@tangle-network/agent-eval 0.115.3 → 0.117.0

package/dist/{provenance-Bibyg1U9.d.ts → provenance-DpjwyseI.d.ts}

@@ -1,9 +1,10 @@
-import { S as Scenario, g as Gate, G as GateResult, n as GateContext, C as CampaignResult, p as Mutator, f as SurfaceProposer, M as MutableSurface, h as GateDecision } from './types-C5gJrOVT.js';
-import { e as RedTeamCase, D as Direction, b as RunCampaignOptions } from './gepa-dne9JDPL.js';
-import { R as RunRecord } from './run-record-B7RTi_ix.js';
-import { a as PairedBootstrapResult } from './statistics-oUbOJe-S.js';
+import { S as Scenario, G as Gate, m as GateResult, l as GateContext, C as CampaignResult, p as Mutator, c as SurfaceProposer, M as MutableSurface, n as GenerationCandidate, d as GateDecision } from './types-BSw1rOUB.js';
+import { n as RedTeamCase, D as Direction, h as RunCampaignOptions } from './gepa-eESocoDi.js';
+import { R as RunRecord } from './run-record-BDH49H2E.js';
+import { a as PairedBootstrapResult } from './statistics-KUnG73jH.js';
+import { P as PolicyEditCandidateRecord } from './policy-edit-wG9uFEFm.js';
 import { HostedClient, TraceSpanEvent } from './hosted/index.js';
-import { C as CampaignStorage } from './storage-Dw_f7WMt.js';
+import { C as CampaignStorage } from './storage-DrX3v_5B.js';
 
 /**
  * Compose multiple `Gate` implementations — every gate must pass for the
@@ -356,7 +357,7 @@ declare function evolutionaryProposer<TFindings = unknown>(opts: EvolutionaryPro
  * Two artifacts, one source of truth:
  *
  *   1. `LoopProvenanceRecord` — a structured JSON record capturing every
- *      candidate (surfaceHash + label + rationale), its measured composite,
+ *      candidate (surfaceHash + label + rationale + structured cause), its measured composite,
  *      the gate decision + reasons + delta, the held-out lift, the explicit
  *      baseline→candidate diff, and BACKEND PROVENANCE (the
  *      `assertRealBackend` verdict + worker call count + model). This is the
@@ -387,6 +388,18 @@ interface LoopProvenanceCandidate {
     /** Proposer rationale — the "because Z". When the proposer returned a bare
      *  surface (blind mutator) this is absent. */
     rationale?: string;
+    /** Exact validated cause when the proposer emitted a structured record. */
+    candidateRecord?: PolicyEditCandidateRecord;
+    /** Exact complete incumbent this candidate mutated. */
+    parentSurfaceHash: string;
+    /** Search-split composite of the exact parent. */
+    parentComposite: number;
+    /** Search-split composite change relative to the exact parent. */
+    observedDeltaFromParent?: number;
+    /** Whether the candidate completed every designed cell and could be selected. */
+    eligibleForPromotion: boolean;
+    /** Designed-denominator receipt retained even for incomplete candidates. */
+    coverage: NonNullable<GenerationCandidate['coverage']>;
     /** Mean composite this candidate scored on the search split. */
     composite: number;
     /** Whether this candidate was promoted out of its generation. */
@@ -409,7 +422,7 @@ interface LoopProvenanceBackend {
  * the bare hosted event) + backend provenance.
  */
 interface LoopProvenanceRecord {
-    schema: 'tangle.loop-provenance.v2';
+    schema: 'tangle.loop-provenance.v3';
     runId: string;
     runDir: string;
     timestamp: string;
@@ -421,8 +434,10 @@ interface LoopProvenanceRecord {
     winnerRationale?: string;
     /** The explicit baseline→winner unified diff the gate decided on. */
     diff: string;
-    /** Every candidate across every generation, each carrying its rationale. */
+    /** Every candidate across every generation, with its rationale and structured cause. */
     candidates: LoopProvenanceCandidate[];
+    /** Baseline composite on the search split that generated the candidates. */
+    baselineSearchComposite: number;
     /** The gate verdict — decision + reasons + contributing gates + delta. */
     gate: {
         decision: GateDecision;
@@ -453,18 +468,15 @@ interface BuildLoopProvenanceArgs<TArtifact, TScenario extends Scenario> {
     winnerLabel?: string;
     winnerRationale?: string;
     diff: string;
+    /** Baseline composite on the search split, distinct from holdout scoring. */
+    baselineSearchComposite: number;
     /** Per-generation candidate records straight off the loop result. */
     generations: Array<{
         generationIndex: number;
-        candidates: Array<{
-            surfaceHash: string;
-            composite: number;
-            label?: string;
-            rationale?: string;
-        }>;
+        candidates: GenerationCandidate[];
         promoted: string[];
-        /** Surfaces measured this generation, keyed positionally to candidates so
-         *  the content hash can be computed from the real surface text. */
+        /** Surfaces measured this generation, keyed by surface hash so the content
+         *  hash can be computed and the loop identity rechecked from real bytes. */
         surfaces: Array<{
             surfaceHash: string;
             surface: MutableSurface;

package/dist/{query-Ck190MOd.d.ts → query-CF7PG61p.d.ts}

@@ -1,5 +1,5 @@
-import { L as LlmSpan, J as JudgeSpan, R as Run, F as FailureClass, T as ToolSpan } from './schema-SGWcK9wa.js';
-import { T as TraceStore } from './store-BsVi7ncX.js';
+import { L as LlmSpan, T as ToolSpan, J as JudgeSpan, R as Run, F as FailureClass } from './schema-B3Q3l9Z_.js';
+import { T as TraceStore } from './store-DGqD0Pyo.js';
 
 /**
  * Typed query helpers over TraceStore.
@@ -19,6 +19,8 @@ declare function judgeSpans(store: TraceStore, runId?: string): Promise<JudgeSpa
 declare function groupBy<T, K extends string | number>(items: T[], key: (t: T) => K): Map<K, T[]>;
 /** Hash tool arguments to an orderless-key-stable string for de-duplication. */
 declare function argHash(args: unknown): string;
+/** Whether argument-based comparisons are valid for this tool call. */
+declare function hasCapturedToolArgs(span: ToolSpan): boolean;
 /** Sum an LLM-span array into aggregate token + cost. */
 declare function aggregateLlm(spans: LlmSpan[]): {
     inputTokens: number;
@@ -30,4 +32,4 @@ declare function aggregateLlm(spans: LlmSpan[]): {
 /** Pick the outcome's failure class when present, else derive 'success' from run status. */
 declare function runFailureClass(run: Run): FailureClass;
 
