@tangle-network/agent-eval 0.217.1 → 0.218.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (172) hide show
  1. package/CHANGELOG.md +36 -1
  2. package/dist/adapters/http.d.ts +2 -2
  3. package/dist/analyst/index.d.ts +1 -1
  4. package/dist/analyst/index.d.ts.map +1 -1
  5. package/dist/analyst/index.js +9 -9
  6. package/dist/analyst/index.js.map +1 -1
  7. package/dist/{analyze-runs-DyeNcenW.d.ts → analyze-runs-DjlZOVXx.d.ts} +2 -2
  8. package/dist/{analyze-runs-DyeNcenW.d.ts.map → analyze-runs-DjlZOVXx.d.ts.map} +1 -1
  9. package/dist/{benchmark-command-xQK8nOI6.js → benchmark-command-BupdHRyf.js} +661 -562
  10. package/dist/benchmark-command-BupdHRyf.js.map +1 -0
  11. package/dist/benchmarks/index.d.ts +4 -4
  12. package/dist/benchmarks/index.js +3 -3
  13. package/dist/campaign/index.d.ts +4 -4
  14. package/dist/campaign/index.js +10 -10
  15. package/dist/{campaign-CJGkxAQ6.js → campaign-CGvUTnmc.js} +312 -281
  16. package/dist/{campaign-CJGkxAQ6.js.map → campaign-CGvUTnmc.js.map} +1 -1
  17. package/dist/{campaign-evidence-CQYw8_0H.js → campaign-evidence-DunzcJGv.js} +2 -2
  18. package/dist/{campaign-evidence-CQYw8_0H.js.map → campaign-evidence-DunzcJGv.js.map} +1 -1
  19. package/dist/{chat-client-ldqxYYTN.js → chat-client-B2pXlPdx.js} +2 -2
  20. package/dist/{chat-client-ldqxYYTN.js.map → chat-client-B2pXlPdx.js.map} +1 -1
  21. package/dist/{chat-json-call-DavzexSp.js → chat-json-call-Dm9c-b41.js} +2 -2
  22. package/dist/{chat-json-call-DavzexSp.js.map → chat-json-call-Dm9c-b41.js.map} +1 -1
  23. package/dist/cli.js +10 -10
  24. package/dist/{completion-verifier-B_jw6C9p.js → completion-verifier-DbLVf3i8.js} +4 -4
  25. package/dist/{completion-verifier-B_jw6C9p.js.map → completion-verifier-DbLVf3i8.js.map} +1 -1
  26. package/dist/contract/index.d.ts +7 -7
  27. package/dist/contract/index.js +90 -10
  28. package/dist/contract/index.js.map +1 -1
  29. package/dist/{default-registry-ClYjq3QE.js → default-registry-BJ-Pq6yL.js} +309 -277
  30. package/dist/{default-registry-ClYjq3QE.js.map → default-registry-BJ-Pq6yL.js.map} +1 -1
  31. package/dist/{define-agent-eval-BDxNlP3D.js → define-agent-eval-BGdFKlrc.js} +165 -105
  32. package/dist/define-agent-eval-BGdFKlrc.js.map +1 -0
  33. package/dist/{define-agent-eval-BIkH_gAR.d.ts → define-agent-eval-BlH10zEw.d.ts} +6 -6
  34. package/dist/{define-agent-eval-BIkH_gAR.d.ts.map → define-agent-eval-BlH10zEw.d.ts.map} +1 -1
  35. package/dist/diagnosis/index.d.ts +1 -1
  36. package/dist/{dspy-rlm-engine-DhdCr1Dc.js → dspy-rlm-engine-CSMwudLO.js} +2 -2
  37. package/dist/{dspy-rlm-engine-DhdCr1Dc.js.map → dspy-rlm-engine-CSMwudLO.js.map} +1 -1
  38. package/dist/eval-campaign-BUMIxKTE.js +393 -0
  39. package/dist/eval-campaign-BUMIxKTE.js.map +1 -0
  40. package/dist/{evaluation-jTjC_QQD.d.ts → evaluation-BOZ38vOH.d.ts} +2 -2
  41. package/dist/{evaluation-jTjC_QQD.d.ts.map → evaluation-BOZ38vOH.d.ts.map} +1 -1
  42. package/dist/evaluation.d.ts +1 -1
  43. package/dist/experiment/index.d.ts +1 -1
  44. package/dist/experiment/index.d.ts.map +1 -1
  45. package/dist/experiment/index.js +110 -71
  46. package/dist/experiment/index.js.map +1 -1
  47. package/dist/{exporters-Cl6IeNAu.js → exporters-DWvM16LE.js} +2 -2
  48. package/dist/{exporters-Cl6IeNAu.js.map → exporters-DWvM16LE.js.map} +1 -1
