@tangle-network/agent-eval 0.144.8 → 0.144.9

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 (117) hide show
  1. package/CHANGELOG.md +6 -0
  2. package/dist/{agent-profile-CgDTo40f.d.ts → agent-profile-DPi7IZg7.d.ts} +3 -3
  3. package/dist/{agent-profile-CgDTo40f.d.ts.map → agent-profile-DPi7IZg7.d.ts.map} +1 -1
  4. package/dist/analyst/index.d.ts +9 -9
  5. package/dist/analyst/index.js +1 -1
  6. package/dist/{analyze-runs-BNNK7irB.js → analyze-runs-C30yljDJ.js} +2 -2
  7. package/dist/{analyze-runs-BNNK7irB.js.map → analyze-runs-C30yljDJ.js.map} +1 -1
  8. package/dist/{benchmark-command-BKENp2s5.js → benchmark-command-Dtym4pA-.js} +2 -2
  9. package/dist/{benchmark-command-BKENp2s5.js.map → benchmark-command-Dtym4pA-.js.map} +1 -1
  10. package/dist/benchmarks/index.d.ts +1 -1
  11. package/dist/benchmarks/index.js +1 -1
  12. package/dist/{benchmarks-BlPmjd88.js → benchmarks-Ce4Ae5fX.js} +3 -3
  13. package/dist/{benchmarks-BlPmjd88.js.map → benchmarks-Ce4Ae5fX.js.map} +1 -1
  14. package/dist/campaign/index.d.ts +6 -6
  15. package/dist/campaign/index.js +2 -2
  16. package/dist/{campaign--HVSuvV0.js → campaign-FG3thH2m.js} +4 -4
  17. package/dist/{campaign--HVSuvV0.js.map → campaign-FG3thH2m.js.map} +1 -1
  18. package/dist/cli.js +1 -1
  19. package/dist/{client-DjXROWpx.d.ts → client-C9gzZE59.d.ts} +4 -4
  20. package/dist/{client-DjXROWpx.d.ts.map → client-C9gzZE59.d.ts.map} +1 -1
  21. package/dist/{completion-verifier-foUCLif_.d.ts → completion-verifier-DJA5BhPb.d.ts} +3 -3
  22. package/dist/{completion-verifier-foUCLif_.d.ts.map → completion-verifier-DJA5BhPb.d.ts.map} +1 -1
  23. package/dist/contract/index.d.ts +9 -9
  24. package/dist/contract/index.js +4 -4
  25. package/dist/contract/index.js.map +1 -1
  26. package/dist/control.d.ts +2 -2
  27. package/dist/control.js +1 -1
  28. package/dist/{default-registry-BZhStdjl.d.ts → default-registry-Bf8Woqmq.d.ts} +6 -6
  29. package/dist/{default-registry-BZhStdjl.d.ts.map → default-registry-Bf8Woqmq.d.ts.map} +1 -1
  30. package/dist/{engine-nB64f48I.d.ts → engine-3hL-XqwJ.d.ts} +7 -7
  31. package/dist/{engine-nB64f48I.d.ts.map → engine-3hL-XqwJ.d.ts.map} +1 -1
  32. package/dist/{eval-campaign-DNjCvAm-.js → eval-campaign-B_7wcnav.js} +2 -2
  33. package/dist/{eval-campaign-DNjCvAm-.js.map → eval-campaign-B_7wcnav.js.map} +1 -1
  34. package/dist/{exact-types-Djvzosly.d.ts → exact-types-DSFFpLLI.d.ts} +2 -2
  35. package/dist/{exact-types-Djvzosly.d.ts.map → exact-types-DSFFpLLI.d.ts.map} +1 -1
  36. package/dist/experiment/index.d.ts +2 -2
  37. package/dist/{feedback-trajectory-Rh280oXo.d.ts → feedback-trajectory-CacHpxsp.d.ts} +3 -3
  38. package/dist/{feedback-trajectory-Rh280oXo.d.ts.map → feedback-trajectory-CacHpxsp.d.ts.map} +1 -1
  39. package/dist/hosted/index.d.ts +3 -3
  40. package/dist/{index-CvXXlyz7.d.ts → index-BnEuDAK2.d.ts} +2 -2
  41. package/dist/{index-CvXXlyz7.d.ts.map → index-BnEuDAK2.d.ts.map} +1 -1
  42. package/dist/{index-Sh2I0DRc.d.ts → index-CBEMKO8d.d.ts} +9 -9
  43. package/dist/{index-Sh2I0DRc.d.ts.map → index-CBEMKO8d.d.ts.map} +1 -1
  44. package/dist/{index-C5HOo4ZF2.d.ts → index-CLKMJTJF2.d.ts} +5 -5
  45. package/dist/{index-C5HOo4ZF2.d.ts.map → index-CLKMJTJF2.d.ts.map} +1 -1
  46. package/dist/{index-CwDrUMe0.d.ts → index-CQTZ-4XN.d.ts} +2 -2
  47. package/dist/{index-CwDrUMe0.d.ts.map → index-CQTZ-4XN.d.ts.map} +1 -1
  48. package/dist/index.d.ts +22 -22
  49. package/dist/index.js +11 -11
  50. package/dist/{insight-report-C6h6F_4L.d.ts → insight-report-CgX_s0Ez.d.ts} +3 -3
  51. package/dist/{insight-report-C6h6F_4L.d.ts.map → insight-report-CgX_s0Ez.d.ts.map} +1 -1
  52. package/dist/meta-eval/index.d.ts +1 -1
  53. package/dist/{mint-CGEkzPLf.js → mint-B2O60ACG.js} +2 -2
  54. package/dist/{mint-CGEkzPLf.js.map → mint-B2O60ACG.js.map} +1 -1
