;EAEf;;EAEA;;EAEA;;cAGW,2CAA2C,SAAS;UAOhD;;EAEf;;EAEA;;cAGW,4CAA4C,SAAS;UAMjD;EACf;EACA;EACA,MAAM,OAAO;;EAEb,SAAS,QAAQ;;KAGP;KAEA,iCAAiC;UAE5B;EACf,WAAW;EACX;;KAqDG,aAAa,KAAK,eAAc,6BACjC,IACA,0BAA0B,gBACf,aAAa,OACtB,+BACc,WAAW,IAAI,aAAa,EAAE,SAC1C;;KAGI,+BAA+B,aACzC,KAAK;EAA0B;;KAGrB;;UAMK;;WAEN;;WAEA,SAAS;;WAET,iBAAiB;WACjB,QAAQ;;;KAIP;WAEG;;WAEA,UAAU;;;;;;;WAOV,SAPUA;;WASV;;WAGA;;WAEA;;WAEA,SATkCA;;WAWlC;;;;;;;;;;;;KAaH,8BACV,SAAS,sCACN,QAAQ;;KAGD;WAEG;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;WAGA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;KAGH;;;;;;KAUA;WAEG;WACA;WACA,WAAW;WACX;;WAGA;WACA;WACA,WAAW;WACX;WACA;;WAEA;WACA;;WAGA;WACA;WACA,QAAQ;WACR,YAAY;WACZ;WACA;;UAGE;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;EAUA;;;;;;EAMA,UAAU;;EAEV;;;UAIe;;EAEf;;EAEA;;UCpQe;EACf;EACA,OAAO;;UAGQ;EACf;EACA;EACA;EACA;EACA,oBAAoB;EACpB;;EAEA,4BAA4B;;KAGlB;UAEK,qBAAqB;EACpC;;EAEA;;EAEA,YAAY;EACZ,OAAO;EACP,yBAAyB;;EAEzB,2BAA2B;EAC3B;EACA,WAAW;;UAGI;EACf;EACA;EACA;EACA;EACA;EACA,mBAAmB;EACnB;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;EACA;;UAGe;EACf,UAAU;EACV;EACA;;UAGe;EACf;EACA;EACA,oBAAoB;EACpB,QAAQ;;EAER;;KAGU,wBAAwB,qBAAqB;EACvD;EACA,WAAW;EACX,UAAU;EACV,SAAS;gBACK;;;;;iBAMA,2BAA2B,QACzC,WAAW,OAAO,WAAW,qBAC5B,wBAAwB;UAoBV;EACf,mBAAmB;EACnB,QAAQ;EACR,WAAW;;;;;EAKX;;EAEA,QAAQ;;UAGO;EACf;EACA;EACA;EACA;;UAGe,uBAAuB;EACtC;EACA,QACE,OAAO,QACP;IAAW;IAAgB;IAAoB,SAAS;MACvD,yBAAyB,QAAQ;;UAGrB;EACf;EACA;EACA;EACA,YAAY;EACZ;EACA;EACA;EACA;EACA,mBAAmB;EACnB,OAAO;EACP,qBAAqB;EACrB;EACA,eAAe;EACf,QAAQ;EACR,iBAAiB;EACjB,QAAQ;;UAGO;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;EACA,WAAW;EACX;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;;UAGe;EACf;EACA,UAAU;EACV;EACA,cAAc;EACd,WAAW;;UAGI,mCAAmC;EAClD;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf,YAAY;EACZ,cAAc;EACd,WAAW;;UAGI,2BAA2B;EAC1C,gBAAgB,qBAAqB;EACrC,kBAAkB,uBAAuB;EACzC;EACA;EACA;EACA,kBAAkB,wBAAwB;EAC1C,YAAY;;EAEZ,+BAA+B;EAC/B,iBAAiB,aAAa,uCAAuC;EACrE,SAAS;;iBAGW,oBAAoB,QACxC,SAAS,2BAA2B,UACnC,QAAQ;;;;;;;;cCnQE,2BAA2B;UAOvB;;;EAGf,MAAM;;EAEN,qBAAqB;;;EAGrB;;;;;;EAMA;;;;;;;cAQW;;;;;;;;;;;;;;;;;;;iBAoBG,kBAAkB,MAAM,kBAAkB;;;;;;;;iBAwD1C,cACd,SAAS,KAAK;EACX,SAAS;EAAa;;;;;;;;;iBAiBX,eAAe,SAAS;;;;;;;;;;;iBAmDxB,iBAAiB,SAAS"}
|
|
@@ -0,0 +1,153 @@
|
|
|
1
|
+
import { s as ValidationError } from "./errors-DeEEV9Vx.js";
|
|
2
|
+
import { r as canonicalString } from "./canonical-CicJlINA.js";
|
|
3
|
+
import { createHash } from "node:crypto";
|
|
4
|
+
import { harnessSupportsModel } from "@tangle-network/agent-interface";
|
|
5
|
+
// src/agent-profile.ts
|
|
6
|
+
/**
|
|
7
|
+
* The agentic coding harnesses an eval sweeps by default — the ones we care about
|
|
8
|
+
* ranking. This is the SINGLE source of that list; consumers import it instead of
|
|
9
|
+
* re-declaring their own (a re-declared list is how the fleet drifts). Pass an
|
|
10
|
+
* explicit `harnesses` (e.g. `harnessTypeSchema.options` for literally every known
|
|
11
|
+
* harness) to widen beyond these.
