@tangle-network/agent-eval 0.183.0 → 0.185.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +28 -0
- package/dist/adapters/http.d.ts +2 -2
- package/dist/{agent-profile-cell-s__adRnK.d.ts → agent-profile-cell-BHcB-mZX.d.ts} +8 -9
- package/dist/{agent-profile-cell-s__adRnK.d.ts.map → agent-profile-cell-BHcB-mZX.d.ts.map} +1 -1
- package/dist/{agent-profile-cell-Cv6UA-W_.js → agent-profile-cell-DcFW4vOc.js} +16 -15
- package/dist/agent-profile-cell-DcFW4vOc.js.map +1 -0
- package/dist/{agent-profile-CivaSsSy.d.ts → agent-profile-uAIQK3Nr.d.ts} +3 -3
- package/dist/{agent-profile-CivaSsSy.d.ts.map → agent-profile-uAIQK3Nr.d.ts.map} +1 -1
- package/dist/analyst/index.d.ts +11 -336
- package/dist/analyst/index.d.ts.map +1 -1
- package/dist/analyst/index.js +10 -8
- package/dist/analyst/index.js.map +1 -1
- package/dist/analyze-runs-CrYhqXop.d.ts +75 -0
- package/dist/analyze-runs-CrYhqXop.d.ts.map +1 -0
- package/dist/{define-agent-eval-CneiWyuA.js → analyze-runs-sTygaDyD.js} +6 -813
- package/dist/analyze-runs-sTygaDyD.js.map +1 -0
- package/dist/{benchmark-DQKzykkO.js → benchmark-BeQJyrRC.js} +3 -3
- package/dist/{benchmark-DQKzykkO.js.map → benchmark-BeQJyrRC.js.map} +1 -1
- package/dist/{benchmark-command-NBrqNwgy.js → benchmark-command-kvq7yFY4.js} +17 -54
- package/dist/benchmark-command-kvq7yFY4.js.map +1 -0
- package/dist/benchmarks/index.d.ts +4 -4
- package/dist/benchmarks/index.js +3 -3
- package/dist/builder-eval/index.js +1 -1
- package/dist/campaign/index.d.ts +6 -5
- package/dist/campaign/index.js +10 -9
- package/dist/{campaign-BwcEBC91.js → campaign-eJgtRs-W.js} +16 -15
- package/dist/{campaign-BwcEBC91.js.map → campaign-eJgtRs-W.js.map} +1 -1
- package/dist/{campaign-evidence-PYPUftRy.js → campaign-evidence-BsrCmBSE.js} +7 -258
- package/dist/campaign-evidence-BsrCmBSE.js.map +1 -0
- package/dist/{canonical-DPyQ_rpt.js → canonical-BsZVgBhW.js} +2 -4
- package/dist/canonical-BsZVgBhW.js.map +1 -0
- package/dist/cli.js +2 -2
- package/dist/{client-kh2jOjTK.d.ts → client-Bc8LANCG.d.ts} +4 -4
- package/dist/{client-kh2jOjTK.d.ts.map → client-Bc8LANCG.d.ts.map} +1 -1
- package/dist/contract/index.d.ts +10 -10
- package/dist/contract/index.js +15 -284
- package/dist/contract/index.js.map +1 -1
- package/dist/{default-registry-BwDSWVzg.d.ts → default-registry-CXv9kHMt.d.ts} +6 -6
- package/dist/{default-registry-BwDSWVzg.d.ts.map → default-registry-CXv9kHMt.d.ts.map} +1 -1
- package/dist/{default-registry-C0-N4PkG.js → default-registry-xiHrldW8.js} +6 -4
- package/dist/{default-registry-C0-N4PkG.js.map → default-registry-xiHrldW8.js.map} +1 -1
- package/dist/define-agent-eval-D4uumjl3.js +813 -0
- package/dist/define-agent-eval-D4uumjl3.js.map +1 -0
- package/dist/{define-agent-eval-CwOWWQt_.d.ts → define-agent-eval-DE67ECH1.d.ts} +7 -7
- package/dist/{define-agent-eval-CwOWWQt_.d.ts.map → define-agent-eval-DE67ECH1.d.ts.map} +1 -1
- package/dist/diagnosis/index.d.ts +319 -0
- package/dist/diagnosis/index.d.ts.map +1 -0
- package/dist/diagnosis/index.js +1328 -0
- package/dist/diagnosis/index.js.map +1 -0
- package/dist/{dspy-rlm-engine-8pgBSiy7.js → dspy-rlm-engine-BE9lCzzL.js} +3 -3
- package/dist/{dspy-rlm-engine-8pgBSiy7.js.map → dspy-rlm-engine-BE9lCzzL.js.map} +1 -1
- package/dist/{engine-DS1cysJy.d.ts → engine-cXUK-WHS.d.ts} +5 -77
- package/dist/engine-cXUK-WHS.d.ts.map +1 -0
- package/dist/equal-terms-Blx9Zk5I.js +25 -0
- package/dist/equal-terms-Blx9Zk5I.js.map +1 -0
- package/dist/{eval-campaign-CCEkwzoH.js → eval-campaign-XFHEpN86.js} +4 -4
- package/dist/{eval-campaign-CCEkwzoH.js.map → eval-campaign-XFHEpN86.js.map} +1 -1
- package/dist/evaluation.d.ts +75 -0
- package/dist/evaluation.d.ts.map +1 -0
- package/dist/evaluation.js +137 -0
- package/dist/evaluation.js.map +1 -0
- package/dist/{exact-types-BZDe0W2D.d.ts → exact-types-BEHN7tnq.d.ts} +2 -2
- package/dist/{exact-types-BZDe0W2D.d.ts.map → exact-types-BEHN7tnq.d.ts.map} +1 -1
- package/dist/experiment/index.d.ts +2 -2
- package/dist/experiment/index.js +6 -5
- package/dist/experiment/index.js.map +1 -1
- package/dist/{experiment-tracker-C7PfnF4b.d.ts → experiment-tracker-URRGP9ek.d.ts} +2 -2
- package/dist/{experiment-tracker-C7PfnF4b.d.ts.map → experiment-tracker-URRGP9ek.d.ts.map} +1 -1
- package/dist/{external-optimizer-process-DIB4YWEe.js → external-optimizer-process-BbhvXTFx.js} +4 -4
- package/dist/{external-optimizer-process-DIB4YWEe.js.map → external-optimizer-process-BbhvXTFx.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-CjguXN2e.js → external-optimizer-subprocess-DMgmqTB6.js} +3 -3
- package/dist/{external-optimizer-subprocess-CjguXN2e.js.map → external-optimizer-subprocess-DMgmqTB6.js.map} +1 -1
- package/dist/{feedback-trajectory-CXmtITBo.d.ts → feedback-trajectory-CytTR7u4.d.ts} +3 -3
- package/dist/{feedback-trajectory-CXmtITBo.d.ts.map → feedback-trajectory-CytTR7u4.d.ts.map} +1 -1
- package/dist/hosted/index.d.ts +2 -2
- package/dist/{index-DoykkxW0.d.ts → index-BQWfAFeH.d.ts} +11 -11
- package/dist/{index-DoykkxW0.d.ts.map → index-BQWfAFeH.d.ts.map} +1 -1
- package/dist/{index-DbG8QbH9.d.ts → index-C8_iU8KL.d.ts} +6 -6
- package/dist/{index-DbG8QbH9.d.ts.map → index-C8_iU8KL.d.ts.map} +1 -1
- package/dist/{index--_8_Nlpn.d.ts → index-CKZa_rnt.d.ts} +4 -121
- package/dist/index-CKZa_rnt.d.ts.map +1 -0
- package/dist/{index-Bp_6sj3x.d.ts → index-ww8736Am.d.ts} +20 -315
- package/dist/index-ww8736Am.d.ts.map +1 -0
- package/dist/index.d.ts +24 -22
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +25 -24
- package/dist/index.js.map +1 -1
- package/dist/{insight-report-D1qa0HWs.d.ts → insight-report-C3Bfl762.d.ts} +3 -3
- package/dist/{insight-report-D1qa0HWs.d.ts.map → insight-report-C3Bfl762.d.ts.map} +1 -1
- package/dist/jev-protocol.d.ts +74 -0
- package/dist/jev-protocol.d.ts.map +1 -0
- package/dist/jev-protocol.js +91 -0
- package/dist/jev-protocol.js.map +1 -0
- package/dist/jev.d.ts +131 -0
- package/dist/jev.d.ts.map +1 -0
- package/dist/jev.js +312 -0
- package/dist/jev.js.map +1 -0
- package/dist/{journal-Cs9f7385.js → journal-BuRLm6ho.js} +2 -2
- package/dist/{journal-Cs9f7385.js.map → journal-BuRLm6ho.js.map} +1 -1
- package/dist/{kind-factory-DxTzg32u.js → kind-factory-V79Hw0AU.js} +7 -84
- package/dist/kind-factory-V79Hw0AU.js.map +1 -0
- package/dist/ledger-core/index.js +2 -2
- package/dist/{llm-judge-FfaRgz9X.js → llm-judge-FWWhEb31.js} +11 -9
- package/dist/{llm-judge-FfaRgz9X.js.map → llm-judge-FWWhEb31.js.map} +1 -1
- package/dist/{matrix-CyhW-vgJ.d.ts → matrix-CPKsRY8Y.d.ts} +2 -2
- package/dist/{matrix-CyhW-vgJ.d.ts.map → matrix-CPKsRY8Y.d.ts.map} +1 -1
- package/dist/meta-eval/index.d.ts +54 -2
- package/dist/meta-eval/index.d.ts.map +1 -1
- package/dist/meta-eval/index.js +52 -3
- package/dist/meta-eval/index.js.map +1 -1
- package/dist/{mint-ySIIkKlV.js → mint-B4zgSAJW.js} +63 -14
- package/dist/mint-B4zgSAJW.js.map +1 -0
- package/dist/mint-BI14xehJ.d.ts +285 -0
- package/dist/mint-BI14xehJ.d.ts.map +1 -0
- package/dist/multishot/golden/index.d.ts +1 -1
- package/dist/multishot/index.d.ts +2 -2
- package/dist/openapi.json +1 -1
- package/dist/otel-spans-CQY1pNiN.js +275 -0
- package/dist/otel-spans-CQY1pNiN.js.map +1 -0
- package/dist/otlp-attributes-BpC_blLF.js +83 -0
- package/dist/otlp-attributes-BpC_blLF.js.map +1 -0
- package/dist/{run-record-DPa2PeHx.js → paired-promotion-decision-ClJh_9nK.js} +2 -215
- package/dist/paired-promotion-decision-ClJh_9nK.js.map +1 -0
- package/dist/pipelines/index.js +2 -2
- package/dist/prime-bridge-transport-Z3cpKGoV.js +43 -0
- package/dist/prime-bridge-transport-Z3cpKGoV.js.map +1 -0
- package/dist/prime-protocol-CDBcL9NM.d.ts +331 -0
- package/dist/prime-protocol-CDBcL9NM.d.ts.map +1 -0
- package/dist/{prime-protocol-6tZTVsWm.js → prime-protocol-DcAK9pC8.js} +2 -23
- package/dist/prime-protocol-DcAK9pC8.js.map +1 -0
- package/dist/{produced-state-CGoVInrl.js → produced-state-CmB02tr2.js} +23 -19
