@tangle-network/agent-eval 0.103.1 → 0.104.0

package/dist/index.d.ts

@@ -1,28 +1,28 @@
-export { A as ActionExecutionPolicy, a as ActionPolicyDecision, C as ControlRunToRunRecordOptions, L as LlmJsonCall, b as LlmReviewerConfig, P as ProposeFn, c as ProposeInput, d as ProposeOutput, e as ProposeReviewConfig, f as ProposeReviewControlAction, g as ProposeReviewControlConfig, h as ProposeReviewControlResult, i as ProposeReviewControlState, j as ProposeReviewReport, k as ProposeReviewShot, R as Review, l as ReviewFn, m as ReviewInput, n as ReviewMemoryEntry, o as ReviewMemoryStore, p as RunEvidenceMetadata, V as Verification, q as VerifyFn, r as controlFailureClassFromVerification, s as controlRunToRunRecord, t as createLlmReviewer, u as evaluateActionPolicy, v as inMemoryReviewStore, w as jsonlReviewStore, x as runProposeReview, y as runProposeReviewAsControlLoop, z as scoreFromEvals } from './control-KofK3gfG.js';
-import { R as RunRecord, b as RunSplitTag } from './run-record-CPfd1ARZ.js';
-export { f as AGENT_PROFILE_KINDS, g as AgentInterfaceProfileLike, A as AgentProfileCell, e as AgentProfileCellInput, h as AgentProfileCellSchemaVersion, i as AgentProfileCellValidationError, j as AgentProfileDimensionValue, k as AgentProfileHarness, a as AgentProfileJson, l as AgentProfileKind, m as AgentProfileSource, n as AgentProfileSourceInput, J as JudgeScoresRecord, d as RunJudgeMetadata, o as RunOutcome, p as RunRecordValidationError, c as RunTokenUsage, q as agentProfileCellHashMaterial, r as agentProfileCellKey, s as assertRunAgentProfileCell, t as buildAgentInterfaceProfileCell, u as buildAgentProfileCell, v as groupRunsByAgentProfileCell, w as isRunRecord, x as modelHasSnapshot, y as parseRunRecordSafe, z as requireAgentProfileCell, B as roundTripRunRecord, C as toAgentProfileJson, D as validateAgentProfileCell, E as validateRunRecord, F as verifyAgentProfileCell } from './run-record-CPfd1ARZ.js';
-import { B as BehavioralMetrics } from './semantic-concept-judge-C6mDEBIo.js';
-export { x as ConceptComplexity, y as ConceptFinding, z as ConceptSpec, A as ConceptWeightStrategy, C as CreateAnalystAiConfig, E as DEFAULT_COMPLEXITY_WEIGHTS, D as DEFAULT_TRACE_ANALYST_KINDS, b as DiffPolicy, F as FAILURE_MODE_KIND_SPEC, e as FindingSubject, f as FindingSubjectKind, h as FindingsDiff, i as FindingsStore, I as IMPROVEMENT_KIND_SPEC, j as KNOWLEDGE_GAP_KIND_SPEC, k as KNOWLEDGE_POISONING_KIND_SPEC, P as PersistedFinding, G as SEMANTIC_CONCEPT_JUDGE_VERSION, l as SKILL_USAGE_ANALYST, a as SemanticConceptJudgeInput, S as SemanticConceptJudgeOptions, H as SemanticConceptJudgeResult, m as SkillUsageAnalyst, J as SuboptimalCode, L as SuboptimalSignal, M as computeTraceMetrics, r as createAnalystAi, N as createSemanticConceptJudge, s as defaultIsMaterial, t as diffFindings, O as runSemanticConceptJudge } from './semantic-concept-judge-C6mDEBIo.js';
-import { l as ChatRequest, p as CreateChatClientOpts } from './types-DA9yj-Jd.js';
-export { a as Analyst, b as AnalystContext, g as AnalystCost, A as AnalystFinding, i as AnalystInputKind, j as AnalystRequirements, f as AnalystRunEvent, e as AnalystRunInputs, d as AnalystRunResult, c as AnalystRunSummary, h as AnalystSeverity, k as ChatCallOpts, C as ChatClient, m as ChatResponse, n as ChatTransport, o as CliBridgeTransportOpts, D as DirectProviderTransportOpts, E as EvidenceRef, M as MockTransportOpts, R as RouterTransportOpts, S as SandboxSdkTransportOpts, q as computeFindingId, r as createChatClient, s as makeFinding } from './types-DA9yj-Jd.js';
-export { a as AnalystHooks, A as AnalystRegistry, b as AnalystRegistryOptions, B as BudgetPolicy, D as DefaultAnalystRegistryOptions, R as RegistryRunOpts, c as buildDefaultAnalystRegistry } from './default-registry-Ez4cxuZ4.js';
-export { C as CreateTraceAnalystKindOpts, a as RawAnalystFinding, c as TraceAnalystGolden, T as TraceAnalystKindSpec, d as createTraceAnalystKind, r as renderPriorFindings } from './kind-factory-BLHxwX71.js';
-export { F as FindingToPolicyEditOptions, P as POLICY_EDIT_AXES, a as POLICY_EDIT_TARGET_SURFACES, b as PolicyEdit, c as PolicyEditAdmission, d as PolicyEditAdmissionOptions, e as PolicyEditAxis, f as PolicyEditChange, g as PolicyEditExpectedGain, h as PolicyEditGainDirection, i as PolicyEditGainUnit, j as PolicyEditInit, k as PolicyEditRisk, l as PolicyEditSchemaVersion, m as PolicyEditSource, n as PolicyEditTarget, o as PolicyEditTargetSurface, p as PolicyEditValidationError, q as admitPolicyEdit, r as applyPolicyEditToSurface, s as computePolicyEditId, t as isPolicyEdit, u as makePolicyEdit, v as policyEditFromFinding, w as policyEditsFromFindings, x as scorePolicyEditReadiness, y as validatePolicyEdit } from './policy-edit-nUhuFtLF.js';
+export { A as ActionExecutionPolicy, a as ActionPolicyDecision, C as ControlRunToRunRecordOptions, L as LlmJsonCall, b as LlmReviewerConfig, P as ProposeFn, c as ProposeInput, d as ProposeOutput, e as ProposeReviewConfig, f as ProposeReviewControlAction, g as ProposeReviewControlConfig, h as ProposeReviewControlResult, i as ProposeReviewControlState, j as ProposeReviewReport, k as ProposeReviewShot, R as Review, l as ReviewFn, m as ReviewInput, n as ReviewMemoryEntry, o as ReviewMemoryStore, p as RunEvidenceMetadata, V as Verification, q as VerifyFn, r as controlFailureClassFromVerification, s as controlRunToRunRecord, t as createLlmReviewer, u as evaluateActionPolicy, v as inMemoryReviewStore, w as jsonlReviewStore, x as runProposeReview, y as runProposeReviewAsControlLoop, z as scoreFromEvals } from './control-B8UthSBL.js';
+import { R as RunRecord, b as RunSplitTag } from './run-record-I-Z3JNvO.js';
+export { f as AGENT_PROFILE_KINDS, g as AgentInterfaceProfileLike, A as AgentProfileCell, e as AgentProfileCellInput, h as AgentProfileCellSchemaVersion, i as AgentProfileCellValidationError, j as AgentProfileDimensionValue, k as AgentProfileHarness, a as AgentProfileJson, l as AgentProfileKind, m as AgentProfileSource, n as AgentProfileSourceInput, J as JudgeScoresRecord, d as RunJudgeMetadata, o as RunOutcome, p as RunRecordValidationError, c as RunTokenUsage, q as agentProfileCellHashMaterial, r as agentProfileCellKey, s as assertRunAgentProfileCell, t as buildAgentInterfaceProfileCell, u as buildAgentProfileCell, v as groupRunsByAgentProfileCell, w as isRunRecord, x as modelHasSnapshot, y as parseRunRecordSafe, z as requireAgentProfileCell, B as roundTripRunRecord, C as toAgentProfileJson, D as validateAgentProfileCell, E as validateRunRecord, F as verifyAgentProfileCell } from './run-record-I-Z3JNvO.js';
