@tangle-network/agent-eval 0.187.2 → 0.189.0
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 +95 -0
- package/dist/adapters/http.d.ts +2 -2
- package/dist/{agent-profile-uAIQK3Nr.d.ts → agent-profile-DNV_U6QR.d.ts} +4 -4
- package/dist/{agent-profile-uAIQK3Nr.d.ts.map → agent-profile-DNV_U6QR.d.ts.map} +1 -1
- package/dist/analyst/index.d.ts +12 -12
- package/dist/analyst/index.d.ts.map +1 -1
- package/dist/analyst/index.js +7 -7
- package/dist/{analyze-runs-DapYZz02.d.ts → analyze-runs-BvK4-wSs.d.ts} +5 -5
- package/dist/{analyze-runs-DapYZz02.d.ts.map → analyze-runs-BvK4-wSs.d.ts.map} +1 -1
- package/dist/{analyze-runs-DCY7LMt4.js → analyze-runs-SCSWYpy0.js} +17 -9
- package/dist/analyze-runs-SCSWYpy0.js.map +1 -0
- package/dist/{benchmark-command-DGokT8zd.js → benchmark-command-BhhAYstU.js} +14 -14
- package/dist/benchmark-command-BhhAYstU.js.map +1 -0
- package/dist/benchmarks/index.d.ts +4 -4
- package/dist/benchmarks/index.js +3 -3
- package/dist/builder-eval/index.d.ts +3 -3
- package/dist/builder-eval/index.js +1 -1
- package/dist/campaign/index.d.ts +5 -5
- package/dist/campaign/index.js +7 -7
- package/dist/{campaign-DlxdTOvw.js → campaign-BWoSQPSU.js} +10 -10
- package/dist/{campaign-DlxdTOvw.js.map → campaign-BWoSQPSU.js.map} +1 -1
- package/dist/{campaign-evidence-IF1p0oDa.js → campaign-evidence-CBlGJ3kC.js} +5 -5
- package/dist/campaign-evidence-CBlGJ3kC.js.map +1 -0
- package/dist/{chat-client-DqPqWzVV.js → chat-client-C4dg7Ng_.js} +2 -2
- package/dist/{chat-client-DqPqWzVV.js.map → chat-client-C4dg7Ng_.js.map} +1 -1
- package/dist/{chat-json-call-C-DnDxEy.js → chat-json-call-JVGbfx95.js} +2 -2
- package/dist/{chat-json-call-C-DnDxEy.js.map → chat-json-call-JVGbfx95.js.map} +1 -1
- package/dist/cli.js +4 -4
- package/dist/{client-DCc7a8ZC.d.ts → client-BhEyAdfe.d.ts} +4 -4
- package/dist/{client-DCc7a8ZC.d.ts.map → client-BhEyAdfe.d.ts.map} +1 -1
- package/dist/{client-CXE-U1SA.js → client-QwJpsNYU.js} +5 -1
- package/dist/client-QwJpsNYU.js.map +1 -0
- package/dist/contract/index.d.ts +11 -11
- package/dist/contract/index.js +22 -13
- package/dist/contract/index.js.map +1 -1
- package/dist/{counterfactual-CLgrwhkY.d.ts → counterfactual-CPaWN99-.d.ts} +4 -4
- package/dist/{counterfactual-CLgrwhkY.d.ts.map → counterfactual-CPaWN99-.d.ts.map} +1 -1
- package/dist/{counterfactual-Bjq1mlUu.js → counterfactual-DtMWvn7D.js} +2 -2
- package/dist/{counterfactual-Bjq1mlUu.js.map → counterfactual-DtMWvn7D.js.map} +1 -1
- package/dist/deep-freeze-CurcOQg3.js +16 -0
- package/dist/deep-freeze-CurcOQg3.js.map +1 -0
- package/dist/{default-registry-DMxeb4GD.d.ts → default-registry-B5HrwrPM.d.ts} +6 -6
- package/dist/{default-registry-DMxeb4GD.d.ts.map → default-registry-B5HrwrPM.d.ts.map} +1 -1
- package/dist/{default-registry-Ca2IWH1X.js → default-registry-m2jhbG6Q.js} +4 -3
- package/dist/default-registry-m2jhbG6Q.js.map +1 -0
- package/dist/{define-agent-eval-C0EqJ9Gz.js → define-agent-eval-98sgUaC1.js} +8 -8
- package/dist/{define-agent-eval-C0EqJ9Gz.js.map → define-agent-eval-98sgUaC1.js.map} +1 -1
- package/dist/{define-agent-eval-CGBwPPaq.d.ts → define-agent-eval-q6890HU4.d.ts} +7 -7
- package/dist/{define-agent-eval-CGBwPPaq.d.ts.map → define-agent-eval-q6890HU4.d.ts.map} +1 -1
- package/dist/diagnosis/index.d.ts +38 -96
- package/dist/diagnosis/index.d.ts.map +1 -1
- package/dist/diagnosis/index.js +84 -135
- package/dist/diagnosis/index.js.map +1 -1
- package/dist/{dspy-rlm-engine-BE9lCzzL.js → dspy-rlm-engine-DLYttKBm.js} +3 -3
- package/dist/{dspy-rlm-engine-BE9lCzzL.js.map → dspy-rlm-engine-DLYttKBm.js.map} +1 -1
- package/dist/emitter-DbMXUIVP.js +334 -0
- package/dist/emitter-DbMXUIVP.js.map +1 -0
- package/dist/{emitter-Cs0egaFd.d.ts → emitter-HcCmFQl_.d.ts} +37 -6
- package/dist/emitter-HcCmFQl_.d.ts.map +1 -0
- package/dist/{engine-DNlKhIm0.d.ts → engine-BkLCnZQX.d.ts} +15 -5
- package/dist/engine-BkLCnZQX.d.ts.map +1 -0
- package/dist/{eval-campaign-XFHEpN86.js → eval-campaign-C0Q1Eeo_.js} +8 -5
- package/dist/{eval-campaign-XFHEpN86.js.map → eval-campaign-C0Q1Eeo_.js.map} +1 -1
- package/dist/evaluation.d.ts +2 -2
- package/dist/{exact-types-BEHN7tnq.d.ts → exact-types-CfpP2A4Z.d.ts} +2 -2
- package/dist/{exact-types-BEHN7tnq.d.ts.map → exact-types-CfpP2A4Z.d.ts.map} +1 -1
- package/dist/experiment/index.d.ts +3 -3
- package/dist/experiment/index.js +2 -3
- package/dist/experiment/index.js.map +1 -1
- package/dist/{external-optimizer-process-BbhvXTFx.js → external-optimizer-process-C3quNeKV.js} +3 -3
- package/dist/{external-optimizer-process-BbhvXTFx.js.map → external-optimizer-process-C3quNeKV.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-DMgmqTB6.js → external-optimizer-subprocess-BDwgKLO1.js} +3 -3
- package/dist/{external-optimizer-subprocess-DMgmqTB6.js.map → external-optimizer-subprocess-BDwgKLO1.js.map} +1 -1
- package/dist/{failure-cluster-OldNRoAt.d.ts → failure-cluster-CbgEDcFJ.d.ts} +3 -3
- package/dist/{failure-cluster-OldNRoAt.d.ts.map → failure-cluster-CbgEDcFJ.d.ts.map} +1 -1
- package/dist/failure-taxonomy-CjabTATj.js +538 -0
- package/dist/failure-taxonomy-CjabTATj.js.map +1 -0
- package/dist/{feedback-trajectory-CytTR7u4.d.ts → feedback-trajectory-B3ZSkfAq.d.ts} +6 -6
- package/dist/{feedback-trajectory-CytTR7u4.d.ts.map → feedback-trajectory-B3ZSkfAq.d.ts.map} +1 -1
- package/dist/hosted/index.d.ts +2 -2
- package/dist/hosted/index.d.ts.map +1 -1
- package/dist/hosted/index.js +1 -1
- package/dist/{index-vYNHCV2G.d.ts → index-2e_MTznj.d.ts} +8 -8
- package/dist/{index-vYNHCV2G.d.ts.map → index-2e_MTznj.d.ts.map} +1 -1
- package/dist/{index-CKZa_rnt.d.ts → index-Bcc9PkOz.d.ts} +3 -3
- package/dist/{index-CKZa_rnt.d.ts.map → index-Bcc9PkOz.d.ts.map} +1 -1
- package/dist/{index-u0d1Jp4F.d.ts → index-BqDJ6fUK.d.ts} +45 -24
- package/dist/index-BqDJ6fUK.d.ts.map +1 -0
- package/dist/{index-BRsrfSHT.d.ts → index-CnJOPR_E.d.ts} +10 -10
- package/dist/{index-BRsrfSHT.d.ts.map → index-CnJOPR_E.d.ts.map} +1 -1
- package/dist/{index-YZrbnBSV.d.ts → index-Dk1dOUfm.d.ts} +11 -11
- package/dist/{index-YZrbnBSV.d.ts.map → index-Dk1dOUfm.d.ts.map} +1 -1
- package/dist/index.d.ts +384 -69
