@tangle-network/agent-eval 0.144.5 → 0.144.6

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 (119) hide show
  1. package/CHANGELOG.md +6 -0
  2. package/dist/{agent-profile-cell-Cw0PVwDr.d.ts → agent-profile-cell-BOP-iA9Q.d.ts} +2 -2
  3. package/dist/{agent-profile-cell-Cw0PVwDr.d.ts.map → agent-profile-cell-BOP-iA9Q.d.ts.map} +1 -1
  4. package/dist/analyst/index.d.ts +12 -12
  5. package/dist/analyst/index.d.ts.map +1 -1
  6. package/dist/analyst/index.js +3 -3
  7. package/dist/{benchmark-Fmo42QVE.d.ts → benchmark-J9Qe6j2_.d.ts} +3 -3
  8. package/dist/{benchmark-Fmo42QVE.d.ts.map → benchmark-J9Qe6j2_.d.ts.map} +1 -1
  9. package/dist/{benchmark-command-CA_NFOmy.js → benchmark-command-CQd78YHt.js} +6 -4
  10. package/dist/{benchmark-command-CA_NFOmy.js.map → benchmark-command-CQd78YHt.js.map} +1 -1
  11. package/dist/benchmarks/index.d.ts +1 -1
  12. package/dist/benchmarks/index.js +1 -1
  13. package/dist/{benchmarks-CWbsj0t4.js → benchmarks-BEOkuvIg.js} +3 -3
  14. package/dist/{benchmarks-CWbsj0t4.js.map → benchmarks-BEOkuvIg.js.map} +1 -1
  15. package/dist/campaign/index.d.ts +6 -6
  16. package/dist/campaign/index.js +2 -2
  17. package/dist/{campaign-ClpnD7Ug.js → campaign-CXsdyym7.js} +3 -3
  18. package/dist/{campaign-ClpnD7Ug.js.map → campaign-CXsdyym7.js.map} +1 -1
  19. package/dist/cli.js +2 -2
  20. package/dist/{client-DAb7MWtL.d.ts → client-0JI64ovJ.d.ts} +4 -4
  21. package/dist/{client-DAb7MWtL.d.ts.map → client-0JI64ovJ.d.ts.map} +1 -1
  22. package/dist/{completion-verifier-VvpHRu78.d.ts → completion-verifier-CBiee74w.d.ts} +5 -5
  23. package/dist/{completion-verifier-VvpHRu78.d.ts.map → completion-verifier-CBiee74w.d.ts.map} +1 -1
  24. package/dist/contract/index.d.ts +10 -10
  25. package/dist/contract/index.js +2 -2
  26. package/dist/control.d.ts +2 -2
  27. package/dist/{cost-ledger-FuQvHxPm.d.ts → cost-ledger-Bv_e8XHY.d.ts} +2 -2
  28. package/dist/{cost-ledger-FuQvHxPm.d.ts.map → cost-ledger-Bv_e8XHY.d.ts.map} +1 -1
  29. package/dist/{dataset-v_Y5902-.d.ts → dataset-C8xaLXdY.d.ts} +2 -2
  30. package/dist/{dataset-v_Y5902-.d.ts.map → dataset-C8xaLXdY.d.ts.map} +1 -1
  31. package/dist/{default-registry-RLNNoeEP.js → default-registry-Dta70shL.js} +2 -2
  32. package/dist/{default-registry-RLNNoeEP.js.map → default-registry-Dta70shL.js.map} +1 -1
  33. package/dist/{default-registry-BwbZ9N9v.d.ts → default-registry-J9m-_tya.d.ts} +5 -5
  34. package/dist/{default-registry-BwbZ9N9v.d.ts.map → default-registry-J9m-_tya.d.ts.map} +1 -1
  35. package/dist/{errors-DkfjIDvD.d.ts → errors-CKPfb2aH.d.ts} +2 -2
  36. package/dist/{errors-DkfjIDvD.d.ts.map → errors-CKPfb2aH.d.ts.map} +1 -1
  37. package/dist/errors-D-LKuDhb.js.map +1 -1
  38. package/dist/{eval-campaign-lZcDIwQM.js → eval-campaign-CfLQQs9B.js} +2 -2
  39. package/dist/{eval-campaign-lZcDIwQM.js.map → eval-campaign-CfLQQs9B.js.map} +1 -1
  40. package/dist/{exact-types-BQ7W90C4.d.ts → exact-types-CBYF5MGd.d.ts} +2 -2
  41. package/dist/{exact-types-BQ7W90C4.d.ts.map → exact-types-CBYF5MGd.d.ts.map} +1 -1
  42. package/dist/{external-optimizer-contracts-CdmX2K2S.d.ts → external-optimizer-contracts-Q9c0L3eY.d.ts} +3 -3
  43. package/dist/{external-optimizer-contracts-CdmX2K2S.d.ts.map → external-optimizer-contracts-Q9c0L3eY.d.ts.map} +1 -1
  44. package/dist/{feedback-trajectory-WK7x4mhy.d.ts → feedback-trajectory-GgoS0-MK.d.ts} +4 -4
  45. package/dist/{feedback-trajectory-WK7x4mhy.d.ts.map → feedback-trajectory-GgoS0-MK.d.ts.map} +1 -1
  46. package/dist/fuzz.d.ts +1 -1
  47. package/dist/hosted/index.d.ts +3 -3
  48. package/dist/{index-CsuAo2-J.d.ts → index-4XwggC10.d.ts} +5 -5
  49. package/dist/{index-CsuAo2-J.d.ts.map → index-4XwggC10.d.ts.map} +1 -1
  50. package/dist/{index-CNOCxBLh.d.ts → index-B6-B0zTB.d.ts} +2 -2
  51. package/dist/{index-CNOCxBLh.d.ts.map → index-B6-B0zTB.d.ts.map} +1 -1
  52. package/dist/{index-CpxZSlB7.d.ts → index-BIL5vxxt.d.ts} +12 -12
  53. package/dist/{index-CpxZSlB7.d.ts.map → index-BIL5vxxt.d.ts.map} +1 -1
  54. package/dist/{index-DEb46kc6.d.ts → index-CvSN3IG1.d.ts} +2 -2
  55. package/dist/{index-DEb46kc6.d.ts.map → index-CvSN3IG1.d.ts.map} +1 -1
  56. package/dist/{index-DGIzNtRv.d.ts → index-Dx1kF3Ez.d.ts} +3 -3
  57. package/dist/{index-DGIzNtRv.d.ts.map → index-Dx1kF3Ez.d.ts.map} +1 -1
  58. package/dist/index.d.ts +68 -77
  59. package/dist/index.d.ts.map +1 -1
  60. package/dist/index.js +56 -120
  61. package/dist/index.js.map +1 -1
  62. package/dist/{insight-report-D5m1z0_n.d.ts → insight-report-DqEsugpr.d.ts} +4 -4
  63. package/dist/{insight-report-D5m1z0_n.d.ts.map → insight-report-DqEsugpr.d.ts.map} +1 -1
  64. package/dist/{integrity-DRXobPEs.d.ts → integrity-CNGUaGBY.d.ts} +3 -3
  65. package/dist/{integrity-DRXobPEs.d.ts.map → integrity-CNGUaGBY.d.ts.map} +1 -1
  66. package/dist/ledger-core/index.d.ts +1 -1
  67. package/dist/{llm-client-D3EoChAU.js → llm-client-Dv5BiKLE.js} +289 -7
  68. package/dist/llm-client-Dv5BiKLE.js.map +1 -0
  69. package/dist/meta-eval/index.d.ts +2 -2
  70. package/dist/multishot/index.d.ts +2 -2
  71. package/dist/openapi.json +1 -1
  72. package/dist/profile-cell.d.ts +1 -1
  73. package/dist/{release-report-BZRWdq_t.d.ts → release-report-ChOgpIoQ.d.ts} +4 -4
  74. package/dist/{release-report-BZRWdq_t.d.ts.map → release-report-ChOgpIoQ.d.ts.map} +1 -1
  75. package/dist/{replay-B7S7Pdbw.d.ts → replay-Krvb114g.d.ts} +7 -7
  76. package/dist/{replay-B7S7Pdbw.d.ts.map → replay-Krvb114g.d.ts.map} +1 -1
  77. package/dist/reporting.d.ts +4 -4
  78. package/dist/{researcher-BSiCoM1s.d.ts → researcher-xLeNcpKX.d.ts} +6 -6
  79. package/dist/{researcher-BSiCoM1s.d.ts.map → researcher-xLeNcpKX.d.ts.map} +1 -1
  80. package/dist/{reward-hacking-DFgkEY4p.d.ts → reward-hacking-RZgnGWlx.d.ts} +2 -2
  81. package/dist/{reward-hacking-DFgkEY4p.d.ts.map → reward-hacking-RZgnGWlx.d.ts.map} +1 -1
  82. package/dist/rl.d.ts +5 -5
  83. package/dist/rl.js +1 -1
  84. package/dist/rollout/index.d.ts +1 -1
  85. package/dist/{rubric-predictive-validity-C4r4Y-q8.d.ts → rubric-predictive-validity-9qAwzkZm.d.ts} +2 -2
  86. package/dist/{rubric-predictive-validity-C4r4Y-q8.d.ts.map → rubric-predictive-validity-9qAwzkZm.d.ts.map} +1 -1
  87. package/dist/{run-evidence-j5Ynww6L.d.ts → run-evidence-C6G41MSI.d.ts} +3 -3
  88. package/dist/{run-evidence-j5Ynww6L.d.ts.map → run-evidence-C6G41MSI.d.ts.map} +1 -1
  89. package/dist/{run-record-ooo9FWns.d.ts → run-record-DdSa93_W.d.ts} +4 -4
  90. package/dist/{run-record-ooo9FWns.d.ts.map → run-record-DdSa93_W.d.ts.map} +1 -1
  91. package/dist/{semantic-concept-judge-DKRtp2sY.js → semantic-concept-judge-DwF6n05O.js} +2 -2
  92. package/dist/{semantic-concept-judge-DKRtp2sY.js.map → semantic-concept-judge-DwF6n05O.js.map} +1 -1
  93. package/dist/{server-Df00sdwz.js → server-D6XJQHw7.js} +2 -2
  94. package/dist/{server-Df00sdwz.js.map → server-D6XJQHw7.js.map} +1 -1
  95. package/dist/{skill-usage-DUvvudWR.d.ts → skill-usage-GlOphAhX.d.ts} +9 -9
  96. package/dist/{skill-usage-DUvvudWR.d.ts.map → skill-usage-GlOphAhX.d.ts.map} +1 -1
  97. package/dist/{skillopt-optimization-method-CGz9ywhM.d.ts → skillopt-optimization-method-7S43rbDB.d.ts} +12 -12
  98. package/dist/{skillopt-optimization-method-CGz9ywhM.d.ts.map → skillopt-optimization-method-7S43rbDB.d.ts.map} +1 -1
  99. package/dist/{skillopt-optimization-method-C4FX42dy.js → skillopt-optimization-method-Bfb-vBKe.js} +2 -2
  100. package/dist/{skillopt-optimization-method-C4FX42dy.js.map → skillopt-optimization-method-Bfb-vBKe.js.map} +1 -1
  101. package/dist/{statistics-B4u_CiFd.d.ts → statistics-C-dm-J6H.d.ts} +2 -2
  102. package/dist/{statistics-B4u_CiFd.d.ts.map → statistics-C-dm-J6H.d.ts.map} +1 -1
  103. package/dist/{summary-report-BOM6dfP7.d.ts → summary-report-B0cAyA7N.d.ts} +3 -3
  104. package/dist/{summary-report-BOM6dfP7.d.ts.map → summary-report-B0cAyA7N.d.ts.map} +1 -1
  105. package/dist/{tool-groups-ByZiqpVk.d.ts → tool-groups-CK0JCkqO.d.ts} +9 -9
  106. package/dist/{tool-groups-ByZiqpVk.d.ts.map → tool-groups-CK0JCkqO.d.ts.map} +1 -1
  107. package/dist/traces.d.ts +6 -6
  108. package/dist/{types-CZt1PBIk.d.ts → types-BhP9q0Fq.d.ts} +4 -4
  109. package/dist/{types-CZt1PBIk.d.ts.map → types-BhP9q0Fq.d.ts.map} +1 -1
  110. package/dist/{types-Dcoaqcsc.d.ts → types-DOZyvsFU.d.ts} +5 -5
  111. package/dist/{types-Dcoaqcsc.d.ts.map → types-DOZyvsFU.d.ts.map} +1 -1
  112. package/dist/{types-BjMFz88h.d.ts → types-XMVEdrE_.d.ts} +207 -6
  113. package/dist/types-XMVEdrE_.d.ts.map +1 -0
  114. package/dist/wire/index.d.ts +3 -3
  115. package/dist/wire/index.js +1 -1
  116. package/docs/building-doctrine.md +15 -0
  117. package/package.json +3 -3
  118. package/dist/llm-client-D3EoChAU.js.map +0 -1
  119. package/dist/types-BjMFz88h.d.ts.map +0 -1
@@ -1,4 +1,4 @@
1
- import { g as JudgeScore } from "./types-BjMFz88h.js";
1
+ import { g as JudgeScore } from "./types-XMVEdrE_.js";
2
2
  //#region src/judge-calibration.d.ts
3
3
  /**
4
4
  * Judge calibration — measure judge quality against human gold + bias.
