@tangle-network/agent-eval 0.144.6 → 0.144.8

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 (240) hide show
  1. package/CHANGELOG.md +30 -0
  2. package/README.md +2 -0
  3. package/dist/{benchmark-J9Qe6j2_.d.ts → agent-profile-CgDTo40f.d.ts} +91 -4
  4. package/dist/agent-profile-CgDTo40f.d.ts.map +1 -0
  5. package/dist/analyst/index.d.ts +470 -88
  6. package/dist/analyst/index.d.ts.map +1 -1
  7. package/dist/analyst/index.js +24 -5
  8. package/dist/analyst/index.js.map +1 -1
  9. package/dist/{analyze-runs-DWIvOAGk.js → analyze-runs-BNNK7irB.js} +3 -3
  10. package/dist/{analyze-runs-DWIvOAGk.js.map → analyze-runs-BNNK7irB.js.map} +1 -1
  11. package/dist/{baseline-D_fT6277.d.ts → baseline-CavEbRyH.d.ts} +2 -23
  12. package/dist/baseline-CavEbRyH.d.ts.map +1 -0
  13. package/dist/{benchmark-command-CQd78YHt.js → benchmark-command-BKENp2s5.js} +662 -605
  14. package/dist/benchmark-command-BKENp2s5.js.map +1 -0
  15. package/dist/benchmarks/index.d.ts +1 -1
  16. package/dist/benchmarks/index.js +1 -1
  17. package/dist/{benchmarks-BEOkuvIg.js → benchmarks-BlPmjd88.js} +4 -4
  18. package/dist/{benchmarks-BEOkuvIg.js.map → benchmarks-BlPmjd88.js.map} +1 -1
  19. package/dist/campaign/index.d.ts +7 -5
  20. package/dist/campaign/index.js +5 -3
  21. package/dist/{campaign-CXsdyym7.js → campaign--HVSuvV0.js} +17 -301
  22. package/dist/campaign--HVSuvV0.js.map +1 -0
  23. package/dist/cli.js +2 -2
  24. package/dist/{client-0JI64ovJ.d.ts → client-DjXROWpx.d.ts} +3 -3
  25. package/dist/{client-0JI64ovJ.d.ts.map → client-DjXROWpx.d.ts.map} +1 -1
  26. package/dist/{completion-verifier-CBiee74w.d.ts → completion-verifier-foUCLif_.d.ts} +5 -5
  27. package/dist/{completion-verifier-CBiee74w.d.ts.map → completion-verifier-foUCLif_.d.ts.map} +1 -1
  28. package/dist/contract/index.d.ts +9 -8
  29. package/dist/contract/index.d.ts.map +1 -1
  30. package/dist/contract/index.js +7 -6
  31. package/dist/contract/index.js.map +1 -1
  32. package/dist/control.d.ts +2 -2
  33. package/dist/control.js +1 -1
  34. package/dist/counterfactual-CWPTrMH7.js +126 -0
  35. package/dist/counterfactual-CWPTrMH7.js.map +1 -0
  36. package/dist/counterfactual-CxmxAONP.d.ts +72 -0
  37. package/dist/counterfactual-CxmxAONP.d.ts.map +1 -0
  38. package/dist/{default-registry-J9m-_tya.d.ts → default-registry-BZhStdjl.d.ts} +6 -5
  39. package/dist/default-registry-BZhStdjl.d.ts.map +1 -0
  40. package/dist/{default-registry-Dta70shL.js → default-registry-BaQXW1Ow.js} +2 -2
  41. package/dist/{default-registry-Dta70shL.js.map → default-registry-BaQXW1Ow.js.map} +1 -1
  42. package/dist/{dspy-rlm-engine-19FQEMBK.js → dspy-rlm-engine-BiN49gK6.js} +4 -3
  43. package/dist/{dspy-rlm-engine-19FQEMBK.js.map → dspy-rlm-engine-BiN49gK6.js.map} +1 -1
  44. package/dist/{tool-groups-CK0JCkqO.d.ts → engine-nB64f48I.d.ts} +18 -31
