@tangle-network/agent-eval 0.102.0 → 0.103.0

package/dist/hosted/index.d.ts

@@ -1,10 +1,10 @@
-import { M as MutableSurface, j as GateDecision } from '../types-fWqEJm7h.js';
-import { I as InsightReport } from '../insight-report-C02J3q4T.js';
-import '../run-record-DEwidcqn.js';
+import { M as MutableSurface, j as GateDecision } from '../types-DFI_Z-ZL.js';
+import { I as InsightReport } from '../insight-report-k0sRTzKg.js';
+import '../run-record-MRdJ-Kq2.js';
 import '@tangle-network/agent-interface';
 import '../errors-CzMUYo7b.js';
 import '../schema-m0gsnbt3.js';
-import '../summary-report-C4uzRWh8.js';
+import '../summary-report-C0nnxOD8.js';
 import '../failure-cluster-DH9Flgcf.js';
 import '../store-BcFXE6LG.js';
 import '../judge-calibration-0p2QcWNE.js';

package/dist/{index-W96macmS.d.ts → index-B-bFgiAF.d.ts}

@@ -1,4 +1,4 @@
-import { b as RunSplitTag } from './run-record-DEwidcqn.js';
+import { b as RunSplitTag } from './run-record-MRdJ-Kq2.js';
 
 /**
  * Shared types for the reference benchmark wrappers under

package/dist/index.d.ts

@@ -1,13 +1,13 @@
-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-DC8TELh0.js';
-import { R as RunRecord, b as RunSplitTag } from './run-record-DEwidcqn.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 parseRunRecordSafe, y as requireAgentProfileCell, z as roundTripRunRecord, B as toAgentProfileJson, C as validateAgentProfileCell, D as validateRunRecord, E as verifyAgentProfileCell } from './run-record-DEwidcqn.js';
-import { B as BehavioralMetrics } from './semantic-concept-judge-J8xvjdc3.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-J8xvjdc3.js';
-import { l as ChatRequest, p as CreateChatClientOpts } from './types-BEzCBMQD.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-BEzCBMQD.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-Dhrc__SE.js';
-export { C as CreateTraceAnalystKindOpts, a as RawAnalystFinding, c as TraceAnalystGolden, T as TraceAnalystKindSpec, d as createTraceAnalystKind, r as renderPriorFindings } from './kind-factory-OgqQSvLi.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-Dccm9tyA.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-C8RmK9H4.js';
+import { R as RunRecord, b as RunSplitTag } from './run-record-MRdJ-Kq2.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-MRdJ-Kq2.js';
+import { B as BehavioralMetrics } from './semantic-concept-judge-D-IlH5v1.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-D-IlH5v1.js';
+import { l as ChatRequest, p as CreateChatClientOpts } from './types-Dz9cKF0g.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-Dz9cKF0g.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-ZhqsTr4K.js';
+export { C as CreateTraceAnalystKindOpts, a as RawAnalystFinding, c as TraceAnalystGolden, T as TraceAnalystKindSpec, d as createTraceAnalystKind, r as renderPriorFindings } from './kind-factory-D0nk7AKV.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-BDQzzsBU.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';
@@ -18,11 +18,11 @@ import { A as AgentEvalError, J as JudgeError, a as ConfigError } from './errors
 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-BjGZf9YA.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-BjGZf9YA.js';
+import { b as CorrectnessChecker } from './pre-registration-mWG2w8d-.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-mWG2w8d-.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-B1tA6pKu.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-B1tA6pKu.js';
+import { h as ReleaseConfidenceThresholds, f as ReleaseConfidenceScorecard } from './release-report-BQ1Ziyu-.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-BQ1Ziyu-.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-xP-cWc5k.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-fWqEJm7h.js';
-import { A as AnalyzeRunsOptions } from './analyze-runs-BlJRBniC.js';
+import { a as JudgeConfig, J as JudgeScore, S as Scenario$1, g as Gate } from './types-DFI_Z-ZL.js';
+import { A as AnalyzeRunsOptions } from './analyze-runs-Cd-A_K4l.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';
@@ -47,7 +47,7 @@ export { D as DEFAULT_REDACTION_RULES, b as REDACTION_VERSION, a as RedactionRep
 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-OJDaTYHN.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-8w0_jmtR.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';
@@ -70,17 +70,17 @@ import { b as Layer, S as Severity, L as LayerResult, c as VerifyContext } from
 export { F as Finding, d as LayerStatus, M as MultiLayerVerifier, a as VerificationReport, V as VerifyOptions, g as gradeSemanticStatus } from './multi-layer-verifier-DUZXrPDA.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-W96macmS.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-Ba2y1Foi.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-C4uzRWh8.js';
+export { B as BENCHMARK_SPLIT_SEED, a as BenchmarkAdapter, b as BenchmarkDatasetItem, c as BenchmarkEvaluation, d as benchmarkDeterministicSplit, i as benchmarks } from './index-B-bFgiAF.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-B_ODTAJs.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-C0nnxOD8.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-bxuDoaO9.js';
+import { j as GepaProposerConstraints, b as RunImprovementLoopResult } from './gepa-DeyPTlvx.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, ProductBenchmarkManifest, ProductBenchmarkProfileRef, ProductBenchmarkRecord, ProductBenchmarkRepoRef, ProductBenchmarkRunInput, ProductBenchmarkScenario, ProductBenchmarkSplit, ProductBenchmarkSubstrateVersions, ProductBenchmarkValidationReport, RuntimeResolution, findProductBenchmarkArtifacts, productBenchmarkIntegrityFailures, productBenchmarkSplits, readProductBenchmarkManifest, readProductBenchmarkRecords, validateProductBenchmarkManifest, validateProductBenchmarkRecord, validateProductBenchmarkRun } from './product-benchmark/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-C02J3q4T.js';
+import './insight-report-k0sRTzKg.js';
 import './outcome-store-rnXLEqSn.js';
 