  55. package/dist/multishot/index.d.ts +1 -1
  56. package/dist/openapi.json +1 -1
  57. package/dist/{promotion-policy-ChWhTDBH.d.ts → promotion-policy-Ckjhzg_4.d.ts} +2 -2
  58. package/dist/{promotion-policy-ChWhTDBH.d.ts.map → promotion-policy-Ckjhzg_4.d.ts.map} +1 -1
  59. package/dist/{propose-review-control-qgWLA6E9.js → propose-review-control-dSNPjFUH.js} +2 -2
  60. package/dist/{propose-review-control-qgWLA6E9.js.map → propose-review-control-dSNPjFUH.js.map} +1 -1
  61. package/dist/{release-report-Dy39cbFF.js → release-report-BUYmoKo2.js} +2 -2
  62. package/dist/{release-report-Dy39cbFF.js.map → release-report-BUYmoKo2.js.map} +1 -1
  63. package/dist/{release-report-CI8uisI1.d.ts → release-report-DA2BCu5a.d.ts} +3 -3
  64. package/dist/{release-report-CI8uisI1.d.ts.map → release-report-DA2BCu5a.d.ts.map} +1 -1
  65. package/dist/{replay-DyBLaKFc.js → replay-B3gACG_H.js} +2 -2
  66. package/dist/{replay-DyBLaKFc.js.map → replay-B3gACG_H.js.map} +1 -1
  67. package/dist/{replay-DFf-teiC.d.ts → replay-DbhZ4Ked.d.ts} +5 -5
  68. package/dist/{replay-DFf-teiC.d.ts.map → replay-DbhZ4Ked.d.ts.map} +1 -1
  69. package/dist/reporting.d.ts +3 -3
  70. package/dist/reporting.js +1 -1
  71. package/dist/{researcher-BoaxeCzP.d.ts → researcher-BchpD55R.d.ts} +3 -3
  72. package/dist/{researcher-BoaxeCzP.d.ts.map → researcher-BchpD55R.d.ts.map} +1 -1
  73. package/dist/{reward-hacking-B2tPL9a3.js → reward-hacking-BDToousL.js} +4 -3
  74. package/dist/reward-hacking-BDToousL.js.map +1 -0
  75. package/dist/{reward-hacking-Cf1PtEOz.d.ts → reward-hacking-BI0OMAlo.d.ts} +2 -2
  76. package/dist/{reward-hacking-Cf1PtEOz.d.ts.map → reward-hacking-BI0OMAlo.d.ts.map} +1 -1
  77. package/dist/rl.d.ts +5 -5
  78. package/dist/rl.js +4 -4
  79. package/dist/rollout/index.d.ts +1 -1
  80. package/dist/rollout/index.js +2 -2
  81. package/dist/{rollout-2ECTXb2N.js → rollout-DRrksrcV.js} +2 -2
  82. package/dist/{rollout-2ECTXb2N.js.map → rollout-DRrksrcV.js.map} +1 -1
  83. package/dist/{rubric-predictive-validity-9qAwzkZm.d.ts → rubric-predictive-validity-DvkjPYCe.d.ts} +2 -2
  84. package/dist/{rubric-predictive-validity-9qAwzkZm.d.ts.map → rubric-predictive-validity-DvkjPYCe.d.ts.map} +1 -1
  85. package/dist/{run-evidence-BDFFai9R.d.ts → run-evidence-Dhi3C81V.d.ts} +3 -3
  86. package/dist/{run-evidence-BDFFai9R.d.ts.map → run-evidence-Dhi3C81V.d.ts.map} +1 -1
  87. package/dist/{run-record-DqOw5X6_.js → run-record-BmSPWXJR.js} +2 -1
  88. package/dist/run-record-BmSPWXJR.js.map +1 -0
  89. package/dist/{run-record-DdSa93_W.d.ts → run-record-CF4Dwpxr.d.ts} +4 -1
  90. package/dist/run-record-CF4Dwpxr.d.ts.map +1 -0
  91. package/dist/{skill-usage-CJlWEUFt.d.ts → skill-usage-CO9OLRBx.d.ts} +6 -6
  92. package/dist/{skill-usage-CJlWEUFt.d.ts.map → skill-usage-CO9OLRBx.d.ts.map} +1 -1
  93. package/dist/{skillopt-optimization-method-BO7NAl3b.d.ts → skillopt-optimization-method-D8wysvU1.d.ts} +6 -6
  94. package/dist/{skillopt-optimization-method-BO7NAl3b.d.ts.map → skillopt-optimization-method-D8wysvU1.d.ts.map} +1 -1
  95. package/dist/{skillopt-optimization-method-CkaI2ly4.js → skillopt-optimization-method-Dtzp7QpP.js} +3 -2
  96. package/dist/skillopt-optimization-method-Dtzp7QpP.js.map +1 -0
  97. package/dist/{statistical-heldout-Dn9ruizm.d.ts → statistical-heldout-TQ-4CYiN.d.ts} +3 -3
  98. package/dist/{statistical-heldout-Dn9ruizm.d.ts.map → statistical-heldout-TQ-4CYiN.d.ts.map} +1 -1
  99. package/dist/{summary-report-DuUS_i7W.d.ts → summary-report-BNR7DWTj.d.ts} +2 -2
  100. package/dist/{summary-report-DuUS_i7W.d.ts.map → summary-report-BNR7DWTj.d.ts.map} +1 -1
  101. package/dist/{tool-groups-DIVBnJyl.d.ts → tool-groups-CZPGGlHf.d.ts} +3 -3
  102. package/dist/tool-groups-CZPGGlHf.d.ts.map +1 -0
  103. package/dist/trace-repair/index.d.ts +2 -2
