@tangle-network/agent-eval 0.198.0 → 0.198.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (51) hide show
  1. package/CHANGELOG.md +10 -0
  2. package/README.md +3 -1
  3. package/dist/adapters/http.d.ts +1 -1
  4. package/dist/analyst/index.d.ts +1 -1
  5. package/dist/analyst/index.js +1 -1
  6. package/dist/{benchmark-command-D8E5TaPC.js → benchmark-command-DKQuLrrD.js} +2 -2
  7. package/dist/{benchmark-command-D8E5TaPC.js.map → benchmark-command-DKQuLrrD.js.map} +1 -1
  8. package/dist/benchmarks/index.d.ts +2 -2
  9. package/dist/benchmarks/index.js +2 -2
  10. package/dist/campaign/index.d.ts +2 -2
  11. package/dist/campaign/index.js +5 -5
  12. package/dist/{campaign-Dlg2iOzH.js → campaign-DRdKwxm-.js} +6 -6
  13. package/dist/{campaign-Dlg2iOzH.js.map → campaign-DRdKwxm-.js.map} +1 -1
  14. package/dist/cli.js +16 -12
  15. package/dist/cli.js.map +1 -1
  16. package/dist/contract/index.d.ts +4 -4
  17. package/dist/contract/index.js +3 -3
  18. package/dist/{define-agent-eval-By9PImpa.d.ts → define-agent-eval-9JYoCkDI.d.ts} +3 -3
  19. package/dist/{define-agent-eval-By9PImpa.d.ts.map → define-agent-eval-9JYoCkDI.d.ts.map} +1 -1
  20. package/dist/{define-agent-eval-DvBgSzAI.js → define-agent-eval-DkfcqaCF.js} +2 -2
  21. package/dist/{define-agent-eval-DvBgSzAI.js.map → define-agent-eval-DkfcqaCF.js.map} +1 -1
  22. package/dist/{index-BhoOCduB.d.ts → index-BXJHJ_jE.d.ts} +2 -2
  23. package/dist/{index-BhoOCduB.d.ts.map → index-BXJHJ_jE.d.ts.map} +1 -1
  24. package/dist/{index-D-tI1vgf.d.ts → index-DzQ6T34s.d.ts} +3 -3
  25. package/dist/{index-D-tI1vgf.d.ts.map → index-DzQ6T34s.d.ts.map} +1 -1
  26. package/dist/index.d.ts +2 -2
  27. package/dist/index.js +4 -4
  28. package/dist/{llm-judge-BYWXsxhZ.js → llm-judge-3ncNcWNT.js} +3 -3
  29. package/dist/{llm-judge-BYWXsxhZ.js.map → llm-judge-3ncNcWNT.js.map} +1 -1
  30. package/dist/openapi.json +1 -1
  31. package/dist/{produced-state-DOfmCEwP.js → produced-state-qIhqN3V5.js} +2 -2
  32. package/dist/{produced-state-DOfmCEwP.js.map → produced-state-qIhqN3V5.js.map} +1 -1
  33. package/dist/search/index.d.ts +39 -14
  34. package/dist/search/index.d.ts.map +1 -1
  35. package/dist/search/index.js +2 -2
  36. package/dist/{search-kernel-D8_kxChy.js → search-kernel-CumYAjA3.js} +55 -27
  37. package/dist/search-kernel-CumYAjA3.js.map +1 -0
  38. package/dist/{search-kernel-CuvbakC_.d.ts → search-kernel-TLoiMMcq.d.ts} +15 -6
  39. package/dist/search-kernel-TLoiMMcq.d.ts.map +1 -0
  40. package/dist/{search-summary-Dmppg8kV.js → search-summary-dJOaZ3Dw.js} +2 -2
  41. package/dist/{search-summary-Dmppg8kV.js.map → search-summary-dJOaZ3Dw.js.map} +1 -1
  42. package/dist/{skillopt-optimization-method-BMP2qV_1.js → skillopt-optimization-method-BL_j6mX8.js} +2 -2
  43. package/dist/{skillopt-optimization-method-BMP2qV_1.js.map → skillopt-optimization-method-BL_j6mX8.js.map} +1 -1
  44. package/dist/{tree-DHHWsH-X.js → tree-DxGfhISA.js} +115 -60
  45. package/dist/tree-DxGfhISA.js.map +1 -0
  46. package/docs/search-ledger.md +6 -4
  47. package/docs/trace-contracts.md +1 -3
  48. package/package.json +1 -1
  49. package/dist/search-kernel-CuvbakC_.d.ts.map +0 -1
  50. package/dist/search-kernel-D8_kxChy.js.map +0 -1
  51. package/dist/tree-DHHWsH-X.js.map +0 -1
@@ -3,10 +3,10 @@ import { a as RunRecord, c as RunSplitTag, n as RunCostProvenance } from "./run-
3
3
  import "./types-CeDfCRXp.js";
4
4
  import "./statistical-heldout-CMpxMjlQ.js";
5
5
  import "./types-EBUav9Bu.js";
6
- import "./index-BhoOCduB.js";
6
+ import "./index-BXJHJ_jE.js";
7
7
  import { p as TraceSpanEvent } from "./client-BzL9niqi.js";
8
8
  import { d as InsightReport, t as AnalyzeRunsOptions } from "./analyze-runs-CPiKLJFs.js";