|
|
12
|
+
*/
|
|
13
|
+
const CODING_HARNESSES = [
|
|
14
|
+
"opencode",
|
|
15
|
+
"claude-code",
|
|
16
|
+
"codex",
|
|
17
|
+
"kimi-code"
|
|
18
|
+
];
|
|
19
|
+
/** Model sentinel for a vendor-locked harness that supports none of the swept models:
|
|
20
|
+
* it carries no provider prefix, so `harnessSupportsModel` accepts it and the harness
|
|
21
|
+
* resolves it to its own native default model at runtime (e.g. kimi-code → its Kimi
|
|
22
|
+
* model). Lets `expandProfileAxes` snap-instead-of-drop without a per-harness flagship
|
|
23
|
+
* table that would rot as router catalogs change. */
|
|
24
|
+
const HARNESS_NATIVE_MODEL = "default";
|
|
25
|
+
/**
|
|
26
|
+
* Expand a base profile across the harness × model matrix into the `AgentProfile[]`
|
|
27
|
+
* that `runProfileMatrix` / `selfImprove` score — the ONE place "which harnesses ×
|
|
28
|
+
* which models do we evaluate" lives, so no product hand-rolls its own harness list
|
|
29
|
+
* or column→profile mapping (the pattern that let those copies drift and silently
|
|
30
|
+
* break the harness pivot).
|
|
31
|
+
*
|
|
32
|
+
* Each cell clones `base`, sets the canonical top-level `harness` and `model.default`,
|
|
33
|
+
* and stamps `metadata.harness` + `metadata.harnessModel` for matrix grouping. Both
|
|
34
|
+
* metadata fields are hash-bearing, so every cell gets a distinct `agentProfileId` row
|
|
35
|
+
* and results join back by harness/model via {@link harnessAxisOf} with no
|
|
36
|
+
* hand-recomputed key. A vendor-locked harness snaps to its family's swept models — or
|
|
37
|
+
* its native default ({@link HARNESS_NATIVE_MODEL}) when it supports none — so every
|
|
38
|
+
* requested harness runs; `keepIncompatible` forces every pair verbatim.
|
|
39
|
+
*
|
|
40
|
+
* Omit `harnesses`/`models` to sweep the full default set — the "turn it on for
|
|
41
|
+
* everything we care about" switch, identical in shape whether one harness or all.