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +1492 -217
- package/dist/index.js.map +1 -1
- package/dist/{insight-report-DpOo-3iL.d.ts → insight-report-CvPkEIcC.d.ts} +14 -7
- package/dist/insight-report-CvPkEIcC.d.ts.map +1 -0
- package/dist/{integrity-Cy9WHAtb.js → integrity-BFP4kjGj.js} +6 -1
- package/dist/integrity-BFP4kjGj.js.map +1 -0
- package/dist/{integrity-rGOfSUle.d.ts → integrity-CaeNeusS.d.ts} +4 -4
- package/dist/integrity-CaeNeusS.d.ts.map +1 -0
- package/dist/jev.d.ts +2 -2
- package/dist/jev.js +2 -1
- package/dist/jev.js.map +1 -1
- package/dist/{journal-BuRLm6ho.js → journal-CDP0Vhx_.js} +262 -115
- package/dist/journal-CDP0Vhx_.js.map +1 -0
- package/dist/{kind-factory-V79Hw0AU.js → kind-factory-BGwN0nE6.js} +17 -5
- package/dist/{kind-factory-V79Hw0AU.js.map → kind-factory-BGwN0nE6.js.map} +1 -1
- package/dist/ledger-core/index.d.ts +2 -2
- package/dist/ledger-core/index.js +1 -1
- package/dist/{llm-client-DDhZ29Vm.js → llm-client-DF9bPNYA.js} +2 -2
- package/dist/{llm-client-DDhZ29Vm.js.map → llm-client-DF9bPNYA.js.map} +1 -1
- package/dist/{llm-judge-BVso5x-u.js → llm-judge-M_l_WXw2.js} +26 -24
- package/dist/llm-judge-M_l_WXw2.js.map +1 -0
- package/dist/{matrix-CPKsRY8Y.d.ts → matrix-CNypzP-v.d.ts} +2 -2
- package/dist/{matrix-CPKsRY8Y.d.ts.map → matrix-CNypzP-v.d.ts.map} +1 -1
- package/dist/meta-eval/index.d.ts +4 -4
- package/dist/{mint-BI14xehJ.d.ts → mint-DsS38v5c.d.ts} +25 -5
- package/dist/{mint-BI14xehJ.d.ts.map → mint-DsS38v5c.d.ts.map} +1 -1
- package/dist/{mint-B4zgSAJW.js → mint-iCD-W43s.js} +30 -12
- package/dist/mint-iCD-W43s.js.map +1 -0
- package/dist/multishot/golden/index.d.ts +1 -1
- package/dist/multishot/index.d.ts +2 -2
- package/dist/openapi.json +1 -1
- package/dist/pipelines/index.d.ts +67 -12
- package/dist/pipelines/index.d.ts.map +1 -1
- package/dist/pipelines/index.js +178 -41
- package/dist/pipelines/index.js.map +1 -1
- package/dist/{prime-protocol-CDBcL9NM.d.ts → prime-protocol-Fwvctmhm.d.ts} +2 -2
- package/dist/{prime-protocol-CDBcL9NM.d.ts.map → prime-protocol-Fwvctmhm.d.ts.map} +1 -1
- package/dist/{produced-state-Cn9ECr4v.js → produced-state-BOY98IJO.js} +4 -4
- package/dist/{produced-state-Cn9ECr4v.js.map → produced-state-BOY98IJO.js.map} +1 -1
- package/dist/{query-D6W6MaGx.d.ts → query-DBUQhehH.d.ts} +3 -3
- package/dist/{query-D6W6MaGx.d.ts.map → query-DBUQhehH.d.ts.map} +1 -1
- package/dist/random-DGEyr3zG.d.ts +83 -0
- package/dist/random-DGEyr3zG.d.ts.map +1 -0
- package/dist/{raw-provider-sink-BQd7mzyT.js → raw-provider-sink-C96CcqeG.js} +19 -56
- package/dist/raw-provider-sink-C96CcqeG.js.map +1 -0
- package/dist/redact-DPoH1-OW.d.ts +158 -0
- package/dist/redact-DPoH1-OW.d.ts.map +1 -0
- package/dist/redact-DxwlUAvh.js +824 -0
- package/dist/redact-DxwlUAvh.js.map +1 -0
- package/dist/{registry-D5Ly3GRV.d.ts → registry-mAQVtuLx.d.ts} +4 -4
- package/dist/{registry-D5Ly3GRV.d.ts.map → registry-mAQVtuLx.d.ts.map} +1 -1
- package/dist/{release-confidence-CSp3MsU4.d.ts → release-confidence-CkoXFx9I.d.ts} +4 -4
- package/dist/{release-confidence-CSp3MsU4.d.ts.map → release-confidence-CkoXFx9I.d.ts.map} +1 -1
- package/dist/{release-confidence-RanRUXRW.js → release-confidence-XQm_xjhR.js} +3 -3
- package/dist/release-confidence-XQm_xjhR.js.map +1 -0
- package/dist/{report-command-CSTjzFGY.js → report-command-DKJ_yAIj.js} +21 -12
- package/dist/{report-command-CSTjzFGY.js.map → report-command-DKJ_yAIj.js.map} +1 -1
- package/dist/reporting.d.ts +3 -3
- package/dist/reporting.js +1 -1
- package/dist/{researcher-D7E92STT.d.ts → researcher-DojEMb7E.d.ts} +7 -7
- package/dist/{researcher-D7E92STT.d.ts.map → researcher-DojEMb7E.d.ts.map} +1 -1
- package/dist/{reward-hacking-BIQ2BWLG.d.ts → reward-hacking-Cyk-QHk9.d.ts} +2 -2
- package/dist/{reward-hacking-BIQ2BWLG.d.ts.map → reward-hacking-Cyk-QHk9.d.ts.map} +1 -1
- package/dist/rl.d.ts +11 -10
- package/dist/rl.d.ts.map +1 -1
- package/dist/rl.js +7 -6
- package/dist/rl.js.map +1 -1
- package/dist/rollout/index.d.ts +4 -4
- package/dist/rollout/index.js +2 -2
- package/dist/{rollout-C0Ifk3vK.js → rollout-B-XBReTC.js} +2 -2
- package/dist/{rollout-C0Ifk3vK.js.map → rollout-B-XBReTC.js.map} +1 -1
- package/dist/{rubric-predictive-validity-CTU8y8Cv.d.ts → rubric-predictive-validity-DIUwJJ9Y.d.ts} +2 -2
- package/dist/{rubric-predictive-validity-CTU8y8Cv.d.ts.map → rubric-predictive-validity-DIUwJJ9Y.d.ts.map} +1 -1
- package/dist/{run-record-tg_5H2IK.js → run-record-BRlM99rX.js} +41 -3
- package/dist/run-record-BRlM99rX.js.map +1 -0
- package/dist/{run-record-D9wS35fK.d.ts → run-record-DF-fuu31.d.ts} +59 -7
- package/dist/run-record-DF-fuu31.d.ts.map +1 -0
- package/dist/{run-record-CNNzWGpB.js → run-record-DnMFyg_g.js} +26 -13
- package/dist/run-record-DnMFyg_g.js.map +1 -0
- package/dist/{schema-CR5cpjQ3.d.ts → schema-2g7trUwH.d.ts} +17 -2
- package/dist/schema-2g7trUwH.d.ts.map +1 -0
- package/dist/{schema-BzWDXhOR.d.ts → schema-B5LtYvn9.d.ts} +7 -5
- package/dist/{schema-BzWDXhOR.d.ts.map → schema-B5LtYvn9.d.ts.map} +1 -1
- package/dist/schema-C1aaAxTf.js.map +1 -1
- package/dist/schema-CSf6qWgZ.js.map +1 -1
- package/dist/{semantic-concept-judge-Cnt0TWp0.js → semantic-concept-judge-Cm2K2wAo.js} +6 -6
- package/dist/{semantic-concept-judge-Cnt0TWp0.js.map → semantic-concept-judge-Cm2K2wAo.js.map} +1 -1
- package/dist/{series-convergence-BnMs_uAr.d.ts → series-convergence-DY7ha0kx.d.ts} +2 -2
- package/dist/{series-convergence-BnMs_uAr.d.ts.map → series-convergence-DY7ha0kx.d.ts.map} +1 -1
- package/dist/{server-DmpcFiVK.js → server-C8XvqCg8.js} +2 -2
- package/dist/{server-DmpcFiVK.js.map → server-C8XvqCg8.js.map} +1 -1
- package/dist/{skillopt-optimization-method-Cc9SVLqi.js → skillopt-optimization-method-BVSB676o.js} +7 -7
- package/dist/{skillopt-optimization-method-Cc9SVLqi.js.map → skillopt-optimization-method-BVSB676o.js.map} +1 -1
- package/dist/spans-W0o8pbEg.d.ts +60 -0
- package/dist/spans-W0o8pbEg.d.ts.map +1 -0
- package/dist/{statistical-heldout-DTmhTW8P.d.ts → statistical-heldout-plAHG82H.d.ts} +3 -3
- package/dist/{statistical-heldout-DTmhTW8P.d.ts.map → statistical-heldout-plAHG82H.d.ts.map} +1 -1