- package/dist/{produced-state-CGoVInrl.js.map → produced-state-CmB02tr2.js.map} +1 -1
- package/dist/profile-cell.d.ts +1 -1
- package/dist/profile-cell.js +1 -1
- package/dist/{promotion-policy-BKFrl7CC.js → promotion-policy-DdrYWHDh.js} +3 -3
- package/dist/{promotion-policy-BKFrl7CC.js.map → promotion-policy-DdrYWHDh.js.map} +1 -1
- package/dist/{query-D1nLIKt7.js → query-DlV0pdx4.js} +2 -2
- package/dist/{query-D1nLIKt7.js.map → query-DlV0pdx4.js.map} +1 -1
- package/dist/{registry-BRbB6Y0v.d.ts → registry-D5Ly3GRV.d.ts} +3 -3
- package/dist/{registry-BRbB6Y0v.d.ts.map → registry-D5Ly3GRV.d.ts.map} +1 -1
- package/dist/{release-confidence-BcqeQTHW.d.ts → release-confidence-DsETiO4f.d.ts} +3 -3
- package/dist/{release-confidence-BcqeQTHW.d.ts.map → release-confidence-DsETiO4f.d.ts.map} +1 -1
- package/dist/{release-confidence-DMg8n18l.js → release-confidence-RanRUXRW.js} +2 -2
- package/dist/{release-confidence-DMg8n18l.js.map → release-confidence-RanRUXRW.js.map} +1 -1
- package/dist/reporting.d.ts +3 -3
- package/dist/reporting.js +2 -2
- package/dist/{researcher-64T49THL.d.ts → researcher-CJkH-6Ty.d.ts} +4 -4
- package/dist/{researcher-64T49THL.d.ts.map → researcher-CJkH-6Ty.d.ts.map} +1 -1
- package/dist/{reward-hacking-uzO_ihep.d.ts → reward-hacking-BIQ2BWLG.d.ts} +2 -2
- package/dist/{reward-hacking-uzO_ihep.d.ts.map → reward-hacking-BIQ2BWLG.d.ts.map} +1 -1
- package/dist/rl.d.ts +10 -6
- package/dist/rl.d.ts.map +1 -1
- package/dist/rl.js +22 -7
- package/dist/rl.js.map +1 -1
- package/dist/rollout/index.d.ts +2 -2
- package/dist/rollout/index.js +2 -2
- package/dist/{rollout-D_54FZVu.js → rollout-C0Ifk3vK.js} +2 -2
- package/dist/{rollout-D_54FZVu.js.map → rollout-C0Ifk3vK.js.map} +1 -1
- package/dist/{rubric-predictive-validity-Bmj2_cll.d.ts → rubric-predictive-validity-CTU8y8Cv.d.ts} +2 -2
- package/dist/{rubric-predictive-validity-Bmj2_cll.d.ts.map → rubric-predictive-validity-CTU8y8Cv.d.ts.map} +1 -1
- package/dist/run-record-CNNzWGpB.js +216 -0
- package/dist/run-record-CNNzWGpB.js.map +1 -0
- package/dist/{run-record-BiTWauyO.d.ts → run-record-D9wS35fK.d.ts} +2 -2
- package/dist/{run-record-BiTWauyO.d.ts.map → run-record-D9wS35fK.d.ts.map} +1 -1
- package/dist/{run-record-DualPTn2.js → run-record-tg_5H2IK.js} +2 -2
- package/dist/{run-record-DualPTn2.js.map → run-record-tg_5H2IK.js.map} +1 -1
- package/dist/search-ledger-types-D7KDVg-o.d.ts +302 -0
- package/dist/search-ledger-types-D7KDVg-o.d.ts.map +1 -0
- package/dist/{semantic-concept-judge-_3Fp6Q_3.js → semantic-concept-judge-Cnt0TWp0.js} +4 -4
- package/dist/{semantic-concept-judge-_3Fp6Q_3.js.map → semantic-concept-judge-Cnt0TWp0.js.map} +1 -1
- package/dist/{sequential-D_xHrBZe.js → sequential-PgLEqcTJ.js} +3 -3
- package/dist/{sequential-D_xHrBZe.js.map → sequential-PgLEqcTJ.js.map} +1 -1
- package/dist/{server-BZQR1f0C.js → server-DmpcFiVK.js} +2 -2
- package/dist/{server-BZQR1f0C.js.map → server-DmpcFiVK.js.map} +1 -1
- package/dist/{skillopt-optimization-method-DY1aQ-aL.js → skillopt-optimization-method-FSk8b5FQ.js} +8 -7
- package/dist/{skillopt-optimization-method-DY1aQ-aL.js.map → skillopt-optimization-method-FSk8b5FQ.js.map} +1 -1
- package/dist/{statistical-heldout-CpVd6FmY.d.ts → statistical-heldout-Bu4oNOJh.d.ts} +3 -3
- package/dist/{statistical-heldout-CpVd6FmY.d.ts.map → statistical-heldout-Bu4oNOJh.d.ts.map} +1 -1
- package/dist/statistical-heldout-CeWT_W7y.js +256 -0
- package/dist/statistical-heldout-CeWT_W7y.js.map +1 -0
- package/dist/{store-otlp-DAaEbMLf.js → store-otlp-2gYCy-v0.js} +2 -2
- package/dist/{store-otlp-DAaEbMLf.js.map → store-otlp-2gYCy-v0.js.map} +1 -1
- package/dist/{store-tool-spans-BZ8MPKFe.js → store-tool-spans-B1e6Auc2.js} +69 -6
- package/dist/store-tool-spans-B1e6Auc2.js.map +1 -0
- package/dist/{store-tool-spans-Dt-YdAuE.d.ts → store-tool-spans-BrdfPaVh.d.ts} +54 -6
- package/dist/store-tool-spans-BrdfPaVh.d.ts.map +1 -0
- package/dist/{summary-report-IGAOtgPd.js → summary-report-BMvou2oZ.js} +2 -2
- package/dist/{summary-report-IGAOtgPd.js.map → summary-report-BMvou2oZ.js.map} +1 -1
- package/dist/{summary-report-D1h4dlrK.d.ts → summary-report-BzFS5UhZ.d.ts} +2 -2
- package/dist/{summary-report-D1h4dlrK.d.ts.map → summary-report-BzFS5UhZ.d.ts.map} +1 -1
- package/dist/{task-failure-attributes-B_JjrOVo.js → task-failure-attributes-B34TiMBe.js} +2 -2
- package/dist/{task-failure-attributes-B_JjrOVo.js.map → task-failure-attributes-B34TiMBe.js.map} +1 -1
- package/dist/{tool-groups-B2bSNaJB.d.ts → tool-groups-eOOgYGaM.d.ts} +3 -3
- package/dist/tool-groups-eOOgYGaM.d.ts.map +1 -0
- package/dist/{tool-waste-C7MU9u1e.js → tool-waste-nbHPAxei.js} +2 -2
- package/dist/{tool-waste-C7MU9u1e.js.map → tool-waste-nbHPAxei.js.map} +1 -1
- package/dist/trace-repair/index.d.ts +2 -2
- package/dist/trace-repair/index.js +5 -4
- package/dist/trace-repair/index.js.map +1 -1
- package/dist/traces.d.ts +5 -5
- package/dist/traces.js +8 -7
- package/dist/traces.js.map +1 -1
- package/dist/trajectory-replay/index.js +1 -1
- package/dist/{types-CS0qk_Yp.d.ts → types-BTRcIqFr.d.ts} +3 -3
- package/dist/{types-CS0qk_Yp.d.ts.map → types-BTRcIqFr.d.ts.map} +1 -1
- package/dist/{types-D7gEdPoQ.d.ts → types-BjB2wZM9.d.ts} +2 -2
- package/dist/{types-D7gEdPoQ.d.ts.map → types-BjB2wZM9.d.ts.map} +1 -1
- package/dist/types-C8VBbiw6.js +91 -0
- package/dist/types-C8VBbiw6.js.map +1 -0
- package/dist/{types-BvZoPTGa.d.ts → types-c3ZRTbw9.d.ts} +3 -3
- package/dist/{types-BvZoPTGa.d.ts.map → types-c3ZRTbw9.d.ts.map} +1 -1
- package/dist/{types-DQ0e2E7y.js → usage-receipt-Cx3xNeyB.js} +2 -89
- package/dist/usage-receipt-Cx3xNeyB.js.map +1 -0
- package/dist/{verdict-BQ3pCFf8.js → verdict-DhPoXApk.js} +2 -2
- package/dist/{verdict-BQ3pCFf8.js.map → verdict-DhPoXApk.js.map} +1 -1
- package/dist/{verdict-cache-B3eCVQtY.js → verdict-cache-BQEsVy9n.js} +2 -2
- package/dist/{verdict-cache-B3eCVQtY.js.map → verdict-cache-BQEsVy9n.js.map} +1 -1
- package/dist/wire/index.d.ts +1 -1
- package/dist/wire/index.js +1 -1
- package/docs/changes/evaluation-definition-identity.md +27 -0
- package/docs/changes/jev-evidence-review.md +18 -0
- package/docs/changes/jev-product-adoption.md +18 -0
- package/docs/changes/probability-policy-audit.md +81 -0
- package/docs/diagnosis.md +42 -0
- package/docs/jev-behavior-review.md +174 -0
- package/docs/jev-integrations.md +50 -0
- package/docs/jev.md +140 -0
- package/docs/produced-state.md +13 -0
- package/docs/profile-json-identity.md +33 -0
- package/package.json +22 -1
- package/dist/agent-profile-cell-Cv6UA-W_.js.map +0 -1
- package/dist/benchmark-command-NBrqNwgy.js.map +0 -1
- package/dist/campaign-evidence-PYPUftRy.js.map +0 -1
- package/dist/canonical-DPyQ_rpt.js.map +0 -1
- package/dist/define-agent-eval-CneiWyuA.js.map +0 -1
- package/dist/engine-DS1cysJy.d.ts.map +0 -1
- package/dist/exporters-D98GPi1n.d.ts +0 -161
- package/dist/exporters-D98GPi1n.d.ts.map +0 -1
- package/dist/index--_8_Nlpn.d.ts.map +0 -1
- package/dist/index-Bp_6sj3x.d.ts.map +0 -1
- package/dist/kind-factory-DxTzg32u.js.map +0 -1
- package/dist/mint-ySIIkKlV.js.map +0 -1
- package/dist/prime-protocol-6tZTVsWm.js.map +0 -1
- package/dist/run-record-DPa2PeHx.js.map +0 -1
- package/dist/store-tool-spans-BZ8MPKFe.js.map +0 -1
- package/dist/store-tool-spans-Dt-YdAuE.d.ts.map +0 -1
- package/dist/tool-groups-B2bSNaJB.d.ts.map +0 -1
- package/dist/types-DQ0e2E7y.js.map +0 -1
package/dist/analyst/index.d.ts
CHANGED
|
@@ -1,13 +1,14 @@
|
|
|
1
1
|
import { b as CustomTokenPricing, c as CostLedgerHandle } from "../cost-ledger-DbQdN3nO.js";
|
|
2
|
-
import { D as TraceAnalysisStore, _ as ProposalFinding, a as AnalystInputKind, b as makeFinding, c as AnalystRunInputs, d as AnalystSeverity, f as AnalystUsageReceipt, g as ExecutionProbeRequest, h as ExecutionProbeOutcome, i as AnalystFinding, l as AnalystRunResult, m as ExecutionProbe, n as AnalystContext, o as AnalystRequirements, p as EvidenceRef, r as AnalystCost, s as AnalystRunEvent, t as Analyst, u as AnalystRunSummary, v as ProposalFindingOrigin, x as makeProposalFinding, y as computeFindingId, z as TraceAnalystSpan } from "../types-
