@tangle-network/agent-eval 0.184.0 → 0.185.0

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 (107) hide show
  1. package/CHANGELOG.md +12 -0
  2. package/dist/adapters/http.d.ts +1 -1
  3. package/dist/analyst/index.d.ts +6 -331
  4. package/dist/analyst/index.d.ts.map +1 -1
  5. package/dist/analyst/index.js +7 -6
  6. package/dist/analyst/index.js.map +1 -1
  7. package/dist/analyze-runs-CrYhqXop.d.ts +75 -0
  8. package/dist/analyze-runs-CrYhqXop.d.ts.map +1 -0
  9. package/dist/{define-agent-eval-RTekxhUE.js → analyze-runs-sTygaDyD.js} +4 -811
  10. package/dist/analyze-runs-sTygaDyD.js.map +1 -0
  11. package/dist/{benchmark-command-DlSePljH.js → benchmark-command-kvq7yFY4.js} +9 -47
  12. package/dist/benchmark-command-kvq7yFY4.js.map +1 -0
  13. package/dist/benchmarks/index.js +2 -2
  14. package/dist/campaign/index.js +7 -6
  15. package/dist/{campaign-D6puiGFO.js → campaign-eJgtRs-W.js} +8 -7
  16. package/dist/{campaign-D6puiGFO.js.map → campaign-eJgtRs-W.js.map} +1 -1
  17. package/dist/{campaign-evidence-aWEmfROB.js → campaign-evidence-BsrCmBSE.js} +4 -255
  18. package/dist/campaign-evidence-BsrCmBSE.js.map +1 -0
  19. package/dist/cli.js +1 -1
  20. package/dist/contract/index.d.ts +3 -3
  21. package/dist/contract/index.js +10 -279
  22. package/dist/contract/index.js.map +1 -1
  23. package/dist/{default-registry-DYB4aHVn.d.ts → default-registry-CXv9kHMt.d.ts} +3 -3
  24. package/dist/{default-registry-DYB4aHVn.d.ts.map → default-registry-CXv9kHMt.d.ts.map} +1 -1
  25. package/dist/{default-registry-CpLko4jB.js → default-registry-xiHrldW8.js} +3 -2
  26. package/dist/{default-registry-CpLko4jB.js.map → default-registry-xiHrldW8.js.map} +1 -1
  27. package/dist/define-agent-eval-D4uumjl3.js +813 -0
  28. package/dist/define-agent-eval-D4uumjl3.js.map +1 -0
  29. package/dist/diagnosis/index.d.ts +319 -0
  30. package/dist/diagnosis/index.d.ts.map +1 -0
  31. package/dist/diagnosis/index.js +1328 -0
  32. package/dist/diagnosis/index.js.map +1 -0
  33. package/dist/{dspy-rlm-engine-Da-l3CpD.js → dspy-rlm-engine-BE9lCzzL.js} +2 -2
  34. package/dist/{dspy-rlm-engine-Da-l3CpD.js.map → dspy-rlm-engine-BE9lCzzL.js.map} +1 -1
  35. package/dist/{engine-D1ykrAKR.d.ts → engine-cXUK-WHS.d.ts} +3 -75
  36. package/dist/engine-cXUK-WHS.d.ts.map +1 -0
  37. package/dist/equal-terms-Blx9Zk5I.js +25 -0
  38. package/dist/equal-terms-Blx9Zk5I.js.map +1 -0
  39. package/dist/experiment/index.js +4 -3
  40. package/dist/experiment/index.js.map +1 -1
  41. package/dist/{index-58wex7KV.d.ts → index-BQWfAFeH.d.ts} +3 -3
  42. package/dist/{index-58wex7KV.d.ts.map → index-BQWfAFeH.d.ts.map} +1 -1
  43. package/dist/{index-DVrtLO5j.d.ts → index-C8_iU8KL.d.ts} +3 -3
  44. package/dist/{index-DVrtLO5j.d.ts.map → index-C8_iU8KL.d.ts.map} +1 -1
  45. package/dist/index.d.ts +5 -4
  46. package/dist/index.d.ts.map +1 -1
  47. package/dist/index.js +12 -11
  48. package/dist/index.js.map +1 -1
