@tangle-network/agent-eval 0.163.2 → 0.171.0
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 +127 -0
- package/README.md +2 -0
- package/dist/adapters/http.d.ts +108 -0
- package/dist/adapters/http.d.ts.map +1 -0
- package/dist/adapters/http.js +208 -0
- package/dist/adapters/http.js.map +1 -0
- package/dist/analyst/index.d.ts +40 -70
- package/dist/analyst/index.d.ts.map +1 -1
- package/dist/analyst/index.js +18 -311
- package/dist/analyst/index.js.map +1 -1
- package/dist/{backend-integrity-DxuQCu_A.d.ts → backend-integrity-e79K3UPD.d.ts} +3 -3
- package/dist/{backend-integrity-DxuQCu_A.d.ts.map → backend-integrity-e79K3UPD.d.ts.map} +1 -1
- package/dist/{benchmark-BhT16ep9.js → benchmark-C4wk_Sjr.js} +10 -3
- package/dist/benchmark-C4wk_Sjr.js.map +1 -0
- package/dist/{benchmark-command-CF-4GEWZ.js → benchmark-command-D8k3Gf0J.js} +236 -251
- package/dist/benchmark-command-D8k3Gf0J.js.map +1 -0
- package/dist/{benchmark-CGPp-kDC.d.ts → benchmark-h-h4bfqj.d.ts} +3 -3
- package/dist/{benchmark-CGPp-kDC.d.ts.map → benchmark-h-h4bfqj.d.ts.map} +1 -1
- package/dist/benchmarks/index.d.ts +5 -5
- 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 +8 -8
- package/dist/campaign/index.js +7 -7
- package/dist/{campaign-DQZmc2Dq.js → campaign-B72njjHj.js} +15 -14
- package/dist/campaign-B72njjHj.js.map +1 -0
- package/dist/{canonical-IL-Bu-14.js → canonical-DPyQ_rpt.js} +22 -2
- package/dist/{canonical-IL-Bu-14.js.map → canonical-DPyQ_rpt.js.map} +1 -1
- package/dist/{chat-client-DlMlAeYI.js → chat-client-DEtybj5i.js} +5 -5
- package/dist/{chat-client-DlMlAeYI.js.map → chat-client-DEtybj5i.js.map} +1 -1
- package/dist/cli.js +2 -2
- package/dist/{client-CX7KqIdB.js → client-BvwNkIRN.js} +2 -2
- package/dist/{client-CX7KqIdB.js.map → client-BvwNkIRN.js.map} +1 -1
- package/dist/{client-L9VVPkim.d.ts → client-CDtcZ3p9.d.ts} +4 -4
- package/dist/{client-L9VVPkim.d.ts.map → client-CDtcZ3p9.d.ts.map} +1 -1
- package/dist/contract/index.d.ts +12 -703
- package/dist/contract/index.js +11 -11
- package/dist/{counterfactual-BaFUWK3H.d.ts → counterfactual-Bee5_BIn.d.ts} +4 -4
- package/dist/{counterfactual-BaFUWK3H.d.ts.map → counterfactual-Bee5_BIn.d.ts.map} +1 -1
- package/dist/{default-registry-G9CKMNkc.d.ts → default-registry-B0s2zU-s.d.ts} +6 -6
- package/dist/{default-registry-G9CKMNkc.d.ts.map → default-registry-B0s2zU-s.d.ts.map} +1 -1
- package/dist/{define-agent-eval-Dx1JnPEa.d.ts → define-agent-eval-Cjy2yhqP.d.ts} +7 -7
- package/dist/{define-agent-eval-Dx1JnPEa.d.ts.map → define-agent-eval-Cjy2yhqP.d.ts.map} +1 -1
- package/dist/{define-agent-eval-D08pWIJb.js → define-agent-eval-Dy8QgxAI.js} +20 -8
- package/dist/{define-agent-eval-D08pWIJb.js.map → define-agent-eval-Dy8QgxAI.js.map} +1 -1
- package/dist/{dspy-rlm-engine-DhA9qKIm.js → dspy-rlm-engine-CS3qcCEk.js} +3 -10
- package/dist/dspy-rlm-engine-CS3qcCEk.js.map +1 -0
- package/dist/{emitter-D_jYSGRd.d.ts → emitter-Bvnu0VzL.d.ts} +3 -3
- package/dist/{emitter-D_jYSGRd.d.ts.map → emitter-Bvnu0VzL.d.ts.map} +1 -1
- package/dist/{engine-Cu5qD5Fc.d.ts → engine-CAmTUk52.d.ts} +7 -7
- package/dist/{engine-Cu5qD5Fc.d.ts.map → engine-CAmTUk52.d.ts.map} +1 -1
- package/dist/{eval-campaign-BfohKmzx.js → eval-campaign-JDTeE6Pl.js} +4 -4
- package/dist/{eval-campaign-BfohKmzx.js.map → eval-campaign-JDTeE6Pl.js.map} +1 -1
- package/dist/{exact-types-qnexxJ1Z.d.ts → exact-types-BEecmnWm.d.ts} +2 -2
- package/dist/{exact-types-qnexxJ1Z.d.ts.map → exact-types-BEecmnWm.d.ts.map} +1 -1
- package/dist/experiment/index.d.ts +5 -5
- package/dist/experiment/index.js +4 -4
- package/dist/{experiment-tracker-DCO6Cz4s.d.ts → experiment-tracker-Dm8yQMqb.d.ts} +2 -2
- package/dist/{experiment-tracker-DCO6Cz4s.d.ts.map → experiment-tracker-Dm8yQMqb.d.ts.map} +1 -1
- package/dist/{external-optimizer-process-BFmh36vW.js → external-optimizer-process-CQxylYeG.js} +4 -11
- package/dist/external-optimizer-process-CQxylYeG.js.map +1 -0
- package/dist/{external-optimizer-subprocess-CqLMW3nh.js → external-optimizer-subprocess-Cex8Da2i.js} +25 -11
- package/dist/external-optimizer-subprocess-Cex8Da2i.js.map +1 -0
- package/dist/{failure-cluster-CXL8NbEw.d.ts → failure-cluster-6YSvsKlp.d.ts} +3 -3
- package/dist/{failure-cluster-CXL8NbEw.d.ts.map → failure-cluster-6YSvsKlp.d.ts.map} +1 -1
- package/dist/{feedback-trajectory-B3ZHaHV_.d.ts → feedback-trajectory-DIqpCyF0.d.ts} +6 -6
- package/dist/{feedback-trajectory-B3ZHaHV_.d.ts.map → feedback-trajectory-DIqpCyF0.d.ts.map} +1 -1
- package/dist/fuzz.d.ts +1 -1
- package/dist/{skillopt-optimization-method-x7TTF23P.d.ts → heldout-gate-Dh2b62w8.d.ts} +111 -111
- package/dist/heldout-gate-Dh2b62w8.d.ts.map +1 -0
- package/dist/hosted/index.d.ts +2 -2
- package/dist/hosted/index.js +1 -1
- package/dist/{index-D-V8gCs_.d.ts → index-8VIogTyS.d.ts} +30 -22
- package/dist/index-8VIogTyS.d.ts.map +1 -0
- package/dist/{index-D-IiQIBB.d.ts → index-DMoxLG8P.d.ts} +3 -3
- package/dist/{index-D-IiQIBB.d.ts.map → index-DMoxLG8P.d.ts.map} +1 -1
- package/dist/{index-D_P7Ye43.d.ts → index-DNgf5gyG.d.ts} +2 -2
- package/dist/{index-D_P7Ye43.d.ts.map → index-DNgf5gyG.d.ts.map} +1 -1
- package/dist/{index-CGtH1piv.d.ts → index-DT73JraI.d.ts} +7 -38
- package/dist/index-DT73JraI.d.ts.map +1 -0
- package/dist/index-fNXZMCzX.d.ts +704 -0
- package/dist/index-fNXZMCzX.d.ts.map +1 -0
- package/dist/index.d.ts +197 -38
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +736 -28
- package/dist/index.js.map +1 -1
- package/dist/{insight-report-DRe8LB6d.d.ts → insight-report-08F022xN.d.ts} +4 -4
- package/dist/{insight-report-DRe8LB6d.d.ts.map → insight-report-08F022xN.d.ts.map} +1 -1
- package/dist/{integrity-DUNX9Fao.d.ts → integrity-B_EDELom.d.ts} +2 -2
- package/dist/{integrity-DUNX9Fao.d.ts.map → integrity-B_EDELom.d.ts.map} +1 -1
- package/dist/internal-BMFSR8Ns.js.map +1 -1
- package/dist/{kind-factory-DY8FdoXf.js → kind-factory-DMeEoMQZ.js} +3 -10
- package/dist/kind-factory-DMeEoMQZ.js.map +1 -0
- package/dist/ledger-core/index.js +2 -2
- package/dist/{ledger-core-BOzlRygb.js → ledger-core-PIfjCbKn.js} +2 -2
- package/dist/{ledger-core-BOzlRygb.js.map → ledger-core-PIfjCbKn.js.map} +1 -1
- package/dist/{llm-judge-Du7WQPh7.js → llm-judge-BtJ2Sfk_.js} +101 -20
- package/dist/llm-judge-BtJ2Sfk_.js.map +1 -0
- package/dist/matrix/index.d.ts +2 -2
- package/dist/{matrix-eXKRMHnL.d.ts → matrix-DiHmUobV.d.ts} +3 -3
- package/dist/{matrix-eXKRMHnL.d.ts.map → matrix-DiHmUobV.d.ts.map} +1 -1
- package/dist/meta-eval/index.d.ts +3 -3
- package/dist/meta-eval/index.js +1 -1
- package/dist/{mint-DfODW1KW.js → mint-DjfDUMHr.js} +2 -2
- package/dist/{mint-DfODW1KW.js.map → mint-DjfDUMHr.js.map} +1 -1
- package/dist/multishot/golden/index.d.ts +1 -1
- package/dist/multishot/index.d.ts +2 -2
- package/dist/openapi.json +4 -4
- package/dist/pipelines/index.d.ts +5 -5
- package/dist/pipelines/index.js +3 -3
- package/dist/{pareto-BqNW3LJR.d.ts → power-preflight-Ptse_Kq7.d.ts} +43 -43
- package/dist/power-preflight-Ptse_Kq7.d.ts.map +1 -0
- package/dist/{pre-registration-KN9jkh58.js → pre-registration-D94b7Of5.js} +2 -2
- package/dist/{pre-registration-KN9jkh58.js.map → pre-registration-D94b7Of5.js.map} +1 -1
- package/dist/{pre-registration-CzFCcwYk.d.ts → pre-registration-DHz6P_6f.d.ts} +2 -2
- package/dist/{pre-registration-CzFCcwYk.d.ts.map → pre-registration-DHz6P_6f.d.ts.map} +1 -1
- package/dist/{produced-state-Be0BK3RN.js → produced-state-Cm6DU_Ao.js} +4 -4
- package/dist/{produced-state-Be0BK3RN.js.map → produced-state-Cm6DU_Ao.js.map} +1 -1
- package/dist/profile-cell.js +1 -1
- package/dist/{promotion-policy-DtnOIZvk.d.ts → promotion-policy-WSXtBgBb.d.ts} +3 -3
- package/dist/{promotion-policy-DtnOIZvk.d.ts.map → promotion-policy-WSXtBgBb.d.ts.map} +1 -1
- package/dist/{run-score-lDzV0X8j.js → proposal-findings-bko3GGy-.js} +2 -31
- package/dist/proposal-findings-bko3GGy-.js.map +1 -0
- package/dist/{transient-failure-DKF5Mofa.d.ts → provenance-CafMdZKM.d.ts} +895 -891
- package/dist/provenance-CafMdZKM.d.ts.map +1 -0
- package/dist/{query-_5g6re3_.js → query-BPGMVlbM.js} +3 -3
- package/dist/{query-_5g6re3_.js.map → query-BPGMVlbM.js.map} +1 -1
- package/dist/{query-CwnHlu5p.d.ts → query-Na5gEIGd.d.ts} +3 -3
- package/dist/{query-CwnHlu5p.d.ts.map → query-Na5gEIGd.d.ts.map} +1 -1
- package/dist/{registry-8You7OK1.d.ts → registry-xEb_xfns.d.ts} +3 -3
- package/dist/{registry-8You7OK1.d.ts.map → registry-xEb_xfns.d.ts.map} +1 -1
- package/dist/{release-confidence-nGDJiiwc.js → release-confidence-CzUHc4z4.js} +3 -3
- package/dist/{release-confidence-nGDJiiwc.js.map → release-confidence-CzUHc4z4.js.map} +1 -1
- package/dist/{release-confidence-Dqt0NFep.d.ts → release-confidence-D6lQw_o7.d.ts} +4 -4
- package/dist/{release-confidence-Dqt0NFep.d.ts.map → release-confidence-D6lQw_o7.d.ts.map} +1 -1
- package/dist/reporting.d.ts +3 -3
- package/dist/reporting.js +2 -2
- package/dist/{researcher-Cz565b7D.d.ts → researcher-CMUTQXD7.d.ts} +6 -6
- package/dist/{researcher-Cz565b7D.d.ts.map → researcher-CMUTQXD7.d.ts.map} +1 -1
- package/dist/{reward-hacking-MBf7qpSB.d.ts → reward-hacking-CgPRUesA.d.ts} +2 -2
- package/dist/{reward-hacking-MBf7qpSB.d.ts.map → reward-hacking-CgPRUesA.d.ts.map} +1 -1
- package/dist/{reward-hacking-O5zKDANP.js → reward-hacking-SkxYgT0x.js} +2 -2
- package/dist/{reward-hacking-O5zKDANP.js.map → reward-hacking-SkxYgT0x.js.map} +1 -1
- package/dist/rl.d.ts +8 -8
- package/dist/rl.d.ts.map +1 -1
- package/dist/rl.js +6 -5
- package/dist/rl.js.map +1 -1
- package/dist/rollout/index.d.ts +1 -1
- package/dist/rollout/index.js +2 -2
- package/dist/{rollout-Dm2tSdiQ.js → rollout-Crypdx8s.js} +2 -2
- package/dist/{rollout-Dm2tSdiQ.js.map → rollout-Crypdx8s.js.map} +1 -1
- package/dist/{rubric-predictive-validity-CxycqzX5.d.ts → rubric-predictive-validity-DluJLCKQ.d.ts} +2 -2
- package/dist/{rubric-predictive-validity-CxycqzX5.d.ts.map → rubric-predictive-validity-DluJLCKQ.d.ts.map} +1 -1
- package/dist/{run-record-BC0ebuRP.js → run-record-DLORoL7t.js} +2 -2
- package/dist/{run-record-BC0ebuRP.js.map → run-record-DLORoL7t.js.map} +1 -1