9
- import "./define-agent-eval-By9PImpa.js";
9
+ import "./define-agent-eval-9JYoCkDI.js";
10
10
  import "./default-registry-oK9hHfX_.js";
11
11
  import { AgentCandidateBenchmarkCellRef, AgentCandidateBenchmarkSuiteInputs, AgentCandidateBenchmarkTask, AgentCandidateBenchmarkTaskMaterial, AgentCandidateBundle, AgentCandidateEvaluationPolicy, AgentCandidateExperiment, AgentCandidateExperimentMaterial, AgentCandidateExperimentMeasurement, AgentImprovementCost, AgentImprovementMeasuredComparison, AgentProfileImprovementExperiment, AgentProfileImprovementExperimentMaterial, AgentProfileImprovementMeasuredComparison, AgentProfileImprovementMeasurement, AgentProfileImprovementRunCell, AgentProfileImprovementRunReceipt, AgentProfileImprovementSuiteInputs, AgentProfileImprovementTask, AgentProfileImprovementTaskMaterial, CandidateExecutionEvidence, Sha256Digest } from "@tangle-network/agent-interface";
12
12
  //#region src/contract/measured-comparison.d.ts
@@ -687,4 +687,4 @@ interface FromOtelSpansOptions {
687
687
  declare function fromOtelSpans(opts: FromOtelSpansOptions): RunRecord[];
688
688
  //#endregion
689
689
  export { FromRunRecordDirResult as $, CodeAgentSessionActionStatus as A, AgentTraceFile as B, fromOpenCodeSession as C, measuredComparisonFromCandidateExperiment as Ct, streamCodeAgentJsonlFile as D, sealCandidateExperiment as Dt, parseCodeAgentJsonlFile as E, sealCandidateBenchmarkTask as Et, CodeAgentSessionTerminalStatus as F, PartitionByAuthoringModelResult as G, AgentTraceRange as H, observeCodeAgentSession as I, EvalReportingSuiteInput as J, parseAgentTrace as K, AgentTraceContributor as L, CodeAgentSessionExecutionReceipt as M, CodeAgentSessionObservation as N, CodeAgentSessionAction as O, verifyCandidateExperiment as Ot, CodeAgentSessionSource as P, FromRunRecordDirOptions as Q, AgentTraceContributorType as R, fromKimiCodeSession as S, evaluatePairedMeasurements as St, parseCodeAgentJsonl as T, sealCandidateBenchmarkSuite as Tt, AgentTraceRecord as U, AgentTraceIndex as V, AuthoringProvenance as W, EvalReportingSuiteResult as X, EvalReportingSuiteOptions as Y, evalReportingSuite as Z, CodeAgentSessionIntakeResult as _, PairedMeasurement as _t, FromFeedbackTableOptions as a, RunAgentProfileImprovementExperimentOptions as at, fromClaudeCodeSession as b, RunCandidateExperimentOptions as bt, CodeAgentStoreObservation as c, runAgentProfileImprovementExperiment as ct, CodeAgentStoreScope as d, sealAgentProfileImprovementTask as dt, RunRecordRejection as et, CodeAgentStoreSession as f, verifyAgentProfileImprovementExperimentComparison as ft, CodeAgentSessionIntakeOptions as g, EvaluatePairedMeasurementsOptions as gt, CodeAgentSessionDiagnostic as h, CompareCandidateExperimentOptions as ht, FeedbackTableRow as i, CompareAgentProfileImprovementExperimentOptions as it, CodeAgentSessionActionSurface as j, CodeAgentSessionActionKind as k, verifyCandidateExperimentComparison as kt, CodeAgentStoreRejection as l, sealAgentProfileImprovementExperiment as lt, CodeAgentJsonlLine as m, CandidateExperimentRun as mt, fromOtelSpans as n, AgentProfileImprovementExperimentExecutionInput as nt, FromFeedbackTableResult as o, SealAgentProfileImprovementSuiteOptions as ot, observeCodeAgentStore as p, CandidateExperimentExecutionInput as pt, partitionRunsByAuthoringModel as q, FeedbackTableMeta as r, AgentProfileImprovementExperimentRun as rt, fromFeedbackTable as s, measuredComparisonFromAgentProfileImprovementExperiment as st, FromOtelSpansOptions as t, fromRunRecordDir as tt, CodeAgentStoreRejections as u, sealAgentProfileImprovementSuite as ut, CodeAgentSessionMetrics as v, PairedMeasurementAdapter as vt, fromPiSession as w, runCandidateExperiment as wt, fromCodexSession as x, SealCandidateBenchmarkSuiteOptions as xt, ParsedCodeAgentJsonl as y, PairedMeasurementEvaluation as yt, AgentTraceConversation as z };