  49. package/dist/{kind-factory-DE088Zhv.js → kind-factory-V79Hw0AU.js} +4 -82
  50. package/dist/kind-factory-V79Hw0AU.js.map +1 -0
  51. package/dist/{llm-judge-C5wbADA5.js → llm-judge-FWWhEb31.js} +5 -3
  52. package/dist/{llm-judge-C5wbADA5.js.map → llm-judge-FWWhEb31.js.map} +1 -1
  53. package/dist/openapi.json +1 -1
  54. package/dist/otel-spans-CQY1pNiN.js +275 -0
  55. package/dist/otel-spans-CQY1pNiN.js.map +1 -0
  56. package/dist/otlp-attributes-BpC_blLF.js +83 -0
  57. package/dist/otlp-attributes-BpC_blLF.js.map +1 -0
  58. package/dist/{run-record-BhMVsHR1.js → paired-promotion-decision-ClJh_9nK.js} +2 -215
  59. package/dist/paired-promotion-decision-ClJh_9nK.js.map +1 -0
  60. package/dist/prime-bridge-transport-Z3cpKGoV.js +43 -0
  61. package/dist/prime-bridge-transport-Z3cpKGoV.js.map +1 -0
  62. package/dist/prime-protocol-CDBcL9NM.d.ts +331 -0
  63. package/dist/prime-protocol-CDBcL9NM.d.ts.map +1 -0
  64. package/dist/{prime-protocol-6tZTVsWm.js → prime-protocol-DcAK9pC8.js} +2 -23
  65. package/dist/prime-protocol-DcAK9pC8.js.map +1 -0
  66. package/dist/{produced-state-D3uWsY8t.js → produced-state-CmB02tr2.js} +3 -3
  67. package/dist/{produced-state-D3uWsY8t.js.map → produced-state-CmB02tr2.js.map} +1 -1
  68. package/dist/{promotion-policy-CD-bUeyR.js → promotion-policy-DdrYWHDh.js} +3 -3
  69. package/dist/{promotion-policy-CD-bUeyR.js.map → promotion-policy-DdrYWHDh.js.map} +1 -1
  70. package/dist/rl.js +2 -1
  71. package/dist/rl.js.map +1 -1
  72. package/dist/run-record-CNNzWGpB.js +216 -0
  73. package/dist/run-record-CNNzWGpB.js.map +1 -0
  74. package/dist/{semantic-concept-judge-5r9hDEBA.js → semantic-concept-judge-Cnt0TWp0.js} +3 -3
  75. package/dist/{semantic-concept-judge-5r9hDEBA.js.map → semantic-concept-judge-Cnt0TWp0.js.map} +1 -1
  76. package/dist/{sequential-BfDkCBGo.js → sequential-PgLEqcTJ.js} +2 -2
  77. package/dist/{sequential-BfDkCBGo.js.map → sequential-PgLEqcTJ.js.map} +1 -1
  78. package/dist/{skillopt-optimization-method-B2zxoDk0.js → skillopt-optimization-method-FSk8b5FQ.js} +4 -3
  79. package/dist/{skillopt-optimization-method-B2zxoDk0.js.map → skillopt-optimization-method-FSk8b5FQ.js.map} +1 -1
  80. package/dist/statistical-heldout-CeWT_W7y.js +256 -0
  81. package/dist/statistical-heldout-CeWT_W7y.js.map +1 -0
  82. package/dist/{store-otlp-rU5PFlOB.js → store-otlp-2gYCy-v0.js} +2 -2
  83. package/dist/{store-otlp-rU5PFlOB.js.map → store-otlp-2gYCy-v0.js.map} +1 -1
  84. package/dist/{store-tool-spans-zrtftcRe.js → store-tool-spans-B1e6Auc2.js} +4 -3
  85. package/dist/{store-tool-spans-zrtftcRe.js.map → store-tool-spans-B1e6Auc2.js.map} +1 -1
  86. package/dist/{store-tool-spans-DwdokbHg.d.ts → store-tool-spans-BrdfPaVh.d.ts} +3 -3
  87. package/dist/{store-tool-spans-DwdokbHg.d.ts.map → store-tool-spans-BrdfPaVh.d.ts.map} +1 -1
  88. package/dist/{task-failure-attributes-BOxjfbhr.js → task-failure-attributes-B34TiMBe.js} +2 -2
