@tangle-network/agent-eval 1.0.12 → 1.0.14
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 +36 -0
- package/dist/adapters/http.d.ts +2 -2
- package/dist/{agent-profile-CoD1Dotf.d.ts → agent-profile-CPv4tENR.d.ts} +2 -2
- package/dist/{agent-profile-CoD1Dotf.d.ts.map → agent-profile-CPv4tENR.d.ts.map} +1 -1
- package/dist/{agreement-irr-w-t3XUTD.js → agreement-irr-DXcXEt4I.js} +20 -23
- package/dist/agreement-irr-DXcXEt4I.js.map +1 -0
- package/dist/analyst/index.d.ts +59 -13
- package/dist/analyst/index.d.ts.map +1 -1
- package/dist/analyst/index.js +144 -57
- package/dist/analyst/index.js.map +1 -1
- package/dist/{analyze-runs-dGYmYNtb.js → analyze-runs-C7vWBDhl.js} +11 -11
- package/dist/{analyze-runs-dGYmYNtb.js.map → analyze-runs-C7vWBDhl.js.map} +1 -1
- package/dist/{analyze-runs-BYyMShsj.d.ts → analyze-runs-pJ4fhnYg.d.ts} +2 -2
- package/dist/{analyze-runs-BYyMShsj.d.ts.map → analyze-runs-pJ4fhnYg.d.ts.map} +1 -1
- package/dist/{benchmark-command-DZQrxqUL.js → benchmark-command-BRRhjX2k.js} +12 -14
- package/dist/benchmark-command-BRRhjX2k.js.map +1 -0
- package/dist/benchmarks/index.d.ts +4 -4
- package/dist/benchmarks/index.js +5 -5
- package/dist/builder-eval/index.js +1 -1
- package/dist/{calibration-DLzckXnn.js → calibration-CbZPApIO.js} +2 -2
- package/dist/{calibration-DLzckXnn.js.map → calibration-CbZPApIO.js.map} +1 -1
- package/dist/campaign/index.d.ts +6 -6
- package/dist/campaign/index.js +12 -12
- package/dist/{campaign-VKtf8HRr.js → campaign-DzLlT0So.js} +17 -17
- package/dist/{campaign-VKtf8HRr.js.map → campaign-DzLlT0So.js.map} +1 -1
- package/dist/{campaign-evidence-BICB3a9Z.js → campaign-evidence-CcA7a67m.js} +3 -3
- package/dist/{campaign-evidence-BICB3a9Z.js.map → campaign-evidence-CcA7a67m.js.map} +1 -1
- package/dist/cli.js +16 -10
- package/dist/cli.js.map +1 -1
- package/dist/{cluster-bootstrap-hNXESoc1.js → cluster-bootstrap-BteuR1hg.js} +2 -2
- package/dist/{cluster-bootstrap-hNXESoc1.js.map → cluster-bootstrap-BteuR1hg.js.map} +1 -1
- package/dist/contract/index.d.ts +9 -9
- package/dist/contract/index.js +14 -14
- package/dist/{decision-point-CErE7GnP.js → decision-point-CNyoKOkz.js} +13 -7
- package/dist/{decision-point-CErE7GnP.js.map → decision-point-CNyoKOkz.js.map} +1 -1
- package/dist/{default-registry-DKlKVXMx.d.ts → default-registry-BNv7eB1C.d.ts} +10 -7
- package/dist/default-registry-BNv7eB1C.d.ts.map +1 -0
- package/dist/{default-registry-gmDeN0KT.js → default-registry-eCkZFfuF.js} +6 -6
- package/dist/default-registry-eCkZFfuF.js.map +1 -0
- package/dist/{define-agent-eval-DyzJ0wOo.js → define-agent-eval-CjF2nTA3.js} +8 -8
- package/dist/{define-agent-eval-DyzJ0wOo.js.map → define-agent-eval-CjF2nTA3.js.map} +1 -1
- package/dist/{define-agent-eval-B9zPWFSK.d.ts → define-agent-eval-DKVYlm-T.d.ts} +7 -7
- package/dist/{define-agent-eval-B9zPWFSK.d.ts.map → define-agent-eval-DKVYlm-T.d.ts.map} +1 -1
- package/dist/{descriptive-BTzLL-aj.js → descriptive-B3Eb7tX0.js} +2 -2
- package/dist/{descriptive-BTzLL-aj.js.map → descriptive-B3Eb7tX0.js.map} +1 -1
- package/dist/diagnosis/index.d.ts +2 -2
- package/dist/diagnosis/index.js +1 -1
- package/dist/{dspy-rlm-engine-Xt3znknf.js → dspy-rlm-engine-_MEe2jIO.js} +3 -3
- package/dist/{dspy-rlm-engine-Xt3znknf.js.map → dspy-rlm-engine-_MEe2jIO.js.map} +1 -1
- package/dist/{engine-CaHaCc3W.d.ts → engine-rfV8Ff8V.d.ts} +2 -2
- package/dist/{engine-CaHaCc3W.d.ts.map → engine-rfV8Ff8V.d.ts.map} +1 -1
- package/dist/{eval-campaign-B-dQl_lA.js → eval-campaign-CZSCjxVb.js} +3 -3
- package/dist/{eval-campaign-B-dQl_lA.js.map → eval-campaign-CZSCjxVb.js.map} +1 -1
- package/dist/{evaluation-B0-KVXaV.d.ts → evaluation-DYAIkPQB.d.ts} +4 -4
- package/dist/{evaluation-B0-KVXaV.d.ts.map → evaluation-DYAIkPQB.d.ts.map} +1 -1
- package/dist/evaluation.d.ts +1 -1
- package/dist/{exact-types-CIViGj98.d.ts → exact-types-CaLLU1Pz.d.ts} +2 -2
- package/dist/{exact-types-CIViGj98.d.ts.map → exact-types-CaLLU1Pz.d.ts.map} +1 -1
- package/dist/experiment/index.d.ts +1 -1
- package/dist/experiment/index.js +9 -9
- package/dist/{external-optimizer-process-BgitfZ-Q.js → external-optimizer-process-za-SsIap.js} +2 -2
- package/dist/{external-optimizer-process-BgitfZ-Q.js.map → external-optimizer-process-za-SsIap.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-CeHuAaMJ.js → external-optimizer-subprocess-B44kOfZb.js} +3 -3
- package/dist/external-optimizer-subprocess-B44kOfZb.js.map +1 -0
- package/dist/{feedback-trajectory-BgomnFOe.d.ts → feedback-trajectory-P37ju-gW.d.ts} +3 -3
- package/dist/{feedback-trajectory-BgomnFOe.d.ts.map → feedback-trajectory-P37ju-gW.d.ts.map} +1 -1
- package/dist/{finding-composition-DlvE1Qul.js → finding-composition-Belh_qU3.js} +12 -3
- package/dist/finding-composition-Belh_qU3.js.map +1 -0
- package/dist/{findings-store-Bp-0Guuv.js → findings-store-DWiBvJVM.js} +22 -5
- package/dist/findings-store-DWiBvJVM.js.map +1 -0
- package/dist/fuzz.js +1 -1
- package/dist/hosted/index.js +1 -1
- package/dist/{index-BwutH5i7.d.ts → index-BJM4rwP2.d.ts} +6 -6
- package/dist/{index-BwutH5i7.d.ts.map → index-BJM4rwP2.d.ts.map} +1 -1
- package/dist/{index-DAk2quYu.d.ts → index-C3t17kHu.d.ts} +16 -21
- package/dist/index-C3t17kHu.d.ts.map +1 -0
- package/dist/{index-Bn-P_XAl.d.ts → index-Dt6E4BHX.d.ts} +6 -6
- package/dist/{index-Bn-P_XAl.d.ts.map → index-Dt6E4BHX.d.ts.map} +1 -1
- package/dist/{index-ByJ20lyu.d.ts → index-chNyKroE.d.ts} +8 -8
- package/dist/{index-ByJ20lyu.d.ts.map → index-chNyKroE.d.ts.map} +1 -1
- package/dist/{index-CtcCSO1Y.d.ts → index-f09MX_H_.d.ts} +2 -2
- package/dist/{index-CtcCSO1Y.d.ts.map → index-f09MX_H_.d.ts.map} +1 -1
- package/dist/index.d.ts +52 -146
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +80 -204
- package/dist/index.js.map +1 -1
- package/dist/{integrity-B1V6klyO.js → integrity-Vx5xgn3s.js} +2 -2
- package/dist/{integrity-B1V6klyO.js.map → integrity-Vx5xgn3s.js.map} +1 -1
- package/dist/{judge-calibration-D93UCm1k.js → judge-calibration-DKR6Xi-w.js} +13 -36
- package/dist/judge-calibration-DKR6Xi-w.js.map +1 -0
- package/dist/judge-calibration-DnrbQaXI.d.ts.map +1 -1
- package/dist/{judge-gate-DMlX38B7.js → judge-gate-CIlBrn6q.js} +2 -2
- package/dist/{judge-gate-DMlX38B7.js.map → judge-gate-CIlBrn6q.js.map} +1 -1
- package/dist/{kind-factory-VKysp1ds.js → kind-factory-Cfz8DUCc.js} +2 -2
- package/dist/{kind-factory-VKysp1ds.js.map → kind-factory-Cfz8DUCc.js.map} +1 -1
- package/dist/{llm-judge-CyDyqb6V.js → llm-judge-CHLvjbVR.js} +14 -14
- package/dist/{llm-judge-CyDyqb6V.js.map → llm-judge-CHLvjbVR.js.map} +1 -1
- package/dist/{matrix-4DcRFJ7Q.d.ts → matrix-Cy3oy7Rq.d.ts} +2 -2
- package/dist/{matrix-4DcRFJ7Q.d.ts.map → matrix-Cy3oy7Rq.d.ts.map} +1 -1
- package/dist/{mean-DDFhslBt.js → mean-D1d6gedE.js} +28 -2
- package/dist/{mean-DDFhslBt.js.map → mean-D1d6gedE.js.map} +1 -1
- package/dist/meta-eval/index.d.ts +1 -1
- package/dist/meta-eval/index.js +6 -6
- package/dist/{mint-D9ZysYtY.js → mint-BH_3Ak2x.js} +3 -3
- package/dist/{mint-D9ZysYtY.js.map → mint-BH_3Ak2x.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 +1 -1
- package/dist/{outcome-observations-ahbNtxuP.js → outcome-observations-DZu5hLYF.js} +3 -3
- package/dist/{outcome-observations-ahbNtxuP.js.map → outcome-observations-DZu5hLYF.js.map} +1 -1
- package/dist/{paired-arms-C0Vd9bGG.js → paired-arms-3-tsXvvC.js} +5 -5
- package/dist/{paired-arms-C0Vd9bGG.js.map → paired-arms-3-tsXvvC.js.map} +1 -1
- package/dist/{paired-binary-B477lksS.js → paired-binary-CMzmSSd5.js} +2 -2
- package/dist/{paired-binary-B477lksS.js.map → paired-binary-CMzmSSd5.js.map} +1 -1
- package/dist/{paired-promotion-decision-5Pl_qJLX.js → paired-promotion-decision-CiueaTTj.js} +3 -3
- package/dist/{paired-promotion-decision-5Pl_qJLX.js.map → paired-promotion-decision-CiueaTTj.js.map} +1 -1
- package/dist/{paired-tests-kASEKpc-.js → paired-tests-Buh6feJ4.js} +2 -2
- package/dist/{paired-tests-kASEKpc-.js.map → paired-tests-Buh6feJ4.js.map} +1 -1
- package/dist/{pareto-DY_x3HHg.js → pareto-B34N-vnn.js} +7 -7
- package/dist/{pareto-DY_x3HHg.js.map → pareto-B34N-vnn.js.map} +1 -1
- package/dist/pipelines/index.js +1 -1
- package/dist/{power-and-mde-CRxepy8y.js → power-and-mde-g3vOn0XK.js} +3 -3
- package/dist/{power-and-mde-CRxepy8y.js.map → power-and-mde-g3vOn0XK.js.map} +1 -1
- package/dist/{prime-protocol-COFkxas6.d.ts → prime-protocol-DKRk4l4U.d.ts} +2 -2
- package/dist/{prime-protocol-COFkxas6.d.ts.map → prime-protocol-DKRk4l4U.d.ts.map} +1 -1
- package/dist/{promotion-policy-f79oklG_.js → promotion-policy-DLPcEnCf.js} +4 -4
- package/dist/{promotion-policy-f79oklG_.js.map → promotion-policy-DLPcEnCf.js.map} +1 -1
- package/dist/{proposal-findings-BclX0PFh.js → proposal-findings-CZJkS_GY.js} +2 -2
- package/dist/{proposal-findings-BclX0PFh.js.map → proposal-findings-CZJkS_GY.js.map} +1 -1
- package/dist/{registry-CMtGAxOw.d.ts → registry-Ck7w-rr5.d.ts} +3 -3
- package/dist/{registry-CMtGAxOw.d.ts.map → registry-Ck7w-rr5.d.ts.map} +1 -1
- package/dist/{release-confidence--1iP-d3x.js → release-confidence-nfQNh3eb.js} +4 -4
- package/dist/{release-confidence--1iP-d3x.js.map → release-confidence-nfQNh3eb.js.map} +1 -1
- package/dist/{render-CFrJ7M0o.js → render-DIQ5nqp_.js} +80 -51
- package/dist/{render-CFrJ7M0o.js.map → render-DIQ5nqp_.js.map} +1 -1
- package/dist/reporting.js +5 -5
- package/dist/{reward-P0VD_Aft.js → reward-DqMfaEax.js} +10 -2
- package/dist/{reward-P0VD_Aft.js.map → reward-DqMfaEax.js.map} +1 -1
- package/dist/{reward-hacking-D-Vi9-fE.js → reward-hacking-B88f-Kh7.js} +4 -4
- package/dist/{reward-hacking-D-Vi9-fE.js.map → reward-hacking-B88f-Kh7.js.map} +1 -1
- package/dist/rl.d.ts +3 -3
- package/dist/rl.js +12 -12
- package/dist/rollout/index.d.ts +2 -2
- package/dist/rollout/index.js +3 -3
- package/dist/{rollout-BC4u-d_I.js → rollout-CbOGxuHl.js} +2 -2
- package/dist/{rollout-BC4u-d_I.js.map → rollout-CbOGxuHl.js.map} +1 -1
- package/dist/{rollout-nodes-BHR9Q5jP.js → rollout-nodes-DKZ3vAYY.js} +2 -2
- package/dist/{rollout-nodes-BHR9Q5jP.js.map → rollout-nodes-DKZ3vAYY.js.map} +1 -1
- package/dist/{rubric-predictive-validity-BdxXDN6h.js → rubric-predictive-validity-D_6ZPTNR.js} +2 -2
- package/dist/{rubric-predictive-validity-BdxXDN6h.js.map → rubric-predictive-validity-D_6ZPTNR.js.map} +1 -1
- package/dist/{run-record-B-G2Tu11.js → run-record-CbUi2adH.js} +3 -3
- package/dist/{run-record-B-G2Tu11.js.map → run-record-CbUi2adH.js.map} +1 -1
- package/dist/{run-record-DEjx43cI.js → run-record-CpLS4Uid.js} +2 -2
- package/dist/{run-record-DEjx43cI.js.map → run-record-CpLS4Uid.js.map} +1 -1
- package/dist/search/index.d.ts +2 -2
- package/dist/search/index.js +2 -2
- package/dist/{search-kernel-BjR2Sr7w.d.ts → search-kernel-kwO-sgCq.d.ts} +2 -2
- package/dist/{search-kernel-BjR2Sr7w.d.ts.map → search-kernel-kwO-sgCq.d.ts.map} +1 -1
- package/dist/{search-ledger-BsVi-Ngv.js → search-ledger-CV_bQQed.js} +2 -2
- package/dist/{search-ledger-BsVi-Ngv.js.map → search-ledger-CV_bQQed.js.map} +1 -1
- package/dist/{search-shipper-pdg4viK7.js → search-shipper-CA3ooBbC.js} +2 -2
- package/dist/{search-shipper-pdg4viK7.js.map → search-shipper-CA3ooBbC.js.map} +1 -1
- package/dist/{search-summary-DedyTdiM.js → search-summary-DBMjdecs.js} +4 -4
- package/dist/{search-summary-DedyTdiM.js.map → search-summary-DBMjdecs.js.map} +1 -1
- package/dist/{sequential-Zp1tsxqI.js → sequential-BPbvgzfe.js} +2 -2
- package/dist/{sequential-Zp1tsxqI.js.map → sequential-BPbvgzfe.js.map} +1 -1
- package/dist/{skillopt-optimization-method-O35aMdxU.js → skillopt-optimization-method-D72XZZfR.js} +9 -9
- package/dist/{skillopt-optimization-method-O35aMdxU.js.map → skillopt-optimization-method-D72XZZfR.js.map} +1 -1
- package/dist/{statistical-heldout-DYF5f10K.js → statistical-heldout-NBxC7jGo.js} +3 -3
- package/dist/{statistical-heldout-DYF5f10K.js.map → statistical-heldout-NBxC7jGo.js.map} +1 -1
- package/dist/{statistical-heldout-bU7FBNjR.d.ts → statistical-heldout-_dXihc9T.d.ts} +3 -3
- package/dist/{statistical-heldout-bU7FBNjR.d.ts.map → statistical-heldout-_dXihc9T.d.ts.map} +1 -1
- package/dist/{store-otlp-DmLm65XW.js → store-otlp-BfPRa2p8.js} +2 -2
- package/dist/{store-otlp-DmLm65XW.js.map → store-otlp-BfPRa2p8.js.map} +1 -1
- package/dist/{store-tool-spans-lWf6INTH.d.ts → store-tool-spans-BgwVHXma.d.ts} +4 -4
- package/dist/{store-tool-spans-lWf6INTH.d.ts.map → store-tool-spans-BgwVHXma.d.ts.map} +1 -1
- package/dist/{store-tool-spans-GGaCijFy.js → store-tool-spans-BrSWHB9O.js} +3 -3