-export { aggregateLlm as a, argHash as b, runsForScenario as c, groupBy as g, judgeSpans as j, llmSpans as l, runFailureClass as r, toolSpans as t };
+export { aggregateLlm as a, argHash as b, runsForScenario as c, groupBy as g, hasCapturedToolArgs as h, judgeSpans as j, llmSpans as l, runFailureClass as r, toolSpans as t };

package/dist/{release-report-CCtzajxP.d.ts → release-report-C8G2i5Xi.d.ts}

@@ -1,6 +1,6 @@
 import { a as DatasetSplit, b as DatasetManifest, D as DatasetScenario } from './dataset-NENEzRgk.js';
-import { m as GateDecision } from './summary-report-BJ5aNwZ1.js';
-import { R as RunRecord, b as RunSplitTag } from './run-record-B7RTi_ix.js';
+import { m as GateDecision } from './summary-report-C5bKFfm-.js';
+import { R as RunRecord, a as RunSplitTag } from './run-record-BDH49H2E.js';
 
 /**
  * Release confidence gate.

package/dist/reporting.d.ts

@@ -1,16 +1,17 @@
-export { R as RubricOutcomePair, a as RubricPredictiveValidityInput, b as RubricPredictiveValidityReport, c as RubricRanking, r as rubricPredictiveValidity } from './rubric-predictive-validity-DYTLjGWu.js';
-export { B as BootstrapOptions, a as BootstrapResult, J as JudgeReplayGateArgs, R as ReleaseConfidenceAxis, b as ReleaseConfidenceAxisName, c as ReleaseConfidenceInput, d as ReleaseConfidenceIssue, e as ReleaseConfidenceMetrics, f as ReleaseConfidenceScorecard, g as ReleaseConfidenceStatus, h as ReleaseConfidenceThresholds, i as ReleaseTraceEvidence, j as RenderReleaseReportOptions, V as Verdict, k as assertReleaseConfidence, l as bootstrapCi, m as evaluateReleaseConfidence, n as judgeReplayGate, r as renderReleaseReport } from './release-report-CCtzajxP.js';
+export { R as RubricOutcomePair, a as RubricPredictiveValidityInput, b as RubricPredictiveValidityReport, c as RubricRanking, r as rubricPredictiveValidity } from './rubric-predictive-validity-p49lLVrE.js';
+export { B as BootstrapOptions, a as BootstrapResult, J as JudgeReplayGateArgs, R as ReleaseConfidenceAxis, b as ReleaseConfidenceAxisName, c as ReleaseConfidenceInput, d as ReleaseConfidenceIssue, e as ReleaseConfidenceMetrics, f as ReleaseConfidenceScorecard, g as ReleaseConfidenceStatus, h as ReleaseConfidenceThresholds, i as ReleaseTraceEvidence, j as RenderReleaseReportOptions, V as Verdict, k as assertReleaseConfidence, l as bootstrapCi, m as evaluateReleaseConfidence, n as judgeReplayGate, r as renderReleaseReport } from './release-report-C8G2i5Xi.js';
 export { I as InterimReleaseConfidence, a as InterimReleaseConfidenceInput, P as PairedEvalueOptions, b as PairedEvalueSequence, c as PairedEvalueStep, S as SequentialDecision, e as evaluateInterimReleaseConfidence, p as pairedEvalueSequence } from './sequential-5iSVfzl2.js';
-export { P as PairedBootstrapOptions, a as PairedBootstrapResult, b as benjaminiHochberg, p as pairedBootstrap, w as wilcoxonSignedRank } from './statistics-oUbOJe-S.js';
-export { G as GainDistributionBin, a as GainDistributionFigureSpec, b as GainDistributionOptions, P as ParetoFigureSpec, c as ParetoPoint, R as RESEARCH_REPORT_HARD_PAIR_FLOOR, d as ResearchReport, e as ResearchReportCandidate, f as ResearchReportDecision, g as ResearchReportMethodology, h as ResearchReportOptions, i as ResearchReportRecommendation, S as SummaryTable, j as SummaryTableOptions, k as SummaryTableRow, l as gainHistogram, p as paretoChart, r as researchReport, s as summaryTable } from './summary-report-BJ5aNwZ1.js';
-import './run-record-B7RTi_ix.js';
+export { P as PairedBootstrapOptions, a as PairedBootstrapResult, b as benjaminiHochberg, p as pairedBootstrap, w as wilcoxonSignedRank } from './statistics-KUnG73jH.js';
+export { G as GainDistributionBin, a as GainDistributionFigureSpec, b as GainDistributionOptions, P as ParetoFigureSpec, c as ParetoPoint, R as RESEARCH_REPORT_HARD_PAIR_FLOOR, d as ResearchReport, e as ResearchReportCandidate, f as ResearchReportDecision, g as ResearchReportMethodology, h as ResearchReportOptions, i as ResearchReportRecommendation, S as SummaryTable, j as SummaryTableOptions, k as SummaryTableRow, l as gainHistogram, p as paretoChart, r as researchReport, s as summaryTable } from './summary-report-C5bKFfm-.js';
+import './run-record-BDH49H2E.js';
 import '@tangle-network/agent-interface';
 import './errors-oeQrLqXC.js';
-import './schema-SGWcK9wa.js';
+import './schema-B3Q3l9Z_.js';
 import './outcome-store-rnXLEqSn.js';
 import './dataset-NENEzRgk.js';
 import './judge-calibration-7C-IDmKr.js';
-import './types-C7DGg5ex.js';
+import './types-BkfcQnxV.js';
+import './cost-ledger-DWy3XdJc.js';
 import '@tangle-network/tcloud';
-import './failure-cluster-C48PiReX.js';
-import './store-BsVi7ncX.js';
+import './failure-cluster-DOAcSJ87.js';
+import './store-DGqD0Pyo.js';

