@tangle-network/agent-eval 0.203.2 → 0.205.0

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 (126) hide show
  1. package/CHANGELOG.md +8 -0
  2. package/dist/analyst/index.d.ts +1 -1
  3. package/dist/analyst/index.d.ts.map +1 -1
  4. package/dist/analyst/index.js +5 -5
  5. package/dist/{analyze-runs-BrIibNd0.js → analyze-runs-d5dcMlxi.js} +9 -9
  6. package/dist/{analyze-runs-BrIibNd0.js.map → analyze-runs-d5dcMlxi.js.map} +1 -1
  7. package/dist/baseline-Dm8966du.js +259 -0
  8. package/dist/baseline-Dm8966du.js.map +1 -0
  9. package/dist/{benchmark-command-6OBcB0sG.js → benchmark-command-BQhi_BVO.js} +9 -6
  10. package/dist/benchmark-command-BQhi_BVO.js.map +1 -0
  11. package/dist/benchmarks/index.js +2 -2
  12. package/dist/builder-eval/index.js +1 -1
  13. package/dist/campaign/index.js +9 -9
  14. package/dist/{campaign-DYhg4SjN.js → campaign-0lmkAMpE.js} +11 -11
  15. package/dist/{campaign-DYhg4SjN.js.map → campaign-0lmkAMpE.js.map} +1 -1
  16. package/dist/{campaign-evidence-Dt_rWPk2.js → campaign-evidence-C__cUpCt.js} +2 -2
  17. package/dist/{campaign-evidence-Dt_rWPk2.js.map → campaign-evidence-C__cUpCt.js.map} +1 -1
  18. package/dist/cli.js +5 -5
  19. package/dist/contract/index.js +10 -10
  20. package/dist/{default-registry-wInsQWlQ.js → default-registry-BthquIoC.js} +2 -2
  21. package/dist/{default-registry-wInsQWlQ.js.map → default-registry-BthquIoC.js.map} +1 -1
  22. package/dist/{define-agent-eval-B25Cc3ki.js → define-agent-eval-CeYrXpl7.js} +4 -4
  23. package/dist/{define-agent-eval-B25Cc3ki.js.map → define-agent-eval-CeYrXpl7.js.map} +1 -1
  24. package/dist/{descriptive-D9LUFk1-.js → descriptive-CnU8Yq29.js} +2 -2
  25. package/dist/{descriptive-D9LUFk1-.js.map → descriptive-CnU8Yq29.js.map} +1 -1
  26. package/dist/diagnosis/index.js +1 -1
  27. package/dist/{dspy-rlm-engine-D0h-P2Mz.js → dspy-rlm-engine-DCQKNQMn.js} +2 -2
  28. package/dist/{dspy-rlm-engine-D0h-P2Mz.js.map → dspy-rlm-engine-DCQKNQMn.js.map} +1 -1
  29. package/dist/{eval-campaign-ha_aWtk7.js → eval-campaign-DsRmYsQI.js} +2 -2
  30. package/dist/{eval-campaign-ha_aWtk7.js.map → eval-campaign-DsRmYsQI.js.map} +1 -1
  31. package/dist/experiment/index.js +9 -9
  32. package/dist/index.d.ts +61 -3
  33. package/dist/index.d.ts.map +1 -1
  34. package/dist/index.js +102 -25
  35. package/dist/index.js.map +1 -1
  36. package/dist/{internal-BOe3zmjQ.js → internal-CRHwS9iz.js} +3 -3
  37. package/dist/internal-CRHwS9iz.js.map +1 -0
  38. package/dist/jev.js +1 -1
  39. package/dist/judge-calibration-DFtEMlde.d.ts.map +1 -1
  40. package/dist/{judge-calibration-C5zgtiJT.js → judge-calibration-N44V04Se.js} +7 -17
  41. package/dist/judge-calibration-N44V04Se.js.map +1 -0
  42. package/dist/{kind-factory-CdPPHtGZ.js → kind-factory-Cn1lbsAd.js} +5 -5
  43. package/dist/kind-factory-Cn1lbsAd.js.map +1 -0
  44. package/dist/{llm-judge-DyZdrp6I.js → llm-judge-Bka5q-ds.js} +8 -8
  45. package/dist/{llm-judge-DyZdrp6I.js.map → llm-judge-Bka5q-ds.js.map} +1 -1
  46. package/dist/matrix/index.js +1 -1
  47. package/dist/{matrix-DnlHn5x1.js → matrix-_BV9-Lk5.js} +12 -9
  48. package/dist/matrix-_BV9-Lk5.js.map +1 -0
  49. package/dist/meta-eval/index.js +2 -2
  50. package/dist/multishot/golden/index.js +1 -1
  51. package/dist/multishot/index.js +1 -1
  52. package/dist/normal-DsJZtrJc.js +25 -0