- package/dist/{store-tool-spans-GGaCijFy.js.map → store-tool-spans-BrSWHB9O.js.map} +1 -1
- package/dist/{summary-report-B4BLQb88.js → summary-report-C3IJvp0i.js} +7 -7
- package/dist/{summary-report-B4BLQb88.js.map → summary-report-C3IJvp0i.js.map} +1 -1
- package/dist/supervisor-run/index.d.ts +6 -0
- package/dist/supervisor-run/index.d.ts.map +1 -1
- package/dist/supervisor-run/index.js +3 -3
- package/dist/{systemone-BDVgb6XV.d.ts → systemone-CBvL1gx-.d.ts} +14 -8
- package/dist/systemone-CBvL1gx-.d.ts.map +1 -0
- package/dist/{systemone-review-S49gsy_y.js → systemone-review-BusEIzsg.js} +2 -2
- package/dist/{systemone-review-S49gsy_y.js.map → systemone-review-BusEIzsg.js.map} +1 -1
- package/dist/{systemone-review-BLgxt-4Q.d.ts → systemone-review-nbuMxxmi.d.ts} +2 -2
- package/dist/{systemone-review-BLgxt-4Q.d.ts.map → systemone-review-nbuMxxmi.d.ts.map} +1 -1
- package/dist/systemone.d.ts +4 -4
- package/dist/systemone.js +3 -3
- package/dist/{tool-groups-C3K-d3gd.d.ts → tool-groups-BQsjZD9F.d.ts} +3 -3
- package/dist/tool-groups-BQsjZD9F.d.ts.map +1 -0
- package/dist/{tool-waste-CRJqOASv.js → tool-waste-Czt3aWID.js} +3 -3
- package/dist/{tool-waste-CRJqOASv.js.map → tool-waste-Czt3aWID.js.map} +1 -1
- package/dist/{trace-questions-ej5AEJ5X.d.ts → trace-questions-CpTrtEcm.d.ts} +3 -3
- package/dist/{trace-questions-ej5AEJ5X.d.ts.map → trace-questions-CpTrtEcm.d.ts.map} +1 -1
- package/dist/trace-repair/index.d.ts +1 -1
- package/dist/trace-repair/index.js +2 -2
- package/dist/traces.d.ts +3 -3
- package/dist/traces.js +4 -4
- package/dist/{tree-C2N31CoI.js → tree-DPwgM2LX.js} +6 -6
- package/dist/{tree-C2N31CoI.js.map → tree-DPwgM2LX.js.map} +1 -1
- package/dist/{types-DzCO-S1m.d.ts → types-B9s9Xo5_.d.ts} +24 -2
- package/dist/types-B9s9Xo5_.d.ts.map +1 -0
- package/dist/{types-BF-xQGQn.d.ts → types-CBkDo7ja.d.ts} +2 -2
- package/dist/{types-BF-xQGQn.d.ts.map → types-CBkDo7ja.d.ts.map} +1 -1
- package/dist/{types-BK0blWhe.js → types-CL1_N7AZ.js} +2 -2
- package/dist/{types-BK0blWhe.js.map → types-CL1_N7AZ.js.map} +1 -1
- package/dist/{types-yofaDmQa.d.ts → types-D91tnhzf.d.ts} +2 -2
- package/dist/{types-yofaDmQa.d.ts.map → types-D91tnhzf.d.ts.map} +1 -1
- package/dist/wire/index.d.ts +3 -3
- package/docs/behavioral-trace-questions.md +23 -0
- package/docs/public-api.md +1 -1
- package/package.json +2 -2
- package/dist/agreement-irr-w-t3XUTD.js.map +0 -1
- package/dist/benchmark-command-DZQrxqUL.js.map +0 -1
- package/dist/default-registry-DKlKVXMx.d.ts.map +0 -1
- package/dist/default-registry-gmDeN0KT.js.map +0 -1
- package/dist/external-optimizer-subprocess-CeHuAaMJ.js.map +0 -1
- package/dist/finding-composition-DlvE1Qul.js.map +0 -1
- package/dist/findings-store-Bp-0Guuv.js.map +0 -1
- package/dist/index-DAk2quYu.d.ts.map +0 -1
- package/dist/judge-calibration-D93UCm1k.js.map +0 -1
- package/dist/systemone-BDVgb6XV.d.ts.map +0 -1
- package/dist/tool-groups-C3K-d3gd.d.ts.map +0 -1
- package/dist/types-DzCO-S1m.d.ts.map +0 -1
package/CHANGELOG.md
CHANGED
|
@@ -4,6 +4,42 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
|
|
|
4
4
|
|
|
5
5
|
---
|
|
6
6
|
|
|
7
|
+
## [1.0.14] — 2026-11-02
|
|
8
|
+
|
|
9
|
+
- fix(supervisor-run): build Discovery rows from execution results (#1189)
|
|
10
|
+
- fix(analyst): pin the benchmark lock digest for harness-sessions 1.0.8 (#1188)
|
|
11
|
+
|
|
12
|
+
## [1.0.13] — 2026-11-02
|
|
13
|
+
|
|
14
|
+
- feat(supervisor-run): write a run's sessions as OTLP spans (#1186)
|
|
15
|
+
- fix(supervisor-run): read whole Discovery episodes and screen rows on calibration (#1185)
|
|
16
|
+
- feat(calibration): precision, recall, AUC and a cluster-bootstrap CI on every calibration row
|
|
17
|
+
- chore(release): 1.0.12
|
|
18
|
+
- fix(statistics): exact retired-contract compat for summarizeDistribution
|
|
19
|
+
- chore(release): 1.0.11
|
|
20
|
+
- fix(statistics): export summarizeDistribution from the package root
|
|
21
|
+
- chore(release): 1.0.10
|
|
22
|
+
- chore(release): 1.0.9
|
|
23
|
+
- feat(statistics): summarizeDistribution + percentileNearestRank (consolidation home)
|
|
24
|
+
- chore(release): 1.0.8
|
|
25
|
+
- chore(analyst): re-pin the implementation manifest on merged main
|
|
26
|
+
- feat(statistics,harness): two-sample Welch evidence + lifted sandbox assertion library
|
|
27
|
+
- chore(release): 1.0.7 (#1172)
|
|
28
|
+
- docs(statistics): keep inference policy and exports readable (#1173)
|
|
29
|
+
- feat(statistics): own independent effect summaries and ROC area (#1171)
|
|
30
|
+
- chore(release): 1.0.6 (#1170)
|
|
31
|
+
- refactor(reporting): share text table layout and simplify summaries (#1169)
|
|
32
|
+
- refactor(harness): retain unavailable verdicts through cleanup (#1168)
|
|
33
|
+
- chore(release): 1.0.5 (#1167)
|
|
34
|
+
- refactor(harness): normalize compact named expectations (#1166)
|
|
35
|
+
- chore(release): 1.0.4 (#1165)
|
|
36
|
+
- refactor(harness): own resource acquisition and readiness cleanup (#1164)
|
|
37
|
+
- chore(release): 1.0.3 (#1163)
|
|
38
|
+
- chore(release): declare the existing deployment runner for workflow lint
|
|
39
|
+
- fix(release): bind bootstrap to the existing qualified action revision
|
|
40
|
+
- fix(release): use the qualified pnpm bootstrap in publication workflows
|
|
41
|
+
- Refactor sandbox integration scenarios into the eval harness (#1161)
|
|
42
|
+
|
|
7
43
|
## [1.0.12] — 2026-10-31
|
|
8
44
|
|
|
9
45
|
- fix(statistics): exact retired-contract compat for summarizeDistribution
|
package/dist/adapters/http.d.ts
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import { d as CostReceipt } from "../cost-ledger-Czro9wmo.js";
|
|
2
|
-
import { R as Scenario, d as DispatchContext, f as DispatchFn } from "../types-
|
|
3
|
-
import "../index-
|
|
2
|
+
import { R as Scenario, d as DispatchContext, f as DispatchFn } from "../types-CBkDo7ja.js";
|
|
3
|
+
import "../index-chNyKroE.js";
|
|
4
4
|
export interface HttpDispatchOptions<TScenario extends Scenario, _TArtifact> {
|
|
5
5
|
/** Static endpoint URL. Mutually exclusive with `resolveUrl`. */
|
|
6
6
|
url?: string;
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import { f as CostReceiptInput, m as CustomTokenPricing } from "./cost-ledger-Czro9wmo.js";
|
|
2
|
-
import {
|
|
2
|
+
import { X as TraceAnalysisStore, i as AnalystFinding, m as EvidenceRef, p as AnalystUsageReceipt } from "./types-B9s9Xo5_.js";
|
|
3
3
|
import { C as ChatResponse, S as ChatRequest } from "./types-0KglnQ9o.js";
|
|
4
4
|
import { AgentProfile, AgentProfile as AgentProfile$1, HarnessType, HarnessType as HarnessType$1 } from "@tangle-network/agent-interface";
|
|
5
5
|
interface ExternalOptimizerProcessLimits {
|
|
@@ -466,4 +466,4 @@ declare function agentProfileId(profile: AgentProfile): string;
|
|
|
466
466
|
*/
|
|
467
467
|
declare function agentProfileHash(profile: AgentProfile): string;
|
|
468
468
|
export { traceStoreEvidenceResolver as A, ExternalOptimizerModelCallRequest as B, AnalystEvidenceResolutionError as C, AnalystLatencyDistribution as D, AnalystIssueExpectation as E, ExternalOptimizerEndpointFormat as F, ExternalOptimizerRunnerCommand as G, ExternalOptimizerModelExecutionObservation as H, ExternalOptimizerEvaluationObservation as I, ExternalTextEvaluationRequest as J, ExternalOptimizerWireCounts as K, ExternalOptimizerEvaluationRefusalReason as L, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS as M, ExternalOptimizerCallbackLimits as N, RunAnalystBenchmarkOptions as O, ExternalOptimizerChatRequest as P, ExternalOptimizerModelBudget as R, AnalystEvidenceResolution as S, AnalystFindingScore as T, ExternalOptimizerProcessLimits as U, ExternalOptimizerModelCallResult as V, ExternalOptimizerResumeMode as W, AnalystBenchmarkProvenance as _, ProfileAxisSpec as a, AnalystBenchmarkSummary as b, expandProfileAxes as c, AnalystBenchmarkDatasetRef as d, AnalystBenchmarkDescriptor as f, AnalystBenchmarkOutput as g, AnalystBenchmarkObservation as h, HarnessType$1 as i, DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS as j, runAnalystBenchmark as k, harnessAxisOf as l, AnalystBenchmarkLabelState as m, CODING_HARNESSES as n, agentProfileHash as o, AnalystBenchmarkError as p, ExternalTextCandidate as q, HARNESS_NATIVE_MODEL as r, agentProfileId as s, AgentProfile$1 as t, AnalystBenchmarkCase as u, AnalystBenchmarkResult as v, AnalystEvidenceResolver as w, AnalystEvidenceExpectation as x, AnalystBenchmarkRunner as y, ExternalOptimizerModelCall as z };
|
|
469
|
-
//# sourceMappingURL=agent-profile-
|
|
469
|
+
//# sourceMappingURL=agent-profile-CPv4tENR.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"agent-profile-
|
|
1
|
+
{"version":3,"file":"agent-profile-CPv4tENR.d.ts","names":["_$___cost_ledger0.CostReceiptInput"],"sources":["../src/campaign/external-optimizer-contracts.ts","../src/analyst/benchmark.ts","../src/agent-profile.ts"],"mappings":";;;;UAciB;;EAEf;;EAEA;;EAEA;;cAGW,2CAA2C,SAAS;UAOhD;;EAEf;;EAEA;;cAGW,4CAA4C,SAAS;UAMjD;EACf;EACA;EACA,MAAM,OAAO;;EAEb,SAAS,QAAQ;;KAGP;KAEA,iCAAiC;UAE5B;EACf,WAAW;EACX;;KAqDG,aAAa,KAAK,eAAc,6BACjC,IACA,0BAA0B,gBACf,aAAa,OACtB,+BACc,WAAW,IAAI,aAAa,EAAE,SAC1C;;KAGI,+BAA+B,aACzC,KAAK;EAA0B;;KAGrB;;UAMK;;WAEN;;WAEA,SAAS;;WAET,iBAAiB;WACjB,QAAQ;;;KAIP;WAEG;;WAEA,UAAU;;;;;;;WAOV,SAPUA;;WASV;;WAGA;;WAEA;;WAEA,SATkCA;;WAWlC;;;;;;;;;;;;KAaH,8BACV,SAAS,sCACN,QAAQ;;KAGD;WAEG;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;WAGA;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;KAGH;;;;;;KAUA;WAEG;WACA;WACA,WAAW;WACX;;WAGA;WACA;WACA,WAAW;WACX;WACA;;WAEA;WACA;;WAGA;WACA;WACA,QAAQ;WACR,YAAY;WACZ;WACA;;UAGE;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;EAUA;;;;;;EAMA,UAAU;;EAEV;;;UAIe;;EAEf;;EAEA;;UCpQe;EACf;EACA,OAAO;;UAGQ;EACf;EACA;EACA;EACA;EACA,oBAAoB;EACpB;;EAEA,4BAA4B;;KAGlB;UAEK,qBAAqB;EACpC;;EAEA;;EAEA,YAAY;EACZ,OAAO;EACP,yBAAyB;;EAEzB,2BAA2B;EAC3B;EACA,WAAW;;UAGI;EACf;EACA;EACA;EACA;EACA;EACA,mBAAmB;EACnB;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;EACA;;UAGe;EACf,UAAU;EACV;EACA;;UAGe;EACf;EACA;EACA,oBAAoB;EACpB,QAAQ;;EAER;;KAGU,wBAAwB,qBAAqB;EACvD;EACA,WAAW;EACX,UAAU;EACV,SAAS;gBACK;;;;;iBAMA,2BAA2B,QACzC,WAAW,OAAO,WAAW,qBAC5B,wBAAwB;UAoBV;EACf,mBAAmB;EACnB,QAAQ;EACR,WAAW;;;;;EAKX;;EAEA,QAAQ;;UAGO;EACf;EACA;EACA;EACA;;UAGe,uBAAuB;EACtC;EACA,QACE,OAAO,QACP;IAAW;IAAgB;IAAoB,SAAS;MACvD,yBAAyB,QAAQ;;UAGrB;EACf;EACA;EACA;EACA,YAAY;EACZ;EACA;EACA;EACA;EACA,mBAAmB;EACnB,OAAO;EACP,qBAAqB;EACrB;EACA,eAAe;EACf,QAAQ;EACR,iBAAiB;EACjB,QAAQ;;UAGO;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;EACA,WAAW;EACX;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;;UAGe;EACf;EACA,UAAU;EACV;EACA,cAAc;EACd,WAAW;;UAGI,mCAAmC;EAClD;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf,YAAY;EACZ,cAAc;EACd,WAAW;;UAGI,2BAA2B;EAC1C,gBAAgB,qBAAqB;EACrC,kBAAkB,uBAAuB;EACzC;EACA;EACA;EACA,kBAAkB,wBAAwB;EAC1C,YAAY;;EAEZ,+BAA+B;EAC/B,iBAAiB,aAAa,uCAAuC;EACrE,SAAS;;iBAGW,oBAAoB,QACxC,SAAS,2BAA2B,UACnC,QAAQ;;;;;;;;cCnQE,2BAA2B;UAOvB;;;EAGf,MAAM;;EAEN,qBAAqB;;;EAGrB;;;;;;EAMA;;;;;;;cAQW;;;;;;;;;;;;;;;;;;;iBAoBG,kBAAkB,MAAM,kBAAkB;;;;;;;;iBAwD1C,cACd,SAAS,KAAK;EACX,SAAS;EAAa;;;;;;;;;iBAiBX,eAAe,SAAS;;;;;;;;;;;iBAmDxB,iBAAiB,SAAS"}
|
|
@@ -1,5 +1,6 @@
|
|
|
1
1
|
import { s as ValidationError } from "./errors-DeEEV9Vx.js";
|
|
2
|
-
import {
|
|
2
|
+
import { c as clusterBootstrap } from "./mean-D1d6gedE.js";
|
|
3
|
+
import { r as continuousAgreement } from "./judge-calibration-DKR6Xi-w.js";
|
|
3
4
|
// src/statistics/agreement-irr.ts
|
|
4
5
|
/**
|
|
5
6
|
* Inter-rater reliability — Krippendorff's α under the squared-difference
|
|
@@ -55,20 +56,16 @@ function interRaterReliability(judgeScores) {
|
|
|
55
56
|
return 1 - observedDisagreement / expectedDisagreement;
|
|
56
57
|
}
|
|
57
58
|
/**
|
|
58
|
-
* Corpus-wide inter-rater agreement across N items × M judges × D dimensions
|
|
59
|
+
* Corpus-wide inter-rater agreement across N items × M judges × D dimensions, the
|
|
60
|
+
* question `interRaterReliability`'s within-item α cannot answer.
|
|
59
61
|
*
|
|
60
62
|
* For each dimension, builds the [n_items][n_judges] matrix of scores
|
|
61
63
|
* (keeping only items every judge rated on that dimension), then runs
|
|
62
64
|
* `continuousAgreement` to get ICC(2,1), κ_w, Pearson, Spearman, and
|
|
63
65
|
* bootstrap CIs. Reports a pooled mean across dimensions as a single
|
|
64
|
-
* "is this judge panel reliable on this corpus?" number.