  45. package/dist/engine-nB64f48I.d.ts.map +1 -0
  46. package/dist/{eval-campaign-CfLQQs9B.js → eval-campaign-DNjCvAm-.js} +7 -6
  47. package/dist/{eval-campaign-CfLQQs9B.js.map → eval-campaign-DNjCvAm-.js.map} +1 -1
  48. package/dist/{exact-types-CBYF5MGd.d.ts → exact-types-Djvzosly.d.ts} +2 -2
  49. package/dist/{exact-types-CBYF5MGd.d.ts.map → exact-types-Djvzosly.d.ts.map} +1 -1
  50. package/dist/exec-BLtYZdWo.js +49 -0
  51. package/dist/exec-BLtYZdWo.js.map +1 -0
  52. package/dist/experiment/index.d.ts +802 -0
  53. package/dist/experiment/index.d.ts.map +1 -0
  54. package/dist/experiment/index.js +1108 -0
  55. package/dist/experiment/index.js.map +1 -0
  56. package/dist/experiment-tracker-CnRICnMl.js +500 -0
  57. package/dist/experiment-tracker-CnRICnMl.js.map +1 -0
  58. package/dist/experiment-tracker-IMntXr6J.d.ts +299 -0
  59. package/dist/experiment-tracker-IMntXr6J.d.ts.map +1 -0
  60. package/dist/{external-optimizer-contracts-Q9c0L3eY.d.ts → external-optimizer-contracts-CZuJNcT5.d.ts} +2 -2
  61. package/dist/{external-optimizer-contracts-Q9c0L3eY.d.ts.map → external-optimizer-contracts-CZuJNcT5.d.ts.map} +1 -1
  62. package/dist/{feedback-trajectory-GgoS0-MK.d.ts → feedback-trajectory-Rh280oXo.d.ts} +3 -3
  63. package/dist/{feedback-trajectory-GgoS0-MK.d.ts.map → feedback-trajectory-Rh280oXo.d.ts.map} +1 -1
  64. package/dist/fuzz.d.ts +1 -1
  65. package/dist/hosted/index.d.ts +3 -3
  66. package/dist/{index-BZUe-ODI.d.ts → index-BZ3-y4YL.d.ts} +3 -3
  67. package/dist/{index-BZUe-ODI.d.ts.map → index-BZ3-y4YL.d.ts.map} +1 -1
  68. package/dist/{index-DSC51roc.d.ts → index-C3ssXVLv.d.ts} +2 -2
  69. package/dist/{index-DSC51roc.d.ts.map → index-C3ssXVLv.d.ts.map} +1 -1
  70. package/dist/{index-4XwggC10.d.ts → index-C5HOo4ZF2.d.ts} +4 -4
  71. package/dist/index-C5HOo4ZF2.d.ts.map +1 -0
  72. package/dist/{index-B6-B0zTB.d.ts → index-CvXXlyz7.d.ts} +2 -2
  73. package/dist/{index-B6-B0zTB.d.ts.map → index-CvXXlyz7.d.ts.map} +1 -1
  74. package/dist/{index-Dx1kF3Ez.d.ts → index-CwDrUMe0.d.ts} +2 -2
  75. package/dist/{index-Dx1kF3Ez.d.ts.map → index-CwDrUMe0.d.ts.map} +1 -1
  76. package/dist/{index-BIL5vxxt.d.ts → index-Sh2I0DRc.d.ts} +11 -645
  77. package/dist/index-Sh2I0DRc.d.ts.map +1 -0
  78. package/dist/index.d.ts +214 -404
  79. package/dist/index.d.ts.map +1 -1
  80. package/dist/index.js +233 -649
  81. package/dist/index.js.map +1 -1
  82. package/dist/{insight-report-DqEsugpr.d.ts → insight-report-C6h6F_4L.d.ts} +3 -3
  83. package/dist/{insight-report-DqEsugpr.d.ts.map → insight-report-C6h6F_4L.d.ts.map} +1 -1
  84. package/dist/{integrity-CNGUaGBY.d.ts → integrity-BuqEKu-x.d.ts} +2 -2
  85. package/dist/{integrity-CNGUaGBY.d.ts.map → integrity-BuqEKu-x.d.ts.map} +1 -1
  86. package/dist/integrity-MLzHOfV9.js +141 -0