+import { B as BehavioralMetrics } from './semantic-concept-judge-BmNZPB_j.js';
+export { x as ConceptComplexity, y as ConceptFinding, z as ConceptSpec, A as ConceptWeightStrategy, C as CreateAnalystAiConfig, E as DEFAULT_COMPLEXITY_WEIGHTS, D as DEFAULT_TRACE_ANALYST_KINDS, b as DiffPolicy, F as FAILURE_MODE_KIND_SPEC, e as FindingSubject, f as FindingSubjectKind, h as FindingsDiff, i as FindingsStore, I as IMPROVEMENT_KIND_SPEC, j as KNOWLEDGE_GAP_KIND_SPEC, k as KNOWLEDGE_POISONING_KIND_SPEC, P as PersistedFinding, G as SEMANTIC_CONCEPT_JUDGE_VERSION, l as SKILL_USAGE_ANALYST, a as SemanticConceptJudgeInput, S as SemanticConceptJudgeOptions, H as SemanticConceptJudgeResult, m as SkillUsageAnalyst, J as SuboptimalCode, L as SuboptimalSignal, M as computeTraceMetrics, r as createAnalystAi, N as createSemanticConceptJudge, s as defaultIsMaterial, t as diffFindings, O as runSemanticConceptJudge } from './semantic-concept-judge-BmNZPB_j.js';
+import { l as ChatRequest, p as CreateChatClientOpts } from './types-D1ytG0Yg.js';
+export { a as Analyst, b as AnalystContext, g as AnalystCost, A as AnalystFinding, i as AnalystInputKind, j as AnalystRequirements, f as AnalystRunEvent, e as AnalystRunInputs, d as AnalystRunResult, c as AnalystRunSummary, h as AnalystSeverity, k as ChatCallOpts, C as ChatClient, m as ChatResponse, n as ChatTransport, o as CliBridgeTransportOpts, D as DirectProviderTransportOpts, E as EvidenceRef, M as MockTransportOpts, R as RouterTransportOpts, S as SandboxSdkTransportOpts, q as computeFindingId, r as createChatClient, s as makeFinding } from './types-D1ytG0Yg.js';
+export { a as AnalystHooks, A as AnalystRegistry, b as AnalystRegistryOptions, B as BudgetPolicy, D as DefaultAnalystRegistryOptions, R as RegistryRunOpts, c as buildDefaultAnalystRegistry } from './default-registry-BswHCXnU.js';
+export { C as CreateTraceAnalystKindOpts, a as RawAnalystFinding, c as TraceAnalystGolden, T as TraceAnalystKindSpec, d as createTraceAnalystKind, r as renderPriorFindings } from './kind-factory-DvIGo_cP.js';
+export { F as FindingToPolicyEditOptions, P as POLICY_EDIT_AXES, a as POLICY_EDIT_TARGET_SURFACES, b as PolicyEdit, c as PolicyEditAdmission, d as PolicyEditAdmissionOptions, e as PolicyEditAxis, f as PolicyEditChange, g as PolicyEditExpectedGain, h as PolicyEditGainDirection, i as PolicyEditGainUnit, j as PolicyEditInit, k as PolicyEditRisk, l as PolicyEditSchemaVersion, m as PolicyEditSource, n as PolicyEditTarget, o as PolicyEditTargetSurface, p as PolicyEditValidationError, q as admitPolicyEdit, r as applyPolicyEditToSurface, s as computePolicyEditId, t as isPolicyEdit, u as makePolicyEdit, v as policyEditFromFinding, w as policyEditsFromFindings, x as scorePolicyEditReadiness, y as validatePolicyEdit } from './policy-edit-DQUXYMDm.js';
 import { TCloud } from '@tangle-network/tcloud';
 import { B as BenchmarkRunnerConfig, S as Scenario, c as BenchmarkReport, P as ProductClientConfig, C as CheckResult, T as TestResult, d as PersonaConfig, D as DriverResult, e as DriverState, b as JudgeFn, f as CollectedArtifacts, g as ScenarioResult, h as TurnMetrics, i as ScenarioFile, j as CompletionCriterion } from './types-C7DGg5ex.js';
 export { A as ArtifactCheck, k as ArtifactResult, E as EvalResult, F as FeedbackPattern, l as JudgeConfig, a as JudgeInput, m as JudgeRubric, J as JudgeScore, n as PersonaRigor, R as RouteMap, o as RubricDimension, p as Turn, q as TurnResult } from './types-C7DGg5ex.js';
 export { c as ControlActionFailureMode, d as ControlActionOutcome, e as ControlBudget, f as ControlContext, g as ControlDecision, C as ControlEvalResult, a as ControlRunResult, h as ControlRuntimeConfig, i as ControlRuntimeError, j as ControlSeverity, b as ControlStep, k as ControlStopPolicies, S as StopDecision, l as allCriticalPassed, o as objectiveEval, r as runAgentControlLoop, s as stopOnNoProgress, m as stopOnRepeatedAction, n as subjectiveEval } from './control-runtime-Acf9CGhw.js';
 import { F as FailureClass, T as ToolSpan, h as BudgetSpec, B as BudgetLedgerEntry, R as Run, L as LlmSpan } from './schema-m0gsnbt3.js';
 export { A as Artifact, E as EventKind, i as FAILURE_CLASSES, G as GenericSpan, J as JudgeSpan, M as Message, d as RetrievalSpan, g as RunLayer, f as RunStatus, e as SandboxSpan, S as Span, j as SpanBase, c as SpanKind, k as SpanStatus, l as TRACE_SCHEMA_VERSION, a as TraceEvent, m as isJudgeSpan, n as isLlmSpan, o as isRetrievalSpan, p as isSandboxSpan, q as isToolSpan } from './schema-m0gsnbt3.js';
-import { A as AgentEvalError, J as JudgeError, a as ConfigError } from './errors-CzMUYo7b.js';
-export { b as AgentEvalErrorCode, C as CaptureIntegrityError, N as NotFoundError, R as ReplayError, V as ValidationError, c as VerificationError } from './errors-CzMUYo7b.js';
-import { b as FeedbackLabel, F as FeedbackTrajectoryStore, a as FeedbackTrajectory } from './feedback-trajectory-BxY0cKfs.js';
-export { c as FeedbackArtifactType, d as FeedbackAttempt, e as FeedbackLabelKind, f as FeedbackLabelSource, g as FeedbackOptimizerRow, h as FeedbackOutcome, i as FeedbackReplayAdapter, j as FeedbackReplayResult, k as FeedbackSeverity, l as FeedbackSplitPolicy, m as FeedbackTask, n as FeedbackTrajectoryFilter, o as FileSystemFeedbackTrajectoryStore, I as InMemoryFeedbackTrajectoryStore, P as PreferenceMemoryEntry, p as ProposedSideEffect, q as assignFeedbackSplit, r as controlRunToFeedbackTrajectory, s as createFeedbackTrajectory, t as feedbackTrajectoriesToDatasetScenarios, u as feedbackTrajectoriesToOptimizerRows, v as feedbackTrajectoryToDatasetScenario, w as feedbackTrajectoryToOptimizerRow, x as parseFeedbackTrajectoriesJsonl, y as renderPreferenceMemoryMarkdown, z as replayFeedbackTrajectories, A as replayFeedbackTrajectory, B as serializeFeedbackTrajectoriesJsonl, C as summarizePreferenceMemory, D as withAssignedFeedbackSplit } from './feedback-trajectory-BxY0cKfs.js';
-import { b as CorrectnessChecker } from './pre-registration-CUOSGAZK.js';