@@ -965,4 +965,4 @@ declare function eProcess(opts?: EProcessOptions): EProcess;
965
965
  declare function mulberry32(seed: number): () => number;
966
966
  //#endregion
967
967
  export { pairedCohensDz as $, WeightedCompositeInput as A, positionalBias as At, eProcess as B, RankTestMethod as C, GoldenItem as Ct, ScoreRiskDifferenceResult as D, calibrateJudge as Dt, RiskDifferenceResult as E, VerbosityBiasResult as Et, cliffsDelta as F, mannWhitneyU as G, interRaterReliability as H, cohensD as I, mcnemarRequiredN as J, mcnemar as K, confidenceInterval as L, WilcoxonSignedRankResult as M, verbosityBias as Mt, benjaminiHochberg as N, SignTestAlternative as O, calibrateJudgeContinuous as Ot, bonferroni as P, pairedBootstrap as Q, corpusInterRaterAgreement as R, ProportionInterval as S, ContinuousCalibrationResult as St, RankTestOptions as T, SelfPreferenceResult as Tt, interpretCliffs as U, holm as V, isBinaryOutcomeVector as W, normalizeScores as X, mulberry32 as Y, pairedBinaryScale as Z, McNemarResult as _, wilson as _t, CorpusAgreementReport as a, pairedSignTest as at, PairedSignTestResult as b, ContinuousAgreement as bt, DEFAULT_PERMUTATIONS as c, passAtK as ct, EProcessState as d, requiredPairedSampleSize as dt, pairedDeltaTieFraction as et, EProcessStep as f, requiredSampleSize as ft, MannWhitneyResult as g, wilcoxonSignedRank as gt, MANN_WHITNEY_EXACT_MAX_WORK as h, weightedMean as ht, CorpusAgreementPerDimension as i, pairedRiskDifferenceScore as it, WeightedCompositeResult as j, selfPreference as jt, WILCOXON_EXACT_MAX_N as k, continuousAgreement as kt, EProcess as l, pearsonR as lt, MANN_WHITNEY_EXACT_MAX_STATES as m, weightedComposite as mt, CliffsMagnitude as n, pairedRiskDifference as nt, CorpusScoreRecord as o, pairedTTest as ot, ExactRiskDifferenceResult as p, spearmanR as pt, mcnemarPower as q, CorpusAgreementOptions as r, pairedRiskDifferenceExact as rt, DECISION_PAIRED_DELTA_STATISTIC as s, partialCredit as st, BOOTSTRAP_GATE_MIN_N as t, pairedMde as tt, EProcessOptions as u, ranks as ut, PairedBootstrapOptions as v, CalibrationResult as vt, RankTestMethodRequest as w, PositionalBiasResult as wt, PairedTTestResult as x, ContinuousAgreementOptions as xt, PairedBootstrapResult as y, CandidateScore as yt, corpusInterRaterAgreementFromJudgeScores as z };
968
- //# sourceMappingURL=statistics-B4u_CiFd.d.ts.map
968
+ //# sourceMappingURL=statistics-C-dm-J6H.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"statistics-B4u_CiFd.d.ts","names":[],"sources":["../src/judge-calibration.ts","../src/statistics.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;;;UAuBiB;EACf;EACA;;EAEA;;UAGe;EACf;EACA;;EAEA;;UAGe;EACf;EACA;;EAEA;;EAEA;;EAEA,YAAY;IAAQ;IAAgB;IAAe;IAAe;;;;;;iBAMpD,eACd,QAAQ,cACR,WAAW,mBACV;UA0Bc;;;;;EAKf;EACA;;;;;;iBAOc,eAAe,QAAQ,mBAAmB;UAgBzC;;EAEf;EACA;;iBAGc,cACd,SAAS;EAAQ;EAAmB;KACnC;UAYc;;EAEf;EACA;EACA;EACA;;;;;;;iBAQc,eACd,SAAS;EAAQ;EAAe;KAC/B;UA8Ec;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;IACE;IACA;;;EAGF;;EAEA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;;;;;;;;iBAUc,oBACd,oBACA,OAAM,6BACL;UAqEc,oCAAoC;;EAEnD;;EAEA;EACA;EACA;IACE;IACA;;;;;;;iBAQY,yBACd,QAAQ,cACR,WAAW,kBACX,OAAM,6BACL;;;;;cCnVU,kBAAe,QAAY,iBAAe;;iBAGvC,aAAa;EAAU;EAAe;;;;;;;;;;iBAoBtC,mBACd,kBACA,qBACA;EAAQ;EAAe;;EACpB;EAAc;EAAe;;;;;;;;;;;;;;;;;;;iBAiDlB,sBAAsB,aAAa;;KAkFvC;;;;;KAMA;UAEK;;EAEf,SAAS;;EAET;;;EAGA;;;cAIW;;cAEA;;cAEA;;cAEA;UAEI;;EAEf;;EAEA;;EAEA;;EAEA,QAAQ;;EAER;;;;;;;;;;;;;;iBAec,aACd,aACA,aACA,OAAM,kBACL;;iBAwFa,cAAc,iBAAiB;UAK9B;;EAEf;EACA;;EAEA;;;;;;;;;;;;;;;;iBAiBc,YAAY,kBAAkB,kBAAkB;UAyB/C;;;EAGf;;EAEA;;EAEA,QAAQ;;EAER;;EAEA;;;;;;;;;;;;;;;iBAgBc,mBACd,kBACA,iBACA,OAAM,kBACL;;;;;;;;;;;;;iBAmFa,QAAQ,aAAa;;;;;;;;;iBAqBrB,eAAe,kBAAkB;KAoBrC;;;;;;;;;;;iBAYI,YAAY,kBAAkB;;;;;;iBAkB9B,gBAAgB,gBAAgB;;;;;iBAuBhC,MAAM;;;;;;iBAmBN,SAAS,aAAa;;;;;iBAuBtB,UAAU,aAAa;UAKtB;;EAEf,MAAM;;;;;EAKN,SAAS;;EAET;;UAGe;EACf;EACA;;;;;;;;;;;iBAYc,kBAAkB,OAAO,yBAAyB;UA4CjD;;EAEf;;EAEA;;EAEA;;EAEA;;UAGe,oCAAoC;EACnD;;EAEA;;EAEA;;UAGe;;EAEf,cAAc;;EAEd;;EAEA;;EAEA;;EAEA;;UAGe,+BAA+B;;;;;;EAM9C;;;;;;EAMA;;;;;;;;;;;;;;;;;;iBAmBc,0BACd,SAAS,qBACT,OAAM,yBACL;;;;;;;;;iBA0Ha,yCACd,aAAa;EAAQ;EAAgB,QAAQ;IAC7C,OAAM,yBACL;;;;;;;iBA8Ba,mBAAmB;EACjC;EACA;EACA;EACA;;;;;;;;;;;iBAsBc,yBAAyB;EACvC;EACA;EACA;EACA;;;;;;;;iBAkBc,UAAU;EACxB;EACA;EACA;EACA;;;;;;;;;;;;;;;;iBAyBc,iBAAiB;EAC/B;EACA;EACA;EACA;EACA;;;;;;;iBAyBc,aAAa;EAC3B;EACA;EACA;EACA;EACA;;;;;;;;;iBAwBc,WACd,4BACA;EACG;EAAoB;;;;;;;;;;iBAgBT,KACd,4BACA;EACG;EAAoB;;;;;;;;;iBA4BT,kBACd,4BACA;EACG;EAAmB;;UAuCP;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;;cAaW;UAEI;;EAEf;;EAEA;;EAEA;;;EAGA;;;;;;;;;;;;iBAac,gBACd,kBACA,iBACA,OAAM,yBACL;;KA0DS;;UAGK;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,aAAa;;EAEb;;;;;;;;;;;;;;;iBAgBc,eACd,gCACA,aAAa,sBACZ;;UAgDc;;EAEf;;EAEA;;EAEA;;;;;;;;;iBAUc,OAAO,mBAAmB,WAAW,sBAAoB;;;;;;;;;;;;;;;;;;;;;;;;;;iBA2CzD,sBAAsB,QAAQ;;UAU7B;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;;;;iBAec,QACd,SAAS,6BACT,WAAW,8BACV;;UAmBc;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;;;;;;;iBAkBc,qBACd,SAAS,6BACT,WAAW,6BACX,sBACC;;;;UAkCc;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA+Bc,0BACd,SAAS,6BACT,WAAW,6BACX,sBACC;;;UAyEc;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA8Dc,0BACd,SAAS,6BACT,WAAW,6BACX,sBACC;;;;;;;;;;;;;;;;;;;;iBA6Ea,kBACd,QAAQ,mBACR,OAAO;;;iBAkBO,uBACd,QAAQ,mBACR,OAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;;cA4CI;;;;;;;;;;;iBAYG,QAAQ,WAAW,WAAW;UAgE7B;;;EAGf;;;;EAIA;;;;EAIA;;UAGe;;EAEf;;EAEA;;EAEA;;UAGe,sBAAsB;EACrC;EACA;EACA;;EAEA;;EAEA;;UAGe;;;EAGf,OAAO,YAAY;EACnB,SAAS;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA8BK,SAAS,OAAM,kBAAuB;;;;;iBAsatC,WAAW"}
1
+ {"version":3,"file":"statistics-C-dm-J6H.d.ts","names":[],"sources":["../src/judge-calibration.ts","../src/statistics.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;;;UAuBiB;EACf;EACA;;EAEA;;UAGe;EACf;EACA;;EAEA;;UAGe;EACf;EACA;;EAEA;;EAEA;;EAEA,YAAY;IAAQ;IAAgB;IAAe;IAAe;;;;;;iBAMpD,eACd,QAAQ,cACR,WAAW,mBACV;UA0Bc;;;;;EAKf;EACA;;;;;;iBAOc,eAAe,QAAQ,mBAAmB;UAgBzC;;EAEf;EACA;;iBAGc,cACd,SAAS;EAAQ;EAAmB;KACnC;UAYc;;EAEf;EACA;EACA;EACA;;;;;;;iBAQc,eACd,SAAS;EAAQ;EAAe;KAC/B;UA8Ec;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;IACE;IACA;;;EAGF;;EAEA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;;;;;;;;iBAUc,oBACd,oBACA,OAAM,6BACL;UAqEc,oCAAoC;;EAEnD;;EAEA;EACA;EACA;IACE;IACA;;;;;;;iBAQY,yBACd,QAAQ,cACR,WAAW,kBACX,OAAM,6BACL;;;;;cCnVU,kBAAe,QAAY,iBAAe;;iBAGvC,aAAa;EAAU;EAAe;;;;;;;;;;iBAoBtC,mBACd,kBACA,qBACA;EAAQ;EAAe;;EACpB;EAAc;EAAe;;;;;;;;;;;;;;;;;;;iBAiDlB,sBAAsB,aAAa;;KAkFvC;;;;;KAMA;UAEK;;EAEf,SAAS;;EAET;;;EAGA;;;cAIW;;cAEA;;cAEA;;cAEA;UAEI;;EAEf;;EAEA;;EAEA;;EAEA,QAAQ;;EAER;;;;;;;;;;;;;;iBAec,aACd,aACA,aACA,OAAM,kBACL;;iBAwFa,cAAc,iBAAiB;UAK9B;;EAEf;EACA;;EAEA;;;;;;;;;;;;;;;;iBAiBc,YAAY,kBAAkB,kBAAkB;UAyB/C;;;EAGf;;EAEA;;EAEA,QAAQ;;EAER;;EAEA;;;;;;;;;;;;;;;iBAgBc,mBACd,kBACA,iBACA,OAAM,kBACL;;;;;;;;;;;;;iBAmFa,QAAQ,aAAa;;;;;;;;;iBAqBrB,eAAe,kBAAkB;KAoBrC;;;;;;;;;;;iBAYI,YAAY,kBAAkB;;;;;;iBAkB9B,gBAAgB,gBAAgB;;;;;iBAuBhC,MAAM;;;;;;iBAmBN,SAAS,aAAa;;;;;iBAuBtB,UAAU,aAAa;UAKtB;;EAEf,MAAM;;;;;EAKN,SAAS;;EAET;;UAGe;EACf;EACA;;;;;;;;;;;iBAYc,kBAAkB,OAAO,yBAAyB;UA4CjD;;EAEf;;EAEA;;EAEA;;EAEA;;UAGe,oCAAoC;EACnD;;EAEA;;EAEA;;UAGe;;EAEf,cAAc;;EAEd;;EAEA;;EAEA;;EAEA;;UAGe,+BAA+B;;;;;;EAM9C;;;;;;EAMA;;;;;;;;;;;;;;;;;;iBAmBc,0BACd,SAAS,qBACT,OAAM,yBACL;;;;;;;;;iBA0Ha,yCACd,aAAa;EAAQ;EAAgB,QAAQ;IAC7C,OAAM,yBACL;;;;;;;iBA8Ba,mBAAmB;EACjC;EACA;EACA;EACA;;;;;;;;;;;iBAsBc,yBAAyB;EACvC;EACA;EACA;EACA;;;;;;;;iBAkBc,UAAU;EACxB;EACA;EACA;EACA;;;;;;;;;;;;;;;;iBAyBc,iBAAiB;EAC/B;EACA;EACA;EACA;EACA;;;;;;;iBAyBc,aAAa;EAC3B;EACA;EACA;EACA;EACA;;;;;;;;;iBAwBc,WACd,4BACA;EACG;EAAoB;;;;;;;;;;iBAgBT,KACd,4BACA;EACG;EAAoB;;;;;;;;;iBA4BT,kBACd,4BACA;EACG;EAAmB;;UAuCP;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;;cAaW;UAEI;;EAEf;;EAEA;;EAEA;;;EAGA;;;;;;;;;;;;iBAac,gBACd,kBACA,iBACA,OAAM,yBACL;;KA0DS;;UAGK;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,aAAa;;EAEb;;;;;;;;;;;;;;;iBAgBc,eACd,gCACA,aAAa,sBACZ;;UAgDc;;EAEf;;EAEA;;EAEA;;;;;;;;;iBAUc,OAAO,mBAAmB,WAAW,sBAAoB;;;;;;;;;;;;;;;;;;;;;;;;;;iBA2CzD,sBAAsB,QAAQ;;UAU7B;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;;;;iBAec,QACd,SAAS,6BACT,WAAW,8BACV;;UAmBc;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;;;;;;;iBAkBc,qBACd,SAAS,6BACT,WAAW,6BACX,sBACC;;;;UAkCc;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA+Bc,0BACd,SAAS,6BACT,WAAW,6BACX,sBACC;;;UAyEc;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA8Dc,0BACd,SAAS,6BACT,WAAW,6BACX,sBACC;;;;;;;;;;;;;;;;;;;;iBA6Ea,kBACd,QAAQ,mBACR,OAAO;;;iBAkBO,uBACd,QAAQ,mBACR,OAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;;cA4CI;;;;;;;;;;;iBAYG,QAAQ,WAAW,WAAW;UAgE7B;;;EAGf;;;;EAIA;;;;EAIA;;UAGe;;EAEf;;EAEA;;EAEA;;UAGe,sBAAsB;EACrC;EACA;EACA;;EAEA;;EAEA;;UAGe;;;EAGf,OAAO,YAAY;EACnB,SAAS;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA8BK,SAAS,OAAM,kBAAuB;;;;;iBAsatC,WAAW"}
@@ -1,5 +1,5 @@
1
- import { a as RunRecord } from "./run-record-ooo9FWns.js";
2
- import { y as PairedBootstrapResult } from "./statistics-B4u_CiFd.js";
1
+ import { a as RunRecord } from "./run-record-DdSa93_W.js";
2
+ import { y as PairedBootstrapResult } from "./statistics-C-dm-J6H.js";
3
3
  import { n as FailureClusterReport } from "./failure-cluster-CqcvCcdR.js";
4
4
  //#region src/paired-promotion-decision.d.ts
5
5
  /** Which paired estimator produced the deciding interval. */
@@ -636,4 +636,4 @@ interface ResearchReport {
636
636
  declare function researchReport(runs: RunRecord[], opts?: ResearchReportOptions): Promise<ResearchReport>;
637
637
  //#endregion
638
638
  export { PairedMcNemarEvidence as A, HeldOutGate as C, PairedDecisionMethod as D, SplitCoverage as E, PairedPromotionDecisionOptions as M, decidePairedPromotion as N, PairedDecisionShape as O, pairedDecisionShape as P, GateEvidence as S, HeldOutGateRejectionCode as T, paretoChart as _, ParetoPoint as a, DeltaStatistic as b, ResearchReportCandidate as c, ResearchReportOptions as d, ResearchReportRecommendation as f, gainHistogram as g, SummaryTableRow as h, ParetoFigureSpec as i, PairedPromotionDecision as j, PairedDecisionStatistic as k, ResearchReportDecision as l, SummaryTableOptions as m, GainDistributionFigureSpec as n, RESEARCH_REPORT_HARD_PAIR_FLOOR as o, SummaryTable as p, GainDistributionOptions as r, ResearchReport as s, GainDistributionBin as t, ResearchReportMethodology as u, researchReport as v, HeldOutGateConfig as w, GateDecision as x, summaryTable as y };
639
- //# sourceMappingURL=summary-report-BOM6dfP7.d.ts.map
639