|
|
2
|
+
import { D as TraceAnalysisStore, _ as ProposalFinding, a as AnalystInputKind, b as makeFinding, c as AnalystRunInputs, d as AnalystSeverity, f as AnalystUsageReceipt, g as ExecutionProbeRequest, h as ExecutionProbeOutcome, i as AnalystFinding, l as AnalystRunResult, m as ExecutionProbe, n as AnalystContext, o as AnalystRequirements, p as EvidenceRef, r as AnalystCost, s as AnalystRunEvent, t as Analyst, u as AnalystRunSummary, v as ProposalFindingOrigin, x as makeProposalFinding, y as computeFindingId, z as TraceAnalystSpan } from "../types-BjB2wZM9.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-CBbLtr2J.js";
|
|
4
|
-
import { A as runAnalystBenchmark, C as AnalystEvidenceResolutionError, D as AnalystLatencyDistribution, E as AnalystIssueExpectation, M as scoreAnalystFindings, O as RunAnalystBenchmarkOptions, S as AnalystEvidenceResolution, T as AnalystFindingScore, V as ExternalOptimizerModelCall, W as ExternalOptimizerModelExecutionObservation, _ as AnalystBenchmarkProvenance, b as AnalystBenchmarkSummary, d as AnalystBenchmarkDatasetRef, f as AnalystBenchmarkDescriptor, g as AnalystBenchmarkOutput, h as AnalystBenchmarkObservation, j as traceStoreEvidenceResolver, k as registryBenchmarkRunner, m as AnalystBenchmarkLabelState, p as AnalystBenchmarkError, q as ExternalOptimizerRunnerCommand, t as AgentProfile, u as AnalystBenchmarkCase, v as AnalystBenchmarkResult, w as AnalystEvidenceResolver, x as AnalystEvidenceExpectation, y as AnalystBenchmarkRunner } from "../agent-profile-
|
|
5
|
-
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-
|
|
6
|
-
import { a as ExactAnalystBudgetPolicy, c as RegistryRunOpts, i as BudgetPolicy, n as AnalystRegistry, o as ExactAnalystRunExecutionError, r as AnalystRegistryOptions, s as ExactRegistryRunOpts, t as AnalystHooks } from "../registry-
|
|
7
|
-
import { a as createTraceAnalyst, c as BehavioralAnalystOptions, i as TraceAnalystDefinition, l as behavioralAnalyst, n as buildDefaultAnalystRegistry, o as renderPriorFindings, r as CreateTraceAnalystOptions, s as runTraceAnalyst, t as DefaultAnalystRegistryOptions, u as deriveEfficiencyFindings } from "../default-registry-
|
|
8
|
-
import { a as TraceAnalystLimits, c as RawAnalystEvidence, d as evidenceRefsFromRawFinding, f as parseRawFinding, i as TraceAnalysisEngineResult, l as RawAnalystFinding, n as TraceAnalysisEngine, o as resolveTraceAnalystLimits, r as TraceAnalysisEngineRequest, s as RAW_FINDING_SCHEMA_PROMPT, t as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, u as RawAnalystFindingSchema } from "../engine-
|
|
9
|
-
import { n as buildTraceToolsForGroup, t as TraceToolGroupName } from "../tool-groups-
|
|
10
|
-
import { A as SemanticConceptJudgeInput, C as DspyRlmTraceEngineOptions, E as createChatTraceEngine, S as renderFindingSubject, T as ChatTraceEngineOptions, _ as FindingSubject, a as FAILURE_MODE_KIND_SPEC, b as findingSubjectGrammarPromptFor, c as emitControlIntegrityFindings, d as FindingsStore, f as PersistedFinding, g as FINDING_SUBJECT_SYNTAX, h as FINDING_SUBJECT_KINDS, i as IMPROVEMENT_KIND_SPEC, j as SemanticConceptJudgeOptions, l as DiffPolicy, m as diffFindings, n as KNOWLEDGE_POISONING_KIND_SPEC, o as CONTROL_INTEGRITY_ANALYST, p as defaultIsMaterial, r as KNOWLEDGE_GAP_KIND_SPEC, s as ControlIntegrityAnalyst, t as DEFAULT_TRACE_ANALYST_KINDS, u as FindingsDiff, v as FindingSubjectKind, w as createDspyRlmTraceEngine, x as parseFindingSubject, y as KIND_EXPECTED_SUBJECTS } from "../index-
|
|
4
|
+
import { A as runAnalystBenchmark, C as AnalystEvidenceResolutionError, D as AnalystLatencyDistribution, E as AnalystIssueExpectation, M as scoreAnalystFindings, O as RunAnalystBenchmarkOptions, S as AnalystEvidenceResolution, T as AnalystFindingScore, V as ExternalOptimizerModelCall, W as ExternalOptimizerModelExecutionObservation, _ as AnalystBenchmarkProvenance, b as AnalystBenchmarkSummary, d as AnalystBenchmarkDatasetRef, f as AnalystBenchmarkDescriptor, g as AnalystBenchmarkOutput, h as AnalystBenchmarkObservation, j as traceStoreEvidenceResolver, k as registryBenchmarkRunner, m as AnalystBenchmarkLabelState, p as AnalystBenchmarkError, q as ExternalOptimizerRunnerCommand, t as AgentProfile, u as AnalystBenchmarkCase, v as AnalystBenchmarkResult, w as AnalystEvidenceResolver, x as AnalystEvidenceExpectation, y as AnalystBenchmarkRunner } from "../agent-profile-uAIQK3Nr.js";
|
|
5
|
+
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-BEHN7tnq.js";
|
|
6
|
+
import { a as ExactAnalystBudgetPolicy, c as RegistryRunOpts, i as BudgetPolicy, n as AnalystRegistry, o as ExactAnalystRunExecutionError, r as AnalystRegistryOptions, s as ExactRegistryRunOpts, t as AnalystHooks } from "../registry-D5Ly3GRV.js";
|
|
7
|
+
import { a as createTraceAnalyst, c as BehavioralAnalystOptions, i as TraceAnalystDefinition, l as behavioralAnalyst, n as buildDefaultAnalystRegistry, o as renderPriorFindings, r as CreateTraceAnalystOptions, s as runTraceAnalyst, t as DefaultAnalystRegistryOptions, u as deriveEfficiencyFindings } from "../default-registry-CXv9kHMt.js";
|
|
8
|
+
import { a as TraceAnalystLimits, c as RawAnalystEvidence, d as evidenceRefsFromRawFinding, f as parseRawFinding, i as TraceAnalysisEngineResult, l as RawAnalystFinding, n as TraceAnalysisEngine, o as resolveTraceAnalystLimits, r as TraceAnalysisEngineRequest, s as RAW_FINDING_SCHEMA_PROMPT, t as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, u as RawAnalystFindingSchema } from "../engine-cXUK-WHS.js";
|
|
9
|
+
import { n as buildTraceToolsForGroup, t as TraceToolGroupName } from "../tool-groups-eOOgYGaM.js";
|
|
10
|
+
import { A as SemanticConceptJudgeInput, C as DspyRlmTraceEngineOptions, E as createChatTraceEngine, S as renderFindingSubject, T as ChatTraceEngineOptions, _ as FindingSubject, a as FAILURE_MODE_KIND_SPEC, b as findingSubjectGrammarPromptFor, c as emitControlIntegrityFindings, d as FindingsStore, f as PersistedFinding, g as FINDING_SUBJECT_SYNTAX, h as FINDING_SUBJECT_KINDS, i as IMPROVEMENT_KIND_SPEC, j as SemanticConceptJudgeOptions, l as DiffPolicy, m as diffFindings, n as KNOWLEDGE_POISONING_KIND_SPEC, o as CONTROL_INTEGRITY_ANALYST, p as defaultIsMaterial, r as KNOWLEDGE_GAP_KIND_SPEC, s as ControlIntegrityAnalyst, t as DEFAULT_TRACE_ANALYST_KINDS, u as FindingsDiff, v as FindingSubjectKind, w as createDspyRlmTraceEngine, x as parseFindingSubject, y as KIND_EXPECTED_SUBJECTS } from "../index-C8_iU8KL.js";
|
|
11
|
+
import { A as ReplyRowDecoded, C as primeProtocolSha256, D as DecodedReply, E as runPrimeExchange, M as decodeReplyRows, O as ReplyContract, S as normalizePrimeUsage, T as projectPrimeTrajectory, _ as buildPrimePrompt, a as PrimeProjectionOutcome, b as extractPrimeJsonObject, c as PrimeProtocolIdentity, d as PrimeRepairPromptSpec, f as PrimeRepairState, g as analystUsageReceiptFromPrimeUsage, h as PrimeTurnRecord, i as PrimeProjectionDelivery, j as ReplyRowRejection, k as ReplyEnvelope, 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 primeReplyDefect, x as mergePrimeRawUsage, y as emptyPrimeRawUsage } from "../prime-protocol-CDBcL9NM.js";
|
|
11
12
|
import { i as nodeHttpPrimeBridgeTransport, n as PrimeBridgeTransportRequest, r as PrimeBridgeTransportResult, t as PrimeBridgeTransport } from "../prime-bridge-transport-6feEglLf.js";
|
|
12
13
|
import { z } from "zod";
|
|
13
14
|
//#region src/analyst/adapters.d.ts
|
|
@@ -454,91 +455,6 @@ declare function preparePublicAnalystBenchmark(options: {
|
|
|
454
455
|
seed: number;
|
|
455
456
|
}): Promise<PreparedPublicAnalystBenchmark>;
|
|
456
457
|
//#endregion
|
|
457
|
-
//#region src/analyst/reply-contract.d.ts
|
|
458
|
-
/**
|
|
459
|
-
* The reply grammar an analyst arm holds a model to, independent of transport.
