@tangle-network/agent-eval 0.131.1 → 0.133.0

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 (133) hide show
  1. package/CHANGELOG.md +14 -0
  2. package/dist/analyst/index.d.ts +5 -5
  3. package/dist/analyst/index.js +3 -3
  4. package/dist/{analyze-runs-CBnCgfse.d.ts → analyze-runs-DZr7JW-m.d.ts} +5 -5
  5. package/dist/{analyze-runs-CBnCgfse.d.ts.map → analyze-runs-DZr7JW-m.d.ts.map} +1 -1
  6. package/dist/benchmarks/index.d.ts +1 -1
  7. package/dist/benchmarks/index.js +1 -1
  8. package/dist/{benchmarks-BDtRiOvE.js → benchmarks-DncrVrSr.js} +3 -3
  9. package/dist/{benchmarks-BDtRiOvE.js.map → benchmarks-DncrVrSr.js.map} +1 -1
  10. package/dist/campaign/index.d.ts +4 -4
  11. package/dist/campaign/index.js +2 -2
  12. package/dist/{campaign-CLG6y9jF.js → campaign-CyQMCCEU.js} +4 -4
  13. package/dist/{campaign-CLG6y9jF.js.map → campaign-CyQMCCEU.js.map} +1 -1
  14. package/dist/cli.js +1 -1
  15. package/dist/{client-w90OcvvR.d.ts → client-D4F9hdzR.d.ts} +4 -4
  16. package/dist/{client-w90OcvvR.d.ts.map → client-D4F9hdzR.d.ts.map} +1 -1
  17. package/dist/contract/index.d.ts +188 -22
  18. package/dist/contract/index.d.ts.map +1 -1
  19. package/dist/contract/index.js +1663 -65
  20. package/dist/contract/index.js.map +1 -1
  21. package/dist/control.d.ts +1 -1
  22. package/dist/{cost-ledger-DIgQUFZZ.js → cost-ledger-BrJxbrMy.js} +278 -9
  23. package/dist/cost-ledger-BrJxbrMy.js.map +1 -0
  24. package/dist/{cost-ledger-Dye6jCgg.d.ts → cost-ledger-fGS_u_O1.d.ts} +18 -2
  25. package/dist/cost-ledger-fGS_u_O1.d.ts.map +1 -0
  26. package/dist/{default-registry-Cj1oUpLN.d.ts → default-registry-Cl3pHo4n.d.ts} +4 -4
  27. package/dist/{default-registry-Cj1oUpLN.d.ts.map → default-registry-Cl3pHo4n.d.ts.map} +1 -1
  28. package/dist/{default-registry-C-vFCSEc.js → default-registry-D3T9XbuY.js} +4 -13
  29. package/dist/default-registry-D3T9XbuY.js.map +1 -0
  30. package/dist/{eval-campaign-C2k-m4aY.js → eval-campaign-CYkUME2T.js} +2 -2
  31. package/dist/{eval-campaign-C2k-m4aY.js.map → eval-campaign-CYkUME2T.js.map} +1 -1
  32. package/dist/{task-failure-attributes-CQZlB3et.js → extract-usage-2j25whHw.js} +154 -2
  33. package/dist/extract-usage-2j25whHw.js.map +1 -0
  34. package/dist/fuzz.d.ts +1 -1
  35. package/dist/fuzz.js +1 -1
  36. package/dist/hosted/index.d.ts +2 -2
  37. package/dist/{index-B11XCkdf2.d.ts → index-3cdlURSk.d.ts} +2 -2
  38. package/dist/index-3cdlURSk.d.ts.map +1 -0
  39. package/dist/{index-VTypFU3t.d.ts → index-B_eUhU9K.d.ts} +7 -7
  40. package/dist/{index-VTypFU3t.d.ts.map → index-B_eUhU9K.d.ts.map} +1 -1
  41. package/dist/{index-DNXoNaFF.d.ts → index-C2fkZhv_.d.ts} +2 -2
  42. package/dist/{index-DNXoNaFF.d.ts.map → index-C2fkZhv_.d.ts.map} +1 -1
  43. package/dist/{index-NPeSWD98.d.ts → index-CiP2DRCj.d.ts} +5 -5
  44. package/dist/{index-NPeSWD98.d.ts.map → index-CiP2DRCj.d.ts.map} +1 -1
  45. package/dist/index.d.ts +69 -23
  46. package/dist/index.d.ts.map +1 -1
  47. package/dist/index.js +100 -12
  48. package/dist/index.js.map +1 -1
  49. package/dist/{llm-client-B_nIBlYo.d.ts → llm-client-BiK4HW0u.d.ts} +2 -2
  50. package/dist/{llm-client-B_nIBlYo.d.ts.map → llm-client-BiK4HW0u.d.ts.map} +1 -1
  51. package/dist/{llm-client--GR4JbZE.js → llm-client-ClPW-dWB.js} +2 -2
  52. package/dist/{llm-client--GR4JbZE.js.map → llm-client-ClPW-dWB.js.map} +1 -1
  53. package/dist/meta-eval/index.d.ts +216 -3
  54. package/dist/{index-D_F6VAKe.d.ts.map → meta-eval/index.d.ts.map} +1 -1
  55. package/dist/meta-eval/index.js +91 -1
  56. package/dist/meta-eval/index.js.map +1 -1
  57. package/dist/multishot/index.d.ts +1 -1
  58. package/dist/openapi.json +1 -1
  59. package/dist/{release-report-Crg9oFJ0.d.ts → release-report-DfmKSIEE.d.ts} +3 -3
  60. package/dist/{release-report-Crg9oFJ0.d.ts.map → release-report-DfmKSIEE.d.ts.map} +1 -1
  61. package/dist/{replay-RE97Ckjl.d.ts → replay-BI6CVKkp.d.ts} +2 -2
  62. package/dist/{replay-RE97Ckjl.d.ts.map → replay-BI6CVKkp.d.ts.map} +1 -1
  63. package/dist/{replay-D18-pBAA.js → replay-BMR4TEYY.js} +2 -3
  64. package/dist/{replay-D18-pBAA.js.map → replay-BMR4TEYY.js.map} +1 -1
  65. package/dist/reporting.d.ts +4 -4
  66. package/dist/{researcher-Q5rpPqZY.d.ts → researcher-DMimgHtN.d.ts} +4 -4
  67. package/dist/{researcher-Q5rpPqZY.d.ts.map → researcher-DMimgHtN.d.ts.map} +1 -1
  68. package/dist/{reward-hacking-CW-3HN0n.d.ts → reward-hacking-D-QqXvg-.d.ts} +2 -2
  69. package/dist/{reward-hacking-CW-3HN0n.d.ts.map → reward-hacking-D-QqXvg-.d.ts.map} +1 -1
  70. package/dist/rl.d.ts +156 -6
  71. package/dist/rl.d.ts.map +1 -1
  72. package/dist/rl.js +217 -2
  73. package/dist/rl.js.map +1 -1
  74. package/dist/rollout/index.d.ts +1 -1
  75. package/dist/{rubric-predictive-validity-lXLmashy.d.ts → rubric-predictive-validity-C1dCLcvb.d.ts} +2 -2
  76. package/dist/{rubric-predictive-validity-lXLmashy.d.ts.map → rubric-predictive-validity-C1dCLcvb.d.ts.map} +1 -1
  77. package/dist/{run-evidence-oByzm-dE.d.ts → run-evidence-DokQtX0-.d.ts} +2 -2
  78. package/dist/{run-evidence-oByzm-dE.d.ts.map → run-evidence-DokQtX0-.d.ts.map} +1 -1
  79. package/dist/run-record-CN8Zd21B.js.map +1 -1
  80. package/dist/{run-record-BJnYdTxO.d.ts → run-record-DcObtIGh.d.ts} +4 -14
  81. package/dist/run-record-DcObtIGh.d.ts.map +1 -0
  82. package/dist/{semantic-concept-judge-b5m3irbR.js → semantic-concept-judge-BypLt6Fw.js} +4 -4
  83. package/dist/{semantic-concept-judge-b5m3irbR.js.map → semantic-concept-judge-BypLt6Fw.js.map} +1 -1
  84. package/dist/{server-m5D9cvnG.js → server-BPqlDBWK.js} +3 -3
  85. package/dist/{server-m5D9cvnG.js.map → server-BPqlDBWK.js.map} +1 -1
  86. package/dist/{skill-usage-D5mlWdAJ.d.ts → skill-usage-BaaxFSJR.d.ts} +5 -5
  87. package/dist/{skill-usage-D5mlWdAJ.d.ts.map → skill-usage-BaaxFSJR.d.ts.map} +1 -1
  88. package/dist/{skillopt-optimization-method-D0MVjJdP.d.ts → skillopt-optimization-method-CAASpcS3.d.ts} +9 -9
  89. package/dist/{skillopt-optimization-method-D0MVjJdP.d.ts.map → skillopt-optimization-method-CAASpcS3.d.ts.map} +1 -1
  90. package/dist/{skillopt-optimization-method-eOJL2570.js → skillopt-optimization-method-DgN7U9iR.js} +3 -3
  91. package/dist/{skillopt-optimization-method-eOJL2570.js.map → skillopt-optimization-method-DgN7U9iR.js.map} +1 -1
  92. package/dist/{statistics-Cmj6nynr.d.ts → statistics-DbvkkDPa.d.ts} +2 -2
  93. package/dist/{statistics-Cmj6nynr.d.ts.map → statistics-DbvkkDPa.d.ts.map} +1 -1
  94. package/dist/{summary-report-CWwB_LiV.d.ts → summary-report-DnUcjVpV.d.ts} +2 -2
  95. package/dist/{summary-report-CWwB_LiV.d.ts.map → summary-report-DnUcjVpV.d.ts.map} +1 -1
  96. package/dist/traces.d.ts +2 -2
  97. package/dist/traces.js +2 -2
  98. package/dist/{types-CsD5nTfV.d.ts → types-BokuXvOG.d.ts} +4 -4
  99. package/dist/{types-CsD5nTfV.d.ts.map → types-BokuXvOG.d.ts.map} +1 -1
  100. package/dist/{types-DGsxbAEd.d.ts → types-Cc3qbqzj.d.ts} +3 -3
  101. package/dist/{types-DGsxbAEd.d.ts.map → types-Cc3qbqzj.d.ts.map} +1 -1
  102. package/dist/wire/index.d.ts +2 -2