-export { A as ArtifactCheckArtifact, d as ArtifactEventLike, e as ArtifactValidator, f as BackendIntegrityError, B as BackendIntegrityReport, C as CompletionRequirement, a as CompletionVerdict, H as HypothesisManifest, g as HypothesisResult, h as LlmCorrectnessCheckerOpts, L as LlmJudgeDimension, c as LlmJudgeOptions, i as ProducedProposal, P as ProducedState, j as ProposalEventLike, k as RequirementCheck, R as RuntimeEventLike, m as SatisfiedBy, S as SignedManifest, n as SignedManifestAlgo, T as TaskGold, o as ToolCallEventLike, V as ValidationContext, p as ValidationIssue, q as ValidationResult, r as assertRealBackend, s as byteLengthRange, t as canonicalize, u as completionVerdict, v as composeValidators, w as containsAll, x as createLlmCorrectnessChecker, y as createTokenRecallChecker, z as evaluateHypothesis, D as extractProducedState, E as hashJson, F as jsonHasKeys, l as llmJudge, G as parseCorrectnessResponse, I as regexMatch, J as signManifest, K as summarizeBackendIntegrity, M as verifyCompletion, N as verifyManifest } from './pre-registration-CUOSGAZK.js';
+import { A as AgentEvalError, J as JudgeError, a as ConfigError } from './errors-oeQrLqXC.js';
+export { b as AgentEvalErrorCode, C as CaptureIntegrityError, N as NotFoundError, R as ReplayError, V as ValidationError, c as VerificationError } from './errors-oeQrLqXC.js';
+import { b as FeedbackLabel, F as FeedbackTrajectoryStore, a as FeedbackTrajectory } from './feedback-trajectory-C9KCo8ag.js';
+export { c as FeedbackArtifactType, d as FeedbackAttempt, e as FeedbackLabelKind, f as FeedbackLabelSource, g as FeedbackOptimizerRow, h as FeedbackOutcome, i as FeedbackReplayAdapter, j as FeedbackReplayResult, k as FeedbackSeverity, l as FeedbackSplitPolicy, m as FeedbackTask, n as FeedbackTrajectoryFilter, o as FileSystemFeedbackTrajectoryStore, I as InMemoryFeedbackTrajectoryStore, P as PreferenceMemoryEntry, p as ProposedSideEffect, q as assignFeedbackSplit, r as controlRunToFeedbackTrajectory, s as createFeedbackTrajectory, t as feedbackTrajectoriesToDatasetScenarios, u as feedbackTrajectoriesToOptimizerRows, v as feedbackTrajectoryToDatasetScenario, w as feedbackTrajectoryToOptimizerRow, x as parseFeedbackTrajectoriesJsonl, y as renderPreferenceMemoryMarkdown, z as replayFeedbackTrajectories, A as replayFeedbackTrajectory, B as serializeFeedbackTrajectoriesJsonl, C as summarizePreferenceMemory, D as withAssignedFeedbackSplit } from './feedback-trajectory-C9KCo8ag.js';
+import { b as CorrectnessChecker } from './pre-registration-D8Zr8-rr.js';
+export { A as ArtifactCheckArtifact, d as ArtifactEventLike, e as ArtifactValidator, f as BackendIntegrityError, B as BackendIntegrityReport, C as CompletionRequirement, a as CompletionVerdict, H as HypothesisManifest, g as HypothesisResult, h as LlmCorrectnessCheckerOpts, L as LlmJudgeDimension, c as LlmJudgeOptions, i as ProducedProposal, P as ProducedState, j as ProposalEventLike, k as RequirementCheck, R as RuntimeEventLike, m as SatisfiedBy, S as SignedManifest, n as SignedManifestAlgo, T as TaskGold, o as ToolCallEventLike, V as ValidationContext, p as ValidationIssue, q as ValidationResult, r as assertRealBackend, s as byteLengthRange, t as canonicalize, u as completionVerdict, v as composeValidators, w as containsAll, x as createLlmCorrectnessChecker, y as createTokenRecallChecker, z as evaluateHypothesis, D as extractProducedState, E as hashJson, F as jsonHasKeys, l as llmJudge, G as parseCorrectnessResponse, I as regexMatch, J as signManifest, K as summarizeBackendIntegrity, M as verifyCompletion, N as verifyManifest } from './pre-registration-D8Zr8-rr.js';
 export { DataAcquisitionPlan, KnowledgeAcquisitionMode, KnowledgeBundle, KnowledgeFallbackPolicy, KnowledgeFreshness, KnowledgeImportance, KnowledgeReadinessReport, KnowledgeRecommendedAction, KnowledgeRequirement, KnowledgeRequirementCategory, KnowledgeResponsibleSurface, KnowledgeSensitivity, ScoreKnowledgeReadinessOptions, UserQuestion, acquisitionPlansForKnowledgeGaps, blockingKnowledgeEval, knowledgeReadinessTracePayload, scoreKnowledgeReadiness, userQuestionsForKnowledgeGaps } from './knowledge/index.js';
-import { h as ReleaseConfidenceThresholds, f as ReleaseConfidenceScorecard } from './release-report-DkaCZ9k4.js';
-export { A as ActionableSideInfo, o as AsiSeverity, B as BootstrapOptions, a as BootstrapResult, J as JudgeReplayGateArgs, R as ReleaseConfidenceAxis, b as ReleaseConfidenceAxisName, c as ReleaseConfidenceInput, d as ReleaseConfidenceIssue, e as ReleaseConfidenceMetrics, g as ReleaseConfidenceStatus, 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-DkaCZ9k4.js';
+import { h as ReleaseConfidenceThresholds, f as ReleaseConfidenceScorecard } from './release-report-DeJpsBiA.js';
+export { A as ActionableSideInfo, o as AsiSeverity, B as BootstrapOptions, a as BootstrapResult, J as JudgeReplayGateArgs, R as ReleaseConfidenceAxis, b as ReleaseConfidenceAxisName, c as ReleaseConfidenceInput, d as ReleaseConfidenceIssue, e as ReleaseConfidenceMetrics, g as ReleaseConfidenceStatus, 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-DeJpsBiA.js';
 export { c as CliffsMagnitude, d as CorpusAgreementOptions, e as CorpusAgreementPerDimension, C as CorpusAgreementReport, f as CorpusScoreRecord, g as EProcess, h as EProcessOptions, E as EProcessState, i as EProcessStep, M as McNemarResult, P as PairedBootstrapOptions, a as PairedBootstrapResult, j as ProportionInterval, R as RiskDifferenceResult, W as WeightedCompositeInput, k as WeightedCompositeResult, b as benjaminiHochberg, l as bonferroni, m as cliffsDelta, n as cohensD, o as confidenceInterval, q as corpusInterRaterAgreement, r as corpusInterRaterAgreementFromJudgeScores, s as eProcess, t as interRaterReliability, u as interpretCliffs, v as mannWhitneyU, x as mcnemar, y as mcnemarPower, z as mcnemarRequiredN, A as mulberry32, B as normalizeScores, p as pairedBootstrap, D as pairedMde, F as pairedRiskDifference, G as pairedTTest, H as partialCredit, I as passAtK, J as pearsonR, K as ranks, L as requiredSampleSize, N as spearmanR, O as weightedComposite, Q as weightedMean, w as wilcoxonSignedRank, S as wilson } from './statistics-D88peojY.js';
 import { OtelExporter, OtelExportConfig } from './traces.js';