  104. package/dist/traces.d.ts +4 -4
  105. package/dist/traces.js +1 -1
  106. package/dist/{types-DYuNHo9R.d.ts → types-BnjdJ70P.d.ts} +3 -3
  107. package/dist/{types-DYuNHo9R.d.ts.map → types-BnjdJ70P.d.ts.map} +1 -1
  108. package/dist/{types-DF_Udrp-.d.ts → types-BvDKaULh.d.ts} +2 -2
  109. package/dist/{types-DF_Udrp-.d.ts.map → types-BvDKaULh.d.ts.map} +1 -1
  110. package/dist/wire/index.d.ts +1 -1
  111. package/docs/concepts.md +1 -0
  112. package/package.json +1 -1
  113. package/dist/reward-hacking-B2tPL9a3.js.map +0 -1
  114. package/dist/run-record-DdSa93_W.d.ts.map +0 -1
  115. package/dist/run-record-DqOw5X6_.js.map +0 -1
  116. package/dist/skillopt-optimization-method-CkaI2ly4.js.map +0 -1
  117. package/dist/tool-groups-DIVBnJyl.d.ts.map +0 -1
package/dist/cli.js CHANGED
@@ -1,6 +1,6 @@
1
1
  #!/usr/bin/env node
2
2
  import { o as runRolloutReleaseCli } from "./hf-dataset-XggBupCr.js";
3
- import { n as runAnalystBenchmarkCommand } from "./benchmark-command-BKENp2s5.js";
3
+ import { n as runAnalystBenchmarkCommand } from "./benchmark-command-Dtym4pA-.js";
4
4
  import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-iu0ede49.js";
5
5
  import { writeFileSync } from "node:fs";
6
6
  //#region src/cli-config.ts
@@ -1,6 +1,6 @@
1
- import { l as RunTerminalOutcome } from "./run-record-DdSa93_W.js";
2
- import { _ as GateDecision, j as MutableSurface } from "./types-DYuNHo9R.js";
3
- import { o as InsightReport } from "./insight-report-C6h6F_4L.js";
1
+ import { l as RunTerminalOutcome } from "./run-record-CF4Dwpxr.js";
2
+ import { _ as GateDecision, j as MutableSurface } from "./types-BnjdJ70P.js";
3
+ import { o as InsightReport } from "./insight-report-CgX_s0Ez.js";
4
4
  //#region src/hosted/types.d.ts
5
5
  declare const HOSTED_WIRE_VERSION: '2026-07-24.v1';
6
6
  type HostedWireVersion = typeof HOSTED_WIRE_VERSION;
@@ -199,4 +199,4 @@ declare function hostedClientFromEnv(overrides?: Partial<HostedTenant> & {
199
199
  }): HostedClient | undefined;
200
200
  //#endregion
201
201
  export { UnixNanoTimestamp as _, hostedTenantFromEnv as a, EvalRunGenerationSnapshot as c, HostedIngestHeaders as d, HostedWireVersion as f, TraceSpanEvent as g, IngestTracesRequest as h, hostedClientFromEnv as i, EvalRunStatus as l, IngestResponse as m, HostedTenant as n, EvalRunCellScore as o, IngestEvalRunsRequest as p, createHostedClient as r, EvalRunEvent as s, HostedClient as t, HOSTED_WIRE_VERSION as u };
202
- //# sourceMappingURL=client-DjXROWpx.d.ts.map
202
+ //# sourceMappingURL=client-C9gzZE59.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"client-DjXROWpx.d.ts","names":[],"sources":["../src/hosted/types.ts","../src/hosted/client.ts"],"mappings":";;;;cAgCa;KACD,2BAA2B;;UAKtB;;EAEf;;EAEA;;EAEA,yBAAyB;;EAEzB;;;KAMU;UAQK;;EAEf;;EAEA;;EAEA;;EAEA,YAAY,eAAe;;EAE3B,iBAAiB;;EAEjB;;EAEA;;UAGe;;EAEf;;;EAGA;;;EAGA,UAAU;;EAEV,OAAO;;EAEP;;EAEA;;EAEA;;;;;;;UAQe;;EAEf;;EAEA;;EAEA;;EAEA,QAAQ;;EAER,QAAQ;;EAER,WAAW;;EAEX,aAAa;;EAEb,eAAe;;EAEf;;EAEA;;EAEA;;EAEA;;;;;;EAMA,gBAAgB;;;;;;KASN;;;;;;UAOK;EACf;EACA;EACA;EACA;EACA,mBAAmB;EACnB,iBAAiB;EACjB,YAAY;EACZ,SAAS;IACP,cAAc;IACd;IACA,aAAa;;EAEf;IAAW;IAAgC;;;EAE3C;;EAEA;;EAEA;;EAEA;;UAKe;EACf,aAAa;EACb,QAAQ;;UAGO;EACf,aAAa;EACb,OAAO;;UAGQ;;EAEf;;EAEA,UAAU;IAAQ;IAAe;;;;;UC1JlB;;EAEf;;EAEA;;EAEA;;EAEA,mBAAmB;;EAEnB;;EAEA;;UAGe;EACf,cAAc,OAAO,cAAc,0BAA0B,QAAQ;EACrE,eAAe,QAAQ,gBAAgB,0BAA0B,QAAQ;EACzE,aAAa,OAAO,kBAAkB,0BAA0B,QAAQ;WAC/D,QAAQ;WACR,aAAa;;iBAyGR,mBAAmB,QAAQ,eAAe;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAsE1C,oBACd,YAAW,QAAQ;EAAkB,MAAM;IAC1C;iBAiBa,oBACd,YAAW,QAAQ;EAAkB,MAAM;IAC1C"}
1
