@tangle-network/agent-eval 0.222.1 → 1.0.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (147) hide show
  1. package/CHANGELOG.md +191 -0
  2. package/dist/adapters/http.d.ts +2 -2
  3. package/dist/{agent-profile-CSEQRC9l.d.ts → agent-profile-CoD1Dotf.d.ts} +2 -2
  4. package/dist/{agent-profile-CSEQRC9l.d.ts.map → agent-profile-CoD1Dotf.d.ts.map} +1 -1
  5. package/dist/analyst/index.d.ts +11 -11
  6. package/dist/analyst/index.d.ts.map +1 -1
  7. package/dist/analyst/index.js +1 -1
  8. package/dist/{analyze-runs-CGLRgZGd.d.ts → analyze-runs-C-pP1sK9.d.ts} +2 -2
  9. package/dist/{analyze-runs-CGLRgZGd.d.ts.map → analyze-runs-C-pP1sK9.d.ts.map} +1 -1
  10. package/dist/{analyze-runs-CTheASYV.js → analyze-runs-Ck68mbgj.js} +7 -9
  11. package/dist/analyze-runs-Ck68mbgj.js.map +1 -0
  12. package/dist/{benchmark-command-Cr7i7YSX.js → benchmark-command-DWkc5mzq.js} +32 -37
  13. package/dist/benchmark-command-DWkc5mzq.js.map +1 -0
  14. package/dist/benchmarks/index.d.ts +4 -4
  15. package/dist/benchmarks/index.js +3 -3
  16. package/dist/campaign/index.d.ts +6 -6
  17. package/dist/campaign/index.js +9 -9
  18. package/dist/{campaign-ddr1N3yD.js → campaign-DdOgQfp8.js} +14 -19
  19. package/dist/campaign-DdOgQfp8.js.map +1 -0
  20. package/dist/{campaign-evidence-DN8rWuBj.js → campaign-evidence-MjCjkb5a.js} +3 -3
  21. package/dist/{campaign-evidence-DN8rWuBj.js.map → campaign-evidence-MjCjkb5a.js.map} +1 -1
  22. package/dist/cli.js +4 -4
  23. package/dist/contract/index.d.ts +9 -9
  24. package/dist/contract/index.js +11 -10
  25. package/dist/contract/index.js.map +1 -1
  26. package/dist/{default-registry-1f4ukAIX.d.ts → default-registry-B1DstPYg.d.ts} +7 -7
  27. package/dist/{default-registry-1f4ukAIX.d.ts.map → default-registry-B1DstPYg.d.ts.map} +1 -1
  28. package/dist/{define-agent-eval-sUYh9BFA.d.ts → define-agent-eval-CAffiFbN.d.ts} +7 -7
  29. package/dist/{define-agent-eval-sUYh9BFA.d.ts.map → define-agent-eval-CAffiFbN.d.ts.map} +1 -1
  30. package/dist/{define-agent-eval-BLoP4Lf9.js → define-agent-eval-Cc15qGik.js} +5 -5
  31. package/dist/{define-agent-eval-BLoP4Lf9.js.map → define-agent-eval-Cc15qGik.js.map} +1 -1
  32. package/dist/diagnosis/index.d.ts +2 -2
  33. package/dist/diagnosis/index.js +1 -1
  34. package/dist/{engine-BITiErg-.d.ts → engine-CaHaCc3W.d.ts} +2 -2
  35. package/dist/{engine-BITiErg-.d.ts.map → engine-CaHaCc3W.d.ts.map} +1 -1
  36. package/dist/{eval-campaign-Cgixrvf7.js → eval-campaign-C_mNDshv.js} +2 -2
  37. package/dist/{eval-campaign-Cgixrvf7.js.map → eval-campaign-C_mNDshv.js.map} +1 -1
  38. package/dist/{evaluation-Djq7-Ybj.d.ts → evaluation-BPqUvZ66.d.ts} +3 -3
  39. package/dist/{evaluation-Djq7-Ybj.d.ts.map → evaluation-BPqUvZ66.d.ts.map} +1 -1