- package/dist/{run-record-VVy4T9OW.d.ts → run-record-DQjRcYwA.d.ts} +3 -3
- package/dist/{run-record-VVy4T9OW.d.ts.map → run-record-DQjRcYwA.d.ts.map} +1 -1
- package/dist/{schema-k6ZBftVv.js → schema-CdIX2aHu.js} +5 -1
- package/dist/{schema-k6ZBftVv.js.map → schema-CdIX2aHu.js.map} +1 -1
- package/dist/{schema-Bjgdsn73.d.ts → schema-DID1Cqct.d.ts} +7 -3
- package/dist/{schema-Bjgdsn73.d.ts.map → schema-DID1Cqct.d.ts.map} +1 -1
- package/dist/{semantic-concept-judge-BsDMOwJr.js → semantic-concept-judge-I36eejJx.js} +2 -2
- package/dist/{semantic-concept-judge-BsDMOwJr.js.map → semantic-concept-judge-I36eejJx.js.map} +1 -1
- package/dist/{sequential-BLMbdrD7.js → sequential-B51qAYE4.js} +2 -2
- package/dist/{sequential-BLMbdrD7.js.map → sequential-B51qAYE4.js.map} +1 -1
- package/dist/{server-BjYiJHoJ.js → server-CCEnywOR.js} +26 -18
- package/dist/server-CCEnywOR.js.map +1 -0
- package/dist/{skillopt-optimization-method-UArRo-nr.js → skillopt-optimization-method-DzlF2RM7.js} +9 -9
- package/dist/{skillopt-optimization-method-UArRo-nr.js.map → skillopt-optimization-method-DzlF2RM7.js.map} +1 -1
- package/dist/{statistical-heldout-Cy3EhjlC.d.ts → statistical-heldout-UhiexnjU.d.ts} +3 -3
- package/dist/{statistical-heldout-Cy3EhjlC.d.ts.map → statistical-heldout-UhiexnjU.d.ts.map} +1 -1
- package/dist/{store-B06JdC56.d.ts → store-Cq9oOrI1.d.ts} +2 -2
- package/dist/{store-B06JdC56.d.ts.map → store-Cq9oOrI1.d.ts.map} +1 -1
- package/dist/{store-otlp-C_Rq5I4D.js → store-otlp-CHjBvWQY.js} +2 -2
- package/dist/{store-otlp-C_Rq5I4D.js.map → store-otlp-CHjBvWQY.js.map} +1 -1
- package/dist/{store-tool-spans-BVga3c37.js → store-tool-spans-B9o6tU8f.js} +3 -3
- package/dist/{store-tool-spans-BVga3c37.js.map → store-tool-spans-B9o6tU8f.js.map} +1 -1
- package/dist/{store-tool-spans-DPUG7UUY.d.ts → store-tool-spans-D_qMl2__.d.ts} +102 -36
- package/dist/store-tool-spans-D_qMl2__.d.ts.map +1 -0
- package/dist/storyboard/index.d.ts +1 -1
- package/dist/{summary-report-BXeQ5Ues.js → summary-report-Bgh8CpNK.js} +2 -2
- package/dist/{summary-report-BXeQ5Ues.js.map → summary-report-Bgh8CpNK.js.map} +1 -1
- package/dist/{summary-report-CC07PhEL.d.ts → summary-report-DRstQNBX.d.ts} +3 -3
- package/dist/{summary-report-CC07PhEL.d.ts.map → summary-report-DRstQNBX.d.ts.map} +1 -1
- package/dist/{task-failure-attributes-DTl-7-Kw.js → task-failure-attributes-CBGtLS_H.js} +3 -3
- package/dist/{task-failure-attributes-DTl-7-Kw.js.map → task-failure-attributes-CBGtLS_H.js.map} +1 -1
- package/dist/{tool-groups-Ci8i9ErB.d.ts → tool-groups-RGYfVWpc.d.ts} +3 -3
- package/dist/tool-groups-RGYfVWpc.d.ts.map +1 -0
- package/dist/{tool-waste-CKc7bYIg.d.ts → tool-waste-BrmLKxMw.d.ts} +4 -4
- package/dist/{tool-waste-CKc7bYIg.d.ts.map → tool-waste-BrmLKxMw.d.ts.map} +1 -1
- package/dist/{tool-waste-8BQiUc8K.js → tool-waste-CwGHzBzX.js} +2 -2
- package/dist/{tool-waste-8BQiUc8K.js.map → tool-waste-CwGHzBzX.js.map} +1 -1
- package/dist/trace-repair/index.d.ts +3 -3
- package/dist/trace-repair/index.d.ts.map +1 -1
- package/dist/trace-repair/index.js +4 -3
- package/dist/trace-repair/index.js.map +1 -1
- package/dist/traces.d.ts +59 -61
- package/dist/traces.d.ts.map +1 -1
- package/dist/traces.js +7 -7
- package/dist/{trajectory-Bi157Gun.d.ts → trajectory-r1bQqvBQ.d.ts} +3 -3
- package/dist/{trajectory-Bi157Gun.d.ts.map → trajectory-r1bQqvBQ.d.ts.map} +1 -1
- package/dist/trajectory-replay/index.d.ts +3 -3
- package/dist/trajectory-replay/index.js +1 -1
- package/dist/{types-BPb2Kf_C.d.ts → types-CMyW4GnH.d.ts} +3 -3
- package/dist/{types-BPb2Kf_C.d.ts.map → types-CMyW4GnH.d.ts.map} +1 -1
- package/dist/{types-BI4fT3HN.js → types-CiWITkGo.js} +11 -2
- package/dist/types-CiWITkGo.js.map +1 -0
- package/dist/{types-D9ssmxKL.d.ts → types-DMoNFDWi.d.ts} +6 -3
- package/dist/{types-D9ssmxKL.d.ts.map → types-DMoNFDWi.d.ts.map} +1 -1
- package/dist/{types-D4s7Z6nq.d.ts → types-JHMOqZI4.d.ts} +13 -3
- package/dist/types-JHMOqZI4.d.ts.map +1 -0
- package/dist/{verdict-B0xltqu6.js → verdict-BQ3pCFf8.js} +2 -2
- package/dist/{verdict-B0xltqu6.js.map → verdict-BQ3pCFf8.js.map} +1 -1
- package/dist/{verdict-cache-CdVVTVmn.js → verdict-cache-B3eCVQtY.js} +2 -2
- package/dist/{verdict-cache-CdVVTVmn.js.map → verdict-cache-B3eCVQtY.js.map} +1 -1
- package/dist/wire/index.d.ts +26 -11
- package/dist/wire/index.d.ts.map +1 -1
- package/dist/wire/index.js +2 -2
- package/docs/campaign-proposers.md +4 -0
- package/docs/code-agent-intake.md +64 -0
- package/docs/concepts.md +1 -1
- package/docs/design/statistics-decisions.md +1 -1
- package/docs/distributed-driver.md +3 -6
- package/docs/public-api.md +122 -106
- package/docs/wire-protocol.md +5 -3
- package/package.json +27 -19
- package/dist/benchmark-BhT16ep9.js.map +0 -1
- package/dist/benchmark-command-CF-4GEWZ.js.map +0 -1
- package/dist/campaign-DQZmc2Dq.js.map +0 -1
- package/dist/capture-fetch-CqwsJkkG.d.ts +0 -68
- package/dist/capture-fetch-CqwsJkkG.d.ts.map +0 -1
- package/dist/contract/index.d.ts.map +0 -1
- package/dist/dspy-rlm-engine-DhA9qKIm.js.map +0 -1
- package/dist/external-optimizer-process-BFmh36vW.js.map +0 -1
- package/dist/external-optimizer-subprocess-CqLMW3nh.js.map +0 -1
- package/dist/index-CGtH1piv.d.ts.map +0 -1
- package/dist/index-D-V8gCs_.d.ts.map +0 -1
- package/dist/index-vrJugRal.d.ts +0 -1
- package/dist/kind-factory-DY8FdoXf.js.map +0 -1
- package/dist/llm-judge-Du7WQPh7.js.map +0 -1
- package/dist/pareto-BqNW3LJR.d.ts.map +0 -1
- package/dist/run-score-lDzV0X8j.js.map +0 -1
- package/dist/server-BjYiJHoJ.js.map +0 -1
- package/dist/skillopt-optimization-method-x7TTF23P.d.ts.map +0 -1
- package/dist/store-tool-spans-DPUG7UUY.d.ts.map +0 -1
- package/dist/tool-groups-Ci8i9ErB.d.ts.map +0 -1
- package/dist/transient-failure-DKF5Mofa.d.ts.map +0 -1
- package/dist/types-BI4fT3HN.js.map +0 -1
- package/dist/types-D4s7Z6nq.d.ts.map +0 -1
|
@@ -1,10 +1,22 @@
|
|
|
1
1
|
import { c as CostLedgerHandle } from "./cost-ledger-DbQdN3nO.js";
|
|
2
|
-
import { a as RunRecord } from "./run-record-
|
|
3
|
-
import {
|
|
2
|
+
import { a as RunRecord } from "./run-record-DQjRcYwA.js";
|
|
3
|
+
import { A as CanaryOptions, F as OptimizationMethod, Qn as ServedModelPolicy, U as OptimizationPackageSource, b as RedTeamCase, yt as SearchRunIdentity } from "./provenance-CafMdZKM.js";
|
|
4
4
|
import { C as LlmCallMetadata, p as ChatClient } from "./types-Bfk0uxRj.js";
|
|
5
|
+
import { R as Scenario, S as JudgeConfig, i as CampaignCostMeter, j as MutableSurface, o as CampaignScenarioIdentity, p as Gate } from "./types-JHMOqZI4.js";
|
|
5
6
|
import { _ as ExternalTextCandidate, c as ExternalOptimizerModelBudget, h as ExternalOptimizerRunnerCommand, l as ExternalOptimizerModelCall, m as ExternalOptimizerResumeMode, r as ExternalOptimizerCallbackLimits, v as ExternalTextEvaluationRequest } from "./external-optimizer-contracts-szBJ_1vh.js";
|
|
6
|
-
import {
|
|
7
|
-
|
|
7
|
+
import { t as DetectRewardHackingInput } from "./reward-hacking-CgPRUesA.js";
|
|
8
|
+
//#region src/campaign/coverage.d.ts
|
|
9
|
+
/** Reject campaign designs whose denominator cannot be identified exactly. */
|
|
10
|
+
declare function assertCampaignDesign<TScenario extends Scenario>(scenarios: readonly TScenario[], reps: number): void;
|
|
11
|
+
/** Redacted but independently verifiable identity of one complete scenario. */
|
|
12
|
+
declare function campaignScenarioIdentity<TScenario extends Scenario>(scenario: TScenario): CampaignScenarioIdentity & Pick<TScenario, 'id' | 'kind'>;
|
|
13
|
+
/** Canonical split identity reconstructed from redacted scenario identities. */
|
|
14
|
+
declare function campaignSplitDigestFromIdentities(scenarios: readonly CampaignScenarioIdentity[], reps: number): `sha256:${string}`;
|
|
15
|
+
/** Canonical identity of the exact scenario payloads and replicate count. */
|
|
16
|
+
declare function campaignSplitDigest<TScenario extends Scenario>(scenarios: readonly TScenario[], reps: number): `sha256:${string}`;
|
|
17
|
+
/** Refuse a campaign whose retained task identities contradict its split digest. */
|
|
18
|
+
declare function assertCampaignSplitIdentity(scenarios: readonly CampaignScenarioIdentity[], reps: number, splitDigest: string): void;
|
|
19
|
+
//#endregion
|
|
8
20
|
//#region src/reference-equivalence-judge.d.ts
|
|
9
21
|
declare const REFERENCE_EQUIVALENCE_JUDGE_VERSION = "reference-equivalence-judge-v1-2026-07-13";
|
|
10
22
|
declare const REFERENCE_EQUIVALENCE_INPUT_LIMITS: {
|
|
@@ -42,18 +54,6 @@ declare function createReferenceEquivalenceJudge(options: ReferenceEquivalenceJu
|
|
|
42
54
|
/** Direct-call adapter over the campaign judge for product callers. */
|
|
43
55
|
declare function runReferenceEquivalenceJudge(input: ReferenceEquivalenceJudgeInput, options: ReferenceEquivalenceJudgeOptions): Promise<ReferenceEquivalenceJudgeResult>;
|
|
44
56
|
//#endregion
|
|
45
|
-
//#region src/campaign/coverage.d.ts
|
|
46
|
-
/** Reject campaign designs whose denominator cannot be identified exactly. */
|
|
47
|
-
declare function assertCampaignDesign<TScenario extends Scenario>(scenarios: readonly TScenario[], reps: number): void;
|
|
48
|
-
/** Redacted but independently verifiable identity of one complete scenario. */
|
|
49
|
-
declare function campaignScenarioIdentity<TScenario extends Scenario>(scenario: TScenario): CampaignScenarioIdentity & Pick<TScenario, 'id' | 'kind'>;
|
|
50
|
-
/** Canonical split identity reconstructed from redacted scenario identities. */
|
|
51
|
-
declare function campaignSplitDigestFromIdentities(scenarios: readonly CampaignScenarioIdentity[], reps: number): `sha256:${string}`;
|
|
52
|
-
/** Canonical identity of the exact scenario payloads and replicate count. */
|
|
53
|
-
declare function campaignSplitDigest<TScenario extends Scenario>(scenarios: readonly TScenario[], reps: number): `sha256:${string}`;
|
|
54
|
-
/** Refuse a campaign whose retained task identities contradict its split digest. */
|
|
55
|
-
declare function assertCampaignSplitIdentity(scenarios: readonly CampaignScenarioIdentity[], reps: number, splitDigest: string): void;
|
|
56
|
-
//#endregion
|