|
|
460
|
-
*
|
|
461
|
-
* One contract serves every arm shape: the inline bridge protocol
|
|
462
|
-
* (`runPrimeExchange`) reads the base fields, and one-shot JSON arms
|
|
463
|
-
* additionally use the strict-envelope and all-rejected knobs. `PrimeReplyContract`
|
|
464
|
-
* in ./prime-protocol is a type alias of this contract, so a consumer written
|
|
465
|
-
* against the prime protocol names the same grammar object.
|
|
466
|
-
*/
|
|
467
|
-
/** Verdict of decoding one raw reply row. */
|
|
468
|
-
type ReplyRowDecoded<TRow> = {
|
|
469
|
-
ok: true;
|
|
470
|
-
row: TRow;
|
|
471
|
-
} | {
|
|
472
|
-
ok: false;
|
|
473
|
-
reason: string;
|
|
474
|
-
};
|
|
475
|
-
/** A strictly parsed reply envelope: the row array plus retained extras. */
|
|
476
|
-
interface ReplyEnvelope {
|
|
477
|
-
rows: readonly unknown[];
|
|
478
|
-
/** Envelope fields beside the rows the arm retains (e.g. a `report` string). */
|
|
479
|
-
extras: Record<string, unknown>;
|
|
480
|
-
}
|
|
481
|
-
interface ReplyContract<TRow> {
|
|
482
|
-
/** Name of the reply's row array, e.g. `blocks` or `findings`. */
|
|
483
|
-
rowsField: string;
|
|
484
|
-
/** Row-grammar lines spliced into the first-turn prompt. */
|
|
485
|
-
contractLines: readonly string[];
|
|
486
|
-
/**
|
|
487
|
-
* The same grammar restated for the repair turn, which never carries the
|
|
488
|
-
* evidence. Empty when the arm declares zero repair turns.
|
|
489
|
-
*/
|
|
490
|
-
repairContractLines: readonly string[];
|
|
491
|
-
/**
|
|
492
|
-
* Validate and convert one raw row in a single pass. One function rather than
|
|
493
|
-
* a separate validator and constructor, so the two can never disagree about
|
|
494
|
-
* what a valid row is.
|
|
495
|
-
*/
|
|
496
|
-
decodeRow(row: unknown, index: number): ReplyRowDecoded<TRow>;
|
|
497
|
-
/**
|
|
498
|
-
* Cap on ACCEPTED rows. The cap is applied after decoding, so malformed rows
|
|
499
|
-
* never consume an accepted slot; the surplus is reported as `overflow`.
|
|
500
|
-
* Omit when the consumer's own expansion owns the cap and records the drop.
|
|
501
|
-
*/
|
|
502
|
-
maxRows?: number;
|
|
503
|
-
/**
|
|
504
|
-
* Strict envelope parse for one-shot JSON arms. Throws on an envelope defect
|
|
505
|
-
* (missing report, unknown key, over-cap row count). When absent, rows are
|
|
506
|
-
* read from `rowsField` on the parsed reply object.
|
|
507
|
-
*/
|
|
508
|
-
parseEnvelope?(value: unknown): ReplyEnvelope;
|
|
509
|
-
/**
|
|
510
|
-
* What a reply whose EVERY reported row was rejected means. `fail` throws a
|
|
511
|
-
* validation error carrying each rejection; `accept-empty` (the default)
|
|
512
|
-
* keeps the honest empty result beside the recorded rejections.
|
|
513
|
-
*/
|
|
514
|
-
whenAllRowsRejected?: 'fail' | 'accept-empty';
|
|
515
|
-
/**
|
|
516
|
-
* Message prefix for the `fail` case, in the contract's own row vocabulary
|
|
517
|
-
* (e.g. "every reported failure block was malformed").
|
|
518
|
-
*/
|
|
519
|
-
allRejectedMessage?: string;
|
|
520
|
-
}
|
|
521
|
-
/** One rejected row, indexed as the model reported it. */
|
|
522
|
-
interface ReplyRowRejection {
|
|
523
|
-
index: number;
|
|
524
|
-
reason: string;
|
|
525
|
-
}
|
|
526
|
-
interface DecodedReply<TRow> {
|
|
527
|
-
rows: TRow[];
|
|
528
|
-
/** Envelope fields beside the rows (empty when the contract has no envelope). */
|
|
529
|
-
extras: Record<string, unknown>;
|
|
530
|
-
rejected: ReplyRowRejection[];
|
|
531
|
-
/** Rows the model reported, before decoding or the count cap. */
|
|
532
|
-
reportedRows: number;
|
|
533
|
-
/** Valid rows dropped by `maxRows`. */
|
|
534
|
-
overflow: number;
|
|
535
|
-
}
|
|
536
|
-
/**
|
|
537
|
-
* Decode a parsed reply value under a contract: strict envelope when declared,
|
|
538
|
-
* then per-row decoding, then the all-rejected policy.
|
|
539
|
-
*/
|
|
540
|
-
declare function decodeReplyRows<TRow>(contract: ReplyContract<TRow>, value: unknown): DecodedReply<TRow>;
|
|
541
|
-
//#endregion
|
|
542
458
|
//#region src/analyst/definition.d.ts
|
|
543
459
|
/**
|
|
544
460
|
* The slice of the canonical `AgentProfile` an analyst definition carries:
|
|
@@ -1265,11 +1181,11 @@ declare function runAnalystBenchmarkCommand(argv: readonly string[], env?: NodeJ
|
|
|
1265
1181
|
declare const ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM = "sha256-canonical-source-manifest";
|
|
1266
1182
|
declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM = "sha256-canonical-file-manifest";
|
|
1267
1183
|
declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES: readonly string[];
|
|
1268
|
-
declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 = "
|
|
1184
|
+
declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 = "3c2fc1a6a3cb07b2c285736ef071f0d2cf02aaffe02d827bd69d3051f65262a9";
|
|
1269
1185
|
declare const ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256 = "1e03f2daed356d60316aabefb407ec1e437ac94d408d61eea4ae096e9c6fbb5b";
|
|
1270
1186
|
declare const ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256 = "4dba263b6256a30d56c7fdb2d992d3a953c0035d731f359b704db806f68f75ac";
|
|
1271
1187
|
declare const ANALYST_BENCHMARK_IMPLEMENTATION_FILES: readonly string[];
|
|
1272
|
-
declare const ANALYST_BENCHMARK_IMPLEMENTATION_SHA256 = "
|
|
1188
|
+
declare const ANALYST_BENCHMARK_IMPLEMENTATION_SHA256 = "dd54cb34dafbb0cab915e4223a387fbd1093ad8daa44b1230979bdeca71bf5ce";
|
|
1273
1189
|
declare function analystBenchmarkImplementationDigest(): string;
|
|
1274
1190
|
declare function analystBenchmarkDependencyLockDigest(): string;
|
|
1275
1191
|
//#endregion
|
|
@@ -1429,247 +1345,6 @@ declare function stripCodeFences(text: string): string;
|
|
|
1429
1345
|
*/
|
|
1430
1346
|
declare function coerceJson(text: string): unknown;
|
|
1431
1347
|
//#endregion
|
|
1432
|
-
//#region src/analyst/prime-protocol.d.ts
|
|
1433
|
-
/**
|
|
1434
|
-
* The prime analyst protocol: a one-shot RLM reached through an
|
|
1435
|
-
* OpenAI-compatible bridge, given the whole trajectory inline because it has no
|
|
1436
|
-
* REPL and no trace tools, answering in one fenced JSON block of short strings.
|
|
1437
|
-
*
|
|
1438
|
-
* This module owns every part of that protocol that does not depend on what the
|
|
1439
|
-
* rows MEAN: prompt and contract composition, the bounded repair turn, reply
|
|
1440
|
-
* extraction, row decoding order, the render-measure-fall-back-fail-loud
|
|
1441
|
-
* projection ladder, usage normalization, and the protocol identity digest.
|
|
1442
|
-
*
|
|
1443
|
-
* The cut: the core speaks RAW ROWS and never names a finding type. It knows a
|
|
1444
|
-
* prime reply carries an `answer` string plus an array of rows under a named
|
|
1445
|
-
* field, and that each row is decoded by a caller-supplied function. Consumers
|
|
1446
|
-
* — the CodeTraceBench benchmark runner here, a span-grounded trace analyzer
|
|
1447
|
-
* elsewhere — map decoded rows into their own finding shape on the way out. So
|
|
1448
|
-
* nothing in this file may import a block type, a finding type, or a trace
|
|
1449
|
-
* store, and no consumer needs another copy of the protocol.