 export { CaptureFetchContext, CaptureFetchOptions, ExportableSpan, ExtractedUsage, 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, OtlpExport, OtlpFileTraceStore, OtlpFileTraceStoreOptions, OtlpFlatLine, OtlpResourceSpans, OtlpSpan, OtlpToRunRecordsOptions, OtlpTraceRunRecord, ProjectedOtlpSpan, ReplayCache, ReplayCacheEntry, ReplayCacheMissError, ReplayCacheStats, ReplayFetchOptions, SPAN_KIND_ATTR_KEYS, SpanNotFoundError, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_SUBAGENT_DESCRIPTION, TraceAggregate, TraceAnalystHookOptions, TraceFileMissingError, TraceInsightContext, TraceInsightFinding, TraceInsightPanelRole, TraceInsightPromptInput, TraceInsightQualityGate, TraceInsightQuestion, TraceInsightReadiness, TraceInsightSuite, TraceInsightTask, TraceNotFoundError, TraceStoreSource, TraceStoreToOtlpOptions, TracesToOtlpResult, 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, scoreTraceInsightReadiness, stringField, tokenizeDomainWords, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceSpanKindToOpenInferenceKind } from './traces.js';
@@ -30,8 +30,8 @@ import { a as AnalyzeTracesInput, A as AnalyzeTracesOptions, b as AnalyzeTracesR
 export { c as AnalyzeTracesTurnSnapshot, d as analyzeTraces } from './analyst-C8HHvfJp.js';
 import { a as TraceAnalystSpan } from './store-C1YxJDEK.js';
 export { D as DEFAULT_TRACE_ANALYST_BUDGETS, b as DatasetOverview, E as ErrorCluster, Q as QueryTracesPage, S as SearchSpanResult, c as SearchTraceResult, d as SpanMatchRecord, e as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, T as TraceAnalysisStore, f as TraceAnalystByteBudgets, g as TraceAnalystFilters, h as TraceAnalystSpanKind, i as TraceAnalystSpanStatus, j as TraceAnalystTraceSummary, V as ViewSpansResult, k as ViewTraceOversized, l as ViewTraceResult } from './store-C1YxJDEK.js';
-import { a as JudgeConfig, J as JudgeScore, S as Scenario$1, g as Gate } from './types-Bihq6-a3.js';
-import { A as AnalyzeRunsOptions } from './analyze-runs-rF2eGxZ_.js';
+import { a as JudgeConfig, J as JudgeScore, S as Scenario$1, g as Gate } from './types-Mh6I44ub.js';
+import { A as AnalyzeRunsOptions } from './analyze-runs-DJYpep3L.js';
 import { S as SteeringBundle } from './harness-optimizer-mOl9XX_O.js';
 export { D as DEFAULT_HARNESS_OBJECTIVES, H as HarnessAdapter, a as HarnessExperimentConfig, b as HarnessExperimentResult, c as HarnessIntervention, d as HarnessRunRequest, e as HarnessRunResult, f as HarnessScenario, g as HarnessSelection, h as HarnessVariant, i as HarnessVariantReport, M as MeasurementPolicy, j as SteeringDelta, k as SteeringRolePrompt, W as WorkflowTopology, m as mergeSteeringBundle, r as renderSteeringText, l as runHarnessExperiment, s as selectHarnessVariant, n as summarizeHarnessResults } from './harness-optimizer-mOl9XX_O.js';
 import { S as SandboxDriver, H as HarnessConfig, a as SandboxHarnessResult } from './test-graded-scenario-DeODGLra.js';
@@ -40,14 +40,14 @@ import { b as RunScoreWeights, R as RunScore } from './run-critic-CmMf05uV.js';
 export { D as DEFAULT_RUN_SCORE_WEIGHTS, c as RunCritic, d as RunCriticOptions, a as RunTrace, e as aggregateRunScore, f as clamp01 } from './run-critic-CmMf05uV.js';
 import { T as TraceEmitter } from './emitter-C2rqGH_l.js';
 export { R as RunCompleteHook, a as RunCompleteHookContext, S as SpanHandle, b as TraceEmitterOptions, l as llmSpanFromProvider } from './emitter-C2rqGH_l.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-D2t12mMw.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-DqGZg3st.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-0aTmbmQe.js';
 export { F as FileSystemRawProviderSink, a as FileSystemRawProviderSinkOptions, I as InMemoryRawProviderSink, b as InMemoryRawProviderSinkOptions, N as NoopRawProviderSink, P as ProviderRedactor, c as RawProviderDirection, d as RawProviderEvent, R as RawProviderSink, e as RawProviderSinkFilter, f as defaultProviderRedactor, p as providerFromBaseUrl } from './raw-provider-sink-C46HDghv.js';
 export { D as DEFAULT_REDACTION_RULES, b as REDACTION_VERSION, a as RedactionReport, R as RedactionRule, r as redactString, c as redactValue } from './redact-B40YG2M_.js';
 import { T as TraceStore, R as RunFilter } from './store-BcFXE6LG.js';
 export { E as EventFilter, F as FileSystemTraceStore, a as FileSystemTraceStoreOptions, I as InMemoryTraceStore, S as SpanFilter } from './store-BcFXE6LG.js';
 export { D as DEFAULT_FAILURE_RULES, b as FailureClassification, c as FailureContext, d as FailureRule, e as classifyFailure } from './failure-cluster-DH9Flgcf.js';
-export { P as ProjectRuntimeTrajectoryEvidenceOptions, a as RuntimeTrajectoryEvidenceProjection, b as RuntimeTrajectoryEvidenceSummary, c as RuntimeTrajectoryHookEvent, R as RuntimeTrajectoryRecord, d as RuntimeTrajectoryRunRecord, p as parseRuntimeTrajectoryHookEvent, e as projectRuntimeTrajectoryEvidence } from './runtime-trajectory-DZ8ei-Jo.js';
+export { P as ProjectRuntimeTrajectoryEvidenceOptions, a as RuntimeTrajectoryEvidenceProjection, b as RuntimeTrajectoryEvidenceSummary, c as RuntimeTrajectoryHookEvent, R as RuntimeTrajectoryRecord, d as RuntimeTrajectoryRunRecord, p as parseRuntimeTrajectoryHookEvent, e as projectRuntimeTrajectoryEvidence } from './runtime-trajectory-iW9IhV3e.js';
 import { a as BaselineReport } from './baseline-Bbid3WoO.js';
 export { B as BaselineOptions, M as MetricSamples, b as MetricVerdict, T as ToolStats, d as ToolUseMetrics, e as ToolUseOptions, f as compareToBaseline, c as computeToolUseMetrics, i as iqr, w as welchsTTest } from './baseline-Bbid3WoO.js';
 import { a as TrajectoryStep, T as Trajectory } from './trajectory-2TkpSEVh.js';
@@ -59,28 +59,28 @@ export { a as CostChannel, c as CostLedgerEntry, d as CostLedgerSummary, e as Co
 export { D as Direction, O as Objective, P as ParetoResult, c as crowdingDistance, d as dominates, p as paretoFrontier, a as paretoFrontierWithCrowding, s as scalarScore } from './pareto-E-pembql.js';
 export { S as SeriesConvergenceOptions, a as SeriesConvergenceResult, b as analyzeSeries } from './series-convergence-D5OWMBg6.js';
 import { D as DefaultVerdict } from './verdict-C9MlYujm.js';
-import { a as DatasetScenario, b as Dataset } from './dataset-BbGkaN2I.js';
-export { d as DatasetDifficulty, c as DatasetManifest, e as DatasetProvenance, D as DatasetSplit, H as HoldoutLockedError, S as SliceOptions, h as hashScenarios } from './dataset-BbGkaN2I.js';
+import { a as DatasetScenario, b as Dataset } from './dataset-DS7ytHZU.js';