|
|
42
|
+
*/
|
|
43
|
+
function expandProfileAxes(spec) {
|
|
44
|
+
const harnesses = spec.harnesses ?? CODING_HARNESSES;
|
|
45
|
+
if (harnesses.length === 0) throw new ValidationError("expandProfileAxes: no harnesses to sweep");
|
|
46
|
+
const baseModel = spec.base.model?.default;
|
|
47
|
+
const models = spec.models ?? (baseModel ? [baseModel] : []);
|
|
48
|
+
if (models.length === 0) throw new ValidationError("expandProfileAxes: no models to sweep — base profile has no model.default and none were supplied");
|
|
49
|
+
const out = [];
|
|
50
|
+
const seen = /* @__PURE__ */ new Set();
|
|
51
|
+
for (const harness of harnesses) {
|
|
52
|
+
const supported = spec.keepIncompatible ? models : models.filter((model) => harnessSupportsModel(harness, model));
|
|
53
|
+
const effective = supported.length > 0 ? supported : [HARNESS_NATIVE_MODEL];
|
|
54
|
+
for (const model of effective) {
|
|
55
|
+
const profile = {
|
|
56
|
+
...spec.base,
|
|
57
|
+
harness,
|
|
58
|
+
name: `${spec.base.name ?? "agent"}/${harness}/${model}`,
|
|
59
|
+
model: {
|
|
60
|
+
...spec.base.model,
|
|
61
|
+
default: model
|
|
62
|
+
},
|
|
63
|
+
metadata: {
|
|
64
|
+
...spec.base.metadata ?? {},
|
|
65
|
+
harness,
|
|
66
|
+
harnessModel: model
|
|
67
|
+
}
|
|
68
|
+
};
|
|
69
|
+
const id = agentProfileId(profile);
|
|
70
|
+
if (seen.has(id)) continue;
|
|
71
|
+
seen.add(id);
|
|
72
|
+
out.push(profile);
|
|
73
|
+
}
|
|
74
|
+
}
|
|
75
|
+
if (out.length === 0) throw new ValidationError(`expandProfileAxes: produced no profiles (harnesses=[${harnesses.join(", ")}], models=[${models.join(", ")}]).`);
|
|
76
|
+
return out;
|
|
77
|
+
}
|
|
78
|
+
/**
|
|
79
|
+
* Read the (harness, model) a matrix cell ran under, off a profile or a result row's
|
|
80
|
+
* profile — the join-back for a `byHarness` pivot. Returns undefined when the profile
|
|
81
|
+
* wasn't produced by {@link expandProfileAxes}. Callers group `result.byProfile` by
|
|
82
|
+
* this instead of recomputing an id (recomputing the wrong key is what broke the pivot
|
|
83
|
+
* in the hand-rolled copies).
|
|
84
|
+
*/
|
|
85
|
+
function harnessAxisOf(profile) {
|
|
86
|
+
const m = profile.metadata;
|
|
87
|
+
const harness = m?.harness;
|
|
88
|
+
const model = m?.harnessModel;
|
|
89
|
+
if (typeof harness === "string" && typeof model === "string") return {
|
|
90
|
+
harness,
|
|
91
|
+
model
|
|
92
|
+
};
|
|
93
|
+
}
|
|
94
|
+
/**
|
|
95
|
+
* Collision-resistant, path-safe, human-readable profile id for eval artifacts.
|
|
96
|
+
* Scorecard joins still use `agentProfileHash`; this id is for run ids, matrix
|
|
97
|
+
* keys, and directory names where two profiles must not collapse onto one row.
|
|
98
|
+
* The suffix is the first 64 bits of the behaviour hash, enough for ordinary
|
|
99
|
+
* eval matrices while keeping filenames readable.
|
|
100
|
+
*/
|
|
101
|
+
function agentProfileId(profile) {
|
|
102
|
+
return `${pathSafeProfileLabel(agentProfileDisplayLabel(profile)) ?? "profile"}-${agentProfileHash(profile).slice(0, 16)}`;
|
|
103
|
+
}
|
|
104
|
+
/**
|
|
105
|
+
* Model snapshot used for `RunRecord.model`. Eval surfaces require a concrete
|
|
106
|
+
* model id because run records reject bare/missing model aliases.
|
|
107
|
+
*/
|
|
108
|
+
function agentProfileModelId(profile) {
|
|
109
|
+
const model = profile.model?.default?.trim();
|
|
110
|
+
if (!model) {
|
|
111
|
+
const label = agentProfileDisplayLabel(profile) ?? "unnamed profile";
|
|
112
|
+
throw new ValidationError(`AgentProfile "${label}" has no model.default — cannot record eval run`);
|
|
113
|
+
}
|
|
114
|
+
return model;
|
|
115
|
+
}
|
|
116
|
+
function agentProfileDisplayLabel(profile) {
|
|
117
|
+
return profile.name?.trim() || profile.version?.trim() || void 0;
|
|
118
|
+
}
|
|
119
|
+
function pathSafeProfileLabel(label) {
|
|
120
|
+
return label?.trim().replace(/[^A-Za-z0-9._-]+/g, "-").replace(/-+/g, "-").replace(/^-|-$/g, "") || void 0;
|
|
121
|
+
}
|
|
122
|
+
function compact(input) {
|
|
123
|
+
const out = {};
|
|
124
|
+
for (const [key, value] of Object.entries(input)) if (value !== void 0) out[key] = value;
|
|
125
|
+
return out;
|
|
126
|
+
}
|
|
127
|
+
/**
|
|
128
|
+
* Deterministic behaviour identity for the canonical
|
|
129
|
+
* `@tangle-network/agent-interface` AgentProfile.