  103. package/dist/wire/index.js +1 -1
  104. package/package.json +3 -8
  105. package/dist/belief-state/index.d.ts +0 -622
  106. package/dist/belief-state/index.d.ts.map +0 -1
  107. package/dist/belief-state/index.js +0 -1819
  108. package/dist/belief-state/index.js.map +0 -1
  109. package/dist/calibration-CNWWA6K8.js +0 -94
  110. package/dist/calibration-CNWWA6K8.js.map +0 -1
  111. package/dist/code-agent-session-BjkMTQ7H.js +0 -1390
  112. package/dist/code-agent-session-BjkMTQ7H.js.map +0 -1
  113. package/dist/code-agent-session-aYa3SGKz.d.ts +0 -143
  114. package/dist/code-agent-session-aYa3SGKz.d.ts.map +0 -1
  115. package/dist/cost-ledger-DIgQUFZZ.js.map +0 -1
  116. package/dist/cost-ledger-Dye6jCgg.d.ts.map +0 -1
  117. package/dist/default-registry-C-vFCSEc.js.map +0 -1
  118. package/dist/extract-usage-BrQ8mCLX.js +0 -155
  119. package/dist/extract-usage-BrQ8mCLX.js.map +0 -1
  120. package/dist/index-B11XCkdf2.d.ts.map +0 -1
  121. package/dist/index-D_F6VAKe.d.ts +0 -217
  122. package/dist/metrics-C9YY1OcL.js +0 -239
  123. package/dist/metrics-C9YY1OcL.js.map +0 -1
  124. package/dist/off-policy-DvgzvtIx.js +0 -220
  125. package/dist/off-policy-DvgzvtIx.js.map +0 -1
  126. package/dist/off-policy-mskQw8Mb.d.ts +0 -153
  127. package/dist/off-policy-mskQw8Mb.d.ts.map +0 -1
  128. package/dist/run-record-BJnYdTxO.d.ts.map +0 -1
  129. package/dist/runtime-trajectory-1gyaTOoC.js +0 -93
  130. package/dist/runtime-trajectory-1gyaTOoC.js.map +0 -1
  131. package/dist/runtime-trajectory-BXxG4lyi.d.ts +0 -50
  132. package/dist/runtime-trajectory-BXxG4lyi.d.ts.map +0 -1
  133. package/dist/task-failure-attributes-CQZlB3et.js.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-DBJXXoY1.js";
3
- import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-m5D9cvnG.js";
3
+ import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-BPqlDBWK.js";
4
4
  import { writeFileSync } from "node:fs";
5
5
  //#region src/cli-config.ts
6
6
  function resolveCliLlmConfig(env = process.env) {
@@ -1,8 +1,8 @@
1
1
  import { o as FailureClass } from "./schema-BtVldJ3T.js";
2
- import { l as RunTerminalOutcome } from "./run-record-BJnYdTxO.js";
2
+ import { l as RunTerminalOutcome } from "./run-record-DcObtIGh.js";
3
+ import { _ as GateDecision, j as MutableSurface } from "./types-BokuXvOG.js";
3
4
  import { r as ContinuousAgreement } from "./judge-calibration-DFtEMlde.js";
4
- import { _ as GateDecision, j as MutableSurface } from "./types-CsD5nTfV.js";
5
- import { i as ParetoFigureSpec, t as GainDistributionBin } from "./summary-report-CWwB_LiV.js";
5
+ import { i as ParetoFigureSpec, t as GainDistributionBin } from "./summary-report-DnUcjVpV.js";
6
6
  //#region src/contract/insight-report.d.ts
7
7
  interface InsightReport {
8
8
  /** Number of runs analyzed. */
@@ -578,4 +578,4 @@ declare function hostedClientFromEnv(overrides?: Partial<HostedTenant> & {
578
578
  }): HostedClient | undefined;
579
579
  //#endregion
580
580
  export { ScalarDistribution as A, InsightReport as C, OutcomeCorrelationInsight as D, LiftInsight as E, Recommendation as O, FailureClusterInsight as S, JudgeInsight as T, UnixNanoTimestamp as _, hostedTenantFromEnv as a, ExecutionInsight as b, EvalRunGenerationSnapshot as c, HostedIngestHeaders as d, HostedWireVersion as f, TraceSpanEvent as g, IngestTracesRequest as h, hostedClientFromEnv as i, TokenUsageInsight as j, ReleaseSummary as k, 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, CostProvenanceSummary as v, InterRaterInsight as w, FailureClassTally as x, ExecutionErrorOutcomeCell as y };
581
- //# sourceMappingURL=client-w90OcvvR.d.ts.map
581
+ //# sourceMappingURL=client-D4F9hdzR.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"client-w90OcvvR.d.ts","names":[],"sources":["../src/contract/insight-report.ts","../src/hosted/types.ts","../src/hosted/client.ts"],"mappings":";;;;;;UAqCiB;;EAEf;;;;EAKA,WAAW;;EAGX,WAAW;;;EAIX,cAAc,eAAe;;EAG7B;IACE,MAAM;IACN,QAAQ;;;;IAIR,aAAa;;;;;IAKb;MAAa;MAAe;;;;;;EAM9B,QAAQ,eAAe;;;;EAKvB,aAAa;;;;EAKb,OAAO;;;EAIP,kBAAkB;;;EAIlB,gBAAgB;;;;;EAMhB,qBAAqB;;;;;EAMrB,SAAS;;;;;;EAOT,wBAAwB;;;;EAKxB,iBAAiB;;;EAIjB,iBAAiB;;UAGF;EACf;IAAY;IAAW;;EACvB;IAAa;IAAW;;EACxB;IAAc;;EACd;;UAGe;;EAEf,YAAY;;EAEZ,SAAS;;;EAGT,YAAY;;;EAGZ;IACE;IACA,YAAY;IACZ,SAAS;IACT;;;EAGF,QAAQ;IAAQ;IAAe;;;;;EAI/B;IACE;IACA;IACA;;;;;EAKF;IACE;;;IAGA;;IAEA;;IAEA;;IAEA;;IAEA;;;;;IAKA,mBAAmB,OAAO,oBAAoB;;;;EAIhD;IACE;IACA;IACA;IACA;IACA;;;UAIa;;EAEf;;EAEA;;EAEA;;UAGe;EACf,OAAO;EACP,QAAQ;EACR,WAAW;EACX,QAAQ;EACR,YAAY;EACZ;IACE;IACA;IACA;IACA;IACA;;;;UAOa;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,WAAW;;;;;EAKX,WAAW;IAAQ;IAAe;;;UAGnB;;EAEf;;EAEA;;;EAGA,cAAc;;;EAGd;;;EAGA;;;EAGA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,SAAS;;EAET,mBAAmB;IACjB;IACA,SAAS;MAAQ;MAAe;;IAChC;;;UAIa;EACf;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;;EAEf,UAAU;IACR;IACA;;IAEA;;IAEA;;IAEA;;EAEF;;;;;UAMe;;EAEf,cAAc;;EAEd;;EAEA;;UAGe;;EAEf;;;EAGA;EACA,UAAU;IAAQ;IAAe;IAAgB;;;UAGlC;;;EAGf;;EAEA;;EAEA;;EAEA;;EAEA;IACE;IACA;IACA;;;UAIa;;;EAGf;EACA,MAAM;IACJ;IACA;IACA;;;;EAIF;;UAGe;;EAEf;;EAEA;;;;EAIA;;;EAGA;;EAEA;;EAEA;;EAEA;EACA;;;;EAIA;;UAGe;;EAEf;EACA;;EAEA;;;;EAIA,SAAS,eAAe;;;EAGxB;;EAEA;;UAGe;EACf;EACA;EACA;EACA;;EAEA;;;;cClYW;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-D4F9hdzR.d.ts","names":[],"sources":["../src/contract/insight-report.ts","../src/hosted/types.ts","../src/hosted/client.ts"],"mappings":";;;;;;UAqCiB;;EAEf;;;;EAKA,WAAW;;EAGX,WAAW;;;EAIX,cAAc,eAAe;;EAG7B;IACE,MAAM;IACN,QAAQ;;;;IAIR,aAAa;;;;;IAKb;MAAa;MAAe;;;;;;EAM9B,QAAQ,eAAe;;;;EAKvB,aAAa;;;;EAKb,OAAO;;;EAIP,kBAAkB;;;EAIlB,gBAAgB;;;;;EAMhB,qBAAqB;;;;;EAMrB,SAAS;;;;;;EAOT,wBAAwB;;;;EAKxB,iBAAiB;;;EAIjB,iBAAiB;;UAGF;EACf;IAAY;IAAW;;EACvB;IAAa;IAAW;;EACxB;IAAc;;EACd;;UAGe;;EAEf,YAAY;;EAEZ,SAAS;;;EAGT,YAAY;;;EAGZ;IACE;IACA,YAAY;IACZ,SAAS;IACT;;;EAGF,QAAQ;IAAQ;IAAe;;;;;EAI/B;IACE;IACA;IACA;;;;;EAKF;IACE;;;IAGA;;IAEA;;IAEA;;IAEA;;IAEA;;;;;IAKA,mBAAmB,OAAO,oBAAoB;;;;EAIhD;IACE;IACA;IACA;IACA;IACA;;;UAIa;;EAEf;;EAEA;;EAEA;;UAGe;EACf,OAAO;EACP,QAAQ;EACR,WAAW;EACX,QAAQ;EACR,YAAY;EACZ;IACE;IACA;IACA;IACA;IACA;;;;UAOa;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,WAAW;;;;;EAKX,WAAW;IAAQ;IAAe;;;UAGnB;;EAEf;;EAEA;;;EAGA,cAAc;;;EAGd;;;EAGA;;;EAGA;;UAGe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA,SAAS;;EAET,mBAAmB;IACjB;IACA,SAAS;MAAQ;MAAe;;IAChC;;;UAIa;EACf;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;UAGe;;EAEf,UAAU;IACR;IACA;;IAEA;;IAEA;;IAEA;;EAEF;;;;;UAMe;;EAEf,cAAc;;EAEd;;EAEA;;UAGe;;EAEf;;;EAGA;EACA,UAAU;IAAQ;IAAe;IAAgB;;;UAGlC;;;EAGf;;EAEA;;EAEA;;EAEA;;EAEA;IACE;IACA;IACA;;;UAIa;;;EAGf;EACA,MAAM;IACJ;IACA;IACA;;;;EAIF;;UAGe;;EAEf;;EAEA;;;;EAIA;;;EAGA;;EAEA;;EAEA;;EAEA;EACA;;;;EAIA;;UAGe;;EAEf;EACA;;EAEA;;;;EAIA,SAAS,eAAe;;;EAGxB;;EAEA;;UAGe;EACf;EACA;EACA;EACA;;EAEA;;;;cClYW;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"}
@@ -1,14 +1,13 @@
1
- import { a as RunRecord, s as RunSplitTag } from "../run-record-BJnYdTxO.js";
2
- import { c as CostLedgerHandle, f as CostLedgerSummary, p as CostReceipt } from "../cost-ledger-Dye6jCgg.js";