+export { d as DatasetDifficulty, c as DatasetManifest, e as DatasetProvenance, D as DatasetSplit, H as HoldoutLockedError, S as SliceOptions, h as hashScenarios } from './dataset-DS7ytHZU.js';
 export { a as CalibrationResult, c as CandidateScore, C as ContinuousAgreement, d as ContinuousAgreementOptions, b as ContinuousCalibrationResult, G as GoldenItem, P as PositionalBiasResult, S as SelfPreferenceResult, V as VerbosityBiasResult, e as calibrateJudge, f as calibrateJudgeContinuous, g as continuousAgreement, p as positionalBias, s as selfPreference, v as verbosityBias } from './judge-calibration-7C-IDmKr.js';
-export { D as DEFAULT_RED_TEAM_CORPUS, R as RedTeamCase, a as RedTeamCategory, b as RedTeamFinding, c as RedTeamPayload, d as RedTeamReport, r as redTeamDataset, e as redTeamReport, s as scoreRedTeamOutput, t as toolNamesForRun } from './red-team-BWdoyleI.js';
+export { D as DEFAULT_RED_TEAM_CORPUS, R as RedTeamCase, a as RedTeamCategory, b as RedTeamFinding, c as RedTeamPayload, d as RedTeamReport, r as redTeamDataset, e as redTeamReport, s as scoreRedTeamOutput, t as toolNamesForRun } from './red-team-KmmiqBlY.js';
 export { c as CounterfactualContext, C as CounterfactualMutation, d as CounterfactualResult, b as CounterfactualRunner, a as attributeCounterfactuals, r as runCounterfactual } from './counterfactual-DlOz8PBx.js';
 import { a as PrmGrader } from './rubric-Cc6UHvUb.js';
 export { EuRiskClass, GovernanceContext, GovernanceFinding, GovernanceReport, UseCaseSignals, classifyEuAiRisk, euAiActReport, nistAiRmfReport, renderMarkdown, soc2Report, summarize } from './governance/index.js';
 import { b as Layer, S as Severity, L as LayerResult, c as VerifyContext } from './multi-layer-verifier-CI4jdX-q.js';
 export { F as Finding, d as LayerStatus, M as MultiLayerVerifier, a as VerificationReport, V as VerifyOptions, g as gradeSemanticStatus } from './multi-layer-verifier-CI4jdX-q.js';
-import { L as LlmClientOptions } from './llm-client-Bj7g0rqu.js';
-export { d as LlmCallError, b as LlmCallRequest, c as LlmCallResult, e as LlmClient, f as LlmMessage, g as LlmRouteAssertionError, a as LlmRouteRequirements, h as LlmUsage, i as assertLlmRoute, j as backoffMs, k as callLlm, l as callLlmJson, m as isTransientLlmError, p as probeLlm, s as stripFencedJson } from './llm-client-Bj7g0rqu.js';
-export { B as BENCHMARK_SPLIT_SEED, a as BenchmarkAdapter, b as BenchmarkDatasetItem, c as BenchmarkEvaluation, d as benchmarkDeterministicSplit, i as benchmarks } from './index-unCSYRJJ.js';
-export { C as CallbackResearcher, d as CallbackResearcherOptions, e as CampaignFactoryParams, f as CampaignIntegrityPolicy, g as CampaignRunContext, h as CampaignRunOutcome, i as CampaignRunner, j as CampaignScenario, k as CampaignVariant, c as EvalCampaignOptions, b as EvalCampaignResult, E as ExperimentPlan, a as ExperimentResult, l as FailedRun, F as FailureMode, N as NoopResearcher, R as Researcher, S as SteeringChange, r as runEvalCampaign } from './researcher-SDAezfML.js';
-export { G as GainDistributionBin, a as GainDistributionFigureSpec, b as GainDistributionOptions, m as GateDecision, n as GateEvidence, H as HeldOutGate, o as HeldOutGateConfig, q as HeldOutGateRejectionCode, 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-Fc_YFJat.js';
+import { L as LlmClientOptions } from './llm-client-DyqEH4jH.js';
+export { d as LlmCallError, b as LlmCallRequest, c as LlmCallResult, e as LlmClient, f as LlmMessage, g as LlmRouteAssertionError, a as LlmRouteRequirements, h as LlmUsage, i as assertLlmRoute, j as backoffMs, k as callLlm, l as callLlmJson, m as isTransientLlmError, p as probeLlm, s as stripFencedJson } from './llm-client-DyqEH4jH.js';
+export { B as BENCHMARK_SPLIT_SEED, a as BenchmarkAdapter, b as BenchmarkDatasetItem, c as BenchmarkEvaluation, d as benchmarkDeterministicSplit, i as benchmarks } from './index-C2CC_dry.js';
+export { C as CallbackResearcher, d as CallbackResearcherOptions, e as CampaignFactoryParams, f as CampaignIntegrityPolicy, g as CampaignRunContext, h as CampaignRunOutcome, i as CampaignRunner, j as CampaignScenario, k as CampaignVariant, c as EvalCampaignOptions, b as EvalCampaignResult, E as ExperimentPlan, a as ExperimentResult, l as FailedRun, F as FailureMode, N as NoopResearcher, R as Researcher, S as SteeringChange, r as runEvalCampaign } from './researcher-Wc7dx6GM.js';
+export { G as GainDistributionBin, a as GainDistributionFigureSpec, b as GainDistributionOptions, m as GateDecision, n as GateEvidence, H as HeldOutGate, o as HeldOutGateConfig, q as HeldOutGateRejectionCode, 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-QMZVe3P-.js';
 export { L as LockedJsonlAppender } from './testing-C21CHsq2.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';
-import { j as GepaProposerConstraints, b as RunImprovementLoopResult } from './gepa-COlCAkHN.js';
+import { j as GepaProposerConstraints, b as RunImprovementLoopResult } from './gepa-Dn0gtj9n.js';
 export { IntegrityResult, IntegrityViolation, JourneySpec, PerfBaseline, PerfGateResult, PerfRegression, PerfScenario, PerfStat, ScenarioAxes, assertRecordIntegrity, checkRecordIntegrity, expandMatrix, gatePerf, scenarioKey, summarizeRecords } from './perf/index.js';
 export { AgentProfileRuntimeReceipt, ProductBenchmarkArm, ProductBenchmarkArtifactPaths, ProductBenchmarkBudgets, ProductBenchmarkExportOptions, ProductBenchmarkExportResult, ProductBenchmarkManifest, ProductBenchmarkProfileRef, ProductBenchmarkRecord, ProductBenchmarkRepoRef, ProductBenchmarkRunInput, ProductBenchmarkScenario, ProductBenchmarkSingleRunExportOptions, ProductBenchmarkSplit, ProductBenchmarkSubstrateVersions, ProductBenchmarkValidationReport, RuntimeResolution, assertProductBenchmarkRun, buildProductBenchmarkManifest, exportProductBenchmark, exportProductBenchmarkRuns, findProductBenchmarkArtifacts, productBenchmarkIntegrityFailures, productBenchmarkMutableSurfaces, productBenchmarkRepoIdentity, productBenchmarkSplits, readProductBenchmarkManifest, readProductBenchmarkRecords, runRecordToProductBenchmarkRecord, validateProductBenchmarkManifest, validateProductBenchmarkRecord, validateProductBenchmarkRun } from './product-benchmark/index.js';
 import '@ax-llm/ax';
 import 'zod';
-import './insight-report-DumCfEur.js';
+import './insight-report-B4xrdwEK.js';
 import './outcome-store-rnXLEqSn.js';
 