|
57
57
|
//#region src/campaign/external-text-evaluation.d.ts
|
|
58
58
|
interface ExternalOptimizationExample {
|
|
59
59
|
id: string;
|
|
@@ -145,99 +145,6 @@ interface ExternalTextOptimizationMethodConfig<TScenario extends Scenario, TArti
|
|
|
145
145
|
*/
|
|
146
146
|
declare function externalTextOptimizationMethod<TScenario extends Scenario, TArtifact>(config: ExternalTextOptimizationMethodConfig<TScenario, TArtifact>): OptimizationMethod<TScenario, TArtifact>;
|
|
147
147
|
//#endregion
|
|
148
|
-
//#region src/campaign/gates/compose.d.ts
|
|
149
|
-
/** Compose gates — all must `ship` for the composite to `ship`. First
|
|
150
|
-
* non-ship verdict short-circuits the composite verdict, but ALL gates run
|
|
151
|
-
* (so the result records every gate's reason — useful for diagnostics). */
|
|
152
|
-
declare function composeGate<TArtifact = unknown, TScenario extends Scenario = Scenario>(...gates: Array<Gate<TArtifact, TScenario>>): Gate<TArtifact, TScenario>;
|
|
153
|
-
//#endregion
|
|
154
|
-
//#region src/campaign/gates/default-production-gate.d.ts
|
|
155
|
-
type DefaultProductionGateCheck = 'dimension-regression' | 'budget' | 'red-team' | 'reward-hacking' | 'canary';
|
|
156
|
-
type DefaultProductionRewardHackingOptions = Omit<DetectRewardHackingInput, 'runs' | 'truthOf'> & {
|
|
157
|
-
truthOf: NonNullable<DetectRewardHackingInput['truthOf']>;
|
|
158
|
-
};
|
|
159
|
-
interface DefaultProductionGateOptions {
|
|
160
|
-
/** Required: scenarios held out from training; substrate compares
|
|
161
|
-
* candidate-on-holdout vs baseline-on-holdout. */
|
|
162
|
-
holdoutScenarios: Scenario[];
|
|
163
|
-
/** Minimum held-out lift the **paired-bootstrap CI lower bound** must clear
|
|
164
|
-
* to ship — NOT a point estimate. Default 0 ⇒ "confidently positive at the
|
|
165
|
-
* confidence level". Interpreted in the judge's native composite scale (set
|
|
166
|
-
* e.g. 2 for a 0-100 rubric to require a ≥2-point significant gain). */
|
|
167
|
-
deltaThreshold?: number;
|
|
168
|
-
/** Confidence level for the held-out + dimension bootstraps. Default 0.95. */
|
|
169
|
-
confidence?: number;
|
|
170
|
-
/** Bootstrap resamples. Default 2000. */
|
|
171
|
-
bootstrapResamples?: number;
|
|
172
|
-
/** Fixed bootstrap seed for a deterministic verdict. Default 1337. */
|
|
173
|
-
bootstrapSeed?: number;
|
|
174
|
-
/** Minimum paired holdout observations (scenarios × reps) before a
|
|
175
|
-
* significance claim is allowed. The exact small-sample test may require
|
|
176
|
-
* more observations at the selected confidence. Default 3. */
|
|
177
|
-
minProductiveRuns?: number;
|
|
178
|
-
/** Ship statistic for the held-out significance test. Default `'mean'`
|
|
179
|
-
* (tie-robust — see `heldoutSignificance`). Pass `'median'` for
|
|
180
|
-
* outlier-robustness at the cost of tie-blindness. */
|
|
181
|
-
heldoutStatistic?: 'mean' | 'median';
|
|
182
|
-
/** Critical judge dimensions that must NOT significantly regress even when
|
|
183
|
-
* the net composite rises (anti-Goodhart). The gate HOLDS if any listed
|
|
184
|
-
* dimension's paired-delta CI lower bound < −`regressionTolerance`. E.g.
|
|
185
|
-
* `['hallucination_free']` for a legal agent. */
|
|
186
|
-
criticalDimensions?: string[];
|
|
187
|
-
/** Tolerance for the per-dimension regression guard, in the dimension's
|
|
188
|
-
* native scale. When omitted it auto-scales off observed magnitudes:
|
|
189
|
-
* 0.05 on [0,1], 5 on 0-100. */
|
|
190
|
-
regressionTolerance?: number;
|
|
191
|
-
/** Total $ budget for the complete improvement run. Requires
|
|
192
|
-
* `GateContext.costLedger`; missing or incomplete accounting holds. */
|
|
193
|
-
budgetUsd?: number;
|
|
194
|
-
/** Static artifact-screening cases. Only `expected: 'ignore'` cases without
|
|
195
|
-
* tool assertions are valid because this check does not dispatch case inputs
|
|
196
|
-
* or observe tool calls. */
|
|
197
|
-
redTeamBattery?: RedTeamCase[];
|
|
198
|
-
/** Shared run history, oldest first. Supplying history does not enable either
|
|
199
|
-
* monitoring check; configure `rewardHacking` and/or `canary` explicitly. */
|
|
200
|
-
recentRuns?: RunRecord[];
|
|
201
|
-
/** Enable reward-hacking monitoring with a caller-owned independent truth channel. */
|
|
202
|
-
rewardHacking?: DefaultProductionRewardHackingOptions;
|
|
203
|
-
/** Enable canary monitoring. Pass `{}` to use the canary defaults. */
|
|
204
|
-
canary?: CanaryOptions;
|
|
205
|
-
/** Optional checks that must be evaluated even when their normal input is
|
|
206
|
-
* absent. Configuring a check's input also makes that check required.
|
|
207
|
-
* Missing evidence always records `not_evaluated`; required unevaluated
|
|
208
|
-
* checks hold the release decision. Held-out significance is always required. */
|
|
209
|
-
requiredChecks?: DefaultProductionGateCheck[];
|
|
210
|
-
}
|
|
211
|
-
/**
|
|
212
|
-
* Opinionated production gate composing held-out significance, red-team, reward-hacking, and canary checks into a single `Gate.decide` decision.
|
|
213
|
-
*/
|
|
214
|
-
declare function defaultProductionGate<TArtifact, TScenario extends Scenario>(options: DefaultProductionGateOptions): Gate<TArtifact, TScenario>;
|
|
215
|
-
//#endregion
|
|
216
|
-
//#region src/campaign/gates/heldout-gate.d.ts
|
|
217
|
-
interface HeldOutGateOptions<TScenario extends Scenario = Scenario> {
|
|
218
|
-
scenarios: TScenario[];
|
|
219
|
-
/** Effect-size threshold the CI lower bound must clear, in the judge's native
|
|
220
|
-
* scale. Default 0.5. Equality holds; CI.low must be greater than this value. */
|
|
221
|
-
deltaThreshold?: number;
|
|
222
|
-
/** Bootstrap CI confidence. Default 0.95. */
|
|
223
|
-
confidence?: number;
|
|
224
|
-
/** Minimum paired holdout observations to claim significance. The exact
|
|
225
|
-
* small-sample test may require more observations at the selected
|
|
226
|
-
* confidence. Default 3. */
|
|
227
|
-
minProductiveRuns?: number;
|
|
228
|
-
/** Bootstrap resamples. Default 2000. */
|
|
229
|
-
resamples?: number;
|
|
230
|
-
/** Fixed bootstrap seed for deterministic verdicts. Default 1337. */
|
|
231
|
-
bootstrapSeed?: number;
|
|
232
|
-
}
|
|
233
|
-
/**
|
|
234
|
-
* Composable held-out gate: ships only when the lower bound of the DECIDING
|
|
235
|
-
* paired interval on the candidate-minus-baseline composite delta clears
|
|
236
|
-
* `deltaThreshold` — Tango's score interval on a pass/fail holdout, the mean
|
|
237
|
-
* bootstrap otherwise. See {@link decidePairedPromotion}.
|
|
238
|
-
*/
|
|
239
|
-
declare function heldOutGate<TArtifact, TScenario extends Scenario>(options: HeldOutGateOptions<TScenario>): Gate<TArtifact, TScenario>;
|
|
240
|
-
//#endregion
|
|
241
148
|
//#region src/campaign/optimizer-model.d.ts
|
|
242
149
|
type OptimizerModelBudget = ExternalOptimizerModelBudget;
|
|
243
150
|
/** One metered model path supplied by the package that owns execution. */
|
|
@@ -449,5 +356,98 @@ interface SkillOptOptimizationMethodConfig<TScenario extends Scenario, TArtifact
|
|
|
449
356
|
/** Run Microsoft's SkillOpt trainer as a complete optimization method. */
|
|
450
357
|
declare function skillOptOptimizationMethod<TScenario extends Scenario, TArtifact>(config: SkillOptOptimizationMethodConfig<TScenario, TArtifact>): OptimizationMethod<TScenario, TArtifact>;
|
|
451
358
|
//#endregion
|
|
452
|
-
|
|
453
|
-
|
|
359
|
+
//#region src/campaign/gates/compose.d.ts
|
|
360
|
+
/** Compose gates — all must `ship` for the composite to `ship`. First
|
|
361
|
+
* non-ship verdict short-circuits the composite verdict, but ALL gates run
|
|
362
|
+
* (so the result records every gate's reason — useful for diagnostics). */
|
|
363
|
+
declare function composeGate<TArtifact = unknown, TScenario extends Scenario = Scenario>(...gates: Array<Gate<TArtifact, TScenario>>): Gate<TArtifact, TScenario>;
|
|
364
|
+
//#endregion
|
|
365
|
+
//#region src/campaign/gates/default-production-gate.d.ts
|
|
366
|
+
type DefaultProductionGateCheck = 'dimension-regression' | 'budget' | 'red-team' | 'reward-hacking' | 'canary';
|
|
367
|
+
type DefaultProductionRewardHackingOptions = Omit<DetectRewardHackingInput, 'runs' | 'truthOf'> & {
|
|
368
|
+
truthOf: NonNullable<DetectRewardHackingInput['truthOf']>;
|
|
369
|
+
};
|
|
370
|
+
interface DefaultProductionGateOptions {
|
|
371
|
+
/** Required: scenarios held out from training; substrate compares
|
|
372
|
+
* candidate-on-holdout vs baseline-on-holdout. */
|
|
373
|
+
holdoutScenarios: Scenario[];
|
|
374
|
+
/** Minimum held-out lift the **paired-bootstrap CI lower bound** must clear
|
|
375
|
+
* to ship — NOT a point estimate. Default 0 ⇒ "confidently positive at the
|
|
376
|
+
* confidence level". Interpreted in the judge's native composite scale (set
|
|
377
|
+
* e.g. 2 for a 0-100 rubric to require a ≥2-point significant gain). */
|
|
378
|
+
deltaThreshold?: number;
|
|
379
|
+
/** Confidence level for the held-out + dimension bootstraps. Default 0.95. */
|
|
380
|
+
confidence?: number;
|
|
381
|
+
/** Bootstrap resamples. Default 2000. */
|
|
382
|
+
bootstrapResamples?: number;
|
|
383
|
+
/** Fixed bootstrap seed for a deterministic verdict. Default 1337. */
|
|
384
|
+
bootstrapSeed?: number;
|
|
385
|
+
/** Minimum paired holdout observations (scenarios × reps) before a
|
|
386
|
+
* significance claim is allowed. The exact small-sample test may require
|
|
387
|
+
* more observations at the selected confidence. Default 3. */
|
|
388
|
+
minProductiveRuns?: number;
|
|
389
|
+
/** Ship statistic for the held-out significance test. Default `'mean'`
|
|
390
|
+
* (tie-robust — see `heldoutSignificance`). Pass `'median'` for
|
|
391
|
+
* outlier-robustness at the cost of tie-blindness. */
|
|
392
|
+
heldoutStatistic?: 'mean' | 'median';
|
|
393
|
+
/** Critical judge dimensions that must NOT significantly regress even when
|
|
394
|
+
* the net composite rises (anti-Goodhart). The gate HOLDS if any listed
|
|
395
|
+
* dimension's paired-delta CI lower bound < −`regressionTolerance`. E.g.
