@tangle-network/agent-eval 1.0.20 → 1.0.22

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 (154) hide show
  1. package/CHANGELOG.md +32 -6
  2. package/dist/adapters/http.d.ts +2 -2
  3. package/dist/{agent-profile-Dngk3LZ0.js → agent-profile-4j6f2n-x.js} +4 -4
  4. package/dist/agent-profile-4j6f2n-x.js.map +1 -0
  5. package/dist/{agent-profile-BO8ruI6R.d.ts → agent-profile-DU5InUT9.d.ts} +4 -4
  6. package/dist/{agent-profile-BO8ruI6R.d.ts.map → agent-profile-DU5InUT9.d.ts.map} +1 -1
  7. package/dist/analyst/index.d.ts +6 -6
  8. package/dist/analyst/index.js +5 -5
  9. package/dist/{analyze-runs-DicleUI_.d.ts → analyze-runs-DYYe18Gw.d.ts} +3 -3
  10. package/dist/{analyze-runs-DicleUI_.d.ts.map → analyze-runs-DYYe18Gw.d.ts.map} +1 -1
  11. package/dist/{analyze-runs-e9vfXY--.js → analyze-runs-jLTRKuoK.js} +5 -5
  12. package/dist/{analyze-runs-e9vfXY--.js.map → analyze-runs-jLTRKuoK.js.map} +1 -1
  13. package/dist/app-grade/index.d.ts +2 -2
  14. package/dist/app-grade/index.js +1 -1
  15. package/dist/{benchmark-command-BGZD77Un.js → benchmark-command-DkUpd9Q6.js} +4 -4
  16. package/dist/{benchmark-command-BGZD77Un.js.map → benchmark-command-DkUpd9Q6.js.map} +1 -1
  17. package/dist/benchmarks/index.d.ts +4 -4
  18. package/dist/benchmarks/index.js +3 -3
  19. package/dist/builder-eval/index.js +1 -1
  20. package/dist/{calibration-Ze7SjhT9.js → calibration-Cf-wpqqO.js} +2 -2
  21. package/dist/{calibration-Ze7SjhT9.js.map → calibration-Cf-wpqqO.js.map} +1 -1
  22. package/dist/campaign/index.d.ts +6 -6
  23. package/dist/campaign/index.js +9 -9
  24. package/dist/{campaign-BfDMncQN.js → campaign-C2mmydmx.js} +12 -12
  25. package/dist/{campaign-BfDMncQN.js.map → campaign-C2mmydmx.js.map} +1 -1
  26. package/dist/{campaign-evidence-BuKlRzSA.js → campaign-evidence-NtH29ctt.js} +3 -3
  27. package/dist/{campaign-evidence-BuKlRzSA.js.map → campaign-evidence-NtH29ctt.js.map} +1 -1
  28. package/dist/cli.js +5 -5
  29. package/dist/{cluster-bootstrap-BYIW7QJ4.js → cluster-bootstrap-B0MTOBRU.js} +2 -2
  30. package/dist/{cluster-bootstrap-BYIW7QJ4.js.map → cluster-bootstrap-B0MTOBRU.js.map} +1 -1
  31. package/dist/contract/index.d.ts +8 -8
  32. package/dist/contract/index.js +10 -10
  33. package/dist/{decision-point-DJWievVw.js → decision-point-Di2LATNH.js} +3 -3
  34. package/dist/{decision-point-DJWievVw.js.map → decision-point-Di2LATNH.js.map} +1 -1
  35. package/dist/{default-registry-CdMtoY-s.js → default-registry-CdFbxPnr.js} +2 -2
  36. package/dist/{default-registry-CdMtoY-s.js.map → default-registry-CdFbxPnr.js.map} +1 -1
  37. package/dist/{default-registry-BpGFsSpZ.d.ts → default-registry-aPvwtzEM.d.ts} +2 -2