package/dist/{researcher-Dq-EtpbE.d.ts → researcher-C8XyxQsu.d.ts}

@@ -1,11 +1,11 @@
-import { b as RunSplitTag, d as RunTokenUsage, e as RunJudgeMetadata, J as JudgeScoresRecord, A as AgentProfileCell, f as AgentProfileCellInput, R as RunRecord } from './run-record-B7RTi_ix.js';
-import { L as LlmClientOptions, a as LlmRouteRequirements } from './llm-client-DyqEH4jH.js';
-import { h as ResearchReportOptions, d as ResearchReport, m as GateDecision } from './summary-report-BJ5aNwZ1.js';
-import { T as TraceEmitter, R as RunCompleteHook } from './emitter-BRchAAAx.js';
-import { R as RunIntegrityExpectations, a as RunIntegrityReport } from './integrity-qemeBAyx.js';
+import { a as RunSplitTag, c as RunTokenUsage, d as RunJudgeMetadata, J as JudgeScoresRecord, A as AgentProfileCell, e as AgentProfileCellInput, R as RunRecord } from './run-record-BDH49H2E.js';
+import { a as LlmClientOptions, b as LlmRouteRequirements } from './llm-client-qoDd18Qz.js';
+import { h as ResearchReportOptions, d as ResearchReport, m as GateDecision } from './summary-report-C5bKFfm-.js';
+import { T as TraceEmitter, R as RunCompleteHook } from './emitter-CjD7vUwv.js';
+import { R as RunIntegrityExpectations, a as RunIntegrityReport } from './integrity-DqlBiLyK.js';
 import { R as RawProviderSink } from './raw-provider-sink-C46HDghv.js';
-import { F as FailureClass } from './schema-SGWcK9wa.js';
-import { T as TraceStore } from './store-BsVi7ncX.js';
+import { F as FailureClass } from './schema-B3Q3l9Z_.js';
+import { T as TraceStore } from './store-DGqD0Pyo.js';
 
 /**
  * EvalCampaign — opinionated matrix runner that wires the four

package/dist/rl.d.ts

@@ -1,25 +1,30 @@
-import { R as RunRecord, b as RunSplitTag } from './run-record-B7RTi_ix.js';
+import { R as RunRecord, a as RunSplitTag } from './run-record-BDH49H2E.js';
 export { A as AdversarialMutation } from './adversarial-B7loGVVX.js';
-import { S as Span } from './schema-SGWcK9wa.js';
-import { T as TraceStore } from './store-BsVi7ncX.js';
+import { S as Span } from './schema-B3Q3l9Z_.js';
+import { T as TraceStore } from './store-DGqD0Pyo.js';
 export { O as OffPolicyEstimate, a as OffPolicyOptions, b as OffPolicyTrajectory, d as doublyRobust, i as inverseProbabilityWeighting, o as offPolicyEstimateAll, s as selfNormalizedImportanceWeighting } from './off-policy-DiwuKKg7.js';
 import { b as OutcomeStore } from './outcome-store-rnXLEqSn.js';
 export { D as DeploymentOutcome, F as FileSystemOutcomeStore, a as FileSystemOutcomeStoreOptions, I as InMemoryOutcomeStore } from './outcome-store-rnXLEqSn.js';
-import { b as RubricPredictiveValidityReport } from './rubric-predictive-validity-DYTLjGWu.js';
-import { R as Researcher, F as FailureMode, S as SteeringChange, E as ExperimentPlan, a as ExperimentResult, b as EvalCampaignResult, c as EvalCampaignOptions } from './researcher-Dq-EtpbE.js';
-export { r as runEvalCampaign } from './researcher-Dq-EtpbE.js';
+import { b as RubricPredictiveValidityReport } from './rubric-predictive-validity-p49lLVrE.js';
+import { R as Researcher, F as FailureMode, S as SteeringChange, E as ExperimentPlan, a as ExperimentResult, b as EvalCampaignResult, c as EvalCampaignOptions } from './researcher-C8XyxQsu.js';
+export { r as runEvalCampaign } from './researcher-C8XyxQsu.js';
 import { a as VerificationReport } from './multi-layer-verifier-BsqKuLyN.js';
 import { I as InterimReleaseConfidence } from './sequential-5iSVfzl2.js';
-import { C as CampaignResult } from './types-C5gJrOVT.js';
+import { C as CampaignResult } from './types-BSw1rOUB.js';
 import '@tangle-network/agent-interface';
 import './errors-oeQrLqXC.js';
-import './llm-client-DyqEH4jH.js';
+import './llm-client-qoDd18Qz.js';
+import './cost-ledger-DWy3XdJc.js';
 import './raw-provider-sink-C46HDghv.js';
-import './summary-report-BJ5aNwZ1.js';
-import './failure-cluster-C48PiReX.js';
-import './emitter-BRchAAAx.js';
-import './integrity-qemeBAyx.js';
+import './summary-report-C5bKFfm-.js';
+import './failure-cluster-DOAcSJ87.js';
+import './emitter-CjD7vUwv.js';
+import './integrity-DqlBiLyK.js';
 import './verdict-C9MlYujm.js';
+import './policy-edit-wG9uFEFm.js';
+import './store-C1YxJDEK.js';
+import './types-BkfcQnxV.js';
+import '@tangle-network/tcloud';
 
 /**
  * Adaptive curriculum / active scenario selection.

package/dist/rl.js

@@ -10,7 +10,7 @@ import {
 } from "./chunk-3RF76KTD.js";
 import {
   runEvalCampaign
-} from "./chunk-ONM6PEAE.js";
+} from "./chunk-GQCZRZ7L.js";
 import {
   detectRewardHacking,
   extractVerifiableReward,
@@ -38,7 +38,7 @@ import "./chunk-TVVP3ZZQ.js";
 import "./chunk-5UF54T55.js";
 import "./chunk-XJYR7XFV.js";
 import "./chunk-VSMTAMNK.js";
-import "./chunk-GY4SYVPJ.js";
+import "./chunk-NJC7U437.js";
 import "./chunk-PC4UYEBM.js";
 import {
   ValidationError

package/dist/{rubric-predictive-validity-DYTLjGWu.d.ts → rubric-predictive-validity-p49lLVrE.d.ts}

@@ -1,4 +1,4 @@
-import { R as RunRecord } from './run-record-B7RTi_ix.js';
+import { R as RunRecord } from './run-record-BDH49H2E.js';
 import { b as OutcomeStore } from './outcome-store-rnXLEqSn.js';
 