+ {"version":3,"file":"client-C9gzZE59.d.ts","names":[],"sources":["../src/hosted/types.ts","../src/hosted/client.ts"],"mappings":";;;;cAgCa;KACD,2BAA2B;;UAKtB;;EAEf;;EAEA;;EAEA,yBAAyB;;EAEzB;;;KAMU;UAQK;;EAEf;;EAEA;;EAEA;;EAEA,YAAY,eAAe;;EAE3B,iBAAiB;;EAEjB;;EAEA;;UAGe;;EAEf;;;EAGA;;;EAGA,UAAU;;EAEV,OAAO;;EAEP;;EAEA;;EAEA;;;;;;;UAQe;;EAEf;;EAEA;;EAEA;;EAEA,QAAQ;;EAER,QAAQ;;EAER,WAAW;;EAEX,aAAa;;EAEb,eAAe;;EAEf;;EAEA;;EAEA;;EAEA;;;;;;EAMA,gBAAgB;;;;;;KASN;;;;;;UAOK;EACf;EACA;EACA;EACA;EACA,mBAAmB;EACnB,iBAAiB;EACjB,YAAY;EACZ,SAAS;IACP,cAAc;IACd;IACA,aAAa;;EAEf;IAAW;IAAgC;;;EAE3C;;EAEA;;EAEA;;EAEA;;UAKe;EACf,aAAa;EACb,QAAQ;;UAGO;EACf,aAAa;EACb,OAAO;;UAGQ;;EAEf;;EAEA,UAAU;IAAQ;IAAe;;;;;UC1JlB;;EAEf;;EAEA;;EAEA;;EAEA,mBAAmB;;EAEnB;;EAEA;;UAGe;EACf,cAAc,OAAO,cAAc,0BAA0B,QAAQ;EACrE,eAAe,QAAQ,gBAAgB,0BAA0B,QAAQ;EACzE,aAAa,OAAO,kBAAkB,0BAA0B,QAAQ;WAC/D,QAAQ;WACR,aAAa;;iBAyGR,mBAAmB,QAAQ,eAAe;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAsE1C,oBACd,YAAW,QAAQ;EAAkB,MAAM;IAC1C;iBAiBa,oBACd,YAAW,QAAQ;EAAkB,MAAM;IAC1C"}
@@ -1,8 +1,8 @@
1
1
  import { t as DefaultVerdict } from "./verdict-DExhxfgR.js";
2
2
  import { c as CostLedgerHandle } from "./cost-ledger-Bv_e8XHY.js";
3
3
  import { A as ChatClient, Ft as RawProviderSink } from "./types-D216SgwM.js";
4
- import { c as AnalystRunInputs, i as AnalystFinding, l as AnalystRunResult, n as AnalystContext, s as AnalystRunEvent, t as Analyst, u as AnalystRunSummary } from "./types-DF_Udrp-.js";
5
- import { i as ExactAnalystRunEvent, o as ExactAnalystRunResult, u as ExactExecutionComponentIdentity } from "./exact-types-Djvzosly.js";
4
+ import { c as AnalystRunInputs, i as AnalystFinding, l as AnalystRunResult, n as AnalystContext, s as AnalystRunEvent, t as Analyst, u as AnalystRunSummary } from "./types-BvDKaULh.js";
5
+ import { i as ExactAnalystRunEvent, o as ExactAnalystRunResult, u as ExactExecutionComponentIdentity } from "./exact-types-DSFFpLLI.js";
6
6
  //#region src/analyst/registry.d.ts
7
7
  interface AnalystHooks {
8
8
  /** Legacy runs may mutate ctx; exact runs provide a frozen observational context. */
@@ -411,4 +411,4 @@ declare function createTokenRecallChecker(opts?: {
411
411
  }): CorrectnessChecker;
412
412
  //#endregion
413
413
  export { ExactAnalystRunExecutionError as A, jsonHasKeys as C, AnalystRegistryOptions as D, AnalystRegistry as E, RegistryRunOpts as M, assertExactRegistryRunOpts as N, BudgetPolicy as O, containsAll as S, AnalystHooks as T, ValidationContext as _, ProducedProposal as a, byteLengthRange as b, SatisfiedBy as c, createLlmCorrectnessChecker as d, createTokenRecallChecker as f, ArtifactValidator as g, Artifact as h, LlmCorrectnessCheckerOpts as i, ExactRegistryRunOpts as j, ExactAnalystBudgetPolicy as k, TaskGold as l, verifyCompletion as m, CompletionVerdict as n, ProducedState as o, parseCorrectnessResponse as p, CorrectnessChecker as r, RequirementCheck as s, CompletionRequirement as t, completionVerdict as u, ValidationIssue as v, regexMatch as w, composeValidators as x, ValidationResult as y };
414
- //# sourceMappingURL=completion-verifier-foUCLif_.d.ts.map
414
+ //# sourceMappingURL=completion-verifier-DJA5BhPb.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"completion-verifier-foUCLif_.d.ts","names":[],"sources":["../src/analyst/registry.ts","../src/artifact-validator.ts","../src/completion-verifier.ts"],"mappings":";;;;;;UAqDiB;;EAEf,iBAAiB;IACf,SAAS;IACT,KAAK;IACL;aACS;;EAEX,gBAAgB;IACd,SAAS;IACT,SAAS;IACT,UAAU;IACV;aACS;;;;;;EAMX,SAAS;IACP,SAAS;IACT,OAAO;IACP;MACE,+BAA+B,QAAQ;;EAE3C,YAAY;IAAQ,QAAQ;aAA4B;;UAGzC;;EAEf;;EAEA,UAAU;;;;;;;EAOV,YAAY;IACV,SAAS;IACT;IACA;IACA;;;UAIa;;EAEf,OAAO;;EAEP,OAAO,aAAa,SAAS;;EAE7B,QAAQ;;EAER,gBAAgB;;EAEhB,gBAAgB;;EAEhB,eAAe;;UAGA;;EAEf;;EAEA;;EAEA,SAAS;;EAET;;EAEA,SAAS;;EAET,aAAa;;EAEb;;EAEA,OAAO;;;;;;;;;EASP,gBAAgB,cAAc,kBAAkB,eAAe,cAAc;;;;;;EAM7E;;;KAIU;WAEG;WACA;;WAGA;WACA;;WAEA,SAAS,SAAS;;;;;;;;;;UAWhB;WACN;WACA,QAAQ;WACR;WACA,QAAQ;WACR,YAAY;WACZ,oBAAoB;WACpB;WACA,MAAM,SAAS;WACf,eACL,cAAc,kBACd,SAAS,eAAe,cAAc;WAEjC;WACA;WACA;WACA;;;cAqCE,sCAAsC;WACxC;WACA,QAAQ;EAEjB,YAAY,iBAAiB,QAAQ,uBAAuB,UAAU;;cAU3D;mBACM;mBACA;EAEjB,YAAY,UAAS;EAIrB,SAAS,SAAS;EAsBlB,QAAQ;IACN;IACA;IACA;IACA,MAAM;;EAUF,IACJ,eACA,QAAQ,kBACR,UAAS,kBACR,QAAQ;;EAUL,SACJ,eACA,QAAQ,kBACR,SAAS,uBACR,QAAQ;;EAQJ,eACL,eACA,QAAQ,kBACR,SAAS,uBACR,eAAe;;;;;;;;;;;;EAmBX,UACL,eACA,QAAQ,kBACR,UAAS,kBACR,eAAe;UAIV;UA8BA;UAwFO;UA2aP;UAaA;UAUA;;;iBAiGM,2BAA2B,yBAAyB,SAAS;;;;;;;;;;;;;;;;UC79B5D;;EAEf;;EAEA;;EAEA;;EAEA,QAAQ;;EAER,WAAW;;UAGI;EACf;EACA;;EAEA,iBAAiB;;EAEjB,QAAQ;;UAGO;EACf;EACA;;EAEA;;UAGe;EACf;;EAEA;EACA,QAAQ;;EAER,WAAW;;UAGI;;EAEf;;EAEA;;EAEA,SAAS,UAAU,UAAU,SAAS,oBAAoB,QAAQ;;;;;;iBAWpD,kBACd,YAAY,qBACZ;EAAY;EAAe;IAC1B;;iBAgCa,WAAW,cAAc,SAAS,SAAS;;iBAkB3C,YAAY,cAAc,0BAA0B;;iBAuCpD,gBAAgB,cAAc,aAAa,cAAc;;iBAoBzD,YACd,cACA,oBACA;EAAY;IACX;;;;KClJS;UAEK;;EAEf;;EAEA;;EAEA;;EAEA,cAAc;;UAGC;EACf;EACA,cAAc;;UAGC;EACf;EACA;EACA;;EAEA;;;UAIe;;EAEf,WAAW;;EAEX,WAAW;;EAEX;;UAGe;EACf;EACA;;EAEA;;;;;;EAMA;;EAEA;;;;;;;;;EASA;;EAEA;;EAEA;;;;;UAMe,0BAA0B;EACzC;EACA,cAAc;;EAEd;;EAEA;;EAEA;;;;;;;;;iBAUc,kBAAkB;EAChC;EACA,cAAc;IACZ;;;;;;KAoCQ,sBACV,aAAa,uBACb,oBACG;EAAU;EAAkB;;;;;;;;;iBAqLX,iBACpB,MAAM,UACN,OAAO,eACP,kBAAkB,qBACjB,QAAQ;UAoFM;EACf;;EAEA,aAAa;EACb;EACA,WAAW;EACX,SAAS;;EAET;;;;;;EAMA;;;;;;EAMA,UAAU;;;;;;;;;iBAUI,yBAAyB;EAAgB;EAAkB;;;;;;;;iBAiC3D,4BACd,MAAM,YACN,OAAM,4BACL;;;;;;;;;;;;;;;iBAyJa,yBACd;EAAQ;EAAoB;IAC3B"}
1
+ {"version":3,"file":"completion-verifier-DJA5BhPb.d.ts","names":[],"sources":["../src/analyst/registry.ts","../src/artifact-validator.ts","../src/completion-verifier.ts"],"mappings":";;;;;;UAqDiB;;EAEf,iBAAiB;IACf,SAAS;IACT,KAAK;IACL;aACS;;EAEX,gBAAgB;IACd,SAAS;IACT,SAAS;IACT,UAAU;IACV;aACS;;;;;;EAMX,SAAS;IACP,SAAS;IACT,OAAO;IACP;MACE,+BAA+B,QAAQ;;EAE3C,YAAY;IAAQ,QAAQ;aAA4B;;UAGzC;;EAEf;;EAEA,UAAU;;;;;;;EAOV,YAAY;IACV,SAAS;IACT;IACA;IACA;;;UAIa;;EAEf,OAAO;;EAEP,OAAO,aAAa,SAAS;;EAE7B,QAAQ;;EAER,gBAAgB;;EAEhB,gBAAgB;;EAEhB,eAAe;;UAGA;;EAEf;;EAEA;;EAEA,SAAS;;EAET;;EAEA,SAAS;;EAET,aAAa;;EAEb;;EAEA,OAAO;;;;;;;;;EASP,gBAAgB,cAAc,kBAAkB,eAAe,cAAc;;;;;;EAM7E;;;KAIU;WAEG;WACA;;WAGA;WACA;;WAEA,SAAS,SAAS;;;;;;;;;;UAWhB;WACN;WACA,QAAQ;WACR;WACA,QAAQ;WACR,YAAY;WACZ,oBAAoB;WACpB;WACA,MAAM,SAAS;WACf,eACL,cAAc,kBACd,SAAS,eAAe,cAAc;WAEjC;WACA;WACA;WACA;;;cAqCE,sCAAsC;WACxC;WACA,QAAQ;EAEjB,YAAY,iBAAiB,QAAQ,uBAAuB,UAAU;;cAU3D;mBACM;mBACA;EAEjB,YAAY,UAAS;EAIrB,SAAS,SAAS;EAsBlB,QAAQ;IACN;IACA;IACA;IACA,MAAM;;EAUF,IACJ,eACA,QAAQ,kBACR,UAAS,kBACR,QAAQ;;EAUL,SACJ,eACA,QAAQ,kBACR,SAAS,uBACR,QAAQ;;EAQJ,eACL,eACA,QAAQ,kBACR,SAAS,uBACR,eAAe;;;;;;;;;;;;EAmBX,UACL,eACA,QAAQ,kBACR,UAAS,kBACR,eAAe;UAIV;UA8BA;UAwFO;UA2aP;UAaA;UAUA;;;iBAiGM,2BAA2B,yBAAyB,SAAS;;;;;;;;;;;;;;;;UC79B5D;;EAEf;;EAEA;;EAEA;;EAEA,QAAQ;;EAER,WAAW;;UAGI;EACf;EACA;;EAEA,iBAAiB;;EAEjB,QAAQ;;UAGO;EACf;EACA;;EAEA;;UAGe;EACf;;EAEA;EACA,QAAQ;;EAER,WAAW;;UAGI;;EAEf;;EAEA;;EAEA,SAAS,UAAU,UAAU,SAAS,oBAAoB,QAAQ;;;;;;iBAWpD,kBACd,YAAY,qBACZ;EAAY;EAAe;IAC1B;;iBAgCa,WAAW,cAAc,SAAS,SAAS;;iBAkB3C,YAAY,cAAc,0BAA0B;;iBAuCpD,gBAAgB,cAAc,aAAa,cAAc;;iBAoBzD,YACd,cACA,oBACA;EAAY;IACX;;;;KClJS;UAEK;;EAEf;;EAEA;;EAEA;;EAEA,cAAc;;UAGC;EACf;EACA,cAAc;;UAGC;EACf;EACA;EACA;;EAEA;;;UAIe;;EAEf,WAAW;;EAEX,WAAW;;EAEX;;UAGe;EACf;EACA;;EAEA;;;;;;EAMA;;EAEA;;;;;;;;;EASA;;EAEA;;EAEA;;;;;UAMe,0BAA0B;EACzC;EACA,cAAc;;EAEd;;EAEA;;EAEA;;;;;;;;;iBAUc,kBAAkB;EAChC;EACA,cAAc;IACZ;;;;;;KAoCQ,sBACV,aAAa,uBACb,oBACG;EAAU;EAAkB;;;;;;;;;iBAqLX,iBACpB,MAAM,UACN,OAAO,eACP,kBAAkB,qBACjB,QAAQ;UAoFM;EACf;;EAEA,aAAa;EACb;EACA,WAAW;EACX,SAAS;;EAET;;;;;;EAMA;;;;;;EAMA,UAAU;;;;;;;;;iBAUI,yBAAyB;EAAgB;EAAkB;;;;;;;;iBAiC3D,4BACd,MAAM,YACN,OAAM,4BACL;;;;;;;;;;;;;;;iBAyJa,yBACd;EAAQ;EAAoB;IAC3B"}
@@ -1,14 +1,14 @@
1
1
  import { c as CostLedgerHandle, f as CostLedgerSummary, m as CostReceipt } from "../cost-ledger-Bv_e8XHY.js";
2
