@tangle-network/agent-eval 0.100.0 → 0.100.1

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 (88) hide show
  1. package/dist/adapters/http.d.ts +2 -2
  2. package/dist/adapters/langchain.d.ts +2 -2
  3. package/dist/adapters/otel.d.ts +4 -4
  4. package/dist/analyst/index.d.ts +8 -7
  5. package/dist/analyst/index.js +35 -27
  6. package/dist/analyst/index.js.map +1 -1
  7. package/dist/{analyze-runs-DtT6F_6T.d.ts → analyze-runs-BlJRBniC.d.ts} +3 -3
  8. package/dist/belief-state/index.d.ts +3 -3
  9. package/dist/benchmarks/index.d.ts +2 -2
  10. package/dist/campaign/index.d.ts +37 -13
  11. package/dist/campaign/index.js +76 -7
  12. package/dist/campaign/index.js.map +1 -1
  13. package/dist/{chunk-WMBLMTUE.js → chunk-2KTBHICD.js} +139 -16
  14. package/dist/chunk-2KTBHICD.js.map +1 -0
  15. package/dist/{chunk-IZCEK2HR.js → chunk-2MLIEQSN.js} +3 -2
  16. package/dist/{chunk-IZCEK2HR.js.map → chunk-2MLIEQSN.js.map} +1 -1
  17. package/dist/{chunk-OKQ2LAT7.js → chunk-4LWD6GC7.js} +7 -5
  18. package/dist/{chunk-OKQ2LAT7.js.map → chunk-4LWD6GC7.js.map} +1 -1
  19. package/dist/{chunk-LO6IOIJ2.js → chunk-ABOIVNXL.js} +2 -240
  20. package/dist/chunk-ABOIVNXL.js.map +1 -0
  21. package/dist/{chunk-NZEQVRH5.js → chunk-BOETF6BU.js} +2 -2
  22. package/dist/{chunk-S4SYLDFX.js → chunk-FRI6RG3P.js} +3 -2
  23. package/dist/chunk-FRI6RG3P.js.map +1 -0
  24. package/dist/chunk-G6S73VA7.js +248 -0
  25. package/dist/chunk-G6S73VA7.js.map +1 -0
  26. package/dist/{chunk-GMGRBNVT.js → chunk-G7IB3GJ5.js} +2 -2
  27. package/dist/chunk-IN3SHQML.js +664 -0
  28. package/dist/chunk-IN3SHQML.js.map +1 -0
  29. package/dist/{chunk-SJISCGWD.js → chunk-JU6ZX3CX.js} +2 -2
  30. package/dist/{chunk-3NHEO6ZC.js → chunk-L5TVEZFT.js} +2 -2
  31. package/dist/{chunk-77T4STFI.js → chunk-PMF5WIBX.js} +3 -3
  32. package/dist/{chunk-OYU4D7FY.js → chunk-VWQ6PO5O.js} +2 -2
  33. package/dist/{code-agent-session-CPHRCb4-.d.ts → code-agent-session-B6ZcDwyA.d.ts} +1 -1
  34. package/dist/contract/index.d.ts +16 -16
  35. package/dist/contract/index.js +6 -5
  36. package/dist/contract/index.js.map +1 -1
  37. package/dist/{control-Doncu-B_.d.ts → control-DC8TELh0.d.ts} +1 -1
  38. package/dist/control.d.ts +2 -2
  39. package/dist/control.js +3 -2
  40. package/dist/{corpus-D4YW9UoJ.d.ts → corpus-ONOzGFmG.d.ts} +1 -1
  41. package/dist/{default-registry-GyE8X5SP.d.ts → default-registry-Dhrc__SE.d.ts} +2 -2
  42. package/dist/diagnose.d.ts +3 -3
  43. package/dist/diagnose.js +2 -1
  44. package/dist/diagnose.js.map +1 -1
  45. package/dist/{gepa-H6mlM0KN.d.ts → gepa-BRgNnmGZ.d.ts} +1 -1
  46. package/dist/hosted/index.d.ts +4 -4
  47. package/dist/{index-_Y4oNOOb.d.ts → index-W96macmS.d.ts} +1 -1
  48. package/dist/index.d.ts +22 -21
  49. package/dist/index.js +47 -21
  50. package/dist/index.js.map +1 -1
  51. package/dist/{insight-report-BnRjTibG.d.ts → insight-report-C02J3q4T.d.ts} +1 -1