  53. package/dist/normal-DsJZtrJc.js.map +1 -0
  54. package/dist/openapi.json +1 -1
  55. package/dist/{paired-arms-DpS_h8ON.js → paired-arms-Dsu-r6xd.js} +6 -27
  56. package/dist/paired-arms-Dsu-r6xd.js.map +1 -0
  57. package/dist/{paired-binary-OMBVRtc-.js → paired-binary-DpwwK1fU.js} +2 -2
  58. package/dist/{paired-binary-OMBVRtc-.js.map → paired-binary-DpwwK1fU.js.map} +1 -1
  59. package/dist/{paired-promotion-decision-DhrmjdV8.js → paired-promotion-decision-CO0kA0bq.js} +3 -3
  60. package/dist/{paired-promotion-decision-DhrmjdV8.js.map → paired-promotion-decision-CO0kA0bq.js.map} +1 -1
  61. package/dist/{paired-tests-Ccr16gcW.js → paired-tests-jbwPdSz5.js} +3 -3
  62. package/dist/{paired-tests-Ccr16gcW.js.map → paired-tests-jbwPdSz5.js.map} +1 -1
  63. package/dist/pipelines/index.d.ts +1 -1
  64. package/dist/pipelines/index.js +2 -2
  65. package/dist/{power-and-mde-bie6i965.js → power-and-mde-fK5Qocwo.js} +3 -3
  66. package/dist/{power-and-mde-bie6i965.js.map → power-and-mde-fK5Qocwo.js.map} +1 -1
  67. package/dist/{produced-state-DLToUXrD.js → produced-state-FdbE-88t.js} +3 -3
  68. package/dist/{produced-state-DLToUXrD.js.map → produced-state-FdbE-88t.js.map} +1 -1
  69. package/dist/{promotion-policy-BI1RxzwD.js → promotion-policy-BSbTAMtE.js} +4 -4
  70. package/dist/{promotion-policy-BI1RxzwD.js.map → promotion-policy-BSbTAMtE.js.map} +1 -1
  71. package/dist/quantile--0yb07pD.js +15 -0
  72. package/dist/quantile--0yb07pD.js.map +1 -0
  73. package/dist/{report-command-D0XL2Cvi.js → report-command-JrjiWReS.js} +2 -2
  74. package/dist/{report-command-D0XL2Cvi.js.map → report-command-JrjiWReS.js.map} +1 -1
  75. package/dist/reporting.js +4 -4
  76. package/dist/{reward-hacking-C-a__ShZ.js → reward-hacking-BFcQgkux.js} +2 -2
  77. package/dist/{reward-hacking-C-a__ShZ.js.map → reward-hacking-BFcQgkux.js.map} +1 -1
  78. package/dist/rl.js +7 -7
  79. package/dist/{rubric-predictive-validity-B0Pg-3I5.js → rubric-predictive-validity-BsgkBfT0.js} +3 -3
  80. package/dist/{rubric-predictive-validity-B0Pg-3I5.js.map → rubric-predictive-validity-BsgkBfT0.js.map} +1 -1
  81. package/dist/search/index.js +2 -2
  82. package/dist/{search-kernel-Dpniloks.js → search-kernel-CsISc5t8.js} +5 -5
  83. package/dist/{search-kernel-Dpniloks.js.map → search-kernel-CsISc5t8.js.map} +1 -1
  84. package/dist/{search-summary-C1nAHgkB.js → search-summary-BZkoLwbB.js} +4 -4
  85. package/dist/{search-summary-C1nAHgkB.js.map → search-summary-BZkoLwbB.js.map} +1 -1
  86. package/dist/{semantic-concept-judge-BgWxa4L_.js → semantic-concept-judge-BbiYuv7U.js} +3 -3
  87. package/dist/{semantic-concept-judge-BgWxa4L_.js.map → semantic-concept-judge-BbiYuv7U.js.map} +1 -1
  88. package/dist/{sequential-DWn_np6P.js → sequential-7uxpO0Vm.js} +2 -2
  89. package/dist/{sequential-DWn_np6P.js.map → sequential-7uxpO0Vm.js.map} +1 -1
  90. package/dist/{skillopt-optimization-method-Dz5CGZb8.js → skillopt-optimization-method-bEm4jYV4.js} +4 -4
  91. package/dist/{skillopt-optimization-method-Dz5CGZb8.js.map → skillopt-optimization-method-bEm4jYV4.js.map} +1 -1
  92. package/dist/{statistical-heldout-CNLOvRq_.js → statistical-heldout-C26hOFWD.js} +3 -3
  93. package/dist/{statistical-heldout-CNLOvRq_.js.map → statistical-heldout-C26hOFWD.js.map} +1 -1
  94. package/dist/{store-otlp-yd9lDziN.js → store-otlp-BA7G1Du-.js} +44 -11