  49. package/dist/{external-optimizer-process-KVFGvYX1.js → external-optimizer-process-C3Wm7Vhn.js} +2 -2
  50. package/dist/{external-optimizer-process-KVFGvYX1.js.map → external-optimizer-process-C3Wm7Vhn.js.map} +1 -1
  51. package/dist/{external-optimizer-subprocess-XIHBOSKt.js → external-optimizer-subprocess-B5gdLRZv.js} +3 -3
  52. package/dist/{external-optimizer-subprocess-XIHBOSKt.js.map → external-optimizer-subprocess-B5gdLRZv.js.map} +1 -1
  53. package/dist/{findings-store-BMVQtEsm.js → findings-store-BtTzPL0d.js} +5 -5
  54. package/dist/{findings-store-BMVQtEsm.js.map → findings-store-BtTzPL0d.js.map} +1 -1
  55. package/dist/{hf-dataset-D9f7-oX7.js → hf-dataset-CKxIJQba.js} +3 -3
  56. package/dist/{hf-dataset-D9f7-oX7.js.map → hf-dataset-CKxIJQba.js.map} +1 -1
  57. package/dist/hosted/index.js +1 -1
  58. package/dist/{index-ZuWx52WG.d.ts → index-B16FNdW5.d.ts} +34 -17
  59. package/dist/index-B16FNdW5.d.ts.map +1 -0
  60. package/dist/{index-9jJ2fKSX.d.ts → index-CalAsXMR.d.ts} +7 -5
  61. package/dist/index-CalAsXMR.d.ts.map +1 -0
  62. package/dist/{index-BiT6RtNH.d.ts → index-DbMOXYnP.d.ts} +6 -6
  63. package/dist/{index-BiT6RtNH.d.ts.map → index-DbMOXYnP.d.ts.map} +1 -1
  64. package/dist/index.d.ts +9 -9
  65. package/dist/index.d.ts.map +1 -1
  66. package/dist/index.js +611 -688
  67. package/dist/index.js.map +1 -1
  68. package/dist/{integrity-DOd9Q86O.js → integrity-C8oUeB0g.js} +322 -197
  69. package/dist/integrity-C8oUeB0g.js.map +1 -0
  70. package/dist/{journal-CR4QikYp.js → journal-C7om8Gum.js} +48 -38
  71. package/dist/journal-C7om8Gum.js.map +1 -0
  72. package/dist/ledger-core/index.d.ts +1 -1
  73. package/dist/ledger-core/index.js +2 -2
  74. package/dist/ledger-core/lock-holder.d.ts +2 -0
  75. package/dist/ledger-core/lock-holder.js +181 -0
  76. package/dist/ledger-core/lock-holder.js.map +1 -0
  77. package/dist/{ledger-core-DrJhbms9.js → ledger-core-CAxUdDk0.js} +2 -2
  78. package/dist/{ledger-core-DrJhbms9.js.map → ledger-core-CAxUdDk0.js.map} +1 -1
  79. package/dist/{llm-client-BshCREn6.js → llm-client-_Y6ic-tf.js} +238 -248
  80. package/dist/llm-client-_Y6ic-tf.js.map +1 -0
  81. package/dist/{llm-judge-C8jMTNNG.js → llm-judge-Djxh-Uhs.js} +646 -458
  82. package/dist/llm-judge-Djxh-Uhs.js.map +1 -0
  83. package/dist/lock-holder-DaJbfpci.d.ts +86 -0
  84. package/dist/lock-holder-DaJbfpci.d.ts.map +1 -0
  85. package/dist/{matrix-DHe-dwVJ.d.ts → matrix-CCSBna4q.d.ts} +2 -2
  86. package/dist/{matrix-DHe-dwVJ.d.ts.map → matrix-CCSBna4q.d.ts.map} +1 -1
  87. package/dist/{mint-zszyYlpA.js → mint-CbdwjTPW.js} +2 -2
  88. package/dist/{mint-zszyYlpA.js.map → mint-CbdwjTPW.js.map} +1 -1
  89. package/dist/multishot/golden/index.d.ts +1 -1
  90. package/dist/multishot/index.d.ts +2 -2
  91. package/dist/openapi.json +1 -1
  92. package/dist/{pareto-BzsxidFD.js → pareto-CTv8vozZ.js} +4 -4
  93. package/dist/{pareto-BzsxidFD.js.map → pareto-CTv8vozZ.js.map} +1 -1
  94. package/dist/{promotion-gate-DaY95yU6.d.ts → promotion-gate-CxEHRsZW.d.ts} +2 -2
  95. package/dist/{promotion-gate-DaY95yU6.d.ts.map → promotion-gate-CxEHRsZW.d.ts.map} +1 -1
  96. package/dist/{proposal-findings-BbWidvUk.js → proposal-findings-n-5WBbEx.js} +2 -2
  97. package/dist/{proposal-findings-BbWidvUk.js.map → proposal-findings-n-5WBbEx.js.map} +1 -1
  98. package/dist/{render-C-w2EZ6p.js → render-Cs34y0iX.js} +404 -208