- import { a as RunRecord, n as RunCostProvenance, s as RunSplitTag } from "../run-record-DdSa93_W.js";
2
+ import { a as RunRecord, n as RunCostProvenance, s as RunSplitTag } from "../run-record-CF4Dwpxr.js";
3
3
  import { A as ChatClient, F as CreateChatClientOpts, V as createChatClient } from "../types-D216SgwM.js";
4
- import { _ as ProposalFinding, i as AnalystFinding, v as ProposalFindingOrigin, x as makeProposalFinding } from "../types-DF_Udrp-.js";
5
- import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-BZhStdjl.js";
6
- import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-DYuNHo9R.js";
7
- import { a as FailureClusterInsight, c as JudgeInsight, d as Recommendation, f as ReleaseSummary, i as FailureClassTally, l as LiftInsight, m as TokenUsageInsight, n as ExecutionErrorOutcomeCell, o as InsightReport, p as ScalarDistribution, r as ExecutionInsight, s as InterRaterInsight, t as CostProvenanceSummary, u as OutcomeCorrelationInsight } from "../insight-report-C6h6F_4L.js";
8
- import { D as analyzeRuns, E as SummarizeExecutionOptions, O as summarizeExecution, T as ExecutionReport, w as AnalyzeRunsOptions } from "../engine-nB64f48I.js";
9
- import { A as RunEvalOptions, At as compareOptimizationMethods, B as OptimizerModelBudget, C as RunImprovementLoopOptions, Ct as OptimizationMethodProvenance, D as RunOptimizationOptions, Dt as OptimizationPackageSource, E as PremeasuredOptimizationBaseline, F as GepaOptimizationMethodConfig, G as DefaultProductionRewardHackingOptions, Gt as inMemoryCampaignStorage, H as heldOutGate, Ht as CampaignStorage, I as GepaOptimizationRecipe, K as defaultProductionGate, L as GepaRunnerCommand, Lt as RunCampaignOptions, M as GepaAdaptiveEngineRun, N as GepaEngineOptions, Nt as CampaignCellFailureReceipt, Ot as OptimizationTokenUsage, P as GepaEngineRun, R as gepaOptimizationMethod, T as runImprovementLoop, U as DefaultProductionGateCheck, V as HeldOutGateOptions, W as DefaultProductionGateOptions, Wt as fsCampaignStorage, Yt as campaignSplitDigest, _t as CompareOptimizationMethodsOptions, an as ReferenceEquivalenceScenario, bt as OptimizationMethodComparison, cn as LlmJudgeDimension, dt as externalTextOptimizationMethod, en as REFERENCE_EQUIVALENCE_INPUT_LIMITS, ft as ExternalTextOptimizationMethodConfig, gt as ExternalTextEvaluationResponse, ht as ExternalOptimizationExample, i as skillOptOptimizationMethod, in as ReferenceEquivalenceJudgeResult, j as runEval, ln as LlmJudgeOptions, mt as ExternalTextOptimizerResult, n as SkillOptRunnerCommand, nn as ReferenceEquivalenceJudgeInput, on as createReferenceEquivalenceJudge, p as LoopProvenanceRecord, pt as ExternalTextOptimizerContext, r as SkillOptTrainerConfig, rn as ReferenceEquivalenceJudgeOptions, sn as runReferenceEquivalenceJudge, t as SkillOptOptimizationMethodConfig, tn as REFERENCE_EQUIVALENCE_JUDGE_VERSION, un as llmJudge, ut as composeGate, vt as ComparisonCost, w as RunImprovementLoopResult, wt as OptimizationMethodResult, xt as OptimizationMethodInput, yt as OptimizationMethod, z as OpenAICompatibleOptimizerModel, zt as runCampaign } from "../skillopt-optimization-method-BO7NAl3b.js";
10
- import { a as ObjectiveSource, b as PowerPreflight, c as PromotionPolicy, d as paretoSignificanceGate, i as EvidenceVector, l as buildEvidenceVector, n as AxisVerdict, o as ParetoSignificanceGateOptions, r as BuildEvidenceVectorOptions, s as PromotionObjective, t as AxisEvidence, u as paretoPolicy } from "../promotion-policy-ChWhTDBH.js";
11
- import { c as EvalRunGenerationSnapshot, g as TraceSpanEvent, n as HostedTenant, o as EvalRunCellScore, s as EvalRunEvent } from "../client-DjXROWpx.js";
4
+ import { _ as ProposalFinding, i as AnalystFinding, v as ProposalFindingOrigin, x as makeProposalFinding } from "../types-BvDKaULh.js";
5
+ import { n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-Bf8Woqmq.js";