690
- //# sourceMappingURL=index-D-tI1vgf.d.ts.map
690
+ //# sourceMappingURL=index-DzQ6T34s.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"index-D-tI1vgf.d.ts","names":[],"sources":["../src/contract/measured-comparison.ts","../src/contract/profile-measured-comparison.ts","../src/contract/intake/run-record-dir.ts","../src/contract/eval-reporting-suite.ts","../src/contract/intake/agent-trace.ts","../src/contract/intake/code-agent-observation.ts","../src/contract/intake/code-agent-session.ts","../src/contract/intake/code-agent-store.ts","../src/contract/intake/feedback-table.ts","../src/contract/intake/otel-spans.ts"],"mappings":";;;;;;;;;;;;UAiCiB;EACf,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,QAAQ;EACR,MAAM;EACN,eAAe;EACf;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QAAQ,OAAO,oCAAoC,QAAQ;;EAE3D;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;UAII,kBAAkB;EACjC;EACA,UAAU;EACV,WAAW;;;UAII,yBAAyB;EACxC,MAAM,KAAK;EACX,WAAW,KAAK;IAAkB;IAAc;;EAChD,QAAQ,KAAK;EACb,eAAe,KAAK,OAAO;EAC3B,UAAU,KAAK;EACf,UAAU,KAAK;EACf,OAAO,KAAK;;UAGG,kCAAkC;EACjD,uBAAuB,kBAAkB;EACzC,QAAQ;EACR,SAAS,yBAAyB;;EAElC;;EAEA,kBAAkB;;EAElB,kBAAkB;;;KAIR,8BAA8B,KACxC;EAGA,iBAAiB;EACjB,WAAW;EACX;;;iBAIc,2BACd,UAAU,sCACT;;iBAQa,4BACd,SAAS,qCACR;;iBAiBa,wBACd,UAAU,mCACT;iBAQa,0BAA0B,iBAAiB;;iBAarC,uBACpB,SAAS,gCACR,QAAQ;;;;;;;;iBAkEK,2BAA2B,MACzC,SAAS,kCAAkC,QAC1C;;iBA8Ya,0CACd,SAAS,oCACR;;iBAuEa,oCACd,iBACC;;;UCjrBc;EACf,aAAa;EACb,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,aAAa;EACb,MAAM;EACN,SAAS;EACT;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QACE,OAAO,kDACN,QAAQ;;EAEX;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;iBAIG,gCACd,UAAU,sCACT;;iBAQa,iCACd,SAAS,0CACR;;iBAqBa,sCACd,UAAU,4CACT;;;;;;iBAsBmB,qCACpB,SAAS,8CACR,QAAQ;;iBA0CK,wDACd,SAAS,kDACR;;iBAuEa,kDACd,iBACC;;;;UC7Oc;;EAEf;;EAEA;;EAEA;;UAGe;;;;;;EAMf;;;;;;;EAOA,WAAW;;;;;;EAMX;;UAGe;;EAEf,MAAM;;EAEN,UAAU;;EAEV;;;;;;;;;;iBAkBoB,iBACpB,cACA,UAAS,0BACR,QAAQ;;;;;KC7CC,0BAA0B;UAErB;;;;;EAKf,UAAU,KAAK;;EAEf,OAAO;;;;;;;;;;EAUP;;;;UAKe;;;EAGf,QAAQ;;EAER;;IAEE;;IAEA;;;IAGA;;IAEA;;;IAGA,UAAU;;;EAGZ;;;;;;;iBAUoB,mBACpB,OAAO,yBACP,UAAS,4BACR,QAAQ;;;KChEC;UAEK;EACf,MAAM;;EAEN;;UAGe;EACf;EACA;EACA;;;EAGA,cAAc;;UAGC;EACf;EACA,cAAc;EACd,QAAQ;;UAGO;EACf;EACA,eAAe;;UAGA;EACf;EACA;EACA;EACA;IAAQ;IAAc;;EACtB;IAAS;IAAe;;EACxB,OAAO;;;;UAOQ;EACf;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;KAGU,kBAAkB,YAAY;;;;;;iBAW1B,gBAAgB,SAAS,qBAAqB;UAmE7C;;;;EAIf,SAAS,YAAY;;;EAGrB,cAAc;;;;;;;;iBASA,8BACd,MAAM,aACN,OAAO,kBACN;;;KCjLS;KAEA;KAEA;KAEA;KASA;UAEK;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA,MAAM;EACN,SAAS;EACT;EACA,QAAQ;EACR;EACA;EACA,UAAU;;UAGK;EACf,QAAQ;EACR;EACA;EACA;IACE,QAAQ;IACR;;EAEF,SAAS;;UAGM;EACf,QAAQ;EACR;EACA;EACA,YAAY;;;;;;;iBAeE,wBACd,SAAS,iCACR;;;UCjCc;EACf;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf,QAAQ;EACR;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;;UAGe;EACf,MAAM;EACN,aAAa;EACb,SAAS;EACT,cAAc;;UAGC;EACf;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;EACA;EACA;EACA;EACA;EACA;;;EAGA,iBAAiB;;;EAGjB,YAAY;;;;KAKF;EACN;EAAe;EAAoB;;EACnC;EAAmB;;iBAcT,oBAAoB,gBAAgB;;;;;;;;;;iBAuB7B,yBAAyB,eAAe,eAAe;;;;;iBA6BxD,wBAAwB,eAAe,QAAQ;iBAUrD,iBACd,SAAS,gCACR;iBAIa,sBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,cACd,SAAS,gCACR;;;;UC/Lc;;;;;WAKN;;WAEA,QAAQ;;;;;WAKR;;;;;WAKA;aACE;;aAEA;;;;KAKD;;UAUK;WACN;WACA;WACA;WACA;WACA,aAAa;;;UAIP;WACN;WACA,UAAU,SAAS,QAAQ,OAAO;;WAElC,QAAQ;aACN;aACA,QAAQ;;;;;;;;;KAUT;WAEG;WACA,OAAO;WACP;WACA,mBAAmB;;WAEnB;;WAEA;;WAEA;WACA,WAAW,SAAS,QAAQ,OAAO;WACnC,YAAY;;WAGZ;WACA,OAAO;WACP;;WAEA;WACA,YAAY;;;;;;;;;iBAaL,sBACpB,OAAO,sBACN,QAAQ;;;UCtHM;;;EAGf;;EAEA;;;EAGA;;;EAGA,WAAW;;UAGI;EACf;;;EAGA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA,WAAW;;;EAGX,SAAS;;UAGM;;EAEf,SAAS;;;EAGT,OAAO;;;;EAIP;IAAU;IAAa;;;;;EAIvB;;UAGe;EACf,MAAM;;;EAGN,aAAa;IAAQ;IAAe;IAAe;;;iBAGrC,kBAAkB,MAAM,2BAA2B;;;UC3BlD;EACf,OAAO;;EAEP,eAAe;;EAEf;;;;;;EAMA,eAAe,eAAe,gBAAgB;;iBAGhC,cAAc,MAAM,uBAAuB"}