  99. package/dist/render-Cs34y0iX.js.map +1 -0
  100. package/dist/reporting.d.ts +2 -2
  101. package/dist/{researcher-0ABVIdGa.d.ts → researcher-GU1tV4dn.d.ts} +2 -2
  102. package/dist/{researcher-0ABVIdGa.d.ts.map → researcher-GU1tV4dn.d.ts.map} +1 -1
  103. package/dist/rl.d.ts +2 -2
  104. package/dist/rl.js +4 -4
  105. package/dist/rollout/index.js +5 -5
  106. package/dist/{rollout-D6C50mfA.js → rollout-Dd1ucT_M.js} +5 -5
  107. package/dist/{rollout-D6C50mfA.js.map → rollout-Dd1ucT_M.js.map} +1 -1
  108. package/dist/schema-C-ziyJD7.d.ts.map +1 -1
  109. package/dist/{schema-DxcKTzbi.js → schema-CvhfLjtZ.js} +77 -47
  110. package/dist/schema-CvhfLjtZ.js.map +1 -0
  111. package/dist/search/index.d.ts +2 -2
  112. package/dist/search/index.d.ts.map +1 -1
  113. package/dist/search/index.js +2 -2
  114. package/dist/{search-kernel-DxqjBxA-.d.ts → search-kernel-p9JB3SmL.d.ts} +3 -3
  115. package/dist/{search-kernel-DxqjBxA-.d.ts.map → search-kernel-p9JB3SmL.d.ts.map} +1 -1
  116. package/dist/{search-ledger-TUvB5gWH.js → search-ledger-ClpN4I3e.js} +4 -4
  117. package/dist/{search-ledger-TUvB5gWH.js.map → search-ledger-ClpN4I3e.js.map} +1 -1
  118. package/dist/{search-ledger-file-UKCO2r8a.js → search-ledger-file-Bd5Tioq6.js} +3 -3
  119. package/dist/{search-ledger-file-UKCO2r8a.js.map → search-ledger-file-Bd5Tioq6.js.map} +1 -1
  120. package/dist/{search-shipper-Cw2z3-td.js → search-shipper-Bg2HW3Uv.js} +3 -3
  121. package/dist/{search-shipper-Cw2z3-td.js.map → search-shipper-Bg2HW3Uv.js.map} +1 -1
  122. package/dist/{search-summary-DQIvJsSB.js → search-summary-CrBuvVnf.js} +3 -3
  123. package/dist/{search-summary-DQIvJsSB.js.map → search-summary-CrBuvVnf.js.map} +1 -1
  124. package/dist/{server-DoVbqqtJ.js → server-C6M0qR8h.js} +2 -2
  125. package/dist/{server-DoVbqqtJ.js.map → server-C6M0qR8h.js.map} +1 -1
  126. package/dist/{single-run-lock-U58yf8N6.js → single-run-lock-Dekd8_Wv.js} +7 -7
  127. package/dist/single-run-lock-Dekd8_Wv.js.map +1 -0
  128. package/dist/{skillopt-optimization-method-kyL94-ri.js → skillopt-optimization-method-5dlex7MJ.js} +1339 -1059
  129. package/dist/skillopt-optimization-method-5dlex7MJ.js.map +1 -0
  130. package/dist/{statistical-heldout-C9AbgREN.d.ts → statistical-heldout-BHfLZHrm.d.ts} +2 -2
  131. package/dist/{statistical-heldout-C9AbgREN.d.ts.map → statistical-heldout-BHfLZHrm.d.ts.map} +1 -1
  132. package/dist/{summary-report-DM7tlml1.d.ts → summary-report-CkbRcSuu.d.ts} +16 -1
  133. package/dist/summary-report-CkbRcSuu.d.ts.map +1 -0
  134. package/dist/supervisor-run/index.d.ts.map +1 -1
  135. package/dist/supervisor-run/index.js +3 -3
  136. package/dist/{systemone-Bp_Ac0p1.d.ts → systemone-BMzDJb1i.d.ts} +3 -3
  137. package/dist/{systemone-Bp_Ac0p1.d.ts.map → systemone-BMzDJb1i.d.ts.map} +1 -1
  138. package/dist/systemone.d.ts +2 -2
  139. package/dist/{terminal-record-78g6E4vh.js → terminal-record-D4Rv-j1W.js} +241 -193
  140. package/dist/{terminal-record-78g6E4vh.js.map → terminal-record-D4Rv-j1W.js.map} +1 -1
  141. package/dist/trajectory-replay/index.d.ts +4 -1
  142. package/dist/trajectory-replay/index.d.ts.map +1 -1
  143. package/dist/trajectory-replay/index.js +834 -824
  144. package/dist/trajectory-replay/index.js.map +1 -1
  145. package/dist/{tree-BGRglkKd.js → tree-B5e1uAeL.js} +122 -14