6
+ import { C as JudgeDimension, H as SurfaceProposer, M as OptimizerConfig, R as Scenario, S as JudgeConfig, V as SessionScript, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, m as GateCheckStatus, n as CampaignArtifactWriter, p as Gate, r as CampaignCellResult, t as CampaignAggregates, v as GateResult, w as JudgeScore, y as GenerationCandidate } from "../types-BnjdJ70P.js";
7
+ import { a as FailureClusterInsight, c as JudgeInsight, d as Recommendation, f as ReleaseSummary, i as FailureClassTally, l as LiftInsight, m as TokenUsageInsight, n as ExecutionErrorOutcomeCell, o as InsightReport, p as ScalarDistribution, r as ExecutionInsight, s as InterRaterInsight, t as CostProvenanceSummary, u as OutcomeCorrelationInsight } from "../insight-report-CgX_s0Ez.js";
8
+ import { D as analyzeRuns, E as SummarizeExecutionOptions, O as summarizeExecution, T as ExecutionReport, w as AnalyzeRunsOptions } from "../engine-3hL-XqwJ.js";
9
+ import { A as RunEvalOptions, At as compareOptimizationMethods, B as OptimizerModelBudget, C as RunImprovementLoopOptions, Ct as OptimizationMethodProvenance, D as RunOptimizationOptions, Dt as OptimizationPackageSource, E as PremeasuredOptimizationBaseline, F as GepaOptimizationMethodConfig, G as DefaultProductionRewardHackingOptions, Gt as inMemoryCampaignStorage, H as heldOutGate, Ht as CampaignStorage, I as GepaOptimizationRecipe, K as defaultProductionGate, L as GepaRunnerCommand, Lt as RunCampaignOptions, M as GepaAdaptiveEngineRun, N as GepaEngineOptions, Nt as CampaignCellFailureReceipt, Ot as OptimizationTokenUsage, P as GepaEngineRun, R as gepaOptimizationMethod, T as runImprovementLoop, U as DefaultProductionGateCheck, V as HeldOutGateOptions, W as DefaultProductionGateOptions, Wt as fsCampaignStorage, Yt as campaignSplitDigest, _t as CompareOptimizationMethodsOptions, an as ReferenceEquivalenceScenario, bt as OptimizationMethodComparison, cn as LlmJudgeDimension, dt as externalTextOptimizationMethod, en as REFERENCE_EQUIVALENCE_INPUT_LIMITS, ft as ExternalTextOptimizationMethodConfig, gt as ExternalTextEvaluationResponse, ht as ExternalOptimizationExample, i as skillOptOptimizationMethod, in as ReferenceEquivalenceJudgeResult, j as runEval, ln as LlmJudgeOptions, mt as ExternalTextOptimizerResult, n as SkillOptRunnerCommand, nn as ReferenceEquivalenceJudgeInput, on as createReferenceEquivalenceJudge, p as LoopProvenanceRecord, pt as ExternalTextOptimizerContext, r as SkillOptTrainerConfig, rn as ReferenceEquivalenceJudgeOptions, sn as runReferenceEquivalenceJudge, t as SkillOptOptimizationMethodConfig, tn as REFERENCE_EQUIVALENCE_JUDGE_VERSION, un as llmJudge, ut as composeGate, vt as ComparisonCost, w as RunImprovementLoopResult, wt as OptimizationMethodResult, xt as OptimizationMethodInput, yt as OptimizationMethod, z as OpenAICompatibleOptimizerModel, zt as runCampaign } from "../skillopt-optimization-method-D8wysvU1.js";
10
+ import { a as ObjectiveSource, b as PowerPreflight, c as PromotionPolicy, d as paretoSignificanceGate, i as EvidenceVector, l as buildEvidenceVector, n as AxisVerdict, o as ParetoSignificanceGateOptions, r as BuildEvidenceVectorOptions, s as PromotionObjective, t as AxisEvidence, u as paretoPolicy } from "../promotion-policy-Ckjhzg_4.js";
11
+ import { c as EvalRunGenerationSnapshot, g as TraceSpanEvent, n as HostedTenant, o as EvalRunCellScore, s as EvalRunEvent } from "../client-C9gzZE59.js";
12
12
  import { i as InMemoryOutcomeStore, n as FileSystemOutcomeStore, o as OutcomeStore, r as FileSystemOutcomeStoreOptions, t as DeploymentOutcome } from "../outcome-store-BYHIuO0e.js";
13
13