|
|
396
|
+
* `['hallucination_free']` for a legal agent. */
|
|
397
|
+
criticalDimensions?: string[];
|
|
398
|
+
/** Tolerance for the per-dimension regression guard, in the dimension's
|
|
399
|
+
* native scale. When omitted it auto-scales off observed magnitudes:
|
|
400
|
+
* 0.05 on [0,1], 5 on 0-100. */
|
|
401
|
+
regressionTolerance?: number;
|
|
402
|
+
/** Total $ budget for the complete improvement run. Requires
|
|
403
|
+
* `GateContext.costLedger`; missing or incomplete accounting holds. */
|
|
404
|
+
budgetUsd?: number;
|
|
405
|
+
/** Static artifact-screening cases. Only `expected: 'ignore'` cases without
|
|
406
|
+
* tool assertions are valid because this check does not dispatch case inputs
|
|
407
|
+
* or observe tool calls. */
|
|
408
|
+
redTeamBattery?: RedTeamCase[];
|
|
409
|
+
/** Shared run history, oldest first. Supplying history does not enable either
|
|
410
|
+
* monitoring check; configure `rewardHacking` and/or `canary` explicitly. */
|
|
411
|
+
recentRuns?: RunRecord[];
|
|
412
|
+
/** Enable reward-hacking monitoring with a caller-owned independent truth channel. */
|
|
413
|
+
rewardHacking?: DefaultProductionRewardHackingOptions;
|
|
414
|
+
/** Enable canary monitoring. Pass `{}` to use the canary defaults. */
|
|
415
|
+
canary?: CanaryOptions;
|
|
416
|
+
/** Optional checks that must be evaluated even when their normal input is
|
|
417
|
+
* absent. Configuring a check's input also makes that check required.
|
|
418
|
+
* Missing evidence always records `not_evaluated`; required unevaluated
|
|
419
|
+
* checks hold the release decision. Held-out significance is always required. */
|
|
420
|
+
requiredChecks?: DefaultProductionGateCheck[];
|
|
421
|
+
}
|
|
422
|
+
/**
|
|
423
|
+
* Opinionated production gate composing held-out significance, red-team, reward-hacking, and canary checks into a single `Gate.decide` decision.
|
|
424
|
+
*/
|
|
425
|
+
declare function defaultProductionGate<TArtifact, TScenario extends Scenario>(options: DefaultProductionGateOptions): Gate<TArtifact, TScenario>;
|
|
426
|
+
//#endregion
|
|
427
|
+
//#region src/campaign/gates/heldout-gate.d.ts
|
|
428
|
+
interface HeldOutGateOptions<TScenario extends Scenario = Scenario> {
|
|
429
|
+
scenarios: TScenario[];
|
|
430
|
+
/** Effect-size threshold the CI lower bound must clear, in the judge's native
|
|
431
|
+
* scale. Default 0.5. Equality holds; CI.low must be greater than this value. */
|
|
432
|
+
deltaThreshold?: number;
|
|
433
|
+
/** Bootstrap CI confidence. Default 0.95. */
|
|
434
|
+
confidence?: number;
|
|
435
|
+
/** Minimum paired holdout observations to claim significance. The exact
|
|
436
|
+
* small-sample test may require more observations at the selected
|
|
437
|
+
* confidence. Default 3. */
|
|
438
|
+
minProductiveRuns?: number;
|
|
439
|
+
/** Bootstrap resamples. Default 2000. */
|
|
440
|
+
resamples?: number;
|
|
441
|
+
/** Fixed bootstrap seed for deterministic verdicts. Default 1337. */
|
|
442
|
+
bootstrapSeed?: number;
|
|
443
|
+
}
|
|
444
|
+
/**
|
|
445
|
+
* Composable held-out gate: ships only when the lower bound of the DECIDING
|
|
446
|
+
* paired interval on the candidate-minus-baseline composite delta clears
|
|
447
|
+
* `deltaThreshold` — Tango's score interval on a pass/fail holdout, the mean
|
|
448
|
+
* bootstrap otherwise. See {@link decidePairedPromotion}.
|
|
449
|
+
*/
|
|
450
|
+
declare function heldOutGate<TArtifact, TScenario extends Scenario>(options: HeldOutGateOptions<TScenario>): Gate<TArtifact, TScenario>;
|
|
451
|
+
//#endregion
|
|
452
|
+
export { ReferenceEquivalenceJudgeInput as A, campaignSplitDigestFromIdentities as B, ExternalTextOptimizerContext as C, decodeExternalTextCandidate as D, ExternalTextEvaluationResponse as E, runReferenceEquivalenceJudge as F, assertCampaignDesign as I, assertCampaignSplitIdentity as L, ReferenceEquivalenceJudgeResult as M, ReferenceEquivalenceScenario as N, REFERENCE_EQUIVALENCE_INPUT_LIMITS as O, createReferenceEquivalenceJudge as P, campaignScenarioIdentity as R, ExternalTextOptimizationMethodConfig as S, ExternalOptimizationExample as T, GepaRunnerCommand as _, DefaultProductionRewardHackingOptions as a, OptimizerModelBudget as b, SkillOptOptimizationMethodConfig as c, skillOptOptimizationMethod as d, GepaAdaptiveEngineRun as f, GepaOptimizationRecipe as g, GepaOptimizationMethodConfig as h, DefaultProductionGateOptions as i, ReferenceEquivalenceJudgeOptions as j, REFERENCE_EQUIVALENCE_JUDGE_VERSION as k, SkillOptRunnerCommand as l, GepaEngineRun as m, heldOutGate as n, defaultProductionGate as o, GepaEngineOptions as p, DefaultProductionGateCheck as r, composeGate as s, HeldOutGateOptions as t, SkillOptTrainerConfig as u, gepaOptimizationMethod as v, ExternalTextOptimizerResult as w, externalTextOptimizationMethod as x, OpenAICompatibleOptimizerModel as y, campaignSplitDigest as z };
|
|
453
|
+
//# sourceMappingURL=heldout-gate-Dh2b62w8.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"heldout-gate-Dh2b62w8.d.ts","names":[],"sources":["../src/campaign/coverage.ts","../src/reference-equivalence-judge.ts","../src/campaign/external-text-evaluation.ts","../src/campaign/external-text-optimization-contract.ts","../src/campaign/external-text-optimization.ts","../src/campaign/optimizer-model.ts","../src/campaign/gepa-optimization-method.ts","../src/campaign/skillopt-optimization-method.ts","../src/campaign/gates/compose.ts","../src/campaign/gates/default-production-gate.ts","../src/campaign/gates/heldout-gate.ts"],"mappings":";;;;;;;;;iBAWgB,qBAAqB,kBAAkB,UACrD,oBAAoB,aACpB;;iBA2Bc,yBAAyB,kBAAkB,UACzD,UAAU,YACT,2BAA2B,KAAK;;iBAUnB,kCACd,oBAAoB,4BACpB;;iBAgBc,oBAAoB,kBAAkB,UACpD,oBAAoB,aACpB;;iBAOc,4BACd,oBAAoB,4BACpB,cACA;;;cC3EW;cAEA;WACX;WACA;WACA;;UAGe,qCAAqC;EACpD;EACA;;UAGe;EACf;EACA;EACA;;UAGe;;EAEf,MAAM;;EAEN;;EAEA,SAAS;;EAET,aAAa;;UAGE,wCAAwC;EACvD;EACA;EACA;EACA;;;iBA0Bc,gCACd,SAAS,mCACR,oBAAoB;;iBA2CD,6BACpB,OAAO,gCACP,SAAS,mCACR,QAAQ;;;UCrGM;EACf;EACA;;UAGe;EACf;EACA;IACE;IACA,YAAY;IACZ;IACA;;;IAGA;IACA;MACE;MACA;;;;iBAwHU,4BAA4B,WAAW,wBAAwB;;;UC3I9D;WACN;WACA;WACA;WACA;WACA;WACA,eAAe;WACf,mBAAmB;WACnB,uBAAuB;WACvB;WACA;;WAEA;WACA;WACA;WACA,QAAQ;;WAER,MAAM;WACN,WACP,SAFa,kCAGV,QAAQ;;UAGE;EACf,eAAe;EACf;EACA;IACM;;IACA;;IACA;IAAkB;;;;;;;;;UAqBT,qCACf,kBAAkB,UAClB;EAEA;EACA,QAAQ,KAAK;EACb;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA,SAAS;EACT;EACA;EACA,oBAAoB,UAAU;EAC9B,oBAAoB,UAAU,WAAW,UAAU;EACnD,MAAM,SAAS,iCAAiC,QAAQ;;;;;;;;;;iBCN1C,+BAA+B,kBAAkB,UAAU,WACzE,QAAQ,qCAAqC,WAAW,aACvD,mBAAmB,WAAW;;;KCvErB,uBAAuB;;UAGlB;EACf;EACA,QAAQ;;EAER,MAAM;;EAEN;;;;;;;EAOA,oBAAoB;;;;;;;EAOpB;;;;;UCyCe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;EAKA,eAAe;;;UAIA,sBAAsB;;EAErC;;;KAIU,wBAAwB;;;;;;;KAQxB;EAEN;EACA,KAAK;;EAGL;EACA,eAAe;;EAGf;EACA,eAAe;;EAEf;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;;EAGA;EACA,eAAe;EACf;;EAGA;EACA,eAAe;EACf;;EAGA;EACA,kBAAkB;EAClB,cAAc;EACd;;;KAIM,oBAAoB;UAEf,6BAA6B,kBAAkB,UAAU;;EAExE;;EAEA,QAAQ;;EAER;;EAEA;;EAEA;;;;;EAKA;;EAEA;;EAEA;;EAEA,2BAA2B,QAAQ;;EAEnC;;;;;;EAMA,YAAY;;;;;;EAMZ,oBAAoB,UAAU;;EAE9B,oBAAoB,UAAU,WAAW,UAAU;;EAEnD,SAAS;;;;;EAKT;EACA,SAAS;;;;;;;;;;EAUT;IAAiB,UAAU;IAAmB;;;iBA6ChC,uBAAuB,kBAAkB,UAAU,WACjE,QAAQ,6BAA6B,WAAW,aAC/C,mBAAmB,WAAW;;;UC7LhB;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;;;;EAKA,YAAY;;KAGF,wBAAwB;UAEnB,iCAAiC,kBAAkB,UAAU;EAC5E;;EAEA;EACA;;EAEA;EACA,SAAS;;;;;EAKT,WAAW;;EAEX;;EAEA;EACA;;EAEA;;EAEA,2BAA2B,QAAQ;EACnC;EACA,oBAAoB,UAAU;EAC9B,oBAAoB,UAAU,WAAW,UAAU;EACnD,SAAS;EACT,SAAS;;;iBAIK,2BAA2B,kBAAkB,UAAU,WACrE,QAAQ,iCAAiC,WAAW,aACnD,mBAAmB,WAAW;;;;;;iBC1GjB,YAAY,qBAAqB,kBAAkB,WAAW,aACzE,OAAO,MAAM,KAAK,WAAW,cAC/B,KAAK,WAAW;;;KCaP;KAOA,wCAAwC,KAClD;EAGA,SAAS,YAAY;;UAGN;;;EAGf,kBAAkB;;;;;EAKlB;;EAEA;;EAEA;;EAEA;;;;EAIA;;;;EAIA;;;;;EAKA;;;;EAIA;;;EAGA;;;;EAIA,iBAAiB;;;EAGjB,aAAa;;EAEb,gBAAgB;;EAEhB,SAAS;;;;;EAKT,iBAAiB;;;;;iBAMH,sBAAsB,WAAW,kBAAkB,UACjE,SAAS,+BACR,KAAK,WAAW;;;UC/EF,mBAAmB,kBAAkB,WAAW;EAC/D,WAAW;;;EAGX;;EAEA;;;;EAIA;;EAEA;;EAEA;;;;;;;;iBASc,YAAY,WAAW,kBAAkB,UACvD,SAAS,mBAAmB,aAC3B,KAAK,WAAW"}
|
package/dist/hosted/index.d.ts
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { o as InsightReport } from "../insight-report-
|
|
2
|
-
import { _ as UnixNanoTimestamp, a as hostedTenantFromEnv, c as EvalRunGenerationSnapshot, d as HostedIngestHeaders, f as HostedWireVersion, g as TraceSpanEvent, h as IngestTracesRequest, i as hostedClientFromEnv, l as EvalRunStatus, m as IngestResponse, n as HostedTenant, o as EvalRunCellScore, p as IngestEvalRunsRequest, r as createHostedClient, s as EvalRunEvent, t as HostedClient, u as HOSTED_WIRE_VERSION } from "../client-
|
|
1
|
+
import { o as InsightReport } from "../insight-report-08F022xN.js";
|
|
2
|
+
import { _ as UnixNanoTimestamp, a as hostedTenantFromEnv, c as EvalRunGenerationSnapshot, d as HostedIngestHeaders, f as HostedWireVersion, g as TraceSpanEvent, h as IngestTracesRequest, i as hostedClientFromEnv, l as EvalRunStatus, m as IngestResponse, n as HostedTenant, o as EvalRunCellScore, p as IngestEvalRunsRequest, r as createHostedClient, s as EvalRunEvent, t as HostedClient, u as HOSTED_WIRE_VERSION } from "../client-CDtcZ3p9.js";