 /**
@@ -5271,6 +5271,10 @@ interface ReflectionProposal {
     rationale: string;
     payload: unknown;
 }
+/**
+ * Parse the model's JSON response back into proposals. Tolerates markdown
+ * fences and surrounding prose. Returns at most `maxProposals`.
+ */
 declare function parseReflectionResponse(raw: string, maxProposals?: number): ReflectionProposal[];
 
 /**

package/dist/index.js

@@ -9,7 +9,7 @@ import {
   checkBehavioralCanary,
   checkCanaries,
   runBehavioralCanaries
-} from "./chunk-JCUREYF5.js";
+} from "./chunk-U3IDYATS.js";
 import {
   classifyEuAiRisk,
   euAiActReport,
@@ -34,15 +34,22 @@ import {
   summarizeRecords
 } from "./chunk-STGVSCDH.js";
 import {
+  assertProductBenchmarkRun,
+  buildProductBenchmarkManifest,
+  exportProductBenchmark,
+  exportProductBenchmarkRuns,
   findProductBenchmarkArtifacts,
   productBenchmarkIntegrityFailures,
+  productBenchmarkMutableSurfaces,
+  productBenchmarkRepoIdentity,
   productBenchmarkSplits,
   readProductBenchmarkManifest,
   readProductBenchmarkRecords,
+  runRecordToProductBenchmarkRecord,
   validateProductBenchmarkManifest,
   validateProductBenchmarkRecord,
   validateProductBenchmarkRun
-} from "./chunk-63MBSQTX.js";
+} from "./chunk-RQNOLV3I.js";
 import {
   CODING_HARNESSES,
   HARNESS_NATIVE_MODEL,
@@ -65,7 +72,7 @@ import {
   llmJudge,
   parseCorrectnessResponse,
   verifyCompletion
-} from "./chunk-CMJSTXUR.js";
+} from "./chunk-IXOV77YF.js";
 import {
   DEFAULT_MUTATION_PRIMITIVES,
   DEFAULT_RED_TEAM_CORPUS,
@@ -89,7 +96,7 @@ import {
   scoreRedTeamOutput,
   surfaceContentHash,
   toolNamesForRun
-} from "./chunk-52CCCXU3.js";
+} from "./chunk-NTVWIH24.js";
 import {
   BackendIntegrityError,
   assertRealBackend,
@@ -99,7 +106,7 @@ import {
   fileVerdictCache,
   inMemoryVerdictCache,
   summarizeBackendIntegrity
-} from "./chunk-LSCBODPQ.js";
+} from "./chunk-XKA6ZGEY.js";
 import {
   MODEL_PRICING,
   MetricsCollector,
@@ -198,7 +205,7 @@ import {
   runProposeReview,
   runProposeReviewAsControlLoop,
   scoreFromEvals
-} from "./chunk-BOETF6BU.js";
+} from "./chunk-B2TMQM62.js";
 import {
   allCriticalPassed,
   errorStreakDetector,
@@ -220,7 +227,7 @@ import {
 } from "./chunk-VDGPPGE3.js";
 import {
   runEvalCampaign
-} from "./chunk-4LWD6GC7.js";
+} from "./chunk-6FIAJHCU.js";
 import {
   LlmCallError,