  import { AgentCandidateBenchmarkCellRef, AgentCandidateBenchmarkSuiteInputs, AgentCandidateBenchmarkTask, AgentCandidateBenchmarkTaskMaterial, AgentCandidateBundle, AgentCandidateEvaluationPolicy, AgentCandidateExperiment, AgentCandidateExperimentMaterial, AgentCandidateExperimentMeasurement, AgentImprovementCost, AgentImprovementMeasuredComparison, AgentProfileImprovementExperiment, AgentProfileImprovementExperimentMaterial, AgentProfileImprovementMeasuredComparison, AgentProfileImprovementMeasurement, AgentProfileImprovementRunCell, AgentProfileImprovementRunReceipt, AgentProfileImprovementSuiteInputs, AgentProfileImprovementTask, AgentProfileImprovementTaskMaterial, CandidateExecutionEvidence, Sha256Digest } from "@tangle-network/agent-interface";
14
14
  //#region src/contract/self-improve.d.ts
@@ -7,13 +7,13 @@ import { nt as classifyOtlpSpanRole, rt as isOtlpModelCall } from "../kind-facto
7
7
  import { s as makeProposalFinding } from "../usage-receipt-EVI8B8Xu.js";
8
8
  import { i as inMemoryCampaignStorage, n as createRunCostLedger, r as fsCampaignStorage } from "../single-run-lock-BMQEv1wG.js";
9
9
  import { m as mapConcurrentRange } from "../ledger-core-DXZIqu17.js";
10
- import { $ as runReferenceEquivalenceJudge, H as resolveRunDir, M as surfaceContentHash, N as surfaceHash, Q as createReferenceEquivalenceJudge, S as assertOptimizationResult, V as runCampaign, X as REFERENCE_EQUIVALENCE_INPUT_LIMITS, Z as REFERENCE_EQUIVALENCE_JUDGE_VERSION, _ as heldOutGate, a as emitLoopProvenance, b as composeGate, ct as llmJudge, d as runImprovementLoop, g as gepaOptimizationMethod, o as loopProvenanceArgsFromResult, p as runEval, q as campaignSplitDigest, t as skillOptOptimizationMethod, v as defaultProductionGate, w as compareOptimizationMethods, x as externalTextOptimizationMethod } from "../skillopt-optimization-method-CkaI2ly4.js";
10
+ import { $ as runReferenceEquivalenceJudge, H as resolveRunDir, M as surfaceContentHash, N as surfaceHash, Q as createReferenceEquivalenceJudge, S as assertOptimizationResult, V as runCampaign, X as REFERENCE_EQUIVALENCE_INPUT_LIMITS, Z as REFERENCE_EQUIVALENCE_JUDGE_VERSION, _ as heldOutGate, a as emitLoopProvenance, b as composeGate, ct as llmJudge, d as runImprovementLoop, g as gepaOptimizationMethod, o as loopProvenanceArgsFromResult, p as runEval, q as campaignSplitDigest, t as skillOptOptimizationMethod, v as defaultProductionGate, w as compareOptimizationMethods, x as externalTextOptimizationMethod } from "../skillopt-optimization-method-Dtzp7QpP.js";
11
11
  import { E as pairedBootstrap } from "../statistics-ByxzSiOM.js";
12
- import { i as parseRunRecordSafe, r as modelHasSnapshot } from "../run-record-DqOw5X6_.js";
12
+ import { i as parseRunRecordSafe, r as modelHasSnapshot } from "../run-record-BmSPWXJR.js";
13
13
  import { c as heldoutSignificance, i as powerPreflight, n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector, u as decidePairedPromotion } from "../promotion-policy-CrLrmys8.js";
14
14
  import { a as recordAggregateMeasurements, i as readTaskFailureLabels, o as summarizeExecutionMeasurements, s as summarizeTraceErrors, t as extractUsage } from "../extract-usage-BW27f3XW.js";
15
- import { n as summarizeExecution, t as analyzeRuns } from "../analyze-runs-BNNK7irB.js";
16
- import { c as campaignCellTaskScore, l as campaignCellToRunRecord, o as campaignCellExecutionEvidence, s as campaignCellJudgeDimensions } from "../reward-hacking-B2tPL9a3.js";
15
+ import { n as summarizeExecution, t as analyzeRuns } from "../analyze-runs-C30yljDJ.js";
16
+ import { c as campaignCellTaskScore, l as campaignCellToRunRecord, o as campaignCellExecutionEvidence, s as campaignCellJudgeDimensions } from "../reward-hacking-BDToousL.js";
17
17
  import { n as InMemoryOutcomeStore, t as FileSystemOutcomeStore } from "../outcome-store-ChBKlTd_.js";
18
18
  import { t as createHostedClient } from "../client-LIuo-KPv.js";
19
19
  import { dirname, join } from "node:path";