  89. package/dist/{task-failure-attributes-BOxjfbhr.js.map → task-failure-attributes-B34TiMBe.js.map} +1 -1
  90. package/dist/{tool-groups-qHMydbAH.d.ts → tool-groups-eOOgYGaM.d.ts} +2 -2
  91. package/dist/tool-groups-eOOgYGaM.d.ts.map +1 -0
  92. package/dist/trace-repair/index.d.ts +1 -1
  93. package/dist/trace-repair/index.js +2 -1
  94. package/dist/trace-repair/index.js.map +1 -1
  95. package/dist/traces.d.ts +2 -2
  96. package/dist/traces.js +5 -4
  97. package/dist/traces.js.map +1 -1
  98. package/docs/diagnosis.md +42 -0
  99. package/package.json +7 -1
  100. package/dist/benchmark-command-DlSePljH.js.map +0 -1
  101. package/dist/campaign-evidence-aWEmfROB.js.map +0 -1
  102. package/dist/define-agent-eval-RTekxhUE.js.map +0 -1
  103. package/dist/engine-D1ykrAKR.d.ts.map +0 -1
  104. package/dist/kind-factory-DE088Zhv.js.map +0 -1
  105. package/dist/prime-protocol-6tZTVsWm.js.map +0 -1
  106. package/dist/run-record-BhMVsHR1.js.map +0 -1
  107. package/dist/tool-groups-qHMydbAH.d.ts.map +0 -1
@@ -1,8 +1,8 @@
1
1
  import { D as TraceAnalysisStore, i as AnalystFinding, n as AnalystContext } from "./types-BjB2wZM9.js";
2
2
  import { l as ExactCapableAnalyst } from "./exact-types-BEHN7tnq.js";
3
3
  import { n as AnalystRegistry, r as AnalystRegistryOptions } from "./registry-D5Ly3GRV.js";
4
- import { a as TraceAnalystLimits, i as TraceAnalysisEngineResult, l as RawAnalystFinding, n as TraceAnalysisEngine } from "./engine-D1ykrAKR.js";
5
- import { t as TraceToolGroupName } from "./tool-groups-qHMydbAH.js";
4
+ import { a as TraceAnalystLimits, i as TraceAnalysisEngineResult, l as RawAnalystFinding, n as TraceAnalysisEngine } from "./engine-cXUK-WHS.js";
5
+ import { t as TraceToolGroupName } from "./tool-groups-eOOgYGaM.js";
6
6
  //#region src/trace-analyst/behavioral-metrics.d.ts
7
7
  type SuboptimalCode = 'monotonic-input-growth' | 'output-length-decay' | 'single-tool-dependency' | 'no-self-verification';
8
8
  interface SuboptimalSignal {
@@ -110,4 +110,4 @@ interface DefaultAnalystRegistryOptions {
110
110
  declare function buildDefaultAnalystRegistry(options?: DefaultAnalystRegistryOptions): AnalystRegistry;
111
111
  //#endregion
112
112
  export { createTraceAnalyst as a, BehavioralAnalystOptions as c, TraceAnalystDefinition as i, behavioralAnalyst as l, buildDefaultAnalystRegistry as n, renderPriorFindings as o, CreateTraceAnalystOptions as r, runTraceAnalyst as s, DefaultAnalystRegistryOptions as t, deriveEfficiencyFindings as u };
113
- //# sourceMappingURL=default-registry-DYB4aHVn.d.ts.map
113