  38. package/dist/{default-registry-BpGFsSpZ.d.ts.map → default-registry-aPvwtzEM.d.ts.map} +1 -1
  39. package/dist/{define-agent-eval-DuQNZajG.js → define-agent-eval-BROmIwn1.js} +5 -5
  40. package/dist/{define-agent-eval-DuQNZajG.js.map → define-agent-eval-BROmIwn1.js.map} +1 -1
  41. package/dist/{define-agent-eval-DcK8mg63.d.ts → define-agent-eval-DGWBgCid.d.ts} +6 -6
  42. package/dist/{define-agent-eval-DcK8mg63.d.ts.map → define-agent-eval-DGWBgCid.d.ts.map} +1 -1
  43. package/dist/diagnosis/index.d.ts +1 -1
  44. package/dist/diagnosis/index.js +1 -1
  45. package/dist/{eval-campaign-LmpC_5fm.js → eval-campaign-Dq8DEKmy.js} +2 -2
  46. package/dist/{eval-campaign-LmpC_5fm.js.map → eval-campaign-Dq8DEKmy.js.map} +1 -1
  47. package/dist/{evaluation-BCejQgkV.d.ts → evaluation-CFKj_IlJ.d.ts} +2 -2
  48. package/dist/{evaluation-BCejQgkV.d.ts.map → evaluation-CFKj_IlJ.d.ts.map} +1 -1
  49. package/dist/evaluation.d.ts +1 -1
  50. package/dist/experiment/index.d.ts +2 -2
  51. package/dist/experiment/index.js +6 -6
  52. package/dist/fuzz.js +1 -1
  53. package/dist/{index-CzfdjM6o.d.ts → index-B62BaRzH.d.ts} +2 -2
  54. package/dist/{index-CzfdjM6o.d.ts.map → index-B62BaRzH.d.ts.map} +1 -1
  55. package/dist/{index-DYNHOTTB.d.ts → index-BlTbxHlY.d.ts} +2 -2
  56. package/dist/{index-DYNHOTTB.d.ts.map → index-BlTbxHlY.d.ts.map} +1 -1
  57. package/dist/{index-Bc8BZfzb.d.ts → index-Bp3w4fWc.d.ts} +7 -7
  58. package/dist/{index-Bc8BZfzb.d.ts.map → index-Bp3w4fWc.d.ts.map} +1 -1
  59. package/dist/{index-BdLmBInV.d.ts → index-DNXFzSpc.d.ts} +23 -21
  60. package/dist/{index-BdLmBInV.d.ts.map → index-DNXFzSpc.d.ts.map} +1 -1
  61. package/dist/{index-gSu9DUiw.d.ts → index-DX87wzX3.d.ts} +7 -7
  62. package/dist/{index-gSu9DUiw.d.ts.map → index-DX87wzX3.d.ts.map} +1 -1
  63. package/dist/{index-D_eAFB3T.d.ts → index-QLDWZTxI.d.ts} +4 -4
  64. package/dist/{index-D_eAFB3T.d.ts.map → index-QLDWZTxI.d.ts.map} +1 -1
  65. package/dist/{index-CeYLLnfi.d.ts → index-YS58LHlx.d.ts} +17 -17
  66. package/dist/{index-CeYLLnfi.d.ts.map → index-YS58LHlx.d.ts.map} +1 -1
  67. package/dist/index.d.ts +17 -17
  68. package/dist/index.js +15 -15
  69. package/dist/{judge-gate-BliY81A0.js → judge-gate-CrN0PaAL.js} +2 -2
  70. package/dist/{judge-gate-BliY81A0.js.map → judge-gate-CrN0PaAL.js.map} +1 -1
  71. package/dist/{llm-judge-Bnxuwjxl.js → llm-judge-C813vmpc.js} +9 -9
  72. package/dist/{llm-judge-Bnxuwjxl.js.map → llm-judge-C813vmpc.js.map} +1 -1
  73. package/dist/{matrix-D6zEwLvE.d.ts → matrix-C-GxIgeJ.d.ts} +2 -2
  74. package/dist/{matrix-D6zEwLvE.d.ts.map → matrix-C-GxIgeJ.d.ts.map} +1 -1
  75. package/dist/meta-eval/index.d.ts +2 -2
  76. package/dist/meta-eval/index.js +6 -6