|
|
1450
|
-
*/
|
|
1451
|
-
/**
|
|
1452
|
-
* The prime reply grammar is the general analyst reply contract: these aliases
|
|
1453
|
-
* keep every prime-protocol consumer compiling against the shared type in
|
|
1454
|
-
* ./reply-contract. The exchange below reads the base fields (`rowsField`,
|
|
1455
|
-
* contract lines, `decodeRow`, `maxRows`); the strict-envelope knobs serve
|
|
1456
|
-
* one-shot JSON arms and are inert here.
|
|
1457
|
-
*/
|
|
1458
|
-
type PrimeRowDecoded<TRow> = ReplyRowDecoded<TRow>;
|
|
1459
|
-
type PrimeReplyContract<TRow> = ReplyContract<TRow>;
|
|
1460
|
-
interface PrimePromptSpec {
|
|
1461
|
-
question: string;
|
|
1462
|
-
/** Task definition spliced under a `TASK DEFINITION:` heading. Omit when the question is the whole task. */
|
|
1463
|
-
taskDefinition?: string;
|
|
1464
|
-
contractLines: readonly string[];
|
|
1465
|
-
/** Line introducing the trajectory, e.g. `TRAJECTORY (trace_id X; 12 assistant step spans; ...):`. */
|
|
1466
|
-
trajectoryHeader: string;
|
|
1467
|
-
renderedTrajectory: string;
|
|
1468
|
-
/** Material appended after the trajectory, e.g. the final-verification spans. */
|
|
1469
|
-
trailer?: string;
|
|
1470
|
-
}
|
|
1471
|
-
declare function buildPrimePrompt(spec: PrimePromptSpec): string;
|
|
1472
|
-
interface PrimeRepairPromptSpec {
|
|
1473
|
-
defect: string;
|
|
1474
|
-
previousReply: string;
|
|
1475
|
-
repairContractLines: readonly string[];
|
|
1476
|
-
}
|
|
1477
|
-
/** Carries the malformed reply and the contract — never the trajectory. */
|
|
1478
|
-
declare function buildPrimeRepairPrompt(spec: PrimeRepairPromptSpec): string;
|
|
1479
|
-
/**
|
|
1480
|
-
* Recover the reply's JSON object.
|
|
1481
|
-
*
|
|
1482
|
-
* Distinct from `extractJsonPayload` in ../llm-client, which serves a response
|
|
1483
|
-
* that DECLARES a JSON root and therefore must not scan onward. A prime reply
|
|
1484
|
-
* is prose plus a fenced block, and when the model emits several fences the
|
|
1485
|
-
* last one is its answer — so fences are scanned in reverse, and only then is a
|
|
1486
|
-
* brace-to-brace slice tried.
|
|
1487
|
-
*/
|
|
1488
|
-
declare function extractPrimeJsonObject(text: string): Record<string, unknown> | null;
|
|
1489
|
-
/** Why the reply cannot be read as a prime answer, or null when it can. */
|
|
1490
|
-
declare function primeReplyDefect(parsed: Record<string, unknown> | null, rowsField: string): string | null;
|
|
1491
|
-
/**
|
|
1492
|
-
* Token accounting exactly as the bridge reported it. Lossless on purpose: a
|
|
1493
|
-
* side the bridge did not report stays null rather than becoming a zero, and a
|
|
1494
|
-
* side it DID report survives even when its partner is missing.
|
|
1495
|
-
*/
|
|
1496
|
-
interface PrimeRawUsage {
|
|
1497
|
-
/**
|
|
1498
|
-
* Model-usage records the bridge observed. NOT a count of bridge round trips:
|
|
1499
|
-
* one prime turn routes several internal model calls, so a single turn
|
|
1500
|
-
* typically reports 3-5.
|
|
1501
|
-
*/
|
|
1502
|
-
calls: number | null;
|
|
1503
|
-
/** Prompt tokens the bridge reported; null when it reported none. */
|
|
1504
|
-
inputTokens: number | null;
|
|
1505
|
-
/** Completion tokens the bridge reported; null when it reported none. */
|
|
1506
|
-
outputTokens: number | null;
|
|
1507
|
-
/**
|
|
1508
|
-
* True when the bridge DERIVED the counts (from character lengths) because
|
|
1509
|
-
* the backend reported no usage of its own. Pricing derived counts produces a
|
|
1510
|
-
* guess about a guess, which a reader must be able to tell apart from a
|
|
1511
|
-
* rate-estimated cost over exact tokens.
|
|
1512
|
-
*/
|
|
1513
|
-
bridgeEstimated: boolean;
|
|
1514
|
-
}
|
|
1515
|
-
declare function emptyPrimeRawUsage(): PrimeRawUsage;
|
|
1516
|
-
/** Read the bridge's OpenAI-shaped `usage` object. */
|
|
1517
|
-
declare function normalizePrimeUsage(raw: unknown): PrimeRawUsage;
|
|
1518
|
-
/**
|
|
1519
|
-
* Sum two turns. Each side poisons independently: two turns that both report
|
|
1520
|
-
* input and neither report output yield a real input total beside a null
|
|
1521
|
-
* output, because discarding a measured count is as wrong as inventing one.
|
|
1522
|
-
*/
|
|
1523
|
-
declare function mergePrimeRawUsage(a: PrimeRawUsage, b: PrimeRawUsage): PrimeRawUsage;
|
|
1524
|
-
/**
|
|
1525
|
-
* Bind raw prime usage to agent-eval's typed receipt.
|
|
1526
|
-
*
|
|
1527
|
-
* `RunTokenUsage.input` and `.output` are non-nullable, so a one-sided count
|
|
1528
|
-
* cannot round-trip through `tokens` without writing a zero nobody measured.
|
|
1529
|
-
* The complete-accounting field therefore stays null, the reported side is
|
|
1530
|
-
* carried verbatim in `partialTokens`, and its price becomes the receipt's
|
|
1531
|
-
* `knownCostUsd` lower bound.
|
|
1532
|
-
*
|
|
1533
|
-
* Only agent-eval calls this; consumers with no pricing table read
|
|
1534
|
-
* `PrimeRawUsage` directly.
|
|
1535
|
-
*/
|
|
1536
|
-
declare function analystUsageReceiptFromPrimeUsage(usage: PrimeRawUsage, pricing: CustomTokenPricing): AnalystUsageReceipt;
|
|
1537
|
-
/**
|
|
1538
|
-
* Why an exchange ended without a readable reply. `deadline` and `aborted` are
|
|
1539
|
-
* separate members from `transport` on purpose: a model that ran past its
|
|
1540
|
-
* deadline, a caller who cancelled, and a bridge that refused the connection
|
|
1541
|
-
* demand three different responses, and a record that conflates them cannot be
|
|
1542
|
-
* acted on.
|
|
1543
|
-
*/
|
|
1544
|
-
type PrimeFailure = {
|
|
1545
|
-
kind: 'transport' | 'unparseable-json' | 'no-content' | 'deadline' | 'malformed-reply';
|
|
1546
|
-
message: string;
|
|
1547
|
-
} | {
|
|
1548
|
-
kind: 'http-status';
|
|
1549
|
-
message: string;
|
|
1550
|
-
status: number;
|
|
1551
|
-
bodySnippet: string;
|
|
1552
|
-
} | {
|
|
1553
|
-
kind: 'aborted';
|
|
1554
|
-
message: string;
|
|
1555
|
-
cause: unknown;
|
|
1556
|
-
};
|
|
1557
|
-
interface PrimeRepairState {
|
|
1558
|
-
attempted: boolean;
|
|
1559
|
-
/** Null when no repair turn ran, or when the repair turn never returned a reply. */
|
|
1560
|
-
succeeded: boolean | null;
|
|
1561
|
-
}
|
|
1562
|
-
/** One completed bridge turn. A turn that failed produces no record. */
|
|
1563
|
-
interface PrimeTurnRecord {
|
|
1564
|
-
turn: 'first' | 'repair';
|
|
1565
|
-
usage: PrimeRawUsage;
|
|
1566
|
-
/** The reply's `usage` object verbatim; null when the bridge sent none. */
|
|
1567
|
-
rawUsage: unknown;
|
|
1568
|
-
}
|
|
1569
|
-
interface PrimeRejectedRow {
|
|
1570
|
-
index: number;
|
|
1571
|
-
reason: string;
|
|
1572
|
-
}
|
|
1573
|
-
type PrimeExchangeOutcome<TRow> = {
|
|
1574
|
-
ok: true;
|
|
1575
|
-
/** The reply's `answer` field; null when absent or not a string. */
|
|
1576
|
-
answer: string | null;
|
|
1577
|
-
rows: TRow[];
|
|
1578
|
-
rejected: PrimeRejectedRow[];
|
|
1579
|
-
/** Rows the model reported, before decoding or the count cap. */
|
|
1580
|
-
reportedRows: number;
|
|
1581
|
-
/** Valid rows dropped by `contract.maxRows`. */
|
|
1582
|
-
overflow: number;
|
|
1583
|
-
usage: PrimeRawUsage;
|
|
1584
|
-
turns: PrimeTurnRecord[];
|
|
1585
|
-
repair: PrimeRepairState;
|
|
1586
|
-
reply: string;
|
|
1587
|
-
} | {
|
|
1588
|
-
ok: false;
|
|
1589
|
-
failure: PrimeFailure;
|
|
1590
|
-
usage: PrimeRawUsage;
|
|
1591
|
-
turns: PrimeTurnRecord[];
|
|
1592
|
-
repair: PrimeRepairState;
|
|
1593
|
-
/** The last reply received, when one arrived — the diagnostic artifact for a malformed case. */
|
|
1594
|
-
reply?: string;
|
|
1595
|
-
};
|
|
1596
|
-
interface PrimeExchangeOptions<TRow> {
|
|
1597
|
-
contract: PrimeReplyContract<TRow>;
|
|
1598
|
-
prompt: string;
|
|
1599
|
-
transport: PrimeBridgeTransport;
|
|
1600
|
-
/** Full endpoint, e.g. `http://localhost:4181/v1/chat/completions`. */
|
|
1601
|
-
url: string;
|
|
1602
|
-
model: string;
|
|
1603
|
-
/** Deadline for ONE turn. Prime analyses routinely exceed five minutes. */
|
|
1604
|
-
timeoutMs: number;
|
|
1605
|
-
/** Whether a structurally malformed reply gets one bounded repair turn. */
|
|
1606
|
-
repair: boolean;
|
|
1607
|
-
signal?: AbortSignal;
|
|
1608
|
-
}
|
|
1609
|
-
/**
|
|
1610
|
-
* Run the protocol: one call, one bounded repair turn on a structurally
|
|
1611
|
-
* malformed reply, then decode. Zero valid rows from a well-formed reply is an
|
|
1612
|
-
* honest null, not a failure.