|
|
65
|
-
*
|
|
66
|
-
*
|
|
67
|
-
* - Empty input throws.
|
|
68
|
-
* - Fewer than 2 judges or fewer than 2 items per dimension throws.
|
|
69
|
-
* - A judge present in some dimensions but with zero scored items on
|
|
70
|
-
* another dimension throws (would silently shrink the matrix).
|
|
71
|
-
* - Duplicate (itemId, judgeName, dimension) records throw.
|
|
66
|
+
* "is this judge panel reliable on this corpus?" number. Empty input, fewer than 2 judges or
|
|
67
|
+
* items, a judge without scores on a dimension and duplicate records all throw: a silently
|
|
68
|
+
* shrunk matrix would corrupt the overall metric.
|
|
72
69
|
*/
|
|
73
70
|
function corpusInterRaterAgreement(records, opts = {}) {
|
|
74
71
|
if (records.length === 0) throw new ValidationError("corpusInterRaterAgreement: no score records supplied");
|
|
@@ -96,17 +93,10 @@ function corpusInterRaterAgreement(records, opts = {}) {
|
|
|
96
93
|
const kappas = [];
|
|
97
94
|
for (const dim of targetDims) {
|
|
98
95
|
const byJudge = grid.get(dim);
|
|
99
|
-
const judgeItemCounts =
|
|
100
|
-
for (const j of targetJudges) judgeItemCounts[j] = byJudge.get(j)?.size ?? 0;
|
|
96
|
+
const judgeItemCounts = Object.fromEntries(targetJudges.map((j) => [j, byJudge.get(j)?.size ?? 0]));
|
|
101
97
|
const emptyJudges = targetJudges.filter((j) => judgeItemCounts[j] === 0);
|
|
102
98
|
if (emptyJudges.length > 0) throw new ValidationError(`corpusInterRaterAgreement: dimension '${dim}' has no scores from judge(s) ${emptyJudges.join(", ")} (counts: ${JSON.stringify(judgeItemCounts)})`);
|
|
103
|
-
|
|
104
|
-
for (const j of targetJudges) {
|
|
105
|
-
const ids = new Set(byJudge.get(j).keys());
|
|
106
|
-
if (commonItems === null) commonItems = ids;
|
|
107
|
-
else commonItems = new Set([...commonItems].filter((x) => ids.has(x)));
|
|
108
|
-
}
|
|
109
|
-
const sortedItems = [...commonItems ?? /* @__PURE__ */ new Set()].sort();
|
|
99
|
+
const sortedItems = targetJudges.map((j) => [...byJudge.get(j).keys()]).reduce((common, ids) => common.filter((x) => ids.includes(x))).sort();
|
|
110
100
|
if (sortedItems.length < 2) throw new ValidationError(`corpusInterRaterAgreement: dimension '${dim}' has ${sortedItems.length} item(s) rated by all ${targetJudges.length} judges (need ≥2)`);
|
|
111
101
|
const matrix = sortedItems.map((itemId) => targetJudges.map((j) => byJudge.get(j).get(itemId)));
|
|
112
102
|
const agreement = continuousAgreement(matrix, opts);
|
|
@@ -155,10 +145,18 @@ function corpusInterRaterAgreementFromJudgeScores(itemsScores, opts = {}) {
|
|
|
155
145
|
* Unweighted Cohen's κ for paired nominal labels. Labels carry no order, so a
|
|
156
146
|
* near miss counts as a miss; use `calibrateJudge` for ordinal scores.
|
|
157
147
|
*/
|
|
158
|
-
function nominalAgreement(pairs) {
|
|
148
|
+
function nominalAgreement(pairs, options = {}) {
|
|
149
|
+
if (options.clusters && options.clusters.length !== pairs.length) throw new ValidationError("nominalAgreement: one cluster per pair");
|
|
150
|
+
const kappaCi = clusterBootstrap(pairs, (_, index) => options.clusters?.[index] ?? String(index), (sample) => cohenKappa(sample).kappa, options);
|
|
151
|
+
return {
|
|
152
|
+
n: pairs.length,
|
|
153
|
+
...cohenKappa(pairs),
|
|
154
|
+
kappaCi
|
|
155
|
+
};
|
|
156
|
+
}
|
|
157
|
+
function cohenKappa(pairs) {
|
|
159
158
|
const n = pairs.length;
|
|
160
159
|
if (n === 0) return {
|
|
161
|
-
n: 0,
|
|
162
160
|
agreement: NaN,
|
|
163
161
|
kappa: NaN
|
|
164
162
|
};
|
|
@@ -175,11 +173,10 @@ function nominalAgreement(pairs) {
|
|
|
175
173
|
let expected = 0;
|
|
176
174
|
for (const [label, count] of left) expected += count / n * ((right.get(label) ?? 0) / n);
|
|
177
175
|
return {
|
|
178
|
-
n,
|
|
179
176
|
agreement: observed,
|
|
180
177
|
kappa: expected === 1 ? NaN : (observed - expected) / (1 - expected)
|
|
181
178
|
};
|
|
182
179
|
}
|
|
183
180
|
export { nominalAgreement as i, corpusInterRaterAgreementFromJudgeScores as n, interRaterReliability as r, corpusInterRaterAgreement as t };
|
|
184
181
|
|
|
185
|
-
//# sourceMappingURL=agreement-irr-
|
|
182
|
+
//# sourceMappingURL=agreement-irr-DXcXEt4I.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agreement-irr-DXcXEt4I.js","names":[],"sources":["../src/statistics/agreement-irr.ts"],"sourcesContent":["import { ValidationError } from '../errors'\nimport {\n type ContinuousAgreement,\n type ContinuousAgreementOptions,\n continuousAgreement,\n} from '../judge-calibration'\nimport type { DimensionJudgeScore } from '../types'\nimport { type BootstrapInterval, clusterBootstrap } from './internal'\n\n/**\n * Inter-rater reliability — Krippendorff's α under the squared-difference\n * metric, pooled across dimensions.\n *\n * Each inner array is one judge's scores. Items are matched by position\n * WITHIN a dimension: the k-th score a judge supplies carrying dimension\n * `d` is item k of `d`, and the ratings compared against each other are\n * the ones different judges gave to the same item. Every judge that scores\n * a dimension at all must supply the same number of scores for it —\n * ragged input cannot be aligned into items and throws rather than\n * comparing mismatched items.\n *\n * α = 1 − D_observed / D_expected: D_observed averages the squared\n * difference over within-item judge pairs, D_expected over every pair of\n * ratings irrespective of item. α = 1 is perfect agreement, 0 is chance,\n * negative is systematic disagreement.\n */\nexport function interRaterReliability(judgeScores: DimensionJudgeScore[][]): number {\n if (judgeScores.length < 2) return 1\n\n // dimension → one score sequence per judge, in that judge's supplied order.\n const perDimension = new Map<string, number[][]>()\n for (let judgeIndex = 0; judgeIndex < judgeScores.length; judgeIndex++) {\n for (const s of judgeScores[judgeIndex]!) {\n let byJudge = perDimension.get(s.dimension)\n if (byJudge === undefined) {\n byJudge = Array.from({ length: judgeScores.length }, () => [] as number[])\n perDimension.set(s.dimension, byJudge)\n }\n byJudge[judgeIndex]!.push(s.score)\n }\n }\n\n const allValues: number[] = []\n const pairDiffs: number[] = []\n\n for (const [dimension, byJudge] of perDimension) {\n const scoring = byJudge.filter((scores) => scores.length > 0)\n if (scoring.length < 2) continue\n const itemCount = scoring[0]!.length\n if (scoring.some((scores) => scores.length !== itemCount)) {\n throw new ValidationError(\n `interRaterReliability: dimension '${dimension}' has judges supplying ` +\n `${scoring.map((scores) => scores.length).join('/')} scores — items cannot be aligned`,\n )\n }\n for (let item = 0; item < itemCount; item++) {\n const ratings = scoring.map((scores) => scores[item]!)\n for (const v of ratings) allValues.push(v)\n for (let i = 0; i < ratings.length; i++) {\n for (let j = i + 1; j < ratings.length; j++) {\n pairDiffs.push((ratings[i]! - ratings[j]!) ** 2)\n }\n }\n }\n }\n\n if (pairDiffs.length === 0 || allValues.length < 2) return 1\n\n const observedDisagreement = pairDiffs.reduce((a, b) => a + b, 0) / pairDiffs.length\n\n // Expected disagreement from all possible pairings of values\n let expectedDisagreement = 0\n let expectedCount = 0\n for (let i = 0; i < allValues.length; i++) {\n for (let j = i + 1; j < allValues.length; j++) {\n expectedDisagreement += (allValues[i]! - allValues[j]!) ** 2\n expectedCount++\n }\n }\n expectedDisagreement = expectedCount > 0 ? expectedDisagreement / expectedCount : 0\n\n if (expectedDisagreement === 0) return 1\n return 1 - observedDisagreement / expectedDisagreement\n}\n\nexport interface CorpusScoreRecord {\n /** Stable identifier for the rated item (scenario, span, turn, …). */\n itemId: string\n /** Identifier for the judge that produced this score. */\n judgeName: string\n /** Dimension name (matches `DimensionJudgeScore.dimension`). */\n dimension: string\n /** Numeric score; must be finite. */\n score: number\n}\n\nexport interface CorpusAgreementPerDimension extends ContinuousAgreement {\n dimension: string\n /** Item IDs that contributed to this dimension's matrix (every judge scored them). */\n itemIds: string[]\n /** Judge IDs that contributed to this dimension's matrix. */\n judgeIds: string[]\n}\n\nexport interface CorpusAgreementReport {\n /** Per-dimension ICC(2,1) + κ_w + Pearson + Spearman + bootstrap CIs. */\n perDimension: CorpusAgreementPerDimension[]\n /** Mean ICC across dimensions (NaN if no dimension yielded a finite ICC). */\n overallIcc: number\n /** Mean weighted κ across dimensions (NaN if none finite). */\n overallWeightedKappa: number\n /** Dimensions evaluated (sorted). */\n dimensions: string[]\n /** Judges seen across the corpus (sorted). */\n judgeIds: string[]\n}\n\nexport interface CorpusAgreementOptions extends ContinuousAgreementOptions {\n /** Restrict the audit to these dimensions; default every one in the input. A named absentee throws. */\n dimensions?: string[]\n /** Restrict the audit to these judges; default every one in the input. A named absentee throws. */\n judges?: string[]\n}\n\n/**\n * Corpus-wide inter-rater agreement across N items × M judges × D dimensions, the\n * question `interRaterReliability`'s within-item α cannot answer.\n *\n * For each dimension, builds the [n_items][n_judges] matrix of scores\n * (keeping only items every judge rated on that dimension), then runs\n * `continuousAgreement` to get ICC(2,1), κ_w, Pearson, Spearman, and\n * bootstrap CIs. Reports a pooled mean across dimensions as a single\n * \"is this judge panel reliable on this corpus?\" number. Empty input, fewer than 2 judges or\n * items, a judge without scores on a dimension and duplicate records all throw: a silently\n * shrunk matrix would corrupt the overall metric.\n */\nexport function corpusInterRaterAgreement(\n records: CorpusScoreRecord[],\n opts: CorpusAgreementOptions = {},\n): CorpusAgreementReport {\n if (records.length === 0) {\n throw new ValidationError('corpusInterRaterAgreement: no score records supplied')\n }\n\n const judgesSeen = new Set<string>()\n const dimsSeen = new Set<string>()\n // dimension → judge → itemId → score\n const grid = new Map<string, Map<string, Map<string, number>>>()\n\n for (const r of records) {\n if (!Number.isFinite(r.score)) {\n throw new ValidationError(\n `corpusInterRaterAgreement: non-finite score for (item=${r.itemId}, judge=${r.judgeName}, dim=${r.dimension})`,\n )\n }\n judgesSeen.add(r.judgeName)\n dimsSeen.add(r.dimension)\n const byJudge = grid.get(r.dimension) ?? new Map<string, Map<string, number>>()\n const byItem = byJudge.get(r.judgeName) ?? new Map<string, number>()\n if (byItem.has(r.itemId)) {\n throw new ValidationError(\n `corpusInterRaterAgreement: duplicate record for (item=${r.itemId}, judge=${r.judgeName}, dim=${r.dimension})`,\n )\n }\n byItem.set(r.itemId, r.score)\n byJudge.set(r.judgeName, byItem)\n grid.set(r.dimension, byJudge)\n }\n\n const targetDims = opts.dimensions ?? [...dimsSeen].sort()\n for (const d of targetDims) {\n if (!dimsSeen.has(d)) {\n throw new ValidationError(\n `corpusInterRaterAgreement: dimension '${d}' was requested but no records carry it`,\n )\n }\n }\n const targetJudges = opts.judges ? [...opts.judges] : [...judgesSeen].sort()\n for (const j of targetJudges) {\n if (!judgesSeen.has(j)) {\n throw new ValidationError(\n `corpusInterRaterAgreement: judge '${j}' was requested but produced no records`,\n )\n }\n }\n if (targetJudges.length < 2) {\n throw new ValidationError(\n `corpusInterRaterAgreement: need ≥2 judges, got ${targetJudges.length}`,\n )\n }\n\n const perDimension: CorpusAgreementPerDimension[] = []\n const iccs: number[] = []\n const kappas: number[] = []\n\n for (const dim of targetDims) {\n const byJudge = grid.get(dim)!\n // Fail loud: every requested judge must have scored ≥1 item on this dim.\n const judgeItemCounts = Object.fromEntries(\n targetJudges.map((j) => [j, byJudge.get(j)?.size ?? 