3
- import { A as ChatClient, F as CreateChatClientOpts, V as createChatClient } from "../types-DGsxbAEd.js";
4
- import { T as AnalystFinding, n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-Cj1oUpLN.js";
1
+ import { c as CostLedgerHandle, f as CostLedgerSummary, m as CostReceipt } from "../cost-ledger-fGS_u_O1.js";
2
+ import { a as RunRecord, n as RunCostProvenance, s as RunSplitTag } from "../run-record-DcObtIGh.js";
3
+ import { A as ChatClient, F as CreateChatClientOpts, V as createChatClient } from "../types-Cc3qbqzj.js";
4
+ import { T as AnalystFinding, n as buildDefaultAnalystRegistry, t as DefaultAnalystRegistryOptions } from "../default-registry-Cl3pHo4n.js";
5
+ 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-BokuXvOG.js";
6
+ import { A as RunEvalOptions, B as OptimizerModelBudget, Bt as ComparisonCost, C as RunImprovementLoopOptions, Cn as ReferenceEquivalenceJudgeResult, D as RunOptimizationOptions, Dn as LlmJudgeDimension, E as PremeasuredOptimizationBaseline, En as runReferenceEquivalenceJudge, F as GepaOptimizationMethodConfig, Ft as ExternalTextOptimizerContext, G as ObjectiveSource, Gt as OptimizationMethodProvenance, H as AxisVerdict, Ht as OptimizationMethodComparison, I as GepaOptimizationRecipe, It as ExternalTextOptimizerResult, J as PromotionPolicy, K as ParetoSignificanceGateOptions, Kt as OptimizationMethodResult, L as GepaRunnerCommand, Lt as ExternalOptimizationExample, M as GepaAdaptiveEngineRun, Mt as composeGate, N as GepaEngineOptions, Nt as externalTextOptimizationMethod, On as LlmJudgeOptions, P as GepaEngineRun, Pt as ExternalTextOptimizationMethodConfig, R as gepaOptimizationMethod, Rt as ExternalTextEvaluationResponse, Sn as ReferenceEquivalenceJudgeOptions, T as runImprovementLoop, Tn as createReferenceEquivalenceJudge, U as BuildEvidenceVectorOptions, Ut as OptimizationMethodInput, V as AxisEvidence, Vt as OptimizationMethod, W as EvidenceVector, X as paretoPolicy, Xt as OptimizationTokenUsage, Y as buildEvidenceVector, Yt as OptimizationPackageSource, Z as paretoSignificanceGate, Zt as compareOptimizationMethods, bn as REFERENCE_EQUIVALENCE_JUDGE_VERSION, dt as DefaultProductionGateCheck, en as CampaignCellFailureReceipt, ft as DefaultProductionGateOptions, i as skillOptOptimizationMethod, in as RunCampaignOptions, j as runEval, kn as llmJudge, ln as fsCampaignStorage, lt as HeldOutGateOptions, mn as campaignSplitDigest, mt as defaultProductionGate, n as SkillOptRunnerCommand, on as runCampaign, ot as PowerPreflight, p as LoopProvenanceRecord, pt as DefaultProductionRewardHackingOptions, q as PromotionObjective, r as SkillOptTrainerConfig, sn as CampaignStorage, t as SkillOptOptimizationMethodConfig, un as inMemoryCampaignStorage, ut as heldOutGate, w as RunImprovementLoopResult, wn as ReferenceEquivalenceScenario, xn as ReferenceEquivalenceJudgeInput, yn as REFERENCE_EQUIVALENCE_INPUT_LIMITS, z as OpenAICompatibleOptimizerModel, zt as CompareOptimizationMethodsOptions } from "../skillopt-optimization-method-CAASpcS3.js";
7
+ import { A as ScalarDistribution, C as InsightReport, D as OutcomeCorrelationInsight, E as LiftInsight, O as Recommendation, S as FailureClusterInsight, T as JudgeInsight, b as ExecutionInsight, c as EvalRunGenerationSnapshot, g as TraceSpanEvent, j as TokenUsageInsight, k as ReleaseSummary, n as HostedTenant, o as EvalRunCellScore, s as EvalRunEvent, v as CostProvenanceSummary, w as InterRaterInsight, x as FailureClassTally, y as ExecutionErrorOutcomeCell } from "../client-D4F9hdzR.js";
5
8
  import { i as InMemoryOutcomeStore, n as FileSystemOutcomeStore, o as OutcomeStore, r as FileSystemOutcomeStoreOptions, t as DeploymentOutcome } from "../outcome-store-BYHIuO0e.js";
6
- import { _ as CodeAgentSessionExecutionReceipt, a as ParsedCodeAgentJsonl, b as CodeAgentSessionTerminalStatus, c as fromKimiCodeSession, d as fromPigraphSession, f as parseCodeAgentJsonl, g as CodeAgentSessionActionSurface, h as CodeAgentSessionActionStatus, i as CodeAgentSessionMetrics, l as fromOpenCodeSession, m as CodeAgentSessionActionKind, n as CodeAgentSessionIntakeOptions, o as fromClaudeCodeSession, p as CodeAgentSessionAction, r as CodeAgentSessionIntakeResult, s as fromCodexSession, t as CodeAgentSessionDiagnostic, u as fromPiSession, v as CodeAgentSessionObservation, x as observeCodeAgentSession, y as CodeAgentSessionSource } from "../code-agent-session-aYa3SGKz.js";
7
- 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-CsD5nTfV.js";
8
- import { A as RunEvalOptions, B as OptimizerModelBudget, Bt as ComparisonCost, C as RunImprovementLoopOptions, Cn as ReferenceEquivalenceJudgeResult, D as RunOptimizationOptions, Dn as LlmJudgeDimension, E as PremeasuredOptimizationBaseline, En as runReferenceEquivalenceJudge, F as GepaOptimizationMethodConfig, Ft as ExternalTextOptimizerContext, G as ObjectiveSource, Gt as OptimizationMethodProvenance, H as AxisVerdict, Ht as OptimizationMethodComparison, I as GepaOptimizationRecipe, It as ExternalTextOptimizerResult, J as PromotionPolicy, K as ParetoSignificanceGateOptions, Kt as OptimizationMethodResult, L as GepaRunnerCommand, Lt as ExternalOptimizationExample, M as GepaAdaptiveEngineRun, Mt as composeGate, N as GepaEngineOptions, Nt as externalTextOptimizationMethod, On as LlmJudgeOptions, P as GepaEngineRun, Pt as ExternalTextOptimizationMethodConfig, R as gepaOptimizationMethod, Rt as ExternalTextEvaluationResponse, Sn as ReferenceEquivalenceJudgeOptions, T as runImprovementLoop, Tn as createReferenceEquivalenceJudge, U as BuildEvidenceVectorOptions, Ut as OptimizationMethodInput, V as AxisEvidence, Vt as OptimizationMethod, W as EvidenceVector, X as paretoPolicy, Xt as OptimizationTokenUsage, Y as buildEvidenceVector, Yt as OptimizationPackageSource, Z as paretoSignificanceGate, Zt as compareOptimizationMethods, bn as REFERENCE_EQUIVALENCE_JUDGE_VERSION, dt as DefaultProductionGateCheck, en as CampaignCellFailureReceipt, ft as DefaultProductionGateOptions, i as skillOptOptimizationMethod, in as RunCampaignOptions, j as runEval, kn as llmJudge, ln as fsCampaignStorage, lt as HeldOutGateOptions, mn as campaignSplitDigest, mt as defaultProductionGate, n as SkillOptRunnerCommand, on as runCampaign, ot as PowerPreflight, p as LoopProvenanceRecord, pt as DefaultProductionRewardHackingOptions, q as PromotionObjective, r as SkillOptTrainerConfig, sn as CampaignStorage, t as SkillOptOptimizationMethodConfig, un as inMemoryCampaignStorage, ut as heldOutGate, w as RunImprovementLoopResult, wn as ReferenceEquivalenceScenario, xn as ReferenceEquivalenceJudgeInput, yn as REFERENCE_EQUIVALENCE_INPUT_LIMITS, z as OpenAICompatibleOptimizerModel, zt as CompareOptimizationMethodsOptions } from "../skillopt-optimization-method-D0MVjJdP.js";
9
- import { A as ScalarDistribution, C as InsightReport, D as OutcomeCorrelationInsight, E as LiftInsight, O as Recommendation, S as FailureClusterInsight, T as JudgeInsight, b as ExecutionInsight, c as EvalRunGenerationSnapshot, g as TraceSpanEvent, j as TokenUsageInsight, k as ReleaseSummary, n as HostedTenant, o as EvalRunCellScore, s as EvalRunEvent, v as CostProvenanceSummary, w as InterRaterInsight, x as FailureClassTally, y as ExecutionErrorOutcomeCell } from "../client-w90OcvvR.js";