1
+ {"version":3,"file":"index-DzQ6T34s.d.ts","names":[],"sources":["../src/contract/measured-comparison.ts","../src/contract/profile-measured-comparison.ts","../src/contract/intake/run-record-dir.ts","../src/contract/eval-reporting-suite.ts","../src/contract/intake/agent-trace.ts","../src/contract/intake/code-agent-observation.ts","../src/contract/intake/code-agent-session.ts","../src/contract/intake/code-agent-store.ts","../src/contract/intake/feedback-table.ts","../src/contract/intake/otel-spans.ts"],"mappings":";;;;;;;;;;;;UAiCiB;EACf,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,QAAQ;EACR,MAAM;EACN,eAAe;EACf;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QAAQ,OAAO,oCAAoC,QAAQ;;EAE3D;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;UAII,kBAAkB;EACjC;EACA,UAAU;EACV,WAAW;;;UAII,yBAAyB;EACxC,MAAM,KAAK;EACX,WAAW,KAAK;IAAkB;IAAc;;EAChD,QAAQ,KAAK;EACb,eAAe,KAAK,OAAO;EAC3B,UAAU,KAAK;EACf,UAAU,KAAK;EACf,OAAO,KAAK;;UAGG,kCAAkC;EACjD,uBAAuB,kBAAkB;EACzC,QAAQ;EACR,SAAS,yBAAyB;;EAElC;;EAEA,kBAAkB;;EAElB,kBAAkB;;;KAIR,8BAA8B,KACxC;EAGA,iBAAiB;EACjB,WAAW;EACX;;;iBAIc,2BACd,UAAU,sCACT;;iBAQa,4BACd,SAAS,qCACR;;iBAiBa,wBACd,UAAU,mCACT;iBAQa,0BAA0B,iBAAiB;;iBAarC,uBACpB,SAAS,gCACR,QAAQ;;;;;;;;iBAkEK,2BAA2B,MACzC,SAAS,kCAAkC,QAC1C;;iBA8Ya,0CACd,SAAS,oCACR;;iBAuEa,oCACd,iBACC;;;UCjrBc;EACf,aAAa;EACb,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,aAAa;EACb,MAAM;EACN,SAAS;EACT;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QACE,OAAO,kDACN,QAAQ;;EAEX;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;iBAIG,gCACd,UAAU,sCACT;;iBAQa,iCACd,SAAS,0CACR;;iBAqBa,sCACd,UAAU,4CACT;;;;;;iBAsBmB,qCACpB,SAAS,8CACR,QAAQ;;iBA0CK,wDACd,SAAS,kDACR;;iBAuEa,kDACd,iBACC;;;;UC7Oc;;EAEf;;EAEA;;EAEA;;UAGe;;;;;;EAMf;;;;;;;EAOA,WAAW;;;;;;EAMX;;UAGe;;EAEf,MAAM;;EAEN,UAAU;;EAEV;;;;;;;;;;iBAkBoB,iBACpB,cACA,UAAS,0BACR,QAAQ;;;;;KC7CC,0BAA0B;UAErB;;;;;EAKf,UAAU,KAAK;;EAEf,OAAO;;;;;;;;;;EAUP;;;;UAKe;;;EAGf,QAAQ;;EAER;;IAEE;;IAEA;;;IAGA;;IAEA;;;IAGA,UAAU;;;EAGZ;;;;;;;iBAUoB,mBACpB,OAAO,yBACP,UAAS,4BACR,QAAQ;;;KChEC;UAEK;EACf,MAAM;;EAEN;;UAGe;EACf;EACA;EACA;;;EAGA,cAAc;;UAGC;EACf;EACA,cAAc;EACd,QAAQ;;UAGO;EACf;EACA,eAAe;;UAGA;EACf;EACA;EACA;EACA;IAAQ;IAAc;;EACtB;IAAS;IAAe;;EACxB,OAAO;;;;UAOQ;EACf;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;KAGU,kBAAkB,YAAY;;;;;;iBAW1B,gBAAgB,SAAS,qBAAqB;UAmE7C;;;;EAIf,SAAS,YAAY;;;EAGrB,cAAc;;;;;;;;iBASA,8BACd,MAAM,aACN,OAAO,kBACN;;;KCjLS;KAEA;KAEA;KAEA;KASA;UAEK;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA,MAAM;EACN,SAAS;EACT;EACA,QAAQ;EACR;EACA;EACA,UAAU;;UAGK;EACf,QAAQ;EACR;EACA;EACA;IACE,QAAQ;IACR;;EAEF,SAAS;;UAGM;EACf,QAAQ;EACR;EACA;EACA,YAAY;;;;;;;iBAeE,wBACd,SAAS,iCACR;;;UCjCc;EACf;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf,QAAQ;EACR;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;;UAGe;EACf,MAAM;EACN,aAAa;EACb,SAAS;EACT,cAAc;;UAGC;EACf;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;EACA;EACA;EACA;EACA;EACA;;;EAGA,iBAAiB;;;EAGjB,YAAY;;;;KAKF;EACN;EAAe;EAAoB;;EACnC;EAAmB;;iBAcT,oBAAoB,gBAAgB;;;;;;;;;;iBAuB7B,yBAAyB,eAAe,eAAe;;;;;iBA6BxD,wBAAwB,eAAe,QAAQ;iBAUrD,iBACd,SAAS,gCACR;iBAIa,sBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,cACd,SAAS,gCACR;;;;UC/Lc;;;;;WAKN;;WAEA,QAAQ;;;;;WAKR;;;;;WAKA;aACE;;aAEA;;;;KAKD;;UAUK;WACN;WACA;WACA;WACA;WACA,aAAa;;;UAIP;WACN;WACA,UAAU,SAAS,QAAQ,OAAO;;WAElC,QAAQ;aACN;aACA,QAAQ;;;;;;;;;KAUT;WAEG;WACA,OAAO;WACP;WACA,mBAAmB;;WAEnB;;WAEA;;WAEA;WACA,WAAW,SAAS,QAAQ,OAAO;WACnC,YAAY;;WAGZ;WACA,OAAO;WACP;;WAEA;WACA,YAAY;;;;;;;;;iBAaL,sBACpB,OAAO,sBACN,QAAQ;;;UCtHM;;;EAGf;;EAEA;;;EAGA;;;EAGA,WAAW;;UAGI;EACf;;;EAGA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA,WAAW;;;EAGX,SAAS;;UAGM;;EAEf,SAAS;;;EAGT,OAAO;;;;EAIP;IAAU;IAAa;;;;;EAIvB;;UAGe;EACf,MAAM;;;EAGN,aAAa;IAAQ;IAAe;IAAe;;;iBAGrC,kBAAkB,MAAM,2BAA2B;;;UC3BlD;EACf,OAAO;;EAEP,eAAe;;EAEf;;;;;;EAMA,eAAe,eAAe,gBAAgB;;iBAGhC,cAAc,MAAM,uBAAuB"}
package/dist/index.d.ts CHANGED
@@ -17,7 +17,7 @@ import { a as RunFilter, i as InMemoryTraceStore, n as FileSystemTraceStore, o a
17
17
  import { a as MintedRolloutLine } from "./schema-B5LtYvn9.js";
18
18
  import { a as mintRolloutRows, c as ScorePreference, i as RolloutSearchLineage, n as MintRolloutResult, t as MintRolloutOptions } from "./mint-C9Lojt1V.js";
19
19
  import { R as Scenario$1, S as JudgeConfig, _ as GateDecision, a as CampaignResult, w as JudgeScore$1 } from "./types-EBUav9Bu.js";
20
- import { Ar as RedTeamFinding, At as completionVerdict, Br as runCanaries, Ct as CorrectnessChecker, Dr as DEFAULT_RED_TEAM_CORPUS, Dt as RequirementCheck, Et as ProducedState, Fr as CanaryAlert, Gi as llmJudge, Ir as CanaryEvaluation, Ji as quotaExhaustedUntil, Lr as CanaryKind, Mr as redTeamDataset, Mt as verifyCompletion, Nr as redTeamReport, Or as RedTeamCase, Ot as SatisfiedBy, Pr as scoreRedTeamOutput, Rr as CanaryOptions, St as CompletionVerdict, Tt as ProducedProposal, Wi as LlmJudgeOptions, Yi as transientDispatchFailure, _t as ProposalEventLike, bt as extractProducedState, da as CampaignCellFailureReceipt, fa as CampaignCellRetryPolicy, ft as BackendIntegrityError, gt as ArtifactEventLike, ht as summarizeBackendIntegrity, jr as RedTeamReport, jt as createLlmCorrectnessChecker, kr as RedTeamCategory, kt as TaskGold, ma as runCampaign, mt as assertRealBackend, pa as RunCampaignOptions, pt as BackendIntegrityReport, vt as RuntimeEventLike, wt as LlmCorrectnessCheckerOpts, xt as CompletionRequirement, yt as ToolCallEventLike, zr as CanaryReport } from "./index-BhoOCduB.js";
20
+ import { Ar as RedTeamFinding, At as completionVerdict, Br as runCanaries, Ct as CorrectnessChecker, Dr as DEFAULT_RED_TEAM_CORPUS, Dt as RequirementCheck, Et as ProducedState, Fr as CanaryAlert, Gi as llmJudge, Ir as CanaryEvaluation, Ji as quotaExhaustedUntil, Lr as CanaryKind, Mr as redTeamDataset, Mt as verifyCompletion, Nr as redTeamReport, Or as RedTeamCase, Ot as SatisfiedBy, Pr as scoreRedTeamOutput, Rr as CanaryOptions, St as CompletionVerdict, Tt as ProducedProposal, Wi as LlmJudgeOptions, Yi as transientDispatchFailure, _t as ProposalEventLike, bt as extractProducedState, da as CampaignCellFailureReceipt, fa as CampaignCellRetryPolicy, ft as BackendIntegrityError, gt as ArtifactEventLike, ht as summarizeBackendIntegrity, jr as RedTeamReport, jt as createLlmCorrectnessChecker, kr as RedTeamCategory, kt as TaskGold, ma as runCampaign, mt as assertRealBackend, pa as RunCampaignOptions, pt as BackendIntegrityReport, vt as RuntimeEventLike, wt as LlmCorrectnessCheckerOpts, xt as CompletionRequirement, yt as ToolCallEventLike, zr as CanaryReport } from "./index-BXJHJ_jE.js";
21
21
  import { a as ProfileAxisSpec, c as expandProfileAxes, i as HarnessType, l as harnessAxisOf, n as CODING_HARNESSES, o as agentProfileHash, r as HARNESS_NATIVE_MODEL, s as agentProfileId, t as AgentProfile } from "./agent-profile-DDfIjdyS.js";
22
22