0]),\n )\n const emptyJudges = targetJudges.filter((j) => judgeItemCounts[j] === 0)\n if (emptyJudges.length > 0) {\n throw new ValidationError(\n `corpusInterRaterAgreement: dimension '${dim}' has no scores from judge(s) ${emptyJudges.join(', ')} (counts: ${JSON.stringify(judgeItemCounts)})`,\n )\n }\n // Items rated by *every* requested judge on this dim.\n const sortedItems = targetJudges\n .map((j) => [...byJudge.get(j)!.keys()])\n .reduce((common, ids) => common.filter((x) => ids.includes(x)))\n .sort()\n if (sortedItems.length < 2) {\n throw new ValidationError(\n `corpusInterRaterAgreement: dimension '${dim}' has ${sortedItems.length} item(s) rated by all ${targetJudges.length} judges (need ≥2)`,\n )\n }\n\n const matrix: number[][] = sortedItems.map((itemId) =>\n targetJudges.map((j) => byJudge.get(j)!.get(itemId)!),\n )\n const agreement = continuousAgreement(matrix, opts)\n perDimension.push({\n ...agreement,\n dimension: dim,\n itemIds: sortedItems,\n judgeIds: [...targetJudges],\n })\n if (Number.isFinite(agreement.icc)) iccs.push(agreement.icc)\n if (Number.isFinite(agreement.weightedKappa)) kappas.push(agreement.weightedKappa)\n }\n\n const mean = (xs: number[]) =>\n xs.length === 0 ? Number.NaN : xs.reduce((a, b) => a + b, 0) / xs.length\n return {\n perDimension,\n overallIcc: mean(iccs),\n overallWeightedKappa: mean(kappas),\n dimensions: targetDims,\n judgeIds: targetJudges,\n }\n}\n\n/**\n * Convenience adapter for `DimensionJudgeScore[]` data keyed externally by item.\n *\n * Use when you have per-item arrays of `DimensionJudgeScore[]` (e.g. one\n * `ScenarioResult.judgeScores` per scenario) and want corpus-wide\n * agreement without manually flattening. `itemId` must be unique per\n * row of `itemsScores`.\n */\nexport function corpusInterRaterAgreementFromJudgeScores(\n itemsScores: Array<{ itemId: string; scores: DimensionJudgeScore[] }>,\n opts: CorpusAgreementOptions = {},\n): CorpusAgreementReport {\n const records: CorpusScoreRecord[] = []\n const seen = new Set<string>()\n for (const { itemId, scores } of itemsScores) {\n if (seen.has(itemId)) {\n throw new ValidationError(\n `corpusInterRaterAgreementFromJudgeScores: duplicate itemId '${itemId}'`,\n )\n }\n seen.add(itemId)\n for (const s of scores) {\n records.push({\n itemId,\n judgeName: s.judgeName,\n dimension: s.dimension,\n score: s.score,\n })\n }\n }\n return corpusInterRaterAgreement(records, opts)\n}\n\n/** Agreement between two raters who each assigned one nominal label per item. */\nexport interface NominalAgreement {\n /** Items both raters labeled. */\n n: number\n /** Share of items with identical labels. */\n agreement: number\n /**\n * Cohen's κ = (pₒ − pₑ) / (1 − pₑ), with pₑ from each rater's own label\n * marginals. NaN when n = 0, or when both raters used one identical label\n * for every item (pₑ = 1): chance and observed agreement are then the same\n * thing, and no number distinguishes skill from luck.\n */\n kappa: number\n /** Seeded percentile bootstrap of κ over clusters; NaN bounds when no resample has a finite κ. */\n kappaCi: BootstrapInterval\n}\n\nexport interface NominalAgreementOptions {\n /** Cluster id per pair; pairs sharing one are resampled together. Default: every pair alone. */\n clusters?: readonly string[]\n /** Default 2000. */\n resamples?: number\n seed?: number\n}\n\n/**\n * Unweighted Cohen's κ for paired nominal labels. Labels carry no order, so a\n * near miss counts as a miss; use `calibrateJudge` for ordinal scores.\n */\nexport function nominalAgreement(\n pairs: ReadonlyArray<readonly [string, string]>,\n options: NominalAgreementOptions = {},\n): NominalAgreement {\n if (options.clusters && options.clusters.length !== pairs.length) {\n throw new ValidationError('nominalAgreement: one cluster per pair')\n }\n const kappaCi = clusterBootstrap(\n pairs,\n (_, index) => options.clusters?.[index] ?? String(index),\n (sample) => cohenKappa(sample).kappa,\n options,\n )\n return { n: pairs.length, ...cohenKappa(pairs), kappaCi }\n}\n\nfunction cohenKappa(pairs: ReadonlyArray<readonly [string, string]>) {\n const n = pairs.length\n if (n === 0) return { agreement: Number.NaN, kappa: Number.NaN }\n const left = new Map<string, number>()\n const right = new Map<string, number>()\n let same = 0\n for (const [a, b] of pairs) {\n if (typeof a !== 'string' || typeof b !== 'string') {\n throw new ValidationError('nominalAgreement: every label must be a string')\n }\n if (a === b) same += 1\n left.set(a, (left.get(a) ?? 0) + 1)\n right.set(b, (right.get(b) ?? 0) + 1)\n }\n const observed = same / n\n let expected = 0\n for (const [label, count] of left) expected += (count / n) * ((right.get(label) ?? 0) / n)\n const kappa = expected === 1 ? Number.NaN : (observed - expected) / (1 - expected)\n return { agreement: observed, kappa }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;AA0BA,SAAgB,sBAAsB,aAA8C;CAClF,IAAI,YAAY,SAAS,GAAG,OAAO;CAGnC,MAAM,+BAAe,IAAI,IAAwB;CACjD,KAAK,IAAI,aAAa,GAAG,aAAa,YAAY,QAAQ,cACxD,KAAK,MAAM,KAAK,YAAY,aAAc;EACxC,IAAI,UAAU,aAAa,IAAI,EAAE,SAAS;EAC1C,IAAI,YAAY,KAAA,GAAW;GACzB,UAAU,MAAM,KAAK,EAAE,QAAQ,YAAY,OAAO,SAAS,CAAC,CAAa;GACzE,aAAa,IAAI,EAAE,WAAW,OAAO;EACvC;EACA,QAAQ,WAAW,CAAE,KAAK,EAAE,KAAK;CACnC;CAGF,MAAM,YAAsB,CAAC;CAC7B,MAAM,YAAsB,CAAC;CAE7B,KAAK,MAAM,CAAC,WAAW,YAAY,cAAc;EAC/C,MAAM,UAAU,QAAQ,QAAQ,WAAW,OAAO,SAAS,CAAC;EAC5D,IAAI,QAAQ,SAAS,GAAG;EACxB,MAAM,YAAY,QAAQ,EAAE,CAAE;EAC9B,IAAI,QAAQ,MAAM,WAAW,OAAO,WAAW,SAAS,GACtD,MAAM,IAAI,gBACR,qCAAqC,UAAU,yBAC1C,QAAQ,KAAK,WAAW,OAAO,MAAM,CAAC,CAAC,KAAK,GAAG,EAAE,kCACxD;EAEF,KAAK,IAAI,OAAO,GAAG,OAAO,WAAW,QAAQ;GAC3C,MAAM,UAAU,QAAQ,KAAK,WAAW,OAAO,KAAM;GACrD,KAAK,MAAM,KAAK,SAAS,UAAU,KAAK,CAAC;GACzC,KAAK,IAAI,IAAI,GAAG,IAAI,QAAQ,QAAQ,KAClC,KAAK,IAAI,IAAI,IAAI,GAAG,IAAI,QAAQ,QAAQ,KACtC,UAAU,MAAM,QAAQ,KAAM,QAAQ,OAAQ,CAAC;EAGrD;CACF;CAEA,IAAI,UAAU,WAAW,KAAK,UAAU,SAAS,GAAG,OAAO;CAE3D,MAAM,uBAAuB,UAAU,QAAQ,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,UAAU;CAG9E,IAAI,uBAAuB;CAC3B,IAAI,gBAAgB;CACpB,KAAK,IAAI,IAAI,GAAG,IAAI,UAAU,QAAQ,KACpC,KAAK,IAAI,IAAI,IAAI,GAAG,IAAI,UAAU,QAAQ,KAAK;EAC7C,yBAAyB,UAAU,KAAM,UAAU,OAAQ;EAC3D;CACF;CAEF,uBAAuB,gBAAgB,IAAI,uBAAuB,gBAAgB;CAElF,IAAI,yBAAyB,GAAG,OAAO;CACvC,OAAO,IAAI,uBAAuB;AACpC;;;;;;;;;;;;;AAqDA,SAAgB,0BACd,SACA,OAA+B,CAAC,GACT;CACvB,IAAI,QAAQ,WAAW,GACrB,MAAM,IAAI,gBAAgB,sDAAsD;CAGlF,MAAM,6BAAa,IAAI,IAAY;CACnC,MAAM,2BAAW,IAAI,IAAY;CAEjC,MAAM,uBAAO,IAAI,IAA8C;CAE/D,KAAK,MAAM,KAAK,SAAS;EACvB,IAAI,CAAC,OAAO,SAAS,EAAE,KAAK,GAC1B,MAAM,IAAI,gBACR,yDAAyD,EAAE,OAAO,UAAU,EAAE,UAAU,QAAQ,EAAE,UAAU,EAC9G;EAEF,WAAW,IAAI,EAAE,SAAS;EAC1B,SAAS,IAAI,EAAE,SAAS;EACxB,MAAM,UAAU,KAAK,IAAI,EAAE,SAAS,qBAAK,IAAI,IAAiC;EAC9E,MAAM,SAAS,QAAQ,IAAI,EAAE,SAAS,qBAAK,IAAI,IAAoB;EACnE,IAAI,OAAO,IAAI,EAAE,MAAM,GACrB,MAAM,IAAI,gBACR,yDAAyD,EAAE,OAAO,UAAU,EAAE,UAAU,QAAQ,EAAE,UAAU,EAC9G;EAEF,OAAO,IAAI,EAAE,QAAQ,EAAE,KAAK;EAC5B,QAAQ,IAAI,EAAE,WAAW,MAAM;EAC/B,KAAK,IAAI,EAAE,WAAW,OAAO;CAC/B;CAEA,MAAM,aAAa,KAAK,cAAc,CAAC,GAAG,QAAQ,CAAC,CAAC,KAAK;CACzD,KAAK,MAAM,KAAK,YACd,IAAI,CAAC,SAAS,IAAI,CAAC,GACjB,MAAM,IAAI,gBACR,yCAAyC,EAAE,wCAC7C;CAGJ,MAAM,eAAe,KAAK,SAAS,CAAC,GAAG,KAAK,MAAM,IAAI,CAAC,GAAG,UAAU,CAAC,CAAC,KAAK;CAC3E,KAAK,MAAM,KAAK,cACd,IAAI,CAAC,WAAW,IAAI,CAAC,GACnB,MAAM,IAAI,gBACR,qCAAqC,EAAE,wCACzC;CAGJ,IAAI,aAAa,SAAS,GACxB,MAAM,IAAI,gBACR,kDAAkD,aAAa,QACjE;CAGF,MAAM,eAA8C,CAAC;CACrD,MAAM,OAAiB,CAAC;CACxB,MAAM,SAAmB,CAAC;CAE1B,KAAK,MAAM,OAAO,YAAY;EAC5B,MAAM,UAAU,KAAK,IAAI,GAAG;EAE5B,MAAM,kBAAkB,OAAO,YAC7B,aAAa,KAAK,MAAM,CAAC,GAAG,QAAQ,IAAI,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC,CACxD;EACA,MAAM,cAAc,aAAa,QAAQ,MAAM,gBAAgB,OAAO,CAAC;EACvE,IAAI,YAAY,SAAS,GACvB,MAAM,IAAI,gBACR,yCAAyC,IAAI,gCAAgC,YAAY,KAAK,IAAI,EAAE,YAAY,KAAK,UAAU,eAAe,EAAE,EAClJ;EAGF,MAAM,cAAc,aACjB,KAAK,MAAM,CAAC,GAAG,QAAQ,IAAI,CAAC,CAAC,CAAE,KAAK,CAAC,CAAC,CAAC,CACvC,QAAQ,QAAQ,QAAQ,OAAO,QAAQ,MAAM,IAAI,SAAS,CAAC,CAAC,CAAC,CAAC,CAC9D,KAAK;EACR,IAAI,YAAY,SAAS,GACvB,MAAM,IAAI,gBACR,yCAAyC,IAAI,QAAQ,YAAY,OAAO,wBAAwB,aAAa,OAAO,kBACtH;EAGF,MAAM,SAAqB,YAAY,KAAK,WAC1C,aAAa,KAAK,MAAM,QAAQ,IAAI,CAAC,CAAC,CAAE,IAAI,MAAM,CAAE,CACtD;EACA,MAAM,YAAY,oBAAoB,QAAQ,IAAI;EAClD,aAAa,KAAK;GAChB,GAAG;GACH,WAAW;GACX,SAAS;GACT,UAAU,CAAC,GAAG,YAAY;EAC5B,CAAC;EACD,IAAI,OAAO,SAAS,UAAU,GAAG,GAAG,KAAK,KAAK,UAAU,GAAG;EAC3D,IAAI,OAAO,SAAS,UAAU,aAAa,GAAG,OAAO,KAAK,UAAU,aAAa;CACnF;CAEA,MAAM,QAAQ,OACZ,GAAG,WAAW,IAAI,MAAa,GAAG,QAAQ,GAAG,MAAM,IAAI,GAAG,CAAC,IAAI,GAAG;CACpE,OAAO;EACL;EACA,YAAY,KAAK,IAAI;EACrB,sBAAsB,KAAK,MAAM;EACjC,YAAY;EACZ,UAAU;CACZ;AACF;;;;;;;;;AAUA,SAAgB,yCACd,aACA,OAA+B,CAAC,GACT;CACvB,MAAM,UAA+B,CAAC;CACtC,MAAM,uBAAO,IAAI,IAAY;CAC7B,KAAK,MAAM,EAAE,QAAQ,YAAY,aAAa;EAC5C,IAAI,KAAK,IAAI,MAAM,GACjB,MAAM,IAAI,gBACR,+DAA+D,OAAO,EACxE;EAEF,KAAK,IAAI,MAAM;EACf,KAAK,MAAM,KAAK,QACd,QAAQ,KAAK;GACX;GACA,WAAW,EAAE;GACb,WAAW,EAAE;GACb,OAAO,EAAE;EACX,CAAC;CAEL;CACA,OAAO,0BAA0B,SAAS,IAAI;AAChD;;;;;AA+BA,SAAgB,iBACd,OACA,UAAmC,CAAC,GAClB;CAClB,IAAI,QAAQ,YAAY,QAAQ,SAAS,WAAW,MAAM,QACxD,MAAM,IAAI,gBAAgB,wCAAwC;CAEpE,MAAM,UAAU,iBACd,QACC,GAAG,UAAU,QAAQ,WAAW,UAAU,OAAO,KAAK,IACtD,WAAW,WAAW,MAAM,CAAC,CAAC,OAC/B,OACF;CACA,OAAO;EAAE,GAAG,MAAM;EAAQ,GAAG,WAAW,KAAK;EAAG;CAAQ;AAC1D;AAEA,SAAS,WAAW,OAAiD;CACnE,MAAM,IAAI,MAAM;CAChB,IAAI,MAAM,GAAG,OAAO;EAAE,WAAW;EAAY,OAAO;CAAW;CAC/D,MAAM,uBAAO,IAAI,IAAoB;CACrC,MAAM,wBAAQ,IAAI,IAAoB;CACtC,IAAI,OAAO;CACX,KAAK,MAAM,CAAC,GAAG,MAAM,OAAO;EAC1B,IAAI,OAAO,MAAM,YAAY,OAAO,MAAM,UACxC,MAAM,IAAI,gBAAgB,gDAAgD;EAE5E,IAAI,MAAM,GAAG,QAAQ;EACrB,KAAK,IAAI,IAAI,KAAK,IAAI,CAAC,KAAK,KAAK,CAAC;EAClC,MAAM,IAAI,IAAI,MAAM,IAAI,CAAC,KAAK,KAAK,CAAC;CACtC;CACA,MAAM,WAAW,OAAO;CACxB,IAAI,WAAW;CACf,KAAK,MAAM,CAAC,OAAO,UAAU,MAAM,YAAa,QAAQ,MAAO,MAAM,IAAI,KAAK,KAAK,KAAK;CAExF,OAAO;EAAE,WAAW;EAAU,OADhB,aAAa,IAAI,OAAc,WAAW,aAAa,IAAI;CACrC;AACtC"}
|
package/dist/analyst/index.d.ts
CHANGED
|
@@ -1,19 +1,20 @@
|
|
|
1
1
|
import { m as CustomTokenPricing, o as CostLedgerHandle } from "../cost-ledger-Czro9wmo.js";
|
|
2
|
-
import { A as
|
|
2
|
+
import { A as FindingParseOutcome, B as FindingConditions, C as CalibrationInterval, D as FindingLayer, E as FindingEvidenceUnit, F as independentUnitCount, H as TREATMENT_LAYERS, I as lineageRoots, L as parseAnalystFinding, M as FindingScope, N as FindingScopeReading, O as FindingMeasure, P as ProposalFindingSchema, R as readFindingScope, S as makeProposalFinding, T as FindingCalibration, U as TreatmentLayer, V as FindingTreatment, X as TraceAnalysisStore, _ as ExecutionProbeRequest, a as AnalystFindingSchemaVersion, b as computeFindingId, c as AnalystRunEvent, d as AnalystRunSummary, f as AnalystSeverity, g as ExecutionProbeOutcome, h as ExecutionProbe, i as AnalystFinding, j as FindingReaderLabel, k as FindingMethod, l as AnalystRunInputs, m as EvidenceRef, n as AnalystContext, o as AnalystInputKind, p as AnalystUsageReceipt, r as AnalystCost, s as AnalystRequirements, st as TraceAnalystSpan, t as Analyst, u as AnalystRunResult, v as ProposalFinding, w as FINDING_LAYERS, x as makeFinding, y as ProposalFindingOrigin, z as BootstrapInterval } from "../types-B9s9Xo5_.js";
|
|
3
3
|
import { A as createChatClient, C as ChatResponse, D as MockTransportOpts, E as CustomTransportOpts, O as OpenAiCompatibleTransportOpts, S as ChatRequest, T as CreateChatClientOpts, b as ChatCallOpts, k as SandboxSdkTransportOpts, w as ChatTransport, x as ChatClient } from "../types-0KglnQ9o.js";
|
|
4
|
-
import {
|
|
5
|
-
import {
|
|
6
|
-
import { a as
|
|
4
|
+
import { cn as NominalAgreementOptions } from "../index-C3t17kHu.js";
|
|
5
|
+
import { A as traceStoreEvidenceResolver, C as AnalystEvidenceResolutionError, D as AnalystLatencyDistribution, E as AnalystIssueExpectation, G as ExternalOptimizerRunnerCommand, H as ExternalOptimizerModelExecutionObservation, O as RunAnalystBenchmarkOptions, S as AnalystEvidenceResolution, T as AnalystFindingScore, _ as AnalystBenchmarkProvenance, b as AnalystBenchmarkSummary, d as AnalystBenchmarkDatasetRef, f as AnalystBenchmarkDescriptor, g as AnalystBenchmarkOutput, h as AnalystBenchmarkObservation, k as runAnalystBenchmark, m as AnalystBenchmarkLabelState, p as AnalystBenchmarkError, t as AgentProfile, u as AnalystBenchmarkCase, v as AnalystBenchmarkResult, w as AnalystEvidenceResolver, x as AnalystEvidenceExpectation, y as AnalystBenchmarkRunner, z as ExternalOptimizerModelCall } from "../agent-profile-CPv4tENR.js";