  87. package/dist/integrity-MLzHOfV9.js.map +1 -0
  88. package/dist/kind-factory-BHIgPmzS.js.map +1 -1
  89. package/dist/{llm-client-Dv5BiKLE.js → llm-client-DzvMUsS_.js} +24 -6
  90. package/dist/llm-client-DzvMUsS_.js.map +1 -0
  91. package/dist/matrix/index.d.ts +2 -2
  92. package/dist/meta-eval/index.d.ts +1 -1
  93. package/dist/{mint-DD-0oQTA.js → mint-CGEkzPLf.js} +2 -2
  94. package/dist/{mint-DD-0oQTA.js.map → mint-CGEkzPLf.js.map} +1 -1
  95. package/dist/{multi-layer-verifier-BHY1gWAc.d.ts → multi-layer-verifier-DnAqwl0h.d.ts} +2 -2
  96. package/dist/{multi-layer-verifier-BHY1gWAc.d.ts.map → multi-layer-verifier-DnAqwl0h.d.ts.map} +1 -1
  97. package/dist/multishot/index.d.ts +2 -2
  98. package/dist/openapi.json +1 -1
  99. package/dist/paired-promotion-decision-B6zJ3gYM.d.ts +114 -0
  100. package/dist/paired-promotion-decision-B6zJ3gYM.d.ts.map +1 -0
  101. package/dist/pipelines/index.d.ts +2 -1
  102. package/dist/pipelines/index.d.ts.map +1 -1
  103. package/dist/pre-registration-DakwTRXk.js +96 -0
  104. package/dist/pre-registration-DakwTRXk.js.map +1 -0
  105. package/dist/prime-bridge-transport-6feEglLf.d.ts +49 -0
  106. package/dist/prime-bridge-transport-6feEglLf.d.ts.map +1 -0
  107. package/dist/prime-protocol-BfSalTfR.js +453 -0
  108. package/dist/prime-protocol-BfSalTfR.js.map +1 -0
  109. package/dist/profile-cell.js +242 -1
  110. package/dist/profile-cell.js.map +1 -0
  111. package/dist/promotion-policy-ChWhTDBH.d.ts +289 -0
  112. package/dist/promotion-policy-ChWhTDBH.d.ts.map +1 -0
  113. package/dist/promotion-policy-CrLrmys8.js +682 -0
  114. package/dist/promotion-policy-CrLrmys8.js.map +1 -0
  115. package/dist/{propose-review-control-BciUCZoh.js → propose-review-control-qgWLA6E9.js} +2 -2
  116. package/dist/{propose-review-control-BciUCZoh.js.map → propose-review-control-qgWLA6E9.js.map} +1 -1
  117. package/dist/{release-report-ChOgpIoQ.d.ts → release-report-CI8uisI1.d.ts} +2 -2
  118. package/dist/{release-report-ChOgpIoQ.d.ts.map → release-report-CI8uisI1.d.ts.map} +1 -1
  119. package/dist/{release-report-Sl0xfkFv.js → release-report-Dy39cbFF.js} +3 -3
  120. package/dist/{release-report-Sl0xfkFv.js.map → release-report-Dy39cbFF.js.map} +1 -1
  121. package/dist/{replay-Krvb114g.d.ts → replay-DFf-teiC.d.ts} +5 -4
  122. package/dist/replay-DFf-teiC.d.ts.map +1 -0
  123. package/dist/{replay-GW61ezMW.js → replay-DyBLaKFc.js} +3 -3
  124. package/dist/{replay-GW61ezMW.js.map → replay-DyBLaKFc.js.map} +1 -1
  125. package/dist/reporting.d.ts +3 -3
  126. package/dist/reporting.js +2 -2
  127. package/dist/{researcher-xLeNcpKX.d.ts → researcher-BoaxeCzP.d.ts} +4 -4
  128. package/dist/{researcher-xLeNcpKX.d.ts.map → researcher-BoaxeCzP.d.ts.map} +1 -1
  129. package/dist/{reward-hacking-CyuzxKly.js → reward-hacking-B2tPL9a3.js} +12 -3