  146. package/dist/tree-B5e1uAeL.js.map +1 -0
  147. package/dist/{types-DtMbFw6U.d.ts → types-BKCzzmit.d.ts} +2 -2
  148. package/dist/{types-DtMbFw6U.d.ts.map → types-BKCzzmit.d.ts.map} +1 -1
  149. package/dist/{types-D8GhJoHk.d.ts → types-DtkO1G9L.d.ts} +2 -2
  150. package/dist/{types-D8GhJoHk.d.ts.map → types-DtkO1G9L.d.ts.map} +1 -1
  151. package/dist/wire/index.js +1 -1
  152. package/docs/design/statistics-decisions.md +1 -1
  153. package/docs/eval-surface-map.md +1 -0
  154. package/docs/profile-json-identity.md +1 -2
  155. package/docs/public-api.md +27 -16
  156. package/package.json +7 -2
  157. package/dist/benchmark-command-xQK8nOI6.js.map +0 -1
  158. package/dist/define-agent-eval-BDxNlP3D.js.map +0 -1
  159. package/dist/eval-campaign-BLEfBGbA.js +0 -343
  160. package/dist/eval-campaign-BLEfBGbA.js.map +0 -1
  161. package/dist/index-9jJ2fKSX.d.ts.map +0 -1
  162. package/dist/index-ZuWx52WG.d.ts.map +0 -1
  163. package/dist/integrity-DOd9Q86O.js.map +0 -1
  164. package/dist/journal-CR4QikYp.js.map +0 -1
  165. package/dist/llm-client-BshCREn6.js.map +0 -1
  166. package/dist/llm-judge-C8jMTNNG.js.map +0 -1
  167. package/dist/render-C-w2EZ6p.js.map +0 -1
  168. package/dist/schema-DxcKTzbi.js.map +0 -1
  169. package/dist/single-run-lock-U58yf8N6.js.map +0 -1
  170. package/dist/skillopt-optimization-method-kyL94-ri.js.map +0 -1
  171. package/dist/summary-report-DM7tlml1.d.ts.map +0 -1
  172. package/dist/tree-BGRglkKd.js.map +0 -1
@@ -3,7 +3,7 @@ import { a as hashCanonical, i as compareCodeUnits, r as canonicalString } from
3
3
  import { n as aggregatePairedHoldout, o as pairHoldout } from "./statistical-heldout-Bksv92Th.js";
4
4
  import { n as contentHash } from "./verdict-cache-CL_7tePD.js";
5
5
  import { r as projectCampaignCellQuality } from "./run-record-CysKLB93.js";
6
- import { t as FileLedgerJournal } from "./journal-CR4QikYp.js";
6
+ import { t as FileLedgerJournal } from "./journal-C7om8Gum.js";
7
7
  import { t as defineEvaluationClaim } from "./claim-BPVuyHuy.js";
8
8
  import { createHash } from "node:crypto";
9
9
  import { z } from "zod";
@@ -1730,4 +1730,4 @@ function createCampaignEvidenceReceipt(input) {
1730
1730
  //#endregion
1731
1731
  export { surfaceHashMatches as A, campaignSplitDigestFromIdentities as B, assertCodeSurfaceIdentity as C, surfaceContentHash as D, renderSurfaceDiff as E, assertCampaignSplitIdentity as F, assertCompleteCampaign as I, campaignCoverage as L, assertRealBackend as M, summarizeBackendIntegrity as N, surfaceDispatchRef as O, assertCampaignDesign as P, campaignScenarioIdentity as R, openFinalEvidenceLedger as S, componentSurfaceIdentityMaterial as T, formatCoverageFailures as V, parseReflectionResponse as _, ATTESTATION_ALGORITHM as a, FinalEvidenceConflictError as b, campaignMeasurementDigest as c, loopProvenanceArgsFromResult as d, campaignBreakdown as f, buildReflectionPrompt as g, pairedCampaignComposites as h, verifyEvidenceReceipt as i, BackendIntegrityError as j, surfaceHash 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, recoverTruncatedJson as v, codeSurfaceIdentityMaterial as w, FinalEvidenceError as x, powerPreflight as y, campaignSplitDigest as z };
1732
1732
 
1733
- //# sourceMappingURL=campaign-evidence-CQYw8_0H.js.map
1733
+ //# sourceMappingURL=campaign-evidence-DunzcJGv.js.map