|
|
6
|
+
import { C as BehavioralWindowPolicy, S as BehavioralTraceWindow, _ as BehavioralTraceEventMap, a as createTraceAnalyst, b as BehavioralTraceQuestionsOptions, c as BehavioralAnalystOptions, d as pinBehavioralFailureClass, f as BEHAVIORAL_TRACE_QUESTIONS, g as BehavioralTraceEvent, h as BehavioralTraceAnswer, i as TraceAnalystDefinition, l as assertFailureClasses, m as BehavioralQuestionCalibration, n as buildDefaultAnalystRegistry, o as runTraceAnalyst, p as BEHAVIORAL_TRACE_QUESTIONS_VERSION, r as CreateTraceAnalystOptions, s as BEHAVIORAL_FAILURE_CLASS, t as DefaultAnalystRegistryOptions, u as behavioralAnalyst, v as BehavioralTraceLayer, w as behavioralTraceQuestionDigest, x as BehavioralTraceSupport, y as BehavioralTraceQuestionId } from "../default-registry-BNv7eB1C.js";
|
|
7
|
+
import { a as ExactAnalystRunPolicySnapshot, c as ExactAnalystSnapshot, d as ExactExecutionComponentSnapshot, i as ExactAnalystRunEvent, l as ExactCapableAnalyst, n as ExactAnalystExecutionPlanSnapshot, o as ExactAnalystRunResult, r as ExactAnalystRunCompletion, s as ExactAnalystRunSummary, t as ExactAnalystBudgetSnapshot, u as ExactExecutionComponentIdentity } from "../exact-types-CaLLU1Pz.js";
|
|
7
8
|
import { g as SystemOneState } from "../systemone-protocol-C4t7JH4Y.js";
|
|
8
|
-
import {
|
|
9
|
-
import { a as RAW_FINDING_SCHEMA_PROMPT, c as RawAnalystFindingSchema, i as TraceAnalystLimits, l as evidenceRefsFromRawFinding, n as TraceAnalysisEngineRequest, o as RawAnalystEvidence, r as TraceAnalysisEngineResult, s as RawAnalystFinding, t as TraceAnalysisEngine, u as parseRawFinding } from "../engine-
|
|
10
|
-
import { n as buildTraceToolsForGroup, t as TraceToolGroupName } from "../tool-groups-
|
|
11
|
-
import { a as BudgetPolicy, c as RegistryRunOpts, i as AnalystRegistryOptions, n as ExactAnalystRunExecutionError, o as ExactAnalystBudgetPolicy, r as AnalystHooks, s as ExactRegistryRunOpts, t as AnalystRegistry } from "../registry-
|
|
12
|
-
import { T as SemanticConceptJudgeOptions, _ as DspyRlmTraceEngineOptions, a as FAILURE_MODE_KIND_SPEC, b as createChatTraceEngine, c as emitControlIntegrityFindings, d as FindingSubject, f as FindingSubjectKind, g as renderFindingSubject, h as parseFindingSubject, i as IMPROVEMENT_KIND_SPEC, l as FINDING_SUBJECT_KINDS, m as findingSubjectGrammarPromptFor, n as KNOWLEDGE_POISONING_KIND_SPEC, o as CONTROL_INTEGRITY_ANALYST, p as KIND_EXPECTED_SUBJECTS, r as KNOWLEDGE_GAP_KIND_SPEC, s as ControlIntegrityAnalyst, t as DEFAULT_TRACE_ANALYST_KINDS, u as FINDING_SUBJECT_SYNTAX, v as createDspyRlmTraceEngine, y as ChatTraceEngineOptions } from "../index-
|
|
13
|
-
import { C as runPrimeExchange, D as ReplyRowDecoded, E as ReplyEnvelope, O as ReplyRowRejection, S as projectPrimeTrajectory, T as ReplyContract, _ as buildPrimePrompt, a as PrimeProjectionOutcome, b as normalizePrimeUsage, c as PrimeProtocolIdentity, d as PrimeRepairPromptSpec, f as PrimeRepairState, g as analystUsageReceiptFromPrimeUsage, h as PrimeTurnRecord, i as PrimeProjectionDelivery, l as PrimeRawUsage, m as PrimeRowDecoded, n as PrimeExchangeOutcome, o as PrimeProjectionSource, p as PrimeReplyContract, r as PrimeFailure, s as PrimePromptSpec, t as PrimeExchangeOptions, u as PrimeRejectedRow, v as buildPrimeRepairPrompt, w as DecodedReply, x as primeProtocolSha256, y as mergePrimeRawUsage } from "../prime-protocol-
|
|
9
|
+
import { B as DiffPolicy, F as FindingsTimeStep, H as FindingsStore, I as LedgerFinding, L as diffFindingsOverTime, M as FindingsLedgerOutcome, N as FindingsLedgerRead, P as FindingsQuery, R as openFindingsLedger, U as PersistedFinding, V as FindingsDiff, W as diffFindings, j as FindingsLedger, r as SystemOneEvaluatorOptions, z as queryFindings } from "../systemone-CBvL1gx-.js";
|
|
10
|
+
import { a as RAW_FINDING_SCHEMA_PROMPT, c as RawAnalystFindingSchema, i as TraceAnalystLimits, l as evidenceRefsFromRawFinding, n as TraceAnalysisEngineRequest, o as RawAnalystEvidence, r as TraceAnalysisEngineResult, s as RawAnalystFinding, t as TraceAnalysisEngine, u as parseRawFinding } from "../engine-rfV8Ff8V.js";
|
|
11
|
+
import { n as buildTraceToolsForGroup, t as TraceToolGroupName } from "../tool-groups-BQsjZD9F.js";
|
|
12
|
+
import { a as BudgetPolicy, c as RegistryRunOpts, i as AnalystRegistryOptions, n as ExactAnalystRunExecutionError, o as ExactAnalystBudgetPolicy, r as AnalystHooks, s as ExactRegistryRunOpts, t as AnalystRegistry } from "../registry-Ck7w-rr5.js";
|
|
13
|
+
import { T as SemanticConceptJudgeOptions, _ as DspyRlmTraceEngineOptions, a as FAILURE_MODE_KIND_SPEC, b as createChatTraceEngine, c as emitControlIntegrityFindings, d as FindingSubject, f as FindingSubjectKind, g as renderFindingSubject, h as parseFindingSubject, i as IMPROVEMENT_KIND_SPEC, l as FINDING_SUBJECT_KINDS, m as findingSubjectGrammarPromptFor, n as KNOWLEDGE_POISONING_KIND_SPEC, o as CONTROL_INTEGRITY_ANALYST, p as KIND_EXPECTED_SUBJECTS, r as KNOWLEDGE_GAP_KIND_SPEC, s as ControlIntegrityAnalyst, t as DEFAULT_TRACE_ANALYST_KINDS, u as FINDING_SUBJECT_SYNTAX, v as createDspyRlmTraceEngine, y as ChatTraceEngineOptions } from "../index-Dt6E4BHX.js";
|
|
14
|
+
import { C as runPrimeExchange, D as ReplyRowDecoded, E as ReplyEnvelope, O as ReplyRowRejection, S as projectPrimeTrajectory, T as ReplyContract, _ as buildPrimePrompt, a as PrimeProjectionOutcome, b as normalizePrimeUsage, c as PrimeProtocolIdentity, d as PrimeRepairPromptSpec, f as PrimeRepairState, g as analystUsageReceiptFromPrimeUsage, h as PrimeTurnRecord, i as PrimeProjectionDelivery, l as PrimeRawUsage, m as PrimeRowDecoded, n as PrimeExchangeOutcome, o as PrimeProjectionSource, p as PrimeReplyContract, r as PrimeFailure, s as PrimePromptSpec, t as PrimeExchangeOptions, u as PrimeRejectedRow, v as buildPrimeRepairPrompt, w as DecodedReply, x as primeProtocolSha256, y as mergePrimeRawUsage } from "../prime-protocol-DKRk4l4U.js";
|
|
14
15
|
import { i as nodeHttpPrimeBridgeTransport, n as PrimeBridgeTransportRequest, r as PrimeBridgeTransportResult, t as PrimeBridgeTransport } from "../prime-bridge-transport-Cfqir_Im.js";
|
|
15
16
|
import { b as SupervisorRunSources } from "../types-DNB9hIXa.js";
|
|
16
|
-
import { a as askTraceQuestions, i as VerifiedTraceFinding, n as TraceQuestionOutcome, r as TraceQuestionSpec, t as AskTraceQuestionsOptions } from "../trace-questions-
|
|
17
|
+
import { a as askTraceQuestions, i as VerifiedTraceFinding, n as TraceQuestionOutcome, r as TraceQuestionSpec, t as AskTraceQuestionsOptions } from "../trace-questions-CpTrtEcm.js";
|
|
17
18
|
import { z } from "zod";
|
|
18
19
|
interface SemanticConceptJudgeAdapterOpts {
|
|
19
20
|
id?: string;
|
|
@@ -59,7 +60,16 @@ interface BehavioralTraceGold {
|
|
|
59
60
|
label: BehavioralTraceAnswer | null;
|
|
60
61
|
source: 'human' | 'agentrx';
|
|
61
62
|
}
|
|
62
|
-
|
|
63
|
+
/** One-vs-rest reading of a class: precision and recall from hard answers; the AUC of its probability
|
|
64
|
+
* (null without a scored member of each class) with a seeded cluster bootstrap; the AUC bar line. */
|
|
65
|
+
interface ClassCalibration {
|
|
66
|
+
precision: number | null;
|
|
67
|
+
recall: number | null;
|
|
68
|
+
auc: number | null;
|
|
69
|
+
ci: CalibrationInterval | null;
|
|
70
|
+
eligibleAuc: boolean;
|
|
71
|
+
}
|
|
72
|
+
interface BehavioralTraceCalibrationRow extends ClassCalibration {
|
|
63
73
|
questionId: BehavioralTraceQuestionId;
|
|
64
74
|
layer: BehavioralTraceLayer;
|
|
65
75
|
method: 'computed' | 'judged';
|
|
@@ -74,6 +84,7 @@ interface BehavioralTraceCalibrationRow {
|
|
|
74
84
|
};
|
|
75
85
|
agreement: number | null;
|
|
76
86
|
kappa: number | null;
|
|
87
|
+
kappaCi: BootstrapInterval | null;
|
|
77
88
|
confusion: Record<'yes' | 'no', Record<'yes' | 'no' | 'unsure' | 'missing', number>>;
|
|
78
89
|
/** Wrong/unsure/missing semantic answers retain the selected event and complete scoped map.
|
|
79
90
|
* Cross-map nominal agreement is separate from exact-definition decision reliability. */
|
|
@@ -87,6 +98,34 @@ interface BehavioralTraceCalibrationRow {
|
|
|
87
98
|
/** Finite measured evidence, even if it misses the activation bar. */
|
|
88
99
|
calibration: BehavioralQuestionCalibration | null;
|
|
89
100
|
}
|
|
101
|
+
interface ChoiceGold {
|
|
102
|
+
id: string;
|
|
103
|
+
label: string | null;
|
|
104
|
+
}
|
|
105
|
+
interface ChoicePrediction {
|
|
106
|
+
id: string;
|
|
107
|
+
cluster?: string;
|
|
108
|
+
choice: string | null;
|
|
109
|
+
probabilities: Record<string, number> | null;
|
|
110
|
+
}
|
|
111
|
+
/** `gold_n` is the gold items labeled with this option, `predicted` the labeled items that chose it. */
|
|
112
|
+
interface ChoiceOptionCalibration extends ClassCalibration {
|
|
113
|
+
option: string;
|
|
114
|
+
gold_n: number;
|
|
115
|
+
predicted: number;
|
|
116
|
+
}
|
|
117
|
+
interface ChoiceCalibration {
|
|
118
|
+
options: string[];
|
|
119
|
+
gold_n: number;
|
|
120
|
+
missingLabels: number;
|
|
121
|
+
missingPredictions: number;
|
|
122
|
+
abstentions: number;
|
|
123
|
+
agreement: number | null;
|
|
124
|
+
/** Multi-class Cohen's κ over the hard choices, with its cluster bootstrap. */
|
|
125
|
+
kappa: number | null;
|
|
126
|
+
kappaCi: BootstrapInterval | null;
|
|
127
|
+
rows: ChoiceOptionCalibration[];
|
|
128
|
+
}
|
|
90
129
|
/** Same nominalAgreement owner as the existing AgentRx/failure-mode calibration path. */
|
|
91
130
|
export declare function calibrateBehavioralTraceQuestions(input: {
|
|
92
131
|
layer: BehavioralTraceLayer;
|
|
@@ -97,7 +136,14 @@ export declare function calibrateBehavioralTraceQuestions(input: {
|
|
|
97
136
|
gold: readonly BehavioralTraceGold[];
|
|
98
137
|
findings: readonly AnalystFinding[];
|
|
99
138
|
evidenceRefs: EvidenceRef[];
|
|
139
|
+
resamples?: number;
|
|
140
|
+
seed?: number;
|
|
100
141
|
}): BehavioralTraceCalibrationRow[];
|
|
142
|
+
/** A Choice question over any option set: a one-vs-rest row per option plus the multi-class κ. A null
|
|
143
|
+
* or out-of-set gold label is an exclusion; an out-of-set choice an abstention, a null choice missing. */
|
|
144
|
+
export declare function calibrateChoiceQuestion(gold: readonly ChoiceGold[], findings: readonly ChoicePrediction[], options?: NominalAgreementOptions & {
|
|
145
|
+
options?: readonly string[];
|
|
146
|
+
}): ChoiceCalibration;
|
|
101
147
|
type ExternalId = string | number;
|
|
102
148
|
interface AgentRxFailure {
|
|
103
149
|
failure_id: ExternalId;
|
|
@@ -1633,5 +1679,5 @@ export declare function classifyTeamShape(input: TeamShapeInput): TeamShapeOutco
|
|
|
1633
1679
|
* timestamps fails the analyst instead of emitting a shape it did not measure.
|
|
1634
1680
|
*/
|
|
1635
1681
|
export declare function teamShapeAnalyst(): Analyst<TeamShapeInput>;
|
|
1636
|
-