  40. package/dist/evaluation.d.ts +1 -1
  41. package/dist/{exact-types-CndDT_h7.d.ts → exact-types-D98BKpxo.d.ts} +2 -2
  42. package/dist/{exact-types-CndDT_h7.d.ts.map → exact-types-D98BKpxo.d.ts.map} +1 -1
  43. package/dist/experiment/index.d.ts +1 -1
  44. package/dist/experiment/index.js +7 -7
  45. package/dist/{feedback-trajectory-DsHR9krG.d.ts → feedback-trajectory-SBGncrCP.d.ts} +3 -3
  46. package/dist/{feedback-trajectory-DsHR9krG.d.ts.map → feedback-trajectory-SBGncrCP.d.ts.map} +1 -1
  47. package/dist/{index-Bdcpx2s8.d.ts → index-B7v2QHlD.d.ts} +6 -6
  48. package/dist/{index-Bdcpx2s8.d.ts.map → index-B7v2QHlD.d.ts.map} +1 -1
  49. package/dist/{index-IzWTz5EP.d.ts → index-CUjeTDPv.d.ts} +6 -6
  50. package/dist/{index-IzWTz5EP.d.ts.map → index-CUjeTDPv.d.ts.map} +1 -1
  51. package/dist/index-DAk2quYu.d.ts.map +1 -1
  52. package/dist/{index-BdKzYYkC.d.ts → index-DUn0GNiZ.d.ts} +2 -2
  53. package/dist/{index-BdKzYYkC.d.ts.map → index-DUn0GNiZ.d.ts.map} +1 -1
  54. package/dist/{index-Cu12qxJb.d.ts → index-aDPVYBIM.d.ts} +8 -8
  55. package/dist/{index-Cu12qxJb.d.ts.map → index-aDPVYBIM.d.ts.map} +1 -1
  56. package/dist/index.d.ts +17 -17
  57. package/dist/index.js +13 -12
  58. package/dist/index.js.map +1 -1
  59. package/dist/{judge-gate-C2BCOt-M.js → judge-gate-Di_6bifh.js} +4 -4
  60. package/dist/judge-gate-Di_6bifh.js.map +1 -0
  61. package/dist/{llm-judge-cgbhTKQp.js → llm-judge-D6JtIqiU.js} +9 -9
  62. package/dist/{llm-judge-cgbhTKQp.js.map → llm-judge-D6JtIqiU.js.map} +1 -1
  63. package/dist/{matrix-8X6i5kJ6.d.ts → matrix-CBUzIZs4.d.ts} +2 -2
  64. package/dist/{matrix-8X6i5kJ6.d.ts.map → matrix-CBUzIZs4.d.ts.map} +1 -1
  65. package/dist/mean-BPuUNznQ.js +16 -0
  66. package/dist/mean-BPuUNznQ.js.map +1 -0
  67. package/dist/meta-eval/index.js +1 -1
  68. package/dist/multishot/golden/index.d.ts +1 -1
  69. package/dist/multishot/index.d.ts +2 -2
  70. package/dist/openapi.json +1 -1
  71. package/dist/{paired-arms-D-54hO-k.js → paired-arms-D5H_58e8.js} +2 -2
  72. package/dist/{paired-arms-D-54hO-k.js.map → paired-arms-D5H_58e8.js.map} +1 -1
  73. package/dist/{paired-promotion-decision-FGvHA4p_.js → paired-promotion-decision-BiVumgvM.js} +2 -2
  74. package/dist/{paired-promotion-decision-FGvHA4p_.js.map → paired-promotion-decision-BiVumgvM.js.map} +1 -1
  75. package/dist/{mean-CdDYG-us.js → paired-tests-C2zUqgUW.js} +2 -7
  76. package/dist/paired-tests-C2zUqgUW.js.map +1 -0
  77. package/dist/{pareto-B9udAt2i.js → pareto-7_LEmYGG.js} +4 -3
  78. package/dist/{pareto-B9udAt2i.js.map → pareto-7_LEmYGG.js.map} +1 -1
  79. package/dist/{power-and-mde-ClI6RWw4.js → power-and-mde-CHvyDO_H.js} +2 -2