- package/dist/{store-BErPvYBr.d.ts → store-C4avFjIi.d.ts} +2 -2
- package/dist/{store-BErPvYBr.d.ts.map → store-C4avFjIi.d.ts.map} +1 -1
- package/dist/{store-otlp-2gYCy-v0.js → store-otlp-D0AV4cvB.js} +2 -2
- package/dist/{store-otlp-2gYCy-v0.js.map → store-otlp-D0AV4cvB.js.map} +1 -1
- package/dist/{store-tool-spans-B3bX9yqn.d.ts → store-tool-spans-8PAyhRqb.d.ts} +9 -54
- package/dist/store-tool-spans-8PAyhRqb.d.ts.map +1 -0
- package/dist/{store-tool-spans-B1e6Auc2.js → store-tool-spans-B_0mXvYl.js} +4 -4
- package/dist/{store-tool-spans-B1e6Auc2.js.map → store-tool-spans-B_0mXvYl.js.map} +1 -1
- package/dist/storyboard/index.d.ts +1 -1
- package/dist/{summary-report-Iy6kXo8x.d.ts → summary-report-DvvesgR_.d.ts} +6 -5
- package/dist/summary-report-DvvesgR_.d.ts.map +1 -0
- package/dist/supervisor-run/index.d.ts +2 -2
- package/dist/supervisor-run/index.d.ts.map +1 -1
- package/dist/supervisor-run/index.js +1 -1
- package/dist/{tool-groups-veQYPiTy.d.ts → tool-groups-W7MJobX_.d.ts} +3 -3
- package/dist/tool-groups-W7MJobX_.d.ts.map +1 -0
- package/dist/{tool-waste-nbHPAxei.js → tool-waste-CJDfxH45.js} +3 -537
- package/dist/tool-waste-CJDfxH45.js.map +1 -0
- package/dist/{tool-waste-D23I0zWm.d.ts → tool-waste-CP5DO383.d.ts} +4 -4
- package/dist/{tool-waste-D23I0zWm.d.ts.map → tool-waste-CP5DO383.d.ts.map} +1 -1
- package/dist/trace-repair/index.d.ts +2 -2
- package/dist/traces.d.ts +12 -11
- package/dist/traces.d.ts.map +1 -1
- package/dist/traces.js +9 -9
- package/dist/{trajectory-D7qrNvaN.d.ts → trajectory-Y1fmdlw7.d.ts} +3 -3
- package/dist/{trajectory-D7qrNvaN.d.ts.map → trajectory-Y1fmdlw7.d.ts.map} +1 -1
- package/dist/trajectory-replay/index.d.ts +3 -3
- package/dist/trajectory-replay/index.js +2 -2
- package/dist/{types-BjB2wZM9.d.ts → types-Bay1dh2o.d.ts} +6 -6
- package/dist/{types-BjB2wZM9.d.ts.map → types-Bay1dh2o.d.ts.map} +1 -1
- package/dist/{types-BTRcIqFr.d.ts → types-BvmWNjjs.d.ts} +4 -4
- package/dist/{types-BTRcIqFr.d.ts.map → types-BvmWNjjs.d.ts.map} +1 -1
- package/dist/{types-CBbLtr2J.d.ts → types-C0bWoSKC.d.ts} +10 -11
- package/dist/types-C0bWoSKC.d.ts.map +1 -0
- package/dist/{types-C8VBbiw6.js → types-CW5b90JH.js} +2 -15
- package/dist/types-CW5b90JH.js.map +1 -0
- package/dist/{types--18oHwVF.d.ts → types-Cvqy7asY.d.ts} +3 -3
- package/dist/{types--18oHwVF.d.ts.map → types-Cvqy7asY.d.ts.map} +1 -1
- package/dist/{types-Tp5WhYL9.d.ts → types-DRNlLbBR.d.ts} +2 -2
- package/dist/{types-Tp5WhYL9.d.ts.map → types-DRNlLbBR.d.ts.map} +1 -1
- package/dist/wire/index.d.ts +3 -3
- package/dist/wire/index.js +1 -1
- package/docs/code-agent-intake.md +1 -1
- package/docs/diagnosis.md +2 -3
- package/docs/eval-surface-map.md +1 -1
- package/docs/experiment.md +0 -1
- package/docs/insight-report.md +3 -2
- package/docs/public-api.md +2 -2
- package/docs/redaction.md +59 -0
- package/docs/rollout.md +5 -5
- package/docs/search-history-receipts.md +1 -0
- package/docs/trace-contracts.md +102 -0
- package/docs/verdicts.md +1 -1
- package/package.json +3 -2
- package/dist/analyze-runs-DCY7LMt4.js.map +0 -1
- package/dist/benchmark-command-DGokT8zd.js.map +0 -1
- package/dist/campaign-evidence-IF1p0oDa.js.map +0 -1
- package/dist/client-CXE-U1SA.js.map +0 -1
- package/dist/default-registry-Ca2IWH1X.js.map +0 -1
- package/dist/emitter-Cs0egaFd.d.ts.map +0 -1
- package/dist/emitter-DeQHiDMm.js +0 -257
- package/dist/emitter-DeQHiDMm.js.map +0 -1
- package/dist/engine-DNlKhIm0.d.ts.map +0 -1
- package/dist/experiment-tracker-BKEumQug.js +0 -280
- package/dist/experiment-tracker-BKEumQug.js.map +0 -1
- package/dist/experiment-tracker-URRGP9ek.d.ts +0 -237
- package/dist/experiment-tracker-URRGP9ek.d.ts.map +0 -1
- package/dist/index-u0d1Jp4F.d.ts.map +0 -1
- package/dist/insight-report-DpOo-3iL.d.ts.map +0 -1
- package/dist/integrity-Cy9WHAtb.js.map +0 -1
- package/dist/integrity-rGOfSUle.d.ts.map +0 -1
- package/dist/journal-BuRLm6ho.js.map +0 -1
- package/dist/llm-judge-BVso5x-u.js.map +0 -1
- package/dist/mint-B4zgSAJW.js.map +0 -1
- package/dist/raw-provider-sink-BQd7mzyT.js.map +0 -1
- package/dist/redact-7Aq1ukl-.js +0 -107
- package/dist/redact-7Aq1ukl-.js.map +0 -1
- package/dist/release-confidence-RanRUXRW.js.map +0 -1
- package/dist/run-record-CNNzWGpB.js.map +0 -1
- package/dist/run-record-D9wS35fK.d.ts.map +0 -1
- package/dist/run-record-tg_5H2IK.js.map +0 -1
- package/dist/schema-CR5cpjQ3.d.ts.map +0 -1
- package/dist/store-tool-spans-B3bX9yqn.d.ts.map +0 -1
- package/dist/summary-report-Iy6kXo8x.d.ts.map +0 -1
- package/dist/tool-groups-veQYPiTy.d.ts.map +0 -1
- package/dist/tool-waste-nbHPAxei.js.map +0 -1
- package/dist/types-C8VBbiw6.js.map +0 -1
- package/dist/types-CBbLtr2J.d.ts.map +0 -1
package/dist/reporting.d.ts
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { t as benjaminiHochberg } from "./multiplicity-DIWHvysC.js";
|
|
2
2
|
import { d as PairedBootstrapResult, h as pairedBootstrap, u as PairedBootstrapOptions } from "./paired-promotion-decision-DPKz8ZnF.js";
|
|
3
|
-
import { _ as bootstrapCi, a as ReleaseConfidenceIssue, c as ReleaseConfidenceStatus, d as assertReleaseConfidence, f as evaluateReleaseConfidence, g as Verdict, h as JudgeReplayGateArgs, i as ReleaseConfidenceInput, k as wilcoxonSignedRank, l as ReleaseConfidenceThresholds, m as BootstrapResult, n as ReleaseConfidenceAxis, o as ReleaseConfidenceMetrics, p as BootstrapOptions, r as ReleaseConfidenceAxisName, s as ReleaseConfidenceScorecard, u as ReleaseTraceEvidence, v as judgeReplayGate } from "./release-confidence-
|
|
4
|
-
import { _ as paretoChart, a as ParetoPoint, c as ResearchReportCandidate, d as ResearchReportOptions, f as ResearchReportRecommendation, g as gainHistogram, h as SummaryTableRow, i as ParetoFigureSpec, l as ResearchReportDecision, m as SummaryTableOptions, n as GainDistributionFigureSpec, o as RESEARCH_REPORT_HARD_PAIR_FLOOR, p as SummaryTable, r as GainDistributionOptions, s as ResearchReport, t as GainDistributionBin, u as ResearchReportMethodology, v as researchReport, y as summaryTable } from "./summary-report-
|
|
3
|
+