  95. package/dist/store-otlp-BA7G1Du-.js.map +1 -0
  96. package/dist/{store-tool-spans-CAF3T-BS.d.ts → store-tool-spans-BYIarVrG.d.ts} +17 -8
  97. package/dist/{store-tool-spans-CAF3T-BS.d.ts.map → store-tool-spans-BYIarVrG.d.ts.map} +1 -1
  98. package/dist/{store-tool-spans-D5kOU0mr.js → store-tool-spans-DKoVbE-w.js} +3 -3
  99. package/dist/{store-tool-spans-D5kOU0mr.js.map → store-tool-spans-DKoVbE-w.js.map} +1 -1
  100. package/dist/{student-t-BBZYka0E.js → student-t-BQeKJx2N.js} +2 -2
  101. package/dist/{student-t-BBZYka0E.js.map → student-t-BQeKJx2N.js.map} +1 -1
  102. package/dist/{summary-report-DLzptLSA.js → summary-report-BG6-hOkm.js} +6 -6
  103. package/dist/{summary-report-DLzptLSA.js.map → summary-report-BG6-hOkm.js.map} +1 -1
  104. package/dist/supervisor-run/index.js +1 -1
  105. package/dist/{tool-waste-CP5DO383.d.ts → tool-waste-DxlebQkx.d.ts} +42 -2
  106. package/dist/tool-waste-DxlebQkx.d.ts.map +1 -0
  107. package/dist/{tool-waste-B-u3rzCz.js → tool-waste-x_E8h9Ih.js} +3 -3
  108. package/dist/{tool-waste-B-u3rzCz.js.map → tool-waste-x_E8h9Ih.js.map} +1 -1
  109. package/dist/trace-repair/index.js +1 -1
  110. package/dist/traces.d.ts +2 -2
  111. package/dist/traces.js +4 -4
  112. package/dist/{tree-D48IJUK0.js → tree-CAQflhLR.js} +5 -5
  113. package/dist/{tree-D48IJUK0.js.map → tree-CAQflhLR.js.map} +1 -1
  114. package/docs/concepts.md +21 -0
  115. package/docs/trace-analysis.md +10 -0
  116. package/package.json +1 -1
  117. package/dist/baseline-CVKsfWTg.js +0 -150
  118. package/dist/baseline-CVKsfWTg.js.map +0 -1
  119. package/dist/benchmark-command-6OBcB0sG.js.map +0 -1
  120. package/dist/internal-BOe3zmjQ.js.map +0 -1
  121. package/dist/judge-calibration-C5zgtiJT.js.map +0 -1
  122. package/dist/kind-factory-CdPPHtGZ.js.map +0 -1
  123. package/dist/matrix-DnlHn5x1.js.map +0 -1
  124. package/dist/paired-arms-DpS_h8ON.js.map +0 -1
  125. package/dist/store-otlp-yd9lDziN.js.map +0 -1
  126. package/dist/tool-waste-CP5DO383.d.ts.map +0 -1
package/dist/cli.js CHANGED
@@ -1,14 +1,14 @@
1
1
  #!/usr/bin/env node
2
- import { g as INSUFFICIENT_FROM, h as operatorYield } from "./search-kernel-Dpniloks.js";
2
+ import { g as INSUFFICIENT_FROM, h as operatorYield } from "./search-kernel-CsISc5t8.js";
3
3
  import { i as replaySearchLedgerText } from "./search-ledger-CXQPwdWz.js";
4
- import { t as renderSearchSummary } from "./search-summary-C1nAHgkB.js";
4
+ import { t as renderSearchSummary } from "./search-summary-BZkoLwbB.js";
5
5
  import { c as hostedTenantFromEnv, g as SearchRunKindSchema, r as shipSearchLedger, t as SearchShipConflictError, u as SEARCH_LEDGER_BATCH_MAX_BYTES } from "./search-shipper-DhhofkDX.js";
6
6
  import { i as runRolloutReleaseCli } from "./hf-dataset-BR2yhuNn.js";
7
7
  import { t as createChatClient } from "./chat-client-5LLl4Ctl.js";
8
- import { t as runAnalystBenchmarkCommand } from "./benchmark-command-6OBcB0sG.js";
8
+ import { t as runAnalystBenchmarkCommand } from "./benchmark-command-BQhi_BVO.js";
9
9
  import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-CoBZO1fj.js";
10
- import { A as editCreditText, E as front, f as renderMetaSearchText, h as landscape, k as editCredit, l as metaSearch, m as formatLandscape, n as taskMatrix, o as skillManifold, r as formatSkillManifold, t as tree, x as surfaceTextEdits } from "./tree-D48IJUK0.js";