export { type AgentRxBenchmarkCaseOptions, type AgentRxCalibrationRunnerSummary, type AgentRxCalibrationSummary, type AgentRxFailure, type AgentRxPrediction, type AgentRxPredictionReport, type AgentRxRow, type Analyst, type AnalystBenchmarkArtifact, type AnalystBenchmarkCase, type AnalystBenchmarkCommandConfig, type AnalystBenchmarkCommandDependencies, type AnalystBenchmarkDatasetRef, type AnalystBenchmarkDescriptor, type AnalystBenchmarkError, type AnalystBenchmarkLabelState, type AnalystBenchmarkLocalRunReceipt, type AnalystBenchmarkObservation, type AnalystBenchmarkOutput, type AnalystBenchmarkProgressRow, type AnalystBenchmarkProvenance, type AnalystBenchmarkResult, type AnalystBenchmarkRunIdentity, type AnalystBenchmarkRunManifest, type AnalystBenchmarkRunner, type AnalystBenchmarkSummary, type AnalystBudgetDeclaration, type AnalystComparisonMetric, type AnalystContext, type AnalystCost, type AnalystDefinition, type AnalystDefinitionAsymmetry, type AnalystDefinitionAsymmetryReport, type AnalystEvidenceBinding, type AnalystEvidenceExpectation, type AnalystEvidenceResolution, type AnalystEvidenceResolutionError, type AnalystEvidenceResolver, type AnalystFinding, type AnalystFindingSchemaVersion, type AnalystFindingScore, type AnalystHooks, type AnalystInputKind, type AnalystInstructionsOverride, type AnalystIssueExpectation, type AnalystLatencyDistribution, type AnalystMetricComparison, type AnalystProfileFragment, AnalystRegistry, type AnalystRegistryOptions, type AnalystRepairDeclaration, type AnalystRequirements, type AnalystRowExpansion, type AnalystRunEvent, type AnalystRunInputs, type AnalystRunResult, type AnalystRunSummary, type AnalystRunnerComparison, type AnalystSeverity, type AnalystTransportBinding, type AnalystUsageReceipt, type AskTraceQuestionsOptions, BEHAVIORAL_FAILURE_CLASS, BEHAVIORAL_TRACE_QUESTIONS, BEHAVIORAL_TRACE_QUESTIONS_VERSION, type BehavioralAnalystOptions, type BehavioralQuestionCalibration, type BehavioralTraceAnswer, type BehavioralTraceCalibrationRow, type BehavioralTraceEvent, type BehavioralTraceEventMap, type BehavioralTraceGold, type BehavioralTraceLayer, type BehavioralTraceQuestionId, type BehavioralTraceQuestionsOptions, type BehavioralTraceSupport, type BehavioralTraceWindow, type BehavioralWindowPolicy, type BudgetPolicy, CONTROL_INTEGRITY_ANALYST, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTraceEngineOptions, type ChatTransport, type ChunkedEvidenceBinding, type CodeTraceBenchCaseOptions, type CodeTraceBenchLabelOptions, type CodeTraceBenchLabelSet, type CodeTraceBenchRow, type CodeTraceBlockDiagnostics, type CodeTraceCalibrationRunnerSummary, type CodeTraceCalibrationSummary, type CodeTraceFailureBlock, type CodeTraceStageAnnotation, type CodeTracerLabelGroup, type CodeTracerPredictionAdapterOptions, type CodeTracerPredictions, type CodeTracerStepLabel, ControlIntegrityAnalyst, type CreateChatClientOpts, type CreateTraceAnalystOptions, type CustomTransportOpts, DEFAULT_TRACE_ANALYST_KINDS, type DecodedReply, type DefaultAnalystRegistryOptions, type DefineCustomAnalystOptions, type DefineExactCustomAnalystOptions, type DiffPolicy, type DomainTaxonomy, type DspyRlmTraceEngineOptions, type EvidenceProjection, type EvidenceRef, type ExactAnalystBudgetPolicy, type ExactAnalystBudgetSnapshot, type ExactAnalystExecutionPlanSnapshot, type ExactAnalystRunCompletion, type ExactAnalystRunEvent, ExactAnalystRunExecutionError, type ExactAnalystRunPolicySnapshot, type ExactAnalystRunResult, type ExactAnalystRunSummary, type ExactAnalystSnapshot, type ExactCapableAnalyst, type ExactExecutionComponentIdentity, type ExactExecutionComponentSnapshot, type ExactRegistryRunOpts, type ExecutionProbe, type ExecutionProbeOutcome, type ExecutionProbeRequest, type ExpandRowsArgs, FAILURE_MODE_KIND_SPEC, FINDING_LAYERS, FINDING_SUBJECT_KINDS, FINDING_SUBJECT_SYNTAX, type FailureModeGold, type FailureModeGoldSources, type FailureModeLabel, type FailureModeLabelItem, type FailureModesAnalystOptions, type FailureModesCalibration, type FailureModesOutcome, type FindingCalibration, type FindingConditions, type FindingEvidenceUnit, type FindingLayer, type FindingMeasure, type FindingMethod, type FindingParseOutcome, type FindingReaderLabel, type FindingScope, type FindingScopeReading, type FindingSubject, type FindingSubjectKind, type FindingTreatment, type FindingsDiff, type FindingsLedger, type FindingsLedgerOutcome, type FindingsLedgerRead, type FindingsQuery, FindingsStore, type FindingsTimeStep, IMPROVEMENT_KIND_SPEC, type InlineEvidenceBinding, KIND_EXPECTED_SUBJECTS, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, type LedgerFinding, type LoadedVerificationArtifacts, type MockTransportOpts, type NotObservable, type OpenAiCompatibleTransportOpts, type PersistedFinding, type PreparedPublicAnalystBenchmark, type PrimeBenchmarkRunnerOptions, type PrimeBridgeTransport, type PrimeBridgeTransportRequest, type PrimeBridgeTransportResult, type PrimeCodeTraceDefinitionArgs, type PrimeExchangeOptions, type PrimeExchangeOutcome, type PrimeFailure, type PrimeProjectionDelivery, type PrimeProjectionOutcome, type PrimeProjectionSource, type PrimePromptSpec, type PrimeProtocolIdentity, type PrimeRawUsage, type PrimeRejectedRow, type PrimeRepairPromptSpec, type PrimeRepairState, type PrimeReplyContract, type PrimeRowDecoded, type PrimeTurnRecord, type PromoteOptions, type PromotionHold, type PromotionResult, type ProposalFinding, type ProposalFindingOrigin, ProposalFindingSchema, type PublicAnalystBenchmarkDataset, type PublicAnalystBenchmarkModelConfig, type PublicAnalystBenchmarkModelOwner, type PublicAnalystBenchmarkModelSettings, type PublicBenchmarkDistributions, type PublicBenchmarkModelPrediction, type PublicBenchmarkSelectionReport, type PublicBenchmarkValueDistribution, type PublicDirectDefinitionArgs, type PublicRlmDefinitionArgs, RAW_FINDING_SCHEMA_PROMPT, type RawAnalystEvidence, type RawAnalystFinding, RawAnalystFindingSchema, type RegistryRunOpts, type ReplVariableConsensusPort, type ReplVariableEvidenceBinding, type ReplyContract, type ReplyEnvelope, type ReplyRowDecoded, type ReplyRowRejection, type RollupInput, type RollupOptions, type RollupResult, type RunAnalystBenchmarkOptions, type SandboxSdkTransportOpts, type SemanticConceptJudgeAdapterOpts, type SkillUsageRecord, type SkillUsageReport, type SkillUsageScanConfig, type StepLabelAdapterOptions, TREATMENT_LAYERS, type TaxonomyAttachment, type TaxonomyNode, type TaxonomyNodeLayer, type TaxonomyOutcome, type TeamIntegration, type TeamShape, type TeamShapeCheck, type TeamShapeClass, type TeamShapeEdge, type TeamShapeFeatures, type TeamShapeInput, type TeamShapeObserved, type TeamShapeOutcome, type TraceAnalysisEngine, type TraceAnalysisEngineRequest, type TraceAnalysisEngineResult, type TraceAnalystDefinition, type TraceAnalystLimits, type TraceQuestionOutcome, type TraceQuestionSpec, type TraceToolGroupName, type TreatmentLayer, type UpstreamPredictionAdapterOptions, type VerificationArtifactFile, type VerificationArtifactManifest, type VerificationArtifactRole, type VerificationAvailabilitySummary, type VerificationOutcome, type VerificationOutcomeSource, type VerificationOutcomeStatus, type VerificationResultFile, type VerifiedTraceFinding, analystUsageReceiptFromPrimeUsage, askTraceQuestions, assertFailureClasses, behavioralAnalyst, behavioralTraceQuestionDigest, buildDefaultAnalystRegistry, buildPrimePrompt, buildPrimeRepairPrompt, buildTraceToolsForGroup, computeFindingId, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createTraceAnalyst, diffFindings, diffFindingsOverTime, emitControlIntegrityFindings, evidenceRefsFromRawFinding, findingSubjectGrammarPromptFor, independentUnitCount, lineageRoots, makeFinding, makeProposalFinding, mergePrimeRawUsage, nodeHttpPrimeBridgeTransport, normalizePrimeUsage, openFindingsLedger, parseAnalystFinding, parseFindingSubject, parseRawFinding, pinBehavioralFailureClass, primeProtocolSha256, projectPrimeTrajectory, queryFindings, readFindingScope, renderFindingSubject, runAnalystBenchmark, runPrimeExchange, runTraceAnalyst, traceStoreEvidenceResolver };
|
|
1682
|
+
export { type AgentRxBenchmarkCaseOptions, type AgentRxCalibrationRunnerSummary, type AgentRxCalibrationSummary, type AgentRxFailure, type AgentRxPrediction, type AgentRxPredictionReport, type AgentRxRow, type Analyst, type AnalystBenchmarkArtifact, type AnalystBenchmarkCase, type AnalystBenchmarkCommandConfig, type AnalystBenchmarkCommandDependencies, type AnalystBenchmarkDatasetRef, type AnalystBenchmarkDescriptor, type AnalystBenchmarkError, type AnalystBenchmarkLabelState, type AnalystBenchmarkLocalRunReceipt, type AnalystBenchmarkObservation, type AnalystBenchmarkOutput, type AnalystBenchmarkProgressRow, type AnalystBenchmarkProvenance, type AnalystBenchmarkResult, type AnalystBenchmarkRunIdentity, type AnalystBenchmarkRunManifest, type AnalystBenchmarkRunner, type AnalystBenchmarkSummary, type AnalystBudgetDeclaration, type AnalystComparisonMetric, type AnalystContext, type AnalystCost, type AnalystDefinition, type AnalystDefinitionAsymmetry, type AnalystDefinitionAsymmetryReport, type AnalystEvidenceBinding, type AnalystEvidenceExpectation, type AnalystEvidenceResolution, type AnalystEvidenceResolutionError, type AnalystEvidenceResolver, type AnalystFinding, type AnalystFindingSchemaVersion, type AnalystFindingScore, type AnalystHooks, type AnalystInputKind, type AnalystInstructionsOverride, type AnalystIssueExpectation, type AnalystLatencyDistribution, type AnalystMetricComparison, type AnalystProfileFragment, AnalystRegistry, type AnalystRegistryOptions, type AnalystRepairDeclaration, type AnalystRequirements, type AnalystRowExpansion, type AnalystRunEvent, type AnalystRunInputs, type AnalystRunResult, type AnalystRunSummary, type AnalystRunnerComparison, type AnalystSeverity, type AnalystTransportBinding, type AnalystUsageReceipt, type AskTraceQuestionsOptions, BEHAVIORAL_FAILURE_CLASS, BEHAVIORAL_TRACE_QUESTIONS, BEHAVIORAL_TRACE_QUESTIONS_VERSION, type BehavioralAnalystOptions, type BehavioralQuestionCalibration, type BehavioralTraceAnswer, type BehavioralTraceCalibrationRow, type BehavioralTraceEvent, type BehavioralTraceEventMap, type BehavioralTraceGold, type BehavioralTraceLayer, type BehavioralTraceQuestionId, type BehavioralTraceQuestionsOptions, type BehavioralTraceSupport, type BehavioralTraceWindow, type BehavioralWindowPolicy, type BudgetPolicy, CONTROL_INTEGRITY_ANALYST, type CalibrationInterval, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTraceEngineOptions, type ChatTransport, type ChoiceCalibration, type ChoiceGold, type ChoiceOptionCalibration, type ChoicePrediction, type ChunkedEvidenceBinding, type ClassCalibration, type CodeTraceBenchCaseOptions, type CodeTraceBenchLabelOptions, type CodeTraceBenchLabelSet, type CodeTraceBenchRow, type CodeTraceBlockDiagnostics, type CodeTraceCalibrationRunnerSummary, type CodeTraceCalibrationSummary, type CodeTraceFailureBlock, type CodeTraceStageAnnotation, type CodeTracerLabelGroup, type CodeTracerPredictionAdapterOptions, type CodeTracerPredictions, type CodeTracerStepLabel, ControlIntegrityAnalyst, type CreateChatClientOpts, type CreateTraceAnalystOptions, type CustomTransportOpts, DEFAULT_TRACE_ANALYST_KINDS, type DecodedReply, type DefaultAnalystRegistryOptions, type DefineCustomAnalystOptions, type DefineExactCustomAnalystOptions, type DiffPolicy, type DomainTaxonomy, type DspyRlmTraceEngineOptions, type EvidenceProjection, type EvidenceRef, type ExactAnalystBudgetPolicy, type ExactAnalystBudgetSnapshot, type ExactAnalystExecutionPlanSnapshot, type ExactAnalystRunCompletion, type ExactAnalystRunEvent, ExactAnalystRunExecutionError, type ExactAnalystRunPolicySnapshot, type ExactAnalystRunResult, type ExactAnalystRunSummary, type ExactAnalystSnapshot, type ExactCapableAnalyst, type ExactExecutionComponentIdentity, type ExactExecutionComponentSnapshot, type ExactRegistryRunOpts, type ExecutionProbe, type ExecutionProbeOutcome, type ExecutionProbeRequest, type ExpandRowsArgs, FAILURE_MODE_KIND_SPEC, FINDING_LAYERS, FINDING_SUBJECT_KINDS, FINDING_SUBJECT_SYNTAX, type FailureModeGold, type FailureModeGoldSources, type FailureModeLabel, type FailureModeLabelItem, type FailureModesAnalystOptions, type FailureModesCalibration, type FailureModesOutcome, type FindingCalibration, type FindingConditions, type FindingEvidenceUnit, type FindingLayer, type FindingMeasure, type FindingMethod, type FindingParseOutcome, type FindingReaderLabel, type FindingScope, type FindingScopeReading, type FindingSubject, type FindingSubjectKind, type FindingTreatment, type FindingsDiff, type FindingsLedger, type FindingsLedgerOutcome, type FindingsLedgerRead, type FindingsQuery, FindingsStore, type FindingsTimeStep, IMPROVEMENT_KIND_SPEC, type