  130. package/dist/reward-hacking-B2tPL9a3.js.map +1 -0
  131. package/dist/{reward-hacking-RZgnGWlx.d.ts → reward-hacking-Cf1PtEOz.d.ts} +33 -3
  132. package/dist/reward-hacking-Cf1PtEOz.d.ts.map +1 -0
  133. package/dist/rl.d.ts +17 -7
  134. package/dist/rl.d.ts.map +1 -1
  135. package/dist/rl.js +16 -7
  136. package/dist/rl.js.map +1 -1
  137. package/dist/rollout/index.d.ts +2 -2
  138. package/dist/rollout/index.js +2 -2
  139. package/dist/{rollout-C2fD1cf4.js → rollout-2ECTXb2N.js} +2 -2
  140. package/dist/{rollout-C2fD1cf4.js.map → rollout-2ECTXb2N.js.map} +1 -1
  141. package/dist/{run-evidence-C6G41MSI.d.ts → run-evidence-BDFFai9R.d.ts} +2 -2
  142. package/dist/{run-evidence-C6G41MSI.d.ts.map → run-evidence-BDFFai9R.d.ts.map} +1 -1
  143. package/dist/{run-record-CWN8-VsV.js → run-record-DqOw5X6_.js} +2 -2
  144. package/dist/{run-record-CWN8-VsV.js.map → run-record-DqOw5X6_.js.map} +1 -1
  145. package/dist/{schema-Cef2cFmb.d.ts → schema-Cef2cFmb2.d.ts} +1 -1
  146. package/dist/schema-Cef2cFmb2.d.ts.map +1 -0
  147. package/dist/{semantic-concept-judge-DwF6n05O.js → semantic-concept-judge-Bmrq6yqU.js} +2 -2
  148. package/dist/{semantic-concept-judge-DwF6n05O.js.map → semantic-concept-judge-Bmrq6yqU.js.map} +1 -1
  149. package/dist/sequential-D-BLJBKU.js +299 -0
  150. package/dist/sequential-D-BLJBKU.js.map +1 -0
  151. package/dist/{server-D6XJQHw7.js → server-iu0ede49.js} +2 -2
  152. package/dist/{server-D6XJQHw7.js.map → server-iu0ede49.js.map} +1 -1
  153. package/dist/{single-run-lock-D5iN0Xzb.js → single-run-lock-BMQEv1wG.js} +7 -161
  154. package/dist/single-run-lock-BMQEv1wG.js.map +1 -0
  155. package/dist/{skill-usage-GlOphAhX.d.ts → skill-usage-CJlWEUFt.d.ts} +10 -10
  156. package/dist/{skill-usage-GlOphAhX.d.ts.map → skill-usage-CJlWEUFt.d.ts.map} +1 -1
  157. package/dist/{skillopt-optimization-method-7S43rbDB.d.ts → skillopt-optimization-method-BO7NAl3b.d.ts} +8 -294
  158. package/dist/skillopt-optimization-method-BO7NAl3b.d.ts.map +1 -0
  159. package/dist/{skillopt-optimization-method-Bfb-vBKe.js → skillopt-optimization-method-CkaI2ly4.js} +19 -654
  160. package/dist/skillopt-optimization-method-CkaI2ly4.js.map +1 -0
  161. package/dist/statistical-heldout-Dn9ruizm.d.ts +555 -0
  162. package/dist/statistical-heldout-Dn9ruizm.d.ts.map +1 -0
  163. package/dist/{statistics-C-dm-J6H.d.ts → statistics-D6Uebe_4.d.ts} +2 -2
  164. package/dist/{statistics-C-dm-J6H.d.ts.map → statistics-D6Uebe_4.d.ts.map} +1 -1
  165. package/dist/steps-BArUxhna.d.ts +51 -0
  166. package/dist/steps-BArUxhna.d.ts.map +1 -0
  167. package/dist/{integrity-fdt8XPAv.js → store-DNe_Uv1Q.js} +2 -140
  168. package/dist/store-DNe_Uv1Q.js.map +1 -0
  169. package/dist/{summary-report-B0cAyA7N.d.ts → summary-report-DuUS_i7W.d.ts} +3 -114
  170. package/dist/summary-report-DuUS_i7W.d.ts.map +1 -0