  52. package/dist/{kind-factory-X3eDYbKn.d.ts → kind-factory-OgqQSvLi.d.ts} +1 -1
  53. package/dist/meta-eval/index.d.ts +2 -2
  54. package/dist/multishot/index.d.ts +2 -2
  55. package/dist/openapi.json +1 -1
  56. package/dist/policy-edit-Dccm9tyA.d.ts +103 -0
  57. package/dist/{pre-registration-CMm8cvrh.d.ts → pre-registration-DB8oDqZJ.d.ts} +3 -3
  58. package/dist/{provenance-Bg_RttR8.d.ts → provenance-B0SZw1z2.d.ts} +3 -3
  59. package/dist/{release-report-pidWUMZ2.d.ts → release-report-B1tA6pKu.d.ts} +2 -2
  60. package/dist/reporting.d.ts +4 -4
  61. package/dist/{researcher-Jr8ME1dZ.d.ts → researcher-Ba2y1Foi.d.ts} +2 -2
  62. package/dist/rl.d.ts +8 -8
  63. package/dist/rl.js +3 -2
  64. package/dist/rl.js.map +1 -1
  65. package/dist/{rubric-predictive-validity-C2hDKM8Z.d.ts → rubric-predictive-validity-w7tun-q3.d.ts} +1 -1
  66. package/dist/{run-record-CP2ObebC.d.ts → run-record-DEwidcqn.d.ts} +1 -1
  67. package/dist/{runtime-trajectory-BOUUjI0y.d.ts → runtime-trajectory-OJDaTYHN.d.ts} +1 -1
  68. package/dist/{semantic-concept-judge-DSBB2Cfp.d.ts → semantic-concept-judge-J8xvjdc3.d.ts} +2 -2
  69. package/dist/{summary-report-CInXwsza.d.ts → summary-report-C4uzRWh8.d.ts} +1 -1
  70. package/dist/traces.d.ts +1 -1
  71. package/dist/traces.js +4 -3
  72. package/dist/{types-B5x54y6n.d.ts → types-BEzCBMQD.d.ts} +2 -2
  73. package/dist/{types-BTI16iFl.d.ts → types-Cv1bo4_a.d.ts} +1 -1
  74. package/dist/workflow/index.d.ts +4 -4
  75. package/dist/workflow/index.js +2 -1
  76. package/dist/workflow/index.js.map +1 -1
  77. package/package.json +1 -1
  78. package/dist/chunk-BUTW4RGG.js +0 -32
  79. package/dist/chunk-BUTW4RGG.js.map +0 -1
  80. package/dist/chunk-LO6IOIJ2.js.map +0 -1
  81. package/dist/chunk-S4SYLDFX.js.map +0 -1
  82. package/dist/chunk-WMBLMTUE.js.map +0 -1
  83. /package/dist/{chunk-NZEQVRH5.js.map → chunk-BOETF6BU.js.map} +0 -0
  84. /package/dist/{chunk-GMGRBNVT.js.map → chunk-G7IB3GJ5.js.map} +0 -0
  85. /package/dist/{chunk-SJISCGWD.js.map → chunk-JU6ZX3CX.js.map} +0 -0
  86. /package/dist/{chunk-3NHEO6ZC.js.map → chunk-L5TVEZFT.js.map} +0 -0
  87. /package/dist/{chunk-77T4STFI.js.map → chunk-PMF5WIBX.js.map} +0 -0
  88. /package/dist/{chunk-OYU4D7FY.js.map → chunk-VWQ6PO5O.js.map} +0 -0
@@ -1,4 +1,4 @@
1
- import { R as RunRecord } from './run-record-CP2ObebC.js';
1
+ import { R as RunRecord } from './run-record-DEwidcqn.js';
2
2
  import { b as OutcomeStore } from './outcome-store-rnXLEqSn.js';
3
3
 
4
4
  /**
@@ -311,4 +311,4 @@ declare function parseRunRecordSafe(input: unknown): {
311
311
  /** Round-trip helper — `JSON.parse(JSON.stringify(record))` then validate. */
312
312
  declare function roundTripRunRecord(record: RunRecord): RunRecord;
313
313
 
314