+ //# sourceMappingURL=summary-report-B0cAyA7N.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"summary-report-BOM6dfP7.d.ts","names":[],"sources":["../src/paired-promotion-decision.ts","../src/held-out-gate.ts","../src/summary-report.ts"],"mappings":";;;;;KA2DY;;KAMA;;UAGK;;EAEf;;EAEA;;EAEA;;EAEA;;UAGe;;;EAGf;;EAEA;;EAEA;;EAEA;;;EAGA;;;;;;;;EAQA;;UAGe;;EAEf;;EAEA;EACA;EACA,WAAW;EACX,QAAQ;;;;;EAKR;;EAEA;;EAEA;;EAEA;;EAEA;;;;EAIA,WAAW;;EAEX,SAAS;;;EAGT;;EAEA;;EAEA;;;EAGA;;EAEA;;EAEA;;;EAGA;;EAEA;;EAEA;;EAEA;;;UAIe;EACf,WAAW;;EAEX;;EAEA;;;;;;;;iBASc,oBACd,kBACA,iBACA,gCACC;;;;;;iBAiBa,sBACd,kBACA,iBACA,UAAS,iCACR;;;KC1DS;;;;;;;;UAiBK;;EAEf;;;EAGA;;;;EAIA;;EAEA;;EAEA;;;EAGA;;UAGe;;;;EAIf;;;;;EAKA;;;;;;;;;;;EAWA;;;;EAIA;;;EAGA;;EAEA;;EAEA;;;EAGA;;;;;;;;;;;;;;;;;EAiBA;;;;;;;;;;;;;EAaA;;;;KAKU,iBAAiB;UAEZ;;EAEf;;EAEA;;EAEA;;;;;;;;EAQA,iBAAiB;;;;;EAKjB,gBAAgB;;;;;EAKhB;;;;;;;;;EASA,gBAAgB;;;EAGhB;;;EAGA;IAAY;IAAa;;;;EAGzB;;;;;;;EAOA;IAAW;IAAW;IAAW;IAAqB;;;;;;EAKtD;;;;;EAKA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;EAKA;;;EAGA;;;;;;;EAOA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA,UAAU;;EAEV;;EAEA,eAAe;;;;;;;cAQJ;mBACM;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;EAEjB,YAAY,QAAQ;;;;EAoCpB,SAAS,WAAW,aAAa,UAAU,cAAc;;;;;;;UA2SjD;;;;UC7oBO;;;EAGf;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;EACf,MAAM;EACN;EACA;;EAEA;;;;;;;iBAQc,aAAa,MAAM,aAAa,OAAM,sBAA2B;UA6IhE;EACf;;EAEA;;EAEA;;EAEA;;;EAGA;;;EAGA;;UAGe;EACf;EACA;EACA,QAAQ;EACR;IAAQ;IAAc;;;;;;;;iBAQR,YACd,MAAM,aACN;EACE;EACA,gBAAgB,eAAe;IAEhC;UAsDc;;EAEf;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;EACA,MAAM;EACN;EACA;IAAM;IAAa;;;UAGJ;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;iBAQc,cACd,MAAM,aACN,qBACA,oBACA,OAAM,0BACL;KA8ES;;;;;;cAYC;UAEI;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;EASA;;;;;;;EAOA;IAAS;IAAa;;;;;;EAKtB;;;;;EAKA;;EAEA,gBAAgB,eAAe;;EAE/B,kBAAkB;;EAElB;;EAEA;;EAEA;;;;;;EAMA;;UAGe;EACf,UAAU;EACV;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;IAAU;IAAa;;;;;;;EAMvB;EACA;;;;;;;EAOA;EACA;EACA,OAAO;EACP,UAAU;EACV;;UAGe;;;;;EAKf;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;EACA;;;;;;EAMA;EACA;EACA;IAAQ;IAAa;;EACrB;EACA,gBAAgB;EAChB,YAAY;EACZ,SAAS;EACT;IACE,QAAQ;IACR,OAAO;;EAET,aAAa;EACb,kBAAkB;EAClB;EACA;;;;;;;;;;;;;;;;;;;;;iBA6IoB,eACpB,MAAM,aACN,OAAM,wBACL,QAAQ"}
1
+ {"version":3,"file":"summary-report-B0cAyA7N.d.ts","names":[],"sources":["../src/paired-promotion-decision.ts","../src/held-out-gate.ts","../src/summary-report.ts"],"mappings":";;;;;KA2DY;;KAMA;;UAGK;;EAEf;;EAEA;;EAEA;;EAEA;;UAGe;;;EAGf;;EAEA;;EAEA;;EAEA;;;EAGA;;;;;;;;EAQA;;UAGe;;EAEf;;EAEA;EACA;EACA,WAAW;EACX,QAAQ;;;;;EAKR;;EAEA;;EAEA;;EAEA;;EAEA;;;;EAIA,WAAW;;EAEX,SAAS;;;EAGT;;EAEA;;EAEA;;;EAGA;;EAEA;;EAEA;;;EAGA;;EAEA;;EAEA;;EAEA;;;UAIe;EACf,WAAW;;EAEX;;EAEA;;;;;;;;iBASc,oBACd,kBACA,iBACA,gCACC;;;;;;iBAiBa,sBACd,kBACA,iBACA,UAAS,iCACR;;;KC1DS;;;;;;;;UAiBK;;EAEf;;;EAGA;;;;EAIA;;EAEA;;EAEA;;;EAGA;;UAGe;;;;EAIf;;;;;EAKA;;;;;;;;;;;EAWA;;;;EAIA;;;EAGA;;EAEA;;EAEA;;;EAGA;;;;;;;;;;;;;;;;;EAiBA;;;;;;;;;;;;;EAaA;;;;KAKU,iBAAiB;UAEZ;;EAEf;;EAEA;;EAEA;;;;;;;;EAQA,iBAAiB;;;;;EAKjB,gBAAgB;;;;;EAKhB;;;;;;;;;EASA,gBAAgB;;;EAGhB;;;EAGA;IAAY;IAAa;;;;EAGzB;;;;;;;EAOA;IAAW;IAAW;IAAW;IAAqB;;;;;;EAKtD;;;;;EAKA;;EAEA;;EAEA;;EAEA;;EAEA;;;;;EAKA;;;EAGA;;;;;;;EAOA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA,UAAU;;EAEV;;EAEA,eAAe;;;;;;;cAQJ;mBACM;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;mBACA;EAEjB,YAAY,QAAQ;;;;EAoCpB,SAAS,WAAW,aAAa,UAAU,cAAc;;;;;;;UA2SjD;;;;UC7oBO;;;EAGf;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;EACf,MAAM;EACN;EACA;;EAEA;;;;;;;iBAQc,aAAa,MAAM,aAAa,OAAM,sBAA2B;UA6IhE;EACf;;EAEA;;EAEA;;EAEA;;;EAGA;;;EAGA;;UAGe;EACf;EACA;EACA,QAAQ;EACR;IAAQ;IAAc;;;;;;;;iBAQR,YACd,MAAM,aACN;EACE;EACA,gBAAgB,eAAe;IAEhC;UAsDc;;EAEf;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;EACA,MAAM;EACN;EACA;IAAM;IAAa;;;UAGJ;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;iBAQc,cACd,MAAM,aACN,qBACA,oBACA,OAAM,0BACL;KA8ES;;;;;;cAYC;UAEI;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;;EASA;;;;;;;EAOA;IAAS;IAAa;;;;;;EAKtB;;;;;EAKA;;EAEA,gBAAgB,eAAe;;EAE/B,kBAAkB;;EAElB;;EAEA;;EAEA;;;;;;EAMA;;UAGe;EACf,UAAU;EACV;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;IAAU;IAAa;;;;;;;EAMvB;EACA;;;;;;;EAOA;EACA;EACA,OAAO;EACP,UAAU;EACV;;UAGe;;;;;EAKf;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;EACA;;;;;;EAMA;EACA;EACA;IAAQ;IAAa;;EACrB;EACA,gBAAgB;EAChB,YAAY;EACZ,SAAS;EACT;IACE,QAAQ;IACR,OAAO;;EAET,aAAa;EACb,kBAAkB;EAClB;EACA;;;;;;;;;;;;;;;;;;;;;iBA6IoB,eACpB,MAAM,aACN,OAAM,wBACL,QAAQ"}
@@ -1,11 +1,11 @@
1
- import { c as CostLedgerHandle } from "./cost-ledger-FuQvHxPm.js";
2
- import { a as RunRecord } from "./run-record-ooo9FWns.js";
3
- import { p as EvidenceRef, x as TraceAnalysisStore, y as BoundedTraceAnalysisStoreOptions } from "./types-CZt1PBIk.js";
4
- import { E as AnalystRegistry, r as CorrectnessChecker } from "./completion-verifier-VvpHRu78.js";
5
- import { S as JudgeConfig } from "./types-Dcoaqcsc.js";
6
- import { a as DatasetScenario } from "./dataset-v_Y5902-.js";
7
- import "./summary-report-BOM6dfP7.js";
8
- import { o as InsightReport, r as ExecutionInsight, t as CostProvenanceSummary } from "./insight-report-D5m1z0_n.js";
1
+ import { c as CostLedgerHandle } from "./cost-ledger-Bv_e8XHY.js";
2
+ import { a as RunRecord } from "./run-record-DdSa93_W.js";
3
+ import { p as EvidenceRef, x as TraceAnalysisStore, y as BoundedTraceAnalysisStoreOptions } from "./types-BhP9q0Fq.js";
4
+ import { E as AnalystRegistry, r as CorrectnessChecker } from "./completion-verifier-CBiee74w.js";
5
+ import { S as JudgeConfig } from "./types-DOZyvsFU.js";
6
+ import { a as DatasetScenario } from "./dataset-C8xaLXdY.js";
7
+ import "./summary-report-B0cAyA7N.js";
8
+ import { o as InsightReport, r as ExecutionInsight, t as CostProvenanceSummary } from "./insight-report-DqEsugpr.js";
9
9
  import { z } from "zod";
10
10
  //#region src/contract/analyze-runs.d.ts
11
11
  interface AnalyzeRunsOptions {
@@ -255,4 +255,4 @@ type TraceToolGroupName =
255
255
  declare function buildTraceToolsForGroup(group: TraceToolGroupName, store: TraceAnalysisStore): TraceAnalysisToolDescriptor[];
256
256
  //#endregion
257
257
  export { summarizeExecution as A, MakeEvalToolsConfig as C, ExecutionReport as D, AnalyzeRunsOptions as E, SummarizeExecutionOptions as O, EvalToolDef as S, toOpenAiTool as T, BuildTraceAnalysisToolsOptions as _, TraceAnalysisEngineRequest as a, buildTraceAnalysisToolDescriptors as b, resolveTraceAnalystLimits as c, RawAnalystEvidence as d, RawAnalystEvidenceSchema as f, parseRawFinding as g, evidenceRefsFromRawFinding as h, TraceAnalysisEngine as i, analyzeRuns as k, ANALYST_SEVERITIES as l, RawAnalystFindingSchema as m, buildTraceToolsForGroup as n, TraceAnalysisEngineResult as o, RawAnalystFinding as p, DEFAULT_TRACE_ANALYST_LIMITS as r, TraceAnalystLimits as s, TraceToolGroupName as t, RAW_FINDING_SCHEMA_PROMPT as u, TRACE_ANALYST_TOOL_NAMESPACE as v, makeEvalTools as w, traceAnalystFunctionGroup as x, TraceAnalysisToolDescriptor as y };
258
- //# sourceMappingURL=tool-groups-ByZiqpVk.d.ts.map
258
+ //# sourceMappingURL=tool-groups-CK0JCkqO.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"tool-groups-ByZiqpVk.d.ts","names":[],"sources":["../src/contract/analyze-runs.ts","../src/eval-tools.ts","../src/trace-analyst/tools.ts","../src/analyst/finding-signature.ts","../src/analyst/engine.ts","../src/analyst/tool-groups.ts"],"mappings":";;;;;;;;;;UAoEiB;;EAEf,MAAM;;;EAGN;;;;;EAKA;EACA;;;EAGA,kBAAkB;;;EAGlB,UAAU;;;;EAIV;IACE;IACA,cAAc;;;;;EAKhB,cAAc;IAAQ;IAAe;IAAe;;;EAEpD;;;;EAIA;;;;;;;;;EASA,eAAe;;;EAGf;;UAGe;EACf,MAAM;EACN;;UAGe;EACf,WAAW;EACX,gBAAgB;;;iBAIF,mBAAmB,MAAM,4BAA4B;iBAgB/C,YAAY,MAAM,qBAAqB,QAAQ;;;;UCpHpD;EACf;EACA;EACA,YAAY;EACZ,UAAU,eAAe;IAAQ,SAAS;QAAkB;;UAG7C;;EAEf,SAAS,MAAM;;;;EAIf;IAAe,kBAAkB;;;;EAGjC,UAAU,KAAK;;;iBAID,aAAa,KAAK;EAChC;EACA;IAAY;IAAc;IAAqB,YAAY;;;;;;;;iBAa7C,cAAc,KAAK,sBAAsB;;;cCjD5C;UAEI,uCAAuC;EACtD,OAAO;;UAGQ,oCAAoC;EACnD,kBAAkB;;;iBAIJ,kCACd,SAAS,iCACR;iBAkHa,0BAA0B,SAAS;EACjD;EACA;EACA;EACA;EACA,WAAW;;;;cCtIA;cAEA,0BAAwB,EAAA;;;GAK1B,EAAA,KAAA;KAEC,qBAAqB,EAAE,aAAa;cAiBnC,yBAAuB,EAAA;;;;;;;;;;;;;;;;;GAKzB,EAAA,KAAA;KAEC,oBAAoB,EAAE,aAAa;;;;;;cAOlC;;iBAYG,2BAA2B,SAAS,oBAAoB;iBAQxD,gBACd,cACA,OAAO,aAAa,SAAS,mCAC5B;;;;UCtEc;;EAEf;;EAEA;;EAEA;;EAEA;;cAGW,8BAA8B,SAAS;UAOnC;EACf;EACA;EACA;EACA,gBAAgB;EAChB,QAAQ;EACR,YAAY;EACZ;EACA,WAAW;EACX,SAAS;EACT,OAAO,iBAAiB,SAAS;;;UAIlB;EACf;EACA,UAAU;;EAEV;;;;;EAKA;;EAEA;EACA,SAAS;;;;;;;;;UAUM;WACN;WACA;WACA;;WAEA;;;;;;;;WAQA,iBAAiB,SAAS;EACnC,QAAQ,SAAS,6BAA6B,QAAQ;;iBAGxC,0BACd,QAAQ,QAAQ,kCACf;;;;KC3DS;;;;;;;;;;;;;;;;;;;;iBAgDI,wBACd,OAAO,oBACP,OAAO,qBACN"}
1
+ {"version":3,"file":"tool-groups-CK0JCkqO.d.ts","names":[],"sources":["../src/contract/analyze-runs.ts","../src/eval-tools.ts","../src/trace-analyst/tools.ts","../src/analyst/finding-signature.ts","../src/analyst/engine.ts","../src/analyst/tool-groups.ts"],"mappings":";;;;;;;;;;UAoEiB;;EAEf,MAAM;;;EAGN;;;;;EAKA;EACA;;;EAGA,kBAAkB;;;EAGlB,UAAU;;;;EAIV;IACE;IACA,cAAc;;;;;EAKhB,cAAc;IAAQ;IAAe;IAAe;;;EAEpD;;;;EAIA;;;;;;;;;EASA,eAAe;;;EAGf;;UAGe;EACf,MAAM;EACN;;UAGe;EACf,WAAW;EACX,gBAAgB;;;iBAIF,mBAAmB,MAAM,4BAA4B;iBAgB/C,YAAY,MAAM,qBAAqB,QAAQ;;;;UCpHpD;EACf;EACA;EACA,YAAY;EACZ,UAAU,eAAe;IAAQ,SAAS;QAAkB;;UAG7C;;EAEf,SAAS,MAAM;;;;EAIf;IAAe,kBAAkB;;;;EAGjC,UAAU,KAAK;;;iBAID,aAAa,KAAK;EAChC;EACA;IAAY;IAAc;IAAqB,YAAY;;;;;;;;iBAa7C,cAAc,KAAK,sBAAsB;;;cCjD5C;UAEI,uCAAuC;EACtD,OAAO;;UAGQ,oCAAoC;EACnD,kBAAkB;;;iBAIJ,kCACd,SAAS,iCACR;iBAkHa,0BAA0B,SAAS;EACjD;EACA;EACA;EACA;EACA,WAAW;;;;cCtIA;cAEA,0BAAwB,EAAA;;;GAK1B,EAAA,KAAA;KAEC,qBAAqB,EAAE,aAAa;cAiBnC,yBAAuB,EAAA;;;;;;;;;;;;;;;;;GAKzB,EAAA,KAAA;KAEC,oBAAoB,EAAE,aAAa;;;;;;cAOlC;;iBAYG,2BAA2B,SAAS,oBAAoB;iBAQxD,gBACd,cACA,OAAO,aAAa,SAAS,mCAC5B;;;;UCtEc;;EAEf;;EAEA;;EAEA;;EAEA;;cAGW,8BAA8B,SAAS;UAOnC;EACf;EACA;EACA;EACA,gBAAgB;EAChB,QAAQ;EACR,YAAY;EACZ;EACA,WAAW;EACX,SAAS;EACT,OAAO,iBAAiB,SAAS;;;UAIlB;EACf;EACA,UAAU;;EAEV;;;;;EAKA;;EAEA;EACA,SAAS;;;;;;;;;UAUM;WACN;WACA;WACA;;WAEA;;;;;;;;WAQA,iBAAiB,SAAS;EACnC,QAAQ,SAAS,6BAA6B,QAAQ;;iBAGxC,0BACd,QAAQ,QAAQ,kCACf;;;;KC3DS;;;;;;;;;;;;;;;;;;;;iBAgDI,wBACd,OAAO,oBACP,OAAO,qBACN"}
package/dist/traces.d.ts CHANGED
@@ -1,12 +1,12 @@
1
1
  import { C as TraceEvent, D as isSandboxSpan, E as isRetrievalSpan, O as isToolSpan, S as ToolSpan, T as isLlmSpan, _ as Span, a as FAILURE_CLASSES, b as SpanStatus, c as JudgeSpan, d as RetrievalSpan, f as Run, g as SandboxSpan, h as RunStatus, i as EventKind, l as LlmSpan, m as RunOutcome, n as BudgetLedgerEntry, o as FailureClass, p as RunLayer, r as BudgetSpec, s as GenericSpan, t as Artifact, u as Message, v as SpanBase, w as isJudgeSpan, x as TRACE_SCHEMA_VERSION, y as SpanKind } from "./schema-BtVldJ3T.js";