|
|
1613
|
-
*/
|
|
1614
|
-
declare function runPrimeExchange<TRow>(options: PrimeExchangeOptions<TRow>): Promise<PrimeExchangeOutcome<TRow>>;
|
|
1615
|
-
/**
|
|
1616
|
-
* Where the inlined trajectory comes from. The protocol needs only two moves —
|
|
1617
|
-
* fetch it all, or fetch a reduced version — so no store, artifact path, or
|
|
1618
|
-
* span type enters the shared signature.
|
|
1619
|
-
*/
|
|
1620
|
-
interface PrimeProjectionSource<TItem> {
|
|
1621
|
-
/** The full projection, or null when the source declares itself oversized. */
|
|
1622
|
-
full(): Promise<readonly TItem[] | null>;
|
|
1623
|
-
/** A reduced projection of the same material. */
|
|
1624
|
-
capped(): Promise<readonly TItem[]>;
|
|
1625
|
-
/** How `capped` reduces, named for the failure message, e.g. `per-attribute cap 1200`. */
|
|
1626
|
-
cappedDescription: string;
|
|
1627
|
-
}
|
|
1628
|
-
interface PrimeProjectionDelivery {
|
|
1629
|
-
mode: 'inline-json';
|
|
1630
|
-
fetch: 'full' | 'capped';
|
|
1631
|
-
renderedChars: number;
|
|
1632
|
-
}
|
|
1633
|
-
type PrimeProjectionOutcome<TItem> = {
|
|
1634
|
-
ok: true;
|
|
1635
|
-
items: readonly TItem[];
|
|
1636
|
-
rendered: string;
|
|
1637
|
-
delivery: PrimeProjectionDelivery;
|
|
1638
|
-
} | {
|
|
1639
|
-
ok: false;
|
|
1640
|
-
reason: string;
|
|
1641
|
-
renderedChars: number;
|
|
1642
|
-
};
|
|
1643
|
-
/**
|
|
1644
|
-
* Render, measure, fall back to the capped projection, re-measure, fail loud.
|
|
1645
|
-
*
|
|
1646
|
-
* Inline is the only delivery prime has, so an oversized trajectory is a
|
|
1647
|
-
* refusal rather than a silent truncation: dropping spans would understate the
|
|
1648
|
-
* trajectory and the analyst would answer a question about a different run.
|
|
1649
|
-
*/
|
|
1650
|
-
declare function projectPrimeTrajectory<TItem>(source: PrimeProjectionSource<TItem>, limits: {
|
|
1651
|
-
maxInlineChars: number;
|
|
1652
|
-
}): Promise<PrimeProjectionOutcome<TItem>>;
|
|
1653
|
-
interface PrimeProtocolIdentity {
|
|
1654
|
-
question: string;
|
|
1655
|
-
taskDefinition?: string;
|
|
1656
|
-
contractLines: readonly string[];
|
|
1657
|
-
repairContractLines: readonly string[];
|
|
1658
|
-
/** Numeric limits the contract states, e.g. row and width caps. */
|
|
1659
|
-
limits: Readonly<Record<string, number>>;
|
|
1660
|
-
}
|
|
1661
|
-
/**
|
|
1662
|
-
* Digest of everything a consumer can send to the bridge under the prime
|
|
1663
|
-
* protocol, recorded per observation so a prime result names the exact contract
|
|
1664
|
-
* that produced it.
|
|
1665
|
-
*
|
|
1666
|
-
* Computed over the ACTUALLY composed contract, so two consumers that both
|
|
1667
|
-
* stamp `analyst_id: 'prime'` while asking materially different questions get
|
|
1668
|
-
* different digests by construction. That is what makes 'prime' a reproducible
|
|
1669
|
-
* claim rather than a label.
|
|
1670
|
-
*/
|
|
1671
|
-
declare function primeProtocolSha256(identity: PrimeProtocolIdentity): string;
|
|
1672
|
-
//#endregion
|
|
1673
1348
|
//#region src/analyst/proposal-findings.d.ts
|
|
1674
1349
|
/** True when a finding names a source candidate generation may learn from. */
|
|
1675
1350
|
declare function isProposalFinding(finding: unknown): finding is ProposalFinding;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/analyst/adapters.ts","../../src/analyst/benchmark-agentrx-calibration.ts","../../src/analyst/benchmark-dataset-types.ts","../../src/analyst/benchmark-dataset-agentrx.ts","../../src/analyst/benchmark-dataset-codetrace.ts","../../src/analyst/benchmark-dataset-utils.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/reply-contract.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/benchmark-instructions-override.ts","../../src/analyst/benchmark-report.ts","../../src/analyst/benchmark-summary.ts","../../src/analyst/bind.ts","../../src/analyst/define.ts","../../src/analyst/kinds/skill-usage.ts","../../src/analyst/parse-tolerant.ts","../../src/analyst/prime-protocol.ts","../../src/analyst/proposal-findings.ts"],"mappings":";;;;;;;;;;;;;UA8CiB;EACf;EACA;;EAEA,SAAS,KAAK;;EAEd;;iBAGc,kCACd,MAAM,kCACL,QAAQ;;;cCtDE;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;;iBAGK,4BACd,QAAQ,wBACR,2BACC;iBAkBa,iCAAiC,SAAS;;;KCtD9C;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;;;iBCtGY,qBAAqB,QACnC,KAAK,YACL,OAAO,QACP,UAAS,8BACR,qBAAqB;;iBAsGR,6BACd,mBAAmB,YACnB,iBACA,UAAS,mCACR;iBA6Ea,yBAAyB;;iBAoNzB,iBAAiB;;;iBC/YjB,mBAAmB,QACjC,KAAK,mBACL,OAAO,QACP,UAAS,4BACR,qBAAqB;;iBAwFR,gCACd,2BACA,aAAa,uBACb,UAAS,qCACR;;;iBClHa,wBAAwB;;;KCA5B;UAEK;EACf;EACA;EACA,QAAQ;;UAGO;EACf,QAAQ;EACR;EAKA;IAAe;IAAe;;EAC9B,SAAS;EACT;EACA;EACA;EACA;;UAGe;EACf;EACA;;iBA8Ec,yBACd,gBAAgB,2BACf;;;cC/FU;KAED;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;;;iBAatB,mCAAmC;EACvD;EACA,KAAK;EACL;IACE,QAAQ;iBAwHI,kCACd,kBACA,iBACA,WAAW,6BACX;;;KC7KU;;;;;;;;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;;iBAGO,8BAA8B,uBAAuB;iBAiB/C,6BAA6B;EACjD,SAAS;EACT;EACA,mBAAmB;EACnB;EACA,OAAO;EACP,SAAS;IACP;EACF,UAAU;EACV,aAAa;;EAEb,aAAa;;;;;;;;;iBA4MO,6BAA6B;EACjD;EACA,iBAAiB;EACjB,OAAO;EACP;EACA;EACA,SAAS;IACP;EACF,UAAU;EACV,aAAa;EACb,YAAY;;;;iBCrQQ,wBACpB,eACC,QAAQ,MAAM;iBA2BD,0BACd,SAAS,+BACT,eAAe,2BACf;EAAW;EAAe;IACzB,MAAM;iBAwBO,6BACd,SAAS,+BACT,eAAe,4BACd;iBAyCa,+BACd,SAAS,+BACT,iBAAiB,2BACjB,mBAAmB,2BACnB,eACC;iBAcmB,8BAA8B;EAClD,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;IACE,QAAQ;;;;;;;;;;;;;KCpKA,gBAAgB;EAAU;EAAU,KAAK;;EAAW;EAAW;;;UAG1D;EACf;;EAEA,QAAQ;;UAGO,cAAc;;EAE7B;;EAEA;;;;;EAKA;;;;;;EAMA,UAAU,cAAc,gBAAgB,gBAAgB;;;;;;EAMxD;;;;;;EAMA,eAAe,iBAAiB;;;;;;EAMhC;;;;;EAKA;;;UAIe;EACf;EACA;;UAGe,aAAa;EAC5B,MAAM;;EAEN,QAAQ;EACR,UAAU;;EAEV;;EAEA;;;;;;iBAmCc,gBAAgB,MAC9B,UAAU,cAAc,OACxB,iBACC,aAAa;;;;;;;;KCrEJ,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;;;;;;;;;cAUjD,mCAAmC;;;;;;;iBAUhC,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;;;;;;;;;iBAUjB,6BACd,aAAa,cAAc,gDAC1B;;;;;;;cCzUU;;;;cAKA;cAMA;;iBA8GG,4BAA4B,SAAS;;iBASrC,+BAA+B,SAAS;;;;iBAcxC,8BAA8B,SAAS;;;;;;;;;;;;;UC5EtC;;EAEf;;EAEA;;EAEA;;;iBAIc,8BACd,SAAS,+BACT,MAAM,6BACL,kBAAkB;;iBAsEL,kCACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;cAwhBpB,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;;;;;;;;;;;;;UCjtBrD;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,cAAc;;;iBAIA,2BACd,SAAS,+BACT,MAAM,0BACL,kBAAkB,mBAAmB;;iBA+GxB,+BACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;;iBAwBV,gBAAgB,QAAQ,oCAAoC;;;UCtI3D;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA,UAAU;;EAEV,YAAY;;;;;;iBA8EE;UAIC;;EAEf;;EAEA;;;;;;iBAOc,gCACd,MAAM,+BACL,kBAAkB;;iBAyDL,2BACd,SAAS,8BACR,uBAAuB;;;;;;;;;;cC5OpB;WACQ;WACC;WACK;WACd;WACkB;WACJ;WACO;WACD;WACJ;WACK;WACd;WACJ;WACM;WACC;WACG;WACH;WACI;WACT;;KAGC,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