|
|
3
3
|
import { z } from "zod";
|
|
4
4
|
//#region src/hosted/schemas.d.ts
|
|
5
5
|
declare const UnixNanoTimestampSchema: z.ZodType<UnixNanoTimestamp>;
|
package/dist/hosted/index.js
CHANGED
|
@@ -1,2 +1,2 @@
|
|
|
1
|
-
import { a as IngestEvalRunsRequestSchema, c as InsightReportSchema, d as HOSTED_WIRE_VERSION, i as EvalRunEventSchema, l as TraceSpanEventSchema, n as hostedClientFromEnv, o as IngestResponseSchema, r as hostedTenantFromEnv, s as IngestTracesRequestSchema, t as createHostedClient, u as UnixNanoTimestampSchema } from "../client-
|
|
1
|
+
import { a as IngestEvalRunsRequestSchema, c as InsightReportSchema, d as HOSTED_WIRE_VERSION, i as EvalRunEventSchema, l as TraceSpanEventSchema, n as hostedClientFromEnv, o as IngestResponseSchema, r as hostedTenantFromEnv, s as IngestTracesRequestSchema, t as createHostedClient, u as UnixNanoTimestampSchema } from "../client-BvwNkIRN.js";
|
|
2
2
|
export { EvalRunEventSchema, HOSTED_WIRE_VERSION, IngestEvalRunsRequestSchema, IngestResponseSchema, IngestTracesRequestSchema, InsightReportSchema, TraceSpanEventSchema, UnixNanoTimestampSchema, createHostedClient, hostedClientFromEnv, hostedTenantFromEnv };
|
|
@@ -1,18 +1,26 @@
|
|
|
1
1
|
import { c as ValidationError, t as AgentEvalError } from "./errors-DEE6u6ot.js";
|
|
2
2
|
import { T as RunPaidCallInput, p as CostProvenance } from "./cost-ledger-DbQdN3nO.js";
|
|
3
|
-
import { a as RunRecord, s as RunSplitTag } from "./run-record-
|
|
4
|
-
import {
|
|
5
|
-
import {
|
|
6
|
-
import {
|
|
7
|
-
import
|
|
8
|
-
import {
|
|
9
|
-
import { d as CompletionVerdict, f as CorrectnessChecker, h as ProducedState, n as BackendIntegrityReport, s as RuntimeEventLike, u as CompletionRequirement } from "./backend-integrity-DxuQCu_A.js";
|
|
10
|
-
import { _ as PairedArmsComparison } from "./pre-registration-CzFCcwYk.js";
|
|
3
|
+
import { a as RunRecord, s as RunSplitTag } from "./run-record-DQjRcYwA.js";
|
|
4
|
+
import { f as AnalystUsageReceipt, i as AnalystFinding, l as AnalystRunResult, w as TraceAnalysisStore } from "./types-DMoNFDWi.js";
|
|
5
|
+
import { Fn as CampaignRunPlan, Ln as PlanCampaignRunOptions, Un as CampaignStorage } from "./provenance-CafMdZKM.js";
|
|
6
|
+
import { A as LabeledScenarioWrite, D as LabeledScenarioSampleArgs, E as LabeledScenarioRecord, O as LabeledScenarioSource, R as Scenario, S as JudgeConfig, T as LabelTrust, a as CampaignResult, d as DispatchContext, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, p as Gate, u as ComponentSurface } from "./types-JHMOqZI4.js";
|
|
7
|
+
import "./heldout-gate-Dh2b62w8.js";
|
|
8
|
+
import { _ as PairedArmsComparison } from "./pre-registration-DHz6P_6f.js";
|
|
11
9
|
import { d as PairedBootstrapResult } from "./paired-promotion-decision-CGzg0cI_.js";
|
|
12
|
-
import "./
|
|
13
|
-
import "./
|
|
14
|
-
import "./
|
|
10
|
+
import "./promotion-policy-WSXtBgBb.js";
|
|
11
|
+
import { d as CompletionVerdict, f as CorrectnessChecker, h as ProducedState, n as BackendIntegrityReport, s as RuntimeEventLike, u as CompletionRequirement } from "./backend-integrity-e79K3UPD.js";
|
|
12
|
+
import { _ as AnalystIssueExpectation, a as AnalystBenchmarkLabelState, t as AnalystBenchmarkCase } from "./benchmark-h-h4bfqj.js";
|
|
13
|
+
import { t as AgentProfile$1 } from "./agent-profile-B9_GGsG8.js";
|
|
14
|
+
import "./statistical-heldout-UhiexnjU.js";
|
|
15
15
|
import { AgentProfile } from "@tangle-network/agent-interface";
|
|
16
|
+
//#region src/campaign/search-ledger-errors.d.ts
|
|
17
|
+
/** Base error for invalid search-ledger input or operations. */
|
|
18
|
+
declare class SearchLedgerError extends ValidationError {}
|
|
19
|
+
/** Error raised when durable search-ledger data fails an integrity check. */
|
|
20
|
+
declare class SearchLedgerIntegrityError extends SearchLedgerError {}
|
|
21
|
+
/** Error raised when an event identifier is reused with different content. */
|
|
22
|
+
declare class SearchLedgerConflictError extends SearchLedgerError {}
|
|
23
|
+
//#endregion
|
|
16
24
|
//#region src/campaign/analyst-surface.d.ts
|
|
17
25
|
interface TraceAnalystScenario extends Scenario {
|
|
18
26
|
kind: 'trace-analyst';
|
|
@@ -306,14 +314,6 @@ interface CrossSurfaceInteractionReport<TRow extends CrossSurfaceTaskRow = Cross
|
|
|
306
314
|
*/
|
|
307
315
|
declare function analyzeCrossSurfaceInteractions<TRow extends CrossSurfaceTaskRow>(input: AnalyzeCrossSurfaceInteractionsInput<TRow>): CrossSurfaceInteractionReport<TRow>;
|
|
308
316
|
//#endregion
|
|
309
|
-
//#region src/campaign/search-ledger-errors.d.ts
|
|
310
|
-
/** Base error for invalid search-ledger input or operations. */
|
|
311
|
-
declare class SearchLedgerError extends ValidationError {}
|
|
312
|
-
/** Error raised when durable search-ledger data fails an integrity check. */
|
|
313
|
-
declare class SearchLedgerIntegrityError extends SearchLedgerError {}
|
|
314
|
-
/** Error raised when an event identifier is reused with different content. */
|
|
315
|
-
declare class SearchLedgerConflictError extends SearchLedgerError {}
|
|
316
|
-
//#endregion
|
|
317
317
|
//#region src/campaign/fixtures.d.ts
|
|
318
318
|
type EvalFixtureValidationMode = 'vitest' | 'none';
|
|
319
319
|
interface EvalFixtureFile {
|
|
@@ -988,7 +988,15 @@ declare function acquireSingleRunLock(opts: SingleRunLockOptions): SingleRunLock
|
|
|
988
988
|
//#region src/campaign/surface-identity.d.ts
|
|
989
989
|
/** Validate the immutable identity shape; the owning executor verifies the Git objects and patch. */
|
|
990
990
|
declare function assertCodeSurfaceIdentity(surface: unknown): asserts surface is CodeSurface;
|
|
991
|
-
/**
|
|
991
|
+
/**
|
|
992
|
+
* Deterministic identity material for a component surface.
|
|
993
|
+
*
|
|
994
|
+
* `canonicalString` orders keys by UTF-16 code unit (RFC 8785), which is a
|
|
995
|
+
* property of the value alone. The previous material ordered them with
|
|
996
|
+
* `localeCompare`, which reads the host's collation — so the same surface
|
|
997
|
+
* could produce two different identities on two machines, and the stored
|
|
998
|
+
* identity would stop matching a recomputation of the identical surface.
|
|
999
|
+
*/
|
|
992
1000
|
declare function componentSurfaceIdentityMaterial(surface: ComponentSurface): string;
|
|
993
1001
|
/** Canonical, location-independent identity of a finalized code candidate.
|
|
994
1002
|
* Commit metadata is excluded: two commits with the same base, final tree,
|
|
@@ -1123,5 +1131,5 @@ declare function verifyCodeSurface(surface: CodeSurface, worktreeDir?: string):
|
|
|
1123
1131
|
* identity against the checkout at `worktreeRef`. */
|
|
1124
1132
|
declare function resolveWorktreePath(surface: CodeSurface, worktreeDir?: string): string;
|
|
1125
1133
|
//#endregion
|
|
1126
|
-
export { makePlaybackDispatch as $,
|
|
1127
|
-
//# sourceMappingURL=index-
|
|
1134
|
+
export { makePlaybackDispatch as $, CrossSurfaceEvidenceBreakdown as $t, scoreDiscrimination as A, LoadEvalFixtureScenariosOptions as At, ProfileMatrixSegmentResult as B, CrossSurfaceAttemptCompleteness as Bt, acquireSingleRunLock as C, SearchLedgerIntegrityError as Cn, neutralizationGate as Ct, compareRankKeys as D, EvalFixtureRunPlan as Dt, campaignMeanComposite as E, EvalFixtureLoadOptions as Et, FinalizeProfileMatrixOptions as F, planEvalFixtureRun as Ft, PlaybackContext as G, CrossSurfaceCandidateEvidence as Gt, createProfileMatrixPlan as H, CrossSurfaceBootstrapPolicy as Ht, FinalizedProfileMatrixResult as I, analyzeCrossSurfaceInteractions as It, ScoreboardRenderOptions as J, CrossSurfaceComponent as Jt, PlaybackDriver as K, CrossSurfaceCandidateOutcome as Kt, ProfileMatrixCoverage as L, AnalyzeCrossSurfaceInteractionsInput as Lt, resolveRunDir as M, discoverEvalFixtures as Mt, tangleTracesRoot as N, loadEvalFixture as Nt, DiscriminationScore as O, EvalFixtureScenario as Ot, CreateProfileMatrixPlanOptions as P, loadEvalFixtureScenarios as Pt, UserStoryVerdict as Q, CrossSurfaceEligibility as Qt, ProfileMatrixPlan as R, CrossSurfaceAdditionDecision as Rt, SingleRunLockOptions as S, SearchLedgerError as Sn, NeutralizationGateOptions as St, campaignBreakdown as T, EvalFixtureFile as Tt, finalizeProfileMatrix as U, CrossSurfaceCandidate as Ut, RunProfileMatrixSegmentOptions as V, CrossSurfaceBestSingleSelection as Vt, runProfileMatrixSegment as W, CrossSurfaceCandidateComparison as Wt, ScoreboardSummary as X, CrossSurfaceCompositionStep as Xt, ScoreboardRow as Y, CrossSurfaceComponentEvidence as Yt, UserStory as Z, CrossSurfaceDistribution as Zt, componentSurfaceIdentityMaterial as _, TraceAnalystArtifact as _n, RolloutCall as _t, WorktreeAdapterError as a, CrossSurfaceInteractionTask as an, ProfileMatrixError as at, surfaceHash as b, traceAnalystQualityJudge as