2
- import { a as ExtractedUsage, c as extractUsageFromResponse, i as ExtractUsageFromSseOptions, l as extractUsageFromSse, n as CaptureFetchOptions, o as SseUsageMode, r as captureFetchToRawSink, s as extractUsage, t as CaptureFetchContext } from "./index-DGIzNtRv.js";
3
- import { _t as RawProviderSink, bt as providerFromBaseUrl, dt as InMemoryRawProviderSink, ft as InMemoryRawProviderSinkOptions, gt as RawProviderEvent, ht as RawProviderDirection, lt as FileSystemRawProviderSink, mt as ProviderRedactor, pt as NoopRawProviderSink, ut as FileSystemRawProviderSinkOptions, vt as RawProviderSinkFilter, yt as defaultProviderRedactor } from "./types-BjMFz88h.js";
2
+ import { a as ExtractedUsage, c as extractUsageFromResponse, i as ExtractUsageFromSseOptions, l as extractUsageFromSse, n as CaptureFetchOptions, o as SseUsageMode, r as captureFetchToRawSink, s as extractUsage, t as CaptureFetchContext } from "./index-Dx1kF3Ez.js";
3
+ import { At as NoopRawProviderSink, Dt as FileSystemRawProviderSinkOptions, Et as FileSystemRawProviderSink, Ft as RawProviderSinkFilter, It as defaultProviderRedactor, Lt as providerFromBaseUrl, Mt as RawProviderDirection, Nt as RawProviderEvent, Ot as InMemoryRawProviderSink, Pt as RawProviderSink, jt as ProviderRedactor, kt as InMemoryRawProviderSinkOptions } from "./types-XMVEdrE_.js";
4
4
  import { a as RunFilter, i as InMemoryTraceStore, n as FileSystemTraceStore, o as SpanFilter, r as FileSystemTraceStoreOptions, s as TraceStore, t as EventFilter } from "./store-CT9YIIve.js";
5
5
  import { a as TraceEmitterOptions, i as TraceEmitter, n as RunCompleteHookContext, o as llmSpanFromProvider, r as SpanHandle, t as RunCompleteHook } from "./emitter-DGQGoLyj.js";
6
- import { a as RunIntegrityReport, i as RunIntegrityIssueCode, n as RunIntegrityExpectations, o as assertRunCaptured, r as RunIntegrityIssue, s as throwIfRunIncomplete, t as RunIntegrityError } from "./integrity-DRXobPEs.js";
7
- import { $ as traceAnalystOnRunComplete, A as stringField, At as OtlpSpanRoleInput, B as TraceInsightReadiness, Bt as exportRunAsOtlp, C as ProjectedOtlpSpan, Ct as createOtelTracingStore, D as inferOtlpKind, Dt as OtelExporter, E as firstStringAttr, Et as OtelExportConfig, F as TraceInsightFinding, Ft as traceSpanKindToOpenInferenceKind, G as defaultTraceInsightPanel, H as TraceInsightTask, I as TraceInsightPanelRole, It as OTEL_AGENT_EVAL_SCOPE, J as inferDomainKeywords, K as describeTraceInsightScope, L as TraceInsightPromptInput, Lt as OtlpExport, M as flattenOtlpExportToNdjson, Mt as applyToolSpanOtlpAttributes, N as OtlpFlatLine, Nt as classifyOtlpSpanRole, O as projectOtlpFlatLine, Ot as createOtelExporter, P as TraceInsightContext, Pt as isOtlpModelCall, Q as TraceAnalystHookOptions, R as TraceInsightQualityGate, Rt as OtlpResourceSpans, S as otlpToTraceRunRecords, St as redactValue, T as extractOtlpAttributes, Tt as ExportableSpan, U as buildTraceInsightContext, V as TraceInsightSuite, W as buildTraceInsightPrompt, X as scoreTraceInsightReadiness, Y as planTraceInsightQuestions, Z as tokenizeDomainWords, _ as OtlpTraceRunRecord, _t as DEFAULT_REDACTION_RULES, a as ReplayFetchOptions, at as TraceFileMissingError, b as otlpRowsToTraceRunRecords, bt as RedactionRule, c as ToolTraceMissingError, ct as AnalyzeTracesInput, d as OtlpFileTraceStoreOptions, dt as analyzeTraces, et as SpanNotFoundError, f as ToolSpansToTraceAnalysisStoreOptions, ft as createBoundedTraceAnalysisStore, g as OtlpToRunRecordsOptions, gt as convertTraceStoresToOtlp, h as TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, ht as TracesToOtlpResult, i as ReplayCacheStats, it as TraceFileMalformedError, j as FlattenOtlpOptions, jt as ToolSpanOtlpInput, k as readOtlpStatus, kt as OtlpSpanRole, l as toolSpansToTraceAnalysisStore, lt as AnalyzeTracesOptions, m as TRACE_ANALYST_ACTOR_DESCRIPTION, mt as TraceStoreToOtlpOptions, n as ReplayCacheEntry, nt as TraceAnalysisStoreContractError, o as createReplayFetch, ot as TraceFileTooLargeError, p as otlpTextToTraceAnalysisStore, pt as TraceStoreSource, q as domainEvidencePattern, r as ReplayCacheMissError, rt as TraceAnalysisValidationError, s as iterateRawCalls, st as TraceNotFoundError, t as ReplayCache, tt as TraceAnalysisLimitError, u as OtlpFileTraceStore, ut as AnalyzeTracesResult, v as TraceAggregate, vt as REDACTION_VERSION, w as asString, wt as otelRunCompleteHook, x as otlpToRunRecords, xt as redactString, y as otlpRowsToRunRecords, yt as RedactionReport, z as TraceInsightQuestion, zt as OtlpSpan } from "./replay-B7S7Pdbw.js";
6
+ import { a as RunIntegrityReport, i as RunIntegrityIssueCode, n as RunIntegrityExpectations, o as assertRunCaptured, r as RunIntegrityIssue, s as throwIfRunIncomplete, t as RunIntegrityError } from "./integrity-CNGUaGBY.js";
7
+ import { $ as traceAnalystOnRunComplete, A as stringField, At as OtlpSpanRoleInput, B as TraceInsightReadiness, Bt as exportRunAsOtlp, C as ProjectedOtlpSpan, Ct as createOtelTracingStore, D as inferOtlpKind, Dt as OtelExporter, E as firstStringAttr, Et as OtelExportConfig, F as TraceInsightFinding, Ft as traceSpanKindToOpenInferenceKind, G as defaultTraceInsightPanel, H as TraceInsightTask, I as TraceInsightPanelRole, It as OTEL_AGENT_EVAL_SCOPE, J as inferDomainKeywords, K as describeTraceInsightScope, L as TraceInsightPromptInput, Lt as OtlpExport, M as flattenOtlpExportToNdjson, Mt as applyToolSpanOtlpAttributes, N as OtlpFlatLine, Nt as classifyOtlpSpanRole, O as projectOtlpFlatLine, Ot as createOtelExporter, P as TraceInsightContext, Pt as isOtlpModelCall, Q as TraceAnalystHookOptions, R as TraceInsightQualityGate, Rt as OtlpResourceSpans, S as otlpToTraceRunRecords, St as redactValue, T as extractOtlpAttributes, Tt as ExportableSpan, U as buildTraceInsightContext, V as TraceInsightSuite, W as buildTraceInsightPrompt, X as scoreTraceInsightReadiness, Y as planTraceInsightQuestions, Z as tokenizeDomainWords, _ as OtlpTraceRunRecord, _t as DEFAULT_REDACTION_RULES, a as ReplayFetchOptions, at as TraceFileMissingError, b as otlpRowsToTraceRunRecords, bt as RedactionRule, c as ToolTraceMissingError, ct as AnalyzeTracesInput, d as OtlpFileTraceStoreOptions, dt as analyzeTraces, et as SpanNotFoundError, f as ToolSpansToTraceAnalysisStoreOptions, ft as createBoundedTraceAnalysisStore, g as OtlpToRunRecordsOptions, gt as convertTraceStoresToOtlp, h as TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, ht as TracesToOtlpResult, i as ReplayCacheStats, it as TraceFileMalformedError, j as FlattenOtlpOptions, jt as ToolSpanOtlpInput, k as readOtlpStatus, kt as OtlpSpanRole, l as toolSpansToTraceAnalysisStore, lt as AnalyzeTracesOptions, m as TRACE_ANALYST_ACTOR_DESCRIPTION, mt as TraceStoreToOtlpOptions, n as ReplayCacheEntry, nt as TraceAnalysisStoreContractError, o as createReplayFetch, ot as TraceFileTooLargeError, p as otlpTextToTraceAnalysisStore, pt as TraceStoreSource, q as domainEvidencePattern, r as ReplayCacheMissError, rt as TraceAnalysisValidationError, s as iterateRawCalls, st as TraceNotFoundError, t as ReplayCache, tt as TraceAnalysisLimitError, u as OtlpFileTraceStore, ut as AnalyzeTracesResult, v as TraceAggregate, vt as REDACTION_VERSION, w as asString, wt as otelRunCompleteHook, x as otlpToRunRecords, xt as redactString, y as otlpRowsToRunRecords, yt as RedactionReport, z as TraceInsightQuestion, zt as OtlpSpan } from "./replay-Krvb114g.js";
8
8
  import { C as TOOL_LATENCY_MS, D as asNumber, E as applyLlmSpanOtlpAttributes, O as contextInputTokens, S as TOOL_ARGS_CAPTURED, T as TOOL_NAME_ATTR_KEYS, _ as LlmSpanOtlpInput, a as LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, b as RUN_COST_ATTR_KEYS, c as LLM_COST_USD, d as LLM_MODEL_ATTR_KEYS, f as LLM_MODEL_NAME, g as LLM_REASONING_TOKEN_ATTR_KEYS, h as LLM_REASONING_TOKENS, i as LLM_CACHE_WRITE_TOKENS, k as firstNumberAttr, l as LLM_INPUT_TOKENS, m as LLM_OUTPUT_TOKEN_ATTR_KEYS, n as LLM_CACHED_TOKENS, o as LLM_CONTEXT_TOKENS, p as LLM_OUTPUT_TOKENS, r as LLM_CACHED_TOKEN_ATTR_KEYS, s as LLM_COST_ATTR_KEYS, t as INPUT_VALUE, u as LLM_INPUT_TOKEN_ATTR_KEYS, v as OPENINFERENCE_SPAN_KIND, w as TOOL_NAME, x as SPAN_KIND_ATTR_KEYS, y as OUTPUT_VALUE } from "./attribute-vocabulary-DLJ6303h.js";
9
9
  import { a as judgeSpans, c as runsForScenario, i as hasCapturedToolArgs, l as toolSpans, n as argHash, o as llmSpans, r as groupBy, s as runFailureClass, t as aggregateLlm } from "./query-CJ_DX8vl.js";
10
- import { A as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, C as DEFAULT_TRACE_ANALYST_BUDGETS, D as SearchSpanResult, E as QueryTracesPage, F as TraceAnalystSpanStatus, I as TraceAnalystTraceSummary, L as ViewSpansResult, M as TraceAnalystFilters, N as TraceAnalystSpan, O as SearchTraceResult, P as TraceAnalystSpanKind, R as ViewTraceOversized, S as TraceAnalysisStoreContext, T as ErrorCluster, b as TRACE_ANALYSIS_LIMITS, j as TraceAnalystByteBudgets, k as SpanMatchRecord, w as DatasetOverview, x as TraceAnalysisStore, y as BoundedTraceAnalysisStoreOptions, z as ViewTraceResult } from "./types-CZt1PBIk.js";
11
- import { _ as BuildTraceAnalysisToolsOptions, b as buildTraceAnalysisToolDescriptors, v as TRACE_ANALYST_TOOL_NAMESPACE, x as traceAnalystFunctionGroup, y as TraceAnalysisToolDescriptor } from "./tool-groups-ByZiqpVk.js";
10
+ import { A as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, C as DEFAULT_TRACE_ANALYST_BUDGETS, D as SearchSpanResult, E as QueryTracesPage, F as TraceAnalystSpanStatus, I as TraceAnalystTraceSummary, L as ViewSpansResult, M as TraceAnalystFilters, N as TraceAnalystSpan, O as SearchTraceResult, P as TraceAnalystSpanKind, R as ViewTraceOversized, S as TraceAnalysisStoreContext, T as ErrorCluster, b as TRACE_ANALYSIS_LIMITS, j as TraceAnalystByteBudgets, k as SpanMatchRecord, w as DatasetOverview, x as TraceAnalysisStore, y as BoundedTraceAnalysisStoreOptions, z as ViewTraceResult } from "./types-BhP9q0Fq.js";
11
+ import { _ as BuildTraceAnalysisToolsOptions, b as buildTraceAnalysisToolDescriptors, v as TRACE_ANALYST_TOOL_NAMESPACE, x as traceAnalystFunctionGroup, y as TraceAnalysisToolDescriptor } from "./tool-groups-CK0JCkqO.js";
12
12
  export { type AnalyzeTracesInput, type AnalyzeTracesOptions, type AnalyzeTracesResult, Artifact, type BoundedTraceAnalysisStoreOptions, BudgetLedgerEntry, BudgetSpec, type BuildTraceAnalysisToolsOptions, CaptureFetchContext, CaptureFetchOptions, DEFAULT_REDACTION_RULES, DEFAULT_TRACE_ANALYST_BUDGETS, type DatasetOverview, type ErrorCluster, EventFilter, EventKind, ExportableSpan, ExtractUsageFromSseOptions, ExtractedUsage, FAILURE_CLASSES, FailureClass, FileSystemRawProviderSink, FileSystemRawProviderSinkOptions, FileSystemTraceStore, FileSystemTraceStoreOptions, type FlattenOtlpOptions, GenericSpan, INPUT_VALUE, InMemoryRawProviderSink, InMemoryRawProviderSinkOptions, InMemoryTraceStore, JudgeSpan, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_CACHE_WRITE_TOKENS, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, LLM_CONTEXT_TOKENS, LLM_COST_ATTR_KEYS, LLM_COST_USD, LLM_INPUT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_MODEL_ATTR_KEYS, LLM_MODEL_NAME, LLM_OUTPUT_TOKENS, LLM_OUTPUT_TOKEN_ATTR_KEYS, LLM_REASONING_TOKENS, LLM_REASONING_TOKEN_ATTR_KEYS, LlmSpan, LlmSpanOtlpInput, Message, NoopRawProviderSink, OPENINFERENCE_SPAN_KIND, OTEL_AGENT_EVAL_SCOPE, OUTPUT_VALUE, OtelExportConfig, OtelExporter, OtlpExport, OtlpFileTraceStore, type OtlpFileTraceStoreOptions, type OtlpFlatLine, OtlpResourceSpans, OtlpSpan, OtlpSpanRole, OtlpSpanRoleInput, type OtlpToRunRecordsOptions, type OtlpTraceRunRecord, type ProjectedOtlpSpan, ProviderRedactor, type QueryTracesPage, REDACTION_VERSION, RUN_COST_ATTR_KEYS, RawProviderDirection, RawProviderEvent, RawProviderSink, RawProviderSinkFilter, RedactionReport, RedactionRule, ReplayCache, ReplayCacheEntry, ReplayCacheMissError, ReplayCacheStats, ReplayFetchOptions, RetrievalSpan, Run, RunCompleteHook, RunCompleteHookContext, RunFilter, RunIntegrityError, RunIntegrityExpectations, RunIntegrityIssue, RunIntegrityIssueCode, RunIntegrityReport, RunLayer, RunOutcome, RunStatus, SPAN_KIND_ATTR_KEYS, SandboxSpan, type SearchSpanResult, type SearchTraceResult, Span, SpanBase, SpanFilter, SpanHandle, SpanKind, type SpanMatchRecord, SpanNotFoundError, SpanStatus, SseUsageMode, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYSIS_LIMITS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_TOOL_NAMESPACE, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TRACE_SCHEMA_VERSION, ToolSpan, ToolSpanOtlpInput, type ToolSpansToTraceAnalysisStoreOptions, ToolTraceMissingError, type TraceAggregate, TraceAnalysisLimitError, type TraceAnalysisStore, type TraceAnalysisStoreContext, TraceAnalysisStoreContractError, type TraceAnalysisToolDescriptor, TraceAnalysisValidationError, type TraceAnalystByteBudgets, type TraceAnalystFilters, type TraceAnalystHookOptions, type TraceAnalystSpan, type TraceAnalystSpanKind, type TraceAnalystSpanStatus, type TraceAnalystTraceSummary, TraceEmitter, TraceEmitterOptions, TraceEvent, TraceFileMalformedError, TraceFileMissingError, TraceFileTooLargeError, type TraceInsightContext, type TraceInsightFinding, type TraceInsightPanelRole, type TraceInsightPromptInput, type TraceInsightQualityGate, type TraceInsightQuestion, type TraceInsightReadiness, type TraceInsightSuite, type TraceInsightTask, TraceNotFoundError, TraceStore, TraceStoreSource, TraceStoreToOtlpOptions, TracesToOtlpResult, type ViewSpansResult, type ViewTraceOversized, type ViewTraceResult, aggregateLlm, analyzeTraces, applyLlmSpanOtlpAttributes, applyToolSpanOtlpAttributes, argHash, asNumber, asString, assertRunCaptured, buildTraceAnalysisToolDescriptors, buildTraceInsightContext, buildTraceInsightPrompt, captureFetchToRawSink, classifyOtlpSpanRole, contextInputTokens, convertTraceStoresToOtlp, createBoundedTraceAnalysisStore, createOtelExporter, createOtelTracingStore, createReplayFetch, defaultProviderRedactor, defaultTraceInsightPanel, describeTraceInsightScope, domainEvidencePattern, exportRunAsOtlp, extractOtlpAttributes, extractUsage, extractUsageFromResponse, extractUsageFromSse, firstNumberAttr, firstStringAttr, flattenOtlpExportToNdjson, groupBy, hasCapturedToolArgs, inferDomainKeywords, inferOtlpKind, isJudgeSpan, isLlmSpan, isOtlpModelCall, isRetrievalSpan, isSandboxSpan, isToolSpan, iterateRawCalls, judgeSpans, llmSpanFromProvider, llmSpans, otelRunCompleteHook, otlpRowsToRunRecords, otlpRowsToTraceRunRecords, otlpTextToTraceAnalysisStore, otlpToRunRecords, otlpToTraceRunRecords, planTraceInsightQuestions, projectOtlpFlatLine, providerFromBaseUrl, readOtlpStatus, redactString, redactValue, runFailureClass, runsForScenario, scoreTraceInsightReadiness, stringField, throwIfRunIncomplete, tokenizeDomainWords, toolSpans, toolSpansToTraceAnalysisStore, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceSpanKindToOpenInferenceKind };
@@ -1,6 +1,6 @@
1
- import { c as CostLedgerHandle } from "./cost-ledger-FuQvHxPm.js";
2
- import { a as RunRecord, n as RunCostProvenance, u as RunTokenUsage } from "./run-record-ooo9FWns.js";
3
- import { A as ChatClient, m as JudgeInput } from "./types-BjMFz88h.js";
1
+ import { c as CostLedgerHandle } from "./cost-ledger-Bv_e8XHY.js";
2
+ import { a as RunRecord, n as RunCostProvenance, u as RunTokenUsage } from "./run-record-DdSa93_W.js";
3
+ import { A as ChatClient, m as JudgeInput } from "./types-XMVEdrE_.js";
4
4
  import { RE2JS } from "re2js";
5
5
  //#region src/trace-analyst/types.d.ts
6
6
  /**
@@ -545,4 +545,4 @@ type AnalystRunEvent = {
545
545
  };
546
546
  //#endregion
547
547
  export { TRACE_ANALYST_TRUNCATION_MARKER_PREFIX as A, DEFAULT_TRACE_ANALYST_BUDGETS as C, SearchSpanResult as D, QueryTracesPage as E, TraceAnalystSpanStatus as F, TraceAnalystTraceSummary as I, ViewSpansResult as L, TraceAnalystFilters as M, TraceAnalystSpan as N, SearchTraceResult as O, TraceAnalystSpanKind as P, ViewTraceOversized as R, TraceAnalysisStoreContext as S, ErrorCluster as T, makeFinding as _, AnalystInputKind as a, TRACE_ANALYSIS_LIMITS as b, AnalystRunInputs as c, AnalystSeverity as d, AnalystUsageReceipt as f, computeFindingId as g, ProposalFindingOrigin as h, AnalystFinding as i, TraceAnalystByteBudgets as j, SpanMatchRecord as k, AnalystRunResult as l, ProposalFinding as m, AnalystContext as n, AnalystRequirements as o, EvidenceRef as p, AnalystCost as r, AnalystRunEvent as s, Analyst as t, AnalystRunSummary as u, makeProposalFinding as v, DatasetOverview as w, TraceAnalysisStore as x, BoundedTraceAnalysisStoreOptions as y, ViewTraceResult as z };
548
- //# sourceMappingURL=types-CZt1PBIk.d.ts.map
548
+ //# sourceMappingURL=types-BhP9q0Fq.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"types-CZt1PBIk.d.ts","names":[],"sources":["../src/trace-analyst/types.ts","../src/trace-analyst/store-contract.ts","../src/analyst/types.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;KAgBY;KAUA;;;;UAKK;EACf;EACA;EACA;EACA;EACA,MAAM;EACN;EACA;EACA;EACA,QAAQ;EACR;EACA;EACA;EACA;EACA;;;EAGA,YAAY;;UAGG;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA;;;;;;;UAQe;;EAEf;;EAEA;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA;IAAU;IAAqB;;;;;;EAK/B,gBAAgB;EAChB;IAAc;IAAkB;;;UAGjB;EACf,QAAQ;EACR;EACA;;;;;UAMe;EACf;EACA,QAAQ;EACR,YAAY;;UAGG;EACf;;EAEA,gBAAgB;;EAEhB;EACA;;UAGe;EACf;EACA,OAAO;;EAEP;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA,WAAW;;;EAGX;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA,MAAM;EACN;;UAGe;EACf;EACA;EACA,MAAM;EACN;;;UAIe;;;EAGf;;;EAGA;;;EAGA;;;EAGA;;cAGW,+BAA+B;;;cAS/B;;;cC7MA;WACX;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;UAGe;EACf,SAAS;;;;;;;;;UAUM;EACf,SAAS,kBAAkB,UAAU,4BAA4B;EAEjE,SACE;IAAS;IAAkB;KAC3B,UAAU,4BACT;EAEH,YACE,UAAU,qBACV,UAAU,4BACT,QAAQ;EAEX,YACE;IAAS,UAAU;IAAqB;IAAe;KACvD,UAAU,4BACT,QAAQ;EAEX,YAAY,UAAU,qBAAqB,UAAU,4BAA4B;EAEjF,UACE;IACE;IACA;KAEF,UAAU,4BACT,QAAQ;EAEX,UACE;IACE;IACA;IACA;KAEF,UAAU,4BACT,QAAQ;EAEX,YACE;IACE;IACA;IACA;KAEF,UAAU,4BACT,QAAQ;EAEX,WACE;IACE;IACA;IACA;IACA;KAEF,UAAU,4BACT,QAAQ;;UAGI;EACf,UAAU,QAAQ;;;;;;;;UC/DH;EACf;;;;;;;EAOA;EACA;EACA;EACA,UAAU;;;;;;;EAOV;EACA;EACA;EACA,eAAe;EACf;EACA;;EAEA;;;;;;EAMA;;;;EAIA;;EAEA,WAAW;;KAGD;;KAGA;;KAGA,kBAAkB;WACnB,iBAAiB;;UAGX;;;;;;;EAOf;EACA;EACA;;;;;;;;KAWU;UAOK;;EAEf;;EAEA;;EAEA;;EAEA;;UAGe;;EAEf;;EAEA;;;;;;UAOe;EACf,aAAa;EACb;EACA,YAAY;EACZ,aAAa;;EAEb,SAAS;;UAGM;EACf;;EAEA;;EAEA;;EAEA;;EAEA,aAAa;;EAEb;;;;;;;EAOA,OAAO;;;;;;;;;;EAUP,gBAAgB,cAAc;;;;;;;EAO9B,mBAAmB,cAAc;;;;;EAKjC,eAAe,SAAS;;EAExB,OAAO;;EAEP,OAAO,aAAa,SAAS;;EAE7B,SAAS;;;;;;;;UASM,QAAQ;;WAEd;;WAEA;WACA,WAAW;WACX,MAAM;WACN,WAAW;;WAEX;EACT,QAAQ,OAAO,QAAQ,KAAK,iBAAiB,QAAQ;;;UAItC;;EAEf;;EAEA,QAAQ;;EAER,MAAM;;EAEN;;;;;;;;EAQA;IAAkB;IAAsB;;;;;;;;EAOxC;;;;;;;;;;iBAac,iBAAiB;EAC/B;EACA;EACA;EACA;;EAEA;;;;;;iBAyBc,YACd,MAAM,KAAK;EACT;EACA;IAED;;iBAiBa,oBACd,MAAM,KAAK;EACT;EACA;IAED;UAOc;EACf;EACA;;EAEA;EACA;EACA;;EAEA,OAAO;;EAEP;IAAU;IAAe;;;UAGV;EACf;EACA;EACA;EACA;EACA,UAAU;EACV,aAAa;;EAEb;;;;;EAKA,wBAAwB;;;;;;;;;;;;;;;KAkBd;EAEN;EACA;EACA;EACA;;EAEA,aAAa;;EAGb;EACA,SAAS;;EAGT;EACA;EACA;;EAGA;;EAEA,SAAS;EACT,UAAU,cAAc;;EAGxB;EACA,QAAQ"}
1
+ {"version":3,"file":"types-BhP9q0Fq.d.ts","names":[],"sources":["../src/trace-analyst/types.ts","../src/trace-analyst/store-contract.ts","../src/analyst/types.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;KAgBY;KAUA;;;;UAKK;EACf;EACA;EACA;EACA;EACA,MAAM;EACN;EACA;EACA;EACA,QAAQ;EACR;EACA;EACA;EACA;EACA;;;EAGA,YAAY;;UAGG;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA;;;;;;;UAQe;;EAEf;;EAEA;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA;IAAU;IAAqB;;;;;;EAK/B,gBAAgB;EAChB;IAAc;IAAkB;;;UAGjB;EACf,QAAQ;EACR;EACA;;;;;UAMe;EACf;EACA,QAAQ;EACR,YAAY;;UAGG;EACf;;EAEA,gBAAgB;;EAEhB;EACA;;UAGe;EACf;EACA,OAAO;;EAEP;;EAEA;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA;EACA,WAAW;;;EAGX;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA,MAAM;EACN;;UAGe;EACf;EACA;EACA,MAAM;EACN;;;UAIe;;;EAGf;;;EAGA;;;EAGA;;;EAGA;;cAGW,+BAA+B;;;cAS/B;;;cC7MA;WACX;WACA;WACA;WACA;WACA;WACA;WACA;WACA;;UAGe;EACf,SAAS;;;;;;;;;UAUM;EACf,SAAS,kBAAkB,UAAU,4BAA4B;EAEjE,SACE;IAAS;IAAkB;KAC3B,UAAU,4BACT;EAEH,YACE,UAAU,qBACV,UAAU,4BACT,QAAQ;EAEX,YACE;IAAS,UAAU;IAAqB;IAAe;KACvD,UAAU,4BACT,QAAQ;EAEX,YAAY,UAAU,qBAAqB,UAAU,4BAA4B;EAEjF,UACE;IACE;IACA;KAEF,UAAU,4BACT,QAAQ;EAEX,UACE;IACE;IACA;IACA;KAEF,UAAU,4BACT,QAAQ;EAEX,YACE;IACE;IACA;IACA;KAEF,UAAU,4BACT,QAAQ;EAEX,WACE;IACE;IACA;IACA;IACA;KAEF,UAAU,4BACT,QAAQ;;UAGI;EACf,UAAU,QAAQ;;;;;;;;UC/DH;EACf;;;;;;;EAOA;EACA;EACA;EACA,UAAU;;;;;;;EAOV;EACA;EACA;EACA,eAAe;EACf;EACA;;EAEA;;;;;;EAMA;;;;EAIA;;EAEA,WAAW;;KAGD;;KAGA;;KAGA,kBAAkB;WACnB,iBAAiB;;UAGX;;;;;;;EAOf;EACA;EACA;;;;;;;;KAWU;UAOK;;EAEf;;EAEA;;EAEA;;EAEA;;UAGe;;EAEf;;EAEA;;;;;;UAOe;EACf,aAAa;EACb;EACA,YAAY;EACZ,aAAa;;EAEb,SAAS;;UAGM;EACf;;EAEA;;EAEA;;EAEA;;EAEA,aAAa;;EAEb;;;;;;;EAOA,OAAO;;;;;;;;;;EAUP,gBAAgB,cAAc;;;;;;;EAO9B,mBAAmB,cAAc;;;;;EAKjC,eAAe,SAAS;;EAExB,OAAO;;EAEP,OAAO,aAAa,SAAS;;EAE7B,SAAS;;;;;;;;UASM,QAAQ;;WAEd;;WAEA;WACA,WAAW;WACX,MAAM;WACN,WAAW;;WAEX;EACT,QAAQ,OAAO,QAAQ,KAAK,iBAAiB,QAAQ;;;UAItC;;EAEf;;EAEA,QAAQ;;EAER,MAAM;;EAEN;;;;;;;;EAQA;IAAkB;IAAsB;;;;;;;;EAOxC;;;;;;;;;;iBAac,iBAAiB;EAC/B;EACA;EACA;EACA;;EAEA;;;;;;iBAyBc,YACd,MAAM,KAAK;EACT;EACA;IAED;;iBAiBa,oBACd,MAAM,KAAK;EACT;EACA;IAED;UAOc;EACf;EACA;;EAEA;EACA;EACA;;EAEA,OAAO;;EAEP;IAAU;IAAe;;;UAGV;EACf;EACA;EACA;EACA;EACA,UAAU;EACV,aAAa;;EAEb;;;;;EAKA,wBAAwB;;;;;;;;;;;;;;;KAkBd;EAEN;EACA;EACA;EACA;;EAEA,aAAa;;EAGb;EACA,SAAS;;EAGT;EACA;EACA;;EAGA;;EAEA,SAAS;EACT,UAAU,cAAc;;EAGxB;EACA,QAAQ"}
@@ -1,7 +1,7 @@
1
- import { S as PaidCallResult, T as RunPaidCallInput, a as CostChannel, c as CostLedgerHandle, f as CostLedgerSummary, p as CostProvenance } from "./cost-ledger-FuQvHxPm.js";
2
- import { u as RunTokenUsage } from "./run-record-ooo9FWns.js";
3
- import { U as LlmCallMetadata } from "./types-BjMFz88h.js";
4
- import { m as ProposalFinding } from "./types-CZt1PBIk.js";
1
+ import { S as PaidCallResult, T as RunPaidCallInput, a as CostChannel, c as CostLedgerHandle, f as CostLedgerSummary, p as CostProvenance } from "./cost-ledger-Bv_e8XHY.js";
2
+ import { u as RunTokenUsage } from "./run-record-DdSa93_W.js";
3
+ import { U as LlmCallMetadata } from "./types-XMVEdrE_.js";
4
+ import { m as ProposalFinding } from "./types-BhP9q0Fq.js";
5
5
  //#region src/campaign/types.d.ts
6
6
  /** Stable identifier + kind tag for any scenario. Consumers
7
7
  * extend with their per-domain payload (persona, task, requirement, ...). */
@@ -624,4 +624,4 @@ interface CampaignResult<TArtifact = unknown, TScenario extends Scenario = Scena
624
624
  }
625
625
  //#endregion
626
626
  export { LabeledScenarioWrite as A, ScoredSurfaceOutcome as B, JudgeDimension as C, LabeledScenarioSampleArgs as D, LabeledScenarioRecord as E, ProposeContext as F, labelTrustRank as G, SurfaceProposer as H, ProposedCandidate as I, RedactionStatus as L, OptimizerConfig as M, ParetoParent as N, LabeledScenarioSource as O, ProposalTrackContext as P, Scenario as R, JudgeConfig as S, LabelTrust as T, TraceSpan as U, SessionScript as V, isProposedCandidate as W, GateDecision as _, CampaignResult as a, GenerationRecord as b, CampaignTraceWriter as c, DispatchContext as d, DispatchFn as f, GateContribution as g, GateContext as h, CampaignCostMeter as i, MutableSurface as j, LabeledScenarioStore as k, CodeSurface as l, GateCheckStatus as m, CampaignArtifactWriter as n, CampaignScenarioIdentity as o, Gate as p, CampaignCellResult as r, CampaignTokenUsage as s, CampaignAggregates as t, ComponentSurface as u, GateResult as v, JudgeScore as w, JudgeAggregate as x, GenerationCandidate as y, ScenarioAggregate as z };
627
- //# sourceMappingURL=types-Dcoaqcsc.d.ts.map
627
+ //# sourceMappingURL=types-DOZyvsFU.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"types-Dcoaqcsc.d.ts","names":[],"sources":["../src/campaign/types.ts"],"mappings":";;;;;;;UAgCiB;EACf;EACA;EACA;;;;;EAKA;;;UAIe,iCAAiC,KAAK;EACrD;;;;;UAMe;EACf;EACA;EACA;EACA;EACA,QAAQ;EACR,OAAO;EACP,WAAW;EACX,MAAM;;EAEN;;EAEA;;;;;;;;EAQA;;;;KAKU,WAAW,kBAAkB,UAAU,cACjD,UAAU,WACV,KAAK,oBACF,QAAQ;;;;UAOI,cAAc,WAAW;EACxC;EACA;EACA;;EAEA;;;EAGA,sBAAsB,UAAU,WAAW,sBAAsB,UAAU,cAAc;;UAK1E;;EAEf;;EAEA;;;;;;;;;UAUe,YAAY,WAAW,kBAAkB,WAAW;EACnE;EACA,YAAY;;;;EAIZ;;;EAGA,MAAM;IACJ,UAAU;IACV,UAAU;IACV,QAAQ;;IAER,aAAa;IACb;IACA,WAAW;MACT,aAAa,QAAQ;EACzB,aAAa,UAAU;;;;;;;;;;;UAYR;EACf,YAAY;EACZ;EACA;;EAEA,UAAU;;;;EAIV;;;EAGA;;EAEA;;EAEA,WAAW,eAAe;;;;;;;UAUX;WACN;;;WAGA;;WAEA;;WAEA;;WAEA;;WAEA;;WAEA;;;WAGA;aACE;aACA;aACA;;;WAGF;;;UAIM;WACN;WACA,YAAY,SAAS;;;;;;;;;;;KAYpB,0BAA0B,mBAAmB;;;;;;;UAQxC;EACf,SAAS;;EAET;;;;EAIA;;;;iBAKc,oBACd,OAAO,iBAAiB,oBACvB,SAAS;;;;;;;;;UAkBK;EACf,SAAS;EACT;;;EAGA,YAAY;;;EAGZ;;EAEA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;;;;;UAMe;;;EAGf;;EAEA;EACA;EACA;EACA,YAAY;;;;;EAKZ,WAAW;IAAQ;IAAoB;IAAmB;IAAgB;;EAC1E;IACE;IACA;;;;;UAMa,eAAe,YAAY;WACjC,gBAAgB;WAChB,SAAS,cAAc;WACvB,UAAU,cAAc;;WAExB;WACA;WACA,QAAQ;;WAER,QAAQ;;;WAGR,kBAAkB;;;WAGlB,mBAAmB;;;;WAInB;;;;;;;WAOA,gBAAgB,cAAc;;WAE9B,aAAa;WACb;;;;;;;;;;;;;;UAeM,gBAAgB,YAAY;EAC3C;;;;;EAKA,QAAQ,KAAK,eAAe,aAAa,QAAQ,MAAM,iBAAiB;;;EAGxE,QAAQ;IAAQ,SAAS,cAAc;;IAAwB;IAAe;;;UAG/D;EACf;EACA;EACA,mBAAmB,qBAAqB;;UAGzB,wBAAwB;EACvC,UAAU;;;KAMA;;KAGA;UAEK;EACf;EACA,QAAQ;EACR;;UAGe,YAAY,WAAW,kBAAkB;EACxD,oBAAoB,YAAY;EAChC,oBAAoB,YAAY;;EAEhC,aAAa,YAAY,eAAe;;;;;EAKxC,sBAAsB,YAAY,eAAe;;;;;;;EAOjD,yBAAyB,YAAY,eAAe;;;EAGpD,uBAAuB,YAAY;EACnC,WAAW;EACX;IAAQ;IAAmB;;;EAE3B,aAAa;EACb;EACA,QAAQ;;UAGO;EACf,UAAU;EACV;EACA,mBAAmB;EACnB;;;UAIe,KAAK,qBAAqB,kBAAkB,WAAW;EACtE;EACA,OAAO,KAAK,YAAY,WAAW,aAAa,QAAQ;;;;UAOzC;EACf,KAAK,cAAc,aAAa,0BAA0B;EAC1D,SAAS;;UAGM;EACf,IAAI,aAAa;EACjB,aAAa,aAAa;;;;UAKX;EACf,MAAM,cAAc,kBAAkB,aAAa;EACnD,UAAU,cAAc,iBAAiB;;;;;;KAO/B,qBAAqB;;;;;UAMhB;;EAEf,YAAY,GACV,OAAO,KAAK,iBAAiB;IAC3B,UAAU;MAEX,QAAQ,eAAe;;;;;KAQhB;KAOA;;;;;;;;;;;;;;;KAgBA;;iBASI,eAAe,OAAO;;;;UAOrB,qBAAqB,kBAAkB,WAAW,UAAU;EAC3E,UAAU;EACV,UAAU;EACV,aAAa,eAAe;EAC5B,QAAQ;EACR;EACA;EACA,iBAAiB;;;;;EAKjB,aAAa;;EAEb;;UAGe,sBAAsB,kBAAkB,WAAW,UAAU,6BACpE,qBAAqB,WAAW;;EAExC;;;EAGA;;UAGe;EACf;;EAEA;;;;EAIA;EACA;IACE;IACA,SAAS,wBAAwB;IACjC;IACA;;;;;IAKA,WAAW;;;UAIE;EACf,QAAQ,OAAO,uBAAuB;EACtC,OAAO,MAAM,4BAA4B,QAAQ;EACjD,QAAQ;IACN;IACA;IACA,UAAU;;;IAGV,SAAS,OAAO;;;UAMH,mBAAmB;;;EAGlC;EACA;EACA;EACA;EACA;EACA,UAAU;EACV,aAAa,eAAe;;;EAG5B;;EAEA,gBAAgB;;EAEhB;;;EAGA,YAAY;;;EAGZ;;;EAGA;EACA;EACA;EACA;;EAEA;;EAEA;EACA;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;;UAGe;EACf;EACA,YAAY;EACZ;;;;;;UAOe;EACf;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA;;;;EAIA;;;;EAIA;IACE;IACA;IACA,iBAAiB;MAAQ;MAAgB;;;;;EAI3C,YAAY;;;;;;;;;;EAUZ,WAAW;IAAQ;IAAoB;IAAmB;IAAgB;;;;EAG1E;;;;EAIA;;UAGe;EACf,SAAS,eAAe;EACxB,YAAY,eAAe;;EAE3B,MAAM;;EAEN;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe,eAAe,qBAAqB,kBAAkB,WAAW;;EAEhF;;EAEA;EACA;;EAEA;EACA;EACA;EACA;EACA,OAAO,MAAM,mBAAmB;EAChC,YAAY;EACZ;IACE,aAAa;IACb;;EAEF,OAAO;EACP;EACA;EACA,iBAAiB;;;EAGjB,WAAW,MAAM,2BAA2B,KAAK"}
1
+ {"version":3,"file":"types-DOZyvsFU.d.ts","names":[],"sources":["../src/campaign/types.ts"],"mappings":";;;;;;;UAgCiB;EACf;EACA;EACA;;;;;EAKA;;;UAIe,iCAAiC,KAAK;EACrD;;;;;UAMe;EACf;EACA;EACA;EACA;EACA,QAAQ;EACR,OAAO;EACP,WAAW;EACX,MAAM;;EAEN;;EAEA;;;;;;;;EAQA;;;;KAKU,WAAW,kBAAkB,UAAU,cACjD,UAAU,WACV,KAAK,oBACF,QAAQ;;;;UAOI,cAAc,WAAW;EACxC;EACA;EACA;;EAEA;;;EAGA,sBAAsB,UAAU,WAAW,sBAAsB,UAAU,cAAc;;UAK1E;;EAEf;;EAEA;;;;;;;;;UAUe,YAAY,WAAW,kBAAkB,WAAW;EACnE;EACA,YAAY;;;;EAIZ;;;EAGA,MAAM;IACJ,UAAU;IACV,UAAU;IACV,QAAQ;;IAER,aAAa;IACb;IACA,WAAW;MACT,aAAa,QAAQ;EACzB,aAAa,UAAU;;;;;;;;;;;UAYR;EACf,YAAY;EACZ;EACA;;EAEA,UAAU;;;;EAIV;;;EAGA;;EAEA;;EAEA,WAAW,eAAe;;;;;;;UAUX;WACN;;;WAGA;;WAEA;;WAEA;;WAEA;;WAEA;;WAEA;;;WAGA;aACE;aACA;aACA;;;WAGF;;;UAIM;WACN;WACA,YAAY,SAAS;;;;;;;;;;;KAYpB,0BAA0B,mBAAmB;;;;;;;UAQxC;EACf,SAAS;;EAET;;;;EAIA;;;;iBAKc,oBACd,OAAO,iBAAiB,oBACvB,SAAS;;;;;;;;;UAkBK;EACf,SAAS;EACT;;;EAGA,YAAY;;;EAGZ;;EAEA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;;;;;UAMe;;;EAGf;;EAEA;EACA;EACA;EACA,YAAY;;;;;EAKZ,WAAW;IAAQ;IAAoB;IAAmB;IAAgB;;EAC1E;IACE;IACA;;;;;UAMa,eAAe,YAAY;WACjC,gBAAgB;WAChB,SAAS,cAAc;WACvB,UAAU,cAAc;;WAExB;WACA;WACA,QAAQ;;WAER,QAAQ;;;WAGR,kBAAkB;;;WAGlB,mBAAmB;;;;WAInB;;;;;;;WAOA,gBAAgB,cAAc;;WAE9B,aAAa;WACb;;;;;;;;;;;;;;UAeM,gBAAgB,YAAY;EAC3C;;;;;EAKA,QAAQ,KAAK,eAAe,aAAa,QAAQ,MAAM,iBAAiB;;;EAGxE,QAAQ;IAAQ,SAAS,cAAc;;IAAwB;IAAe;;;UAG/D;EACf;EACA;EACA,mBAAmB,qBAAqB;;UAGzB,wBAAwB;EACvC,UAAU;;;KAMA;;KAGA;UAEK;EACf;EACA,QAAQ;EACR;;UAGe,YAAY,WAAW,kBAAkB;EACxD,oBAAoB,YAAY;EAChC,oBAAoB,YAAY;;EAEhC,aAAa,YAAY,eAAe;;;;;EAKxC,sBAAsB,YAAY,eAAe;;;;;;;EAOjD,yBAAyB,YAAY,eAAe;;;EAGpD,uBAAuB,YAAY;EACnC,WAAW;EACX;IAAQ;IAAmB;;;EAE3B,aAAa;EACb;EACA,QAAQ;;UAGO;EACf,UAAU;EACV;EACA,mBAAmB;EACnB;;;UAIe,KAAK,qBAAqB,kBAAkB,WAAW;EACtE;EACA,OAAO,KAAK,YAAY,WAAW,aAAa,QAAQ;;;;UAOzC;EACf,KAAK,cAAc,aAAa,0BAA0B;EAC1D,SAAS;;UAGM;EACf,IAAI,aAAa;EACjB,aAAa,aAAa;;;;UAKX;EACf,MAAM,cAAc,kBAAkB,aAAa;EACnD,UAAU,cAAc,iBAAiB;;;;;;KAO/B,qBAAqB;;;;;UAMhB;;EAEf,YAAY,GACV,OAAO,KAAK,iBAAiB;IAC3B,UAAU;MAEX,QAAQ,eAAe;;;;;KAQhB;KAOA;;;;;;;;;;;;;;;KAgBA;;iBASI,eAAe,OAAO;;;;UAOrB,qBAAqB,kBAAkB,WAAW,UAAU;EAC3E,UAAU;EACV,UAAU;EACV,aAAa,eAAe;EAC5B,QAAQ;EACR;EACA;EACA,iBAAiB;;;;;EAKjB,aAAa;;EAEb;;UAGe,sBAAsB,kBAAkB,WAAW,UAAU,6BACpE,qBAAqB,WAAW;;EAExC;;;EAGA;;UAGe;EACf;;EAEA;;;;EAIA;EACA;IACE;IACA,SAAS,wBAAwB;IACjC;IACA;;;;;IAKA,WAAW;;;UAIE;EACf,QAAQ,OAAO,uBAAuB;EACtC,OAAO,MAAM,4BAA4B,QAAQ;EACjD,QAAQ;IACN;IACA;IACA,UAAU;;;IAGV,SAAS,OAAO;;;UAMH,mBAAmB;;;EAGlC;EACA;EACA;EACA;EACA;EACA,UAAU;EACV,aAAa,eAAe;;;EAG5B;;EAEA,gBAAgB;;EAEhB;;;EAGA,YAAY;;;EAGZ;;;EAGA;EACA;EACA;EACA;;EAEA;;EAEA;EACA;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;;UAGe;EACf;EACA,YAAY;EACZ;;;;;;UAOe;EACf;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA;;;;EAIA;;;;EAIA;IACE;IACA;IACA,iBAAiB;MAAQ;MAAgB;;;;;EAI3C,YAAY;;;;;;;;;;EAUZ,WAAW;IAAQ;IAAoB;IAAmB;IAAgB;;;;EAG1E;;;;EAIA;;UAGe;EACf,SAAS,eAAe;EACxB,YAAY,eAAe;;EAE3B,MAAM;;EAEN;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe,eAAe,qBAAqB,kBAAkB,WAAW;;EAEhF;;EAEA;EACA;;EAEA;EACA;EACA;EACA;EACA,OAAO,MAAM,mBAAmB;EAChC,YAAY;EACZ;IACE,aAAa;IACb;;EAEF,OAAO;EACP;EACA;EACA,iBAAiB;;;EAGjB,WAAW,MAAM,2BAA2B,KAAK"}
@@ -1,5 +1,5 @@
1
- import { r as CaptureIntegrityError, t as AgentEvalError } from "./errors-DkfjIDvD.js";
2
- import { b as CustomTokenPricing, c as CostLedgerHandle, f as CostLedgerSummary, g as CostReceiptInput, x as MaximumCharge } from "./cost-ledger-FuQvHxPm.js";
1
+ import { r as CaptureIntegrityError, t as AgentEvalError } from "./errors-CKPfb2aH.js";
2
+ import { b as CustomTokenPricing, c as CostLedgerHandle, f as CostLedgerSummary, g as CostReceiptInput, x as MaximumCharge } from "./cost-ledger-Bv_e8XHY.js";
3
3
  //#region src/trace/raw-provider-sink.d.ts
4
4
  /**
5
5
  * RawProviderSink — first-class persistence for the actual HTTP-level
@@ -132,6 +132,169 @@ declare class FileSystemRawProviderSink implements RawProviderSink {
132
132
  */
133
133
  declare function providerFromBaseUrl(baseUrl: string): string;
134
134
  //#endregion
135
+ //#region src/judge-families.d.ts
136
+ /**
137
+ * Judge model-family classification + cross-family enforcement.
138
+ *
139
+ * A judge ensemble built entirely from one provider family shares that
140
+ * family's blind spots and self-preference — its "agreement" is correlated
141
+ * bias, not independent signal. `assertCrossFamily` makes the consumer prove
142
+ * the ensemble spans ≥2 families; `judgeFamily` is the single regex map that
143
+ * replaces the per-consumer copies (tax/legal/creative/gtm each ship one).
144
+ */
145
+ /** Provider family a model belongs to. `unknown` when no rule matches. */
146
+ type JudgeFamily = 'anthropic' | 'openai' | 'google' | 'meta' | 'mistral' | 'deepseek' | 'xai' | 'qwen' | 'cohere' | 'amazon' | 'moonshot' | 'zhipu' | 'unknown';
147
+ /**
148
+ * Classify a model id into its provider family. Strips a `@snapshot` suffix
149
+ * and prefers an explicit `provider/...` prefix; otherwise matches the model
150
+ * name. Returns `unknown` when nothing matches (callers decide whether that's
151
+ * acceptable — `assertCrossFamily` counts it as its own family).
152
+ */
153
+ declare function judgeFamily(modelId: string): JudgeFamily;
154
+ interface AssertCrossFamilyOptions {
155
+ /** Minimum number of distinct families the ensemble must span. Default 2. */
156
+ minFamilies?: number;
157
+ /** When false (default), `unknown`-family models do NOT count toward the
158
+ * family total — an ensemble of all-unclassifiable models is not provably
159
+ * cross-family. Set true to count `unknown` as one shared family. */
160
+ allowUnknown?: boolean;
161
+ }
162
+ declare class CrossFamilyError extends Error {
163
+ readonly families: JudgeFamily[];
164
+ readonly models: string[];
165
+ constructor(message: string, families: JudgeFamily[], models: string[]);
166
+ }
167
+ /**
168
+ * Throw unless the judge models span at least `minFamilies` distinct provider
169
+ * families. Pass the model ids backing your judge ensemble. Fail-loud by
170
+ * design — a correlated single-family ensemble silently inflates agreement.
171
+ *
172
+ * Scope: this reads the ids you REQUEST. It proves the panel was configured
173
+ * across families; it cannot prove the panel RAN across families, because a
174
+ * routing gateway may answer several different ids from one provider. Where
175
+ * the diversity claim is load-bearing (a published leaderboard, a
176
+ * certification, a non-self-judging exclusion), assert on the ids the
177
+ * provider echoed instead: `assertCrossFamilyServed` in
178
+ * ./integrity/served-model.
179
+ */
180
+ declare function assertCrossFamily(models: string[], opts?: AssertCrossFamilyOptions): JudgeFamily[];
181
+ //#endregion
182
+ //#region src/integrity/served-model.d.ts
183
+ /** How a served id relates to the id that was requested. */
184
+ type ServedModelVerdict =
185
+ /** Byte-identical after normalisation — the requested model answered. */
186
+ 'exact' |
187
+ /** Same model, different spelling (provider prefix, snapshot, tier suffix). */
188
+ 'alias' |
189
+ /** A different model of the SAME provider family answered. */
190
+ 'substituted-within-family' |
191
+ /** A different provider's model answered. */
192
+ 'substituted-cross-family' |
193
+ /** The response carried no model id — identity is unproven either way. */
194
+ 'unreported';
195
+ interface ServedModelCheck {
196
+ /** The id the caller asked for. */
197
+ requested: string;
198
+ /** The id echoed on the response; `null` when the response omitted it. */
199
+ served: string | null;
200
+ requestedFamily: JudgeFamily;
201
+ /** `null` when `served` is null. */
202
+ servedFamily: JudgeFamily | null;
203
+ verdict: ServedModelVerdict;
204
+ /** True for every verdict except `exact` and `alias`. */
205
+ substituted: boolean;
206
+ }
207
+ /**
208
+ * Reduce a model id to its comparable core: lowercase, no surrounding space,
209
+ * no `provider/` prefix, no `@snapshot` / `:batch` / `:free` tier suffix, no
210
+ * trailing build date, and `.`/`_` folded to `-` so one version is spelled one
211
+ * way.
212
+ *
213
+ * Dropping the build date is what makes snapshot resolution legible as the
214
+ * non-event it is: a router answering `gpt-4o-mini` with
215
+ * `gpt-4o-mini-2024-07-18` pinned a floating alias to a reproducible build —
216
+ * the same model, which is the behaviour we want. Only routing decoration is
217
+ * stripped; version digits are load-bearing, so `deepseek-v3.2` and
218
+ * `deepseek-v4-flash` stay distinct, and comparison is EXACT equality rather
219
+ * than a prefix test (a prefix rule would accept `gpt-5` → `gpt-5-mini`, a
220
+ * silent downgrade wearing the right vendor name).
221
+ */
222
+ declare function normalizeModelId(modelId: string): string;
223
+ /**
224
+ * Classify one requested/served pair. Pure — no I/O — so it is safe inside
225
+ * response handlers, reducers, and CI gates.
226
+ *
227
+ * `served` is the id echoed by the provider (OpenAI-compatible bodies put it
228
+ * at `model`). `null`/`undefined` means the body omitted it; that is
229
+ * `unreported`, NOT a pass — a provider that does not name what answered has
230
+ * not proven identity, and a transport that drops the field must not read as
231
+ * agreement.
232
+ */
233
+ declare function checkServedModel(requested: string, served: string | null | undefined): ServedModelCheck;
234
+ declare class ModelSubstitutionError extends AgentEvalError {
235
+ readonly checks: ReadonlyArray<ServedModelCheck>;
236
+ constructor(message: string, checks: ReadonlyArray<ServedModelCheck>);
237
+ }
238
+ interface AssertServedModelOptions {
239
+ /**
240
+ * Accept a different model of the same provider family (e.g. requested
241
+ * `deepseek-v3.2`, served `deepseek-v4-flash`). Default false. Setting this
242
+ * keeps family-level claims valid and forfeits per-model claims.
243
+ */
244
+ allowWithinFamily?: boolean;
245
+ /**
246
+ * Accept a response that carried no model id. Default false — an
247
+ * unidentified response cannot support a per-model or per-family claim.
248
+ */
249
+ allowUnreported?: boolean;
250
+ /** Prefixed to the thrown message, e.g. the judge or campaign cell name. */
251
+ context?: string;
252
+ }
253
+ /**
254
+ * The one place the accept/reject policy lives, so a caller that reports
255
+ * substitution (a preflight table, a run record) and a caller that throws on it
256
+ * can never drift apart. A cross-family substitution is never acceptable.
257
+ */
258
+ declare function servedModelAcceptable(check: ServedModelCheck, opts?: AssertServedModelOptions): boolean;
259
+ /**
260
+ * Throw `ModelSubstitutionError` unless the served id is the requested model.
261
+ * Returns the check on success so callers can record the served id alongside
262
+ * the result.
263
+ */
264
+ declare function assertServedModel(requested: string, served: string | null | undefined, opts?: AssertServedModelOptions): ServedModelCheck;
265
+ /**
266
+ * Batch form: check every pair and throw naming EVERY substitution, so one
267
+ * failure does not hide the rest. Returns all checks on success.
268
+ */
269
+ declare function assertServedModels(pairs: ReadonlyArray<{
270
+ requested: string;
271
+ served: string | null | undefined;
272
+ }>, opts?: AssertServedModelOptions): ServedModelCheck[];
273
+ interface AssertCrossFamilyServedOptions extends AssertServedModelOptions {
274
+ /** Minimum distinct SERVED families required. Default 2. */
275
+ minFamilies?: number;
276
+ /** Count `unknown`-family served ids toward the total. Default false. */
277
+ allowUnknown?: boolean;
278
+ }
279
+ declare class ServedCrossFamilyError extends AgentEvalError {
280
+ readonly families: JudgeFamily[];
281
+ readonly checks: ReadonlyArray<ServedModelCheck>;
282
+ constructor(message: string, families: JudgeFamily[], checks: ReadonlyArray<ServedModelCheck>);
283
+ }
284
+ /**
285
+ * Family-diversity rule over the models that actually ANSWERED.
286
+ *
287
+ * `assertCrossFamily` (../judge-families) reads the requested ids and so
288
+ * cannot see a gateway that answers three "different" requests from one
289
+ * provider. This one asserts no substitution first, then counts families from
290
+ * the served ids — a panel that collapsed to one family under the hood fails
291
+ * here even though its request list looked diverse.
292
+ */
293
+ declare function assertCrossFamilyServed(pairs: ReadonlyArray<{
294
+ requested: string;
295
+ served: string | null | undefined;
296
+ }>, opts?: AssertCrossFamilyServedOptions): JudgeFamily[];
297
+ //#endregion
135
298
  //#region src/llm-client.d.ts
136
299
  interface LlmMessage {
137
300
  role: 'system' | 'user' | 'assistant';
@@ -193,8 +356,27 @@ interface LlmCallResult {
193
356
  * caller-supplied token pricing. `null` when neither is available.
194
357
  */
195
358
  costUsd: number | null;
196
- /** Model name actually used (echoed from response). */
359
+ /**
360
+ * Model id used for attribution (cost, pricing, log lines). The response's
361
+ * echoed id when the provider sent one, else the requested id.
362
+ *
363
+ * NOT evidence of which model answered — read `servedModel` for that. A
364
+ * provider that omits `model` makes this equal to the request, which is
365
+ * exactly the case an identity check must be able to distinguish.
366
+ */
197
367
  model: string;
368
+ /**
369
+ * The model id the provider echoed on the response, verbatim; `null` when
370
+ * the body carried none. This is the only field that can witness a gateway
371
+ * substituting a different model for the one requested — compare it with
372
+ * `assertServedModel` / `checkServedModel` (src/integrity/served-model.ts).
373
+ *
374
+ * Optional so hand-built results (mock/custom transports) still typecheck,
375
+ * but omitting it is not a pass: the identity checks read `undefined` as
376
+ * `unreported` and reject it by default. A transport that knows which model
377
+ * answered should say so.
378
+ */
379
+ servedModel?: string | null;
198
380
  /** Wall-clock duration of the HTTP call (last attempt, if retried). */
199
381
  durationMs: number;
200
382
  /**
@@ -307,6 +489,16 @@ interface LlmClientOptions {
307
489
  };
308
490
  /** Override the redaction strategy for this call. Defaults to `defaultProviderRedactor`. */
309
491
  redactor?: ProviderRedactor;
492
+ /**
493
+ * Reject a response whose echoed model is not the model that was requested.
494
+ * A routing gateway can accept one id and answer from another, which
495
+ * silently invalidates every per-model and per-family claim downstream.
496
+ * `true` uses the strict default (aliases pass, substitutions and
497
+ * unidentified responses throw `ModelSubstitutionError`); pass an options
498
+ * object to relax a specific case. Off by default — turning it on for a
499
+ * measurement run is the point.
500
+ */
501
+ assertServedModel?: boolean | AssertServedModelOptions;
310
502
  }
311
503
  /**
312
504
  * True when an error is a transient transport/network fault worth retrying,
@@ -392,7 +584,12 @@ declare function assertLlmRoute(opts: LlmClientOptions, req?: LlmRouteRequiremen
392
584
  * Sends a tiny `ping` message with `maxTokens=64`. Reasoning models
393
585
  * (glm-5.1, deepseek-v4) can burn the entire budget on internal reasoning
394
586
  * for short prompts, so don't tighten this further. We don't validate
395
- * content; HTTP 200 means reachable.
587
+ * content.
588
+ *
589
+ * Reachability and identity are separate answers: `ok` means the route
590
+ * answered, `servedModel` / `substituted` say WHICH model answered. A gateway
591
+ * that serves another provider's model returns `ok: true` with
592
+ * `substituted: true` — inspect both before treating the id as measured.
396
593
  */
397
594
  declare function probeLlm(model: string, opts?: LlmClientOptions & {
398
595
  timeoutMs?: number;
@@ -400,6 +597,10 @@ declare function probeLlm(model: string, opts?: LlmClientOptions & {
400
597
  ok: boolean;
401
598
  latencyMs: number;
402
599
  error: string | null;
600
+ /** Id echoed by the provider; `null` when it sent none or the probe failed. */
601
+ servedModel: string | null;
602
+ /** True when the echoed id is a different model than `model` (or absent). */
603
+ substituted: boolean;
403
604
  }>;
404
605
  /**
405
606
  * Stateful client — construct once with defaults, call many times.
@@ -800,5 +1001,5 @@ interface EvalResult {
800
1001
  artifact?: string;
801
1002
  }
802
1003
  //#endregion
803
- export { assertLlmRoute as $, ChatClient as A, SandboxSdkTransportOpts as B, ScenarioFile as C, TurnMetrics as D, Turn as E, CreateChatClientOpts as F, LlmCallResult as G, LlmCallError as H, CustomTransportOpts as I, LlmMessage as J, LlmClient as K, DirectProviderTransportOpts as L, ChatResponse as M, ChatTransport as N, TurnResult as O, CliBridgeTransportOpts as P, LlmUsage as Q, MockTransportOpts as R, Scenario as S, TestResult as T, LlmCallMetadata as U, createChatClient as V, LlmCallRequest as W, LlmRouteAssertionError as X, LlmResponseError as Y, LlmRouteRequirements as Z, PersonaConfig as _, RawProviderSink as _t, CheckResult as a, isTransientLlmError as at, RouteMap as b, providerFromBaseUrl as bt, DriverResult as c, stripFencedJson as ct, FeedbackPattern as d, InMemoryRawProviderSink as dt, backoffMs as et, JudgeConfig as f, InMemoryRawProviderSinkOptions as ft, JudgeScore as g, RawProviderEvent as gt, JudgeRubric as h, RawProviderDirection as ht, BenchmarkRunnerConfig as i, costReceiptFromLlmError as it, ChatRequest as j, ChatCallOpts as k, DriverState as l, FileSystemRawProviderSink as lt, JudgeInput as m, ProviderRedactor as mt, ArtifactResult as n, callLlmJson as nt, CollectedArtifacts as o, maximumChargeForLlmRequest as ot, JudgeFn as p, NoopRawProviderSink as pt, LlmClientOptions as q, BenchmarkReport as r, costReceiptFromLlm as rt, CompletionCriterion as s, probeLlm as st, ArtifactCheck as t, callLlm as tt, EvalResult as u, FileSystemRawProviderSinkOptions as ut, PersonaRigor as v, RawProviderSinkFilter as vt, ScenarioResult as w, RubricDimension as x, ProductClientConfig as y, defaultProviderRedactor as yt, RouterTransportOpts as z };
804
- //# sourceMappingURL=types-BjMFz88h.d.ts.map
1004