  171. package/dist/{summary-report-9A5y7EsK.js → summary-report-Lf-5I7xh.js} +2 -2
  172. package/dist/{summary-report-9A5y7EsK.js.map → summary-report-Lf-5I7xh.js.map} +1 -1
  173. package/dist/supervisor-run/index.d.ts +2 -2
  174. package/dist/supervisor-run/index.js +1 -1
  175. package/dist/{supervisor-run-DiyQVczd.js → supervisor-run-D_sokXcO.js} +32 -8
  176. package/dist/{supervisor-run-DiyQVczd.js.map → supervisor-run-D_sokXcO.js.map} +1 -1
  177. package/dist/tool-groups-DIVBnJyl.d.ts +28 -0
  178. package/dist/tool-groups-DIVBnJyl.d.ts.map +1 -0
  179. package/dist/trace-repair/index.d.ts +2102 -0
  180. package/dist/trace-repair/index.d.ts.map +1 -0
  181. package/dist/trace-repair/index.js +3878 -0
  182. package/dist/trace-repair/index.js.map +1 -0
  183. package/dist/traces.d.ts +6 -6
  184. package/dist/traces.js +3 -2
  185. package/dist/trajectory-YC15QDYQ.d.ts +24 -0
  186. package/dist/trajectory-YC15QDYQ.d.ts.map +1 -0
  187. package/dist/trajectory-replay/index.d.ts +781 -0
  188. package/dist/trajectory-replay/index.d.ts.map +1 -0
  189. package/dist/trajectory-replay/index.js +2103 -0
  190. package/dist/trajectory-replay/index.js.map +1 -0
  191. package/dist/{types-XMVEdrE_.d.ts → types-D216SgwM.d.ts} +24 -6
  192. package/dist/types-D216SgwM.d.ts.map +1 -0
  193. package/dist/{types-yLK8gXE9.d.ts → types-D4mog56g.d.ts} +2 -2
  194. package/dist/{types-yLK8gXE9.d.ts.map → types-D4mog56g.d.ts.map} +1 -1
  195. package/dist/{types-BhP9q0Fq.d.ts → types-DF_Udrp-.d.ts} +52 -3
  196. package/dist/{types-BhP9q0Fq.d.ts.map → types-DF_Udrp-.d.ts.map} +1 -1
  197. package/dist/{types-DOZyvsFU.d.ts → types-DYuNHo9R.d.ts} +3 -3
  198. package/dist/{types-DOZyvsFU.d.ts.map → types-DYuNHo9R.d.ts.map} +1 -1
  199. package/dist/usage-receipt-EVI8B8Xu.js.map +1 -1
  200. package/dist/verdict-DExhxfgR.d.ts +201 -0
  201. package/dist/verdict-DExhxfgR.d.ts.map +1 -0
  202. package/dist/verdict-cache-BCcOh0kF.js +159 -0
  203. package/dist/verdict-cache-BCcOh0kF.js.map +1 -0
  204. package/dist/wire/index.d.ts +2 -2
  205. package/dist/wire/index.js +1 -1
  206. package/docs/campaign-proposers.md +5 -0
  207. package/docs/charter.md +112 -0
  208. package/docs/experiment.md +104 -0
  209. package/docs/prime-analyst.md +1 -0
  210. package/docs/trace-analysis.md +26 -0
  211. package/docs/trace-repair-admission.md +194 -0
  212. package/docs/trace-repair-analyst-arms.md +121 -0
  213. package/docs/trace-repair-continuation.md +107 -0
  214. package/docs/trace-repair-grader.md +163 -0
  215. package/docs/trajectory-replay.md +110 -0
  216. package/docs/verification-strategies.md +103 -0
  217. package/package.json +19 -2
  218. package/dist/agent-profile-cell-CbfBm2g6.js +0 -335
  219. package/dist/agent-profile-cell-CbfBm2g6.js.map +0 -1
  220. package/dist/baseline-D_fT6277.d.ts.map +0 -1