 /**

package/dist/{run-campaign-UADIM77S.js → run-campaign-IM26A6PD.js}

@@ -1,12 +1,14 @@
 import {
   planCampaignRun,
   runCampaign
-} from "./chunk-FAOEFFRT.js";
+} from "./chunk-IDZTTFRR.js";
 import "./chunk-PJQFMIOX.js";
+import "./chunk-VCTY3W6J.js";
+import "./chunk-VI2UW6B6.js";
 import "./chunk-ONWEPEDO.js";
 import "./chunk-PZ5AY32C.js";
 export {
   planCampaignRun,
   runCampaign
 };
-//# sourceMappingURL=run-campaign-UADIM77S.js.map
+//# sourceMappingURL=run-campaign-IM26A6PD.js.map

package/dist/{run-record-B7RTi_ix.d.ts → run-record-BDH49H2E.d.ts}

@@ -1,6 +1,6 @@
 import { AgentProfile } from '@tangle-network/agent-interface';
 import { V as ValidationError } from './errors-oeQrLqXC.js';
-import { F as FailureClass } from './schema-SGWcK9wa.js';
+import { F as FailureClass } from './schema-B3Q3l9Z_.js';
 
 type AgentProfileCellSchemaVersion = 'agent-profile-cell/v1';
 type AgentProfileJson = string | number | boolean | null | AgentProfileJson[] | {
@@ -357,4 +357,4 @@ declare function roundTripRunRecord(record: RunRecord): RunRecord;
  */
 declare function modelHasSnapshot(model: string): boolean;
 
-export { type AgentProfileCell as A, requireAgentProfileCell as B, resolveRunCostProvenance as C, roundTripRunRecord as D, toAgentProfileJson as E, validateAgentProfileCell as F, validateRunRecord as G, verifyAgentProfileCell as H, type JudgeScoresRecord as J, type RunRecord as R, type AgentProfileJson as a, type RunSplitTag as b, type RunCostProvenance as c, type RunTokenUsage as d, type RunJudgeMetadata as e, type AgentProfileCellInput as f, AGENT_PROFILE_KINDS as g, type AgentInterfaceProfileLike as h, type AgentProfileCellSchemaVersion as i, AgentProfileCellValidationError as j, type AgentProfileDimensionValue as k, type AgentProfileHarness as l, type AgentProfileKind as m, type AgentProfileSource as n, type AgentProfileSourceInput as o, type RunOutcome as p, RunRecordValidationError as q, agentProfileCellHashMaterial as r, agentProfileCellKey as s, assertRunAgentProfileCell as t, buildAgentInterfaceProfileCell as u, buildAgentProfileCell as v, groupRunsByAgentProfileCell as w, isRunRecord as x, modelHasSnapshot as y, parseRunRecordSafe as z };
+export { type AgentProfileCell as A, requireAgentProfileCell as B, resolveRunCostProvenance as C, roundTripRunRecord as D, toAgentProfileJson as E, validateAgentProfileCell as F, validateRunRecord as G, verifyAgentProfileCell as H, type JudgeScoresRecord as J, type RunRecord as R, type RunSplitTag as a, type RunCostProvenance as b, type RunTokenUsage as c, type RunJudgeMetadata as d, type AgentProfileCellInput as e, type AgentProfileJson as f, AGENT_PROFILE_KINDS as g, type AgentInterfaceProfileLike as h, type AgentProfileCellSchemaVersion as i, AgentProfileCellValidationError as j, type AgentProfileDimensionValue as k, type AgentProfileHarness as l, type AgentProfileKind as m, type AgentProfileSource as n, type AgentProfileSourceInput as o, type RunOutcome as p, RunRecordValidationError as q, agentProfileCellHashMaterial as r, agentProfileCellKey as s, assertRunAgentProfileCell as t, buildAgentInterfaceProfileCell as u, buildAgentProfileCell as v, groupRunsByAgentProfileCell as w, isRunRecord as x, modelHasSnapshot as y, parseRunRecordSafe as z };

package/dist/{runtime-trajectory-Dws7Kpgi.d.ts → runtime-trajectory-DGBIUt4B.d.ts}

@@ -1,4 +1,4 @@
-import { b as RunSplitTag } from './run-record-B7RTi_ix.js';
+import { a as RunSplitTag } from './run-record-BDH49H2E.js';
 
 interface RuntimeTrajectoryHookEvent {
     id: string;

package/dist/{schema-SGWcK9wa.d.ts → schema-B3Q3l9Z_.d.ts}

@@ -116,6 +116,8 @@ interface ToolSpan extends SpanBase {
     kind: 'tool';
     toolName: string;
     args: unknown;
+    /** False when the source observed the call but did not capture its arguments. */
+    argsCaptured?: boolean;
     result?: unknown;
     latencyMs?: number;
 }

package/dist/{semantic-concept-judge-DxJmRkyJ.d.ts → semantic-concept-judge-CXnPEJbf.d.ts}

@@ -1,10 +1,12 @@
 import { AxAIService } from '@ax-llm/ax';
 import { z } from 'zod';
-import { A as AnalystFinding, T as TraceAnalystKindSpec, a as Analyst, b as AnalystContext } from './kind-factory-DcNg13sZ.js';
+import { c as AnalystFinding, A as Analyst, a as AnalystContext } from './policy-edit-wG9uFEFm.js';
+import { T as TraceAnalystKindSpec } from './kind-factory-ClZmO25A.js';
 import { a as TraceAnalystSpan } from './store-C1YxJDEK.js';
-import { R as Run, S as Span, e as TraceEvent, A as Artifact, B as BudgetLedgerEntry } from './schema-SGWcK9wa.js';
-import { T as TraceStore } from './store-BsVi7ncX.js';
-import { L as LlmClientOptions } from './llm-client-DyqEH4jH.js';
+import { R as Run, S as Span, e as TraceEvent, A as Artifact, B as BudgetLedgerEntry } from './schema-B3Q3l9Z_.js';
+import { T as TraceStore } from './store-DGqD0Pyo.js';
+import { C as CostLedger } from './cost-ledger-DWy3XdJc.js';
+import { a as LlmClientOptions } from './llm-client-qoDd18Qz.js';
 import { S as Severity } from './multi-layer-verifier-BsqKuLyN.js';
 