   LlmClient,
@@ -323,7 +330,7 @@ import {
   scoreTraceInsightReadiness,
   tokenizeDomainWords,
   traceAnalystOnRunComplete
-} from "./chunk-PMF5WIBX.js";
+} from "./chunk-7RBJANJD.js";
 import {
   TRACE_ANALYST_ACTOR_DESCRIPTION,
   TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
@@ -404,10 +411,11 @@ import {
 import {
   RunRecordValidationError,
   isRunRecord,
+  modelHasSnapshot,
   parseRunRecordSafe,
   roundTripRunRecord,
   validateRunRecord
-} from "./chunk-G6S73VA7.js";
+} from "./chunk-2NSLDY4B.js";
 import {
   TraceEmitter,
   llmSpanFromProvider
@@ -5489,7 +5497,7 @@ function matchText(actual, matcher) {
 }
 function resolveToolName(span) {
   if (typeof span.toolName === "string") return span.toolName;
-  const fromAttr = span.attributes?.["tool.name"] ?? span.attributes?.["toolName"];
+  const fromAttr = span.attributes?.["tool.name"] ?? span.attributes?.toolName;
   if (typeof fromAttr === "string") return fromAttr;
   if (span.kind === "tool" && typeof span.name === "string") return span.name;
   return void 0;
@@ -10143,6 +10151,7 @@ export {
   assertLlmRoute,
   assertModelsServed,
   assertNoHiddenLeak,
+  assertProductBenchmarkRun,
   assertRealBackend,
   assertRecordIntegrity,
   assertReleaseConfidence,
@@ -10168,6 +10177,7 @@ export {
   buildAgreementJudge,
   buildDefaultAnalystRegistry,
   buildDriverSystemPrompt,
+  buildProductBenchmarkManifest,
   buildReflectionPrompt,
   buildReviewerPrompt,
   buildTraceAnalystTools,
@@ -10282,6 +10292,8 @@ export {
   expandMatrix,
   expandProfileAxes,
   expectAgent,
+  exportProductBenchmark,
+  exportProductBenchmarkRuns,
   exportRewardModel,
   exportRunAsOtlp,
   extractAssetUrls,
@@ -10392,6 +10404,7 @@ export {
   mergeLayerResults,
   mergeSteeringBundle,
   modelDescriptionBits,
+  modelHasSnapshot,
   modelPriceKey,
   mulberry32,
   multiToolchainLayer,
@@ -10435,6 +10448,8 @@ export {
   printDriverSummary,
   probeLlm,
   productBenchmarkIntegrityFailures,
+  productBenchmarkMutableSurfaces,
+  productBenchmarkRepoIdentity,
   productBenchmarkSplits,
   profile_exports as profile,
   projectOtlpFlatLine,
@@ -10496,6 +10511,7 @@ export {
   runLiveProof,
   runProposeReview,
   runProposeReviewAsControlLoop,
+  runRecordToProductBenchmarkRecord,
   runReferenceReplay,
   runScore,
   runSelfPlay,