11
- import { t as runSupervisorRunCommand } from "./report-command-D0XL2Cvi.js";
10
+ import { A as editCreditText, E as front, f as renderMetaSearchText, h as landscape, k as editCredit, l as metaSearch, m as formatLandscape, n as taskMatrix, o as skillManifold, r as formatSkillManifold, t as tree, x as surfaceTextEdits } from "./tree-CAQflhLR.js";
11
+ import { t as runSupervisorRunCommand } from "./report-command-JrjiWReS.js";
12
12
  import { createHash } from "node:crypto";
13
13
  import { existsSync, readFileSync, writeFileSync } from "node:fs";
14
14
  import { dirname, join } from "node:path";
@@ -1,23 +1,23 @@
1
1
  import { a as hashCanonical } from "../canonical-BsZVgBhW.js";
2
- import { r as pairedBootstrap } from "../paired-tests-Ccr16gcW.js";
3
- import { t as decidePairedPromotion } from "../paired-promotion-decision-DhrmjdV8.js";
4
- import { a as heldoutSignificance } from "../statistical-heldout-CNLOvRq_.js";
5
- import { n as summarizeExecution, t as analyzeRuns } from "../analyze-runs-BrIibNd0.js";
2
+ import { r as pairedBootstrap } from "../paired-tests-jbwPdSz5.js";
3
+ import { t as decidePairedPromotion } from "../paired-promotion-decision-CO0kA0bq.js";
4
+ import { a as heldoutSignificance } from "../statistical-heldout-C26hOFWD.js";
5
+ import { n as summarizeExecution, t as analyzeRuns } from "../analyze-runs-d5dcMlxi.js";
6
6
  import { a as parseRunRecordSafe } from "../run-record-BRlM99rX.js";
7
- import { K as defaultProductionGate, a as transientDispatchFailure, c as compareOptimizationMethods, h as runImprovementLoop, nt as runEval, rt as runCampaign, t as llmJudge } from "../llm-judge-DyZdrp6I.js";
7
+ import { K as defaultProductionGate, a as transientDispatchFailure, c as compareOptimizationMethods, h as runImprovementLoop, nt as runEval, rt as runCampaign, t as llmJudge } from "../llm-judge-Bka5q-ds.js";
8
8
  import { i as isModelPriced, n as estimateCost } from "../metrics-DOMpkoeb.js";
9
9
  import { i as CostLedger } from "../cost-ledger-_IbkejCF.js";
10
- import { q as campaignSplitDigest } from "../campaign-evidence-Dt_rWPk2.js";
10
+ import { q as campaignSplitDigest } from "../campaign-evidence-C__cUpCt.js";
11
11
  import { h as mapConcurrentRange } from "../journal-DvKCvjty.js";
12
12
  import { C as inMemoryCampaignStorage, S as fsCampaignStorage } from "../external-optimizer-subprocess-BmKtH3T6.js";
13
- import { n as SelfImproveRunError, r as selfImprove, t as defineAgentEval } from "../define-agent-eval-B25Cc3ki.js";
14
- import { t as buildDefaultAnalystRegistry } from "../default-registry-wInsQWlQ.js";
13
+ import { n as SelfImproveRunError, r as selfImprove, t as defineAgentEval } from "../define-agent-eval-CeYrXpl7.js";
14
+ import { t as buildDefaultAnalystRegistry } from "../default-registry-BthquIoC.js";
15
15
  import { t as extractUsage } from "../extract-usage-BrnzRJ6H.js";
16
16
  import { i as makeProposalFinding } from "../types-CW5b90JH.js";
17
17
  import { t as createChatClient } from "../chat-client-5LLl4Ctl.js";
18
18
  import { n as InMemoryOutcomeStore, t as FileSystemOutcomeStore } from "../outcome-store-DUcMj-3D.js";
19
- import { c as heldOutGate, g as runReferenceEquivalenceJudge, h as createReferenceEquivalenceJudge, l as composeGate, m as REFERENCE_EQUIVALENCE_JUDGE_VERSION, n as gepaOptimizationMethod, p as REFERENCE_EQUIVALENCE_INPUT_LIMITS, t as skillOptOptimizationMethod, u as externalTextOptimizationMethod } from "../skillopt-optimization-method-Dz5CGZb8.js";