import { _ as bootstrapCi, a as ReleaseConfidenceIssue, c as ReleaseConfidenceStatus, d as assertReleaseConfidence, f as evaluateReleaseConfidence, g as Verdict, h as JudgeReplayGateArgs, i as ReleaseConfidenceInput, k as wilcoxonSignedRank, l as ReleaseConfidenceThresholds, m as BootstrapResult, n as ReleaseConfidenceAxis, o as ReleaseConfidenceMetrics, p as BootstrapOptions, r as ReleaseConfidenceAxisName, s as ReleaseConfidenceScorecard, u as ReleaseTraceEvidence, v as judgeReplayGate } from "./release-confidence-CkoXFx9I.js";
|
|
4
|
+
import { _ as paretoChart, a as ParetoPoint, c as ResearchReportCandidate, d as ResearchReportOptions, f as ResearchReportRecommendation, g as gainHistogram, h as SummaryTableRow, i as ParetoFigureSpec, l as ResearchReportDecision, m as SummaryTableOptions, n as GainDistributionFigureSpec, o as RESEARCH_REPORT_HARD_PAIR_FLOOR, p as SummaryTable, r as GainDistributionOptions, s as ResearchReport, t as GainDistributionBin, u as ResearchReportMethodology, v as researchReport, y as summaryTable } from "./summary-report-DvvesgR_.js";
|
|
5
5
|
import { a as PairedEvalueStep, c as SequentialDecision, i as PairedEvalueSequence, l as evaluateInterimReleaseConfidence, n as InterimReleaseConfidenceInput, r as PairedEvalueOptions, t as InterimReleaseConfidence, u as pairedEvalueSequence } from "./sequential-BhsrMupG.js";
|
|
6
|
-
import { a as RubricPredictiveValidityReport, i as RubricPredictiveValidityInput, o as RubricRanking, r as RubricOutcomePair, s as rubricPredictiveValidity } from "./rubric-predictive-validity-
|
|
6
|
+
import { a as RubricPredictiveValidityReport, i as RubricPredictiveValidityInput, o as RubricRanking, r as RubricOutcomePair, s as rubricPredictiveValidity } from "./rubric-predictive-validity-DIUwJJ9Y.js";
|
|
7
7
|
export { type BootstrapOptions, type BootstrapResult, type GainDistributionBin, type GainDistributionFigureSpec, type GainDistributionOptions, type InterimReleaseConfidence, type InterimReleaseConfidenceInput, type JudgeReplayGateArgs, type PairedBootstrapOptions, type PairedBootstrapResult, type PairedEvalueOptions, type PairedEvalueSequence, type PairedEvalueStep, type ParetoFigureSpec, type ParetoPoint, RESEARCH_REPORT_HARD_PAIR_FLOOR, type ReleaseConfidenceAxis, type ReleaseConfidenceAxisName, type ReleaseConfidenceInput, type ReleaseConfidenceIssue, type ReleaseConfidenceMetrics, type ReleaseConfidenceScorecard, type ReleaseConfidenceStatus, type ReleaseConfidenceThresholds, type ReleaseTraceEvidence, type ResearchReport, type ResearchReportCandidate, type ResearchReportDecision, type ResearchReportMethodology, type ResearchReportOptions, type ResearchReportRecommendation, type RubricOutcomePair, type RubricPredictiveValidityInput, type RubricPredictiveValidityReport, type RubricRanking, type SequentialDecision, type SummaryTable, type SummaryTableOptions, type SummaryTableRow, type Verdict, assertReleaseConfidence, benjaminiHochberg, bootstrapCi, evaluateInterimReleaseConfidence, evaluateReleaseConfidence, gainHistogram, judgeReplayGate, pairedBootstrap, pairedEvalueSequence, paretoChart, researchReport, rubricPredictiveValidity, summaryTable, wilcoxonSignedRank };
|
package/dist/reporting.js
CHANGED
|
@@ -3,6 +3,6 @@ import { l as wilcoxonSignedRank } from "./paired-arms-JSDcs2zx.js";
|
|
|
3
3
|
import { r as pairedBootstrap } from "./paired-tests-C8iCsioC.js";
|
|
4
4
|
import { a as summaryTable, i as researchReport, n as gainHistogram, r as paretoChart, t as RESEARCH_REPORT_HARD_PAIR_FLOOR } from "./summary-report-BMvou2oZ.js";
|
|
5
5
|
import { n as pairedEvalueSequence, t as evaluateInterimReleaseConfidence } from "./sequential-CzK5DarL.js";
|
|
6
|
-
import { i as judgeReplayGate, n as evaluateReleaseConfidence, r as bootstrapCi, t as assertReleaseConfidence } from "./release-confidence-
|
|
6
|
+
import { i as judgeReplayGate, n as evaluateReleaseConfidence, r as bootstrapCi, t as assertReleaseConfidence } from "./release-confidence-XQm_xjhR.js";
|
|
7
7
|
import { t as rubricPredictiveValidity } from "./rubric-predictive-validity-CCK-1B7w.js";
|
|
8
8
|
export { RESEARCH_REPORT_HARD_PAIR_FLOOR, assertReleaseConfidence, benjaminiHochberg, bootstrapCi, evaluateInterimReleaseConfidence, evaluateReleaseConfidence, gainHistogram, judgeReplayGate, pairedBootstrap, pairedEvalueSequence, paretoChart, researchReport, rubricPredictiveValidity, summaryTable, wilcoxonSignedRank };
|
|
@@ -1,10 +1,10 @@
|
|
|
1
1
|
import { i as AgentProfileCellInput, r as AgentProfileCell } from "./agent-profile-cell-BHcB-mZX.js";
|
|
2
|
-
import { a as RunRecord, c as
|
|
3
|
-
import { rt as RawProviderSink, x as ChatClient } from "./types-
|
|
4
|
-
import { s as TraceStore } from "./store-
|
|
5
|
-
import { b as GateDecision, d as ResearchReportOptions, s as ResearchReport } from "./summary-report-
|
|
6
|
-
import {
|
|
7
|
-
import { a as RunIntegrityReport, n as RunIntegrityExpectations } from "./integrity-
|
|
2
|
+
import { a as RunRecord, c as RunSplitTag, d as RunTokenUsage, l as RunTaskFailure, n as RunCostProvenance, r as RunJudgeMetadata, t as JudgeScoresRecord } from "./run-record-DF-fuu31.js";
|
|
3
|
+
import { rt as RawProviderSink, x as ChatClient } from "./types-C0bWoSKC.js";
|
|
4
|
+
import { s as TraceStore } from "./store-C4avFjIi.js";
|
|
5
|
+
import { b as GateDecision, d as ResearchReportOptions, s as ResearchReport } from "./summary-report-DvvesgR_.js";
|
|
6
|
+
import { o as TraceEmitter, r as RunCompleteHook } from "./emitter-HcCmFQl_.js";
|
|
7
|
+
import { a as RunIntegrityReport, n as RunIntegrityExpectations } from "./integrity-CaeNeusS.js";
|
|
8
8
|
//#region src/eval-campaign.d.ts
|
|
9
9
|
interface CampaignVariant<V> {
|
|
10
10
|
id: string;
|
|
@@ -290,4 +290,4 @@ interface Researcher {
|
|
|
290
290
|
}
|
|
291
291
|
//#endregion
|
|
292
292
|
export { SteeringChange as a, CampaignRunContext as c, CampaignVariant as d, EvalCampaignOptions as f, runEvalCampaign as h, Researcher as i, CampaignRunOutcome as l, FailedRun as m, ExperimentResult as n, CampaignFactoryParams as o, EvalCampaignResult as p, FailureMode as r, CampaignIntegrityPolicy as s, ExperimentPlan as t, CampaignScenario as u };
|
|
293
|
-
//# sourceMappingURL=researcher-
|
|
293