 /**

package/dist/index.js

@@ -9,7 +9,7 @@ import {
   checkBehavioralCanary,
   checkCanaries,
   runBehavioralCanaries
-} from "./chunk-TCPP4Z5Q.js";
+} from "./chunk-GDZAWO2I.js";
 import {
   classifyEuAiRisk,
   euAiActReport,
@@ -49,7 +49,7 @@ import {
   validateProductBenchmarkManifest,
   validateProductBenchmarkRecord,
   validateProductBenchmarkRun
-} from "./chunk-RQNOLV3I.js";
+} from "./chunk-FOUG2VVS.js";
 import {
   CODING_HARNESSES,
   HARNESS_NATIVE_MODEL,
@@ -72,7 +72,7 @@ import {
   llmJudge,
   parseCorrectnessResponse,
   verifyCompletion
-} from "./chunk-YFYIOSNV.js";
+} from "./chunk-BBQILVBH.js";
 import {
   DEFAULT_MUTATION_PRIMITIVES,
   DEFAULT_RED_TEAM_CORPUS,
@@ -96,7 +96,7 @@ import {
   scoreRedTeamOutput,
   surfaceContentHash,
   toolNamesForRun
-} from "./chunk-2B4HPFXN.js";
+} from "./chunk-ZKIL3VOD.js";
 import {
   BackendIntegrityError,
   assertRealBackend,
@@ -106,7 +106,7 @@ import {
   fileVerdictCache,
   inMemoryVerdictCache,
   summarizeBackendIntegrity
-} from "./chunk-VNM52AGA.js";
+} from "./chunk-3PFZBGMR.js";
 import {
   MODEL_PRICING,
   MetricsCollector,
@@ -128,7 +128,7 @@ import {
   computeToolUseMetrics,
   iqr,
   welchsTTest
-} from "./chunk-BXZVBX3D.js";
+} from "./chunk-DBDRR6GF.js";
 import {
   analyzeSeries
 } from "./chunk-BOD4O7OF.js";
@@ -145,7 +145,7 @@ import {
   pytestTestParser,
   runTestGradedScenario,
   vitestTestParser
-} from "./chunk-ZOPLCJSY.js";
+} from "./chunk-HZHNRYHK.js";
 import {
   DEFAULT_COMPLEXITY_WEIGHTS,
   FindingsStore,
@@ -159,7 +159,7 @@ import {
   defaultIsMaterial,
   diffFindings,
   runSemanticConceptJudge
-} from "./chunk-SMCACT4Z.js";
+} from "./chunk-5PK3626Q.js";
 import {
   LockedJsonlAppender,
   Mutex
@@ -167,7 +167,7 @@ import {
 import {
   buildDefaultAnalystRegistry,
   computeTraceMetrics
-} from "./chunk-L5TVEZFT.js";
+} from "./chunk-QRVS7MX4.js";
 import {
   DEFAULT_RUN_SCORE_WEIGHTS,
   POLICY_EDIT_AXES,
@@ -184,7 +184,7 @@ import {
   policyEditsFromFindings,
   scorePolicyEditReadiness,
   validatePolicyEdit
-} from "./chunk-LMSQ6EFA.js";
+} from "./chunk-LOJ2QVCE.js";
 import {
   AnalystRegistry,
   DEFAULT_TRACE_ANALYST_KINDS,
@@ -194,7 +194,7 @@ import {
   KNOWLEDGE_POISONING_KIND_SPEC,
   createTraceAnalystKind,
   renderPriorFindings
-} from "./chunk-FRI6RG3P.js";
+} from "./chunk-2OGPXHOB.js";
 import {
   controlFailureClassFromVerification,
   controlRunToRunRecord,
@@ -205,7 +205,7 @@ import {
   runProposeReview,
   runProposeReviewAsControlLoop,
   scoreFromEvals
-} from "./chunk-BHCFJGL4.js";
+} from "./chunk-K7QEIHHJ.js";
 import {
   allCriticalPassed,
   errorStreakDetector,
@@ -224,10 +224,10 @@ import {
   evaluateReleaseConfidence,
   judgeReplayGate,
   renderReleaseReport
-} from "./chunk-UMEAR2FI.js";
+} from "./chunk-6HFYGEZ3.js";
 import {
   runEvalCampaign
-} from "./chunk-6YVAN6R3.js";
+} from "./chunk-LIEJUH2I.js";
 import {
   LlmCallError,
   LlmClient,
@@ -239,7 +239,7 @@ import {
   isTransientLlmError,
   probeLlm,
   stripFencedJson
-} from "./chunk-CWNP4DV4.js";
+} from "./chunk-FUCQVFMU.js";
 import {
   evaluateInterimReleaseConfidence,
   pairedEvalueSequence
@@ -250,11 +250,11 @@ import {
   paretoChart,
   researchReport,
   summaryTable
-} from "./chunk-CLS3374R.js";
+} from "./chunk-6MFFKPZ4.js";
 import {
   attributeCounterfactuals,
   runCounterfactual
-} from "./chunk-REVYNR6C.js";
+} from "./chunk-6SK5VFYK.js";
 import {
   buildTrajectory
 } from "./chunk-RZTMDUO7.js";
@@ -299,7 +299,7 @@ import {
   weightedMean,
   wilcoxonSignedRank,
   wilson
-} from "./chunk-HYGRFL7C.js";
+} from "./chunk-XMSYF4A7.js";
 import {
   FileSystemTraceStore,
   InMemoryTraceStore,
@@ -330,13 +330,13 @@ import {
   scoreTraceInsightReadiness,
   tokenizeDomainWords,
   traceAnalystOnRunComplete
-} from "./chunk-3722PKPB.js";
+} from "./chunk-V7HNA47Z.js";
 import {
   TRACE_ANALYST_ACTOR_DESCRIPTION,
   TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
   TRACE_ANALYST_SUBAGENT_DESCRIPTION,
   analyzeTraces
-} from "./chunk-2MLIEQSN.js";
+} from "./chunk-RPDDVKI7.js";
 import {
   DEFAULT_REDACTION_RULES,
   REDACTION_VERSION,
@@ -395,12 +395,12 @@ import {
   stringField,
   traceAnalystFunctionGroup,
   traceSpanKindToOpenInferenceKind
-} from "./chunk-YBIGNSCZ.js";
+} from "./chunk-LNQEP766.js";
 import {
   RunIntegrityError,
   assertRunCaptured,
   throwIfRunIncomplete
-} from "./chunk-SBCB6VZY.js";
+} from "./chunk-TT4KNT67.js";
 import {
   FileSystemRawProviderSink,
   InMemoryRawProviderSink,
@@ -415,7 +415,7 @@ import {
   parseRunRecordSafe,
   roundTripRunRecord,
   validateRunRecord
-} from "./chunk-J5MUFXEY.js";
+} from "./chunk-VK6HBGAE.js";
 import {
   TraceEmitter,
   llmSpanFromProvider
@@ -433,7 +433,7 @@ import {
   toAgentProfileJson,
   validateAgentProfileCell,
   verifyAgentProfileCell
-} from "./chunk-AGYDMORK.js";
+} from "./chunk-XJYR7XFV.js";
 import {
   canonicalize,
   evaluateHypothesis,
@@ -450,7 +450,7 @@ import {
   ReplayError,
   ValidationError,
   VerificationError
-} from "./chunk-3BFEG2F6.js";
+} from "./chunk-ONWEPEDO.js";
 import {
   __export
 } from "./chunk-PZ5AY32C.js";

package/dist/{insight-report-DumCfEur.d.ts → insight-report-B4xrdwEK.d.ts}

@@ -1,4 +1,4 @@
-import { G as GainDistributionBin, P as ParetoFigureSpec } from './summary-report-Fc_YFJat.js';
+import { G as GainDistributionBin, P as ParetoFigureSpec } from './summary-report-QMZVe3P-.js';
 import { C as ContinuousAgreement } from './judge-calibration-7C-IDmKr.js';
 