- export { type AgentProfileCell as A, toAgentProfileJson as B, validateAgentProfileCell as C, validateRunRecord as D, verifyAgentProfileCell as E, type JudgeScoresRecord as J, type RunRecord as R, type RunSplitTag as a, type RunTokenUsage as b, type RunJudgeMetadata as c, type AgentProfileCellInput as d, AGENT_PROFILE_KINDS as e, type AgentInterfaceProfileLike as f, type AgentProfileCellSchemaVersion as g, AgentProfileCellValidationError as h, type AgentProfileDimensionValue as i, type AgentProfileHarness as j, type AgentProfileJson as k, type AgentProfileKind as l, type AgentProfileSource as m, type AgentProfileSourceInput as n, type RunOutcome as o, RunRecordValidationError as p, agentProfileCellHashMaterial as q, agentProfileCellKey as r, assertRunAgentProfileCell as s, buildAgentInterfaceProfileCell as t, buildAgentProfileCell as u, groupRunsByAgentProfileCell as v, isRunRecord as w, parseRunRecordSafe as x, requireAgentProfileCell as y, roundTripRunRecord as z };
314
+ export { type AgentProfileCell as A, toAgentProfileJson as B, validateAgentProfileCell as C, validateRunRecord as D, verifyAgentProfileCell as E, type JudgeScoresRecord as J, type RunRecord as R, type AgentProfileJson as a, type RunSplitTag as b, type RunTokenUsage as c, type RunJudgeMetadata as d, type AgentProfileCellInput as e, AGENT_PROFILE_KINDS as f, type AgentInterfaceProfileLike as g, type AgentProfileCellSchemaVersion as h, AgentProfileCellValidationError as i, type AgentProfileDimensionValue as j, type AgentProfileHarness as k, type AgentProfileKind as l, type AgentProfileSource as m, type AgentProfileSourceInput as n, type RunOutcome as o, RunRecordValidationError as p, agentProfileCellHashMaterial as q, agentProfileCellKey as r, assertRunAgentProfileCell as s, buildAgentInterfaceProfileCell as t, buildAgentProfileCell as u, groupRunsByAgentProfileCell as v, isRunRecord as w, parseRunRecordSafe as x, requireAgentProfileCell as y, roundTripRunRecord as z };
@@ -1,4 +1,4 @@
1
- import { a as RunSplitTag } from './run-record-CP2ObebC.js';
1
+ import { b as RunSplitTag } from './run-record-DEwidcqn.js';
2
2
 
3
3
  interface RuntimeTrajectoryHookEvent {
4
4
  id: string;
@@ -1,7 +1,7 @@
1
1
  import { AxAIService } from '@ax-llm/ax';
2
2
  import { z } from 'zod';
3
- import { c as AnalystFinding, A as Analyst, a as AnalystContext } from './types-B5x54y6n.js';
4
- import { T as TraceAnalystKindSpec } from './kind-factory-X3eDYbKn.js';
3
+ import { A as AnalystFinding, a as Analyst, b as AnalystContext } from './types-BEzCBMQD.js';
4
+ import { T as TraceAnalystKindSpec } from './kind-factory-OgqQSvLi.js';
5
5
  import { a as TraceAnalystSpan } from './store-C1YxJDEK.js';
6
6
  import { L as LlmClientOptions } from './llm-client-Bj7g0rqu.js';
7
7
  import { S as Severity } from './multi-layer-verifier-DUZXrPDA.js';
@@ -1,4 +1,4 @@
1
- import { R as RunRecord } from './run-record-CP2ObebC.js';
1
+ import { R as RunRecord } from './run-record-DEwidcqn.js';
2
2
  import { F as FailureClusterReport } from './failure-cluster-DH9Flgcf.js';
3
3
 
4
4
  /**
package/dist/traces.d.ts CHANGED