10
- import { a as summarizeExecution, i as analyzeRuns, n as ExecutionReport, r as SummarizeExecutionOptions, t as AnalyzeRunsOptions } from "../analyze-runs-CBnCgfse.js";
11
- import { AgentCandidateBenchmarkCellRef, AgentCandidateBenchmarkSuiteInputs, AgentCandidateBenchmarkTask, AgentCandidateBenchmarkTaskMaterial, AgentCandidateBundle, AgentCandidateEvaluationPolicy, AgentCandidateExperiment, AgentCandidateExperimentMaterial, AgentCandidateExperimentMeasurement, AgentImprovementMeasuredComparison, AgentProfileImprovementExperiment, AgentProfileImprovementExperimentMaterial, AgentProfileImprovementMeasuredComparison, AgentProfileImprovementMeasurement, AgentProfileImprovementRunCell, AgentProfileImprovementRunReceipt, AgentProfileImprovementSuiteInputs, AgentProfileImprovementTask, AgentProfileImprovementTaskMaterial, CandidateExecutionEvidence, Sha256Digest } from "@tangle-network/agent-interface";
9
+ import { a as summarizeExecution, i as analyzeRuns, n as ExecutionReport, r as SummarizeExecutionOptions, t as AnalyzeRunsOptions } from "../analyze-runs-DZr7JW-m.js";
10
+ 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";
12
11
  //#region src/contract/self-improve.d.ts
13
12
  interface SelfImproveBudget {
14
13
  /** Hard spend cap across the full run. Each paid call reserves its enforced
@@ -408,16 +407,28 @@ interface RunCandidateExperimentOptions {
408
407
  execute(input: CandidateExperimentExecutionInput): Promise<CandidateExecutionEvidence>;
409
408
  /** Maximum number of simultaneous execute calls across both arms. */
410
409
  maxConcurrency?: number;
410
+ /** Shared run budget that also accounts for analysis and candidate search. */
411
+ costLedger?: CostLedgerHandle;
411
412
  signal?: AbortSignal;
412
413
  }
414
+ interface CandidateExperimentRun {
415
+ measurements: AgentCandidateExperimentMeasurement[];
416
+ measurement: {
417
+ wallDurationMs: number;
418
+ cost: AgentImprovementCost;
419
+ };
420
+ }
413
421
  interface CompareCandidateExperimentOptions {
414
422
  experiment: AgentCandidateExperiment;
415
423
  measurements: AgentCandidateExperimentMeasurement[];
424
+ preparation: {
425
+ wallDurationMs: number;
426
+ cost: AgentImprovementCost;
427
+ };
428
+ measurement: CandidateExperimentRun['measurement'];
416
429
  runId: string;
417
430
  candidate?: AgentImprovementMeasuredComparison['candidate'];
418
431
  generationsExplored?: number;
419
- searchDurationMs?: number;
420
- searchCostUsd?: number;
421
432
  metadata?: AgentImprovementMeasuredComparison['metadata'];
422
433
  }
423
434
  /** One exact baseline/candidate observation of the same held-out cell. */
@@ -434,6 +445,7 @@ interface PairedMeasurementAdapter<TRun> {
434
445
  score: number;
435
446
  }[];
436
447
  costUsd(run: TRun): number;
448
+ costProvenance(run: TRun): AgentImprovementCost['provenance'];
437
449
  latencyMs(run: TRun): number;
438
450
  completed(run: TRun): boolean;
439
451
  passed(run: TRun): boolean;
@@ -444,14 +456,16 @@ interface EvaluatePairedMeasurementsOptions<TRun> {
444
456
  adapter: PairedMeasurementAdapter<TRun>;
445
457
  /** Whether both arms use the same scorer family as the promotion decision. */
446
458
  sharedScorerChannel: boolean;
447
- /** Search or preparation spend that belongs to the same frozen budget. */
448
- additionalCostUsd?: number;
459
+ /** Analysis and candidate-search spend that belongs to the same frozen budget. */
460
+ preparationCost?: AgentImprovementCost;
461
+ /** Settled aggregate receipt for this exact paired suite, when an executor provides one. */
462
+ measurementCost?: AgentImprovementCost;
449
463
  }
450
464
  /** Statistical and operational result derived from complete paired receipts. */
451
465
  type PairedMeasurementEvaluation = Pick<AgentImprovementMeasuredComparison, 'overall' | 'objectives' | 'decision' | 'power'> & {
452
- executionCostUsd: number;
453
- totalCostUsd: number;
454
- executionDurationMs: number;
466
+ measurementCost: AgentImprovementCost;
467
+ totalCost: AgentImprovementCost;
468
+ measurementWorkDurationMs: number;
455
469
  };
456
470
  /** Content-address one task before any measured execution can see it. */
457
471
  declare function sealCandidateBenchmarkTask(material: AgentCandidateBenchmarkTaskMaterial): AgentCandidateBenchmarkTask;
@@ -461,7 +475,7 @@ declare function sealCandidateBenchmarkSuite(options: SealCandidateBenchmarkSuit
461
475
  declare function sealCandidateExperiment(material: AgentCandidateExperimentMaterial): AgentCandidateExperiment;
462
476
  declare function verifyCandidateExperiment(input: unknown): AgentCandidateExperiment;
463
477
  /** Execute each signed cell for both arms. The callback is Runtime's one executor. */
464
- declare function runCandidateExperiment(options: RunCandidateExperimentOptions): Promise<AgentCandidateExperimentMeasurement[]>;
478
+ declare function runCandidateExperiment(options: RunCandidateExperimentOptions): Promise<CandidateExperimentRun>;
465
479
  /**
466
480
  * Calculate the shared paired decision from any complete receipt shape.
467
481
  *
@@ -506,16 +520,28 @@ interface RunAgentProfileImprovementExperimentOptions {
506
520
  execute(input: AgentProfileImprovementExperimentExecutionInput): Promise<AgentProfileImprovementRunReceipt>;
507
521
  /** Maximum number of simultaneous execute calls across both arms. */
508
522
  maxConcurrency?: number;
523
+ /** Shared run budget that also accounts for analysis and candidate search. */
524
+ costLedger?: CostLedgerHandle;
509
525
  signal?: AbortSignal;
510
526
  }
527
+ interface AgentProfileImprovementExperimentRun {
528
+ measurements: AgentProfileImprovementMeasurement[];
529
+ measurement: {
530
+ wallDurationMs: number;
531
+ cost: AgentImprovementCost;
532
+ };
533
+ }
511
534
  interface CompareAgentProfileImprovementExperimentOptions {
512
535
  experiment: AgentProfileImprovementExperiment;
513
536
  measurements: AgentProfileImprovementMeasurement[];
537
+ preparation: {
538
+ wallDurationMs: number;
539
+ cost: AgentImprovementCost;
540
+ };
541
+ measurement: AgentProfileImprovementExperimentRun['measurement'];
514
542
  runId: string;
515
543
  candidate?: AgentProfileImprovementMeasuredComparison['candidate'];
516
544
  generationsExplored?: number;
517
- searchDurationMs?: number;
518
- searchCostUsd?: number;
519
545
  metadata?: AgentProfileImprovementMeasuredComparison['metadata'];
520
546
  }
521
547
  /** Content-address one held-out profile task before either state can execute it. */
@@ -532,7 +558,7 @@ declare function verifyAgentProfileImprovementExperiment(input: unknown): AgentP
532
558
  * Eval owns only the cell schedule and receipt checks; the host resolves each
533
559
  * state digest and captures its own run, billing, trace, and grader evidence.