|
+
//# sourceMappingURL=researcher-DojEMb7E.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"researcher-
|
|
1
|
+
{"version":3,"file":"researcher-DojEMb7E.d.ts","names":[],"sources":["../src/eval-campaign.ts","../src/researcher.ts"],"mappings":";;;;;;;;UAuEiB,gBAAgB;EAC/B;EACA,SAAS;;UAGM;EACf;;EAEA,OAAO;;UAGQ,mBAAmB;;EAElC;;EAEA;EACA,SAAS;EACT;EACA;EACA,cAAc;EACd;EACA,UAAU;;;;;;;EAOV,SAAS;EACT,OAAO;EACP,SAAS;;;;;EAKT,MAAM;;;UAIS;;;;;;EAMf,SAAS;EACT;;UAGQ;;EAER;;EAEA;;EAEA;;EAEA,gBAAgB;EAChB,YAAY;;EAEZ;;EAEA;;EAEA;;EAEA,MAAM;;EAEN,gBAAgB;;;;;;EAMhB,cAAc;;;;;;EAMd,eAAe,mBAAmB;;;KAIxB,qBAAqB,2BAA2B;KAEhD,eAAe,MAAM,KAAK,mBAAmB,OAAO,QAAQ;KAE5D;UAEK,oBAAoB;;;;;EAKnC;EACA,UAAU,gBAAgB;EAC1B,WAAW;;EAEX;;EAEA,WAAW;;EAEX;;;;;;;;EAQA,cAAc,QAAQ,uBAAuB;;;;;;;EAO7C;;;;;;EAMA,eAAe,QAAQ,0BAA0B;;;;;;;EAOjD,kBAAkB,QAAQ,0BAA0B;;;;;EAKpD;;;;;EAKA,gBAAgB;;;;;;EAMhB,YAAY;;EAEZ,qBAAqB;;;;;EAKrB,QAAQ,eAAe;;;;;EAKvB;IAAW;MAAwB,KACjC;;;;;EAOF;;EAEA;;;;EAIA;;EAEA,SAAS,QAAQ;;;;;;;EAOjB,eACI,mBACA,0BAEE,QAAQ;IACN,SAAS;IACT,cAAc;QAGd,mBACA,wBACA,QAAQ,mBAAmB;;UAGpB;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;;EAEA;EACA;;EAEA,MAAM;;EAEN,kBAAkB;EAClB,YAAY;;EAEZ,SAAS;EACT;EACA;;iBAWoB,gBAAgB,GACpC,MAAM,oBAAoB,KACzB,QAAQ;;;;UClRM;;;EAGf;;EAEA;EACA;;;IAGE;;IAEA;;;;UAKa;EACf;;;;EAIA;;;EAGA;;EAEA;;;UAIe;EACf;EACA;EACA,SAAS;;;EAGT;;EAEA;IAAU;IAAkB;;;;UAIb;EACf,MAAM;EACN,MAAM;EACN,cAAc;;;;;;;;;;;;;;;;;;;UAoBC;EACf,gBAAgB,MAAM,cAAc,QAAQ;EAC5C,cAAc,UAAU,gBAAgB,QAAQ;EAChD,YAAY,SAAS,kBAAkB,UAAU,iBAAiB,QAAQ;EAC1E,eAAe,MAAM,iBAAiB,QAAQ"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { a as RunRecord } from "./run-record-
|
|
1
|
+
import { a as RunRecord } from "./run-record-DF-fuu31.js";
|
|
2
2
|
import { x as VerificationStrategySource } from "./verdict-E4eRNf7-.js";
|
|
3
3
|
import { s as VerificationReport } from "./multi-layer-verifier-BUaQ4C17.js";
|
|
4
4
|
//#region src/rl/verifiable-reward.d.ts
|
|
@@ -251,4 +251,4 @@ interface DetectRewardHackingInput {
|
|
|
251
251
|
declare function detectRewardHacking(input: DetectRewardHackingInput): RewardHackingReport;
|
|
252
252
|
//#endregion
|
|
253
253
|
export { detectRewardHacking as a, VerifiableRewardSource as c, filterDeterministicallyRewarded as d, RewardHackingSignal as i, extractVerifiableReward as l, RewardHackingFinding as n, VerifiableReward as o, RewardHackingReport as r, VerifiableRewardExtractionOptions as s, DetectRewardHackingInput as t, extractVerifiableRewardsFromRecords as u };
|
|
254
|
-
//# sourceMappingURL=reward-hacking-
|
|
254
|
+
//# sourceMappingURL=reward-hacking-Cyk-QHk9.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"reward-hacking-
|
|
1
|
+
{"version":3,"file":"reward-hacking-Cyk-QHk9.d.ts","names":[],"sources":["../src/rl/verifiable-reward.ts","../src/rl/reward-hacking.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KA+DY,yBAAyB;UAEpB;;EAEf;;EAEA,QAAQ;;;;;;;;EAQR;;;;;;EAMA;;EAEA;;;;;;;EAOA,YAAY;;;;;;;;;;;EAWZ;;;;;;;;;;;;;;;;;;EAkBA;;UAGe;;;;;;;;EAQf;;;;EAIA,aAAa,sBAAsB;;;;;;;EAOnC;;;;;EAKA;;;;;;;;;;;;EAYA;;;;;;;;;;;;;;;iBA+Dc,wBACd,QAAQ,oBACR,OAAM,oCACL;;;;;;;;;;;;;;;;;;;;;;;iBAqGa,oCACd,MAAM,aACN,OAAM,oCACL;EAAQ;EAAe,QAAQ;;;;;;;;;;;iBA4GlB,gCACd,MAAM,aACN,OAAM,oCACL;EAAQ,KAAK;EAAW,QAAQ;;;;KC1YvB;UAMK;EACf,QAAQ;;EAER;EACA;;EAEA,QAAQ;;UAGO;EACf,UAAU;;EAEV,kBAAkB;;;;;;EAMlB;;EAEA;;EAEA;;UAGe;;;;;EAKf,MAAM;;;;;EAKN,WAAW,KAAK;;;;;;EAMhB,WAAW,KAAK;;;;;;EAMhB,qBAAqB,KAAK;;;;;EAK1B;;;;;EAKA;IAAe;IAAkB;;;;;EAIjC,0BAA0B;;iBAgBZ,oBAAoB,OAAO,2BAA2B"}
|
package/dist/rl.d.ts
CHANGED
|
@@ -1,17 +1,17 @@
|
|
|
1
|
-
import {
|
|
2
|
-
import { a as RunRecord,
|
|
1
|
+
import { v as Span } from "./schema-2g7trUwH.js";
|
|
2
|
+
import { a as RunRecord, c as RunSplitTag } from "./run-record-DF-fuu31.js";
|
|
3
3
|
import { a as PairedPromotionDecision, d as PairedBootstrapResult } from "./paired-promotion-decision-DPKz8ZnF.js";
|
|
4
|
-
import { a as CampaignResult } from "./types-
|
|
5
|
-
import { s as TraceStore } from "./store-
|
|
4
|
+
import { a as CampaignResult } from "./types-BvmWNjjs.js";
|
|
5
|
+
import { s as TraceStore } from "./store-C4avFjIi.js";
|
|
6
6
|
import { s as VerificationReport } from "./multi-layer-verifier-BUaQ4C17.js";
|
|
7
|
-
import { a as detectRewardHacking, c as VerifiableRewardSource, d as filterDeterministicallyRewarded, i as RewardHackingSignal, l as extractVerifiableReward, n as RewardHackingFinding, o as VerifiableReward, r as RewardHackingReport, s as VerifiableRewardExtractionOptions, t as DetectRewardHackingInput, u as extractVerifiableRewardsFromRecords } from "./reward-hacking-
|
|
7
|
+
import { a as detectRewardHacking, c as VerifiableRewardSource, d as filterDeterministicallyRewarded, i as RewardHackingSignal, l as extractVerifiableReward, n as RewardHackingFinding, o as VerifiableReward, r as RewardHackingReport, s as VerifiableRewardExtractionOptions, t as DetectRewardHackingInput, u as extractVerifiableRewardsFromRecords } from "./reward-hacking-Cyk-QHk9.js";
|
|
8
8
|
import { i as InMemoryOutcomeStore, n as FileSystemOutcomeStore, o as OutcomeStore, r as FileSystemOutcomeStoreOptions, t as DeploymentOutcome } from "./outcome-store-CNt4iZ67.js";
|
|
9
|
-
import { a as MintedRolloutLine, t as ChatMessage, y as RolloutSplit } from "./schema-
|
|
9
|
+
import { a as MintedRolloutLine, t as ChatMessage, y as RolloutSplit } from "./schema-B5LtYvn9.js";
|
|
10
10
|
import { t as InterimReleaseConfidence } from "./sequential-BhsrMupG.js";
|
|
11
11
|
import { t as AdversarialMutation } from "./adversarial-smnADNFS.js";
|
|
12
|
-
import { a as SteeringChange, f as EvalCampaignOptions, h as runEvalCampaign, i as Researcher, n as ExperimentResult, p as EvalCampaignResult, r as FailureMode, t as ExperimentPlan } from "./researcher-
|
|
13
|
-
import {
|
|
14
|
-
import { a as RubricPredictiveValidityReport, t as OutcomeMetricSpec } from "./rubric-predictive-validity-
|
|
12
|
+
import { a as SteeringChange, f as EvalCampaignOptions, h as runEvalCampaign, i as Researcher, n as ExperimentResult, p as EvalCampaignResult, r as FailureMode, t as ExperimentPlan } from "./researcher-DojEMb7E.js";
|
|
13
|
+