|
|
130
|
+
*
|
|
131
|
+
* `name` and `description` are labels and do not affect the hash. Profile
|
|
132
|
+
* `version`, prompt, model hints, tools, resources, hooks, modes, permissions,
|
|
133
|
+
* and extensions do affect the hash. Resource array order is hash-bearing
|
|
134
|
+
* because mount order can change agent behaviour. Undefined fields are treated
|
|
135
|
+
* as absent; explicit `null` fields remain hash-bearing.
|
|
136
|
+
*/
|
|
137
|
+
function agentProfileHash(profile) {
|
|
138
|
+
const model = agentProfileModelId(profile);
|
|
139
|
+
const behaviour = compact({
|
|
140
|
+
...profile,
|
|
141
|
+
name: void 0,
|
|
142
|
+
description: void 0,
|
|
143
|
+
tags: profile.tags ? [...profile.tags].sort() : void 0,
|
|
144
|
+
model: compact({
|
|
145
|
+
...profile.model,
|
|
146
|
+
default: model
|
|
147
|
+
})
|
|
148
|
+
});
|
|
149
|
+
return createHash("sha256").update(canonicalString(behaviour)).digest("hex");
|
|
150
|
+
}
|
|
151
|
+
export { agentProfileModelId as a, agentProfileId as i, HARNESS_NATIVE_MODEL as n, expandProfileAxes as o, agentProfileHash as r, harnessAxisOf as s, CODING_HARNESSES as t };
|
|
152
|
+
|
|
153
|
+
//# sourceMappingURL=agent-profile-Dngk3LZ0.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agent-profile-Dngk3LZ0.js","names":[],"sources":["../src/agent-profile.ts"],"sourcesContent":["import { createHash } from 'node:crypto'\nimport type { AgentProfile } from '@tangle-network/agent-interface'\nimport { type HarnessType, harnessSupportsModel } from '@tangle-network/agent-interface'\nimport { ValidationError } from './errors'\nimport { canonicalString } from './ledger-core/canonical'\n\nexport type { AgentProfile, HarnessType } from '@tangle-network/agent-interface'\n\n/**\n * The agentic coding harnesses an eval sweeps by default — the ones we care about\n * ranking. This is the SINGLE source of that list; consumers import it instead of\n * re-declaring their own (a re-declared list is how the fleet drifts). Pass an\n * explicit `harnesses` (e.g. `harnessTypeSchema.options` for literally every known\n * harness) to widen beyond these.\n */\nexport const CODING_HARNESSES: readonly HarnessType[] = [\n 'opencode',\n 'claude-code',\n 'codex',\n 'kimi-code',\n]\n\nexport interface ProfileAxisSpec {\n /** The domain profile to sweep. Its prompt/tools/skills are held fixed; only the\n * harness and model vary. `model.default` is the fallback model. */\n base: AgentProfile\n /** Harnesses to cross. Default: {@link CODING_HARNESSES}. */\n harnesses?: readonly HarnessType[]\n /** Models to cross. Default: `[base.model.default]` — one model, i.e. today's\n * single-model behaviour, so omitting this never changes an existing run. */\n models?: readonly string[]\n /** Force every (harness, model) pair verbatim, even ones the harness can't run —\n * for deliberately testing failure modes. Default (false): SNAP instead — a\n * vendor-locked harness runs only the swept models in its family, or its native\n * default when it supports none, so no harness is dropped and none gets a\n * guaranteed-failing foreign-model cell. */\n keepIncompatible?: boolean\n}\n\n/** Model sentinel for a vendor-locked harness that supports none of the swept models:\n * it carries no provider prefix, so `harnessSupportsModel` accepts it and the harness\n * resolves it to its own native default model at runtime (e.g. kimi-code → its Kimi\n * model). Lets `expandProfileAxes` snap-instead-of-drop without a per-harness flagship\n * table that would rot as router catalogs change. */\nexport const HARNESS_NATIVE_MODEL = 'default'\n\n/**\n * Expand a base profile across the harness × model matrix into the `AgentProfile[]`\n * that `runProfileMatrix` / `selfImprove` score — the ONE place \"which harnesses ×\n * which models do we evaluate\" lives, so no product hand-rolls its own harness list\n * or column→profile mapping (the pattern that let those copies drift and silently\n * break the harness pivot).\n *\n * Each cell clones `base`, sets the canonical top-level `harness` and `model.default`,\n * and stamps `metadata.harness` + `metadata.harnessModel` for matrix grouping. Both\n * metadata fields are hash-bearing, so every cell gets a distinct `agentProfileId` row\n * and results join back by harness/model via {@link harnessAxisOf} with no\n * hand-recomputed key. A vendor-locked harness snaps to its family's swept models — or\n * its native default ({@link HARNESS_NATIVE_MODEL}) when it supports none — so every\n * requested harness runs; `keepIncompatible` forces every pair verbatim.\n *\n * Omit `harnesses`/`models` to sweep the full default set — the \"turn it on for\n * everything we care about\" switch, identical in shape whether one harness or all.\n */\nexport function expandProfileAxes(spec: ProfileAxisSpec): AgentProfile[] {\n const harnesses = spec.harnesses ?? CODING_HARNESSES\n if (harnesses.length === 0) throw new ValidationError('expandProfileAxes: no harnesses to sweep')\n const baseModel = spec.base.model?.default\n const models = spec.models ?? (baseModel ? [baseModel] : [])\n if (models.length === 0) {\n throw new ValidationError(\n 'expandProfileAxes: no models to sweep — base profile has no model.default and none were supplied',\n )\n }\n const out: AgentProfile[] = []\n const seen = new Set<string>()\n for (const harness of harnesses) {\n // A universal (router-backed) harness — opencode/pi/claudish, and claude-code/codex\n // since agent-interface 5.6.0 — runs every swept model. A vendor-locked harness\n // (kimi-code) runs only the swept models in its own family; when it supports NONE\n // of them it snaps to its native default (the `HARNESS_NATIVE_MODEL` sentinel it\n // resolves at runtime) rather than being dropped, so every requested harness still appears in the\n // sweep on a model it can actually run — e.g. sweeping `deepseek/x` puts opencode\n // on deepseek and kimi-code on its own Kimi model, a real head-to-head.\n // `keepIncompatible` forces every (harness, model) pair verbatim (failure-mode runs).\n const supported = spec.keepIncompatible\n ? models\n : models.filter((model) => harnessSupportsModel(harness, model))\n const effective = supported.length > 0 ? supported : [HARNESS_NATIVE_MODEL]\n for (const model of effective) {\n const profile: AgentProfile = {\n ...spec.base,\n harness,\n name: `${spec.base.name ?? 'agent'}/${harness}/${model}`,\n model: { ...spec.base.model, default: model },\n metadata: { ...(spec.base.metadata ?? {}), harness, harnessModel: model },\n }\n const id = agentProfileId(profile)\n if (seen.has(id)) continue\n seen.add(id)\n out.push(profile)\n }\n }\n if (out.length === 0) {\n // Unreachable in normal use — snapping guarantees ≥1 cell per harness — but keep a\n // fail-closed guard so a future refactor can't silently produce an empty sweep.