20
- import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-BI1RxzwD.js";
19
+ import { c as heldOutGate, g as runReferenceEquivalenceJudge, h as createReferenceEquivalenceJudge, l as composeGate, m as REFERENCE_EQUIVALENCE_JUDGE_VERSION, n as gepaOptimizationMethod, p as REFERENCE_EQUIVALENCE_INPUT_LIMITS, t as skillOptOptimizationMethod, u as externalTextOptimizationMethod } from "../skillopt-optimization-method-bEm4jYV4.js";
20
+ import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-BSbTAMtE.js";
21
21
  import { t as fromOtelSpans } from "../otel-spans-C8YIJWy1.js";
22
22
  import { createHash } from "node:crypto";
23
23
  import { agentCandidateBenchmarkSuiteSchema, agentCandidateBenchmarkTaskSchema, agentCandidateBundleSchema, agentCandidateEvaluationPolicySchema, agentCandidateExperimentSchema, agentImprovementMeasuredComparisonSchema, agentProfileImprovementExperimentSchema, agentProfileImprovementMeasuredComparisonSchema, agentProfileImprovementRunCellSchema, agentProfileImprovementRunReceiptSchema, agentProfileImprovementSuiteInputsSchema, agentProfileImprovementSuiteSchema, agentProfileImprovementTaskSchema, candidateExecutionEvidenceSchema, canonicalCandidateDigest, canonicalCandidateJson, numbersApproximatelyEqual, omitTopLevelDigest } from "@tangle-network/agent-interface";
@@ -1,7 +1,7 @@
1
1
  import { a as hashCanonical, r as canonicalString } from "./canonical-BsZVgBhW.js";
2
2
  import { t as deepFreezeCanonicalJson } from "./deep-freeze-CurcOQg3.js";
3
3
  import { C as combineAbortSignals } from "./llm-client-DmCNzrc9.js";
4
- import { A as spanEpochMillis, H as findingSubjectGrammarPromptFor, J as snapshotExactExecutionPlan, q as snapshotExactExecutionComponentIdentity, t as createTraceAnalyst } from "./kind-factory-CdPPHtGZ.js";
4
+ import { J as snapshotExactExecutionComponentIdentity, U as findingSubjectGrammarPromptFor, Y as snapshotExactExecutionPlan, j as spanEpochMillis, t as createTraceAnalyst } from "./kind-factory-Cn1lbsAd.js";
5
5
  import { i as validateUsageSettlementTimeout, t as assertValidAnalystUsageReceipt } from "./usage-receipt-Cx3xNeyB.js";
6
6
  import { LLM_CONTEXT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_OUTPUT_TOKEN_ATTR_KEYS, TOOL_NAME_ATTR_KEYS } from "./trace-attributes.js";
7
7
  import "./otlp-attributes-D9ptWaIJ.js";
@@ -2366,4 +2366,4 @@ function buildDefaultAnalystRegistry(options = {}) {
2366
2366
  //#endregion
2367
2367
  export { completedAnalystReviewQuality as _, KNOWLEDGE_POISONING_KIND_SPEC as a, validateAnalystReviewDecisions as b, FAILURE_MODE_KIND_SPEC as c, emitControlIntegrityFindings as d, behavioralAnalyst as f, assertUniqueFindingIds as g, analystRunDigest as h, DEFAULT_TRACE_ANALYST_KINDS as i, CONTROL_INTEGRITY_ANALYST as l, analystFindingDigest as m, AnalystRegistry as n, KNOWLEDGE_GAP_KIND_SPEC as o, deriveEfficiencyFindings as p, ExactAnalystRunExecutionError as r, IMPROVEMENT_KIND_SPEC as s, buildDefaultAnalystRegistry as t, ControlIntegrityAnalyst as u, readAnalystReview as v, snapshotAnalystRun as y };
2368
2368
 
2369
- //# sourceMappingURL=default-registry-wInsQWlQ.js.map
2369
+ //# sourceMappingURL=default-registry-BthquIoC.js.map