import { b as TrainingExportOptions, t as MintRolloutOptions } from "./mint-DsS38v5c.js";
|
|
14
|
+
import { a as RubricPredictiveValidityReport, t as OutcomeMetricSpec } from "./rubric-predictive-validity-DIUwJJ9Y.js";
|
|
15
15
|
//#region src/rl/active-curriculum.d.ts
|
|
16
16
|
interface CellObservation {
|
|
17
17
|
variantId: string;
|
|
@@ -1282,7 +1282,8 @@ interface AdapterContext {
|
|
|
1282
1282
|
* Errored or unjudged cells remain unlabeled while retaining explicit terminal
|
|
1283
1283
|
* outcome, execution-error count, token usage, cost, and failure detail.
|
|
1284
1284
|
* `candidateId` identifies the measured surface and defaults to the campaign
|
|
1285
|
-
* manifest hash.
|
|
1285
|
+
* manifest hash. A run id is `${candidateId}:${cellId}`: a campaign cell id is
|
|
1286
|
+
* `${scenario}:${rep}`, which every candidate run on that scenario shares.
|
|
1286
1287
|
*/
|
|
1287
1288
|
declare function campaignToRunRecords(campaign: CampaignResult, ctx: AdapterContext & {
|
|
1288
1289
|
candidateId?: string;
|
package/dist/rl.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"rl.d.ts","names":[],"sources":["../src/rl/active-curriculum.ts","../src/rl/adaptation-eval.ts","../src/rl/compute-curves.ts","../src/rl/contamination.ts","../src/rl/rollout-input.ts","../src/rl/preferences.ts","../src/rl/process-reward.ts","../src/rl/exporters.ts","../src/rl/dataset.ts","../src/rl/corpus.ts","../src/rl/off-policy.ts","../src/rl/predictive-validity-researcher.ts","../src/rl/rl-campaign.ts","../src/rl/run-record-adapters.ts","../src/rl/sim-fidelity.ts","../src/rl/tournament.ts","../src/rl/verified-findings-dataset.ts"],"mappings":";;;;;;;;;;;;;;;UAqCiB;EACf;EACA;;EAEA;;EAEA;;UAGe;EACf;EACA;;EAEA;;EAEA;;UAGe;;EAEf;;;;;;EAMA;;;;;EAKA;;;;;;;;;iBAUc,wBACd,cAAc,mBACd,gBAAgB;EAAQ;EAAmB;IAC3C,MAAM,4BACL;UAoDc;EACf;;;;;;EAMA;;EAEA;EACA;;;EAGA;;;;;;;;;;;;;iBAcc,mBACd,cAAc,mBACd,gBAAgB;EAAQ;EAAmB;IAC3C,MAAM,4BACL;;iBAyDa,2BACd,MAAM,aACN;EAAQ;EAAwB;IAC/B;;;UCpNc,iBAAiB;;;;;;EAMhC,IAAI;IAAQ,UAAU;IAAG;IAAW;MAAgB;;UAGrC,0BAA0B;EACzC,WAAW;;EAEX;;EAEA;EACA,QAAQ,iBAAiB;;EAEzB;;UAGe;EACf;EACA;EACA;EACA;EACA;;EAEA,aAAa;IAAQ;IAAoB;IAAmB;IAAgB;;;UAG7D;EACf,QAAQ;;;;;EAKR;;;;;EAKA;;iBAGoB,mBAAmB;EAAY;GACnD,MAAM,0BAA0B,KAC/B,QAAQ;UAmEM;EACf,MAAM;IACJ;;IAEA,OAAO;;;EAGT,WAAW;;EAEX,cAAc;EACd,cAAc;;EAEd;;EAEA;;EAEA;;;;;;;;;iBAUc,wBACd,GAAG,iBACH,GAAG,iBACH;EACE;EACA;EACA;;EAEA;IAED;;iBA0Ea,WAAW,OAAO,iBAAiB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;UC3MlC;;EAEf;;EAEA;;EAEA,OAAO;;UAGQ;EACf;EACA;EACA;;EAEA;;EAEA;;EAEA,UAAU;;UAGK;EACf;EACA,QAAQ;;EAER;;EAEA,MAAM;;UAGS;EACf;EACA,SAAS;;;;;EAKT,cAAc,QAAQ,uBAAuB;IAC3C;IACA;IACA;IACA,UAAU;;;iBAIQ,gBAAgB,MAAM,yBAAyB,QAAQ;UAmB5D,sBAAsB;;EAErC;;EAEA,SAAS,sBAAsB,QAAQ;;EAEvC,UAAU,SAAS,MAAM;;UAGV,qBAAqB;EACpC,MAAM;EACN;EACA;EACA;;EAEA;;;iBAIoB,QAAQ,GAAG,MAAM,sBAAsB,KAAK,QAAQ,qBAAqB;UAqB9E,uBAAuB;EACtC;EACA,SAAS,sBAAsB,QAAQ;;EAEvC,YAAY,SAAS;;UAGN,sBAAsB;;EAErC;;EAEA;;EAEA,WAAW;;EAEX,gBAAgB;;EAEhB,UAAU;;;;;;iBAOU,gBAAgB,GACpC,MAAM,uBAAuB,KAC5B,QAAQ,sBAAsB;;;;;;;UAkChB;EACf;EACA;EACA;EACA;;iBAGc,eAAe,QAAQ,qBAAqB;;;;;;;;;;;;KC7LhD;UAOK,qBAAqB;EACpC,MAAM;;EAEN,QAAQ,UAAU,MAAM,QAAQ,KAAK;;EAErC;;UAGe,wBAAwB;;EAEvC,aAAa,GAAG;;EAEhB,WAAW;;;;;EAKX,YAAY;EACZ,eAAe,qBAAqB;;;;;;EAMpC,UAAU,GAAG,MAAM;;UAGJ;;EAEf;;EAEA;;;;;EAKA;;UAGe;EACf,aAAa;IACX;IACA;IACA;IACA;;;EAGF;IAAc;IAAW;;;EAEzB;EACA;EACA;EACA;;EAEA;;EAEA;;iBAGoB,sBAAsB,GAC1C,OAAO,wBAAwB,IAC/B,OAAM,4BACL,QAAQ;;;;;;;iBAyGK,gBAAgB;EAAY;GAC1C,uBACA,UAAS,cAAc,yBACtB,qBAAqB;;;;;;iBAoBR,aAAa;EAAY;GACvC,iBAAiB,gBAAgB,8BACjC,eACC,qBAAqB;;;;;;iBAgBR,uBAAuB;EAAY;GACjD,gBACA,iCACC,qBAAqB;;;;;;;;;;;;;UClMP;;;;;;;;;;;EAWf,OAAO;;;;KCKG;UAKK;;EAEf;;EAEA;EACA;;EAEA;EACA;;EAEA;;;;;;EAMA;IAAW;IAAgB;;;EAE3B;;;;;EAKA;IACE;IACA;IACA;IACA;IACA;IACA;;;UAIa;EACf,WAAW;;;;;EAKX;;;;;;EAMA,QAAQ;;EAER;;UAGe;EACf,OAAO;;EAEP;;EAEA;;EAEA;;EAEA,UAAU;;;;;;EAMV;;;;;;;;;;;;;;;;;;;;iBAuCc,mBACd,OAAO,qBACP,OAAM,4BACL;;;;;;;;;;;;;;;;;;iBAqMmB,YACpB,SAAS,oBACT,SAAS,YACT,SAAS,qBACR,QAAQ;EAAQ;EAAgB;EAAgB;;;;;;;;;;;iBAkCnC,kBACd,SAAS,oBACT,SAAS,qBACR;EAAQ;EAAoB;EAAqB;EAAuB;;;;UChX1D;;EAEf;EACA;;EAEA;;EAEA,MAAM;;EAEN;;EAEA;;;;;EAKA;;EAEA;;EAEA;;UAGe;;EAEf,WAAW;;EAEX,MAAM,MAAM,OAAO,QAAQ,KAAK;;UAGjB;;;;;;EAMf,SAAS;;EAET,aAAa,MAAM;;iBAGC,mBACpB,OAAO,YACP,eACA,MAAM,4BACL,QAAQ;UAgCM;EACf;EACA;EACA;;EAEA;;EAEA;;EAEA;EACA;;iBAGc,yBAAyB,aAAa,eAAe;UA+CpD;;EAEf;EACA;;EAEA;EACA;;EAEA;EACA;;EAEA;EACA;;;;;;;;;;;;;;;;;iBAkBc,iBACd,kBAAkB,YAAY,eAC9B;EAAQ;EAAoB;IAC3B;;;UCrJc;;EAEf,WAAW,2BAA2B;;EAEtC,eAAe,2BAA2B;;UAG3B;EACf;EACA;EACA;;EAEA;;EAEA,OAAO;;;KAIG,iBAAiB;;;;;;;;;;;;;;;;iBAwBP,UACpB,SAAS,oBACT,SAAS,YACT,SAAS,iBACR,QAAQ;;iBAwCK,WAAW,MAAM;;KAOrB,+BAA+B;UAE1B,oBACP,KAAK;;EAEb,WAAW,2BAA2B;;EAEtC,eAAe,2BAA2B;;UAG3B;EACf;EACA;EACA;;EAEA;EACA,OAAO;;;;;;;;;;;;;;;;;;;iBAoBa,WACpB,OAAO,qBACP,SAAS,cACR,QAAQ;iBAyDK,YAAY,MAAM;UAMjB,mBAAmB;;;;;EAKlC,WAAW,2BAA2B;;EAEtC,eAAe,2BAA2B;;EAE1C,YAAY,MAAM;;EAElB,WAAW,MAAM;;UAGF;;EAEf,UAAU;EACV,OAAO;;;;;;;;;;;;;;;;;;;iBAoBa,UACpB,OAAO,qBACP,SAAS,aACR,QAAQ;iBAwDK,WAAW,MAAM;UAMhB;;EAEf,WAAW,2BAA2B;;EAEtC,aAAa,eAAe,4BAA4B;;EAExD,YAAY,eAAe,uCAAuC;;UAGnD;EACf;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,OAAO;;UAGQ,uBAAuB;;;;;;;;;;EAUtC;;;;;;;;;;;;;;;;;;;;;iBAsBoB,UACpB,SAAS,qBACT,SAAS,YACT,SAAS,iBACR,QAAQ;iBA8FK,WAAW,MAAM;UAMhB;EACf;EACA;EACA;EACA;EACA;EACA;;;;;;;;;;;;iBAoBc,mBAAmB,aAAa,cAAc,SAAS;;;KCpf3D;cACN;KACM,wBAAwB;iBAIpB,uBAAuB,iBAAiB;;;UA2BvC;EACf;EACA;;EAEA;;;EAGA;;;EAGA;IAAU,MAAM;IAAY;IAAgB;;EAC5C;EACA;EACA;;EAEA;;EAEA,UAAU;;EAEV;IACE;IACA;IACA;;;UAIa;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;;EAEA;;EAEA,QAAQ,OAAO;EACf,QAAQ;;EAER;;EAEA;EACA;EACA;IAAe;IAAe;;EAC9B;;;;;;EAMA;;UAGe,0BAA0B;EACzC,SAAS;EACT,WAAW,QAAQ,OAAO;EAC1B,OAAO;;UAGQ;EACf,UAAU;;EAEV,OAAO;;;;;;;;;;iBA0Da,eACpB,OAAO,qBACP,SAAS,cAAc,YACvB,QAAQ,iBACR;EAAgB,SAAS;EAAoB,SAAS;IACrD,QAAQ;;iBA2DK,oBAAoB,GAAG;;;;;KC1O3B,eAAe;EAAc;EAAiB;;UAEzC;EACf;;;EAGA;EACA;;;;;;;;iBASc,eAAe,SAAS,gBAAgB,qBAAqB;;iBAqB7D,WAAW,qBAAqB;UAmB/B;;EAEf;;EAEA,SAAS;;EAET;;;;;;;;;;;;;;;iBAgBoB,uBACpB,oBACA,QAAQ,iBACR,OAAM,iBACL,QAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;UCnEM;;EAEf;;EAEA;;;;;;EAMA;;;;;;EAMA;;;;;;EAMA;;;;;;;EAOA;;UAGe;;EAEf;;EAEA