\n throw new ValidationError(\n `expandProfileAxes: produced no profiles (harnesses=[${harnesses.join(', ')}], models=[${models.join(', ')}]).`,\n )\n }\n return out\n}\n\n/**\n * Read the (harness, model) a matrix cell ran under, off a profile or a result row's\n * profile — the join-back for a `byHarness` pivot. Returns undefined when the profile\n * wasn't produced by {@link expandProfileAxes}. Callers group `result.byProfile` by\n * this instead of recomputing an id (recomputing the wrong key is what broke the pivot\n * in the hand-rolled copies).\n */\nexport function harnessAxisOf(\n profile: Pick<AgentProfile, 'metadata'>,\n): { harness: HarnessType; model: string } | undefined {\n const m = profile.metadata as Record<string, unknown> | undefined\n const harness = m?.harness\n const model = m?.harnessModel\n if (typeof harness === 'string' && typeof model === 'string') {\n return { harness: harness as HarnessType, model }\n }\n return undefined\n}\n\n/**\n * Collision-resistant, path-safe, human-readable profile id for eval artifacts.\n * Scorecard joins still use `agentProfileHash`; this id is for run ids, matrix\n * keys, and directory names where two profiles must not collapse onto one row.\n * The suffix is the first 64 bits of the behaviour hash, enough for ordinary\n * eval matrices while keeping filenames readable.\n */\nexport function agentProfileId(profile: AgentProfile): string {\n const label = pathSafeProfileLabel(agentProfileDisplayLabel(profile)) ?? 'profile'\n return `${label}-${agentProfileHash(profile).slice(0, 16)}`\n}\n\n/**\n * Model snapshot used for `RunRecord.model`. Eval surfaces require a concrete\n * model id because run records reject bare/missing model aliases.\n */\nexport function agentProfileModelId(profile: AgentProfile): string {\n const model = profile.model?.default?.trim()\n if (!model) {\n const label = agentProfileDisplayLabel(profile) ?? 'unnamed profile'\n throw new ValidationError(\n `AgentProfile \"${label}\" has no model.default — cannot record eval run`,\n )\n }\n return model\n}\n\nfunction agentProfileDisplayLabel(profile: AgentProfile): string | undefined {\n return profile.name?.trim() || profile.version?.trim() || undefined\n}\n\nfunction pathSafeProfileLabel(label: string | undefined): string | undefined {\n const safe = label\n ?.trim()\n .replace(/[^A-Za-z0-9._-]+/g, '-')\n .replace(/-+/g, '-')\n .replace(/^-|-$/g, '')\n return safe || undefined\n}\n\nfunction compact<T extends Record<string, unknown>>(input: T): Partial<T> {\n const out: Record<string, unknown> = {}\n for (const [key, value] of Object.entries(input)) {\n if (value !== undefined) out[key] = value\n }\n return out as Partial<T>\n}\n\n/**\n * Deterministic behaviour identity for the canonical\n * `@tangle-network/agent-interface` AgentProfile.\n *\n * `name` and `description` are labels and do not affect the hash. Profile\n * `version`, prompt, model hints, tools, resources, hooks, modes, permissions,\n * and extensions do affect the hash. Resource array order is hash-bearing\n * because mount order can change agent behaviour. Undefined fields are treated\n * as absent; explicit `null` fields remain hash-bearing.