 interface CreateAnalystAiConfig {
@@ -499,7 +501,12 @@ interface SuboptimalSignal {
     evidence: Record<string, number | string | boolean>;
 }
 interface BehavioralMetrics {
+    /** The only trace represented by these metrics; null when spans are empty. */
+    traceId: string | null;
     llmCallCount: number;
+    /** Causally serial LLM timelines. Parallel branches are never joined. */
+    tokenSequences: BehavioralTokenSequence[];
+    /** Token values from the longest serial timeline, retained for convenience. */
     inputTokenTrajectory: number[];
     outputTokenTrajectory: number[];
     toolHistogram: Record<string, number>;
@@ -510,6 +517,12 @@ interface BehavioralMetrics {
     hasSelfVerification: boolean;
     signals: SuboptimalSignal[];
 }
+interface BehavioralTokenSequence {
+    scopeId: string;
+    spanIds: string[];
+    inputTokenTrajectory: Array<number | null>;
+    outputTokenTrajectory: Array<number | null>;
+}
 /**
  * Reduce a span list to behavioral metrics + fired suboptimality signals.
  * Pure + deterministic: same spans → same output, on any machine, no model.
@@ -672,6 +685,8 @@ interface SemanticConceptJudgeOptions {
     model?: string;
     /** Per-call timeout. Default 300s. */
     timeoutMs?: number;
+    /** Provider-enforced output limit. Default 16000. */
+    maxTokens?: number;
     /** Pipeline budget for the prompt (source blob truncation). Default 45000. */
     maxSourceChars?: number;
     /** Per-file cap before inclusion. Default 20000. */
@@ -680,6 +695,9 @@ interface SemanticConceptJudgeOptions {
     maxHtmlChars?: number;
     /** LlmClient config (baseUrl, apiKey, authHeader, …). */
     llm?: LlmClientOptions;
+    costLedger?: CostLedger;
+    costPhase?: string;
+    signal?: AbortSignal;
     /**
      * Score aggregation strategy. Default `mean` — uniform average across
      * concepts. Cross-vertical comparisons should use `complexity` to
@@ -702,4 +720,4 @@ declare function runSemanticConceptJudge(input: SemanticConceptJudgeInput, optio
  */
 declare function createSemanticConceptJudge(options?: SemanticConceptJudgeOptions): (input: SemanticConceptJudgeInput) => Promise<SemanticConceptJudgeResult>;
 
-export { type RunScore as A, type BehavioralMetrics as B, type CreateAnalystAiConfig as C, DEFAULT_TRACE_ANALYST_KINDS as D, type RunScoreWeights as E, FAILURE_MODE_KIND_SPEC as F, type ConceptComplexity as G, type ConceptFinding as H, IMPROVEMENT_KIND_SPEC as I, type ConceptSpec as J, KIND_EXPECTED_SUBJECTS as K, type ConceptWeightStrategy as L, DEFAULT_COMPLEXITY_WEIGHTS as M, DEFAULT_RUN_SCORE_WEIGHTS as N, type RunCriticOptions as O, type PersistedFinding as P, SEMANTIC_CONCEPT_JUDGE_VERSION as Q, RunCritic as R, type SemanticConceptJudgeOptions as S, type SemanticConceptJudgeResult as T, type SuboptimalCode as U, type SuboptimalSignal as V, aggregateRunScore as W, clamp01 as X, computeTraceMetrics as Y, createSemanticConceptJudge as Z, runSemanticConceptJudge as _, type RunTrace as a, type SemanticConceptJudgeInput as b, type DiffPolicy as c, FINDING_SUBJECT_GRAMMAR_PROMPT as d, FINDING_SUBJECT_KINDS as e, FINDING_SUBJECT_SYNTAX as f, type FindingSubject as g, type FindingSubjectKind as h, FindingSubjectStringSchema as i, type FindingsDiff as j, FindingsStore as k, KNOWLEDGE_GAP_KIND_SPEC as l, KNOWLEDGE_POISONING_KIND_SPEC as m, SKILL_USAGE_ANALYST as n, SkillUsageAnalyst as o, type SkillUsageRecord as p, type SkillUsageReport as q, type SkillUsageScanConfig as r, buildSkillUsageReport as s, createAnalystAi as t, defaultIsMaterial as u, diffFindings as v, emitSkillUsageFindings as w, findingSubjectGrammarPromptFor as x, parseFindingSubject as y, renderFindingSubject as z };
+export { runSemanticConceptJudge as $, type RunScore as A, type BehavioralMetrics as B, type CreateAnalystAiConfig as C, DEFAULT_TRACE_ANALYST_KINDS as D, type RunScoreWeights as E, FAILURE_MODE_KIND_SPEC as F, type BehavioralTokenSequence as G, type ConceptComplexity as H, IMPROVEMENT_KIND_SPEC as I, type ConceptFinding as J, KIND_EXPECTED_SUBJECTS as K, type ConceptSpec as L, type ConceptWeightStrategy as M, DEFAULT_COMPLEXITY_WEIGHTS as N, DEFAULT_RUN_SCORE_WEIGHTS as O, type PersistedFinding as P, type RunCriticOptions as Q, RunCritic as R, type SemanticConceptJudgeOptions as S, SEMANTIC_CONCEPT_JUDGE_VERSION as T, type SemanticConceptJudgeResult as U, type SuboptimalCode as V, type SuboptimalSignal as W, aggregateRunScore as X, clamp01 as Y, computeTraceMetrics as Z, createSemanticConceptJudge as _, type RunTrace as a, type SemanticConceptJudgeInput as b, type DiffPolicy as c, FINDING_SUBJECT_GRAMMAR_PROMPT as d, FINDING_SUBJECT_KINDS as e, FINDING_SUBJECT_SYNTAX as f, type FindingSubject as g, type FindingSubjectKind as h, FindingSubjectStringSchema as i, type FindingsDiff as j, FindingsStore as k, KNOWLEDGE_GAP_KIND_SPEC as l, KNOWLEDGE_POISONING_KIND_SPEC as m, SKILL_USAGE_ANALYST as n, SkillUsageAnalyst as o, type SkillUsageRecord as p, type SkillUsageReport as q, type SkillUsageScanConfig as r, buildSkillUsageReport as s, createAnalystAi as t, defaultIsMaterial as u, diffFindings as v, emitSkillUsageFindings as w, findingSubjectGrammarPromptFor as x, parseFindingSubject as y, renderFindingSubject as z };

package/dist/{statistics-oUbOJe-S.d.ts → statistics-KUnG73jH.d.ts}

@@ -1,5 +1,5 @@
 import { d as ContinuousAgreementOptions, C as ContinuousAgreement } from './judge-calibration-7C-IDmKr.js';
-import { J as JudgeScore } from './types-C7DGg5ex.js';
+import { J as JudgeScore } from './types-BkfcQnxV.js';
 
 /** Identity: dimensions already follow "higher = better" by prompt convention
  *  (inverted dims like hallucination are scored 10 = best at the source). */

package/dist/{storage-Dw_f7WMt.d.ts → storage-DrX3v_5B.d.ts}

@@ -1,3 +1,5 @@
+import { C as CostLedger } from './cost-ledger-DWy3XdJc.js';
+
 /**
  * `CampaignStorage` — the filesystem seam `runCampaign` writes through
  * (run/cell dirs, the resumability cache, per-cell artifacts, trace spans).
@@ -22,6 +24,9 @@ interface CampaignStorage {
     read(path: string): string | undefined;
     /** Write a file (string or bytes). Parent dir is assumed ensured. */
     write(path: string, content: string | Uint8Array): void;
+    /** Append only when the current UTF-8 byte length matches `expectedBytes`.
+     * Returns the new length, or undefined when another writer won. */
+    append?(path: string, content: string, expectedBytes: number): number | undefined;
 }
 /** Node-filesystem storage — the default. Lazily requires `node:fs` so the
  *  module imports cleanly in non-Node runtimes (where the caller passes
@@ -35,5 +40,11 @@ declare function fsCampaignStorage(): CampaignStorage;
  *  live in a `Map` for the duration of the run; the `CampaignResult` is
  *  fully populated, but nothing is persisted to disk. */
 declare function inMemoryCampaignStorage(): CampaignStorage;
+/** Open the durable spend account stored beside a logical run. */
+declare function createRunCostLedger(input: {
+    storage: CampaignStorage;
+    runDir: string;
+    costCeilingUsd?: number;
+}): CostLedger;
 
-export { type CampaignStorage as C, fsCampaignStorage as f, inMemoryCampaignStorage as i };
+export { type CampaignStorage as C, createRunCostLedger as c, fsCampaignStorage as f, inMemoryCampaignStorage as i };

package/dist/{store-BsVi7ncX.d.ts → store-DGqD0Pyo.d.ts}

@@ -1,4 +1,4 @@
-import { R as Run, S as Span, e as TraceEvent, A as Artifact, B as BudgetLedgerEntry, f as RunStatus, g as RunLayer, b as SpanKind, E as EventKind } from './schema-SGWcK9wa.js';
+import { R as Run, S as Span, e as TraceEvent, A as Artifact, B as BudgetLedgerEntry, f as RunStatus, g as RunLayer, b as SpanKind, E as EventKind } from './schema-B3Q3l9Z_.js';
 
 interface RunFilter {
     scenarioId?: string;

package/dist/storyboard/index.d.ts

@@ -1,4 +1,4 @@
-import { S as Span } from '../schema-SGWcK9wa.js';
+import { S as Span } from '../schema-B3Q3l9Z_.js';
 
 /**
  * Code-edit extraction + the "watch the agent write code" animation.

package/dist/{summary-report-BJ5aNwZ1.d.ts → summary-report-C5bKFfm-.d.ts}

@@ -1,5 +1,5 @@
-import { R as RunRecord } from './run-record-B7RTi_ix.js';
-import { F as FailureClusterReport } from './failure-cluster-C48PiReX.js';
+import { R as RunRecord } from './run-record-BDH49H2E.js';
+import { F as FailureClusterReport } from './failure-cluster-DOAcSJ87.js';
 
 /**
  * HeldOutGate — first-class held-out paired-delta promotion gate.

package/dist/{test-graded-scenario-mzYBKspu.d.ts → test-graded-scenario-B0ybnPY7.d.ts}

@@ -1,6 +1,6 @@
-import { T as TraceEmitter } from './emitter-BRchAAAx.js';
-import { R as Run, F as FailureClass } from './schema-SGWcK9wa.js';
-import { T as TraceStore } from './store-BsVi7ncX.js';
+import { T as TraceEmitter } from './emitter-CjD7vUwv.js';
+import { R as Run, F as FailureClass } from './schema-B3Q3l9Z_.js';
+import { T as TraceStore } from './store-DGqD0Pyo.js';
 