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;;;;EAKA;;EAEA,qBAAqB;;UAGN;;;;;;EAMf;;EAEA;IAAe;IAAa;;;;;;;;iBAQd,4BACd,cAAc,uBACd,OAAM,mBACL;;;;;;iBA4Ca,kCACd,cAAc,uBACd,OAAM,mBACL;;;;;;;;;;;;;;;;;;;;;iBAwDa,aACd,cAAc,uBACd,OAAM,mBACL;;;;;;iBAuEa,qBACd,cAAc,uBACd,OAAM;EACH,KAAK;EAAmB,OAAO;EAAmB,IAAI;;;;UCvQ1C;EACf,UAAU;;EAEV,eAAe;;EAEf;;EAEA;;;;;;EAMA,YAAY,QAAQ,0CAA0C;;;;;;cAOnD,wCAAwC;mBAClC;UACT;EAER,YAAY,MAAM;EAaZ,gBAAgB,MAAM,cAAc,QAAQ;EA0C5C,cAAc,UAAU,gBAAgB,QAAQ;EA+DhD,YAAY,SAAS,kBAAkB,UAAU,iBAAiB,QAAQ;EAS1E,eAAe,MAAM,iBAAiB,QAAQ;;;;;;EAiD9C,iBAAiB,MAAM,cAAc,QAAQ;;;;;;EAiBnD,UAAU,QAAQ;EAUlB,iBAAiB;;;;UCxMF,qBAAqB,WAAW,oBAAoB;;EAEnE,UAAU;;EAEV,cAAc;;EAEd,mBAAmB;;EAEnB,eAAe;EACf,0BAA0B;;EAE1B;IACE;IACA;IACA;MAAS;MAAa;;;;;;;;;;;;;IAYtB;;;EAGF;IACE,MAAM;IACN,OAAO;IACP,MAAM;;;;;;;;;;;UAYO;EACf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;EACf,UAAU;;EAEV,cAAc;;EAEd,eAAe;IAAQ;IAAe,QAAQ;;;EAE9C,aAAa;;;;;EAKb,mBAAmB;;;;EAInB,eAAe;;EAEf,eAAe;;EAEf,oBAAoB;;EAEpB;IACE,MAAM;IACN,OAAO;IACP,MAAM;;;;;EAKR;;;;EAIA;;iBAGoB,cAAc,GAAG,MAAM,qBAAqB,KAAK,QAAQ;;;UC5I9D;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,WAAW;;EAEX
|
|
1
|
+
{"version":3,"file":"rl.d.ts","names":[],"sources":["../src/rl/active-curriculum.ts","../src/rl/adaptation-eval.ts","../src/rl/compute-curves.ts","../src/rl/contamination.ts","../src/rl/rollout-input.ts","../src/rl/preferences.ts","../src/rl/process-reward.ts","../src/rl/exporters.ts","../src/rl/dataset.ts","../src/rl/corpus.ts","../src/rl/off-policy.ts","../src/rl/predictive-validity-researcher.ts","../src/rl/rl-campaign.ts","../src/rl/run-record-adapters.ts","../src/rl/sim-fidelity.ts","../src/rl/tournament.ts","../src/rl/verified-findings-dataset.ts"],"mappings":";;;;;;;;;;;;;;;UAqCiB;EACf;EACA;;EAEA;;EAEA;;UAGe;EACf;EACA;;EAEA;;EAEA;;UAGe;;EAEf;;;;;;EAMA;;;;;EAKA;;;;;;;;;iBAUc,wBACd,cAAc,mBACd,gBAAgB;EAAQ;EAAmB;IAC3C,MAAM,4BACL;UAoDc;EACf;;;;;;EAMA;;EAEA;EACA;;;EAGA;;;;;;;;;;;;;iBAcc,mBACd,cAAc,mBACd,gBAAgB;EAAQ;EAAmB;IAC3C,MAAM,4BACL;;iBAyDa,2BACd,MAAM,aACN;EAAQ;EAAwB;IAC/B;;;UCpNc,iBAAiB;;;;;;EAMhC,IAAI;IAAQ,UAAU;IAAG;IAAW;MAAgB;;UAGrC,0BAA0B;EACzC,WAAW;;EAEX;;EAEA;EACA,QAAQ,iBAAiB;;EAEzB;;UAGe;EACf;EACA;EACA;EACA;EACA;;EAEA,aAAa;IAAQ;IAAoB;IAAmB;IAAgB;;;UAG7D;EACf,QAAQ;;;;;EAKR;;;;;EAKA;;iBAGoB,mBAAmB;EAAY;GACnD,MAAM,0BAA0B,KAC/B,QAAQ;UAmEM;EACf,MAAM;IACJ;;IAEA,OAAO;;;EAGT,WAAW;;EAEX,cAAc;EACd,cAAc;;EAEd;;EAEA;;EAEA;;;;;;;;;iBAUc,wBACd,GAAG,iBACH,GAAG,iBACH;EACE;EACA;EACA;;EAEA;IAED;;iBA0Ea,WAAW,OAAO,iBAAiB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;UC3MlC;;EAEf;;EAEA;;EAEA,OAAO;;UAGQ;EACf;EACA;EACA;;EAEA;;EAEA;;EAEA,UAAU;;UAGK;EACf;EACA,QAAQ;;EAER;;EAEA,MAAM;;UAGS;EACf;EACA,SAAS;;;;;EAKT,cAAc,QAAQ,uBAAuB;IAC3C;IACA;IACA;IACA,UAAU;;;iBAIQ,gBAAgB,MAAM,yBAAyB,QAAQ;UAmB5D,sBAAsB;;EAErC;;EAEA,SAAS,sBAAsB,QAAQ;;EAEvC,UAAU,SAAS,MAAM;;UAGV,qBAAqB;EACpC,MAAM;EACN;EACA;EACA;;EAEA;;;iBAIoB,QAAQ,GAAG,MAAM,sBAAsB,KAAK,QAAQ,qBAAqB;UAqB9E,uBAAuB;EACtC;EACA,SAAS,sBAAsB,QAAQ;;EAEvC,YAAY,SAAS;;UAGN,sBAAsB;;EAErC;;EAEA;;EAEA,WAAW;;EAEX,gBAAgB;;EAEhB,UAAU;;;;;;iBAOU,gBAAgB,GACpC,MAAM,uBAAuB,KAC5B,QAAQ,sBAAsB;;;;;;;UAkChB;EACf;EACA;EACA;EACA;;iBAGc,eAAe,QAAQ,qBAAqB;;;;;;;;;;;;KC7LhD;UAOK,qBAAqB;EACpC,MAAM;;EAEN,QAAQ,UAAU,MAAM,QAAQ,KAAK;;EAErC;;UAGe,wBAAwB;;EAEvC,aAAa,GAAG;;EAEhB,WAAW;;;;;EAKX,YAAY;EACZ,eAAe,qBAAqB;;;;;;EAMpC,UAAU,GAAG,MAAM;;UAGJ;;EAEf;;EAEA;;;;;EAKA;;UAGe;EACf,aAAa;IACX;IACA;IACA;IACA;;;EAGF;IAAc;IAAW;;;EAEzB;EACA;EACA;EACA;;EAEA;;EAEA;;iBAGoB,sBAAsB,GAC1C,OAAO,wBAAwB,IAC/B,OAAM,4BACL,QAAQ;;;;;;;iBAyGK,gBAAgB;EAAY;GAC1C,uBACA,UAAS,cAAc,yBACtB,qBAAqB;;;;;;iBAoBR,aAAa;EAAY;GACvC,iBAAiB,gBAAgB,8BACjC,eACC,qBAAqB;;;;;;iBAgBR,uBAAuB;EAAY;GACjD,gBACA,iCACC,qBAAqB;;;;;;;;;;;;;UClMP;;;;;;;;;;;EAWf,OAAO;;;;KCKG;UAKK;;EAEf;;EAEA;EACA;;EAEA;EACA;;EAEA;;;;;;EAMA;IAAW;IAAgB;;;EAE3B;;;;;EAKA;IACE;IACA;IACA;IACA;IACA;IACA;;;UAIa;EACf,WAAW;;;;;EAKX;;;;;;EAMA,QAAQ;;EAER;;UAGe;EACf,OAAO;;EAEP;;EAEA;;EAEA;;EAEA,UAAU;;;;;;EAMV;;;;;;;;;;;;;;;;;;;;iBAuCc,mBACd,OAAO,qBACP,OAAM,4BACL;;;;;;;;;;;;;;;;;;iBAqMmB,YACpB,SAAS,oBACT,SAAS,YACT,SAAS,qBACR,QAAQ;EAAQ;EAAgB;EAAgB;;;;;;;;;;;iBAkCnC,kBACd,SAAS,oBACT,SAAS,qBACR;EAAQ;EAAoB;EAAqB;EAAuB;;;;UChX1D;;EAEf;EACA;;EAEA;;EAEA,MAAM;;EAEN;;EAEA;;;;;EAKA;;EAEA;;EAEA;;UAGe;;EAEf,WAAW;;EAEX,MAAM,MAAM,OAAO,QAAQ,KAAK;;UAGjB;;;;;;EAMf,SAAS;;EAET,aAAa,MAAM;;iBAGC,mBACpB,OAAO,YACP,eACA,MAAM,4BACL,QAAQ;UAgCM;EACf;EACA;EACA;;EAEA;;EAEA;;EAEA;EACA;;iBAGc,yBAAyB,aAAa,eAAe;UA+CpD;;EAEf;EACA;;EAEA;EACA;;EAEA;EACA;;EAEA;EACA;;;;;;;;;;;;;;;;;iBAkBc,iBACd,kBAAkB,YAAY,eAC9B;EAAQ;EAAoB;IAC3B;;;UCrJc;;EAEf,WAAW,2BAA2B;;EAEtC,eAAe,2BAA2B;;UAG3B;EACf;EACA;EACA;;EAEA;;EAEA,OAAO;;;KAIG,iBAAiB;;;;;;;;;;;;;;;;iBAwBP,UACpB,SAAS,oBACT,SAAS,YACT,SAAS,iBACR,QAAQ;;iBAwCK,WAAW,MAAM;;KAOrB,+BAA+B;UAE1B,oBACP,KAAK;;EAEb,WAAW,2BAA2B;;EAEtC,eAAe,2BAA2B;;UAG3B;EACf;EACA;EACA;;EAEA;EACA,OAAO;;;;;;;;;;;;;;;;;;;iBAoBa,WACpB,OAAO,qBACP,SAAS,cACR,QAAQ;iBAyDK,YAAY,MAAM;UAMjB,mBAAmB;;;;;EAKlC,WAAW,2BAA2B;;EAEtC,eAAe,2BAA2B;;EAE1C,YAAY,MAAM;;EAElB,WAAW,MAAM;;UAGF;;EAEf,UAAU;EACV,OAAO;;;;;;;;;;;;;;;;;;;iBAoBa,UACpB,OAAO,qBACP,SAAS,aACR,QAAQ;iBAwDK,WAAW,MAAM;UAMhB;;EAEf,WAAW,2BAA2B;;EAEtC,aAAa,eAAe,4BAA4B;;EAExD,YAAY,eAAe,uCAAuC;;UAGnD;EACf;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,OAAO;;UAGQ,uBAAuB;;;;;;;;;;EAUtC;;;;;;;;;;;;;;;;;;;;;iBAsBoB,UACpB,SAAS,qBACT,SAAS,YACT,SAAS,iBACR,QAAQ;iBA8FK,WAAW,MAAM;UAMhB;EACf;EACA;EACA;EACA;EACA;EACA;;;;;;;;;;;;iBAoBc,mBAAmB,aAAa,cAAc,SAAS;;;KCpf3D;cACN;KACM,wBAAwB;iBAIpB,uBAAuB,iBAAiB;;;UA2BvC;EACf;EACA;;EAEA;;;EAGA;;;EAGA;IAAU,MAAM;IAAY;IAAgB;;EAC5C;EACA;EACA;;EAEA;;EAEA,UAAU;;EAEV;IACE;IACA;IACA;;;UAIa;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;;EAEA;;EAEA,QAAQ,OAAO;EACf,QAAQ;;EAER;;EAEA;EACA;EACA;IAAe;IAAe;;EAC9B;;;;;;EAMA;;UAGe,0BAA0B;EACzC,SAAS;EACT,WAAW,QAAQ,OAAO;EAC1B,OAAO;;UAGQ;EACf,UAAU;;EAEV,OAAO;;;;;;;;;;iBA0Da,eACpB,OAAO,qBACP,SAAS,cAAc,YACvB,QAAQ,iBACR;EAAgB,SAAS;EAAoB,SAAS;IACrD,QAAQ;;iBA2DK,oBAAoB,GAAG;;;;;KC1O3B,eAAe;EAAc;EAAiB;;UAEzC;EACf;;;EAGA;EACA;;;;;;;;iBASc,eAAe,SAAS,gBAAgB,qBAAqB;;iBAqB7D,WAAW,qBAAqB;UAmB/B;;EAEf;;EAEA,SAAS;;EAET;;;;;;;;;;;;;;;iBAgBoB,uBACpB,oBACA,QAAQ,iBACR,OAAM,iBACL,QAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;UCnEM;;EAEf