534
560
  */
535
- declare function runAgentProfileImprovementExperiment(options: RunAgentProfileImprovementExperimentOptions): Promise<AgentProfileImprovementMeasurement[]>;
561
+ declare function runAgentProfileImprovementExperiment(options: RunAgentProfileImprovementExperimentOptions): Promise<AgentProfileImprovementExperimentRun>;
536
562
  /** Build the only publishable profile comparison from complete host receipts. */
537
563
  declare function measuredComparisonFromAgentProfileImprovementExperiment(options: CompareAgentProfileImprovementExperimentOptions): AgentProfileImprovementMeasuredComparison;
538
564
  /** Recompute a profile comparison from the exact sealed experiment and receipts. */
@@ -811,6 +837,146 @@ interface PartitionByAuthoringModelResult {
811
837
  */
812
838
  declare function partitionRunsByAuthoringModel(runs: RunRecord[], index: AgentTraceIndex): PartitionByAuthoringModelResult;
813
839
  //#endregion
840
+ //#region src/contract/intake/code-agent-observation.d.ts
841
+ type CodeAgentSessionSource = 'codex' | 'claude-code' | 'opencode' | 'kimi-code' | 'pi';
842
+ type CodeAgentSessionTerminalStatus = 'completed' | 'failed' | 'unknown';
843
+ type CodeAgentSessionActionKind = 'tool' | 'patch' | 'terminal' | 'graph-completion';
844
+ type CodeAgentSessionActionSurface = 'tool' | 'mcp' | 'subagent' | 'skill' | 'hook' | 'web' | 'code';
845
+ type CodeAgentSessionActionStatus = 'started' | 'completed' | 'failed' | 'unknown';
846
+ interface CodeAgentSessionExecutionReceipt {
847
+ exitCode: number;
848
+ startedAtMs?: number;
849
+ completedAtMs?: number;
850
+ }
851
+ interface CodeAgentSessionAction {
852
+ id: string;
853
+ stepIndex: number;
854
+ kind: CodeAgentSessionActionKind;
855
+ surface: CodeAgentSessionActionSurface;
856
+ name: string;
857
+ status: CodeAgentSessionActionStatus;
858
+ timestampMs?: number;
859
+ costUsd?: number;
860
+ metadata: Record<string, unknown>;
861
+ }
862
+ interface CodeAgentSessionObservation {
863
+ source: CodeAgentSessionSource;
864
+ sessionId: string;
865
+ finalText: string | null;
866
+ terminal: {
867
+ status: CodeAgentSessionTerminalStatus;
868
+ explicit: boolean;
869
+ };
870
+ actions: CodeAgentSessionAction[];
871
+ }
872
+ interface ObserveCodeAgentSessionOptions {
873
+ source: CodeAgentSessionSource;
874
+ entries: unknown[];
875
+ sourcePath?: string;
876
+ execution?: CodeAgentSessionExecutionReceipt;
877
+ }
878
+ /**
879
+ * Project one provider session into the exact user-visible answer and a
880
+ * provider-neutral action stream. Raw prompts, tool inputs, and tool outputs
881
+ * stay out of this projection; callers retain the source JSONL as evidence.
882
+ */
883
+ declare function observeCodeAgentSession(options: ObserveCodeAgentSessionOptions): CodeAgentSessionObservation;
884
+ //#endregion
885
+ //#region src/contract/intake/code-agent-session.d.ts
886
+ interface ParsedCodeAgentJsonl {
887
+ entries: unknown[];
888
+ malformedLines: number;
889
+ }
890
+ interface CodeAgentSessionMetrics {
891
+ entries: number;
892
+ userMessages: number;
893
+ assistantMessages: number;
894
+ reasoningItems: number;
895
+ toolCalls: number;
896
+ toolOutputs: number;
897
+ toolErrors: number;
898
+ unclassifiedErrors: number;
899
+ patchAttempts: number;
900
+ patchSuccesses: number;
901
+ patchFailures: number;
902
+ turnsStarted: number;
903
+ turnsCompleted: number;
904
+ turnsAborted: number;
905
+ contextCompactions: number;
906
+ mcpCalls: number;
907
+ subagentCalls: number;
908
+ skillCalls: number;
909
+ hookCalls: number;
910
+ webCalls: number;
911
+ codeActions: number;
912
+ prLinks: number;
913
+ fileSnapshots: number;
914
+ graphNodes: number;
915
+ graphEdges: number;
916
+ actionCandidates: number;
917
+ verificationReports: number;
918
+ completionDecisions: number;
919
+ reliabilityRows: number;
920
+ reliabilityLift: number;
921
+ inputTokens: number;
922
+ outputTokens: number;
923
+ reasoningTokens: number;
924
+ cachedTokens: number;
925
+ cacheWriteTokens: number;
926
+ observedCostUsd: number;
927
+ observedCostCaptured?: boolean;
928
+ wallMs: number;
929
+ processScore: number;
930
+ }
931
+ interface CodeAgentSessionDiagnostic {
932
+ source: CodeAgentSessionSource;
933
+ sessionId: string;
934
+ sourcePath?: string;
935
+ entries: number;
936
+ malformedLines: number;
937
+ hasExplicitTerminalSignal: boolean;
938
+ hasFinalOutput: boolean;
939
+ hasQualityLabel: boolean;
940
+ hasTokenUsage: boolean;
941
+ hasCost: boolean;
942
+ costKind?: RunCostProvenance['kind'];
943
+ warnings: string[];
944
+ }
945
+ interface CodeAgentSessionIntakeResult {
946
+ runs: RunRecord[];
947
+ diagnostics: CodeAgentSessionDiagnostic[];
948
+ metrics: CodeAgentSessionMetrics[];
949
+ observations: CodeAgentSessionObservation[];
950
+ }
951
+ interface CodeAgentSessionIntakeOptions {
952
+ entries: unknown[];
953
+ malformedLines?: number;
954
+ sourcePath?: string;
955
+ experimentId?: string;
956
+ candidateId?: string;
957
+ seed?: number;
958
+ splitTag?: RunSplitTag;
959
+ scenarioId?: string;
960
+ model?: string;
961
+ promptHash?: string;
962
+ configHash?: string;
963
+ commitSha?: string;
964
+ score?: number;
965
+ /** Explicit cost receipt. When omitted, source-reported cost wins, then a
966
+ * token-priced estimate, then uncaptured. */
967
+ costProvenance?: RunCostProvenance;
968
+ /** Exact executor-owned process result. This is required when a provider's
969
+ * JSON stream has no terminal event, as with `opencode run --format json`. */
970
+ execution?: CodeAgentSessionExecutionReceipt;
971
+ }
972
+ declare function parseCodeAgentJsonl(jsonl: string): ParsedCodeAgentJsonl;
973
+ declare function fromCodexSession(options: CodeAgentSessionIntakeOptions): CodeAgentSessionIntakeResult;
974
+ declare function fromClaudeCodeSession(options: CodeAgentSessionIntakeOptions): CodeAgentSessionIntakeResult;
975
+ declare function fromOpenCodeSession(options: CodeAgentSessionIntakeOptions): CodeAgentSessionIntakeResult;
976
+ declare function fromKimiCodeSession(options: CodeAgentSessionIntakeOptions): CodeAgentSessionIntakeResult;
977
+ declare function fromPiSession(options: CodeAgentSessionIntakeOptions): CodeAgentSessionIntakeResult;
978
+ declare const fromPigraphSession: typeof fromPiSession;
979
+ //#endregion
814
980
  //#region src/contract/intake/feedback-table.d.ts
815
981
  interface FeedbackTableRow {
816
982
  /** Stable id for this run — the unit a rater scored. Drives pairing
@@ -899,5 +1065,5 @@ interface FromOtelSpansOptions {
899
1065
  }
900
1066
  declare function fromOtelSpans(opts: FromOtelSpansOptions): RunRecord[];
901
1067
  //#endregion
902