  80. package/dist/{power-and-mde-ClI6RWw4.js.map → power-and-mde-CHvyDO_H.js.map} +1 -1
  81. package/dist/{prime-protocol-BpQlTowq.d.ts → prime-protocol-COFkxas6.d.ts} +2 -2
  82. package/dist/{prime-protocol-BpQlTowq.d.ts.map → prime-protocol-COFkxas6.d.ts.map} +1 -1
  83. package/dist/{promotion-policy-CBYeD4zi.js → promotion-policy-D7B4_jEZ.js} +4 -4
  84. package/dist/{promotion-policy-CBYeD4zi.js.map → promotion-policy-D7B4_jEZ.js.map} +1 -1
  85. package/dist/{registry-D9A7Tybn.d.ts → registry-dhdDoAV8.d.ts} +3 -3
  86. package/dist/{registry-D9A7Tybn.d.ts.map → registry-dhdDoAV8.d.ts.map} +1 -1
  87. package/dist/reporting.js +3 -3
  88. package/dist/{reward-hacking-DZzHZlS1.js → reward-hacking-Bjk3MAjh.js} +7 -9
  89. package/dist/{reward-hacking-DZzHZlS1.js.map → reward-hacking-Bjk3MAjh.js.map} +1 -1
  90. package/dist/reward-hacking-C9hPVT0Q.d.ts.map +1 -1
  91. package/dist/rl.d.ts +3 -3
  92. package/dist/rl.js +6 -6
  93. package/dist/rollout/index.d.ts +2 -2
  94. package/dist/{run-record-CZy4-TnZ.js → run-record-BUm-Jggo.js} +2 -2
  95. package/dist/{run-record-CZy4-TnZ.js.map → run-record-BUm-Jggo.js.map} +1 -1
  96. package/dist/search/index.d.ts +2 -2
  97. package/dist/search/index.js +2 -2
  98. package/dist/{search-kernel-B77eE7Ac.d.ts → search-kernel-CBdwFZ05.d.ts} +2 -2
  99. package/dist/{search-kernel-B77eE7Ac.d.ts.map → search-kernel-CBdwFZ05.d.ts.map} +1 -1
  100. package/dist/{search-summary-Ct7SrmVc.js → search-summary-Bt3uuK53.js} +4 -4
  101. package/dist/{search-summary-Ct7SrmVc.js.map → search-summary-Bt3uuK53.js.map} +1 -1
  102. package/dist/{sequential-B1I0tRfL.js → sequential-De7BTAlV.js} +2 -2
  103. package/dist/{sequential-B1I0tRfL.js.map → sequential-De7BTAlV.js.map} +1 -1
  104. package/dist/{skillopt-optimization-method-qB1M554K.js → skillopt-optimization-method-BKFbJEyz.js} +9 -12
  105. package/dist/skillopt-optimization-method-BKFbJEyz.js.map +1 -0
  106. package/dist/{statistical-heldout-TP6ndL0B.js → statistical-heldout-C9hz_AF9.js} +3 -3
  107. package/dist/{statistical-heldout-TP6ndL0B.js.map → statistical-heldout-C9hz_AF9.js.map} +1 -1
  108. package/dist/{statistical-heldout-BYxHe3KO.d.ts → statistical-heldout-Cei036nE.d.ts} +3 -3
  109. package/dist/{statistical-heldout-BYxHe3KO.d.ts.map → statistical-heldout-Cei036nE.d.ts.map} +1 -1
  110. package/dist/{store-tool-spans-DMM6wMwB.d.ts → store-tool-spans-B3i3fl0A.d.ts} +4 -4
  111. package/dist/{store-tool-spans-DMM6wMwB.d.ts.map → store-tool-spans-B3i3fl0A.d.ts.map} +1 -1
  112. package/dist/storyboard/index.d.ts +17 -0
  113. package/dist/storyboard/index.d.ts.map +1 -1
  114. package/dist/storyboard/index.js +154 -107
  115. package/dist/storyboard/index.js.map +1 -1