\n */\nexport function agentProfileHash(profile: AgentProfile): string {\n const model = agentProfileModelId(profile)\n const behaviour = compact({\n ...profile,\n name: undefined,\n description: undefined,\n tags: profile.tags ? [...profile.tags].sort() : undefined,\n model: compact({ ...profile.model, default: model }),\n })\n return createHash('sha256').update(canonicalString(behaviour)).digest('hex')\n}\n"],"mappings":";;;;;;;;;;;;AAeA,MAAa,mBAA2C;CACtD;CACA;CACA;CACA;AACF;;;;;;AAwBA,MAAa,uBAAuB;;;;;;;;;;;;;;;;;;;AAoBpC,SAAgB,kBAAkB,MAAuC;CACvE,MAAM,YAAY,KAAK,aAAa;CACpC,IAAI,UAAU,WAAW,GAAG,MAAM,IAAI,gBAAgB,0CAA0C;CAChG,MAAM,YAAY,KAAK,KAAK,OAAO;CACnC,MAAM,SAAS,KAAK,WAAW,YAAY,CAAC,SAAS,IAAI,CAAC;CAC1D,IAAI,OAAO,WAAW,GACpB,MAAM,IAAI,gBACR,kGACF;CAEF,MAAM,MAAsB,CAAC;CAC7B,MAAM,uBAAO,IAAI,IAAY;CAC7B,KAAK,MAAM,WAAW,WAAW;EAS/B,MAAM,YAAY,KAAK,mBACnB,SACA,OAAO,QAAQ,UAAU,qBAAqB,SAAS,KAAK,CAAC;EACjE,MAAM,YAAY,UAAU,SAAS,IAAI,YAAY,CAAC,oBAAoB;EAC1E,KAAK,MAAM,SAAS,WAAW;GAC7B,MAAM,UAAwB;IAC5B,GAAG,KAAK;IACR;IACA,MAAM,GAAG,KAAK,KAAK,QAAQ,QAAQ,GAAG,QAAQ,GAAG;IACjD,OAAO;KAAE,GAAG,KAAK,KAAK;KAAO,SAAS;IAAM;IAC5C,UAAU;KAAE,GAAI,KAAK,KAAK,YAAY,CAAC;KAAI;KAAS,cAAc;IAAM;GAC1E;GACA,MAAM,KAAK,eAAe,OAAO;GACjC,IAAI,KAAK,IAAI,EAAE,GAAG;GAClB,KAAK,IAAI,EAAE;GACX,IAAI,KAAK,OAAO;EAClB;CACF;CACA,IAAI,IAAI,WAAW,GAGjB,MAAM,IAAI,gBACR,uDAAuD,UAAU,KAAK,IAAI,EAAE,aAAa,OAAO,KAAK,IAAI,EAAE,IAC7G;CAEF,OAAO;AACT;;;;;;;;AASA,SAAgB,cACd,SACqD;CACrD,MAAM,IAAI,QAAQ;CAClB,MAAM,UAAU,GAAG;CACnB,MAAM,QAAQ,GAAG;CACjB,IAAI,OAAO,YAAY,YAAY,OAAO,UAAU,UAClD,OAAO;EAAW;EAAwB;CAAM;AAGpD;;;;;;;;AASA,SAAgB,eAAe,SAA+B;CAE5D,OAAO,GADO,qBAAqB,yBAAyB,OAAO,CAAC,KAAK,UACzD,GAAG,iBAAiB,OAAO,CAAC,CAAC,MAAM,GAAG,EAAE;AAC1D;;;;;AAMA,SAAgB,oBAAoB,SAA+B;CACjE,MAAM,QAAQ,QAAQ,OAAO,SAAS,KAAK;CAC3C,IAAI,CAAC,OAAO;EACV,MAAM,QAAQ,yBAAyB,OAAO,KAAK;EACnD,MAAM,IAAI,gBACR,iBAAiB,MAAM,gDACzB;CACF;CACA,OAAO;AACT;AAEA,SAAS,yBAAyB,SAA2C;CAC3E,OAAO,QAAQ,MAAM,KAAK,KAAK,QAAQ,SAAS,KAAK,KAAK,KAAA;AAC5D;AAEA,SAAS,qBAAqB,OAA+C;CAM3E,OALa,OACT,KAAK,CAAC,CACP,QAAQ,qBAAqB,GAAG,CAAC,CACjC,QAAQ,OAAO,GAAG,CAAC,CACnB,QAAQ,UAAU,EAAE,KACR,KAAA;AACjB;AAEA,SAAS,QAA2C,OAAsB;CACxE,MAAM,MAA+B,CAAC;CACtC,KAAK,MAAM,CAAC,KAAK,UAAU,OAAO,QAAQ,KAAK,GAC7C,IAAI,UAAU,KAAA,GAAW,IAAI,OAAO;CAEtC,OAAO;AACT;;;;;;;;;;;AAYA,SAAgB,iBAAiB,SAA+B;CAC9D,MAAM,QAAQ,oBAAoB,OAAO;CACzC,MAAM,YAAY,QAAQ;EACxB,GAAG;EACH,MAAM,KAAA;EACN,aAAa,KAAA;EACb,MAAM,QAAQ,OAAO,CAAC,GAAG,QAAQ,IAAI,CAAC,CAAC,KAAK,IAAI,KAAA;EAChD,OAAO,QAAQ;GAAE,GAAG,QAAQ;GAAO,SAAS;EAAM,CAAC;CACrD,CAAC;CACD,OAAO,WAAW,QAAQ,CAAC,CAAC,OAAO,gBAAgB,SAAS,CAAC,CAAC,CAAC,OAAO,KAAK;AAC7E"}
|
|
@@ -1,6 +1,5 @@
|
|
|
1
|
-
import { c as ValidationError } from "./errors-
|
|
1
|
+
import { c as ValidationError } from "./errors-E9I2hldP.js";
|
|
2
2
|
import { AgentProfile } from "@tangle-network/agent-interface";
|
|
3
|
-
//#region src/agent-profile-cell.d.ts
|
|
4
3
|
type AgentProfileCellSchemaVersion = 'agent-profile-cell/v1';
|
|
5
4
|
type AgentProfileJsonObject = {
|
|
6
5
|
[key: string]: AgentProfileJson;
|
|
@@ -71,7 +70,7 @@ declare const AGENT_PROFILE_KINDS: {
|
|
|
71
70
|
/** A profile declared via `defineAgentProfile(...)` from
|
|
72
71
|
* `@tangle-network/agent-interface`. The default kind for router-backed
|
|
73
72
|
* and sandbox-backed products. */
|
|
74
|
-
readonly AGENT_INTERFACE_PROFILE:
|
|
73
|
+
readonly AGENT_INTERFACE_PROFILE: "agent-interface-profile";
|
|
75
74
|
};
|
|
76
75
|
type AgentProfileKind = (typeof AGENT_PROFILE_KINDS)[keyof typeof AGENT_PROFILE_KINDS];
|
|
77
76
|
/** Convert a profile's JSON document form using the shared canonical encoder.