- export { type AgentEvalAgent, type AgentEvalEvaluateOptions, type AgentEvalImproveOptions, type AgentProfileImprovementExperimentExecutionInput, type AgentTraceContributor, type AgentTraceContributorType, type AgentTraceConversation, type AgentTraceFile, type AgentTraceIndex, type AgentTraceRange, type AgentTraceRecord, type AnalystFinding, type AnalyzeRunsOptions, type AuthoringProvenance, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type CampaignAggregates, type CampaignArtifactWriter, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCostMeter, type CampaignResult, type CampaignStorage, type CampaignTraceWriter, type CandidateExperimentExecutionInput, type ChatClient, type CodeAgentSessionAction, type CodeAgentSessionActionKind, type CodeAgentSessionActionStatus, type CodeAgentSessionActionSurface, type CodeAgentSessionDiagnostic, type CodeAgentSessionExecutionReceipt, type CodeAgentSessionIntakeOptions, type CodeAgentSessionIntakeResult, type CodeAgentSessionMetrics, type CodeAgentSessionObservation, type CodeAgentSessionSource, type CodeAgentSessionTerminalStatus, type CodeSurface, type CompareAgentProfileImprovementExperimentOptions, type CompareCandidateExperimentOptions, type CompareOptimizationMethodsOptions, type ComparisonCost, type CostLedgerHandle, type CostProvenanceSummary, type CreateChatClientOpts, type DefaultAnalystRegistryOptions, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DefineAgentEvalOptions, type DefinedAgentEval, type DeploymentOutcome, type DispatchFn as Dispatch, type DispatchContext, type EvalCellScoreDelta, type EvalDimensionDelta, type EvalGenerationDiff, type EvalReportingSuiteInput, type EvalReportingSuiteOptions, type EvalReportingSuiteResult, type EvalRunDiff, type EvaluatePairedMeasurementsOptions, type EvidenceVector, type ExecutionErrorOutcomeCell, type ExecutionInsight, type ExecutionReport, type ExternalOptimizationExample, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, type FailureClassTally, type FailureClusterInsight, type FeedbackTableMeta, type FeedbackTableRow, FileSystemOutcomeStore, type FileSystemOutcomeStoreOptions, type FromFeedbackTableOptions, type FromFeedbackTableResult, type FromOtelSpansOptions, type FromRunRecordDirOptions, type FromRunRecordDirResult, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaEngineOptions, type GepaEngineRun, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaRunnerCommand, type HeldOutGateOptions, type HostedTenant, InMemoryOutcomeStore, type InsightReport, type InterRaterInsight, type JudgeConfig, type JudgeDimension, type JudgeInsight, type JudgeScore, type LiftInsight, type LlmJudgeDimension, type LlmJudgeOptions, type MutableSurface, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodInput, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type OutcomeCorrelationInsight, type OutcomeStore, type PairedMeasurement, type PairedMeasurementAdapter, type PairedMeasurementEvaluation, type ParetoSignificanceGateOptions, type ParsedCodeAgentJsonl, type PartitionByAuthoringModelResult, type PromotionObjective, type PromotionPolicy, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, type Recommendation, type ReferenceEquivalenceJudgeInput, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceJudgeResult, type ReferenceEquivalenceScenario, type ReleaseSummary, type RunAgentProfileImprovementExperimentOptions, type RunCampaignOptions, type RunCandidateExperimentOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunRecordRejection, type ScalarDistribution, type Scenario, type SealAgentProfileImprovementSuiteOptions, type SealCandidateBenchmarkSuiteOptions, type SelfImproveBudget, type SelfImproveOptions, type SelfImproveProgressEvent, type SelfImproveResult, SelfImproveRunError, type SessionScript, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SummarizeExecutionOptions, type SurfaceProposer, type TokenUsageInsight, analyzeRuns, buildDefaultAnalystRegistry, buildEvidenceVector, campaignSplitDigest, compareOptimizationMethods, composeGate, createChatClient, createReferenceEquivalenceJudge, defaultProductionGate, defineAgentEval, diffGenerations, diffRunBaselineToWinner, diffRuns, evalReportingSuite, evaluatePairedMeasurements, externalTextOptimizationMethod, fromClaudeCodeSession, fromCodexSession, fromFeedbackTable, fromKimiCodeSession, fromOpenCodeSession, fromOtelSpans, fromPiSession, fromPigraphSession, fromRunRecordDir, fsCampaignStorage, gepaOptimizationMethod, heldOutGate, inMemoryCampaignStorage, llmJudge, measuredComparisonFromAgentProfileImprovementExperiment, measuredComparisonFromCandidateExperiment, observeCodeAgentSession, paretoPolicy, paretoSignificanceGate, parseAgentTrace, parseCodeAgentJsonl, partitionRunsByAuthoringModel, runAgentProfileImprovementExperiment, runCampaign, runCandidateExperiment, runEval, runImprovementLoop, runReferenceEquivalenceJudge, sealAgentProfileImprovementExperiment, sealAgentProfileImprovementSuite, sealAgentProfileImprovementTask, sealCandidateBenchmarkSuite, sealCandidateBenchmarkTask, sealCandidateExperiment, selfImprove, skillOptOptimizationMethod, summarizeExecution, verifyAgentProfileImprovementExperiment, verifyAgentProfileImprovementExperimentComparison, verifyAgentProfileImprovementSuiteInputs, verifyAgentProfileImprovementTask, verifyCandidateBenchmarkSuite, verifyCandidateBenchmarkSuiteInputs, verifyCandidateBenchmarkTask, verifyCandidateExperiment, verifyCandidateExperimentComparison };
1068
+ export { type AgentEvalAgent, type AgentEvalEvaluateOptions, type AgentEvalImproveOptions, type AgentProfileImprovementExperimentExecutionInput, type AgentProfileImprovementExperimentRun, type AgentTraceContributor, type AgentTraceContributorType, type AgentTraceConversation, type AgentTraceFile, type AgentTraceIndex, type AgentTraceRange, type AgentTraceRecord, type AnalystFinding, type AnalyzeRunsOptions, type AuthoringProvenance, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type CampaignAggregates, type CampaignArtifactWriter, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCostMeter, type CampaignResult, type CampaignStorage, type CampaignTraceWriter, type CandidateExperimentExecutionInput, type CandidateExperimentRun, type ChatClient, type CodeAgentSessionAction, type CodeAgentSessionActionKind, type CodeAgentSessionActionStatus, type CodeAgentSessionActionSurface, type CodeAgentSessionDiagnostic, type CodeAgentSessionExecutionReceipt, type CodeAgentSessionIntakeOptions, type CodeAgentSessionIntakeResult, type CodeAgentSessionMetrics, type CodeAgentSessionObservation, type CodeAgentSessionSource, type CodeAgentSessionTerminalStatus, type CodeSurface, type CompareAgentProfileImprovementExperimentOptions, type CompareCandidateExperimentOptions, type CompareOptimizationMethodsOptions, type ComparisonCost, type CostLedgerHandle, type CostProvenanceSummary, type CreateChatClientOpts, type DefaultAnalystRegistryOptions, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DefineAgentEvalOptions, type DefinedAgentEval, type DeploymentOutcome, type DispatchFn as Dispatch, type DispatchContext, type EvalCellScoreDelta, type EvalDimensionDelta, type EvalGenerationDiff, type EvalReportingSuiteInput, type EvalReportingSuiteOptions, type EvalReportingSuiteResult, type EvalRunDiff, type EvaluatePairedMeasurementsOptions, type EvidenceVector, type ExecutionErrorOutcomeCell, type ExecutionInsight, type ExecutionReport, type ExternalOptimizationExample, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, type FailureClassTally, type FailureClusterInsight, type FeedbackTableMeta, type FeedbackTableRow, FileSystemOutcomeStore, type FileSystemOutcomeStoreOptions, type FromFeedbackTableOptions, type FromFeedbackTableResult, type