;;EAEA;;;;;;EAMA;;;;;;EAMA;;;;;;EAMA;;;;;;;EAOA;;UAGe;;EAEf;;EAEA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;;;;EAKA;;EAEA,qBAAqB;;UAGN;;;;;;EAMf;;EAEA;IAAe;IAAa;;;;;;;;iBAQd,4BACd,cAAc,uBACd,OAAM,mBACL;;;;;;iBA4Ca,kCACd,cAAc,uBACd,OAAM,mBACL;;;;;;;;;;;;;;;;;;;;;iBAwDa,aACd,cAAc,uBACd,OAAM,mBACL;;;;;;iBAuEa,qBACd,cAAc,uBACd,OAAM;EACH,KAAK;EAAmB,OAAO;EAAmB,IAAI;;;;UCvQ1C;EACf,UAAU;;EAEV,eAAe;;EAEf;;EAEA;;;;;;EAMA,YAAY,QAAQ,0CAA0C;;;;;;cAOnD,wCAAwC;mBAClC;UACT;EAER,YAAY,MAAM;EAaZ,gBAAgB,MAAM,cAAc,QAAQ;EA0C5C,cAAc,UAAU,gBAAgB,QAAQ;EA+DhD,YAAY,SAAS,kBAAkB,UAAU,iBAAiB,QAAQ;EAS1E,eAAe,MAAM,iBAAiB,QAAQ;;;;;;EAiD9C,iBAAiB,MAAM,cAAc,QAAQ;;;;;;EAiBnD,UAAU,QAAQ;EAUlB,iBAAiB;;;;UCxMF,qBAAqB,WAAW,oBAAoB;;EAEnE,UAAU;;EAEV,cAAc;;EAEd,mBAAmB;;EAEnB,eAAe;EACf,0BAA0B;;EAE1B;IACE;IACA;IACA;MAAS;MAAa;;;;;;;;;;;;;IAYtB;;;EAGF;IACE,MAAM;IACN,OAAO;IACP,MAAM;;;;;;;;;;;UAYO;EACf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;EACf,UAAU;;EAEV,cAAc;;EAEd,eAAe;IAAQ;IAAe,QAAQ;;;EAE9C,aAAa;;;;;EAKb,mBAAmB;;;;EAInB,eAAe;;EAEf,eAAe;;EAEf,oBAAoB;;EAEpB;IACE,MAAM;IACN,OAAO;IACP,MAAM;;;;;EAKR;;;;EAIA;;iBAGoB,cAAc,GAAG,MAAM,qBAAqB,KAAK,QAAQ;;;UC5I9D;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,WAAW;;EAEX;;;;;;;;;;;iBAYc,qBACd,UAAU,gBACV,KAAK;EAAmB;IACvB;;;;;;;;iBAyBa,8BACd,QAAQ,oBACR,KAAK;EAAmB;EAAqB;GAC7C;EAAQ;IACP;;;;;;;KCpCS,oBAAoB,QAAQ,cAAc;;;;;cAMzC;;;;;;;;;;;;;;;;;;;;;;cAuCA,yBAAyB;;;;;;;;;iBAmDtB,aAAa,GAAG,wBAAwB,GAAG;;;;;;;iBAqC3C,cAAc,kBAAkB;;;iBAsBhC,YAAY,eAAe;UAU1B;;EAEf;;;EAGA;;EAEA;;UAGe;EACf;;EAEA;;EAEA,WAAW;;EAEX;;EAEA;;KAGU;UAEK;EACf,cAAc;;;;EAId;;;EAGA;;;EAGA,SAAS;;UAGM;;EAEf,WAAW;;EAEX;;;;;;;;iBASc,kBACd,WAAW,aACX,YAAY,aACZ,OAAM,qBACL;UAuHc;;;EAGf;;;EAGA;;UAGe;EACf;EACA;;EAEA;;;EAGA;;;;;;;;;iBAUc,cACd,WAAW,aACX,YAAY,aACZ,OAAM,kBACL;;;;;;;;;;;;;;;;;;;;;;;;;;;;;UC/Xc;;EAEf;;EAEA;;;;;EAKA;;;;;;EAMA;;UAGe;EACf;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;EACf,SAAS;;EAET;;EAEA;EACA;;;;;;;;;;;iBAYc,gBACd,UAAU,mBACV;EAAQ;EAAoB;EAAwB;IACnD;;;;;;;;;UA+Gc;;EAEf;;EAEA;;iBAGc,eACd,SAAS,qBACT,SAAS,iBACT,OAAM;EACH;EAAqB;;;;;;;;UA2BT;EACf,MAAM;IACJ;;IAEA;IACA;;;;;;EAMF;;iBAGc,0BACd,OAAO,iCACN;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;cC/MU;;UAKI;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA,KAAK;EACL;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA,OAAO;;;UAIQ;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA,kBAAkB;IAAQ;IAAkB;;;;UAI7B;EACf;EACA;EACA;;UAGe;EACf;;;EAGA;;EAEA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA,mBAAmB;EACnB;EACA;IAAU;IAAyB;;;UAKpB;EACf;EACA;EACA;;EAEA;EACA;;KAGU;UAEK;EACf,SAAS;EACT;EACA;EACA;EACA;;UAGe;EACf,eAAe;;EAEf;EACA;EACA;EACA;IACE;IACA;IACA;IACA;IACA;IACA;;EAEF;;IAEE;;IAEA;;IAEA;;IAEA;IACA;;;EAGF;IACE;MAAU;MAAe;;IACzB,OAAO;;EAET;IACE;;IAEA;IACA;IACA;IACA;IACA;IACA;IACA;;;IAGA;IACA,wBAAwB;IACxB;IACA;IACA;IACA;;EAEF,KAAK;EACL;IACE;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;;;UAIa;EACf;EACA;;;;EAIA;EACA,KAAK,OAAO;EACZ,UAAU;IAAiB;IAAc;IAAoB;;;UAO9C;EACf,WAAW;EACX,OAAO;EACP,OAAO;EACP;EACA;EACA;EACA;EACA;EACA;EACA;EACA,SAAS;EACT;;;;;;iBAiEc,wBAAwB,MAAM,8BAA8B;iBA8I5D,0BAA0B,MAAM,uBAAuB;iBAwBvD,wBAAwB,MAAM;UAM7B;EACf;;EAEA;;UAGe;EACf;;EAEA;;EAEA,SAAS,eAAe;;EAExB;EACA;;UAGe;EACf,MAAM;EACN,SAAS;EACT;IACE;IACA;IACA;IACA;IACA,SAAS;MAAiB;MAAoB;MAAsB;;;;;;;;;iBAuExD,4BACd,QAAQ,yBACP"}
|
package/dist/rl.js
CHANGED
|
@@ -6,14 +6,14 @@ import { l as wilcoxonSignedRank } from "./paired-arms-JSDcs2zx.js";
|
|
|
6
6
|
import { r as pairedBootstrap } from "./paired-tests-C8iCsioC.js";
|
|
7
7
|
import { t as decidePairedPromotion } from "./paired-promotion-decision-6cn3bLc3.js";
|
|
8
8
|
import { r as observedSplitScore, s as trainingScore } from "./reward-nw2xZGZG.js";
|
|
9
|
-
import {
|
|
10
|
-
import { a as campaignCellToRunRecord } from "./run-record-
|
|
9
|
+
import { c as runTaskScore } from "./run-record-BRlM99rX.js";
|
|
10
|
+
import { a as campaignCellToRunRecord } from "./run-record-DnMFyg_g.js";
|
|
11
11
|
import { i as filterDeterministicallyRewarded, n as extractVerifiableReward, r as extractVerifiableRewardsFromRecords, t as detectRewardHacking } from "./reward-hacking-D0XwhVWE.js";
|
|
12
12
|
import { n as InMemoryTraceStore } from "./store-DNe_Uv1Q.js";
|
|
13
|
-
import { t as runEvalCampaign } from "./eval-campaign-
|
|
13
|
+
import { t as runEvalCampaign } from "./eval-campaign-C0Q1Eeo_.js";
|
|
14
14
|
import { s as assertRewardGate } from "./schema-C1aaAxTf.js";
|
|
15
15
|
import { n as isTrainingLineEligible, s as toSftRows$1, t as isSplitEligible } from "./exporters-DEcXbvgO.js";
|
|
16
|
-
import { t as mintRolloutRows } from "./mint-
|
|
16
|
+
import { t as mintRolloutRows } from "./mint-iCD-W43s.js";
|
|
17
17
|
import { t as evaluateInterimReleaseConfidence } from "./sequential-CzK5DarL.js";
|
|
18
18
|
import { n as assertUniqueObservationIds, s as validateOutcomeMetricSpecifications, t as rubricPredictiveValidity } from "./rubric-predictive-validity-CCK-1B7w.js";
|
|
19
19
|
import { n as thompsonCurriculum, r as varianceBasedCurriculum, t as observationsFromRunRecords } from "./active-curriculum-OjIWgrUJ.js";
|
|
@@ -2272,13 +2272,14 @@ function buildSummary(args) {
|
|
|
2272
2272
|
* Errored or unjudged cells remain unlabeled while retaining explicit terminal
|
|
2273
2273
|
* outcome, execution-error count, token usage, cost, and failure detail.
|
|
2274
2274
|
* `candidateId` identifies the measured surface and defaults to the campaign
|
|
2275
|
-
* manifest hash.
|
|
2275
|
+
* manifest hash. A run id is `${candidateId}:${cellId}`: a campaign cell id is
|
|
2276
|
+
* `${scenario}:${rep}`, which every candidate run on that scenario shares.
|
|
2276
2277
|
*/
|
|
2277
2278
|
function campaignToRunRecords(campaign, ctx) {
|
|
2278
2279
|
const splitTag = ctx.splitTag ?? "search";
|
|
2279
2280
|
const candidateId = ctx.candidateId ?? campaign.manifestHash;
|
|
2280
2281
|
return campaign.cells.map((cell) => campaignCellToRunRecord(cell, {
|
|
2281
|
-
runId: cell.cellId
|
|
2282
|
+
runId: `${candidateId}:${cell.cellId}`,
|
|
2282
2283
|
experimentId: ctx.experimentId,
|
|
2283
2284
|
candidateId,
|
|
2284
2285
|
model: ctx.model,
|