InlineEvidenceBinding, KIND_EXPECTED_SUBJECTS, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, type LedgerFinding, type LoadedVerificationArtifacts, type MockTransportOpts, type NotObservable, type OpenAiCompatibleTransportOpts, type PersistedFinding, type PreparedPublicAnalystBenchmark, type PrimeBenchmarkRunnerOptions, type PrimeBridgeTransport, type PrimeBridgeTransportRequest, type PrimeBridgeTransportResult, type PrimeCodeTraceDefinitionArgs, type PrimeExchangeOptions, type PrimeExchangeOutcome, type PrimeFailure, type PrimeProjectionDelivery, type PrimeProjectionOutcome, type PrimeProjectionSource, type PrimePromptSpec, type PrimeProtocolIdentity, type PrimeRawUsage, type PrimeRejectedRow, type PrimeRepairPromptSpec, type PrimeRepairState, type PrimeReplyContract, type PrimeRowDecoded, type PrimeTurnRecord, type PromoteOptions, type PromotionHold, type PromotionResult, type ProposalFinding, type ProposalFindingOrigin, ProposalFindingSchema, type PublicAnalystBenchmarkDataset, type PublicAnalystBenchmarkModelConfig, type PublicAnalystBenchmarkModelOwner, type PublicAnalystBenchmarkModelSettings, type PublicBenchmarkDistributions, type PublicBenchmarkModelPrediction, type PublicBenchmarkSelectionReport, type PublicBenchmarkValueDistribution, type PublicDirectDefinitionArgs, type PublicRlmDefinitionArgs, RAW_FINDING_SCHEMA_PROMPT, type RawAnalystEvidence, type RawAnalystFinding, RawAnalystFindingSchema, type RegistryRunOpts, type ReplVariableConsensusPort, type ReplVariableEvidenceBinding, type ReplyContract, type ReplyEnvelope, type ReplyRowDecoded, type ReplyRowRejection, type RollupInput, type RollupOptions, type RollupResult, type RunAnalystBenchmarkOptions, type SandboxSdkTransportOpts, type SemanticConceptJudgeAdapterOpts, type SkillUsageRecord, type SkillUsageReport, type SkillUsageScanConfig, type StepLabelAdapterOptions, TREATMENT_LAYERS, type TaxonomyAttachment, type TaxonomyNode, type TaxonomyNodeLayer, type TaxonomyOutcome, type TeamIntegration, type TeamShape, type TeamShapeCheck, type TeamShapeClass, type TeamShapeEdge, type TeamShapeFeatures, type TeamShapeInput, type TeamShapeObserved, type TeamShapeOutcome, type TraceAnalysisEngine, type TraceAnalysisEngineRequest, type TraceAnalysisEngineResult, type TraceAnalystDefinition, type TraceAnalystLimits, type TraceQuestionOutcome, type TraceQuestionSpec, type TraceToolGroupName, type TreatmentLayer, type UpstreamPredictionAdapterOptions, type VerificationArtifactFile, type VerificationArtifactManifest, type VerificationArtifactRole, type VerificationAvailabilitySummary, type VerificationOutcome, type VerificationOutcomeSource, type VerificationOutcomeStatus, type VerificationResultFile, type VerifiedTraceFinding, analystUsageReceiptFromPrimeUsage, askTraceQuestions, assertFailureClasses, behavioralAnalyst, behavioralTraceQuestionDigest, buildDefaultAnalystRegistry, buildPrimePrompt, buildPrimeRepairPrompt, buildTraceToolsForGroup, computeFindingId, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createTraceAnalyst, diffFindings, diffFindingsOverTime, emitControlIntegrityFindings, evidenceRefsFromRawFinding, findingSubjectGrammarPromptFor, independentUnitCount, lineageRoots, makeFinding, makeProposalFinding, mergePrimeRawUsage, nodeHttpPrimeBridgeTransport, normalizePrimeUsage, openFindingsLedger, parseAnalystFinding, parseFindingSubject, parseRawFinding, pinBehavioralFailureClass, primeProtocolSha256, projectPrimeTrajectory, queryFindings, readFindingScope, renderFindingSubject, runAnalystBenchmark, runPrimeExchange, runTraceAnalyst, traceStoreEvidenceResolver };
|
|
1637
1683
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/analyst/adapters.ts","../../src/analyst/benchmark-agentrx-calibration.ts","../../src/analyst/benchmark-behavioral-calibration.ts","../../src/analyst/benchmark-dataset-types.ts","../../src/analyst/benchmark-dataset-agentrx.ts","../../src/analyst/benchmark-dataset-codetrace.ts","../../src/analyst/benchmark-verification-outcome.ts","../../src/analyst/benchmark-verification-artifacts.ts","../../src/analyst/benchmark-public-types.ts","../../src/analyst/benchmark-public-adapters.ts","../../src/analyst/benchmark-public-data.ts","../../src/analyst/definition.ts","../../src/analyst/benchmark-public-prompt.ts","../../src/analyst/benchmark-public-model.ts","../../src/analyst/benchmark-public-rlm.ts","../../src/analyst/benchmark-runner-prime.ts","../../src/analyst/benchmark-comparison.ts","../../src/analyst/benchmark-public-calibration.ts","../../src/analyst/benchmark-command-artifact.ts","../../src/analyst/benchmark-command-result.ts","../../src/analyst/benchmark-command.ts","../../src/analyst/benchmark-implementation.ts","../../src/analyst/bind.ts","../../src/analyst/define.ts","../../src/analyst/domain-taxonomy.ts","../../src/analyst/failure-modes.ts","../../src/analyst/kinds/skill-usage.ts","../../src/analyst/parse-tolerant.ts","../../src/analyst/promote.ts","../../src/analyst/proposal-findings.ts","../../src/analyst/rollup.ts","../../src/analyst/team-shape.ts"],"mappings":";;;;;;;;;;;;;;;;;UA+BiB;EACf;EACA;;EAEA,SAAS,KAAK;;EAEd;;qBCjCW;UAEI;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA,SAAS;;;UCbM;EACf,YAAY;EACZ;EACA,OAAO;EACP;;UAEe;EACf,YAAY;EACZ,OAAO;EACP;EACA;EACA;EACA;EACA;EACA;EACA;IAAgB;IAAe;;EAC/B;EACA;EACA,WAAW,qBAAqB;;;EAGhC,eAAe;IACb;IACA;IACA,WAAW;IACX,SAAS;;EAEX;;EAEA,aAAa;;;wBA+BC,kCAAkC;EAChD,OAAO;EACP,cAAc;EACd;EACA;EACA,mBAAmB;EACnB,eAAe;EACf,mBAAmB;EACnB,cAAc;IACZ;KCvFQ;UAEK;EACf,YAAY;EACZ;EACA;EACA;EACA;EACA;;UAGe;EACf,eAAe;EACf,mBAAmB;EACnB;IAAe,YAAY;IAAY;;EACvC,wBAAwB;EACxB;EACA;EACA;;UAGe;EACf,UAAU;EACV;EACA;EACA;EACA;;UAGe;EACf,UAAU;EACV,mBAAmB;EACnB;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA,iBAAiB;;KAGP,0CAEC,sCACA,iCACA;UAEI;EACf;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA,oCAAoC;;UAGrB;EACf,eAAe;EACf,WAAW,sBAAsB;;KAGvB;UAEK;;;;;EAKf,WAAW;;UAGI,kCACP,yBACN;UAEa,oCAAoC;EACnD;;EAEA;;UAGe,yCAAyC;EACxD;EACA;EACA;EACA;;UAGe,2CACP,kCACN;wBCtGY,qBAAqB,QACnC,KAAK,YACL,OAAO,QACP,UAAS,8BACR,qBAAqB;;wBAsGR,6BACd,mBAAmB,YACnB,iBACA,UAAS,mCACR;wBC9Ga,mBAAmB,QACjC,KAAK,mBACL,OAAO,QACP,UAAS,4BACR,qBAAqB;;wBAwFR,gCACd,2BACA,aAAa,uBACb,UAAS,qCACR;KCjHS;UAEK;EACf;EACA;EACA,QAAQ;;UAGO;EACf,QAAQ;EACR;EAKA;IAAe;IAAe;;EAC9B,SAAS;EACT;EACA;EACA;EACA;;UAGe;EACf;EACA;;KCdU;UAEK;EACf,MAAM;EACN;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;EACT;EACA;EACA;EACA;EACA;EACA,OAAO;EACP,cAAc;EACd,UAAU,OAAO;;UAGF;EACf,UAAU;EACV,SAAS;EACT,OAAO,MAAM;IAA6B;;;KChChC;;;;;;;;UASK;;WAEN;;WAEA;;;UAIM;EACf,MAAM;EACN;EACA,kBAAkB,aAAa;;EAE/B,UAAU;;UAGK;;EAEf,MAAM;;EAEN;;EAEA,kBAAkB,aAAa;EAC/B;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,UAAU;;EAEV;;EAEA,uBAAuB;EACvB;IACE,SAAS;IACT;IACA;IACA;IACA;IACA;IACA;IACA;IACA;;;;;;IAMA;;EAEF,aAAa;EACb;IACE;IACA;;;;;;;;;KAUQ,sCAAsC,KAChD,iEAGA,QAAQ,KAAK;UAEE;EACf,OAAO,qBAAqB;EAC5B;EACA;EACA;EACA,YAAY;IACV;IACA;IACA;;EAEF,uBAAuB;EACvB,WAAW;;UAGI;EACf;EACA;EACA,QAAQ;;UAGO;EACf,OAAO;EACP,OAAO;EACP,OAAO;EACP,YAAY;EACZ,QAAQ;;UAGO;EACf;EACA;EACA;EACA;EACA;EACA;EACA,QAAQ;EACR,UAAU;;;;;;;;UChHK;EACf;EACA;;EAEA;EACA;EACA,UAAU;EACV;EACA;EACA;EACA;EACA,WAAW;;;;;;;;;UAUI;EACf;EACA;;EAEA,kCAAkC;;EAElC;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;UASe;EACf;EACA,OAAO;;;;;;;;;wBA2Oa,6BAA6B;EACjD;EACA,iBAAiB;EACjB,OAAO;EACP;EACA;EACA,SAAS;IACP;EACF,UAAU;EACV,aAAa;EACb,YAAY;;wBCrQQ,wBACpB,eACC,QAAQ,MAAM;;;;;;KCLL,yBAAyB,KAAK;;;;;;KAS9B;WAEG;;WAEA;;;;;;WAMA;;WAGA;;WAEA;;;WAIA;WACA,WAAW;;;WAIX;WACA,WAAW;;UAKT;;WAEN;;WAEA;;WAEA;;WAEA,eAAe;;UAGT;;;;;WAKN;;UAKM;EACf,UAAU;;EAEV;;UAGe,eAAe;;EAE9B;EACA,eAAe;EACf,OAAO;EACP;EACA;;EAEA;EACA,SAAS;;;UAIM,sBAAsB;WAC5B;EACT,kBAAkB;;WAET,cAAc,SAAS;;;;;EAKhC,OAAO,iBAAiB,gBAAgB;;EAExC,QAAQ,iBAAiB,gBAAgB;EACzC,WAAW,MAAM,eAAe,QAAQ,QAAQ;;;UAIjC,uBAAuB;WAC7B;EACT,kBAAkB;WACT,cAAc,SAAS;;WAEvB;;WAEA;;EAET,YAAY;EACZ,WAAW,MAAM,eAAe,QAAQ,QAAQ;;EAEhD,gBAAgB;IACd;IACA,mBAAmB;IACnB,OAAO;IACP,SAAS;MACP;;;UAIW,0BAA0B,aAAa;;EAEtD,KAAK,SAAS,uBAAuB;IACnC,iBAAiB;IACjB;;;EAGF,OAAO;IACL;IACA,iBAAiB;IACjB,OAAO;IACP;IACA;IACA,SAAS;MACP,QAAQ;;EAEZ,aAAa,sBAAsB,gBAAgB;;;UAIpC,4BAA4B,uBAAuB;WACzD;;WAEA;EACT,kBAAkB;WACT,cAAc,SAAS;;WAEvB,qBAAqB,SAAS;;WAE9B;;EAET,qBAAqB,8BAA8B;;EAEnD,MAAM;IACJ;IACA,mBAAmB;IACnB;IACA,OAAO;IACP,SAAS;MACP;IAAU,UAAU;IAAkB,aAAa;IAAe;;;EAEtE,YAAY,0BAA0B,aAAa;;EAEnD,mBACE,QAAQ,oCACP,uBAAuB;;KAGhB,uBAAuB,MAAM,uBAAuB,oBAC5D,sBAAsB,QACtB,uBAAuB,QACvB,4BAA4B,aAAa;UAI5B,kBAAkB,gBAAgB,uBAAuB;;WAE/D;WACA;WACA;;WAEA;WACA,SAAS;;WAET;;WAEA;WACA,YAAY;WACZ,eAAe,cAAc;;;;;;WAM7B,gBAAgB,SAAS;WACzB,QAAQ;WACR,QAAQ;;;;;;WAMR;WACA,SAAS,uBAAuB,MAAM,aAAa;;;;;;;;;qBAUjD,mCAAmC;;;;;;;wBAUhC,gCAAgC,MAAM,aAAa,QACjE,YAAY,kBAAkB,MAAM,aAAa;;UAwClC;WACN;WACA,gBAAgB;WAChB,iBAAiB,YAAY;WAC7B;WACA;WACA;;WAEA;;WAEA;;UAGM;WACN;;WAEA;;WAEA,sBAAsB;WACtB,sBAAsB;;;;;;;;;wBAUjB,6BACd,aAAa,cAAc,gDAC1B;qBC9TU;;;;wBAqIG,8BAA8B,SAAS;;;;;;;;;;;UCvGtC;;EAEf;;EAEA;;EAEA;;;wBAIc,8BACd,SAAS,+BACT,MAAM,6BACL,kBAAkB;;wBAsEL,kCACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;cAmBpB,yBAAuB,EAAA;;;;;;;;;;;;;GASlB,EAAA,KAAA;cACL,gCAA8B,EAAA;;;;;;;;;;;;;;;;;;;GAsChC,EAAA,KAAA;cACE,uBAAqB,EAAA;;;;;;;;;;;;KA2BtB,yBAAyB,EAAE,aAAa;EAC3C,WAAW,EAAE,aAAa;;KAEvB,2BAA2B,EAAE,aAAa;KACnC,iCAAiC,yBAAyB;;;;;;;;;;;UC9KrD;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,cAAc;;;wBAIA,2BACd,SAAS,+BACT,MAAM,0BACL,kBAAkB,mBAAmB;;wBA+GxB,+BACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;UCjHT;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA,UAAU;;EAEV,YAAY;;;;;;wBA8EE;UAIC;;EAEf;;EAEA;;;;;;wBAOc,gCACd,MAAM,+BACL,kBAAkB;;wBAyDL,2BACd,SAAS,8BACR,uBAAuB;;;;;;;;cC3OpB;;;;;;;;;;;;;;;;;;;;KAqBM,uCAAuC;UAgBlC;EACf,QAAQ;EACR;;EAEA;;EAEA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;;UCzEM;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;;UClBM;EACf;EACA;EACA;IACE,SAAS;IACT;IACA;IACA;IACA;IACA,YAAY;MAAQ;MAAiB;MAAsB;;IAC3D,uBAAuB;IACvB,0BAA0B;IAC1B;MACE;MACA;MACA;MACA,QAAQ;;IAEV;MACE;MACA;;MAEA;MACA;;MAEA;MACA;MACA;MACA;MACA;MACA;MACA;MACA,UAAU;MACV;QACE;QACA;QACA;QACA;QACA;QACA;QACA;QACA;;MAEF;QACE;QACA;QACA;;MAEF;MACA;MACA;;MAEA;MACA;MACA;;;EAGJ,QAAQ;EACR,aAAa;EACb,uBAAuB;EACvB,qBAAqB;;UAGN;EACf;EACA;EACA;EACA;IACE;IACA;IACA;;;UAIa;EACf;IACE,SAAS;IACT;IACA;IACA;MACE;MACA;MACA;MACA;MACA;MACA;MACA;MACA;MACA,SAAS;MACT;QACE;QACA;QACA;QACA;QACA;QACA;QACA;QACA;;MAEF;QACE;QACA;QACA;;;IAGJ;IACA;IACA;IACA;;IAEA;IACA;IACA;IACA;;IAEA;IACA;IACA;IACA;;EAEF;IACE;IACA;IACA;IACA,YAAY;MAAQ;MAAiB;MAAsB;;IAC3D;IACA;;;UAIa;EACf;EACA;EACA;EACA;EACA,UAAU;;UAGK;EACf;EACA;EACA;EACA;IACE;IACA;IACA;IACA;;IAEA;;EAEF;EACA;IACE;IACA;IACA;;EAEF;IACE;IACA;IACA;IACA;IACA;IACA;;;UAIa;EACf;EACA;EACA;EACA,aAAa;EACb;;qBAGW;qBACA;qBACA;qBACA;wBC5KS,6BACpB,eACC,QAAQ;UCuEM;EACf,uBACE,SAAS,+BACT,QAAQ,wCACL,uBAAuB;EAC5B,2BACE,mBACA;IACE;IACA,aAAa,SAAS,OAAO;QAE5B,QAAQ;;;;;;;;;KAUV;;UAGK;EACR;EACA;;UAGe;EACf,SAAS;EACT,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;EACA,OAAO;EACP;EACA;EACA;EACA;;EAEA;EACA;EACA;;EAEA;;EAEA,QAAQ;EACR;EACA;;wBAKoB,2BACpB,yBACA,MAAK,OAAO,YACZ,eAAc,sCACb;qBC1IU;qBAGA;qBAGA;;KCKD;;WAGG;WACA;WACA;WACA,YAAY;WACZ,UAAU;;;WAIV;WACA,QAAQ;;;;;;wBAOP,YAAY,MAAM,aAAa,QAC7C,YAAY,kBAAkB,MAAM,aAAa,SACjD,YAAY,0BACX,uBAAuB;;;;;;;;wBCpCV,mBAAmB,YAAY,yBAAyB;UAkBvD;EACf;EACA;EACA;EACA,MAAM;EACN,SAAS,QAAQ;;UAGF,wCAAwC;;EAEvD,iBAAiB,SAAS;;;wBAIZ,oBACd,SAAS,kCACR,oBAAoB;wBACP,oBACd,SAAS,6BACR,QAAQ;;;;;;;;;;;;;;;cClCL;;qBAGO;;cAGP;KAEM,4BAA4B;UAEvB;EACf;EACA;EACA,OAAO;;EAEP;;EAEA;;UAGe;EACf,eAAe;;EAEf;EACA,OAAO;;KAGG,gBAAgB;EACtB;EAAiB,OAAO;;EACxB;EAAkB;IAAS;;;;;;;wBAkBjB,qBAAqB;EACnC;EACA,OAAO,cAAc;IACnB,gBAAgB;UAoDH;EACf;EACA;EACA,OAAO,QAAQ;;;;;EAKf;;EAEA;EACA;;;wBAIc,iBACd,UAAU,gBACV,YAAY,qBACX,gBAAgB;;wBA6BH,kBACd,UAAU,gBACV,0BACC,yBAAyB;;wBAiBZ,gBACd,UAAU,gBACV,aACC;;wBAca,iBAAiB,UAAU,gBAAgB;;wBAkB3C,aAAa,UAAU,gBAAgB;;;;;wBAWvC,qBACd,UAAU,gBACV;qBC5MW;qBACA;;qBAGA;qBAEA;;;;;;;;;;;;;;;;;;;;;KA0BD,gCAAgC;qBAuB/B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;UA6BI;EACf;EACA,OAAO;;UAGQ,gCAAgC;;EAE/C;;KAGU,oBAAoB;EAC1B;EAAiB,OAAO;;EACxB;EAAkB;IAAS;IAAkC;;;UA+FlD;;EAEf;;EAEA;;UAGe;EACf,MAAM;;EAEN;;EAEA;;EAEA;IAAkB;IAAiB;;EACnC,UAAU;IACR;IACA;IACA;IACA;MAAQ;MAAmB;MAAc;;;;;;;;;;;wBAWvB,oBACpB,SAAS,yBACR,QAAQ,oBAAoB;;;;;wBA+Cf,sBACd,eAAe,wBACf,sBAAsB,yBACrB;UA2Bc,2BAA2B,WAAW;EACrD;;EAEA;EACA,aAAa;EACb,cAAc,OAAO,GAAG,SAAS,mBAAmB,iBAAiB,QAAQ;;EAE7E,QAAQ,OAAO,GAAG,SAAS,mBAAmB;EAC9C,WAAW,OAAO,GAAG,SAAS;;;;;;;wBAQhB,0BAA0B,GACxC,SAAS,2BAA2B,KACnC,oBAAoB,QAAQ;;;;;;;;;;;;;;;;;;;;;KC7T1B;;UAGY;EACf;EACA,MAAM;;EAEN;EACA;;EAEA;;EAEA;;;EAGA;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;UAGe;EACf;EACA,SAAS;;UAGM;;EAEf;;EAEA;IAAc;IAAc,MAAM;;;EAElC;;;EAGA,kBAAkB;;EAElB;;;wBAyEc,sBAAsB,QAAQ,uBAAuB;;;;;;;;;;;;;wBCrIrD,gBAAgB;UC6Cf;;EAEf;;EAEA;;;UAIe;EACf;EACA;;EAEA;EACA;;UAGe;;EAEf,UAAU;EACV,UAAU;;EAEV,eAAe;;EAEf,UAAU;;wBAWI,QACd,mBAAmB,kBACnB,UAAU,gBACV,UAAS,iBACR;;;;;;wBCnFa,uBACd,mBACA,mBACC,cAAc;UCuBA;;EAEf;;EAEA;EACA;;EAEA;EACA;;UAGe;;EAEf,UAAU;EACV,eAAe;;EAEf,SAAS;;wBAWK,eACd,mBAAmB,kBACnB,UAAU,gBACV,UAAS,gBACR;UAiKc;EACf,mBAAmB;EACnB,UAAU;EACV,UAAU;;qBC3LC;;qBAYA;qBAEA;KAUD,yBAAyB;qBAExB;KACD,uBAAuB;KACvB,kBAAkB,KAAK,IAAI;UAEtB;EACf;;EAEA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA,wBAAwB;;UAGT;EACf;EACA;EACA;;UAGe;EACf;;EAEA;;EAEA;EACA;EACA;IAAU;IAAc;;;EAExB;;EAEA;EACA,uBAAuB;EACvB;IAAU;IAAc;;;EAExB;;EAEA,eAAe;;EAEf,iBAAiB;;EAEjB,sBAAsB;;UAGP;EACf,uBAAuB;EACvB;;UAGe;EACf,OAAO;EACP;;EAEA;;EAEA;EACA;EACA,UAAU;;EAEV,QAAQ,OAAO,gBAAgB;EAC/B;IAAS,SAAS;IAAiB,YAAY;IAAiB,eAAe;;EAC/E;IAAe;IAAe;;EAC9B;IAAU;IAAc;;;KAGd;EACN;EAAiB,OAAO;;EACxB;EAAkB;IAAS;;;;wBAuBjB,0BACd,SAAS,sBACT,wBAAwB;EACrB;EAAiB,OAAO;;EAAqB;EAAkB;IAAS;;;wBAkB7D,kBAAkB,OAAO,iBAAiB;;;;;;wBAqa1C,oBAAoB,QAAQ"}