FromOtelSpansOptions, type FromRunRecordDirOptions, type FromRunRecordDirResult, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaEngineOptions, type GepaEngineRun, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaRunnerCommand, type HeldOutGateOptions, type HostedTenant, InMemoryOutcomeStore, type InsightReport, type InterRaterInsight, type JudgeConfig, type JudgeDimension, type JudgeInsight, type JudgeScore, type LiftInsight, type LlmJudgeDimension, type LlmJudgeOptions, type MutableSurface, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodInput, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type OutcomeCorrelationInsight, type OutcomeStore, type PairedMeasurement, type PairedMeasurementAdapter, type PairedMeasurementEvaluation, type ParetoSignificanceGateOptions, type ParsedCodeAgentJsonl, type PartitionByAuthoringModelResult, type PromotionObjective, type PromotionPolicy, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, type Recommendation, type ReferenceEquivalenceJudgeInput, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceJudgeResult, type ReferenceEquivalenceScenario, type ReleaseSummary, type RunAgentProfileImprovementExperimentOptions, type RunCampaignOptions, type RunCandidateExperimentOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunRecordRejection, type ScalarDistribution, type Scenario, type SealAgentProfileImprovementSuiteOptions, type SealCandidateBenchmarkSuiteOptions, type SelfImproveBudget, type SelfImproveOptions, type SelfImproveProgressEvent, type SelfImproveResult, SelfImproveRunError, type SessionScript, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SummarizeExecutionOptions, type SurfaceProposer, type TokenUsageInsight, analyzeRuns, buildDefaultAnalystRegistry, buildEvidenceVector, campaignSplitDigest, compareOptimizationMethods, composeGate, createChatClient, createReferenceEquivalenceJudge, defaultProductionGate, defineAgentEval, diffGenerations, diffRunBaselineToWinner, diffRuns, evalReportingSuite, evaluatePairedMeasurements, externalTextOptimizationMethod, fromClaudeCodeSession, fromCodexSession, fromFeedbackTable, fromKimiCodeSession, fromOpenCodeSession, fromOtelSpans, fromPiSession, fromPigraphSession, fromRunRecordDir, fsCampaignStorage, gepaOptimizationMethod, heldOutGate, inMemoryCampaignStorage, llmJudge, measuredComparisonFromAgentProfileImprovementExperiment, measuredComparisonFromCandidateExperiment, observeCodeAgentSession, paretoPolicy, paretoSignificanceGate, parseAgentTrace, parseCodeAgentJsonl, partitionRunsByAuthoringModel, runAgentProfileImprovementExperiment, runCampaign, runCandidateExperiment, runEval, runImprovementLoop, runReferenceEquivalenceJudge, sealAgentProfileImprovementExperiment, sealAgentProfileImprovementSuite, sealAgentProfileImprovementTask, sealCandidateBenchmarkSuite, sealCandidateBenchmarkTask, sealCandidateExperiment, selfImprove, skillOptOptimizationMethod, summarizeExecution, verifyAgentProfileImprovementExperiment, verifyAgentProfileImprovementExperimentComparison, verifyAgentProfileImprovementSuiteInputs, verifyAgentProfileImprovementTask, verifyCandidateBenchmarkSuite, verifyCandidateBenchmarkSuiteInputs, verifyCandidateBenchmarkTask, verifyCandidateExperiment, verifyCandidateExperimentComparison };
903
1069
  //# sourceMappingURL=index.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","names":[],"sources":["../../src/contract/self-improve.ts","../../src/contract/define-agent-eval.ts","../../src/contract/measured-comparison.ts","../../src/contract/profile-measured-comparison.ts","../../src/contract/intake/run-record-dir.ts","../../src/contract/eval-reporting-suite.ts","../../src/contract/diff.ts","../../src/contract/intake/agent-trace.ts","../../src/contract/intake/feedback-table.ts","../../src/contract/intake/otel-spans.ts"],"mappings":";;;;;;;;;;;;UA8DiB;;;EAGf;;;;EAIA;;EAEA;;EAEA;;;EAGA;;;EAGA;;;;EAIA;;EAEA,mBAAmB;;;;;;;;;EASnB;;EAEA;;;;;EAKA;;KAGU;EACN;EAA0B;;EAC1B;EAA4B;EAAuB;;EACnD;EAA4B;EAAe;;EAC3C;EAA8B;EAAe;EAAuB;;EAGpE;EAAsB;EAAkB;;EACxC;EAAyB;EAAW;EAAY;EAAa;;UAElD,mBAAmB,kBAAkB,UAAU;;;;;;;;;;;;;;EAc9D,QAAQ,SAAS,gBAAgB,UAAU,WAAW,KAAK,oBAAoB,QAAQ;;;;;;;EAQvF;;;EAIA,WAAW;;;EAIX,OAAO,YAAY,WAAW;;;EAI9B,iBAAiB;;EAGjB,SAAS;;;;;;;;;;;;EAaT,sBAAsB,gCAAgC,WAAW;;;;;EAMjE,WAAW;;;;;;EAOX,SAAS,mBAAmB,WAAW;;;EAIvC,qBAAqB;;;EAIrB,OAAO,KAAK,WAAW;;;;;;;;EASvB,cAAc,eAAe,gBAAgB,iBAAiB,mBAAmB;;;;EAKjF,UAAU;;;;EAKV;;;EAIA,gBAAgB,QAAQ;;;;;EAMxB,iBAAiB;IACf,UAAU;IACV;IACA;;;;EAKF;;;EAIA,cAAc,OAAO;;;;EAKrB;EACA;EACA;;;;;;;;;;;;;;EAeA,eAAe;;;EAIf,eAAe;;;;EAKf,eAAe;;EAGf;;;;;;;;EASA;;;;;;;;;EAUA,oBAAoB,uBAAuB,WAAW;;;EAItD;;;;;;;EAQA,mBAAmB,uBAAuB,WAAW;;UAGtC,kBAAkB,kBAAkB,UAAU;;;;EAI7D;IACE;IACA,aAAa;;;;;EAKf;IACE;IACA,aAAa;IACb,SAAS;;;IAGT;;;;IAIA;;;;;;EAMF;;;EAGA;;;;EAIA,YAAY;;EAEZ;;;EAGA;;EAEA;;EAEA;;EAEA,MAAM;;;EAGN,UAAU;;EAEV;IACE;IACA,MAAM;IACN;IACA,aAAa;;;;;;;;EAQf,SAAS;;;;;EAKT,QAAQ;;;;;;EAMR,KAAK,yBAAyB,WAAW;;;cAI9B,4BAA4B;WAC9B,MAAM;WACN,UAAU;EAEnB,YAAY,gBAAgB,QAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA+LhB,YAAY,kBAAkB,UAAU,WAC5D,MAAM,mBAAmB,WAAW,aACnC,QAAQ,kBAAkB,WAAW;;;KCtiB5B,eAAe,kBAAkB,UAAU,cACrD,SAAS,gBACT,UAAU,WACV,KAAK,oBACF,QAAQ;KAED,uBAAuB,kBAAkB,UAAU,aAAa,mBAC1E,WACA;UAGe,yBAAyB,kBAAkB,UAAU,mBAC5D,KACN,eAAe,WAAW;;EAI5B,YAAY;;EAEZ,UAAU;;EAEV,QAAQ,eAAe,WAAW;;EAElC,QAAQ,YAAY,WAAW;;EAE/B,SAAS,YAAY,WAAW;;EAEhC;;KAGU,wBAAwB,kBAAkB,UAAU,aAAa,KAC3E,QAAQ,mBAAmB,WAAW;EAGtC,SAAS,QAAQ;EACjB,eAAe,QAAQ;;UAGR,iBAAiB,kBAAkB,UAAU;;WAEnD,oBAAoB;;WAEpB,iBAAiB;;;;;EAK1B,SACE,OAAO,yBAAyB,WAAW,aAC1C,QAAQ,eAAe,WAAW;;;;;;;EAOrC,QACE,OAAO,wBAAwB,WAAW,aACzC,QAAQ,kBAAkB,WAAW;;;;;;;;;iBAU1B,gBAAgB,kBAAkB,UAAU,WAC1D,UAAU,uBAAuB,WAAW,aAC3C,iBAAiB,WAAW;;;UC7Dd;EACf,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,QAAQ;EACR,MAAM;EACN,eAAe;EACf;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QAAQ,OAAO,oCAAoC,QAAQ;;EAE3D;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,cAAc;EACd;EACA,YAAY;EACZ;EACA;EACA;EACA,WAAW;;;UAII,kBAAkB;EACjC;EACA,UAAU;EACV,WAAW;;;UAII,yBAAyB;EACxC,MAAM,KAAK;EACX,WAAW,KAAK;IAAkB;IAAc;;EAChD,QAAQ,KAAK;EACb,UAAU,KAAK;EACf,UAAU,KAAK;EACf,OAAO,KAAK;;UAGG,kCAAkC;EACjD,uBAAuB,kBAAkB;EACzC,QAAQ;EACR,SAAS,yBAAyB;;EAElC;;EAEA;;;KAIU,8BAA8B,KACxC;EAGA;EACA;EACA;;;iBAIc,2BACd,UAAU,sCACT;;iBAQa,4BACd,SAAS,qCACR;;iBAiBa,wBACd,UAAU,mCACT;iBAQa,0BAA0B,iBAAiB;;iBAarC,uBACpB,SAAS,gCACR,QAAQ;;;;;;;;iBA0CK,2BAA2B,MACzC,SAAS,kCAAkC,QAC1C;;iBA2Na,0CACd,SAAS,oCACR;;iBAoEa,oCACd,iBACC;iBAwCa,6BAA6B,iBAAiB;iBAM9C,oCACd,iBACC;iBAkBa,8BAA8B;;;;;;;;;;UCnhB7B;EACf,aAAa;EACb,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,aAAa;EACb,MAAM;EACN,SAAS;EACT;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QACE,OAAO,kDACN,QAAQ;;EAEX;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,cAAc;EACd;EACA,YAAY;EACZ;EACA;EACA;EACA,WAAW;;;iBAIG,gCACd,UAAU,sCACT;;iBAQa,iCACd,SAAS,0CACR;;iBAqBa,sCACd,UAAU,4CACT;iBAOa,kCAAkC,iBAAiB;iBAInD,yCACd,iBACC;iBAIa,wCACd,iBACC;;;;;;iBASmB,qCACpB,SAAS,8CACR,QAAQ;;iBAmBK,wDACd,SAAS,kDACR;;iBAoEa,kDACd,iBACC;;;;UCxMc;;EAEf;;EAEA;;EAEA;;UAGe;;;;;;EAMf;;;;;;;EAOA,WAAW;;;;;;EAMX;;UAGe;;EAEf,MAAM;;EAEN,UAAU;;EAEV;;;;;;;;;;iBAkBoB,iBACpB,cACA,UAAS,0BACR,QAAQ;;;;;KC7CC,0BAA0B;UAErB;;;;;EAKf,UAAU,KAAK;;EAEf,OAAO;;;;;;;;;;EAUP;;;;UAKe;;;