  116. package/dist/{summary-report-CPfsrLac.js → summary-report-BLnTyQKH.js} +5 -4
  117. package/dist/{summary-report-CPfsrLac.js.map → summary-report-BLnTyQKH.js.map} +1 -1
  118. package/dist/{systemone-UHbKD6LH.d.ts → systemone-Cw2txK5m.d.ts} +5 -5
  119. package/dist/{systemone-UHbKD6LH.d.ts.map → systemone-Cw2txK5m.d.ts.map} +1 -1
  120. package/dist/{systemone-review-Cn9Uj0xG.d.ts → systemone-review-BLgxt-4Q.d.ts} +3 -3
  121. package/dist/{systemone-review-Cn9Uj0xG.d.ts.map → systemone-review-BLgxt-4Q.d.ts.map} +1 -1
  122. package/dist/systemone.d.ts +3 -3
  123. package/dist/{tool-groups-CFWmAGCV.d.ts → tool-groups-C3K-d3gd.d.ts} +3 -3
  124. package/dist/tool-groups-C3K-d3gd.d.ts.map +1 -0
  125. package/dist/{trace-questions-BFCQAzoT.d.ts → trace-questions-ej5AEJ5X.d.ts} +3 -3
  126. package/dist/{trace-questions-BFCQAzoT.d.ts.map → trace-questions-ej5AEJ5X.d.ts.map} +1 -1
  127. package/dist/trace-repair/index.d.ts +1 -1
  128. package/dist/trace-repair/index.js +2 -1
  129. package/dist/trace-repair/index.js.map +1 -1
  130. package/dist/traces.d.ts +3 -3
  131. package/dist/{tree-DrIChF4a.js → tree-DnBkWXun.js} +5 -4
  132. package/dist/{tree-DrIChF4a.js.map → tree-DnBkWXun.js.map} +1 -1
  133. package/dist/{types-DUR0dqET.d.ts → types-0oMg7zee.d.ts} +2 -2
  134. package/dist/{types-DUR0dqET.d.ts.map → types-0oMg7zee.d.ts.map} +1 -1
  135. package/dist/{types-e3HGaI7P.d.ts → types-C54Uk_a1.d.ts} +2 -2
  136. package/dist/{types-e3HGaI7P.d.ts.map → types-C54Uk_a1.d.ts.map} +1 -1
  137. package/dist/{types-BEvjDziM.d.ts → types-DzCO-S1m.d.ts} +8 -8
  138. package/dist/{types-BEvjDziM.d.ts.map → types-DzCO-S1m.d.ts.map} +1 -1
  139. package/dist/wire/index.d.ts +1 -1
  140. package/package.json +5 -5
  141. package/dist/analyze-runs-CTheASYV.js.map +0 -1
  142. package/dist/benchmark-command-Cr7i7YSX.js.map +0 -1
  143. package/dist/campaign-ddr1N3yD.js.map +0 -1
  144. package/dist/judge-gate-C2BCOt-M.js.map +0 -1
  145. package/dist/mean-CdDYG-us.js.map +0 -1
  146. package/dist/skillopt-optimization-method-qB1M554K.js.map +0 -1
  147. package/dist/tool-groups-CFWmAGCV.d.ts.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-TP6ndL0B.js";
3
+ import { n as aggregatePairedHoldout, o as pairHoldout } from "./statistical-heldout-C9hz_AF9.js";
4
4
  import { t as FileLedgerJournal } from "./journal-C1uLJi0w.js";
5
5
  import { n as contentHash } from "./verdict-cache-BVH-fl18.js";
6
- import { r as projectCampaignCellQuality } from "./run-record-CZy4-TnZ.js";
6
+ import { r as projectCampaignCellQuality } from "./run-record-BUm-Jggo.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-DN8rWuBj.js.map
1431
+ //# sourceMappingURL=campaign-evidence-MjCjkb5a.js.map