bn, classifyUngroundedLiterals as bt, verifyCodeSurface as c, CrossSurfacePairEvidence as cn, RunProfileMatrixResult as ct, PhoenixEvaluationResultLike as d, CrossSurfaceRankedSingle as dn, neutralizeText as dt, CrossSurfaceIneligibilityReason as en, renderScoreboardMarkdown as et, PhoenixEvaluatorLike as f, CrossSurfaceRelativeCost as fn, FsLabeledScenarioStore as ft, codeSurfaceIdentityMaterial as g, BuildTraceAnalystSurfaceDispatchOptions as gn, RolloutArgumentDiffOptions as gt, assertCodeSurfaceIdentity as h, CrossSurfaceTaskRow as hn, RolloutArgumentDiff as ht, WorktreeAdapter as i, CrossSurfaceInteractionReport as in, ProfileDispatchFn as it, selectDiscriminative as j, PlanEvalFixtureRunOptions as jt, ScenarioSignal as k, EvalFixtureValidationMode as kt, AutoevalsScoreLike as l, CrossSurfacePairIncompatibilityReason as ln, ScenarioRollup as lt, phoenixEvaluatorJudge as m, CrossSurfaceSelections as mn, LabeledScenarioStoreError as mt, GitWorktreeAdapterOptions as n, CrossSurfaceInteractionEffect as nn, scoreboardSummary as nt, gitWorktreeAdapter as o, CrossSurfaceNaiveStackSelection as on, ProfileSummary as ot, autoevalsScorerJudge as p, CrossSurfaceSelectionPolicy as pn, FsLabeledScenarioStoreOptions as pt, PlaybackStep as q, CrossSurfaceCandidateSummary as qt, Worktree as r, CrossSurfaceInteractionPath as rn, userStoryScoreboard as rt, resolveWorktreePath as s, CrossSurfacePairCompatibility as sn, RunProfileMatrixOptions as st, CodeSurfaceVerification as t, CrossSurfaceInteractionAwareSelection as tn, scoreUserStory as tt, AutoevalsScorerLike as u, CrossSurfacePairwiseEntry as un, runProfileMatrix as ut, renderSurfaceDiff as v, TraceAnalystScenario as vn, ScoredRollout as vt, CampaignBreakdown as w, EvalFixture as wt, SingleRunLock as x, SearchLedgerConflictError as xn, rolloutArgumentDiff as xt, surfaceContentHash as y, buildTraceAnalystSurfaceDispatch as yn, UngroundedLiteralReport as yt, ProfileMatrixRow as z, CrossSurfaceAdditionRejectionReason as zt };
|
|
1135
|
+
//# sourceMappingURL=index-8VIogTyS.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index-8VIogTyS.d.ts","names":[],"sources":["../src/campaign/search-ledger-errors.ts","../src/campaign/analyst-surface.ts","../src/campaign/cross-surface-types.ts","../src/campaign/cross-surface-interaction.ts","../src/campaign/fixtures.ts","../src/campaign/gates/neutralization-gate.ts","../src/campaign/grounded-reflection.ts","../src/campaign/labeled-store/fs-adapter.ts","../src/campaign/neutralize.ts","../src/campaign/presets/run-profile-matrix.ts","../src/campaign/presets/playback.ts","../src/campaign/presets/segmented-profile-matrix.ts","../src/campaign/run-dir.ts","../src/campaign/scenario-selection.ts","../src/campaign/score-utils.ts","../src/campaign/single-run-lock.ts","../src/campaign/surface-identity.ts","../src/campaign/upstream-evaluators.ts","../src/campaign/worktree/index.ts"],"mappings":";;;;;;;;;;;;;;;;;cAGa,0BAA0B;;cAG1B,mCAAmC;;cAGnC,kCAAkC;;;UCE9B,6BAA6B;EAC5C;EACA,YAAY;EACZ,YAAY;EACZ,yBAAyB;EACzB,kBAAkB;;UAGH;EACf,mBAAmB;EACnB,QAAQ;EACR,MAAM;;UAGS;EACf,QAAQ;IACN;IACA,YAAY;IACZ;IACA,QAAQ;MACN,QAAQ;;iBAGE,iCACd,SAAS,2CAET,SAAS,gBACT,UAAU,sBACV,SAAS,oBACN,QAAQ;iBAgBG,4BAA4B,YAC1C,sBACA;;;;KCtDU;;UAGK;EACf;EACA;;EAEA;;;UAIe;EACf;EACA;EACA;EACA;;;UAIe;EACf;;EAEA;;EAEA;;;;;;;UAQe;EACf;EACA;;EAEA;EACA,cAAc;EACd;EACA;;;;;;EAMA,MAAM;EACN,mBAAmB;;EAEnB;;UAGe;EACf;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;;EAEA,kCAAkC;;EAElC;;UAGe,qCACf,aAAa,sBAAsB;EAEnC,qBAAqB;EACrB,qBAAqB;EACrB,eAAe;EACf;;EAEA;;EAEA;;EAEA;;EAEA;EACA,WAAW;EACX,WAAW;;UAGI;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf,aAAa;EACb;EACA;EACA;EACA;;KAGU;UAWK;EACf;EACA,SAAS;;UAGM;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf,WAAW;EACX,SAAS;EACT,OAAO;EACP,OAAO,eAAe;EACtB,QAAQ;EACR,QAAQ;;EAER,sBAAsB;;EAEtB,aAAa;;UAGE;EACf;EACA;EACA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA,QAAQ;EACR,cAAc,eAAe;;UAGd;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;;EAEA;EACA;;UAGe;EACf,SAAS;EACT;EACA;EACA;EACA;EACA,eAAe;EACf,gBAAgB;;KAGN;UAYK;EACf;EACA,SAAS;EACT;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA,4BAA4B,iCAAiC;EAC7D,wBAAwB,eAAe;EACvC,QAAQ;EACR,QAAQ;EACR,aAAa;EACb,eAAe;;UAGA;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;;UAGM;;EAEf;EACA;;KAGU;UAaK;EACf;EACA;EACA;EACA;EACA;EACA,uBAAuB;EACvB;EACA;EACA,SAAS;;UAGM;EACf;EACA;EACA,YAAY;EACZ;;;UAIe;EACf;EACA;EACA;EACA;EACA,OAAO;;UAGQ;;EAEf;;EAEA;EACA;;EAEA;EACA;;EAEA,gBAAgB;;EAEhB,OAAO;;UAGQ;EACf,YAAY;EACZ,YAAY;EACZ,kBAAkB;;UAGH,8BACf,aAAa,sBAAsB;EAEnC;EACA;EACA;EACA;;EAEA,MAAM;EACN,iBAAiB;EACjB,iBAAiB;EACjB,YAAY;EACZ,UAAU;EACV,YAAY;;;;;;;;;iBCnRE,gCAAgC,aAAa,qBAC3D,OAAO,qCAAqC,QAC3C,8BAA8B;;;KC7CrB;UAEK;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA,OAAO;EACP;;UAGe,4BAA4B;EAC3C;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;;EAEf,aAAa;;EAEb;;UAGe,wCAAwC;EACvD;;UAGe,0BAA0B,6BACjC,KACN,uBAAuB,qBAAqB;EAG9C;EACA,aAAa;EACb;EACA;EACA,UAAU;;KAGA,qBAAqB;EAC/B,UAAU,MAAM,KAAK;;;iBAIP,qBAAqB;;;;;iBA4BrB,gBACd,kBACA,cACA,UAAS,yBACR;;iBA4Ca,yBACd,kBACA,UAAS,kCACR;;;;;iBAuBa,mBAAmB,qBACjC,SAAS,0BAA0B,aAClC;;;UChJc,0BAA0B,kBAAkB,WAAW;EACtE,WAAW;;;;;EAKX;;;;;;iBAuBc,mBAAmB,WAAW,kBAAkB,UAC9D,SAAS,0BAA0B,aAClC,KAAK,WAAW;;;;;;;;;;;;;;;;;;;;;;;;;;;;;UC9BF;WACN;WACA,MAAM,SAAS;;;UAIT;;WAEN;;WAEA;WACA,gBAAgB;;UAGV;;WAEN;;WAEA;;UAGM;;WAEN;;WAEA,eAAe;;WAEf,eAAe;;;;;;;;iBASV,oBACd,mBAAmB,iBACnB,OAAM,6BACL;UAsCc;;WAEN;;WAEA;;;;;;;;;iBAUK,2BACd,cACA,MAAM,KAAK,0DACV;;;UChFc;;EAEf;;;EAGA;;EAEA;;;cAIW,kCAAkC;WAE3B;EADlB,YACkB,cAChB;;;;;;cAiBS,kCAAkC;mBAIhB;mBAHZ;mBACA;EAEjB,YAA6B,SAAS;EAKhC,QAAQ,OAAO,uBAAuB;EAYtC,OAAO,MAAM,4BAA4B,QAAQ;EA8CjD,QAAQ;IACZ;IACA;IACA,UAAU;IACV,SAAS,OAAO;;UAkCV;UAiCA;UAmBA;UAkBA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBCvNM,eAAe;;;;;;cC8ClB,2BAA2B;EACtC,YAAY;;;;;KAQF,kBAAkB,kBAAkB,UAAU,cACxD,SAAS,gBACT,UAAU,WACV,KAAK,oBACF,QAAQ;UAEI,wBAAwB,kBAAkB,UAAU;;EAEnE,UAAU;;EAEV,WAAW;;EAEX,UAAU,kBAAkB,WAAW;;EAEvC,SAAS,YAAY,WAAW;;;EAGhC;;;EAGA;;;EAGA;;;;EAIA;;EAEA,WAAW;;EAEX;;EAEA;;;;;;;;EAQA;;;EAGA;;EAEA;;;EAGA;;EAEA;;EAEA,eAAe;EACf,gBAAgB;;;EAGhB,UAAU;;EAEV,YAAY;;;EAGZ,aAAa,UAAU;;;;EAIvB;;;;;;;EAOA,cACE,UAAU,WACV,UAAU;IACL;IAAgB;;;;;;;EAMvB,aAAa;IAAS,SAAS;IAAc,UAAU;IAAW;;;EAElE;;EAEA;;UAGe;EACf;EACA;EACA;;EAEA;;EAEA;;EAEA;EACA,gBAAgB;;;EAGhB,WAAW;;UAGI;EACf;EACA;;UAGe,uBAAuB,WAAW,kBAAkB;EACnE;EACA;;;;EAIA,SAAS;EACT,WAAW,eAAe;EAC1B,YAAY,eAAe;;EAE3B,YAAY,eAAe;;EAE3B,WAAW;;EAEX,WAAW,eAAe,eAAe,WAAW;;;;;iBAoOhC,iBAAiB,kBAAkB,UAAU,WACjE,MAAM,wBAAwB,WAAW,aACxC,QAAQ,uBAAuB,WAAW;;;;;UC9Z5B;;EAEf;;EAEA,UAAU;;;;;;;UAQK,kBAAkB;;EAEjC;;EAEA,OAAO;;EAEP,cAAc;;;UAIC,wBAAwB;EACvC,SAAS;;;;;;;;;;;UAYM,eAAe,eAAe,YAAY;EACzD,IAAI,OAAO,QAAQ,KAAK,kBAAkB,iBAAiB;;;;;;;iBAQ7C,qBAAqB,eAAe,WAClD,QAAQ,eAAe,UACtB,kBAAkB,QAAQ;;UAQZ,yBAAyB;EACxC;;;;;;;;iBASoB,eACpB,OAAO,WACP,OAAO,eACP,kBAAkB,qBACjB,QAAQ;;UAUM;EACf;EACA;EACA;EACA;EACA;EACA;;;;;;;iBAQc,oBAAoB,mBAAmB,qBAAqB;;UAkB3D;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;iBAIc,kBAAkB,eAAe,kBAAkB;UAwBlD;;EAEf;;EAEA,OAAO;;EAEP;;;;;;;;;iBAkBc,yBACd,eAAe,iBACf,OAAM;;;UCpKS;EACf;EACA;EACA;EACA;EACA;;UAGe,kBAAkB,kBAAkB,UAAU;WACpD;WACA;WACA;WACA;WACA,mBAAmB;WACnB,oBAAoB;WACpB,iBAAiB,YAAY,WAAW;WACxC;WACA;WACA,UAAU,YAAY,wBAAwB,WAAW;WACzD;;WAEA;WACA,WAAW,YAAY,wBAAwB,WAAW;WAC1D;WACA;WACA,aAAa,UAAU;WACvB,cACP,UAAU,WACV,UAAU;IACL;IAAgB;;WACd,eAAe;;UAGT,+BAA+B,kBAAkB,UAAU,mBAClE,KACN,wBAAwB,WAAW;;EAgBrC;;UAGe;EACf;EACA;EACA;EACA;EACA;;UAGe,+BAA+B,kBAAkB,UAAU;EAC1E,MAAM,kBAAkB,WAAW;;EAEnC;;EAEA,yBAAyB;EACzB,UAAU,kBAAkB,WAAW;EACvC;EACA,UAAU;EACV;EACA;EACA;EACA,eAAe,wBAAwB,WAAW;EAClD,gBAAgB,wBAAwB,WAAW;EACnD,YAAY;;UAGG,2BAA2B,WAAW,kBAAkB;EACvE;EACA;EACA,QAAQ,uBAAuB,WAAW;EAC1C,UAAU;;UAGK,6BAA6B,kBAAkB,UAAU;EACxE,MAAM,kBAAkB,WAAW;EACnC;EACA,UAAU;EACV;EACA;EACA,YAAY;;UAGG,6BAA6B,WAAW,kBAAkB,kBACjE,uBAAuB,WAAW;EAC1C,UAAU;;iBAGI,wBAAwB,kBAAkB,UAAU,WAClE,MAAM,+BAA+B,WAAW,aAC/C,kBAAkB,WAAW;iBA0HV,wBAAwB,kBAAkB,UAAU,WACxE,MAAM,+BAA+B,WAAW,aAC/C,QAAQ,2BAA2B,WAAW;iBAkD3B,sBAAsB,kBAAkB,UAAU,WACtE,MAAM,6BAA6B,WAAW,aAC7C,QAAQ,6BAA6B,WAAW;;;;;;;;iBCrTnC;;;;;iBAQA,cAAc,gBAAgB;;;;;;;;;;;;;;;;;UCD7B;EACf;;EAEA;;UAGe;EACf;;EAEA;;EAEA;EACA;;EAEA;;;;;;;;;;iBA4Cc,oBACd,SAAS,kBACT;EAAS;IACR;;;;;;;;;;;iBA0Ba,qBACd,SAAS,kBACT,WACA;EAAS;;;;;;;iBC3FK,sBAAsB,WAAW,kBAAkB,UACjE,UAAU,eAAe,WAAW;;;;iBA0CtB,gBAAgB,sBAAsB;UAWrC;;EAEf,YAAY;;;;;EAKZ,WAAW;IAAQ;IAAoB;IAAmB;IAAgB;;;;;iBAK5D,kBAAkB,WAAW,kBAAkB,UAC7D,UAAU,eAAe,WAAW,aACnC;;;;;;;;;;;;;;;;;;UC5Dc;;WAEN;;WAEA;;WAEA;;WAEA;;UAGM;;EAEf;;;;;;;iBAwBc,qBAAqB,MAAM,uBAAuB;;;;iBCnDlD,0BAA0B,2BAA2B,WAAW;;;;;;;;;;iBAyEhE,iCAAiC,SAAS;;;;iBAoC1C,4BAA4B,SAAS;;iBAgBrC,mBAAmB,SAAS;;iBAW5B,YAAY,SAAS;;iBA4BrB,kBACd,eAAe,gBACf,iBAAiB;;;UCvKF;EACf;EACA;EACA;;UAGe,qBAAqB,gBAAgB;EACpD;EACA;EACA;EACA,SACE,QAAQ,SACR,SAAS,4BACR,QAAQ;;UAGI;EACf;EACA;EACA,WAAW;;KAGD,oBAAoB,eAAe,4BAC7C,OAAO,QACP,SAAS,8BACN,qBAAqB,QAAQ;UAEjB;WACN,QAAQ;WACR;;KAGN,sBAAsB,WAAW,KACpC,iBAAiB;EAGjB;;UAGQ,qBAAqB,kBAAkB;EAC/C;EACA;EACA;EACA,aAAa,UAAU;;EAEvB,eAAe;;iBAGD,sBACd,gBAAgB,yBAChB,WACA,kBAAkB,WAAW,UAE7B,WAAW,qBAAqB,UAChC,SAAS,qBAAqB;EAC5B,SAAS;IAAS,UAAU;IAAW,UAAU;MAAc;EAC/D,WAAW,sBAAsB;IAElC,YAAY,WAAW;iBA8CV,qBACd,eAAe,yBACf,WACA,kBAAkB,WAAW,UAE7B,QAAQ,oBAAoB,SAC5B,SAAS,qBAAqB;EAC5B;EACA,SAAS;IAAS,UAAU;IAAW,UAAU;MAAc;;EAEzD;EAAc;;EACd;EAAa,UAAU,sBAAsB;KAEpD,YAAY,WAAW;;;KCxFrB,qBAAqB;KACrB,iBAAiB,SAAS;KAC1B,aAAa,gBAAgB,aAAa,MAAM,mBAAmB;UAcvD;;WAEN;;WAEA;;WAEA;;WAEA;;WAEA;;UAGM;;EAEf,OAAO;IAAQ;IAAiB;MAAkB,QAAQ;;;EAG1D,SAAS,UAAU,UAAU,kBAAkB,QAAQ;;EAEvD,QAAQ,UAAU,WAAW;;;cAIlB,6BAA6B;WAG7B;EAFX,YACE,iBACS;;UAOI;;EAEf;;EAEA;;EAEA;;;;EAIA,MAAM;;UAobS;;EAEf;;EAEA;;EAEA;;;EAGA,YAAY;;;;;iBAoJE,mBAAmB,MAAM,4BAA4B;;;;iBA0FrD,kBACd,SAAS,aACT,uBACC;;;iBAOa,oBAAoB,SAAS,aAAa"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { a as RunRecord } from "./run-record-