+ //# sourceMappingURL=default-registry-CXv9kHMt.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"default-registry-DYB4aHVn.d.ts","names":[],"sources":["../src/trace-analyst/behavioral-metrics.ts","../src/analyst/behavioral-analyst.ts","../src/analyst/kind-factory.ts","../src/analyst/default-registry.ts"],"mappings":";;;;;;KAoBY;UAMK;EACf,MAAM;EACN;;EAEA;;EAEA,UAAU;;UAGK;;EAEf;EACA;;EAEA,gBAAgB;;EAEhB;EACA;EACA,eAAe;EACf;EACA;;EAEA;EACA;EACA,SAAS;;UAGM;EACf;EACA;EACA,sBAAsB;EACtB,uBAAuB;;;;UC1BR;;EAEf;;EAEA;;;;;;iBA+Dc,yBACd,SAAS,mBACT;EAAQ;EAAoB;IAC3B;;iBAkCa,kBACd,UAAS,2BACR,oBAAoB;;;;UCrHN;EACf;EACA;EACA;EACA;;EAEA,sBAAsB,SAAS;;EAE/B;;EAEA,WAAW;;EAEX,kBACE,OAAO,oBACP,SAAS,wCACe;EAC1B,SAAS,QAAQ;EACjB,eAAe,KAAK,mBAAmB,SAAS,mBAAmB;EACnE;EACA;;UAGe;EACf,QAAQ;;EAER;EACA;;;iBAIoB,gBAAgB;EACpC,YAAY;EACZ,QAAQ;EACR,OAAO;EACP,SAAS;EACT;IACE,QAAQ;;iBAgHI,mBACd,YAAY,wBACZ,SAAS,4BACR,oBAAoB;iBA2RP,oBAAoB,OAAO;;;UChc1B;;;;;;EAMf,SAAS;EACT,uBAAuB;;EAEvB;EACA,aAAa;EACb,WAAW;;iBAGG,4BACd,UAAS,gCACR"}
1
+ {"version":3,"file":"default-registry-CXv9kHMt.d.ts","names":[],"sources":["../src/trace-analyst/behavioral-metrics.ts","../src/analyst/behavioral-analyst.ts","../src/analyst/kind-factory.ts","../src/analyst/default-registry.ts"],"mappings":";;;;;;KAoBY;UAMK;EACf,MAAM;EACN;;EAEA;;EAEA,UAAU;;UAGK;;EAEf;EACA;;EAEA,gBAAgB;;EAEhB;EACA;EACA,eAAe;EACf;EACA;;EAEA;EACA;EACA,SAAS;;UAGM;EACf;EACA;EACA,sBAAsB;EACtB,uBAAuB;;;;UC1BR;;EAEf;;EAEA;;;;;;iBA+Dc,yBACd,SAAS,mBACT;EAAQ;EAAoB;IAC3B;;iBAkCa,kBACd,UAAS,2BACR,oBAAoB;;;;UCrHN;EACf;EACA;EACA;EACA;;EAEA,sBAAsB,SAAS;;EAE/B;;EAEA,WAAW;;EAEX,kBACE,OAAO,oBACP,SAAS,wCACe;EAC1B,SAAS,QAAQ;EACjB,eAAe,KAAK,mBAAmB,SAAS,mBAAmB;EACnE;EACA;;UAGe;EACf,QAAQ;;EAER;EACA;;;iBAIoB,gBAAgB;EACpC,YAAY;EACZ,QAAQ;EACR,OAAO;EACP,SAAS;EACT;IACE,QAAQ;;iBAgHI,mBACd,YAAY,wBACZ,SAAS,4BACR,oBAAoB;iBA2RP,oBAAoB,OAAO;;;UChc1B;;;;;;EAMf,SAAS;EACT,uBAAuB;;EAEvB;EACA,aAAa;EACb,WAAW;;iBAGG,4BACd,UAAS,gCACR"}
@@ -1,9 +1,10 @@
1
1
  import { a as hashCanonical, r as canonicalString } from "./canonical-BsZVgBhW.js";
2
2
  import { t as combineAbortSignals } from "./abort-signal-CtzAM_sJ.js";
3
3
  import { a as deepFreezeCanonicalJson, r as makeFinding } from "./types-C8VBbiw6.js";
4
- import { B as findingSubjectGrammarPromptFor, X as snapshotExactExecutionPlan, Y as snapshotExactExecutionComponentIdentity, k as spanEpochMillis, t as createTraceAnalyst } from "./kind-factory-DE088Zhv.js";
4
+ import { B as findingSubjectGrammarPromptFor, G as snapshotExactExecutionComponentIdentity, K as snapshotExactExecutionPlan, k as spanEpochMillis, t as createTraceAnalyst } from "./kind-factory-V79Hw0AU.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
+ import "./otlp-attributes-BpC_blLF.js";
7
8
  import { t as executionTrackByLane } from "./execution-tracks-CpgFPpS5.js";
8
9
  import { t as analyzeSupervisorRunIntegrity } from "./integrity-CgQce0FV.js";
9
10
  import { randomUUID } from "node:crypto";
@@ -2364,4 +2365,4 @@ function buildDefaultAnalystRegistry(options = {}) {
2364
2365
  //#endregion
2365
2366
  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 };
2366
2367
 
2367
- //# sourceMappingURL=default-registry-CpLko4jB.js.map
2368
+ //# sourceMappingURL=default-registry-xiHrldW8.js.map