@@ -14,7 +14,7 @@ import { A as AnalyzeTracesOptions, b as AnalyzeTracesResult } from './analyst-C
14
14
  export { a as AnalyzeTracesInput, c as AnalyzeTracesTurnSnapshot, d as analyzeTraces } from './analyst-C8HHvfJp.js';
15
15
  import { h as TraceAnalystSpanKind, i as TraceAnalystSpanStatus, T as TraceAnalysisStore, g as TraceAnalystFilters, b as DatasetOverview, Q as QueryTracesPage, l as ViewTraceResult, V as ViewSpansResult, c as SearchTraceResult, S as SearchSpanResult } from './store-C1YxJDEK.js';
16
16
  export { D as DEFAULT_TRACE_ANALYST_BUDGETS, E as ErrorCluster, d as SpanMatchRecord, e as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, f as TraceAnalystByteBudgets, a as TraceAnalystSpan, j as TraceAnalystTraceSummary, k as ViewTraceOversized } from './store-C1YxJDEK.js';
17
- import { a as RunSplitTag, b as RunTokenUsage, R as RunRecord } from './run-record-CP2ObebC.js';
17
+ import { b as RunSplitTag, c as RunTokenUsage, R as RunRecord } from './run-record-DEwidcqn.js';
18
18
  import { AxFunction } from '@ax-llm/ax';
19
19
  import '@tangle-network/agent-interface';
20
20
 
package/dist/traces.js CHANGED
@@ -28,13 +28,13 @@ import {
28
28
  scoreTraceInsightReadiness,
29
29
  tokenizeDomainWords,
30
30
  traceAnalystOnRunComplete
31
- } from "./chunk-77T4STFI.js";
31
+ } from "./chunk-PMF5WIBX.js";
32
32
  import {
33
33
  TRACE_ANALYST_ACTOR_DESCRIPTION,
34
34
  TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
35
35
  TRACE_ANALYST_SUBAGENT_DESCRIPTION,
36
36
  analyzeTraces
37
- } from "./chunk-IZCEK2HR.js";
37
+ } from "./chunk-2MLIEQSN.js";
38
38
  import {
39
39
  DEFAULT_REDACTION_RULES,
40
40
  REDACTION_VERSION,
@@ -106,11 +106,12 @@ import {
106
106
  defaultProviderRedactor,
107
107
  providerFromBaseUrl
108
108
  } from "./chunk-PC4UYEBM.js";
109
- import "./chunk-LO6IOIJ2.js";
109
+ import "./chunk-G6S73VA7.js";
110
110
  import {
111
111
  TraceEmitter,
112
112
  llmSpanFromProvider
113
113
  } from "./chunk-TVVP3ZZQ.js";
114
+ import "./chunk-ABOIVNXL.js";
114
115
  import "./chunk-VSMTAMNK.js";
115
116
  import "./chunk-3BFEG2F6.js";
116
117
  import "./chunk-PZ5AY32C.js";
@@ -1,4 +1,4 @@
1
- import { R as RunRecord } from './run-record-CP2ObebC.js';
1
+ import { R as RunRecord } from './run-record-DEwidcqn.js';
2
2
  import { T as TraceAnalysisStore } from './store-C1YxJDEK.js';
3
3
  import { a as JudgeInput } from './types-C7DGg5ex.js';
4
4
  import { b as LlmCallRequest, c as LlmCallResult } from './llm-client-Bj7g0rqu.js';
@@ -338,4 +338,4 @@ type AnalystRunEvent = {
338
338
  result: AnalystRunResult;
339
339
  };
340
340
 
341
- export { type Analyst as A, type ChatClient as C, type DirectProviderTransportOpts as D, type EvidenceRef as E, type MockTransportOpts as M, type RouterTransportOpts as R, type SandboxSdkTransportOpts as S, type AnalystContext as a, type AnalystRunSummary as b, type AnalystFinding as c, type AnalystRunResult as d, type AnalystRunInputs as e, type AnalystRunEvent as f, type AnalystCost as g, type AnalystSeverity as h, type AnalystInputKind as i, type AnalystRequirements as j, type ChatCallOpts as k, type ChatRequest as l, type ChatResponse as m, type ChatTransport as n, type CliBridgeTransportOpts as o, type CreateChatClientOpts as p, computeFindingId as q, createChatClient as r, makeFinding as s };