 /**

package/dist/{integrity-D2t12mMw.d.ts → integrity-DqGZg3st.d.ts}

@@ -1,4 +1,4 @@
-import { C as CaptureIntegrityError } from './errors-CzMUYo7b.js';
+import { C as CaptureIntegrityError } from './errors-oeQrLqXC.js';
 import { R as RawProviderSink } from './raw-provider-sink-C46HDghv.js';
 import { T as TraceStore } from './store-BcFXE6LG.js';
 

package/dist/{kind-factory-BLHxwX71.d.ts → kind-factory-DvIGo_cP.d.ts}

@@ -1,7 +1,7 @@
 import { AxAIService, AxFunction } from '@ax-llm/ax';
 import { T as TraceAnalysisStore } from './store-C1YxJDEK.js';
 import { z } from 'zod';
-import { g as AnalystCost, b as AnalystContext, a as Analyst } from './types-DA9yj-Jd.js';
+import { g as AnalystCost, b as AnalystContext, a as Analyst } from './types-D1ytG0Yg.js';
 
 /**
  * Typed Ax output for analyst findings.

package/dist/{llm-client-Bj7g0rqu.d.ts → llm-client-DyqEH4jH.d.ts}

@@ -1,4 +1,4 @@
-import { A as AgentEvalError, C as CaptureIntegrityError } from './errors-CzMUYo7b.js';
+import { A as AgentEvalError, C as CaptureIntegrityError } from './errors-oeQrLqXC.js';
 import { R as RawProviderSink, P as ProviderRedactor } from './raw-provider-sink-C46HDghv.js';
 
 /**

package/dist/meta-eval/index.d.ts

@@ -1,14 +1,14 @@
 export { C as CalibrationBin, a as CalibrationOptions, b as CalibrationPair, c as CalibrationReport, d as CorrelationResult, e as CorrelationStudyOptions, f as CorrelationStudyResult, E as EvalMetricSpec, O as OutcomePair, g as calibrationCurve, h as calibrationFromPairs, i as correlationStudy } from '../calibration-BPmzuVPk.js';
 export { D as DeploymentOutcome, F as FileSystemOutcomeStore, a as FileSystemOutcomeStoreOptions, I as InMemoryOutcomeStore, O as OutcomeFilter, b as OutcomeStore } from '../outcome-store-rnXLEqSn.js';
-export { R as RubricOutcomePair, a as RubricPredictiveValidityInput, b as RubricPredictiveValidityReport, c as RubricRanking, r as rubricPredictiveValidity } from '../rubric-predictive-validity-ZGIlJNce.js';
+export { R as RubricOutcomePair, a as RubricPredictiveValidityInput, b as RubricPredictiveValidityReport, c as RubricRanking, r as rubricPredictiveValidity } from '../rubric-predictive-validity-DPnyG-CE.js';
 import { C as ContinuousAgreement, a as CalibrationResult, b as ContinuousCalibrationResult, c as CandidateScore, G as GoldenItem } from '../judge-calibration-7C-IDmKr.js';
 import { S as SeriesConvergenceOptions, a as SeriesConvergenceResult } from '../series-convergence-D5OWMBg6.js';
 import { C as CorpusAgreementReport } from '../statistics-D88peojY.js';
 import '../store-BcFXE6LG.js';
 import '../schema-m0gsnbt3.js';
-import '../run-record-CPfd1ARZ.js';
+import '../run-record-I-Z3JNvO.js';
 import '@tangle-network/agent-interface';
-import '../errors-CzMUYo7b.js';
+import '../errors-oeQrLqXC.js';
 import '../types-C7DGg5ex.js';
 import '@tangle-network/tcloud';
 

package/dist/meta-eval/index.js

@@ -11,11 +11,11 @@ import {
 } from "../chunk-3RF76KTD.js";
 import {
   rubricPredictiveValidity
-} from "../chunk-QCBB6ZIU.js";
+} from "../chunk-X6NIXVOD.js";
 import {
   pearsonR,
   spearmanR
-} from "../chunk-HYGRFL7C.js";
+} from "../chunk-XMSYF4A7.js";
 import {
   aggregateLlm,
   llmSpans
@@ -23,7 +23,7 @@ import {
 import "../chunk-5BKGXME7.js";
 import {
   ValidationError
-} from "../chunk-3BFEG2F6.js";
+} from "../chunk-ONWEPEDO.js";
 import "../chunk-PZ5AY32C.js";
 
 // src/meta-eval/correlation-study.ts

package/dist/multishot/index.d.ts

@@ -1,8 +1,8 @@
-import { J as JudgeScore } from '../types-Bihq6-a3.js';
+import { J as JudgeScore } from '../types-Mh6I44ub.js';
 import { AgentProfile } from '@tangle-network/agent-interface';
 import { M as MatrixResult } from '../types-BUxNaJ8c.js';
-import '../run-record-CPfd1ARZ.js';
-import '../errors-CzMUYo7b.js';
+import '../run-record-I-Z3JNvO.js';
+import '../errors-oeQrLqXC.js';
 import '../schema-m0gsnbt3.js';
 import '../verdict-C9MlYujm.js';
 
@@ -40,6 +40,45 @@ interface MultishotToolDefinition {
         parameters: Record<string, unknown>;
     };
 }
+/** One chat-completion request the multishot loop issues for a single agent
+ *  (or driver) inference step. Mirrors the OpenAI-compat body the loop would
+ *  otherwise POST to the Tangle router. */
+interface MultishotTransportRequest {
+    model: string;
+    messages: Array<Record<string, unknown>>;
+    tools?: MultishotToolDefinition[];
+    temperature?: number;
+    maxTokens?: number;
+    signal?: AbortSignal;
+}
+interface MultishotTransportToolCall {
+    id: string;
+    type: 'function';
+    function: {
+        name: string;
+        arguments: string;
+    };
+}
+interface MultishotTransportResponse {
+    message: {
+        content?: string | null;
+        tool_calls?: MultishotTransportToolCall[];
+    };
+    usage?: {
+        prompt_tokens?: number;
+        completion_tokens?: number;
+    };
+    /** Actual spend for this call. When omitted, the loop meters cost from
+     *  `usage` via the per-model router estimator (estimateRouterCost). */
+    costUsd?: number;
+}
+/** Execution seam for one leg of the multishot loop. When provided, it
+ *  replaces the internal router HTTP call for that leg — the loop still owns
+ *  turn scheduling, tool dispatch, transcript capture, and cost metering.
+ *  agent-eval has no dependency on agent-runtime; adapt agent-runtime's
+ *  resolveAgentBackend (or any sandbox/cli-bridge/router client) into this
+ *  signature product-side. */
+type MultishotTransport = (req: MultishotTransportRequest) => Promise<MultishotTransportResponse>;
 type MultishotToolExecutor = (args: Record<string, unknown>, ctx: {
     apiKey: string;
     baseUrl: string;
@@ -251,6 +290,12 @@ interface RunMultishotMatrixOptions<TPersona extends MultishotPersona> {
     driverMaxTokens?: number;
     /** Maximum output tokens for each judge response. */
     judgeMaxTokens?: number;
+    /** Execution seam for the agent leg of every cell — replaces the router
+     *  HTTP call when provided (see RunMultishotOptions.agentTransport).
+     *  Judges are unaffected; configure those via MultishotJudges. */
+    agentTransport?: MultishotTransport;
+    /** Execution seam for the simulated-user driver leg of every cell. */
+    driverTransport?: MultishotTransport;
     /** Pass-thru fields. */
     apiKey?: string;
     baseUrl?: string;
@@ -308,10 +353,20 @@ interface RunMultishotOptions<TPersona extends MultishotPersona> {
     driverMaxTokens?: number;
     /** Maximum tool calls the agent may dispatch inside one assistant turn. */
     maxToolDispatches?: number;
+    /** Execution seam for the agent leg. When provided, every agent inference
+     *  step goes through this function instead of the router HTTP call; the
+     *  string levers (agentModel, apiKey, baseUrl) stop applying to that leg.
+     *  apiKey/baseUrl are still resolved for tool executors and any leg
+     *  without an injected transport. */
+    agentTransport?: MultishotTransport;
+    /** Execution seam for the simulated-user driver leg (symmetric to
+     *  agentTransport). Driver model fallback rotation still applies — the
+     *  transport receives each candidate model in turn. */
+    driverTransport?: MultishotTransport;
     apiKey?: string;
     baseUrl?: string;
     signal?: AbortSignal;
 }
 declare function runMultishot<TPersona extends MultishotPersona>(opts: RunMultishotOptions<TPersona>): Promise<MultishotResult>;
 