|
|
@@ -83,6 +82,5 @@ type AgentInterfaceProfileLike = AgentProfile & {
|
|
|
83
82
|
name: string;
|
|
84
83
|
version: string;
|
|
85
84
|
};
|
|
86
|
-
//#endregion
|
|
87
85
|
export { toAgentProfileJson as _, AgentProfileCellSchemaVersion as a, AgentProfileHarness as c, AgentProfileKind as d, AgentProfileSource as f, groupRunsByAgentProfileCell as g, buildAgentProfileCell as h, AgentProfileCellInput as i, AgentProfileJson as l, agentProfileCellKey as m, AgentInterfaceProfileLike as n, AgentProfileCellValidationError as o, AgentProfileSourceInput as p, AgentProfileCell as r, AgentProfileDimensionValue as s, AGENT_PROFILE_KINDS as t, AgentProfileJsonObject as u, validateAgentProfileCell as v, verifyAgentProfileCell as y };
|
|
88
|
-
//# sourceMappingURL=agent-profile-cell-
|
|
86
|
+
//# sourceMappingURL=agent-profile-cell-3kYxUsK2.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agent-profile-cell-3kYxUsK2.d.ts","names":[],"sources":["../src/agent-profile-cell.ts"],"mappings":";;KAKY;KAEA;GAA4B,cAAc;;KAE1C,sDAKR,qBACA;KAEQ;UAEK;;EAEf;;EAEA;;UAGe;EACf;;EAEA;;EAEA,UAAU;;UAGK;EACf;EACA;EACA;;UAGe;EACf;EACA,eAAe;EACf,UAAU;EACV;EACA;EACA,aAAa,eAAe;;UAGb;EACf,eAAe;EACf;EACA;EACA,eAAe;EACf,UAAU;EACV;EACA;EACA,aAAa,eAAe;;cAGjB,wCAAwC;WAC1C;EACG,YAAA,iBAAiB;;iBAWT,sBACpB,OAAO,wBACN,QAAQ;;;;;;iBAiBW,uBAAuB,MAAM,mBAAmB;iBAMtD,yBAAyB,iBAAiB;iBAkC1C,oBAAoB;EAClC;EACA,eAAe;;iBAOD,4BACd;EAAY;EAAe,eAAe;GAC1C,kBAAkB,MAAM,YAAY;;;;;cA0NzB;;;;;;KAOD,2BAA2B,kCAAkC;;;;iBAKzD,mBAAmB,iBAAiB;;KAYxC,4BAA4B;EAAiB;EAAc"}
|
|
@@ -1,6 +1,6 @@
|
|
|
1
|
-
import { s as ValidationError } from "./errors-
|
|
2
|
-
import { a as hashCanonical, o as jsonDocument, r as canonicalString } from "./canonical-
|
|
3
|
-
|
|
1
|
+
import { s as ValidationError } from "./errors-DeEEV9Vx.js";
|
|
2
|
+
import { a as hashCanonical, o as jsonDocument, r as canonicalString } from "./canonical-CicJlINA.js";
|
|
3
|
+
// src/pre-registration.ts
|
|
4
4
|
/**
|
|
5
5
|
* Pre-registered hypotheses — declare what you're testing BEFORE the
|
|
6
6
|
* run, check it AFTER. Prevents p-hacking, optional stopping, and the
|
|
@@ -68,8 +68,7 @@ async function verifyManifest(m) {
|
|
|
68
68
|
return false;
|
|
69
69
|
}
|
|
70
70
|
}
|
|
71
|
-
|
|
72
|
-
//#region src/agent-profile-cell.ts
|
|
71
|
+
// src/agent-profile-cell.ts
|
|
73
72
|
var AgentProfileCellValidationError = class extends ValidationError {
|
|
74
73
|
path;
|
|
75
74
|
constructor(message, path = "") {
|
|
@@ -283,7 +282,6 @@ function toAgentProfileJson(value) {
|
|
|
283
282
|
throw new AgentProfileCellValidationError(`agent profile must be JSON-serializable: ${err instanceof Error ? err.message : String(err)}`, "sourceProfile.profile");
|
|
284
283
|
}
|
|
285
284
|
}
|
|
286
|
-
//#endregion
|
|
287
285
|
export { groupRunsByAgentProfileCell as a, verifyAgentProfileCell as c, signManifest as d, verifyManifest as f, buildAgentProfileCell as i, hashJson as l, AgentProfileCellValidationError as n, toAgentProfileJson as o, agentProfileCellKey as r, validateAgentProfileCell as s, AGENT_PROFILE_KINDS as t, manifestContentDigest as u };
|
|
288
286
|
|
|
289
|
-
//# sourceMappingURL=agent-profile-cell-
|
|
287
|
+
//# sourceMappingURL=agent-profile-cell-CT5JYueM.js.map
|