341
+ export { type AnalystFinding as A, type ChatClient as C, type DirectProviderTransportOpts as D, type EvidenceRef as E, type MockTransportOpts as M, type RouterTransportOpts as R, type SandboxSdkTransportOpts as S, type Analyst as a, type AnalystContext as b, type AnalystRunSummary as c, type AnalystRunResult as d, type AnalystRunInputs as e, type AnalystRunEvent as f, type AnalystCost as g, type AnalystSeverity as h, type AnalystInputKind as i, type AnalystRequirements as j, type ChatCallOpts as k, type ChatRequest as l, type ChatResponse as m, type ChatTransport as n, type CliBridgeTransportOpts as o, type CreateChatClientOpts as p, computeFindingId as q, createChatClient as r, makeFinding as s };
@@ -1,4 +1,4 @@
1
- import { b as RunTokenUsage } from './run-record-CP2ObebC.js';
1
+ import { c as RunTokenUsage } from './run-record-DEwidcqn.js';
2
2
 
3
3
  /**
4
4
  * Pass A substrate types — `runCampaign` is the one primitive every
@@ -1,7 +1,7 @@
1
1
  import { W as WorkflowTopology } from '../harness-optimizer-mOl9XX_O.js';
2
- import { a as RunSplitTag, b as RunTokenUsage, R as RunRecord } from '../run-record-CP2ObebC.js';
3
- import { c as AnalystFinding, h as AnalystSeverity, E as EvidenceRef } from '../types-B5x54y6n.js';
4
- import { F as FailureClusterInsight } from '../insight-report-BnRjTibG.js';
2
+ import { b as RunSplitTag, c as RunTokenUsage, R as RunRecord } from '../run-record-DEwidcqn.js';
3
+ import { A as AnalystFinding, h as AnalystSeverity, E as EvidenceRef } from '../types-BEzCBMQD.js';
4
+ import { F as FailureClusterInsight } from '../insight-report-C02J3q4T.js';
5
5
  import { a as VerificationReport, L as LayerResult } from '../multi-layer-verifier-DUZXrPDA.js';
6
6
  import { F as FailureClusterReport } from '../failure-cluster-DH9Flgcf.js';
7
7
  import { R as RedactionRule, a as RedactionReport } from '../redact-B40YG2M_.js';
@@ -19,7 +19,7 @@ import '../types-C7DGg5ex.js';
19
19
  import '@tangle-network/tcloud';
20
20
  import '../llm-client-Bj7g0rqu.js';
21
21
  import '../raw-provider-sink-C46HDghv.js';
22
- import '../summary-report-CInXwsza.js';
22
+ import '../summary-report-C4uzRWh8.js';
23
23
  import '../judge-calibration-0p2QcWNE.js';
24
24
  import '../verdict-C9MlYujm.js';
25
25
  import '../control-runtime-Acf9CGhw.js';
@@ -7,7 +7,8 @@ import {
7
7
  } from "../chunk-GGE4NNQT.js";
8
8
  import {
9
9
  validateRunRecord
10
- } from "../chunk-LO6IOIJ2.js";
10
+ } from "../chunk-G6S73VA7.js";
11
+ import "../chunk-ABOIVNXL.js";
11
12
  import "../chunk-VSMTAMNK.js";
12
13
  import {
13
14
  ValidationError