  77. package/dist/multishot/golden/index.d.ts +1 -1
  78. package/dist/multishot/index.d.ts +2 -2
  79. package/dist/openapi.json +1 -1
  80. package/dist/{outcome-observations-BpicIVQi.js → outcome-observations-DPkH-LHQ.js} +2 -2
  81. package/dist/{outcome-observations-BpicIVQi.js.map → outcome-observations-DPkH-LHQ.js.map} +1 -1
  82. package/dist/{paired-arms-Dep5tbsw.js → paired-arms-DjMEtYI9.js} +2 -2
  83. package/dist/{paired-arms-Dep5tbsw.js.map → paired-arms-DjMEtYI9.js.map} +1 -1
  84. package/dist/{paired-promotion-decision-Nt4CkkOT.js → paired-promotion-decision-Bi9__RQS.js} +2 -2
  85. package/dist/{paired-promotion-decision-Nt4CkkOT.js.map → paired-promotion-decision-Bi9__RQS.js.map} +1 -1
  86. package/dist/{paired-promotion-decision-BNzPfXDw.d.ts → paired-promotion-decision-DJU3fZxH.d.ts} +2 -2
  87. package/dist/{paired-promotion-decision-BNzPfXDw.d.ts.map → paired-promotion-decision-DJU3fZxH.d.ts.map} +1 -1
  88. package/dist/{pareto-C445lnEH.js → pareto-fJZg8pXm.js} +3 -3
  89. package/dist/{pareto-C445lnEH.js.map → pareto-fJZg8pXm.js.map} +1 -1
  90. package/dist/pipelines/index.js +1 -1
  91. package/dist/{promotion-gate-DjAsmBSZ.d.ts → promotion-gate-Da0xV4sa.d.ts} +2 -2
  92. package/dist/{promotion-gate-DjAsmBSZ.d.ts.map → promotion-gate-Da0xV4sa.d.ts.map} +1 -1
  93. package/dist/{promotion-policy-CUWOv01m.js → promotion-policy-DCgftbnd.js} +4 -4
  94. package/dist/{promotion-policy-CUWOv01m.js.map → promotion-policy-DCgftbnd.js.map} +1 -1
  95. package/dist/{release-confidence-DDrkmWxT.js → release-confidence-DGtuPDxJ.js} +2 -2
  96. package/dist/{release-confidence-DDrkmWxT.js.map → release-confidence-DGtuPDxJ.js.map} +1 -1
  97. package/dist/reporting.d.ts +3 -3
  98. package/dist/reporting.js +4 -4
  99. package/dist/{researcher-D35yo0B8.d.ts → researcher-CSlc84Bp.d.ts} +2 -2
  100. package/dist/{researcher-D35yo0B8.d.ts.map → researcher-CSlc84Bp.d.ts.map} +1 -1
  101. package/dist/{reward-hacking-DRb8K6N1.js → reward-hacking-BYiDFbIl.js} +2 -2
  102. package/dist/{reward-hacking-DRb8K6N1.js.map → reward-hacking-BYiDFbIl.js.map} +1 -1
  103. package/dist/rl.d.ts +4 -4
  104. package/dist/rl.js +7 -7
  105. package/dist/rollout/index.d.ts +2 -2
  106. package/dist/{rubric-predictive-validity-BMS7SUyh.js → rubric-predictive-validity-DL7gJ51F.js} +2 -2
  107. package/dist/{rubric-predictive-validity-BMS7SUyh.js.map → rubric-predictive-validity-DL7gJ51F.js.map} +1 -1
  108. package/dist/{run-record-CIflWk6i.js → run-record-CFbnkE1K.js} +2 -2
  109. package/dist/{run-record-CIflWk6i.js.map → run-record-CFbnkE1K.js.map} +1 -1
  110. package/dist/{runtime-rollouts-CfRyjk60.js → runtime-rollouts-BNrm5nC_.js} +2 -2