EAGf,QAAQ;;EAER;;IAEE;;IAEA;;;IAGA;;IAEA;;;IAGA,UAAU;;;EAGZ;;;;;;;iBAUoB,mBACpB,OAAO,yBACP,UAAS,4BACR,QAAQ;;;;;;UC5DM;EACf;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;;;EAGA,YAAY,eAAe,eAAe;;;;UAK3B;EACf;EACA;EACA;EACA;EACA;;EAEA,SAAS;;EAET,SAAS;;EAET,OAAO;;EAEP;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;;;;UAMe;EACf;EACA;EACA;EACA;EACA,oBAAoB;EACpB,mBAAmB;EACnB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA,cAAc;;;;EAId,aAAa;;;;;;;;;iBAoDC,gBACd,QAAQ,2BACR,OAAO,4BACN;;;;;;iBAqEa,SAAS,QAAQ,cAAc,OAAO,eAAe;;;;;;;iBAsCrD,wBAAwB,KAAK,eAAe;;;KC1OhD;UAEK;EACf,MAAM;;EAEN;;UAGe;EACf;EACA;EACA;;;EAGA,cAAc;;UAGC;EACf;EACA,cAAc;EACd,QAAQ;;UAGO;EACf;EACA,eAAe;;UAGA;EACf;EACA;EACA;EACA;IAAQ;IAAc;;EACtB;IAAS;IAAe;;EACxB,OAAO;;;;UAOQ;EACf;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;KAGU,kBAAkB,YAAY;;;;;;iBAW1B,gBAAgB,SAAS,qBAAqB;UAmE7C;;;;EAIf,SAAS,YAAY;;;EAGrB,cAAc;;;;;;;;iBASA,8BACd,MAAM,aACN,OAAO,kBACN;;;UC/Jc;;;EAGf;;EAEA;;;EAGA;;;EAGA,WAAW;;UAGI;EACf;;;EAGA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA,WAAW;;;EAGX,SAAS;;UAGM;;EAEf,SAAS;;;EAGT,OAAO;;;;EAIP;IAAU;IAAa;;;;;EAIvB;;UAGe;EACf,MAAM;;;EAGN,aAAa;IAAQ;IAAe;IAAe;;;iBAGrC,kBAAkB,MAAM,2BAA2B;;;UCvBlD;EACf,OAAO;;EAEP,eAAe;;EAEf;;;;;;EAMA,eAAe,eAAe,gBAAgB;;iBAGhC,cAAc,MAAM,uBAAuB"}
1
+ {"version":3,"file":"index.d.ts","names":[],"sources":["../../src/contract/self-improve.ts","../../src/contract/define-agent-eval.ts","../../src/contract/measured-comparison.ts","../../src/contract/profile-measured-comparison.ts","../../src/contract/intake/run-record-dir.ts","../../src/contract/eval-reporting-suite.ts","../../src/contract/diff.ts","../../src/contract/intake/agent-trace.ts","../../src/contract/intake/code-agent-observation.ts","../../src/contract/intake/code-agent-session.ts","../../src/contract/intake/feedback-table.ts","../../src/contract/intake/otel-spans.ts"],"mappings":";;;;;;;;;;;UA8DiB;;;EAGf;;;;EAIA;;EAEA;;EAEA;;;EAGA;;;EAGA;;;;EAIA;;EAEA,mBAAmB;;;;;;;;;EASnB;;EAEA;;;;;EAKA;;KAGU;EACN;EAA0B;;EAC1B;EAA4B;EAAuB;;EACnD;EAA4B;EAAe;;EAC3C;EAA8B;EAAe;EAAuB;;EAGpE;EAAsB;EAAkB;;EACxC;EAAyB;EAAW;EAAY;EAAa;;UAElD,mBAAmB,kBAAkB,UAAU;;;;;;;;;;;;;;EAc9D,QAAQ,SAAS,gBAAgB,UAAU,WAAW,KAAK,oBAAoB,QAAQ;;;;;;;EAQvF;;;EAIA,WAAW;;;EAIX,OAAO,YAAY,WAAW;;;EAI9B,iBAAiB;;EAGjB,SAAS;;;;;;;;;;;;EAaT,sBAAsB,gCAAgC,WAAW;;;;;EAMjE,WAAW;;;;;;EAOX,SAAS,mBAAmB,WAAW;;;EAIvC,qBAAqB;;;EAIrB,OAAO,KAAK,WAAW;;;;;;;;EASvB,cAAc,eAAe,gBAAgB,iBAAiB,mBAAmB;;;;EAKjF,UAAU;;;;EAKV;;;EAIA,gBAAgB,QAAQ;;;;;EAMxB,iBAAiB;IACf,UAAU;IACV;IACA;;;;EAKF;;;EAIA,cAAc,OAAO;;;;EAKrB;EACA;EACA;;;;;;;;;;;;;;EAeA,eAAe;;;EAIf,eAAe;;;;EAKf,eAAe;;EAGf;;;;;;;;EASA;;;;;;;;;EAUA,oBAAoB,uBAAuB,WAAW;;;EAItD;;;;;;;EAQA,mBAAmB,uBAAuB,WAAW;;UAGtC,kBAAkB,kBAAkB,UAAU;;;;EAI7D;IACE;IACA,aAAa;;;;;EAKf;IACE;IACA,aAAa;IACb,SAAS;;;IAGT;;;;IAIA;;;;;;EAMF;;;EAGA;;;;EAIA,YAAY;;EAEZ;;;EAGA;;EAEA;;EAEA;;EAEA,MAAM;;;EAGN,UAAU;;EAEV;IACE;IACA,MAAM;IACN;IACA,aAAa;;;;;;;;EAQf,SAAS;;;;;EAKT,QAAQ;;;;;;EAMR,KAAK,yBAAyB,WAAW;;;cAI9B,4BAA4B;WAC9B,MAAM;WACN,UAAU;EAEnB,YAAY,gBAAgB,QAAQ;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA+LhB,YAAY,kBAAkB,UAAU,WAC5D,MAAM,mBAAmB,WAAW,aACnC,QAAQ,kBAAkB,WAAW;;;KCtiB5B,eAAe,kBAAkB,UAAU,cACrD,SAAS,gBACT,UAAU,WACV,KAAK,oBACF,QAAQ;KAED,uBAAuB,kBAAkB,UAAU,aAAa,mBAC1E,WACA;UAGe,yBAAyB,kBAAkB,UAAU,mBAC5D,KACN,eAAe,WAAW;;EAI5B,YAAY;;EAEZ,UAAU;;EAEV,QAAQ,eAAe,WAAW;;EAElC,QAAQ,YAAY,WAAW;;EAE/B,SAAS,YAAY,WAAW;;EAEhC;;KAGU,wBAAwB,kBAAkB,UAAU,aAAa,KAC3E,QAAQ,mBAAmB,WAAW;EAGtC,SAAS,QAAQ;EACjB,eAAe,QAAQ;;UAGR,iBAAiB,kBAAkB,UAAU;;WAEnD,oBAAoB;;WAEpB,iBAAiB;;;;;EAK1B,SACE,OAAO,yBAAyB,WAAW,aAC1C,QAAQ,eAAe,WAAW;;;;;;;EAOrC,QACE,OAAO,wBAAwB,WAAW,aACzC,QAAQ,kBAAkB,WAAW;;;;;;;;;iBAU1B,gBAAgB,kBAAkB,UAAU,WAC1D,UAAU,uBAAuB,WAAW,aAC3C,iBAAiB,WAAW;;;UCxDd;EACf,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,QAAQ;EACR,MAAM;EACN,eAAe;EACf;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QAAQ,OAAO,oCAAoC,QAAQ;;EAE3D;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;UAII,kBAAkB;EACjC;EACA,UAAU;EACV,WAAW;;;UAII,yBAAyB;EACxC,MAAM,KAAK;EACX,WAAW,KAAK;IAAkB;IAAc;;EAChD,QAAQ,KAAK;EACb,eAAe,KAAK,OAAO;EAC3B,UAAU,KAAK;EACf,UAAU,KAAK;EACf,OAAO,KAAK;;UAGG,kCAAkC;EACjD,uBAAuB,kBAAkB;EACzC,QAAQ;EACR,SAAS,yBAAyB;;EAElC;;EAEA,kBAAkB;;EAElB,kBAAkB;;;KAIR,8BAA8B,KACxC;EAGA,iBAAiB;EACjB,WAAW;EACX;;;iBAIc,2BACd,UAAU,sCACT;;iBAQa,4BACd,SAAS,qCACR;;iBAiBa,wBACd,UAAU,mCACT;iBAQa,0BAA0B,iBAAiB;;iBAarC,uBACpB,SAAS,gCACR,QAAQ;;;;;;;;iBAkEK,2BAA2B,MACzC,SAAS,kCAAkC,QAC1C;;iBAuPa,0CACd,SAAS,oCACR;;iBAuEa,oCACd,iBACC;iBAkHa,6BAA6B,iBAAiB;iBAM9C,oCACd,iBACC;iBAkBa,8BAA8B;;;;;;;;;;UCrqB7B;EACf,aAAa;EACb,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,aAAa;EACb,MAAM;EACN,SAAS;EACT;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QACE,OAAO,kDACN,QAAQ;;EAEX;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;iBAIG,gCACd,UAAU,sCACT;;iBAQa,iCACd,SAAS,0CACR;;iBAqBa,sCACd,UAAU,4CACT;iBAOa,kCAAkC,iBAAiB;iBAInD,yCACd,iBACC;iBAIa,wCACd,iBACC;;;;;;iBASmB,qCACpB,SAAS,8CACR,QAAQ;;iBA0CK,wDACd,SAAS,kDACR;;iBAuEa,kDACd,iBACC;;;;UCnPc;;EAEf;;EAEA;;EAEA;;UAGe;;;;;;EAMf;;;;;;;EAOA,WAAW;;;;;;EAMX;;UAGe;;EAEf,MAAM;;EAEN,UAAU;;EAEV;;;;;;;;;;iBAkBoB,iBACpB,cACA,UAAS,0BACR,QAAQ;;;;;KC7CC,0BAA0B;UAErB;;;;;EAKf,UAAU,KAAK;;EAEf,OAAO;;;;;;;;;;EAUP;;;;UAKe;;;EAGf,QAAQ;;EAER;;IAEE;;IAEA;;;IAGA;;IAEA;;;IAGA,UAAU;;;EAGZ;;;;;;;iBAUoB,mBACpB,OAAO,yBACP,UAAS,4BACR,QAAQ;;;;;;UC5DM;EACf;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;;;EAGA,YAAY,eAAe,eAAe;;;;UAK3B;EACf;EACA;EACA;EACA;EACA;;EAEA,SAAS;;EAET,SAAS;;EAET,OAAO;;EAEP;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;;;;UAMe;EACf;EACA;EACA;EACA;EACA,oBAAoB;EACpB,mBAAmB;EACnB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA,cAAc;;;;EAId,aAAa;;;;;;;;;iBAoDC,gBACd,QAAQ,2BACR,OAAO,4BACN;;;;;;iBAqEa,SAAS,QAAQ,cAAc,OAAO,eAAe;;;;;;;iBAsCrD,wBAAwB,KAAK,eAAe;;;KC1OhD;UAEK;EACf,MAAM;;EAEN;;UAGe;EACf;EACA;EACA;;;EAGA,cAAc;;UAGC;EACf;EACA,cAAc;EACd,QAAQ;;UAGO;EACf;EACA,eAAe;;UAGA;EACf;EACA;EACA;EACA;IAAQ;IAAc;;EACtB;IAAS;IAAe;;EACxB,OAAO;;;;UAOQ;EACf;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;KAGU,kBAAkB,YAAY;;;;;;iBAW1B,gBAAgB,SAAS,qBAAqB;UAmE7C;;;;EAIf,SAAS,YAAY;;;EAGrB,cAAc;;;;;;;;iBASA,8BACd,MAAM,aACN,OAAO,kBACN;;;KCjLS;KAEA;KAEA;KAEA;KASA;UAEK;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA,MAAM;EACN,SAAS;EACT;EACA,QAAQ;EACR;EACA;EACA,UAAU;;UAGK;EACf,QAAQ;EACR;EACA;EACA;IACE,QAAQ;IACR;;EAEF,SAAS;;UAGM;EACf,QAAQ;EACR;EACA;EACA,YAAY;;;;;;;iBAeE,wBACd,SAAS,iCACR;;;UCrCc;EACf;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf,QAAQ;EACR;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;;UAGe;EACf,MAAM;EACN,aAAa;EACb,SAAS;EACT,cAAc;;UAGC;EACf;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;EACA;EACA;EACA;EACA;EACA;;;EAGA,iBAAiB;;;EAGjB,YAAY;;iBAGE,oBAAoB,gBAAgB;iBAepC,iBACd,SAAS,gCACR;iBAIa,sBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,cACd,SAAS,gCACR;cAIU,2BAAkB;;;UCpJd;;;EAGf;;EAEA;;;EAGA;;;EAGA,WAAW;;UAGI;EACf;;;EAGA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA,WAAW;;;EAGX,SAAS;;UAGM;;EAEf,SAAS;;;EAGT,OAAO;;;;EAIP;IAAU;IAAa;;;;;EAIvB;;UAGe;EACf,MAAM;;;EAGN,aAAa;IAAQ;IAAe;IAAe;;;iBAGrC,kBAAkB,MAAM,2BAA2B;;;UCvBlD;EACf,OAAO;;EAEP,eAAe;;EAEf;;;;;;EAMA,eAAe,eAAe,gBAAgB;;iBAGhC,cAAc,MAAM,uBAAuB"}