;;iBASK,sBACd,QAAQ,wBACR;EACE;EACA;EACA;EACA;EACA;IAED;;;UC3Fc;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;;iBAGK,8BACd,QAAQ,yBACP;iBAca,mCAAmC,SAAS;;;UCpC3C;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;;cAGW;cACA;cACA;cACA;;;iBC5KS,6BACpB,eACC,QAAQ;;;UCuEM;EACf,uBACE,SAAS,+BACT,QAAQ,wCACL,uBAAuB;EAC5B,2BACE,mBACA;IACE;IACA,aAAa,SAAS,OAAO;QAE5B,QAAQ;;;;;;;;;KAUH;;UAGK;EACf;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;;iBAKoB,2BACpB,yBACA,MAAK,OAAO,YACZ,eAAc,sCACb;;;cCzJU;cAEA;cAEA;cAOA;cAYA;cAGA;cAGA;cA+GA;iBAGG;iBAIA;;;;iBC1IA,oCAAoC,eAAe;;iBAQnD,gCAAgC,eAAe;;;;;;;;;;;iBAyB/C,+BACd,SAAS,+BACT,WAAW,KAAK;;;iBCzCF,+BACd,QAAQ,wBACR,uBAAsB;;;iBCIR,gCACd,kBACA,uBAAuB,gCACtB;;;;KCcS;;WAGG;WACA;WACA;WACA,YAAY;WACZ,UAAU;;;WAIV;WACA,QAAQ;;;;;;iBAOP,YAAY,MAAM,aAAa,QAC7C,YAAY,kBAAkB,MAAM,aAAa,SACjD,YAAY,0BACX,uBAAuB;;;;;;;;;;iBCpCV,mBAAmB,YAAY,yBAAyB;UAkBvD;EACf;EACA;EACA;EACA,MAAM;EACN,SAAS,QAAQ;;UAGF,wCAAwC;;EAEvD,iBAAiB,SAAS;;;iBAIZ,oBACd,SAAS,kCACR,oBAAoB;iBACP,oBACd,SAAS,6BACR,QAAQ;;;KCpBC;;UAGK;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;;;iBA2Ec,sBAAsB,QAAQ,uBAAuB;;iBAiHrD,uBACd,QAAQ,kBACR,qBACC;cA2HG,6BAA6B,oBAAoB;WAC5C;WACA;WAEA;WACA;IAAS;IAAgC;;WACzC;WACA;aACP;aACA;aACA;;EAGI,QAAQ,OAAO,kBAAkB,KAAK,iBAAiB,QAAQ;;cAS1D,qBAAmB;;;;;;;;;;;;;;;iBC5YhB,gBAAgB;;;;;iBAgBhB,WAAW;;;;;;;;;;;;;;;;;;;;;;;;;;;;KCKf,gBAAgB,QAAQ,gBAAgB;KAExC,mBAAmB,QAAQ,cAAc;UAIpC;EACf;;EAEA;EACA;;EAEA;EACA;;EAEA;;iBAGc,iBAAiB,MAAM;UAatB;EACf;EACA;EACA;;;iBAIc,uBAAuB,MAAM;;;;;;;;;;iBAuB7B,uBAAuB,eAAe;;iBAkBtC,iBACd,QAAQ,gCACR;;;;;;UAyBe;;;;;;EAMf;;EAEA;;EAEA;;;;;;;EAOA;;iBAGc,sBAAsB;;iBAKtB,oBAAoB,eAAe;;;;;;iBAgBnC,mBAAmB,GAAG,eAAe,GAAG,gBAAgB;;;;;;;;;;;;;iBA6BxD,kCACd,OAAO,eACP,SAAS,qBACR;;;;;;;;KAqCS;EAEN;EACA;;EAEA;EAAqB;EAAiB;EAAgB;;EACtD;EAAiB;EAAiB;;UAEvB;EACf;;EAEA;;;UAIe;EACf;EACA,OAAO;;EAEP;;UAGe;EACf;EACA;;KAGU,qBAAqB;EAE3B;;EAEA;EACA,MAAM;EACN,UAAU;;EAEV;;EAEA;EACA,OAAO;EACP,OAAO;EACP,QAAQ;EACR;;EAGA;EACA,SAAS;EACT,OAAO;EACP,OAAO;EACP,QAAQ;;EAER;;UAGW,qBAAqB;EACpC,UAAU,mBAAmB;EAC7B;EACA,WAAW;;EAEX;EACA;;EAEA;;EAEA;EACA,SAAS;;;;;;;iBAQW,iBAAiB,MACrC,SAAS,qBAAqB,QAC7B,QAAQ,qBAAqB;;;;;;UAyLf,sBAAsB;;EAErC,QAAQ,iBAAiB;;EAEzB,UAAU,iBAAiB;;EAE3B;;UAGe;EACf;EACA;EACA;;KAGU,uBAAuB;EAE7B;EACA,gBAAgB;EAChB;EACA,UAAU;;EAEV;EAAW;EAAgB;;;;;;;;;iBASX,uBAAuB,OAC3C,QAAQ,sBAAsB,QAC9B;EAAU;IACT,QAAQ,uBAAuB;UA8BjB;EACf;EACA;EACA;EACA;;EAEA,QAAQ,SAAS;;;;;;;;;;;;iBAaH,oBAAoB,UAAU;;;;iBC/iB9B,kBAAkB,mBAAmB,WAAW;;;;;;iBAShD,uBACd,mBACA,mBACC,cAAc"}
|
|
1
|
+
{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/analyst/adapters.ts","../../src/analyst/benchmark-agentrx-calibration.ts","../../src/analyst/benchmark-dataset-types.ts","../../src/analyst/benchmark-dataset-agentrx.ts","../../src/analyst/benchmark-dataset-codetrace.ts","../../src/analyst/benchmark-dataset-utils.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/benchmark-instructions-override.ts","../../src/analyst/benchmark-report.ts","../../src/analyst/benchmark-summary.ts","../../src/analyst/bind.ts","../../src/analyst/define.ts","../../src/analyst/kinds/skill-usage.ts","../../src/analyst/parse-tolerant.ts","../../src/analyst/proposal-findings.ts"],"mappings":";;;;;;;;;;;;;;UA8CiB;EACf;EACA;;EAEA,SAAS,KAAK;;EAEd;;iBAGc,kCACd,MAAM,kCACL,QAAQ;;;cCtDE;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;;iBAGK,4BACd,QAAQ,wBACR,2BACC;iBAkBa,iCAAiC,SAAS;;;KCtD9C;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;;;iBCtGY,qBAAqB,QACnC,KAAK,YACL,OAAO,QACP,UAAS,8BACR,qBAAqB;;iBAsGR,6BACd,mBAAmB,YACnB,iBACA,UAAS,mCACR;iBA6Ea,yBAAyB;;iBAoNzB,iBAAiB;;;iBC/YjB,mBAAmB,QACjC,KAAK,mBACL,OAAO,QACP,UAAS,4BACR,qBAAqB;;iBAwFR,gCACd,2BACA,aAAa,uBACb,UAAS,qCACR;;;iBClHa,wBAAwB;;;KCA5B;UAEK;EACf;EACA;EACA,QAAQ;;UAGO;EACf,QAAQ;EACR;EAKA;IAAe;IAAe;;EAC9B,SAAS;EACT;EACA;EACA;EACA;;UAGe;EACf;EACA;;iBA8Ec,yBACd,gBAAgB,2BACf;;;cC/FU;KAED;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;;;iBAatB,mCAAmC;EACvD;EACA,KAAK;EACL;IACE,QAAQ;iBAwHI,kCACd,kBACA,iBACA,WAAW,6BACX;;;KC7KU;;;;;;;;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;;iBAGO,8BAA8B,uBAAuB;iBAiB/C,6BAA6B;EACjD,SAAS;EACT;EACA,mBAAmB;EACnB;EACA,OAAO;EACP,SAAS;IACP;EACF,UAAU;EACV,aAAa;;EAEb,aAAa;;;;;;;;;iBA4MO,6BAA6B;EACjD;EACA,iBAAiB;EACjB,OAAO;EACP;EACA;EACA,SAAS;IACP;EACF,UAAU;EACV,aAAa;EACb,YAAY;;;;iBCrQQ,wBACpB,eACC,QAAQ,MAAM;iBA2BD,0BACd,SAAS,+BACT,eAAe,2BACf;EAAW;EAAe;IACzB,MAAM;iBAwBO,6BACd,SAAS,+BACT,eAAe,4BACd;iBAyCa,+BACd,SAAS,+BACT,iBAAiB,2BACjB,mBAAmB,2BACnB,eACC;iBAcmB,8BAA8B;EAClD,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;IACE,QAAQ;;;;;;;;KCnIA,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;;;;;;;;;cAUjD,mCAAmC;;;;;;;iBAUhC,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;;;;;;;;;iBAUjB,6BACd,aAAa,cAAc,gDAC1B;;;;;;;cCzUU;;;;cAKA;cAMA;;iBA8GG,4BAA4B,SAAS;;iBASrC,+BAA+B,SAAS;;;;iBAcxC,8BAA8B,SAAS;;;;;;;;;;;;;UC5EtC;;EAEf;;EAEA;;EAEA;;;iBAIc,8BACd,SAAS,+BACT,MAAM,6BACL,kBAAkB;;iBAsEL,kCACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;cAwhBpB,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;;;;;;;;;;;;;UCjtBrD;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,cAAc;;;iBAIA,2BACd,SAAS,+BACT,MAAM,0BACL,kBAAkB,mBAAmB;;iBA+GxB,+BACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;;iBAwBV,gBAAgB,QAAQ,oCAAoC;;;UCtI3D;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA,UAAU;;EAEV,YAAY;;;;;;iBA8EE;UAIC;;EAEf;;EAEA;;;;;;iBAOc,gCACd,MAAM,+BACL,kBAAkB;;iBAyDL,2BACd,SAAS,8BACR,uBAAuB;;;;;;;;;;cC5OpB;WACQ;WACC;WACK;WACd;WACkB;WACJ;WACO;WACD;WACJ;WACK;WACd;WACJ;WACM;WACC;WACG;WACH;WACI;WACT;;KAGC,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;;iBASK,sBACd,QAAQ,wBACR;EACE;EACA;EACA;EACA;EACA;IAED;;;UC3Fc;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;;iBAGK,8BACd,QAAQ,yBACP;iBAca,mCAAmC,SAAS;;;UCpC3C;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;;cAGW;cACA;cACA;cACA;;;iBC5KS,6BACpB,eACC,QAAQ;;;UCuEM;EACf,uBACE,SAAS,+BACT,QAAQ,wCACL,uBAAuB;EAC5B,2BACE,mBACA;IACE;IACA,aAAa,SAAS,OAAO;QAE5B,QAAQ;;;;;;;;;KAUH;;UAGK;EACf;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;;iBAKoB,2BACpB,yBACA,MAAK,OAAO,YACZ,eAAc,sCACb;;;cCzJU;cAEA;cAEA;cAOA;cAYA;cAGA;cAGA;cA+GA;iBAGG;iBAIA;;;;iBC1IA,oCAAoC,eAAe;;iBAQnD,gCAAgC,eAAe;;;;;;;;;;;iBAyB/C,+BACd,SAAS,+BACT,WAAW,KAAK;;;iBCzCF,+BACd,QAAQ,wBACR,uBAAsB;;;iBCIR,gCACd,kBACA,uBAAuB,gCACtB;;;;KCcS;;WAGG;WACA;WACA;WACA,YAAY;WACZ,UAAU;;;WAIV;WACA,QAAQ;;;;;;iBAOP,YAAY,MAAM,aAAa,QAC7C,YAAY,kBAAkB,MAAM,aAAa,SACjD,YAAY,0BACX,uBAAuB;;;;;;;;;;iBCpCV,mBAAmB,YAAY,yBAAyB;UAkBvD;EACf;EACA;EACA;EACA,MAAM;EACN,SAAS,QAAQ;;UAGF,wCAAwC;;EAEvD,iBAAiB,SAAS;;;iBAIZ,oBACd,SAAS,kCACR,oBAAoB;iBACP,oBACd,SAAS,6BACR,QAAQ;;;KCpBC;;UAGK;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;;;iBA2Ec,sBAAsB,QAAQ,uBAAuB;;iBAiHrD,uBACd,QAAQ,kBACR,qBACC;cA2HG,6BAA6B,oBAAoB;WAC5C;WACA;WAEA;WACA;IAAS;IAAgC;;WACzC;WACA;aACP;aACA;aACA;;EAGI,QAAQ,OAAO,kBAAkB,KAAK,iBAAiB,QAAQ;;cAS1D,qBAAmB;;;;;;;;;;;;;;;iBC5YhB,gBAAgB;;;;;iBAgBhB,WAAW;;;;iBCIX,kBAAkB,mBAAmB,WAAW;;;;;;iBAShD,uBACd,mBACA,mBACC,cAAc"}
|
package/dist/analyst/index.js
CHANGED
|
@@ -1,14 +1,16 @@
|
|
|
1
1
|
import { i as CostLedger } from "../cost-ledger-_IbkejCF.js";
|
|
2
2
|
import { n as assertProposalFindings, r as isProposalFinding } from "../abort-signal-CtzAM_sJ.js";
|
|
3
|
-
import {
|
|
4
|
-
import { B as findingSubjectGrammarPromptFor, F as coerceJson, H as renderFindingSubject, I as stripCodeFences, L as FINDING_SUBJECT_KINDS, M as RawAnalystFindingSchema, N as evidenceRefsFromRawFinding, P as parseRawFinding, R as FINDING_SUBJECT_SYNTAX, U as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, V as parseFindingSubject, W as resolveTraceAnalystLimits, i as buildTraceToolsForGroup, j as RAW_FINDING_SCHEMA_PROMPT, n as renderPriorFindings, r as runTraceAnalyst, t as createTraceAnalyst, z as KIND_EXPECTED_SUBJECTS } from "../kind-factory-
|
|
5
|
-
import {
|
|
6
|
-
import { t as
|
|
7
|
-
import {
|
|
3
|
+
import { i as makeProposalFinding, n as computeFindingId, r as makeFinding } from "../types-C8VBbiw6.js";
|
|
4
|
+
import { B as findingSubjectGrammarPromptFor, F as coerceJson, H as renderFindingSubject, I as stripCodeFences, L as FINDING_SUBJECT_KINDS, M as RawAnalystFindingSchema, N as evidenceRefsFromRawFinding, P as parseRawFinding, R as FINDING_SUBJECT_SYNTAX, U as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, V as parseFindingSubject, W as resolveTraceAnalystLimits, i as buildTraceToolsForGroup, j as RAW_FINDING_SCHEMA_PROMPT, n as renderPriorFindings, r as runTraceAnalyst, t as createTraceAnalyst, z as KIND_EXPECTED_SUBJECTS } from "../kind-factory-V79Hw0AU.js";
|
|
5
|
+
import { i as validateUsageSettlementTimeout, n as settleUsageReceiptFromCostLedger } from "../usage-receipt-Cx3xNeyB.js";
|
|
6
|
+
import { a as KNOWLEDGE_POISONING_KIND_SPEC, c as FAILURE_MODE_KIND_SPEC, d as emitControlIntegrityFindings, f as behavioralAnalyst, i as DEFAULT_TRACE_ANALYST_KINDS, l as CONTROL_INTEGRITY_ANALYST, n as AnalystRegistry, o as KNOWLEDGE_GAP_KIND_SPEC, p as deriveEfficiencyFindings, r as ExactAnalystRunExecutionError, s as IMPROVEMENT_KIND_SPEC, t as buildDefaultAnalystRegistry, u as ControlIntegrityAnalyst } from "../default-registry-xiHrldW8.js";
|
|
7
|
+
import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-BE9lCzzL.js";
|
|
8
|
+
import { a as diffFindings, i as defaultIsMaterial, n as runSemanticConceptJudge, o as createChatTraceEngine, r as FindingsStore, t as SEMANTIC_CONCEPT_JUDGE_VERSION } from "../semantic-concept-judge-Cnt0TWp0.js";
|
|
8
9
|
import { t as createChatClient } from "../chat-client-DqPqWzVV.js";
|
|
9
|
-
import { a as scoreAnalystFindings, i as summarizeAnalystBenchmarkRunner, n as runAnalystBenchmark, r as traceStoreEvidenceResolver, t as registryBenchmarkRunner } from "../benchmark-
|
|
10
|
-
import { $ as
|
|
11
|
-
import { a as extractPrimeJsonObject, c as primeProtocolSha256, d as runPrimeExchange, f as decodeReplyRows, i as emptyPrimeRawUsage, l as primeReplyDefect, n as buildPrimePrompt, o as mergePrimeRawUsage, r as buildPrimeRepairPrompt, s as normalizePrimeUsage, t as analystUsageReceiptFromPrimeUsage, u as projectPrimeTrajectory } from "../prime-protocol-
|
|
10
|
+
import { a as scoreAnalystFindings, i as summarizeAnalystBenchmarkRunner, n as runAnalystBenchmark, r as traceStoreEvidenceResolver, t as registryBenchmarkRunner } from "../benchmark-BeQJyrRC.js";
|
|
11
|
+
import { $ as ANALYST_BENCHMARK_MANIFEST_FILE, A as readAnalystInstructionsOverride, B as parseVerificationOutcome, C as emptyPublicBenchmarkRunner, D as summarizeCodeTraceCalibration, E as renderCodeTraceCalibrationMarkdown, F as publicBenchmarkRlmInstructions, G as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, H as ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES, I as publicBenchmarkSystemPrompt, J as ANALYST_BENCHMARK_IMPLEMENTATION_SHA256, K as ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM, L as DEFAULT_MAX_VERIFICATION_ARTIFACT_BYTES, M as MAX_INCORRECT_BLOCKS, N as MAX_INCORRECT_BLOCK_STEPS, O as analystInstructionsOverrideFromText, P as publicBenchmarkProtocolSha256, Q as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, R as appendVerificationArtifactsToOtlp, S as adaptPublicBenchmarkFindings, T as readAnalystBenchmarkArtifact, U as ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256, V as ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM, W as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, X as analystBenchmarkImplementationDigest, Y as analystBenchmarkDependencyLockDigest, Z as ANALYST_BENCHMARK_COST_LEDGER_FILE, _ as publicDirectAnalystDefinition, a as primeCodeTraceAnalystDefinition, at as codeTraceBenchCase, b as analystDefinitionAsymmetries, c as preparePublicAnalystBenchmark, ct as agentRxPredictionsToFindings, d as selectPublicBenchmarkRows, dt as normalizeBenchmarkLabel, et as ANALYST_BENCHMARK_OBSERVATIONS_FILE, f as createPublicBenchmarkRlmRunner, g as createPublicBenchmarkDirectRunner, h as runReplVariableAnalystDefinition, i as primeAnalystProtocolSha256, it as summarizeAgentRxCalibration, j as CODE_TRACE_BENCH_ANALYST_PROMPT, k as effectiveAnalystProtocolSha256, l as publicBenchmarkDistributions, lt as normalizeAgentRxCategory, m as rlmEngineLimits, n as renderAnalystBenchmarkMarkdown, nt as AGENT_RX_UPSTREAM_REVISION, o as runInlineAnalystDefinition, ot as codeTracerPredictionsToFindings, p as publicRlmAnalystDefinition, q as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, r as createPrimeBenchmarkRunner, rt as renderAgentRxCalibrationMarkdown, s as loadPublicBenchmarkRows, st as agentRxBenchmarkCase, t as runAnalystBenchmarkCommand, tt as compareAnalystRunners, u as publicBenchmarkSelectionReport, ut as roundAgentRxStep, v as runChunkedAnalystDefinition, w as expandCodeTraceFailureBlocks, x as analystDefinitionProtocolSha256, y as AnalystExpressivenessError, z as loadCodeTraceVerificationArtifacts } from "../benchmark-command-kvq7yFY4.js";
|
|
12
|
+
import { a as extractPrimeJsonObject, c as primeProtocolSha256, d as runPrimeExchange, f as decodeReplyRows, i as emptyPrimeRawUsage, l as primeReplyDefect, n as buildPrimePrompt, o as mergePrimeRawUsage, r as buildPrimeRepairPrompt, s as normalizePrimeUsage, t as analystUsageReceiptFromPrimeUsage, u as projectPrimeTrajectory } from "../prime-protocol-DcAK9pC8.js";
|
|
13
|
+
import { t as nodeHttpPrimeBridgeTransport } from "../prime-bridge-transport-Z3cpKGoV.js";
|
|
12
14
|
import { existsSync, readFileSync, readdirSync, statSync } from "node:fs";
|
|
13
15
|
import { join } from "node:path";
|
|
14
16
|
//#region src/analyst/adapters.ts
|