  111. package/dist/{runtime-rollouts-CfRyjk60.js.map → runtime-rollouts-BNrm5nC_.js.map} +1 -1
  112. package/dist/search/index.d.ts +2 -2
  113. package/dist/search/index.js +2 -2
  114. package/dist/{search-kernel-Ds2zILpY.d.ts → search-kernel-CQGLbmgn.d.ts} +3 -3
  115. package/dist/{search-kernel-Ds2zILpY.d.ts.map → search-kernel-CQGLbmgn.d.ts.map} +1 -1
  116. package/dist/{search-summary-BSftVrGY.js → search-summary-sPWiYtSl.js} +4 -4
  117. package/dist/{search-summary-BSftVrGY.js.map → search-summary-sPWiYtSl.js.map} +1 -1
  118. package/dist/{sequential-CqG6aLWS.js → sequential-9TVb5Nu5.js} +3 -3
  119. package/dist/{sequential-CqG6aLWS.js.map → sequential-9TVb5Nu5.js.map} +1 -1
  120. package/dist/{skillopt-optimization-method-DYwEndON.js → skillopt-optimization-method-QO1nU8Kz.js} +6 -6
  121. package/dist/{skillopt-optimization-method-DYwEndON.js.map → skillopt-optimization-method-QO1nU8Kz.js.map} +1 -1
  122. package/dist/{statistical-heldout--0b1lgx4.d.ts → statistical-heldout-CvOwNqt9.d.ts} +4 -4
  123. package/dist/{statistical-heldout--0b1lgx4.d.ts.map → statistical-heldout-CvOwNqt9.d.ts.map} +1 -1
  124. package/dist/{statistical-heldout-D6i58CJ5.js → statistical-heldout-DLO4Ebos.js} +3 -3
  125. package/dist/{statistical-heldout-D6i58CJ5.js.map → statistical-heldout-DLO4Ebos.js.map} +1 -1
  126. package/dist/{statistics-DCZIC-qT.js → statistics-CvvC-MPv.js} +28 -21
  127. package/dist/statistics-CvvC-MPv.js.map +1 -0
  128. package/dist/statistics.d.ts +2 -2
  129. package/dist/statistics.js +2 -2
  130. package/dist/{summary-report-D_uussI3.js → summary-report-BEhqoxrA.js} +3 -3
  131. package/dist/{summary-report-D_uussI3.js.map → summary-report-BEhqoxrA.js.map} +1 -1
  132. package/dist/{summary-report-Cv1AC-iv.d.ts → summary-report-DyZquTaH.d.ts} +2 -2
  133. package/dist/{summary-report-Cv1AC-iv.d.ts.map → summary-report-DyZquTaH.d.ts.map} +1 -1
  134. package/dist/supervisor-run/index.d.ts +1 -1
  135. package/dist/supervisor-run/index.js +1 -1
  136. package/dist/{systemone-C3xyEKvG.d.ts → systemone-DVk-Crtu.d.ts} +4 -4
  137. package/dist/{systemone-C3xyEKvG.d.ts.map → systemone-DVk-Crtu.d.ts.map} +1 -1
  138. package/dist/systemone.d.ts +2 -2
  139. package/dist/systemone.js +1 -1
  140. package/dist/{tool-waste-DEhKfiIE.js → tool-waste-BqSmmPUz.js} +2 -2
  141. package/dist/{tool-waste-DEhKfiIE.js.map → tool-waste-BqSmmPUz.js.map} +1 -1
  142. package/dist/trace-repair/index.js +1 -1
  143. package/dist/{tree-B2y1UR8o.js → tree-C-N2kP-8.js} +4 -4
  144. package/dist/{tree-B2y1UR8o.js.map → tree-C-N2kP-8.js.map} +1 -1
  145. package/dist/{types-BwF--ZRz.d.ts → types-BsdcSWnf.d.ts} +2 -2