  import { i as DatasetSplit, n as DatasetManifest, r as DatasetScenario, t as Dataset } from "./dataset-DQqhOCPt.js";
23
23
  import { C as defineEquivalenceCheck, S as buildEquivalenceRecord, _ as StrategyChecker, a as CheckerIdentity, b as VerificationStrategyProfile, c as EquivalenceCheckDefinition, d as EquivalenceCheckerInput, f as EquivalenceCheckerResult, g as EquivalenceRecord, h as EquivalenceProtocolError, i as equivalenceVerdict, l as EquivalenceCheckSpec, m as EquivalenceObligationStatus, n as VerdictCertification, o as CheckerOutcome, p as EquivalenceObligation, r as certificationEvidenceDigest, s as EquivalenceArm, t as DefaultVerdict, u as EquivalenceChecker, v as VERIFICATION_STRATEGIES, w as runEquivalenceCheck, x as VerificationStrategySource, y as VERIFICATION_STRATEGY_SOURCES } from "./verdict-E4eRNf7-.js";
@@ -27,7 +27,7 @@ import { a as FAILURE_BLAME, c as FailureContext, d as INFRA_FAILURE_BLAMES, f a
27
27
  import { d as InsightReport, i as analyzeRuns, t as AnalyzeRunsOptions } from "./analyze-runs-CPiKLJFs.js";
28
28
  import { i as ExactAnalystRunEvent, l as ExactCapableAnalyst, o as ExactAnalystRunResult } from "./exact-types-KGxBWH4b.js";
29
29
  import { c as RegistryRunOpts, i as BudgetPolicy, n as AnalystRegistry, r as AnalystRegistryOptions, s as ExactRegistryRunOpts } from "./registry-Bu3lNole.js";
30
- import { _ as SelfImproveMethodResult, a as DefinedAgentEval, c as SelfImproveMethodOptions, d as SelfImproveProposerOptions, f as SelfImproveProposerResult, g as SelfImproveMethodProvenance, h as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveOptions, n as AgentEvalEvaluateOptions, o as defineAgentEval, p as SelfImproveResult, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveProgressEvent } from "./define-agent-eval-By9PImpa.js";
30
+ import { _ as SelfImproveMethodResult, a as DefinedAgentEval, c as SelfImproveMethodOptions, d as SelfImproveProposerOptions, f as SelfImproveProposerResult, g as SelfImproveMethodProvenance, h as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveOptions, n as AgentEvalEvaluateOptions, o as defineAgentEval, p as SelfImproveResult, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveProgressEvent } from "./define-agent-eval-9JYoCkDI.js";
31
31
  import { a as createTraceAnalyst, i as TraceAnalystDefinition, n as buildDefaultAnalystRegistry, r as CreateTraceAnalystOptions, t as DefaultAnalystRegistryOptions } from "./default-registry-oK9hHfX_.js";
32
32
  import { i as TraceAnalysisEngineResult, l as RawAnalystFinding, n as TraceAnalysisEngine, t as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS } from "./engine-Dc996kAg.js";
33
33
  import { A as SemanticConceptJudgeInput, C as DspyRlmTraceEngineOptions, D as ConceptFinding, E as createChatTraceEngine, M as SemanticConceptJudgeResult, N as runSemanticConceptJudge, O as ConceptSpec, T as ChatTraceEngineOptions, _ as FindingSubject, a as FAILURE_MODE_KIND_SPEC, d as FindingsStore, f as PersistedFinding, j as SemanticConceptJudgeOptions, k as SEMANTIC_CONCEPT_JUDGE_VERSION, l as DiffPolicy, m as diffFindings, t as DEFAULT_TRACE_ANALYST_KINDS, u as FindingsDiff, v as FindingSubjectKind, w as createDspyRlmTraceEngine } from "./index-lbwvO9vJ.js";
package/dist/index.js CHANGED
@@ -1,7 +1,7 @@
1
1
  import { t as __exportAll } from "./rolldown-runtime-8H4AJuhK.js";
2
2
  import { i as JudgeError, n as CaptureIntegrityError, o as NotFoundError, r as ConfigError, s as ValidationError, t as AgentEvalError } from "./errors-Dngq5h35.js";
3
3
  import { a as hashCanonical, r as canonicalString } from "./canonical-BsZVgBhW.js";
4
- import { a as CODING_HARNESSES, c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, n as completionVerdict, o as HARNESS_NATIVE_MODEL, r as createLlmCorrectnessChecker, s as agentProfileHash, t as extractProducedState, u as expandProfileAxes } from "./produced-state-DOfmCEwP.js";
4
+ import { a as CODING_HARNESSES, c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, n as completionVerdict, o as HARNESS_NATIVE_MODEL, r as createLlmCorrectnessChecker, s as agentProfileHash, t as extractProducedState, u as expandProfileAxes } from "./produced-state-qIhqN3V5.js";
5
5
  import { c as groupRunsByAgentProfileCell, d as validateAgentProfileCell, f as verifyAgentProfileCell, h as manifestContentDigest, i as agentProfileCellKey, m as hashJson, r as agentProfileCellHashMaterial, s as buildAgentProfileCell, t as AGENT_PROFILE_KINDS, u as toAgentProfileJson } from "./agent-profile-cell-DcFW4vOc.js";
6
6
  import { t as mulberry32 } from "./random-Dn5fPWkt.js";
7
7
  import { a as spearmanR, c as weightedMean, i as ranks, n as partialCredit, o as summarizeNumberSeries, r as pearsonR, s as weightedComposite, t as confidenceInterval } from "./descriptive-1V17A-qa.js";
@@ -23,13 +23,13 @@ import { r as observedSplitScore, t as isRealnessGated } from "./reward-nw2xZGZG
23
23
  import { a as parseRunRecordSafe, c as runTaskScore, i as modelHasSnapshot, l as searchCellRunId, n as UNKNOWN_MODEL, o as roundTripRunRecord, r as isRunRecord, s as runCostFloorUsd, t as RunRecordValidationError, u as validateRunRecord } from "./run-record-BRlM99rX.js";
24
24
  import { a as summaryTable, n as gainHistogram, r as paretoChart } from "./summary-report-BeYnvddy.js";
25
25
  import { n as contentHash, r as fileVerdictCache, t as canonicalJson } from "./verdict-cache-BQEsVy9n.js";
26
- import { J as redTeamDataset, X as scoreRedTeamOutput, Y as redTeamReport, Z as runCanaries, a as transientDispatchFailure, i as quotaExhaustedUntil, o as aggregateRunScore, q as DEFAULT_RED_TEAM_CORPUS, rt as runCampaign, s as clamp01, t as llmJudge } from "./llm-judge-BYWXsxhZ.js";
26
+ import { J as redTeamDataset, X as scoreRedTeamOutput, Y as redTeamReport, Z as runCanaries, a as transientDispatchFailure, i as quotaExhaustedUntil, o as aggregateRunScore, q as DEFAULT_RED_TEAM_CORPUS, rt as runCampaign, s as clamp01, t as llmJudge } from "./llm-judge-3ncNcWNT.js";
27
27
  import { a as resolveModelPricing, i as isModelPriced, n as estimateCost, r as estimateTokens, t as MODEL_PRICING } from "./metrics-DOMpkoeb.js";
28
28
  import { a as CostLedgerPersistenceError, c as costForTokenPricing, i as CostLedger, l as costForUsage, n as CostCallConflictError, o as CostReceiptCaptureError, r as CostCeilingReachedError, s as CostReservationExceededError, t as CostAccountingIncompleteError, u as modelPriceKey } from "./cost-ledger-_IbkejCF.js";
29
29
  import { B as assertRealBackend, F as surfaceContentHash, T as parseReflectionResponse, V as summarizeBackendIntegrity, c as ATTESTATION_ALGORITHM, l as attest, u as verifyAttestation, w as buildReflectionPrompt, z as BackendIntegrityError } from "./campaign-evidence-jL_U7Ikv.js";
30
30
  import { d as redactForShare, f as redactText, n as REDACTION_VERSION, p as shareAllowed, r as assessShareSafety, u as redact } from "./redact-Dk1ZdJEm.js";
31
- import { E as paretoFrontier, T as dominates, d as pairedPromotionPower, f as requiredPairsForPairedPromotion } from "./search-kernel-D8_kxChy.js";
32
- import { r as selfImprove, t as defineAgentEval } from "./define-agent-eval-DvBgSzAI.js";
31
+ import { D as paretoFrontier, E as dominates, d as pairedPromotionPower, f as requiredPairsForPairedPromotion } from "./search-kernel-CumYAjA3.js";
32
+ import { r as selfImprove, t as defineAgentEval } from "./define-agent-eval-DkfcqaCF.js";
33
33
  import { n as equivalenceVerdict, t as certificationEvidenceDigest } from "./verdict-DhPoXApk.js";
34
34
  import { t as TraceEmitter } from "./emitter-DbMXUIVP.js";
35
35
  import { t as buildTrajectory } from "./trajectory-D_7rLrvE.js";
@@ -16,9 +16,9 @@ import { r as summarizeEvaluationUnits, t as defineEvaluationClaim } from "./cla
16
16
  import { f as redactText, n as REDACTION_VERSION, o as detectCredential, s as detectPersonalData } from "./redact-Dk1ZdJEm.js";
17
17
  import { t as detectRewardHacking } from "./reward-hacking-D0XwhVWE.js";
18
18
  import { n as assertProposalFindings, t as combineAbortSignals } from "./abort-signal-CtzAM_sJ.js";
19
- import { D as paretoFrontierWithCrowding, E as paretoFrontier, a as searchPolicyView, b as estimateNode, i as searchExpansionIndex, n as runSearch, p as EXPANSION_OPERATORS, t as SEARCH_KERNEL_SOURCE } from "./search-kernel-D8_kxChy.js";
19
+ import { D as paretoFrontier, O as paretoFrontierWithCrowding, a as searchPolicyView, b as estimateNode, i as searchExpansionIndex, n as runSearch, p as EXPANSION_OPERATORS, t as SEARCH_KERNEL_SOURCE } from "./search-kernel-CumYAjA3.js";
20
20
  import { a as validateSearchLedgerEvent, d as searchTaskSetDigest, i as replaySearchLedgerText, l as searchCellId, n as openSearchLedger, u as searchNodeId } from "./search-ledger-BnUytR8X.js";
21
- import { n as searchProposerView, t as renderSearchSummary } from "./search-summary-Dmppg8kV.js";
21
+ import { n as searchProposerView, t as renderSearchSummary } from "./search-summary-dJOaZ3Dw.js";
22
22
  import { d as stripFencedJson, o as costReceiptFromLlm, s as costReceiptFromLlmError, u as maximumChargeForLlmRequest } from "./llm-client-c_bKHTXz.js";
23
23
  import { createHash } from "node:crypto";
24
24
  import { writeFileSync } from "node:fs";
@@ -6118,4 +6118,4 @@ function firstString(value) {
6118
6118
  //#endregion
6119
6119
  export { captureFinalEvidencePolicy as $, assertCompleteSearchHistory as A, searchReceiptAccounting as B, incumbentWithOperatorBandit as C, surfaceDiff as D, searchEdgeId as E, verifySearchHistoryArtifact as F, captureJudge as G, uniform as H, verifySearchHistoryReceipt as I, redTeamDataset as J, defaultProductionGate as K, campaignCellSearchResult as L, assertSearchHistoryMatchesState as M, createSearchHistoryReceipt as N, surfaceNode as O, searchHistoryCoverageRow as P, assertIndependentEvaluationSplit as Q, searchModelIdentity as R, incumbent as S, developmentClaim as T, runFinalComparison as U, asha as V, openAutoPr as W, scoreRedTeamOutput as X, redTeamReport as Y, runCanaries as Z, isProposedCandidate as _, transientDispatchFailure as a, resolveRunDir as at, beam as b, compareOptimizationMethods as c, buildCellSchedule as ct, combineComparisonCosts as d, evaluationUnitMap as et, costFromLedgerSummary as f, runOptimization as g, runImprovementLoop as h, quotaExhaustedUntil as i, planCampaignRun as it, assertSearchHistoryAdmissionOptions as j, SearchHistoryRequiredError as k, optimizationTokenUsageFromSummary as l, cellCachePath as lt, readGepaCandidatePopulationArtifact as m, JudgeParseError as n, runEval as nt, aggregateRunScore as o, tangleTracesRoot as ot, assertGepaCandidatePopulationSummary as p, DEFAULT_RED_TEAM_CORPUS as q, isTransientTransportFailure as r, runCampaign as rt, clamp01 as s, readCachedCell as st, llmJudge as t, reserveFinalEvidence as tt, executeOptimizationMethod as u, computeManifestHash as ut, labelTrustRank as v, SearchRecorder as w, crowdedFrontierParent as x, aide as y, searchProposalExecution as z };
6120
6120
 
6121
- //# sourceMappingURL=llm-judge-BYWXsxhZ.js.map
6121
+ //# sourceMappingURL=llm-judge-3ncNcWNT.js.map