-export { type ArtifactJudgeInput, type CellCompositeInput, type CellCompositeScore, type ConversationJudgeInput, DEFAULT_CODER_MODEL, DEFAULT_DELEGATE_CODE_TOOL, DEFAULT_DELEGATE_RESEARCH_TOOL, DEFAULT_JUDGE_MODEL, DEFAULT_RESEARCHER_MODEL, type DefaultCoderConfig, type DefaultResearcherConfig, type DefaultToolsBundle, type DefaultToolsConfig, type JudgeConfig, type JudgeDimension, JudgeScore, type MultishotArtifact, MultishotDriverEmptyError, MultishotFatalToolError, type MultishotJudges, type MultishotMessage, type MultishotPersona, type MultishotResult, type MultishotShape, type MultishotToolDefinition, type MultishotToolExecutor, type RouterCompletionRequest, type RouterCompletionResponse, type RouterToolCall, type RunMultishotMatrixOptions, type RunMultishotMatrixResult, type RunMultishotOptions, computeCellComposite, createCodeExecutor, createResearchExecutor, defaultDelegationTools, defaultRouterBaseUrl, estimateRouterCost, renderDimensions, renderJsonFooter, requireRouterApiKey, routerCompletion, runJudge, runMultishot, runMultishotMatrix };
+export { type ArtifactJudgeInput, type CellCompositeInput, type CellCompositeScore, type ConversationJudgeInput, DEFAULT_CODER_MODEL, DEFAULT_DELEGATE_CODE_TOOL, DEFAULT_DELEGATE_RESEARCH_TOOL, DEFAULT_JUDGE_MODEL, DEFAULT_RESEARCHER_MODEL, type DefaultCoderConfig, type DefaultResearcherConfig, type DefaultToolsBundle, type DefaultToolsConfig, type JudgeConfig, type JudgeDimension, JudgeScore, type MultishotArtifact, MultishotDriverEmptyError, MultishotFatalToolError, type MultishotJudges, type MultishotMessage, type MultishotPersona, type MultishotResult, type MultishotShape, type MultishotToolDefinition, type MultishotToolExecutor, type MultishotTransport, type MultishotTransportRequest, type MultishotTransportResponse, type MultishotTransportToolCall, type RouterCompletionRequest, type RouterCompletionResponse, type RouterToolCall, type RunMultishotMatrixOptions, type RunMultishotMatrixResult, type RunMultishotOptions, computeCellComposite, createCodeExecutor, createResearchExecutor, defaultDelegationTools, defaultRouterBaseUrl, estimateRouterCost, renderDimensions, renderJsonFooter, requireRouterApiKey, routerCompletion, runJudge, runMultishot, runMultishotMatrix };

package/dist/multishot/index.js

@@ -259,6 +259,9 @@ async function runMultishot(opts) {
   const tools = opts.tools ?? bundle.tools;
   const executors = opts.toolExecutors ?? bundle.executors;
   const artifactTypeFor = opts.artifactTypeFor ?? bundle.artifactTypeFor;
+  const routerTransport = (req) => routerCompletion({ apiKey, baseUrl, ...req });
+  const agentTransport = opts.agentTransport ?? routerTransport;
+  const driverTransport = opts.driverTransport ?? routerTransport;
   const start = Date.now();
   const transcript = [];
   const artifacts = [];
@@ -275,9 +278,11 @@ async function runMultishot(opts) {
     if (opts.signal?.aborted) throw new Error("multishot aborted");
     let dispatchesThisTurn = 0;
     while (true) {
-      const { message: agentMsg, usage: agentUsage } = await routerCompletion({
-        apiKey,
-        baseUrl,
+      const {
+        message: agentMsg,
+        usage: agentUsage,
+        costUsd: agentCostUsd
+      } = await agentTransport({
         model: agentModel,
         messages: agentMessages,
         tools,
@@ -285,7 +290,7 @@ async function runMultishot(opts) {
         maxTokens: dispatchesThisTurn === 0 ? agentMaxTokens : toolFollowupMaxTokens,
         signal: opts.signal
       });
-      totalCostUsd += estimateRouterCost(agentModel, agentUsage);
+      totalCostUsd += agentCostUsd ?? estimateRouterCost(agentModel, agentUsage);
       const agentText = (agentMsg.content ?? "").trim();
       const agentToolCalls = (agentMsg.tool_calls ?? []).map((tc) => ({
         id: tc.id,
@@ -346,8 +351,7 @@ async function runMultishot(opts) {
     }
     if (turn < maxTurns - 1) {
       const driver = await driverTurn({
-        apiKey,
-        baseUrl,
+        transport: driverTransport,
         persona: opts.persona,
         shape: opts.shape,
         transcript,
@@ -375,9 +379,7 @@ async function driverTurn(opts) {
   }
   for (const model of opts.models) {
     for (let attempt = 0; attempt < 2; attempt++) {
-      const { message, usage } = await routerCompletion({
-        apiKey: opts.apiKey,
-        baseUrl: opts.baseUrl,
+      const { message, usage, costUsd } = await opts.transport({
         model,
         messages: driverMessages,
         temperature: 0.9,
@@ -385,7 +387,8 @@ async function driverTurn(opts) {
         signal: opts.signal
       });
       const content = (message.content ?? "").trim();
-      if (content.length > 0) return { content, costUsd: estimateRouterCost(model, usage) };
+      if (content.length > 0)
+        return { content, costUsd: costUsd ?? estimateRouterCost(model, usage) };
     }
   }
   throw new MultishotDriverEmptyError(opts.turn);
@@ -452,6 +455,8 @@ async function runMultishotMatrix(opts) {
         agentMaxTokens: opts.agentMaxTokens,
         toolFollowupMaxTokens: opts.toolFollowupMaxTokens,
         driverMaxTokens: opts.driverMaxTokens,
+        agentTransport: opts.agentTransport,
+        driverTransport: opts.driverTransport,
         apiKey: opts.apiKey,
         baseUrl: opts.baseUrl
       });