|
|
2
|
-
import { s as TraceStore } from "./store-
|
|
1
|
+
import { a as RunRecord } from "./run-record-DQjRcYwA.js";
|
|
2
|
+
import { s as TraceStore } from "./store-Cq9oOrI1.js";
|
|
3
3
|
import { S as ToolDef, a as MintedRolloutLine, b as RolloutStep, h as RolloutOutcome, i as GatedEvidence, m as RolloutLine, t as ChatMessage, v as RolloutRole, y as RolloutSplit } from "./schema-BzWDXhOR.js";
|
|
4
4
|
import { DatabaseSync } from "node:sqlite";
|
|
5
5
|
//#region src/rollout/exporters.d.ts
|
|
@@ -931,4 +931,4 @@ declare function parseRolloutReleaseArgs(argv: string[]): RolloutReleaseCliArgs;
|
|
|
931
931
|
declare function runRolloutReleaseCli(argv: string[]): Promise<number>;
|
|
932
932
|
//#endregion
|
|
933
933
|
export { observedSplitScore as $, measureFormatGate as A, GateCheckDisposition as At, DEFAULT_CLAUDE_PROJECTS_DIR as B, SftExportOptions as Bt, FORMAT_GATE_DISPOSITION as C, relabelImportedSplit as Ct, ReleaseRowRef as D, GATE_CHECK_IDS as Dt, GateReport as E, GATE_CHECKS as Et, openOpencodeDb as F, gateErrors as Ft, MintRolloutResult as G, toRewardRows as Gt, findClaudeTranscripts as H, VerifiersRolloutOutput as Ht, readOpencodeSessionMessages as I, gatedEvidenceOf as It, unmintableReasons as J, toSftRows as Jt, RolloutScrubber as K, toRftItem as Kt, ClaudeTranscript as L, RealnessLabels as Lt, DEFAULT_OPENCODE_DB as M, GateCheckedOutcome as Mt, OpencodeSessionRow as N, GateEntryPoint as Nt, assertGateReport as O, GATE_POLICIES as Ot, findOpencodeSessionsByDirectory as P, GatePolicy as Pt, observedScore as Q, ClaudeTranscriptRef as R, RewardRow as Rt, EmittedEvidence as S, fromHarborTrajectory as St, GateDisposition as T, toHarborTrajectory as Tt, readClaudeTranscript as U, VerifiersTokenUsage as Ut, claudeProjectSlug as V, SftRow as Vt, MintRolloutOptions as W, toJsonl as Wt, ScorePreference as X, toVerifiersRolloutOutputs as Xt, ScoreOrigin as Y, toVerifiersRolloutOutput as Yt, isRealnessGated as Z, defaultRolloutScrubber as _, HarborStep as _t, buildHfDataset as a, readRolloutLedger as at, scrubRolloutLine as b, HarborToolCall as bt, runRolloutReleaseCli as c, FromHarborOptions as ct, RELEASE_FORMATS as d, HarborContentPart as dt, scoreOrigin as et, ReleaseFormat as f, HarborFinalMetrics as ft, addScrubCounts as g, HarborObservationResult as gt, ScrubRule as h, HarborObservation as ht, ScrubReport as i, readRolloutJournal as it, releaseRowRefs as j, GateCheckId as jt, gatedRolloutIds as k, GateCheck as kt, DatasetCardInputs as l, HARBOR_IMPORT_GAP as lt, ScrubCounts as m, HarborMetrics as mt, BuildSummary as n, trainingScore as nt, parseRolloutReleaseArgs as o, writeRolloutLedger as ot, buildDatasetCard as p, HarborImageSource as pt, mintRolloutRows as q, toRftItems as qt, RolloutReleaseCliArgs as r, appendRolloutLines as rt, planPushCommand as s, ATIF_SCHEMA_VERSION as st, BuildOptions as t, trainingReward as tt, FORMAT_FILES as u, HarborAgent as ut, emptyScrubCounts as v, HarborStepSource as vt, FormatGateCounts as w, toHarborTrajectories as wt, scrubText as x, HarborTrajectory as xt, scrubLines as y, HarborSubagentTrajectoryRef as yt, ClaudeUsageTotals as z, RftItem as zt };
|
|
934
|
-
//# sourceMappingURL=index-
|
|
934
|
+
//# sourceMappingURL=index-DMoxLG8P.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index-
|
|
1
|
+
{"version":3,"file":"index-DMoxLG8P.d.ts","names":[],"sources":["../src/rollout/exporters.ts","../src/rollout/gate-checks.ts","../src/rollout/interchange/harbor.ts","../src/rollout/ledger.ts","../src/rollout/reward.ts","../src/rollout/mint.ts","../src/rollout/readers/claude-jsonl.ts","../src/rollout/readers/opencode-sqlite.ts","../src/rollout/release/gate-report.ts","../src/rollout/release/scrub.ts","../src/rollout/release/card.ts","../src/rollout/release/hf-dataset.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;;UAiDiB;;EAEf;;;;;;EAMA;;UAiBe;;EAEf;;EAEA;;;;;;;;;;;;;;;;;;;;;;KAuBU;UAEK,kCAAkC;;EAEjD,aAAa;;KAGH,mBAAmB;UAEd;EACf,UAAU;EACV;IACE;IACA;IACA;IACA;IACA;MACE;;;;;;;;;;;;;;;;;iBAkBU,UAAU,OAAO,qBAAqB,UAAS,mBAAwB;UAyBtE;;EAEf;EACA,OAAO;EACP;EACA;IACE;IACA;IACA;IACA;IACA,OAAO;MAUL;;;;;iBAMU,aACd,OAAO,qBACP,UAAS,wBACR;UAgCc;EACf;EACA;EACA;EACA;EACA;;UAGe;;EAEf,QAAQ;;EAER,YAAY;EACZ;EACA,SAAS;EACT,WAAW;EACX,aAAa;EACb;IACE,MAAM;IACN,QAAQ;IACR;IACA;IACA;IACA;IACA;IACA;IACA,MAAM;MACJ;;iBASU,yBAAyB,MAAM,oBAAoB;iBA+BnD,0BACd,OAAO,qBACP,UAAS,4BACR;UAoBc;;EAEf,UAAU;;EAEV;IACE;IACA;IACA;IACA;IACA;IACA,OAAO;IACP;MACE;;iBAGU,UAAU,MAAM,oBAAoB;;iBAmBpC,WACd,OAAO,qBACP,UAAS,4BACR;iBAkBa,QAAQ,MAAM;;;;;;;;;;cCpUjB;KAOD,sBAAsB;;;;;;;;;KAUtB,qBAAqB,QAAQ,kBAAkB,SAAS;;;;;;;;;;;;;;;;;UAkBnD;EACf,SAAS;;EAET;;UAOe;EACf,IAAI;;EAEJ;;EAEA,SAAS,SAAS;;;;;;;;;;;;EAYlB,WAAW;;;;;;;;;;;;;iBA6LG,gBAAgB,SAAS,cAAc;;;;;;cA6I1C,yBAAyB,KAAK,cAAc;;;;;;;;;KAe7C;WACG;;;;WAEA;WAAyB;;;;WAEzB;WAAuB;;;KAU1B,yBAAyB,KAAK,cAAc;;;;;;;;;cAU3C;;;;;;;aAWT;eAnCiB;;aAoCjB,kBAAgB;aAMhB,2BAAyB;aAMzB,qBAAmB;;;;;;;;aAenB;;;aACA,kBAAgB;aAChB,2BAAyB;aACzB;;;;;;;;;;;aASA;;;aACA;;;aACA;;;aACA;;;;;;;;;;;aASA;;;aACA;;;aACA;;;aACA;;;;;;KAKQ,8BAA8B;;;;;;iBAO1B,WAAW,SAAS,aAAa,QAAQ;;;cCxd5C;;cAGA;KAwBD;UAEK;EACf;EACA;;UAGe;EACf;EACA;EACA,SAAS;;UAGM;EACf;EACA;;EAEA,WAAW;EACX,QAAQ;;UAGO;EACf;EACA;;EAEA;EACA,QAAQ;;UAGO;EACf;EACA,mBAAmB;EACnB,0BAA0B;EAC1B,QAAQ;;UAGO;EACf,SAAS;;UAGM;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA,QAAQ;;UAGO;;EAEf;EACA;EACA,QAAQ;EACR;EACA;EACA,kBAAkB;EAClB;EACA,aAAa;EACb,cAAc;EACd,UAAU;EACV;EACA;EACA,QAAQ;;UAGO;EACf;EACA;EACA;;EAEA,mBAAmB;EACnB,QAAQ;;UAGO;EACf;EACA;EACA;EACA;EACA;EACA,QAAQ;;UAGO;EACf;EACA;;EAEA;EACA,OAAO;EACP,OAAO;EACP;EACA,gBAAgB;EAChB;EACA,wBAAwB;EACxB,QAAQ;;;;;;;;;;;;;;;;;;iBAmXM,mBAAmB,OAAO,gBAAgB;;iBAa1C,qBAAqB,OAAO,gBAAgB;UA2O3C;;EAEf,YAAY;;;;;;;;;;;iBAwGE,qBACd,YAAY,kBACZ,UAAS,oBACR;;;;;;;;;;;;;;;iBAiCa,qBACd,gBAAgB,eAChB,OAAO,eACN;;;;iBC/6BmB,mBAAmB,cAAc,OAAO,gBAAgB;;iBAOxD,mBAAmB,cAAc,OAAO,gBAAgB;;;;;;;;;;;iBAiBxD,kBAAkB,eAAe,QAAQ;;;;;;;;;;;;;;;iBAkBzC,mBAAmB,eAAe,QAAQ;;;;;;;;KCzCpD;;KAGP,SAAS,KAAK;;iBAGH,gBAAgB,QAAQ;;;;;;;;;;;;;;iBAiBxB,mBAAmB,QAAQ,QAAQ,OAAO;;;;;;;;;;;;;;;;iBAmB1C,cACd,QAAQ,QACR,SAAQ;;KAOE;;;;;;;iBAQI,YAAY,QAAQ,QAAQ,SAAQ,kBAA8B;;;;;;;;;iBAelE,cACd,QAAQ,QACR,SAAQ;;;;;;;;;;;;iBAiBM,eAAe,QAAQ;EAAW;EAAuB;;;;;KCvE7D,mBAAmB;UAEd;EACf,QAAQ;;EAER;;EAEA,OAAO;;EAEP;;EAEA,YAAY;;UAGG;EACf,MAAM;;EAEN;;;;;;;;;;;;;;iBAkLc,kBAAkB,QAAQ;;;;;;;iBA+JpB,gBACpB,SAAS,aACT,OAAO,YACP,UAAS,qBACR,QAAQ;;;cCrYE;;iBAGG,kBAAkB;UAIjB;EACf;EACA;;;iBAIoB,sBACpB,aACA,uBACC,QAAQ;UASM;EACf;EACA;EACA;EACA;;UAGe;EACf,UAAU;EACV,OAAO;;EAEP;EACA;EACA;;UAyDe;;;;;;WAMN;;;iBAgBW,qBACpB,cACA,UAAS,8BACR,QAAQ;;;cCnHE;UAEI;EACf;EACA;EACA;EACA;;EAEA;IAAS;IAAa;;EACtB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;;iBAYoB,eACpB,gBACC,QAAQ;;iBAiDK,gCACd,IAAI,cACJ,oBACC;;;;;;;iBA2Ba,4BAA4B,IAAI,cAAc,oBAAoB;;;;KC5EtE;cAEC,yBAAyB,OAAO,eAAe;;UAQ3C;EACf;EACA;;;;;;;;;;;;EAYA;;;;;;;;EAQA;;;;;;;;;;;;EAYA;;;UAIe;;EAEf;EACA;;UAGe;;EAEf;;EAEA;;;;;;EAMA;;EAEA;;;;;;;;;;;;;;EAcA,oBAAoB;;;;;;;;;;;;EAYpB;;EAEA;;;;;;;;;;EAUA,wBAAwB;;UAGT;;EAEf;EACA,UAAU,QAAQ,OAAO,eAAe;;;iBAI1B,gBAAgB,gBAAgB,sBAAsB;;;;;;;cAUzD;EAIX,MAAG,eAAkB,aAAW;EAShC,YAAS,eAAkB,6BAA2B;EAStD,MAAG,eAAkB,cAAY;EAOjC,MAAG,gBAAmB,wBAAsB;;;iBAgC9B,kBACd,OAAO,qBACP,eAAe,kBACd;;;;;;;;;;;;;;;;iBA+Fa,iBAAiB,QAAQ;;;UCrUxB;EACf;EACA,SAAS;;EAET,UAAU,eAAe;;;KAqEf,cAAc;iBAEV,oBAAoB;iBAIpB,eAAe,MAAM,aAAa,MAAM,cAAc;iBAKtD,UAAU,cAAc,QAAQ;;;;;;;;;iBA8BhC,iBAAiB,MAAM,mBAAmB,QAAQ,cAAc;iBAIhE,WAAW,OAAO;EAChC,OAAO;EACP,QAAQ;;;;;;;;;iBAaM,uBAAuB;;;cCrI1B;KACD,wBAAwB;;cAGvB,cAAc,OAAO;UAejB;;EAEf,OAAO;EACP,SAAS;EACT;;EAEA;EACA,aAAa;EACb;IAAY;IAAmB;;EAC/B,cAAc,QAAQ,OAAO;;;;;;;EAO7B,MAAM;;iBA4GQ,iBAAiB,QAAQ;;;UC1HxB;EACf;EACA,SAAS;EACT;;UAGe;;EAEf,OAAO,eAAe;EACtB,QAAQ;EACR;IAAY;IAAmB;;;UAGhB;EACf;EACA;EACA;EACA,OAAO;EACP,cAAc,QAAQ,OAAO;;EAE7B,MAAM;EACN;;iBAGoB,eACpB,kBACA,SAAS,eACR,QAAQ;iBAuGK,gBAAgB,cAAc;UAoB7B,8BAA8B;EAC7C;EACA;;iBAMc,wBAAwB,iBAAiB;;iBA0CnC,qBAAqB,iBAAiB"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { t as DefaultVerdict } from "./verdict-E4eRNf7-.js";
|
|
2
1
|
import { p as CostProvenance } from "./cost-ledger-DbQdN3nO.js";
|
|
2
|
+
import { t as DefaultVerdict } from "./verdict-E4eRNf7-.js";
|
|
3
3
|
//#region src/matrix/cell-spend.d.ts
|
|
4
4
|
/**
|
|
5
5
|
* Spend carrier for a cell that fails after it has already spent money.
|
|
@@ -195,4 +195,4 @@ interface RunAgentMatrixOptions<Output> {
|
|
|
195
195
|
declare function runAgentMatrix<Output>(opts: RunAgentMatrixOptions<Output>): Promise<MatrixResult<Output>>;
|
|
196
196
|
//#endregion
|
|
197
197
|
export { MatrixCell as a, CellSpend as c, MatrixAxis as i, readCellSpend as l, AxisSummary as n, MatrixResult as o, CellResult as r, RunAgentMatrixOptions as s, runAgentMatrix as t, withCellSpend as u };
|
|
198
|
-
//# sourceMappingURL=index-
|
|
198
|
+
//# sourceMappingURL=index-DNgf5gyG.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index-
|
|
1
|
+
{"version":3,"file":"index-DNgf5gyG.d.ts","names":[],"sources":["../src/matrix/cell-spend.ts","../src/matrix/types.ts","../src/matrix/runner.ts"],"mappings":";;;;;;;;;;;;;;;;;;UAoBiB;;;;EAIf;;EAEA;;;;EAIA;;;;;;;;;;;;;;;;iBAiBc,cAAc,gBAAgB,OAAO;;;;iBA+CrC,cAAc,iBAAiB;;;;;UC5E9B,WAAW;;EAE1B;;EAEA,QAAQ;IAAQ;IAAY,OAAO;;;;EAGnC,SAAS,OAAO,GAAG;;;UAIJ;EACf,MAAM;IAAiB;IAAY;;;EAEnC;;EAEA;;UAGe,WAAW;EAC1B,QAAQ;EACR,SAAS;;;;EAIT;EACA;EACA;;;;;;;;;;;EAWA,iBAAiB;;;;EAIjB;IAAU;IAAiB;;;UAGZ;EACf;EACA;EACA;EACA;EACA;EACA;EACA;;;EAGA;;EAEA;EACA;;UAGe,aAAa;EAC5B,OAAO;IAAQ,MAAM;IAAY,MAAM,WAAW;;;;EAGlD,QAAQ,eAAe,eAAe;EACtC;IACE;IACA;;;IAGA;IACA;IACA;;;;IAIA;;;;IAIA;;;;IAIA;IACA;;;EAGF;;UAGe,sBAAsB;EACrC,MAAM;;;;;;EAMN,UAAU,MAAM,eAAe,QAAQ,WAAW;;EAElD;;EAEA,UAAU,MAAM,KAAK;;EAErB;;EAEA;;;;;EAKA;;;;;;;;;;;;;EAaA;;EAEA,kBAAkB,MAAM,YAAY,QAAQ,WAAW;;;EAGvD,SAAS;;;;iBC1BW,eAAe,QACnC,MAAM,sBAAsB,UAC3B,QAAQ,aAAa"}
|
|
@@ -1,43 +1,12 @@
|
|
|
1
|
-
import { o as Severity } from "./multi-layer-verifier-BUaQ4C17.js";
|
|
2
1
|
import { b as CustomTokenPricing, c as CostLedgerHandle } from "./cost-ledger-DbQdN3nO.js";
|
|
2
|
+
import { i as AnalystFinding, n as AnalystContext } from "./types-DMoNFDWi.js";
|
|
3
3
|
import { p as ChatClient } from "./types-Bfk0uxRj.js";
|
|
4
|
-
import { i as AnalystFinding, n as AnalystContext } from "./types-D9ssmxKL.js";
|
|
5
|
-
import { i as TraceAnalystDefinition } from "./default-registry-G9CKMNkc.js";
|
|
6
|
-
import { l as ExactCapableAnalyst } from "./exact-types-qnexxJ1Z.js";
|
|
7
4
|
import { f as ExternalOptimizerModelExecutionObservation, h as ExternalOptimizerRunnerCommand, l as ExternalOptimizerModelCall, r as ExternalOptimizerCallbackLimits } from "./external-optimizer-contracts-szBJ_1vh.js";
|
|
5
|
+
import { o as Severity } from "./multi-layer-verifier-BUaQ4C17.js";
|
|
8
6
|
import { S as SupervisorRunTree, x as SupervisorRunSources } from "./types-DeIUdzNd.js";
|
|
9
|
-
import {
|
|
10
|
-
|
|
11
|
-
|
|
12
|
-
success: number;
|
|
13
|
-
goalProgress: number;
|
|
14
|
-
repoGroundedness: number;
|
|
15
|
-
driftPenalty: number;
|
|
16
|
-
toolUseQuality: number;
|
|
17
|
-
patchQuality: number;
|
|
18
|
-
testReality: number;
|
|
19
|
-
finalGate: number;
|
|
20
|
-
reviewerBlockers: number;
|
|
21
|
-
costUsd: number;
|
|
22
|
-
wallSeconds: number;
|
|
23
|
-
notes?: string[];
|
|
24
|
-
}
|
|
25
|
-
interface RunScoreWeights {
|
|
26
|
-
success: number;
|
|
27
|
-
goalProgress: number;
|
|
28
|
-
repoGroundedness: number;
|
|
29
|
-
driftPenalty: number;
|
|
30
|
-
toolUseQuality: number;
|
|
31
|
-
patchQuality: number;
|
|
32
|
-
testReality: number;
|
|
33
|
-
finalGate: number;
|
|
34
|
-
reviewerBlockers: number;
|
|
35
|
-
costUsd: number;
|
|
36
|
-
wallSeconds: number;
|
|
37
|
-
}
|
|
38
|
-
declare function aggregateRunScore(score: RunScore, weights?: Partial<RunScoreWeights>): number;
|
|
39
|
-
declare function clamp01(value: number): number;
|
|
40
|
-
//#endregion
|
|
7
|
+
import { i as TraceAnalystDefinition } from "./default-registry-B0s2zU-s.js";
|
|
8
|
+
import { l as ExactCapableAnalyst } from "./exact-types-BEecmnWm.js";
|
|
9
|
+
import { t as TraceAnalysisEngine } from "./engine-CAmTUk52.js";
|
|
41
10
|
//#region src/semantic-concept-judge.d.ts
|
|
42
11
|
/**
|
|
43
12
|
* Implementation complexity class for weighted scoring.
|
|
@@ -470,5 +439,5 @@ declare const KNOWLEDGE_POISONING_KIND_SPEC: TraceAnalystDefinition;
|
|
|
470
439
|
*/
|
|
471
440
|
declare const DEFAULT_TRACE_ANALYST_KINDS: readonly TraceAnalystDefinition[];
|
|
472
441
|
//#endregion
|
|
473
|
-
export { SemanticConceptJudgeResult as A, DspyRlmTraceEngineOptions as C, SEMANTIC_CONCEPT_JUDGE_VERSION as D, ConceptSpec as E,
|
|
474
|
-
//# sourceMappingURL=index-
|
|
442
|
+
export { SemanticConceptJudgeResult as A, DspyRlmTraceEngineOptions as C, SEMANTIC_CONCEPT_JUDGE_VERSION as D, ConceptSpec as E, SemanticConceptJudgeInput as O, renderFindingSubject as S, ConceptFinding as T, FindingSubject as _, FAILURE_MODE_KIND_SPEC as a, findingSubjectGrammarPromptFor as b, emitControlIntegrityFindings as c, FindingsStore as d, PersistedFinding as f, FINDING_SUBJECT_SYNTAX as g, FINDING_SUBJECT_KINDS as h, IMPROVEMENT_KIND_SPEC as i, runSemanticConceptJudge as j, SemanticConceptJudgeOptions as k, DiffPolicy as l, diffFindings as m, KNOWLEDGE_POISONING_KIND_SPEC as n, CONTROL_INTEGRITY_ANALYST as o, defaultIsMaterial as p, KNOWLEDGE_GAP_KIND_SPEC as r, ControlIntegrityAnalyst as s, DEFAULT_TRACE_ANALYST_KINDS as t, FindingsDiff as u, FindingSubjectKind as v, createDspyRlmTraceEngine as w, parseFindingSubject as x, KIND_EXPECTED_SUBJECTS as y };
|
|
443
|
+
//# sourceMappingURL=index-DT73JraI.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index-DT73JraI.d.ts","names":[],"sources":["../src/semantic-concept-judge.ts","../src/analyst/trace-tool-callback.ts","../src/analyst/dspy-rlm-engine.ts","../src/analyst/finding-subject.ts","../src/analyst/findings-store.ts","../src/analyst/kinds/control-integrity.ts","../src/analyst/kinds/failure-mode.ts","../src/analyst/kinds/improvement.ts","../src/analyst/kinds/knowledge-gap.ts","../src/analyst/kinds/knowledge-poisoning.ts","../src/analyst/kinds/index.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KAuDY;UAEK;EACf;;EAEA;;EAEA;;EAEA,aAAa;;UAGE;EACf;EACA;;EAEA;EACA;EACA,UAAU;;UAGK;;EAEf;;EAEA;;EAEA,aAAa;IAAQ;IAAc;;;EAEnC,kBAAkB;;EAElB;EACA;;UAGe;EACf;EACA;;EAEA;EACA;EACA;EACA,UAAU;EACV;EACA;EACA;;EAEA;EACA;;;;;;;;KASU;UAQK;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,MAAM;;EAEN,UAAU;EACV,aAAa;EACb;EACA,WAAW;EACX,SAAS;;;;;;EAMT,iBAAiB;;EAEjB,oBAAoB,QAAQ,OAAO;;cAKxB;;;;;;iBAkGS,wBACpB,OAAO,2BACP,SAAS,8BACR,QAAQ;;;KCzOC,0BAA0B;;;UCerB;;EAEf,MAAM;;EAEN;;EAEA,kBAAkB,aAAa;EAC/B;;EAEA,UAAU;;EAEV;;EAEA;;;;;;EAMA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,kBAAkB,QAAQ;;EAE1B;;;;;;;;;;EAUA;;EAEA,SAAS;;EAET;;;iBAIc,yBAAyB,SAAS,4BAA4B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KCzClE;EAEN;EAAwB;EAAc;;EACtC;EAAyB;;EACzB;EAAuB;;EACvB;EAAyB;;EAGzB;EAAuB;;EACvB;EAAe;;EACf;EAAkB;EAAc;;EAChC;EAAkB;;EAClB;EAAa;EAAgB;;EAC7B;EAAc;;EACd;EAAkB;;EAClB;EAAkB;;EAClB;EAAwB;;EACxB;EAAuB;;EACvB;EAAc;;EACd;EAAa;EAAgB;;EAC7B;EAAgB;;EAChB;EAAqB;;EACrB;EAAuB;;EAEvB;EAA4B;;EAC5B;EAA2B;;EAE3B;EAAiB;;KAEX,qBAAqB;cAEpB,uBAAuB,cAAc;;;;;;;;;;;;;;;;;iBAgGlC,oBAAoB,iCAAiC;;;;;;;iBAoHrD,qBAAqB,GAAG;;;;;;;;;;cA8D3B,wBAAwB,SAAS,OAAO;;;;;;;;;cAoExC,wBAAwB,eAAe,cAAc;;iBAqElD,+BAA+B;;;;;;;;UCxc9B,yBAAyB;EACxC;;cAGW;WAGiB;mBAFX;EAEjB,YAA4B;EAItB,OAAO,eAAe,UAAU,mBAAmB;;EAQzD,WAAW;;EAkBX,QAAQ,gBAAgB;;UAOT;;EAEf,UAAU;;EAEV,aAAa;;EAEb,WAAW;;;;;EAKX,SAAS;IAAQ,UAAU;IAAkB,SAAS;;;UAGvC;;;;;;;;EAQf,cAAc,UAAU,gBAAgB,SAAS;;;;;;iBAOnC,kBAAkB,GAAG,gBAAgB,GAAG;;;;;iBAWxC,aACd,UAAU,oBACV,SAAS,oBACT,SAAQ,aACP;;;;iBC7Fa,6BACd,OAAO,uBAAuB,mBAC9B,qBACC;;cA4BU,mCACA,oBAAoB,uBAAuB;WAE7C;WACA;WAEA;WACA;IAAS;IAAgC;;WACzC;WACA;aACP;aACA;;EAGI,QACJ,OAAO,uBAAuB,mBAC9B,KAAK,iBACJ,QAAQ;;cAUA,2BAAyB;;;cCHzB,wBAAwB;;;cC3BxB,uBAAuB;;;cCGvB,yBAAyB;;;cCPzB,+BAA+B;;;;;;;;;;;;cCP/B,sCAAsC"}
|