|
|
1
|
+
{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/analyst/adapters.ts","../../src/analyst/benchmark-agentrx-calibration.ts","../../src/analyst/benchmark-behavioral-calibration.ts","../../src/analyst/benchmark-dataset-types.ts","../../src/analyst/benchmark-dataset-agentrx.ts","../../src/analyst/benchmark-dataset-codetrace.ts","../../src/analyst/benchmark-verification-outcome.ts","../../src/analyst/benchmark-verification-artifacts.ts","../../src/analyst/benchmark-public-types.ts","../../src/analyst/benchmark-public-adapters.ts","../../src/analyst/benchmark-public-data.ts","../../src/analyst/definition.ts","../../src/analyst/benchmark-public-prompt.ts","../../src/analyst/benchmark-public-model.ts","../../src/analyst/benchmark-public-rlm.ts","../../src/analyst/benchmark-runner-prime.ts","../../src/analyst/benchmark-comparison.ts","../../src/analyst/benchmark-public-calibration.ts","../../src/analyst/benchmark-command-artifact.ts","../../src/analyst/benchmark-command-result.ts","../../src/analyst/benchmark-command.ts","../../src/analyst/benchmark-implementation.ts","../../src/analyst/bind.ts","../../src/analyst/define.ts","../../src/analyst/domain-taxonomy.ts","../../src/analyst/failure-modes.ts","../../src/analyst/kinds/skill-usage.ts","../../src/analyst/parse-tolerant.ts","../../src/analyst/promote.ts","../../src/analyst/proposal-findings.ts","../../src/analyst/rollup.ts","../../src/analyst/team-shape.ts"],"mappings":";;;;;;;;;;;;;;;;;;UA+BiB;EACf;EACA;;EAEA,SAAS,KAAK;;EAEd;;qBCjCW;UAEI;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA,SAAS;;;UCVM;EACf,YAAY;EACZ;EACA,OAAO;EACP;;;;UAIe;EACf;EACA;EACA;EACA,IAAI;EACJ;;UAEe,sCAAsC;EACrD,YAAY;EACZ,OAAO;EACP;EACA;EACA;EACA;EACA;EACA;EACA;IAAgB;IAAe;;EAC/B;EACA;EACA,SAAS;EACT,WAAW,qBAAqB;;;EAGhC,eAAe;IACb;IACA;IACA,WAAW;IACX,SAAS;;EAEX;;EAEA,aAAa;;UAEE;EACf;EACA;;UAEe;EACf;EACA;EACA;EACA,eAAe;;;UAGA,gCAAgC;EAC/C;EACA;EACA;;UAEe;EACf;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA,SAAS;EACT,MAAM;;;wBAmCQ,kCAAkC;EAChD,OAAO;EACP,cAAc;EACd;EACA;EACA,mBAAmB;EACnB,eAAe;EACf,mBAAmB;EACnB,cAAc;EACd;EACA;IACE;;;wBAuHY,wBACd,eAAe,cACf,mBAAmB,oBACnB,UAAS;EAA4B;IACpC;KCjQS;UAEK;EACf,YAAY;EACZ;EACA;EACA;EACA;EACA;;UAGe;EACf,eAAe;EACf,mBAAmB;EACnB;IAAe,YAAY;IAAY;;EACvC,wBAAwB;EACxB;EACA;EACA;;UAGe;EACf,UAAU;EACV;EACA;EACA;EACA;;UAGe;EACf,UAAU;EACV,mBAAmB;EACnB;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA,iBAAiB;;KAGP,0CAEC,sCACA,iCACA;UAEI;EACf;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA,oCAAoC;;UAGrB;EACf,eAAe;EACf,WAAW,sBAAsB;;KAGvB;UAEK;;;;;EAKf,WAAW;;UAGI,kCACP,yBACN;UAEa,oCAAoC;EACnD;;EAEA;;UAGe,yCAAyC;EACxD;EACA;EACA;EACA;;UAGe,2CACP,kCACN;wBCtGY,qBAAqB,QACnC,KAAK,YACL,OAAO,QACP,UAAS,8BACR,qBAAqB;;wBAsGR,6BACd,mBAAmB,YACnB,iBACA,UAAS,mCACR;wBC9Ga,mBAAmB,QACjC,KAAK,mBACL,OAAO,QACP,UAAS,4BACR,qBAAqB;;wBAwFR,gCACd,2BACA,aAAa,uBACb,UAAS,qCACR;KCjHS;UAEK;EACf;EACA;EACA,QAAQ;;UAGO;EACf,QAAQ;EACR;EAKA;IAAe;IAAe;;EAC9B,SAAS;EACT;EACA;EACA;EACA;;UAGe;EACf;EACA;;KCdU;UAEK;EACf,MAAM;EACN;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;EACT;EACA;EACA;EACA;EACA;EACA,OAAO;EACP,cAAc;EACd,UAAU,OAAO;;UAGF;EACf,UAAU;EACV,SAAS;EACT,OAAO,MAAM;IAA6B;;;KChChC;;;;;;;;UASK;;WAEN;;WAEA;;;UAIM;EACf,MAAM;EACN;EACA,kBAAkB,aAAa;;EAE/B,UAAU;;UAGK;;EAEf,MAAM;;EAEN;;EAEA,kBAAkB,aAAa;EAC/B;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,UAAU;;EAEV;;EAEA,uBAAuB;EACvB;IACE,SAAS;IACT;IACA;IACA;IACA;IACA;IACA;IACA;IACA;;;;;;IAMA;;EAEF,aAAa;EACb;IACE;IACA;;;;;;;;;KAUQ,sCAAsC,KAChD,iEAGA,QAAQ,KAAK;UAEE;EACf,OAAO,qBAAqB;EAC5B;EACA;EACA;EACA,YAAY;IACV;IACA;IACA;;EAEF,uBAAuB;EACvB,WAAW;;UAGI;EACf;EACA;EACA,QAAQ;;UAGO;EACf,OAAO;EACP,OAAO;EACP,OAAO;EACP,YAAY;EACZ,QAAQ;;UAGO;EACf;EACA;EACA;EACA;EACA;EACA;EACA,QAAQ;EACR,UAAU;;;;;;;;UChHK;EACf;EACA;;EAEA;EACA;EACA,UAAU;EACV;EACA;EACA;EACA;EACA,WAAW;;;;;;;;;UAUI;EACf;EACA;;EAEA,kCAAkC;;EAElC;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;UASe;EACf;EACA,OAAO;;;;;;;;;wBA2Oa,6BAA6B;EACjD;EACA,iBAAiB;EACjB,OAAO;EACP;EACA;EACA,SAAS;IACP;EACF,UAAU;EACV,aAAa;EACb,YAAY;;wBCrQQ,wBACpB,eACC,QAAQ,MAAM;;;;;;KCLL,yBAAyB,KAAK;;;;;;KAS9B;WAEG;;WAEA;;;;;;WAMA;;WAGA;;WAEA;;;WAIA;WACA,WAAW;;;WAIX;WACA,WAAW;;UAKT;;WAEN;;WAEA;;WAEA;;WAEA,eAAe;;UAGT;;;;;WAKN;;UAKM;EACf,UAAU;;EAEV;;UAGe,eAAe;;EAE9B;EACA,eAAe;EACf,OAAO;EACP;EACA;;EAEA;EACA,SAAS;;;UAIM,sBAAsB;WAC5B;EACT,kBAAkB;;WAET,cAAc,SAAS;;;;;EAKhC,OAAO,iBAAiB,gBAAgB;;EAExC,QAAQ,iBAAiB,gBAAgB;EACzC,WAAW,MAAM,eAAe,QAAQ,QAAQ;;;UAIjC,uBAAuB;WAC7B;EACT,kBAAkB;WACT,cAAc,SAAS;;WAEvB;;WAEA;;EAET,YAAY;EACZ,WAAW,MAAM,eAAe,QAAQ,QAAQ;;EAEhD,gBAAgB;IACd;IACA,mBAAmB;IACnB,OAAO;IACP,SAAS;MACP;;;UAIW,0BAA0B,aAAa;;EAEtD,KAAK,SAAS,uBAAuB;IACnC,iBAAiB;IACjB;;;EAGF,OAAO;IACL;IACA,iBAAiB;IACjB,OAAO;IACP;IACA;IACA,SAAS;MACP,QAAQ;;EAEZ,aAAa,sBAAsB,gBAAgB;;;UAIpC,4BAA4B,uBAAuB;WACzD;;WAEA;EACT,kBAAkB;WACT,cAAc,SAAS;;WAEvB,qBAAqB,SAAS;;WAE9B;;EAET,qBAAqB,8BAA8B;;EAEnD,MAAM;IACJ;IACA,mBAAmB;IACnB;IACA,OAAO;IACP,SAAS;MACP;IAAU,UAAU;IAAkB,aAAa;IAAe;;;EAEtE,YAAY,0BAA0B,aAAa;;EAEnD,mBACE,QAAQ,oCACP,uBAAuB;;KAGhB,uBAAuB,MAAM,uBAAuB,oBAC5D,sBAAsB,QACtB,uBAAuB,QACvB,4BAA4B,aAAa;UAI5B,kBAAkB,gBAAgB,uBAAuB;;WAE/D;WACA;WACA;;WAEA;WACA,SAAS;;WAET;;WAEA;WACA,YAAY;WACZ,eAAe,cAAc;;;;;;WAM7B,gBAAgB,SAAS;WACzB,QAAQ;WACR,QAAQ;;;;;;WAMR;WACA,SAAS,uBAAuB,MAAM,aAAa;;;;;;;;;qBAUjD,mCAAmC;;;;;;;wBAUhC,gCAAgC,MAAM,aAAa,QACjE,YAAY,kBAAkB,MAAM,aAAa;;UAwClC;WACN;WACA,gBAAgB;WAChB,iBAAiB,YAAY;WAC7B;WACA;WACA;;WAEA;;WAEA;;UAGM;WACN;;WAEA;;WAEA,sBAAsB;WACtB,sBAAsB;;;;;;;;;wBAUjB,6BACd,aAAa,cAAc,gDAC1B;qBC9TU;;;;wBAqIG,8BAA8B,SAAS;;;;;;;;;;;UCvGtC;;EAEf;;EAEA;;EAEA;;;wBAIc,8BACd,SAAS,+BACT,MAAM,6BACL,kBAAkB;;wBAsEL,kCACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;cAmBpB,yBAAuB,EAAA;;;;;;;;;;;;;GASlB,EAAA,KAAA;cACL,gCAA8B,EAAA;;;;;;;;;;;;;;;;;;;GAsChC,EAAA,KAAA;cACE,uBAAqB,EAAA;;;;;;;;;;;;KA2BtB,yBAAyB,EAAE,aAAa;EAC3C,WAAW,EAAE,aAAa;;KAEvB,2BAA2B,EAAE,aAAa;KACnC,iCAAiC,yBAAyB;;;;;;;;;;;UC9KrD;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,cAAc;;;wBAIA,2BACd,SAAS,+BACT,MAAM,0BACL,kBAAkB,mBAAmB;;wBA+GxB,+BACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;UCjHT;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA,UAAU;;EAEV,YAAY;;;;;;wBA8EE;UAIC;;EAEf;;EAEA;;;;;;wBAOc,gCACd,MAAM,+BACL,kBAAkB;;wBAyDL,2BACd,SAAS,8BACR,uBAAuB;;;;;;;;cC3OpB;;;;;;;;;;;;;;;;;;;;KAqBM,uCAAuC;UAgBlC;EACf,QAAQ;EACR;;EAEA;;EAEA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;;UCzEM;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;;UClBM;EACf;EACA;EACA;IACE,SAAS;IACT;IACA;IACA;IACA;IACA,YAAY;MAAQ;MAAiB;MAAsB;;IAC3D,uBAAuB;IACvB,0BAA0B;IAC1B;MACE;MACA;MACA;MACA,QAAQ;;IAEV;MACE;MACA;;MAEA;MACA;;MAEA;MACA;MACA;MACA;MACA;MACA;MACA;MACA,UAAU;MACV;QACE;QACA;QACA;QACA;QACA;QACA;QACA;QACA;;MAEF;QACE;QACA;QACA;;MAEF;MACA;MACA;;MAEA;MACA;MACA;;;EAGJ,QAAQ;EACR,aAAa;EACb,uBAAuB;EACvB,qBAAqB;;UAGN;EACf;EACA;EACA;EACA;IACE;IACA;IACA;;;UAIa;EACf;IACE,SAAS;IACT;IACA;IACA;MACE;MACA;MACA;MACA;MACA;MACA;MACA;MACA;MACA,SAAS;MACT;QACE;QACA;QACA;QACA;QACA;QACA;QACA;QACA;;MAEF;QACE;QACA;QACA;;;IAGJ;IACA;IACA;IACA;;IAEA;IACA;IACA;IACA;;IAEA;IACA;IACA;IACA;;EAEF;IACE;IACA;IACA;IACA,YAAY;MAAQ;MAAiB;MAAsB;;IAC3D;IACA;;;UAIa;EACf;EACA;EACA;EACA;EACA,UAAU;;UAGK;EACf;EACA;EACA;EACA;IACE;IACA;IACA;IACA;;IAEA;;EAEF;EACA;IACE;IACA;IACA;;EAEF;IACE;IACA;IACA;IACA;IACA;IACA;;;UAIa;EACf;EACA;EACA;EACA,aAAa;EACb;;qBAGW;qBACA;qBACA;qBACA;wBC5KS,6BACpB,eACC,QAAQ;UCuEM;EACf,uBACE,SAAS,+BACT,QAAQ,wCACL,uBAAuB;EAC5B,2BACE,mBACA;IACE;IACA,aAAa,SAAS,OAAO;QAE5B,QAAQ;;;;;;;;;KAUV;;UAGK;EACR;EACA;;UAGe;EACf,SAAS;EACT,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;EACA,OAAO;EACP;EACA;EACA;EACA;;EAEA;EACA;EACA;;EAEA;;EAEA,QAAQ;EACR;EACA;;wBAKoB,2BACpB,yBACA,MAAK,OAAO,YACZ,eAAc,sCACb;qBC1IU;qBAGA;qBAGA;;KCKD;;WAGG;WACA;WACA;WACA,YAAY;WACZ,UAAU;;;WAIV;WACA,QAAQ;;;;;;wBAOP,YAAY,MAAM,aAAa,QAC7C,YAAY,kBAAkB,MAAM,aAAa,SACjD,YAAY,0BACX,uBAAuB;;;;;;;;wBCpCV,mBAAmB,YAAY,yBAAyB;UAkBvD;EACf;EACA;EACA;EACA,MAAM;EACN,SAAS,QAAQ;;UAGF,wCAAwC;;EAEvD,iBAAiB,SAAS;;;wBAIZ,oBACd,SAAS,kCACR,oBAAoB;wBACP,oBACd,SAAS,6BACR,QAAQ;;;;;;;;;;;;;;;cClCL;;qBAGO;;cAGP;KAEM,4BAA4B;UAEvB;EACf;EACA;EACA,OAAO;;EAEP;;EAEA;;UAGe;EACf,eAAe;;EAEf;EACA,OAAO;;KAGG,gBAAgB;EACtB;EAAiB,OAAO;;EACxB;EAAkB;IAAS;;;;;;;wBAkBjB,qBAAqB;EACnC;EACA,OAAO,cAAc;IACnB,gBAAgB;UAoDH;EACf;EACA;EACA,OAAO,QAAQ;;;;;EAKf;;EAEA;EACA;;;wBAIc,iBACd,UAAU,gBACV,YAAY,qBACX,gBAAgB;;wBA6BH,kBACd,UAAU,gBACV,0BACC,yBAAyB;;wBAiBZ,gBACd,UAAU,gBACV,aACC;;wBAca,iBAAiB,UAAU,gBAAgB;;wBAkB3C,aAAa,UAAU,gBAAgB;;;;;wBAWvC,qBACd,UAAU,gBACV;qBC5MW;qBACA;;qBAGA;qBAEA;;;;;;;;;;;;;;;;;;;;;KA0BD,gCAAgC;qBAuB/B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;UA6BI;EACf;EACA,OAAO;;UAGQ,gCAAgC;;EAE/C;;KAGU,oBAAoB;EAC1B;EAAiB,OAAO;;EACxB;EAAkB;IAAS;IAAkC;;;UA+FlD;;EAEf;;EAEA;;UAGe;EACf,MAAM;;EAEN;;EAEA;;EAEA;IAAkB;IAAiB;;EACnC,UAAU;IACR;IACA;IACA;IACA;MAAQ;MAAmB;MAAc;;;;;;;;;;;wBAWvB,oBACpB,SAAS,yBACR,QAAQ,oBAAoB;;;;;wBA+Cf,sBACd,eAAe,wBACf,sBAAsB,yBACrB;UA2Bc,2BAA2B,WAAW;EACrD;;EAEA;EACA,aAAa;EACb,cAAc,OAAO,GAAG,SAAS,mBAAmB,iBAAiB,QAAQ;;EAE7E,QAAQ,OAAO,GAAG,SAAS,mBAAmB;EAC9C,WAAW,OAAO,GAAG,SAAS;;;;;;;wBAQhB,0BAA0B,GACxC,SAAS,2BAA2B,KACnC,oBAAoB,QAAQ;;;;;;;;;;;;;;;;;;;;;KC7T1B;;UAGY;EACf;EACA,MAAM;;EAEN;EACA;;EAEA;;EAEA;;;EAGA;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;UAGe;EACf;EACA,SAAS;;UAGM;;EAEf;;EAEA;IAAc;IAAc,MAAM;;;EAElC;;;EAGA,kBAAkB;;EAElB;;;wBAyEc,sBAAsB,QAAQ,uBAAuB;;;;;;;;;;;;;wBCrIrD,gBAAgB;UC6Cf;;EAEf;;EAEA;;;UAIe;EACf;EACA;;EAEA;EACA;;UAGe;;EAEf,UAAU;EACV,UAAU;;EAEV,eAAe;;EAEf,UAAU;;wBAWI,QACd,mBAAmB,kBACnB,UAAU,gBACV,UAAS,iBACR;;;;;;wBCnFa,uBACd,mBACA,mBACC,cAAc;UCuBA;;EAEf;;EAEA;EACA;;EAEA;EACA;;UAGe;;EAEf,UAAU;EACV,eAAe;;EAEf,SAAS;;wBAWK,eACd,mBAAmB,kBACnB,UAAU,gBACV,UAAS,gBACR;UAiKc;EACf,mBAAmB;EACnB,UAAU;EACV,UAAU;;qBC3LC;;qBAYA;qBAEA;KAUD,yBAAyB;qBAExB;KACD,uBAAuB;KACvB,kBAAkB,KAAK,IAAI;UAEtB;EACf;;EAEA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA,wBAAwB;;UAGT;EACf;EACA;EACA;;UAGe;EACf;;EAEA;;EAEA;EACA;EACA;IAAU;IAAc;;;EAExB;;EAEA;EACA,uBAAuB;EACvB;IAAU;IAAc;;;EAExB;;EAEA,eAAe;;EAEf,iBAAiB;;EAEjB,sBAAsB;;UAGP;EACf,uBAAuB;EACvB;;UAGe;EACf,OAAO;EACP;;EAEA;;EAEA;EACA;EACA,UAAU;;EAEV,QAAQ,OAAO,gBAAgB;EAC/B;IAAS,SAAS;IAAiB,YAAY;IAAiB,eAAe;;EAC/E;IAAe;IAAe;;EAC9B;IAAU;IAAc;;;KAGd;EACN;EAAiB,OAAO;;EACxB;EAAkB;IAAS;;;;wBAuBjB,0BACd,SAAS,sBACT,wBAAwB;EACrB;EAAiB,OAAO;;EAAqB;EAAkB;IAAS;;;wBAkB7D,kBAAkB,OAAO,iBAAiB;;;;;;wBAqa1C,oBAAoB,QAAQ"}
|