 /**
  * SandboxHarness — executes a scenario in an isolated environment and

package/dist/traces.d.ts

@@ -1,23 +1,24 @@
 import { N as NotFoundError, R as ReplayError } from './errors-oeQrLqXC.js';
 import { P as ProviderRedactor, R as RawProviderSink, d as RawProviderEvent } from './raw-provider-sink-C46HDghv.js';
 export { F as FileSystemRawProviderSink, a as FileSystemRawProviderSinkOptions, I as InMemoryRawProviderSink, b as InMemoryRawProviderSinkOptions, N as NoopRawProviderSink, c as RawProviderDirection, e as RawProviderSinkFilter, f as defaultProviderRedactor, p as providerFromBaseUrl } from './raw-provider-sink-C46HDghv.js';
-import { a as RunCompleteHookContext, R as RunCompleteHook } from './emitter-BRchAAAx.js';
-export { S as SpanHandle, T as TraceEmitter, b as TraceEmitterOptions, l as llmSpanFromProvider } from './emitter-BRchAAAx.js';
-export { b as RunIntegrityError, R as RunIntegrityExpectations, c as RunIntegrityIssue, d as RunIntegrityIssueCode, a as RunIntegrityReport, e as assertRunCaptured, t as throwIfRunIncomplete } from './integrity-qemeBAyx.js';
-import { T as TraceStore } from './store-BsVi7ncX.js';
-export { E as EventFilter, F as FileSystemTraceStore, a as FileSystemTraceStoreOptions, I as InMemoryTraceStore, R as RunFilter, S as SpanFilter } from './store-BsVi7ncX.js';
-export { a as aggregateLlm, b as argHash, g as groupBy, j as judgeSpans, l as llmSpans, r as runFailureClass, c as runsForScenario, t as toolSpans } from './query-Ck190MOd.js';
-import { R as Run } from './schema-SGWcK9wa.js';
-export { A as Artifact, B as BudgetLedgerEntry, h as BudgetSpec, E as EventKind, i as FAILURE_CLASSES, F as FailureClass, G as GenericSpan, J as JudgeSpan, L as LlmSpan, M as Message, c as RetrievalSpan, g as RunLayer, a as RunOutcome, f as RunStatus, d as SandboxSpan, S as Span, j as SpanBase, b as SpanKind, k as SpanStatus, l as TRACE_SCHEMA_VERSION, T as ToolSpan, e as TraceEvent, m as isJudgeSpan, n as isLlmSpan, o as isRetrievalSpan, p as isSandboxSpan, q as isToolSpan } from './schema-SGWcK9wa.js';
+import { a as RunCompleteHookContext, R as RunCompleteHook } from './emitter-CjD7vUwv.js';
+export { S as SpanHandle, T as TraceEmitter, b as TraceEmitterOptions, l as llmSpanFromProvider } from './emitter-CjD7vUwv.js';
+export { b as RunIntegrityError, R as RunIntegrityExpectations, c as RunIntegrityIssue, d as RunIntegrityIssueCode, a as RunIntegrityReport, e as assertRunCaptured, t as throwIfRunIncomplete } from './integrity-DqlBiLyK.js';
+import { T as TraceStore } from './store-DGqD0Pyo.js';
+export { E as EventFilter, F as FileSystemTraceStore, a as FileSystemTraceStoreOptions, I as InMemoryTraceStore, R as RunFilter, S as SpanFilter } from './store-DGqD0Pyo.js';
+import { T as ToolSpan, R as Run } from './schema-B3Q3l9Z_.js';
+export { A as Artifact, B as BudgetLedgerEntry, h as BudgetSpec, E as EventKind, i as FAILURE_CLASSES, F as FailureClass, G as GenericSpan, J as JudgeSpan, L as LlmSpan, M as Message, c as RetrievalSpan, g as RunLayer, a as RunOutcome, f as RunStatus, d as SandboxSpan, S as Span, j as SpanBase, b as SpanKind, k as SpanStatus, l as TRACE_SCHEMA_VERSION, e as TraceEvent, m as isJudgeSpan, n as isLlmSpan, o as isRetrievalSpan, p as isSandboxSpan, q as isToolSpan } from './schema-B3Q3l9Z_.js';
+export { a as aggregateLlm, b as argHash, g as groupBy, h as hasCapturedToolArgs, j as judgeSpans, l as llmSpans, r as runFailureClass, c as runsForScenario, t as toolSpans } from './query-CF7PG61p.js';
 import { A as AnalyzeTracesOptions, b as AnalyzeTracesResult } from './analyst-C8HHvfJp.js';
 export { a as AnalyzeTracesInput, c as AnalyzeTracesTurnSnapshot, d as analyzeTraces } from './analyst-C8HHvfJp.js';
 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';
 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';
-import { b as RunSplitTag, d as RunTokenUsage, R as RunRecord } from './run-record-B7RTi_ix.js';
+import { a as RunSplitTag, c as RunTokenUsage, R as RunRecord } from './run-record-BDH49H2E.js';
 import { AxFunction } from '@ax-llm/ax';
 import '@tangle-network/agent-interface';
 
 /** Canonical OpenInference-over-OTLP attribute vocabulary used at the trace boundary. */
+
 declare const OPENINFERENCE_SPAN_KIND = "openinference.span.kind";
 declare const LLM_MODEL_NAME = "llm.model_name";
 declare const LLM_INPUT_TOKENS = "llm.token_count.prompt";
@@ -25,6 +26,10 @@ declare const LLM_OUTPUT_TOKENS = "llm.token_count.completion";
 declare const LLM_CACHED_TOKENS = "llm.token_count.prompt_cache_hit";
 declare const LLM_COST_USD = "llm.cost_usd";
 declare const TOOL_NAME = "tool.name";
+declare const TOOL_ARGS_CAPTURED = "tool.args_captured";
+declare const TOOL_LATENCY_MS = "tool.latency_ms";
+declare const INPUT_VALUE = "input.value";
+declare const OUTPUT_VALUE = "output.value";
 declare const SPAN_KIND_ATTR_KEYS: readonly ["openinference.span.kind", "inference.observation_kind"];
 declare const LLM_MODEL_ATTR_KEYS: readonly ["llm.model_name", "inference.llm.model_name", "llm.model", "gen_ai.request.model", "gen_ai.response.model"];
 declare const LLM_INPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt", "inference.llm.input_tokens", "llm.input_tokens", "gen_ai.usage.input_tokens", "gen_ai.usage.prompt_tokens"];
@@ -32,6 +37,8 @@ declare const LLM_OUTPUT_TOKEN_ATTR_KEYS: readonly ["llm.token_count.completion"
 declare const LLM_CACHED_TOKEN_ATTR_KEYS: readonly ["llm.token_count.prompt_cache_hit", "inference.llm.cached_tokens", "llm.cached_tokens", "gen_ai.usage.cached_tokens"];
 declare const LLM_COST_ATTR_KEYS: readonly ["llm.cost_usd", "inference.llm.cost.total", "llm.cost.total", "gen_ai.usage.cost"];
 declare const TOOL_NAME_ATTR_KEYS: readonly ["tool.name", "inference.tool.name"];
+type ToolSpanOtlpInput = Pick<ToolSpan, 'toolName' | 'args' | 'argsCaptured' | 'result' | 'latencyMs'>;
+declare function applyToolSpanOtlpAttributes(attributes: Record<string, unknown>, span: ToolSpanOtlpInput): void;
 declare function traceSpanKindToOpenInferenceKind(kind: string): string;
 
 /**
@@ -710,6 +717,7 @@ declare function captureFetchToRawSink(fetch: typeof globalThis.fetch, sink: Raw
  * or when the batch fills. No @opentelemetry SDK dependency — minimal
  * OTLP/JSON serializer (~120 LOC) using the existing otel.ts helpers.
  */
+
 interface OtelExportConfig {
     /** OTLP endpoint. Reads OTEL_EXPORTER_OTLP_ENDPOINT env by default. */
     endpoint?: string;
@@ -746,6 +754,7 @@ interface ExportableSpan {
     inputTokens?: number;
     outputTokens?: number;
     costUsd?: number;
+    tool?: ToolSpanOtlpInput;
     attributes?: Record<string, unknown>;
 }
 /**
@@ -1016,4 +1025,4 @@ declare function iterateRawCalls(sink: RawProviderSink, filter?: {
     spanId?: string;
 }): AsyncGenerator<ReplayCacheEntry>;
 