+ export { assertLlmRoute as $, ChatClient as A, NoopRawProviderSink as At, SandboxSdkTransportOpts as B, ScenarioFile as C, JudgeFamily as Ct, TurnMetrics as D, FileSystemRawProviderSinkOptions as Dt, Turn as E, FileSystemRawProviderSink as Et, CreateChatClientOpts as F, RawProviderSinkFilter as Ft, LlmCallResult as G, LlmCallError as H, CustomTransportOpts as I, defaultProviderRedactor as It, LlmMessage as J, LlmClient as K, DirectProviderTransportOpts as L, providerFromBaseUrl as Lt, ChatResponse as M, RawProviderDirection as Mt, ChatTransport as N, RawProviderEvent as Nt, TurnResult as O, InMemoryRawProviderSink as Ot, CliBridgeTransportOpts as P, RawProviderSink as Pt, LlmUsage as Q, MockTransportOpts as R, Scenario as S, CrossFamilyError as St, TestResult as T, judgeFamily as Tt, LlmCallMetadata as U, createChatClient as V, LlmCallRequest as W, LlmRouteAssertionError as X, LlmResponseError as Y, LlmRouteRequirements as Z, PersonaConfig as _, assertServedModels as _t, CheckResult as a, isTransientLlmError as at, RouteMap as b, servedModelAcceptable as bt, DriverResult as c, stripFencedJson as ct, FeedbackPattern as d, ModelSubstitutionError as dt, backoffMs as et, JudgeConfig as f, ServedCrossFamilyError as ft, JudgeScore as g, assertServedModel as gt, JudgeRubric as h, assertCrossFamilyServed as ht, BenchmarkRunnerConfig as i, costReceiptFromLlmError as it, ChatRequest as j, ProviderRedactor as jt, ChatCallOpts as k, InMemoryRawProviderSinkOptions as kt, DriverState as l, AssertCrossFamilyServedOptions as lt, JudgeInput as m, ServedModelVerdict as mt, ArtifactResult as n, callLlmJson as nt, CollectedArtifacts as o, maximumChargeForLlmRequest as ot, JudgeFn as p, ServedModelCheck as pt, LlmClientOptions as q, BenchmarkReport as r, costReceiptFromLlm as rt, CompletionCriterion as s, probeLlm as st, ArtifactCheck as t, callLlm as tt, EvalResult as u, AssertServedModelOptions as ut, PersonaRigor as v, checkServedModel as vt, ScenarioResult as w, assertCrossFamily as wt, RubricDimension as x, AssertCrossFamilyOptions as xt, ProductClientConfig as y, normalizeModelId as yt, RouterTransportOpts as z };
1005
+ //# sourceMappingURL=types-XMVEdrE_.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"types-XMVEdrE_.d.ts","names":[],"sources":["../src/trace/raw-provider-sink.ts","../src/judge-families.ts","../src/integrity/served-model.ts","../src/llm-client.ts","../src/analyst/chat-client.ts","../src/types.ts"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;KA8BY;UAEK;;EAEf;;EAEA;EACA;;;;;;EAMA;EACA;;EAEA;;EAEA;;EAEA;EACA,WAAW;;EAEX;;EAEA;EACA;EACA,iBAAiB;EACjB;EACA,kBAAkB;EAClB;;EAEA;;EAEA;;UAGe;EACf;EACA;EACA,YAAY;EACZ;;UAGe;EACf,OAAO,OAAO,mBAAmB;;EAEjC,MAAM,SAAS,wBAAwB,QAAQ;;EAE/C,UAAU;;KAGA,oBAAoB,OAAO,qBAAqB;;;;;;iBAmB5C,wBAAwB,OAAO,mBAAmB;UA8CjD;EACf,WAAW;;cAGA,mCAAmC;UACtC;UACA;EAER,YAAY,OAAM;EAIZ,OAAO,OAAO,mBAAmB;EAIjC,KAAK,SAAQ,wBAA6B,QAAQ;EAUxD;;cAKW,+BAA+B;EACpC,UAAU;;;;;;;EASV,QAAQ,QAAQ;;UAOP;;EAEf;;EAEA;;EAEA;EACA,WAAW;;cAGA,qCAAqC;UACxC;UACA;UACA;UACA;UACA;UACA;UACA;EAER,YAAY,MAAM;UAOJ;UAON;EAKF,OAAO,OAAO,mBAAmB;EAYjC,KAAK,SAAQ,wBAA6B,QAAQ;;;;;;iBAkC1C,oBAAoB;;;;;;;;;;;;;KC5QxB;;;;;;;iBAkEI,YAAY,kBAAkB;UAc7B;;EAEf;;;;EAIA;;cAGW,yBAAyB;WAGlB,UAAU;WACV;EAHlB,YACE,iBACgB,UAAU,eACV;;;;;;;;;;;;;;;iBAoBJ,kBACd,kBACA,OAAM,2BACL;;;;KCrGS;;;;;;;;;;;UAYK;;EAEf;;EAEA;EACA,iBAAiB;;EAEjB,cAAc;EACd,SAAS;;EAET;;;;;;;;;;;;;;;;;iBAkBc,iBAAiB;;;;;;;;;;;iBA0BjB,iBACd,mBACA,oCACC;cA4CU,+BAA+B;WAGxB,QAAQ,cAAc;EAFxC,YACE,iBACgB,QAAQ,cAAc;;UAOzB;;;;;;EAMf;;;;;EAKA;;EAEA;;;;;;;iBAQc,sBACd,OAAO,kBACP,OAAM;;;;;;iBA6BQ,kBACd,mBACA,mCACA,OAAM,2BACL;;;;;iBAea,mBACd,OAAO;EAAgB;EAAmB;IAC1C,OAAM,2BACL;UAec,uCAAuC;;EAEtD;;EAEA;;cAGW,+BAA+B;WAGxB,UAAU;WACV,QAAQ,cAAc;EAHxC,YACE,iBACgB,UAAU,eACV,QAAQ,cAAc;;;;;;;;;;;iBAgB1B,wBACd,OAAO;EAAgB;EAAmB;IAC1C,OAAM,iCACL;;;UChOc;EACf;;;;;EAKA,kBAEI;IACM;IAAc;;IACd;IAAmB;MAAa;MAAa;;;;KAI7C;UAEK;EACf;EACA,UAAU;;EAEV;;EAEA;IAAe;IAAc,QAAQ;;EACrC;EACA;;EAEA,WAAW;;EAEX;;;;;;iBAOc,2BACd,SAAS,KAAK,iFACd,UAAS,mBACR;UAgCc;EACf;EACA;EACA;;EAEA;;EAEA;;EAEA;;UAGe;;EAEf;EACA,OAAO;;;;;EAKP;;;;;;;;;EASA;;;;;;;;;;;;EAYA;;EAEA;;;;;;;;;EASA;;;;;;;EAOA;;EAEA,KAAK;;KAGK,kBAAkB,KAAK;;iBAGnB,mBACd,QAAQ,eACR,qBAAqB,qBACpB;;iBA2Ba,wBACd,OAAO,OACP,qBAAqB,qBACpB;cAMU,qBAAqB;WAGd;WACA;WACA;EAJlB,YACE,iBACgB,gBACA,cACA;;;;;cASP,yBAAyB;WAGlB,QAAQ;EAF1B,YACE,iBACgB,QAAQ,eACxB;IAAY;;;UAMC;;EAEf;;EAEA;EACA;;EAEA;IAAe;IAAc;;;EAE7B;;EAEA;;;;;;;EAOA,SAAS;;;;;;;;EAQT;;EAEA;;EAEA,qBAAqB;;;;;;;EAOrB;;;;;;EAMA;;EAEA,WAAW;;EAEX,eAAe;;;;;;;;EAQf,UAAU;;;;;EAKV;;EAEA;IAAiB;IAAgB;;;EAEjC,WAAW;;;;;;;;;;EAUX,8BAA8B;;;;;;;;;;;;;iBAsEhB,oBAAoB;;iBAgCpB,UAAU;;;;;;iBA6GV,gBAAgB;;;;;;iBA6EV,QACpB,KAAK,gBACL,OAAM,mBACL,QAAQ;;;;;;;iBAoVW,YAAY,aAChC,KAAK,gBACL,OAAM,mBACL;EAAU,OAAO;EAAG,QAAQ;;KA4DnB;cAOC,+BAA+B;WAGxB,QAAQ;WACR;EAHlB,YACE,iBACgB,QAAQ,yBACR;;UAMH;;;;;;;EAOf;;;;;EAKA,kBAAkB,eAAe;;EAEjC,kBAAkB,eAAe;;EAEjC;;;;;EAKA;;;;;;;;;;;iBAYc,eAAe,MAAM,kBAAkB,MAAK;;;;;;;;;;;;;;;;;iBA4EtC,SACpB,eACA,OAAM;EAAqB;IAC1B;EACD;EACA;EACA;;EAEA;;EAEA;;;;;;;cAoCW;WACF;mBACQ;EAEjB,YAAY,OAAM;EAKlB,KAAK,KAAK,gBAAgB,MAAM,mBAAmB,QAAQ;EAK3D,SAAS,aACP,KAAK,gBACL,MAAM,mBACL;IAAU,OAAO;IAAG,QAAQ;;;;;;;;UC/pChB;;WAEN,WAAW;;WAEX;;WAEA;;EAGT,KAAK,KAAK,aAAa,OAAO,eAAe,QAAQ;;KAG3C;UAQK,oBAAoB,KAAK;;EAExC;;KAGU,eAAe;UAEV;;EAEf,SAAS;;EAET;;EAEA;;EAEA;;KAKU,uBACR,sBACA,yBACA,8BACA,0BACA,sBACA;UAEM;EACR;;EAEA;;UAGe,4BAA4B;EAC3C;EACA;EACA;;UAGe,+BAA+B;EAC9C;EACA;EACA;;UAGe,oCAAoC;EACnD;EACA;EACA;;;;;;UAOe,gCAAgC;EAC/C;EACA,OAAO,KAAK,aAAa,OAAO,iBAAiB,QAAQ;;;UAI1C,4BAA4B;EAC3C;EACA,OAAO,KAAK,aAAa,OAAO,iBAAiB,QAAQ;;;;;;UAO1C,0BAA0B;EACzC;EACA,UAAU,KAAK,aAAa,OAAO,iBAAiB,QAAQ;;;;;;iBAO9C,iBAAiB,MAAM,uBAAuB;;;UCjH7C;EACf;EACA;EACA;EACA;EACA;EACA,OAAO;EACP,gBAAgB;EAChB;;UAGe;EACf;EACA;EACA;EACA;;UAKe;EACf;EAQA;EACA;EACA;EACA;;UAKe;EACf;EACA;EACA,QAAQ;;UAGO;EACf;EACA;EACA,YAAY;;UAGG;EACf;EACA;EACA;EACA;EACA;;UAKe;EACf;EACA;EACA,OAAO;EACP,iBAAiB;EACjB,aAAa;EACb;EACA;EACA;EACA,WAAW;;EAEX,OAAO;;UAGQ;EACf;EACA;EACA;EACA;EACA;IAAmB;IAAc;;EACjC;EACA;;UAGe;EACf,OAAO;EACP;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;;UAGe;EACf;IAAc;IAAc;;EAC5B;IAAmB;IAAc,QAAQ;;EACzC;IAAc;IAAkB;;EAChC;;UAKe;EACf;EACA;EACA;EACA;EACA,SAAS;EACT,OAAO;EACP;IACE;IACA,WAAW;MAAiB;MAAa;MAAgB;;IACzD,aAAa;MAAiB;MAAa;;IAC3C;MAAW;MAAkB;MAAe;;IAC5C;MAAa;MAAkB;MAAe;;;;UAMjC;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;GACC;;UAGc;EACf;EACA,QAAQ;;EAER;;UAKe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;IACE;MACE;MACA;MACA;;;EAGJ,OAAO;EACP,gBAAgB;;UAKD;EACf;EACA,QAAQ,OAAO;EACf,YAAY,OAAO;;UAGJ;EACf;EACA;;;;;;;;;;KAWU;UAEK;EACf;EACA;EACA;EACA,oBAAoB;EACpB,mBAAmB;EACnB;EACA;;EAEA,QAAQ;;;;;;;EAOR;;;;;;;EAOA;;;;;;EAMA;;UAGe;EACf;EACA;EACA;IAAa;IAAiB;IAAkB;;EAChD;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;IAAa;IAAiB;IAAkB;;EAChD;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;;EAEA;;EAEA;;;;;;EAMA;EACA;EACA,SAAS;EACT,YAAY;EACZ;EACA;EACA;;UAKe;EACf,WAAW;EACX,QAAQ;EACR;EACA;EACA;EACA;EACA;EACA;;EAEA,aAAa;;UAGE;EACf,UAAU;EACV,OAAO;EACP,WAAW;;EAEX,aAAa;EACb;EACA,WAAW;EACX,SAAS;;KAGC,WAAW,MAAM,YAAY,OAAO,eAAe,QAAQ;UAItD;EACf;EACA;EACA;EACA;EACA,QAAQ;;UAGO;EACf;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA"}
@@ -1,7 +1,7 @@
1
- import { c as CostLedgerHandle } from "../cost-ledger-FuQvHxPm.js";
2
- import { Z as LlmRouteRequirements, q as LlmClientOptions } from "../types-BjMFz88h.js";
1
+ import { c as CostLedgerHandle } from "../cost-ledger-Bv_e8XHY.js";
2
+ import { Z as LlmRouteRequirements, q as LlmClientOptions } from "../types-XMVEdrE_.js";
3
3
  import { s as TraceStore } from "../store-CT9YIIve.js";
4
- import { _ as FeedbackTrajectoryStore } from "../feedback-trajectory-WK7x4mhy.js";
4
+ import { _ as FeedbackTrajectoryStore } from "../feedback-trajectory-GgoS0-MK.js";
5
5
  import { z } from "zod";
6
6
  import { ServerType } from "@hono/node-server";
7
7
  import { Hono } from "hono";
@@ -1,2 +1,2 @@
1
- import { A as TraceEventSchema, C as HealthResponseSchema, D as RubricDimensionSchema, E as ListRubricsResponseSchema, F as hashRubric, M as TracesIngestResponseSchema, N as VersionResponseSchema, O as RubricInfoSchema, P as WIRE_VERSION, S as FeedbackTrajectorySchema, T as JudgeResultSchema, _ as ErrorResponseSchema, a as runRpcBatch, b as FeedbackIngestResponseSchema, c as WireError, d as handleListRubrics, f as handleTracesIngest, g as listBuiltinRubrics, h as getBuiltinRubric, i as dispatchRpc, j as TracesIngestRequestSchema, k as RubricSchema, l as handleFeedbackIngest, m as BUILTIN_RUBRICS, n as startServer, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi, t as createApp, u as handleJudge, v as FailureModeSchema, w as JudgeRequestSchema, x as FeedbackLabelSchema, y as FeedbackAttemptSchema } from "../server-Df00sdwz.js";
1
+ import { A as TraceEventSchema, C as HealthResponseSchema, D as RubricDimensionSchema, E as ListRubricsResponseSchema, F as hashRubric, M as TracesIngestResponseSchema, N as VersionResponseSchema, O as RubricInfoSchema, P as WIRE_VERSION, S as FeedbackTrajectorySchema, T as JudgeResultSchema, _ as ErrorResponseSchema, a as runRpcBatch, b as FeedbackIngestResponseSchema, c as WireError, d as handleListRubrics, f as handleTracesIngest, g as listBuiltinRubrics, h as getBuiltinRubric, i as dispatchRpc, j as TracesIngestRequestSchema, k as RubricSchema, l as handleFeedbackIngest, m as BUILTIN_RUBRICS, n as startServer, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi, t as createApp, u as handleJudge, v as FailureModeSchema, w as JudgeRequestSchema, x as FeedbackLabelSchema, y as FeedbackAttemptSchema } from "../server-D6XJQHw7.js";
2
2
  export { BUILTIN_RUBRICS, ErrorResponseSchema, FailureModeSchema, FeedbackAttemptSchema, FeedbackIngestResponseSchema, FeedbackLabelSchema, FeedbackTrajectorySchema, HealthResponseSchema, JudgeRequestSchema, JudgeResultSchema, ListRubricsResponseSchema, RubricDimensionSchema, RubricInfoSchema, RubricSchema, TraceEventSchema, TracesIngestRequestSchema, TracesIngestResponseSchema, VersionResponseSchema, WIRE_VERSION, WireError, buildOpenApi, createApp, dispatchRpc, getBuiltinRubric, handleFeedbackIngest, handleJudge, handleListRubrics, handleTracesIngest, handleVersion, hashRubric, listBuiltinRubrics, runRpcBatch, runRpcOnce, startServer, startServerAsync };
@@ -8,6 +8,21 @@ Every hard-coded model id or endpoint default is verifiable against the live rou
8
8
 
9
9
  Enforced by: `preflightModels` (membership + optional probe) and `assertModelsServed` (gate that names every unreachable id with status + detail).
10
10
 
11
+ ## 1a. Reachable is not the same as identified
12
+
13
+ A 200 proves something answered, never that the requested model answered.
14
+ A routing gateway can accept one id and reply from another, and the swap is silent: same status, same shape, only the response's `model` field betrays it.
15
+ Every claim that names a model — a leaderboard row, a per-model cost, a cross-family judge panel, a non-self-judging exclusion — must be built from the id the provider echoed, not the id we sent.
16
+ Requested is intent; served is evidence.
17
+
18
+ Two failure shapes, and only one is a defect.
19
+ Snapshot resolution (`gpt-4o-mini` → `gpt-4o-mini-2024-07-18`) pins a floating alias to a reproducible build and is the behaviour we want.
20
+ Substitution (`gpt-4.1-mini` → `gemini-2.5-flash-lite`) mislabels every number the call produces.
21
+ A response that echoes no model id at all is unproven, which fails closed rather than defaulting to agreement.
22
+
23
+ Enforced by: `assertServedModel` / `assertServedModels` per call, `assertCrossFamilyServed` for panel diversity computed over the ids that answered, `LlmClientOptions.assertServedModel` to enforce it at the transport, and `assertModelsServed` (probe mode), which now fails a substituted id exactly as it fails a dead one.
24
+ `assertCrossFamily` reads requested ids and therefore proves configuration only — reach for the served-side check wherever the diversity claim is load-bearing.
25
+
11
26
  ## 2. Probe the platform before peeling client layers
12
27
 
13
28
  When a request fails, one direct call against the live endpoint bisects platform-versus-client before any code-level debugging begins. A 401 from the router on a `model_not_found` is the platform telling you the default is dead; a connection refused is the platform being unreachable. Establish which side is at fault with a probe first, then debug only the side that is actually broken.
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@tangle-network/agent-eval",
3
- "version": "0.144.5",
3
+ "version": "0.144.6",
4
4
  "description": "Evaluate and improve AI agents from runs, traces, judges, and feedback. Compare candidates, cluster failures, measure lift, and gate releases.",
5
5
  "homepage": "https://github.com/tangle-network/agent-eval#readme",
6
6
  "repository": {
@@ -168,8 +168,8 @@
168
168
  "dependencies": {
169
169
  "@asteasolutions/zod-to-openapi": "^9.1.0",
170
170
  "@hono/node-server": "^2.0.12",
171
- "@tangle-network/agent-core": "0.5.3",
172
- "@tangle-network/agent-interface": "0.46.0",
171
+ "@tangle-network/agent-core": "0.5.4",
172
+ "@tangle-network/agent-interface": "0.46.1",
173
173
  "@tangle-network/agent-trace-contract": "^1.0.2",
174
174
  "hono": "^4.12.32",
175
175
  "linear-sum-assignment": "1.0.9",