  221. package/dist/benchmark-J9Qe6j2_.d.ts.map +0 -1
  222. package/dist/benchmark-command-CQd78YHt.js.map +0 -1
  223. package/dist/campaign-CXsdyym7.js.map +0 -1
  224. package/dist/default-registry-J9m-_tya.d.ts.map +0 -1
  225. package/dist/index-4XwggC10.d.ts.map +0 -1
  226. package/dist/index-BIL5vxxt.d.ts.map +0 -1
  227. package/dist/integrity-fdt8XPAv.js.map +0 -1
  228. package/dist/llm-client-Dv5BiKLE.js.map +0 -1
  229. package/dist/replay-Krvb114g.d.ts.map +0 -1
  230. package/dist/reward-hacking-CyuzxKly.js.map +0 -1
  231. package/dist/reward-hacking-RZgnGWlx.d.ts.map +0 -1
  232. package/dist/schema-Cef2cFmb.d.ts.map +0 -1
  233. package/dist/single-run-lock-D5iN0Xzb.js.map +0 -1
  234. package/dist/skillopt-optimization-method-7S43rbDB.d.ts.map +0 -1
  235. package/dist/skillopt-optimization-method-Bfb-vBKe.js.map +0 -1
  236. package/dist/summary-report-B0cAyA7N.d.ts.map +0 -1
  237. package/dist/tool-groups-CK0JCkqO.d.ts.map +0 -1
  238. package/dist/types-XMVEdrE_.d.ts.map +0 -1
  239. package/dist/verdict-Dps8_okt.d.ts +0 -37
  240. package/dist/verdict-Dps8_okt.d.ts.map +0 -1
@@ -0,0 +1,28 @@
1
+ import { w as TraceAnalysisStore } from "./types-DF_Udrp-.js";
2
+ import { _ as TraceAnalysisToolDescriptor } from "./engine-nB64f48I.js";
3
+ //#region src/analyst/tool-groups.d.ts
4
+ /** Named tool sets. Kinds pass `tools: TRACE_TOOL_GROUPS.failureForensics` etc. */
5
+ type TraceToolGroupName =
6
+ /** All seven tools. Use for open-ended discovery kinds. */
7
+ 'all' |
8
+ /** Overview + paginated query + count. No deep reads. Cheap. */
9
+ 'discovery' |
10
+ /** Discovery + viewTrace + viewSpans. Deep-read but no regex search. */
11
+ 'discoveryAndRead' |
12
+ /** Discovery + search tools. For pattern-matching across many traces. */
13
+ 'discoveryAndSearch' |
14
+ /** Discovery + viewSpans + searchSpan. Targeted-span work after another kind narrows down. */
15
+ 'targeted' |
16
+ /** One known-small trace: overview, bounded reads, and in-trace search. */
17
+ 'singleTrace';
18
+ /**
19
+ * Build the tool set for a named group bound to a specific trace store.
20
+ *
21
+ * `all` returns every tool. Other groups filter the canonical descriptors
22
+ * by name to the documented subset. An unrecognised group name throws —
23
+ * silently returning all tools would defeat the cost-control point.
24
+ */
25
+ declare function buildTraceToolsForGroup(group: TraceToolGroupName, store: TraceAnalysisStore): TraceAnalysisToolDescriptor[];
26
+ //#endregion
27
+ export { buildTraceToolsForGroup as n, TraceToolGroupName as t };
28
+ //# sourceMappingURL=tool-groups-DIVBnJyl.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"tool-groups-DIVBnJyl.d.ts","names":[],"sources":["../src/analyst/tool-groups.ts"],"mappings":";;;;KAmBY;;;;;;;;;;;;;;;;;;;;iBAgDI,wBACd,OAAO,oBACP,OAAO,qBACN"}