  146. package/dist/{types-BwF--ZRz.d.ts.map → types-BsdcSWnf.d.ts.map} +1 -1
  147. package/dist/{types-BAJ2Hb-d.d.ts → types-DUVhW5V1.d.ts} +2 -2
  148. package/dist/{types-BAJ2Hb-d.d.ts.map → types-DUVhW5V1.d.ts.map} +1 -1
  149. package/dist/{types-CeMZYsNF.d.ts → types-ip2D3-Ib.d.ts} +2 -2
  150. package/dist/{types-CeMZYsNF.d.ts.map → types-ip2D3-Ib.d.ts.map} +1 -1
  151. package/docs/design/statistics-decisions.md +7 -0
  152. package/package.json +5 -5
  153. package/dist/agent-profile-Dngk3LZ0.js.map +0 -1
  154. package/dist/statistics-DCZIC-qT.js.map +0 -1
@@ -1,9 +1,9 @@
1
1
  import { t as AgentEvalError } from "./errors-DeEEV9Vx.js";
2
2
  import { a as hashCanonical, i as compareCodeUnits, r as canonicalString } from "./canonical-CicJlINA.js";
3
- import { n as aggregatePairedHoldout, o as pairHoldout } from "./statistical-heldout-D6i58CJ5.js";
3
+ import { n as aggregatePairedHoldout, o as pairHoldout } from "./statistical-heldout-DLO4Ebos.js";
4
4
  import { t as FileLedgerJournal } from "./journal-CTX5uK-n.js";
5
5
  import { n as contentHash } from "./verdict-cache-BVH-fl18.js";
6
- import { r as projectCampaignCellQuality } from "./run-record-CIflWk6i.js";
6
+ import { r as projectCampaignCellQuality } from "./run-record-CFbnkE1K.js";
7
7
  import { l as surfaceHash, o as renderSurfaceDiff, s as surfaceContentHash, t as autoCloseTruncatedJson, u as surfaceHashMatches } from "./json-recovery-DK4MDm5r.js";
8
8
  import { t as defineEvaluationClaim } from "./claim-Dod_mjk3.js";
9
9
  import { createHash } from "node:crypto";
@@ -1428,4 +1428,4 @@ function createCampaignEvidenceReceipt(input) {
1428
1428
  }
1429
1429
  export { campaignSplitDigest as A, assertRealBackend as C, assertCompleteCampaign as D, assertCampaignSplitIdentity as E, formatCoverageFailures as M, campaignCoverage as O, BackendIntegrityError as S, assertCampaignDesign as T, parseReflectionResponse as _, ATTESTATION_ALGORITHM as a, FinalEvidenceError as b, campaignMeasurementDigest as c, loopProvenanceArgsFromResult as d, campaignBreakdown as f, buildReflectionPrompt as g, pairedCampaignComposites as h, verifyEvidenceReceipt as i, campaignSplitDigestFromIdentities as j, campaignScenarioIdentity as k, canonicalDigest as l, campaignMeanCompositeOrNull as m, createEvidenceReceipt as n, attest as o, campaignMeanComposite as p, isIndependentEvidence as r, verifyAttestation as s, createCampaignEvidenceReceipt as t, emitLoopProvenance as u, powerPreflight as v, summarizeBackendIntegrity as w, openFinalEvidenceLedger as x, FinalEvidenceConflictError as y };
1430
1430
 
1431
- //# sourceMappingURL=campaign-evidence-BuKlRzSA.js.map
1431
+ //# sourceMappingURL=campaign-evidence-NtH29ctt.js.map