-export { AnalyzeTracesOptions, AnalyzeTracesResult, type CaptureFetchContext, type CaptureFetchOptions, DEFAULT_REDACTION_RULES, DatasetOverview, type ExportableSpan, type ExtractedUsage, type FlattenOtlpOptions, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_COST_ATTR_KEYS, LLM_COST_USD, LLM_INPUT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_MODEL_ATTR_KEYS, LLM_MODEL_NAME, LLM_OUTPUT_TOKENS, LLM_OUTPUT_TOKEN_ATTR_KEYS, OPENINFERENCE_SPAN_KIND, OTEL_AGENT_EVAL_SCOPE, type OtelExportConfig, type OtelExporter, type OtlpExport, OtlpFileTraceStore, type OtlpFileTraceStoreOptions, type OtlpFlatLine, type OtlpResourceSpans, type OtlpSpan, type OtlpToRunRecordsOptions, type OtlpTraceRunRecord, type ProjectedOtlpSpan, ProviderRedactor, QueryTracesPage, REDACTION_VERSION, RawProviderEvent, RawProviderSink, type RedactionReport, type RedactionRule, ReplayCache, type ReplayCacheEntry, ReplayCacheMissError, type ReplayCacheStats, type ReplayFetchOptions, Run, RunCompleteHook, RunCompleteHookContext, SPAN_KIND_ATTR_KEYS, SearchSpanResult, SearchTraceResult, SpanNotFoundError, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_SUBAGENT_DESCRIPTION, type TraceAggregate, TraceAnalysisStore, TraceAnalystFilters, type TraceAnalystHookOptions, TraceAnalystSpanKind, TraceAnalystSpanStatus, TraceFileMissingError, type TraceInsightContext, type TraceInsightFinding, type TraceInsightPanelRole, type TraceInsightPromptInput, type TraceInsightQualityGate, type TraceInsightQuestion, type TraceInsightReadiness, type TraceInsightSuite, type TraceInsightTask, TraceNotFoundError, TraceStore, type TraceStoreSource, type TraceStoreToOtlpOptions, type TracesToOtlpResult, ViewSpansResult, ViewTraceResult, asNumber, asString, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, captureFetchToRawSink, convertTraceStoresToOtlp, createOtelExporter, createOtelTracingStore, createReplayFetch, defaultTraceInsightPanel, describeTraceInsightScope, domainEvidencePattern, exportRunAsOtlp, extractOtlpAttributes, extractUsage, extractUsageFromResponse, extractUsageFromSse, firstNumberAttr, firstStringAttr, flattenOtlpExportToNdjson, inferDomainKeywords, inferOtlpKind, iterateRawCalls, otelRunCompleteHook, otlpToRunRecords, otlpToTraceRunRecords, planTraceInsightQuestions, projectOtlpFlatLine, readOtlpStatus, redactString, redactValue, scoreTraceInsightReadiness, stringField, tokenizeDomainWords, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceSpanKindToOpenInferenceKind };
+export { AnalyzeTracesOptions, AnalyzeTracesResult, type CaptureFetchContext, type CaptureFetchOptions, DEFAULT_REDACTION_RULES, DatasetOverview, type ExportableSpan, type ExtractedUsage, type FlattenOtlpOptions, INPUT_VALUE, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_COST_ATTR_KEYS, LLM_COST_USD, LLM_INPUT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_MODEL_ATTR_KEYS, LLM_MODEL_NAME, LLM_OUTPUT_TOKENS, LLM_OUTPUT_TOKEN_ATTR_KEYS, OPENINFERENCE_SPAN_KIND, OTEL_AGENT_EVAL_SCOPE, OUTPUT_VALUE, type OtelExportConfig, type OtelExporter, type OtlpExport, OtlpFileTraceStore, type OtlpFileTraceStoreOptions, type OtlpFlatLine, type OtlpResourceSpans, type OtlpSpan, type OtlpToRunRecordsOptions, type OtlpTraceRunRecord, type ProjectedOtlpSpan, ProviderRedactor, QueryTracesPage, REDACTION_VERSION, RawProviderEvent, RawProviderSink, type RedactionReport, type RedactionRule, ReplayCache, type ReplayCacheEntry, ReplayCacheMissError, type ReplayCacheStats, type ReplayFetchOptions, Run, RunCompleteHook, RunCompleteHookContext, SPAN_KIND_ATTR_KEYS, SearchSpanResult, SearchTraceResult, SpanNotFoundError, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_SUBAGENT_DESCRIPTION, ToolSpan, type ToolSpanOtlpInput, type TraceAggregate, TraceAnalysisStore, TraceAnalystFilters, type TraceAnalystHookOptions, TraceAnalystSpanKind, TraceAnalystSpanStatus, TraceFileMissingError, type TraceInsightContext, type TraceInsightFinding, type TraceInsightPanelRole, type TraceInsightPromptInput, type TraceInsightQualityGate, type TraceInsightQuestion, type TraceInsightReadiness, type TraceInsightSuite, type TraceInsightTask, TraceNotFoundError, TraceStore, type TraceStoreSource, type TraceStoreToOtlpOptions, type TracesToOtlpResult, ViewSpansResult, ViewTraceResult, applyToolSpanOtlpAttributes, asNumber, asString, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, captureFetchToRawSink, convertTraceStoresToOtlp, createOtelExporter, createOtelTracingStore, createReplayFetch, defaultTraceInsightPanel, describeTraceInsightScope, domainEvidencePattern, exportRunAsOtlp, extractOtlpAttributes, extractUsage, extractUsageFromResponse, extractUsageFromSse, firstNumberAttr, firstStringAttr, flattenOtlpExportToNdjson, inferDomainKeywords, inferOtlpKind, iterateRawCalls, otelRunCompleteHook, otlpToRunRecords, otlpToTraceRunRecords, planTraceInsightQuestions, projectOtlpFlatLine, readOtlpStatus, redactString, redactValue, scoreTraceInsightReadiness, stringField, tokenizeDomainWords, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceSpanKindToOpenInferenceKind };