@tangle-network/agent-eval 0.136.0 → 0.138.0

package/dist/index.js

@@ -1,46 +1,48 @@
 import { t as __exportAll } from "./rolldown-runtime-8H4AJuhK.js";
-import { _ as observeAll, a as jsonlReviewStore, c as scoreFromEvals, d as runAgentControlLoop, f as stopOnNoProgress, g as noProgressDetector, h as errorStreakDetector, i as inMemoryReviewStore, l as allCriticalPassed, m as subjectiveEval, n as runProposeReviewAsControlLoop, o as runProposeReview, p as stopOnRepeatedAction, r as createLlmReviewer, s as controlRunToRunRecord, t as controlFailureClassFromVerification, u as objectiveEval, v as repeatedActionDetector, y as evaluateActionPolicy } from "./propose-review-control-SQ-n9-We.js";
-import { a as NotFoundError, c as VerificationError, i as JudgeError, n as CaptureIntegrityError, o as ReplayError, r as ConfigError, s as ValidationError, t as AgentEvalError } from "./errors-8YnH8WlF.js";
-import { _ as verifyManifest, a as assertRunAgentProfileCell, c as groupRunsByAgentProfileCell, d as validateAgentProfileCell, f as verifyAgentProfileCell, g as signManifest, h as hashJson, i as agentProfileCellKey, l as requireAgentProfileCell, m as evaluateHypothesis, n as AgentProfileCellValidationError, o as buildAgentInterfaceProfileCell, p as canonicalize, r as agentProfileCellHashMaterial, s as buildAgentProfileCell, t as AGENT_PROFILE_KINDS, u as toAgentProfileJson } from "./agent-profile-cell-OhuTee9n.js";
-import { F as createChatClient, I as createAnalystAi, P as computeTraceMetrics, a as KNOWLEDGE_GAP_KIND_SPEC, d as renderUpstreamFindings, i as KNOWLEDGE_POISONING_KIND_SPEC, l as createTraceAnalystKind, n as AnalystRegistry, o as IMPROVEMENT_KIND_SPEC, r as DEFAULT_TRACE_ANALYST_KINDS, s as FAILURE_MODE_KIND_SPEC, t as buildDefaultAnalystRegistry, u as renderPriorFindings } from "./default-registry-CHmdy2An.js";
+import { _ as observeAll, a as jsonlReviewStore, c as scoreFromEvals, d as runAgentControlLoop, f as stopOnNoProgress, g as noProgressDetector, h as errorStreakDetector, i as inMemoryReviewStore, l as allCriticalPassed, m as subjectiveEval, n as runProposeReviewAsControlLoop, o as runProposeReview, p as stopOnRepeatedAction, r as createLlmReviewer, s as controlRunToRunRecord, t as controlFailureClassFromVerification, u as objectiveEval, v as repeatedActionDetector, y as evaluateActionPolicy } from "./propose-review-control-DLXz4FCX.js";
+import { a as LimitExceededError, c as ValidationError, i as JudgeError, l as VerificationError, n as CaptureIntegrityError, o as NotFoundError, r as ConfigError, s as ReplayError, t as AgentEvalError } from "./errors-D-LKuDhb.js";
+import { _ as verifyManifest, a as assertRunAgentProfileCell, c as groupRunsByAgentProfileCell, d as validateAgentProfileCell, f as verifyAgentProfileCell, g as signManifest, h as hashJson, i as agentProfileCellKey, l as requireAgentProfileCell, m as evaluateHypothesis, n as AgentProfileCellValidationError, o as buildAgentInterfaceProfileCell, p as canonicalize, r as agentProfileCellHashMaterial, s as buildAgentProfileCell, t as AGENT_PROFILE_KINDS, u as toAgentProfileJson } from "./agent-profile-cell-CbfBm2g6.js";
+import { C as renderUpstreamFindings, G as createChatClient, K as createAnalystAi, S as renderPriorFindings, W as computeTraceMetrics, a as analystFindingDigest, b as emitControlIntegrityFindings, c as completedAnalystReviewQuality, d as validateAnalystReviewDecisions, f as DEFAULT_TRACE_ANALYST_KINDS, g as FAILURE_MODE_KIND_SPEC, h as IMPROVEMENT_KIND_SPEC, i as assertExactRegistryRunOpts, l as readAnalystReview, m as KNOWLEDGE_GAP_KIND_SPEC, n as AnalystRegistry, o as analystRunDigest, p as KNOWLEDGE_POISONING_KIND_SPEC, r as ExactAnalystRunExecutionError, s as assertUniqueFindingIds, t as buildDefaultAnalystRegistry, u as snapshotAnalystRun, v as CONTROL_INTEGRITY_ANALYST, x as createTraceAnalystKind, y as ControlIntegrityAnalyst } from "./default-registry-lp5R0lve.js";
 import { a as estimateTokens, i as estimateCost, n as MetricsCollector, o as isModelPriced, r as TokenCounter, s as resolveModelPricing, t as MODEL_PRICING } from "./metrics-C9YY1OcL.js";
-import { a as CostLedgerPersistenceError, c as costForTokenPricing, i as CostLedger, l as costForUsage, n as CostCallConflictError, o as CostReceiptCaptureError, r as CostCeilingReachedError, s as CostReservationExceededError, t as CostAccountingIncompleteError, u as modelPriceKey } from "./cost-ledger-DHAjwNj7.js";
+import { a as CostLedgerPersistenceError, c as costForTokenPricing, i as CostLedger, l as costForUsage, n as CostCallConflictError, o as CostReceiptCaptureError, r as CostCeilingReachedError, s as CostReservationExceededError, t as CostAccountingIncompleteError, u as modelPriceKey } from "./cost-ledger-CHDLA0Ss.js";
 import { a as providerFromBaseUrl, i as defaultProviderRedactor, n as InMemoryRawProviderSink, r as NoopRawProviderSink, t as FileSystemRawProviderSink } from "./raw-provider-sink-BQd7mzyT.js";
-import { a as assertLlmRoute, c as callLlmJson, d as isTransientLlmError, f as maximumChargeForLlmRequest, i as LlmRouteAssertionError, l as costReceiptFromLlm, m as stripFencedJson, n as LlmClient, o as backoffMs, p as probeLlm, r as LlmResponseError, s as callLlm, t as LlmCallError, u as costReceiptFromLlmError } from "./llm-client-DHx8pzyJ.js";
+import { a as assertLlmRoute, c as callLlmJson, d as isTransientLlmError, f as maximumChargeForLlmRequest, i as LlmRouteAssertionError, l as costReceiptFromLlm, m as stripFencedJson, n as LlmClient, o as backoffMs, p as probeLlm, r as LlmResponseError, s as callLlm, t as LlmCallError, u as costReceiptFromLlmError } from "./llm-client-Cj3c7PEm.js";
 import { INPUT_VALUE, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_CACHE_WRITE_TOKENS, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, LLM_CONTEXT_TOKENS, 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, LLM_REASONING_TOKENS, LLM_REASONING_TOKEN_ATTR_KEYS, OPENINFERENCE_SPAN_KIND, OUTPUT_VALUE, RUN_COST_ATTR_KEYS, SPAN_KIND_ATTR_KEYS, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, applyLlmSpanOtlpAttributes, asNumber, contextInputTokens, firstNumberAttr } from "./trace-attributes.js";
-import { C as stringField, D as traceSpanKindToOpenInferenceKind, E as isOtlpModelCall, T as classifyOtlpSpanRole, _ as extractOtlpAttributes, a as SpanNotFoundError, b as projectOtlpFlatLine, c as TraceNotFoundError, d as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, h as asString, i as OtlpFileTraceStore, l as toolSpansToTraceAnalysisStore, n as traceAnalystFunctionGroup, o as ToolTraceMissingError, s as TraceFileMissingError, t as buildTraceAnalystTools, u as DEFAULT_TRACE_ANALYST_BUDGETS, v as firstStringAttr, w as applyToolSpanOtlpAttributes, x as readOtlpStatus, y as inferOtlpKind } from "./tools-D8yTtNSN.js";
-import { a as clamp01, c as makeFinding, i as aggregateRunScore, l as makeProposalFinding, r as DEFAULT_RUN_SCORE_WEIGHTS, s as computeFindingId } from "./proposal-findings-DCawte-y.js";
-import { n as mapConcurrent, t as Mutex } from "./concurrency-MUjT7VjM.js";
-import { a as RunCritic, d as defaultIsMaterial, f as diffFindings, i as runSemanticConceptJudge, n as SEMANTIC_CONCEPT_JUDGE_VERSION, o as SKILL_USAGE_ANALYST, p as LockedJsonlAppender, r as createSemanticConceptJudge, s as SkillUsageAnalyst, t as DEFAULT_COMPLEXITY_WEIGHTS, u as FindingsStore } from "./semantic-concept-judge-Btozx3Vc.js";
-import { At as dominates, B as surfaceContentHash, Bt as assertRealAgentReceipts, Ct as llmJudge, D as runCanaries, Dt as fileVerdictCache, Et as contentHash, Ft as decidePairedPromotion, G as DEFAULT_MUTATION_PRIMITIVES, Ht as summarizeAgentReceiptIntegrity, It as pairedDecisionShape, K as buildReflectionPrompt, Lt as minimumPairsForPairedDeltaTest, Mt as paretoFrontierWithCrowding, Nt as scalarScore, Ot as inMemoryVerdictCache, Rt as pairedDeltaTest, St as hashScenarios, Tt as canonicalJson, Ut as summarizeBackendIntegrity, Vt as assertRealBackend, Wt as JudgeParseError, _t as redTeamReport, bt as Dataset, dt as REFERENCE_EQUIVALENCE_INPUT_LIMITS, ft as REFERENCE_EQUIVALENCE_JUDGE_VERSION, gt as redTeamDataset, ht as DEFAULT_RED_TEAM_CORPUS, jt as paretoFrontier, kt as crowdingDistance, mt as runReferenceEquivalenceJudge, pt as createReferenceEquivalenceJudge, q as parseReflectionResponse, vt as scoreRedTeamOutput, wt as cachedJudge, xt as HoldoutLockedError, yt as toolNamesForRun, zt as BackendIntegrityError } from "./skillopt-optimization-method-0UmPD6aP.js";
-import { $ as selfPreference, A as pairedRiskDifference, B as requiredSampleSize, C as mulberry32, D as pairedCohensDz, E as pairedBootstrap, F as partialCredit, G as wilson, H as weightedComposite, I as passAtK, J as normalCdf, K as studentTCdf, L as pearsonR, M as pairedRiskDifferenceScore, N as pairedSignTest, O as pairedDeltaTieFraction, P as pairedTTest, Q as positionalBias, R as ranks, S as mcnemarRequiredN, T as pairedBinaryScale, U as weightedMean, V as spearmanR, W as wilcoxonSignedRank, X as calibrateJudgeContinuous, Y as calibrateJudge, Z as continuousAgreement, _ as interpretCliffs, a as MANN_WHITNEY_EXACT_MAX_WORK, b as mcnemar, c as bonferroni, d as confidenceInterval, et as verbosityBias, f as corpusInterRaterAgreement, g as interRaterReliability, h as holm, i as MANN_WHITNEY_EXACT_MAX_STATES, j as pairedRiskDifferenceExact, k as pairedMde, l as cliffsDelta, m as eProcess, n as DECISION_PAIRED_DELTA_STATISTIC, o as WILCOXON_EXACT_MAX_N, p as corpusInterRaterAgreementFromJudgeScores, r as DEFAULT_PERMUTATIONS, s as benjaminiHochberg, t as BOOTSTRAP_GATE_MIN_N, u as cohensD, v as isBinaryOutcomeVector, w as normalizeScores, x as mcnemarPower, y as mannWhitneyU, z as requiredPairedSampleSize } from "./statistics-CnGCLLqc.js";
-import { n as pairArms, r as pairRunRecords, t as comparePairedArms } from "./paired-arms-BbFKrAU-.js";
-import { n as llmSpanFromProvider, t as TraceEmitter } from "./emitter-CPBAhxum.js";
+import { A as classifyOtlpSpanRole, C as firstStringAttr, E as readOtlpStatus, M as traceSpanKindToOpenInferenceKind, O as stringField, S as extractOtlpAttributes, T as projectOtlpFlatLine, _ as TraceFileMissingError, a as createBoundedTraceAnalysisStore, b as asString, d as TRACE_ANALYSIS_LIMITS, f as SpanNotFoundError, g as TraceFileMalformedError, h as TraceAnalysisValidationError, i as otlpTextToTraceAnalysisStore, j as isOtlpModelCall, k as applyToolSpanOtlpAttributes, l as DEFAULT_TRACE_ANALYST_BUDGETS, m as TraceAnalysisStoreContractError, n as OtlpFileTraceStore, p as TraceAnalysisLimitError, u as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, v as TraceFileTooLargeError, w as inferOtlpKind, y as TraceNotFoundError } from "./store-otlp-BenKynPE.js";
+import { h as makeProposalFinding, m as makeFinding, p as computeFindingId } from "./single-run-lock-D_bS5xhj.js";
+import { f as Mutex, p as mapConcurrent } from "./ledger-core-C0Yx1I14.js";
+import { i as traceAnalystFunctionGroup, n as buildTraceAnalysisToolDescriptors, r as buildTraceAnalystTools, t as TRACE_ANALYST_TOOL_NAMESPACE } from "./tools-DZGdROtG.js";
+import { c as SUPERVISOR_RUN_INTEGRITY_SCHEMA, n as supervisorRunRolloutLines, t as analyzeSupervisorRunIntegrity } from "./integrity-CCXTftiL.js";
+import { c as assertMintedLines, f as isRolloutLine, i as ROLLOUT_SCHEMA, m as validateRolloutLine, p as isTrainableSplit, s as assertMinted, u as assertRolloutLine } from "./schema-C6DW4ZHR.js";
 import { a as scoreOrigin, n as observedScore, o as trainingReward, r as observedSplitScore, s as trainingScore, t as isRealnessGated } from "./reward-nw2xZGZG.js";
+import { a as clamp01, i as aggregateRunScore, r as DEFAULT_RUN_SCORE_WEIGHTS } from "./proposal-findings-2GIUo1et.js";
+import { a as RunCritic, d as defaultIsMaterial, f as diffFindings, i as runSemanticConceptJudge, n as SEMANTIC_CONCEPT_JUDGE_VERSION, o as SKILL_USAGE_ANALYST, p as LockedJsonlAppender, r as createSemanticConceptJudge, s as SkillUsageAnalyst, t as DEFAULT_COMPLEXITY_WEIGHTS, u as FindingsStore } from "./semantic-concept-judge-DYXDPZW0.js";
+import { At as pairedDeltaTest, Ct as paretoFrontier, D as runCanaries, Dt as decidePairedPromotion, Ft as summarizeBackendIntegrity, G as buildReflectionPrompt, It as JudgeParseError, K as parseReflectionResponse, Mt as assertRealAgentReceipts, Nt as assertRealBackend, Ot as pairedDecisionShape, Pt as summarizeAgentReceiptIntegrity, St as dominates, Tt as scalarScore, W as DEFAULT_MUTATION_PRIMITIVES, _t as canonicalJson, at as createReferenceEquivalenceJudge, bt as inMemoryVerdictCache, ct as redTeamDataset, dt as toolNamesForRun, ft as Dataset, gt as cachedJudge, ht as llmJudge, it as REFERENCE_EQUIVALENCE_JUDGE_VERSION, jt as BackendIntegrityError, kt as minimumPairsForPairedDeltaTest, lt as redTeamReport, mt as hashScenarios, ot as runReferenceEquivalenceJudge, pt as HoldoutLockedError, rt as REFERENCE_EQUIVALENCE_INPUT_LIMITS, st as DEFAULT_RED_TEAM_CORPUS, ut as scoreRedTeamOutput, vt as contentHash, wt as paretoFrontierWithCrowding, xt as crowdingDistance, yt as fileVerdictCache, z as surfaceContentHash } from "./skillopt-optimization-method-CjKMZy0d.js";
+import { $ as selfPreference, A as pairedRiskDifference, B as requiredSampleSize, C as mulberry32, D as pairedCohensDz, E as pairedBootstrap, F as partialCredit, G as wilson, H as weightedComposite, I as passAtK, J as normalCdf, K as studentTCdf, L as pearsonR, M as pairedRiskDifferenceScore, N as pairedSignTest, O as pairedDeltaTieFraction, P as pairedTTest, Q as positionalBias, R as ranks, S as mcnemarRequiredN, T as pairedBinaryScale, U as weightedMean, V as spearmanR, W as wilcoxonSignedRank, X as calibrateJudgeContinuous, Y as calibrateJudge, Z as continuousAgreement, _ as interpretCliffs, a as MANN_WHITNEY_EXACT_MAX_WORK, b as mcnemar, c as bonferroni, d as confidenceInterval, et as verbosityBias, f as corpusInterRaterAgreement, g as interRaterReliability, h as holm, i as MANN_WHITNEY_EXACT_MAX_STATES, j as pairedRiskDifferenceExact, k as pairedMde, l as cliffsDelta, m as eProcess, n as DECISION_PAIRED_DELTA_STATISTIC, o as WILCOXON_EXACT_MAX_N, p as corpusInterRaterAgreementFromJudgeScores, r as DEFAULT_PERMUTATIONS, s as benjaminiHochberg, t as BOOTSTRAP_GATE_MIN_N, u as cohensD, v as isBinaryOutcomeVector, w as normalizeScores, x as mcnemarPower, y as mannWhitneyU, z as requiredPairedSampleSize } from "./statistics-ByxzSiOM.js";
+import { n as pairArms, r as pairRunRecords, t as comparePairedArms } from "./paired-arms-iZ08VFMN.js";
+import { n as llmSpanFromProvider, t as TraceEmitter } from "./emitter-CPBAhxum.js";
 import { a as isRetrievalSpan, i as isLlmSpan, n as TRACE_SCHEMA_VERSION, o as isSandboxSpan, r as isJudgeSpan, s as isToolSpan, t as FAILURE_CLASSES } from "./schema-CRhEY1SO.js";
-import { a as roundTripRunRecord, i as parseRunRecordSafe, n as isRunRecord, o as runTaskScore, r as modelHasSnapshot, s as validateRunRecord, t as RunRecordValidationError } from "./run-record-BIwU2wdV.js";
-import { a as evaluateReleaseConfidence, i as assertReleaseConfidence, n as bootstrapCi, r as judgeReplayGate, t as renderReleaseReport } from "./release-report-DooPguBc.js";
-import { c as assertMintedLines, f as isRolloutLine, i as ROLLOUT_SCHEMA, m as validateRolloutLine, p as isTrainableSplit, s as assertMinted, u as assertRolloutLine } from "./schema-C6DW4ZHR.js";
+import { a as roundTripRunRecord, i as parseRunRecordSafe, n as isRunRecord, o as runTaskScore, r as modelHasSnapshot, s as validateRunRecord, t as RunRecordValidationError } from "./run-record-vRgqWmJw.js";
+import { a as evaluateReleaseConfidence, i as assertReleaseConfidence, n as bootstrapCi, r as judgeReplayGate, t as renderReleaseReport } from "./release-report-B5XPBvAU.js";
 import { i as toRewardRows, r as toJsonl, s as toSftRows } from "./exporters-q9iL-2Jf.js";
-import { a as toHarborTrajectories, i as relabelImportedSplit, n as HARBOR_IMPORT_GAP, o as toHarborTrajectory, r as fromHarborTrajectory, t as ATIF_SCHEMA_VERSION } from "./rollout-DLSUIWLu.js";
+import { a as toHarborTrajectories, i as relabelImportedSplit, n as HARBOR_IMPORT_GAP, o as toHarborTrajectory, r as fromHarborTrajectory, t as ATIF_SCHEMA_VERSION } from "./rollout-DQFl0UXA.js";
 import { t as buildTrajectory } from "./trajectory-D_7rLrvE.js";
-import { n as unmintableReasons, t as mintRolloutRows } from "./mint-DyRUc9k6.js";
-import { D as showMeasured, E as isUnavailable, S as rollupSupervisorRuns, T as SUPERVISOR_RUN_SCHEMA, _ as claudeCodeSupervisorRunReader, f as renderSupervisorRunHeadline, l as writeSupervisorRunReport, n as analyzeSupervisorRun, p as renderSupervisorRunMarkdown, t as supervisorRunRolloutLines, v as readClaudeCodeSupervisorRun, y as analyzeSupervisorRunSources } from "./supervisor-run-Dr5HnTup.js";
-import { n as TRACE_ANALYST_ACTOR_DESCRIPTION, r as TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, t as analyzeTraces } from "./analyst-j5je5J7c.js";
-import { C as planTraceInsightQuestions, E as traceAnalystOnRunComplete, S as inferDomainKeywords, T as tokenizeDomainWords, _ as buildTraceInsightContext, a as convertTraceStoresToOtlp, b as describeTraceInsightScope, c as otelRunCompleteHook, d as captureFetchToRawSink, f as otlpRowsToRunRecords, g as flattenOtlpExportToNdjson, h as otlpToTraceRunRecords, i as iterateRawCalls, l as OTEL_AGENT_EVAL_SCOPE, m as otlpToRunRecords, n as ReplayCacheMissError, o as createOtelExporter, p as otlpRowsToTraceRunRecords, r as createReplayFetch, s as createOtelTracingStore, t as ReplayCache, u as exportRunAsOtlp, v as buildTraceInsightPrompt, w as scoreTraceInsightReadiness, x as domainEvidencePattern, y as defaultTraceInsightPanel } from "./replay-C6wRg47C.js";
-import { n as extractUsageFromResponse, r as extractUsageFromSse, t as extractUsage } from "./extract-usage-DIQpN-ww.js";
-import { B as harnessAxisOf, F as HARNESS_NATIVE_MODEL, G as parseCorrectnessResponse, H as completionVerdict, I as agentProfileHash, K as verifyCompletion, L as agentProfileId, P as CODING_HARNESSES, R as agentProfileModelId, U as createLlmCorrectnessChecker, V as extractProducedState, W as createTokenRecallChecker, z as expandProfileAxes } from "./campaign-Dz8uQnhC.js";
-import { n as iqr, r as welchsTTest, t as compareToBaseline } from "./baseline-BUeFcgrn.js";
+import { n as unmintableReasons, t as mintRolloutRows } from "./mint-Ctwk079K.js";
+import { C as SUPERVISOR_RUN_SCHEMA, T as showMeasured, _ as readClaudeCodeSupervisorRun, b as rollupSupervisorRuns, c as writeSupervisorRunReport, d as renderSupervisorRunHeadline, f as renderSupervisorRunMarkdown, g as claudeCodeSupervisorRunReader, t as analyzeSupervisorRun, v as analyzeSupervisorRunSources, w as isUnavailable } from "./supervisor-run-B2EWUmQY.js";
+import { A as TRACE_ANALYST_ACTOR_DESCRIPTION, C as domainEvidencePattern, D as tokenizeDomainWords, E as scoreTraceInsightReadiness, O as traceAnalystOnRunComplete, S as describeTraceInsightScope, T as planTraceInsightQuestions, _ as otlpToTraceRunRecords, a as convertTraceStoresToOtlp, b as buildTraceInsightPrompt, c as otelRunCompleteHook, d as captureFetchToRawSink, f as ToolTraceMissingError, g as otlpToRunRecords, h as otlpRowsToTraceRunRecords, i as iterateRawCalls, j as TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, k as analyzeTraces, l as OTEL_AGENT_EVAL_SCOPE, m as otlpRowsToRunRecords, n as ReplayCacheMissError, o as createOtelExporter, p as toolSpansToTraceAnalysisStore, r as createReplayFetch, s as createOtelTracingStore, t as ReplayCache, u as exportRunAsOtlp, v as flattenOtlpExportToNdjson, w as inferDomainKeywords, x as defaultTraceInsightPanel, y as buildTraceInsightContext } from "./replay-Cb-4Vf0k.js";
+import { n as extractUsageFromResponse, r as extractUsageFromSse, t as extractUsage } from "./extract-usage-CS391dOE.js";
+import { B as agentProfileModelId, G as createLlmCorrectnessChecker, H as harnessAxisOf, I as CODING_HARNESSES, J as verifyCompletion, K as createTokenRecallChecker, L as HARNESS_NATIVE_MODEL, R as agentProfileHash, U as extractProducedState, V as expandProfileAxes, W as completionVerdict, q as parseCorrectnessResponse, z as agentProfileId } from "./campaign-BIBS-NHV.js";
+import { n as iqr, r as welchsTTest, t as compareToBaseline } from "./baseline-C-GocmIW.js";
 import { a as judgeSpans, c as runsForScenario, i as hasCapturedToolArgs, l as toolSpans, n as argHash, o as llmSpans, r as groupBy, s as runFailureClass, t as aggregateLlm } from "./query-Di7eEQ79.js";
-import { a as checkBehavioralCanary, i as canaryLeakView, o as checkCanaries, r as HoldoutAuditor, s as runBehavioralCanaries, t as analyzeRuns } from "./analyze-runs-jjCmF8pU.js";
-import { a as summaryTable, i as researchReport, n as gainHistogram, r as paretoChart, t as RESEARCH_REPORT_HARD_PAIR_FLOOR } from "./summary-report-BEk8OFLs.js";
-import { a as composeParsers, c as vitestTestParser, i as SubprocessSandboxDriver, n as DockerSandboxDriver, o as jestTestParser, r as SandboxHarness, s as pytestTestParser, t as runTestGradedScenario } from "./test-graded-scenario-BsqWLmPt.js";
-import { a as InMemoryTraceStore, i as FileSystemTraceStore, n as assertRunCaptured, r as throwIfRunIncomplete, t as RunIntegrityError } from "./integrity-BzRbCHzi.js";
+import { a as checkBehavioralCanary, i as canaryLeakView, o as checkCanaries, r as HoldoutAuditor, s as runBehavioralCanaries, t as analyzeRuns } from "./analyze-runs-BScZvqMV.js";
+import { a as summaryTable, i as researchReport, n as gainHistogram, r as paretoChart, t as RESEARCH_REPORT_HARD_PAIR_FLOOR } from "./summary-report-9A5y7EsK.js";
+import { a as composeParsers, c as vitestTestParser, i as SubprocessSandboxDriver, n as DockerSandboxDriver, o as jestTestParser, r as SandboxHarness, s as pytestTestParser, t as runTestGradedScenario } from "./test-graded-scenario-JHcKQNpq.js";
+import { a as InMemoryTraceStore, i as FileSystemTraceStore, n as assertRunCaptured, r as throwIfRunIncomplete, t as RunIntegrityError } from "./integrity-fdt8XPAv.js";
 import { i as redactValue, n as REDACTION_VERSION, r as redactString, t as DEFAULT_REDACTION_RULES } from "./redact-7Aq1ukl-.js";
 import { n as DEFAULT_RULES, r as classifyFailure, t as computeToolUseMetrics } from "./tool-use-metrics-DEGMKycK.js";
 import { t as analyzeSeries } from "./series-convergence-CjO2QdRW.js";
-import { _ as deterministicSplit, g as BENCHMARK_SPLIT_SEED, t as benchmarks_exports } from "./benchmarks-Dfgm9ts5.js";
-import { t as runEvalCampaign } from "./eval-campaign-Cc8WZJ6b.js";
+import { _ as deterministicSplit, g as BENCHMARK_SPLIT_SEED, t as benchmarks_exports } from "./benchmarks-BJ_xK5rQ.js";
+import { t as runEvalCampaign } from "./eval-campaign-9MozgKL7.js";
 import { n as pairedEvalueSequence, t as evaluateInterimReleaseConfidence } from "./sequential-Br0mAPHA.js";
 import { AxGEPA, AxSignature, ax } from "@ax-llm/ax";
 import { accessSync, appendFileSync, constants, cpSync, existsSync, mkdirSync, promises, readFileSync, readdirSync, statSync, writeFileSync } from "node:fs";
@@ -1735,6 +1737,117 @@ function createFeedbackTrajectory(input) {
 		metadata: input.metadata
 	};
 }
+/** Preserve a trace-analysis run for review, whether or not review is complete. */
+function analystRunToFeedbackTrajectory(run, options) {
+	const archivedRun = snapshotAnalystRun(run, `analyst run "${run.run_id}"`);
+	const runDigest = analystRunDigest(archivedRun);
+	const findings = archivedRun.findings;
+	assertUniqueFindingIds(findings.map((finding) => finding.finding_id));
+	const analystIds = [.../* @__PURE__ */ new Set([...findings.map((finding) => finding.analyst_id), ...archivedRun.per_analyst.map((summary) => summary.analyst_id)])];
+	const reviewDecisions = validateAnalystReviewDecisions({
+		runId: archivedRun.run_id,
+		runDigest,
+		findings,
+		analystIds,
+		decisions: options.reviewDecisions ?? []
+	});
+	const reviewRequests = options.reviewRequests === void 0 ? void 0 : validateAnalystReviewRequests(archivedRun, options.reviewRequests);
+	const createdAt = options.createdAt ?? archivedRun.started_at;
+	const knownCostUsd = archivedRun.total_cost_usd;
+	const capturedCost = archivedRun.total_cost_provenance?.kind === "uncaptured" ? void 0 : archivedRun.total_cost_usd;
+	return createFeedbackTrajectory({
+		id: options.id,
+		projectId: options.projectId,
+		scenarioId: options.scenarioId,
+		task: options.task,
+		attempts: [{
+			id: `analysis:${archivedRun.run_id}`,
+			stepIndex: 0,
+			artifactType: "research",
+			artifact: {
+				type: "analyst-run",
+				analystRunId: archivedRun.run_id,
+				runDigest,
+				correlationId: archivedRun.correlation_id,
+				analystIds,
+				findings,
+				..."execution_plan" in archivedRun ? { executionPlan: archivedRun.execution_plan } : {},
+				..."completion" in archivedRun ? { completion: archivedRun.completion } : {}
+			},
+			createdAt,
+			metadata: {
+				perAnalyst: archivedRun.per_analyst,
+				trace: options.trace,
+				reviewRequests
+			}
+		}],
+		labels: [...options.labels ?? []],
+		outcome: options.outcome ? {
+			...options.outcome,
+			costUsd: options.outcome.costUsd ?? capturedCost
+		} : capturedCost === void 0 ? void 0 : {
+			costUsd: capturedCost,
+			observedAt: archivedRun.ended_at
+		},
+		split: options.split,
+		tags: options.tags,
+		createdAt,
+		metadata: {
+			...options.metadata,
+			analysis: {
+				kind: "analyst-run",
+				runId: archivedRun.run_id,
+				runDigest,
+				startedAt: archivedRun.started_at,
+				endedAt: archivedRun.ended_at,
+				reviewDecisions,
+				costProvenance: archivedRun.total_cost_provenance,
+				knownCostUsd
+			}
+		}
+	});
+}
+/** Convert one immutable analyst run into revision requests for an independent reviewer. */
+function analystRunToReviewRequests(run, options = {}) {
+	const archivedRun = snapshotAnalystRun(run, `analyst run "${run.run_id}"`);
+	const runDigest = analystRunDigest(archivedRun);
+	const createdAt = options.createdAt ?? (/* @__PURE__ */ new Date()).toISOString();
+	return archivedRun.findings.map((finding) => ({
+		id: analystReviewRequestId(runDigest, finding.finding_id),
+		runId: archivedRun.run_id,
+		runDigest,
+		findingId: finding.finding_id,
+		findingDigest: analystFindingDigest(finding),
+		analystId: finding.analyst_id,
+		area: finding.area,
+		claim: finding.claim,
+		evidence: finding.evidence_refs,
+		recommendedAction: finding.recommended_action,
+		validationPlan: finding.validation_plan,
+		severity: feedbackSeverityFromFinding(finding),
+		confidence: finding.confidence,
+		createdAt
+	}));
+}
+function validateAnalystReviewRequests(run, requests) {
+	const runDigest = analystRunDigest(run);
+	const findingsById = new Map(run.findings.map((finding) => [finding.finding_id, finding]));
+	const seenFindingIds = /* @__PURE__ */ new Set();
+	return requests.map((request, index) => {
+		if (request.runId !== run.run_id) throw new TypeError(`analyst review request ${index} run id mismatch`);
+		if (request.runDigest !== runDigest) throw new TypeError(`analyst review request ${index} run digest mismatch`);
+		const finding = findingsById.get(request.findingId);
+		if (!finding) throw new TypeError(`analyst review request ${index} references unknown finding id "${request.findingId}"`);
+		if (seenFindingIds.has(request.findingId)) throw new TypeError(`duplicate analyst review request for finding id "${request.findingId}"`);
+		seenFindingIds.add(request.findingId);
+		if (request.findingDigest !== analystFindingDigest(finding)) throw new TypeError(`analyst review request ${index} finding digest mismatch`);
+		if (request.id !== analystReviewRequestId(runDigest, request.findingId)) throw new TypeError(`analyst review request ${index} id mismatch`);
+		return request;
+	});
+}
+function analystReviewRequestId(runDigest, findingId) {
+	return `analyst:${runDigest}:${findingId}`;
+}
 function assignFeedbackSplit(trajectory, policy = {}) {
 	const split = {
 		...DEFAULT_SPLIT_POLICY,
@@ -1772,18 +1885,37 @@ function feedbackTrajectoriesToDatasetScenarios(trajectories) {
 }
 function feedbackTrajectoryToOptimizerRow(trajectory) {
 	const labels = allLabels(trajectory);
+	const analystReview = readAnalystReview(trajectory);
+	const analystQuality = analystReview ? completedAnalystReviewQuality(analystReview) : void 0;
+	const reviewLabelKinds = analystReview ? analystReview.reviewDecisions.map((decision) => {
+		if (decision.verdict === "rejected") return "reject";
+		if (decision.verdict === "completeness_assessed" && decision.missedIssues.length > 0) return "revision_request";
+		return "approve";
+	}) : [];
 	return {
 		scenarioId: trajectory.scenarioId ?? trajectory.id,
 		trajectoryId: trajectory.id,
-		labelKinds: [...new Set(labels.map((label) => label.kind))],
-		score: trajectory.outcome?.score ?? scoreFromLabels(labels),
+		labelKinds: [.../* @__PURE__ */ new Set([...labels.map((label) => label.kind), ...reviewLabelKinds])],
+		score: analystQuality?.f1 ?? trajectory.outcome?.score ?? scoreFromLabels(labels),
 		metadata: {
 			projectId: trajectory.projectId,
 			split: trajectory.split,
 			intent: trajectory.task.intent,
 			attempts: trajectory.attempts.length,
 			outcome: trajectory.outcome,
-			labels
+			labels,
+			...analystReview ? { analystReview: {
+				findingIds: analystReview.findingIds,
+				findingDigests: analystReview.findings.map((finding) => ({
+					findingId: finding.finding_id,
+					findingDigest: analystFindingDigest(finding)
+				})),
+				decisions: analystReview.reviewDecisions,
+				precision: analystQuality?.precision,
+				recall: analystQuality?.recall,
+				f1: analystQuality?.f1,
+				counts: analystQuality?.counts
+			} } : {}
 		}
 	};
 }
@@ -1923,6 +2055,14 @@ function allLabels(trajectory) {
 		return true;
 	});
 }
+function feedbackSeverityFromFinding(finding) {
+	switch (finding.severity) {
+		case "critical": return "critical";
+		case "high": return "error";
+		case "medium": return "warning";
+		default: return "info";
+	}
+}
 function scoreFromLabels(labels) {
 	if (!labels.length) return void 0;
 	const scored = labels.map((label) => {
@@ -12372,6 +12512,6 @@ function assertProductBenchmarkRun(runDir) {
 	return report;
 }
 //#endregion
-export { AGENT_PROFILE_KINDS, ATIF_SCHEMA_VERSION, ATTESTATION_ALGORITHM, AgentDriver, AgentEvalError, AgentProfileCellValidationError, AnalystRegistry, AxGepaSteeringOptimizer, BENCHMARK_SPLIT_SEED, BOOTSTRAP_GATE_MIN_N, BackendIntegrityError, BenchmarkRunner, BudgetBreachError, BudgetGuard, CODING_HARNESSES, CallbackResearcher, CaptureIntegrityError, ConfigError, ConvergenceTracker, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, CostLedger, CostLedgerPersistenceError, CostReceiptCaptureError, CostReservationExceededError, CostTracker, CrossFamilyError, DECISION_PAIRED_DELTA_STATISTIC, DEFAULT_AGENT_SLOS, DEFAULT_COMPLEXITY_WEIGHTS, DEFAULT_RULES as DEFAULT_FAILURE_RULES, DEFAULT_FINDERS, DEFAULT_HARNESS_OBJECTIVES, DEFAULT_MUTATION_PRIMITIVES, DEFAULT_MUTATORS, DEFAULT_PERMUTATIONS, DEFAULT_PR_REVIEW_SCORE_WEIGHTS, DEFAULT_REDACTION_RULES, DEFAULT_RED_TEAM_CORPUS, DEFAULT_RUN_SCORE_WEIGHTS, DEFAULT_SEVERITY_WEIGHTS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, Dataset, DescriptionLengthGate, DockerSandboxDriver, DualAgentBench, ERROR_COUNT_PATTERNS, EvalTraceStore, ExperimentTracker, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, FileSystemTraceStore, FindingsStore, HARBOR_IMPORT_GAP, HARNESS_NATIVE_MODEL, HeldOutGate, HoldoutAuditor, HoldoutLockedError, IMPROVEMENT_KIND_SPEC, INPUT_VALUE, INTENT_MATCH_JUDGE_VERSION, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, InMemoryTraceStore, InMemoryWorkspaceInspector, JudgeError, JudgeParseError, JudgeRunner, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_CACHE_WRITE_TOKENS, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, LLM_CONTEXT_TOKENS, 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, LLM_REASONING_TOKENS, LLM_REASONING_TOKEN_ATTR_KEYS, LlmCallError, LlmClient, LlmResponseError, LlmRouteAssertionError, LockedJsonlAppender, MANN_WHITNEY_EXACT_MAX_STATES, MANN_WHITNEY_EXACT_MAX_WORK, MODEL_PRICING, MetricsCollector, ModelsUnreachableError, MultiLayerVerifier, Mutex, NoopRawProviderSink, NoopResearcher, NotFoundError, OPENINFERENCE_SPAN_KIND, OTEL_AGENT_EVAL_SCOPE, OUTPUT_VALUE, OtlpFileTraceStore, PairwiseSteeringOptimizer, ProductClient, PromptRegistry, REDACTION_VERSION, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, RESEARCH_REPORT_HARD_PAIR_FLOOR, ROLLOUT_SCHEMA, RUN_COST_ATTR_KEYS, ReplayCache, ReplayCacheMissError, ReplayError, RunCritic, RunIntegrityError, RunRecordValidationError, SEMANTIC_CONCEPT_JUDGE_VERSION, SKILL_USAGE_ANALYST, SPAN_KIND_ATTR_KEYS, SUPERVISOR_RUN_SCHEMA, SandboxHarness, ScenarioRegistry, SeatUnsetError, SingleBackendError, SkillUsageAnalyst, SpanNotFoundError, SubprocessSandboxDriver, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TRACE_SCHEMA_VERSION, TokenCounter, ToolTraceMissingError, TraceContractBuilder, TraceEmitter, TraceFileMissingError, TraceNotFoundError, UI_FINDING_SEVERITIES, UI_LENSES, UNIVERSAL_FINDERS, ValidationError, VerificationError, WILCOXON_EXACT_MAX_N, acquisitionPlansForKnowledgeGaps, agentProfileCellHashMaterial, agentProfileCellKey, agentProfileHash, agentProfileId, agentProfileModelId, agentVisibleFields, aggregateJudgeVerdicts, aggregateLlm, aggregatePrReviewScore, aggregateRunScore, allCriticalPassed, analyzeAntiSlop, analyzeSeries, analyzeSupervisorRun, analyzeSupervisorRunSources, analyzeTraces, appendScorecard, applyLlmSpanOtlpAttributes, applyToolSpanOtlpAttributes, argHash, asNumber, asString, assertCapabilityHeadroom, assertCrossFamily, assertLlmRoute, assertMinted, assertMintedLines, assertModelsServed, assertNoHiddenLeak, assertProductBenchmarkRun, assertRealAgentReceipts, assertRealBackend, assertReleaseConfidence, assertRolloutLine, assertRunAgentProfileCell, assertRunCaptured, assertSingleBackend, assignFeedbackSplit, assignHeldOutTag, attachCostToReport, attest, attributeCounterfactuals, backoffMs, deterministicSplit as benchmarkDeterministicSplit, benchmarks_exports as benchmarks, benjaminiHochberg, bisect, blendHeldout, blockingKnowledgeEval, bonferroni, bootstrapCi, buildAgentInterfaceProfileCell, buildAgentProfileCell, buildDefaultAnalystRegistry, buildDriverSystemPrompt, buildProductBenchmarkManifest, buildReflectionPrompt, buildReviewerPrompt, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, buildWorkerDriverSystemPrompt, byteLengthRange, cachedJudge, calibrateJudge, calibrateJudgeContinuous, callLlm, callLlmJson, canaryLeakView, canonicalJson, canonicalize, capabilityHeadroom, captureFetchToRawSink, causalAttribution, checkBehavioralCanary, checkCanaries, checkSlos, checkTraceContracts, clamp01, classifyFailure, classifyOtlpSpanRole, classifyTreatment, claudeCodeSupervisorRunReader, cliffsDelta, clusteredPairedBinary, cohensD, collectionPreserved, commentsForSource, commitBisect, comparePairedArms, compareReferenceReplay, compareToBaseline, compilerJudge, completionVerdict, composeParsers, composeValidators, computeExperimentStats, computeFindingId, computeToolUseMetrics, computeTraceMetrics, confidenceInterval, containsAll, contentHash, contextInputTokens, continuousAgreement, contractJudge, controlFailureClassFromVerification, controlRunToFeedbackTrajectory, controlRunToRunRecord, convertTraceStoresToOtlp, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, costForTokenPricing, costForUsage, costReceiptFromLlm, costReceiptFromLlmError, costReport, createAnalystAi, createAntiSlopJudge, createChatClient, createDefaultReviewer, createFeedbackTrajectory, createIntentMatchJudge, createLlmCorrectnessChecker, createLlmReviewer, createOtelExporter, createOtelTracingStore, createReferenceEquivalenceJudge, createReplayFetch, createSandboxPool, createSemanticConceptJudge, createTokenRecallChecker, createTraceAnalystKind, crossTraceDiff, crowdingDistance, dataDescriptionBits, decideNextUserTurn, decidePairedPromotion, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultBlendWeights, defaultIsMaterial, defaultProviderRedactor, defaultReferenceReplayMatcher, defaultTraceInsightPanel, deployGateLayer, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, distillPlaybook, domainEvidencePattern, dominates, eProcess, ensembleJudge, errorStreakDetector, estimateCost, estimateTokens, evaluateActionPolicy, evaluateContract, evaluateHypothesis, evaluateInterimReleaseConfidence, evaluateOracles, evaluateReleaseConfidence, evaluateTraceContract, executeScenario, expandProfileAxes, expectAgent, exportProductBenchmark, exportProductBenchmarkRuns, exportRewardModel, exportRunAsOtlp, extractAssetUrls, extractErrorCount, extractOtlpAttributes, extractProducedState, extractUsage, extractUsageFromResponse, extractUsageFromSse, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToDatasetScenario, feedbackTrajectoryToOptimizerRow, fileContains, fileExists, fileExperimentStore, fileVerdictCache, findAutoMatchNoExpectation, findConstructorCwdDropped, findFallbackToPass, findLiteralTruePass, findProductBenchmarkArtifacts, findSkipCountsAsPass, firstNumberAttr, firstStringAttr, flattenOtlpExportToNdjson, flowLayer, fnv1a32, formatBenchmarkReport, formatDriverReport, formatFindings, formatScorecardDiff, fromHarborTrajectory, gainHistogram, gateTreatmentApplied, gateTreatmentFromMetrics, gateTreatmentFromSpans, gateTreatmentFromToolSpans, ghCliClient, gitProvenanceReader, precision as goldenPrecision, gradeOnHidden, gradeSemanticStatus, groupBy, groupRunsByAgentProfileCell, harnessAxisOf, hasCapturedToolArgs, hashContent, hashJson, hashScenarios, hashToUnit, hiddenGrade, holm, htmlContainsElement, httpGithubClient, improvementVerdict, inMemoryExperimentStore, inMemoryReferenceReplayStore, inMemoryReviewStore, inMemoryRunRecordBackend, inMemoryVerdictCache, inferDomainKeywords, inferOtlpKind, interRaterReliability, interpretCliffs, iqr, isBinaryOutcomeVector, isHiddenDestination, isJudgeSpan, isLlmSpan, isModelPriced, isOtelConfigured, isOtlpModelCall, isRealnessGated, isRetrievalSpan, isRolloutLine, isRunRecord, isSandboxSpan, isToolSpan, isTrainableSplit, isTransientLlmError, isUnavailable, iterateRawCalls, jestTestParser, jsonHasKeys, jsonShape, jsonlReferenceReplayStore, jsonlReviewStore, jsonlRunRecordBackend, judgeFamily, judgeReplayGate, judgeSpans, keyPreserved, knowledgeReadinessTracePayload, leaderboard, linterJudge, llmJudge, llmSpanFromProvider, llmSpans, loadScorecard, loadScorerFromGrader, localCommandRunner, lowercaseMutator, makeEvalTools, makeFinding, makeProposalFinding, mannWhitneyU, mapConcurrent, matchGoldens, matchSpan, maximumChargeForLlmRequest, mcnemar, mcnemarPower, mcnemarRequiredN, mergeLayerResults, mergeSteeringBundle, minimumPairsForPairedDeltaTest, mintRolloutRows, modelDescriptionBits, modelHasSnapshot, modelPriceKey, mulberry32, multiToolchainLayer, noProgressDetector, normalCdf, normalizeScores, notBlocked, objectiveEval, observeAll, observedScore, observedSplitScore, otelRunCompleteHook, otlpRowsToRunRecords, otlpRowsToTraceRunRecords, otlpToRunRecords, otlpToTraceRunRecords, pairArms, pairRunRecords, pairedBinaryScale, pairedBootstrap, pairedCohensDz, pairedDecisionShape, pairedDeltaTest, pairedDeltaTieFraction, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedRiskDifferenceExact, pairedRiskDifferenceScore, pairedSignTest, pairedTTest, paraphraseRobustness, paraphraseRobustnessScenarios, paretoChart, paretoFrontier, paretoFrontierWithCrowding, parseCorrectnessResponse, parseFeedbackTrajectoriesJsonl, parseReflectionResponse, parseRunRecordSafe, parseRuntimeTrajectoryHookEvent, partialCredit, partitionHeldOut, passAtK, passOrthogonality, pearsonR, pixelDeltaRatio, planTraceInsightQuestions, politenessPrefixMutator, positionalBias, preflightModels, printDriverSummary, probeLlm, productBenchmarkIntegrityFailures, productBenchmarkMutableSurfaces, productBenchmarkRepoIdentity, productBenchmarkSplits, profile_exports as profile, projectOtlpFlatLine, projectRuntimeTrajectoryEvidence, promptBisect, proposeSynthesisTargets, providerFromBaseUrl, pytestTestParser, ranks, readClaudeCodeSupervisorRun, readOtlpStatus, readProductBenchmarkManifest, readProductBenchmarkRecords, recordRuns, recordRunsToScorecard, redTeamDataset, redTeamReport, redactString, redactValue, referenceReplayRunsToSteeringRows, referenceReplayScenarioToRunScore, regexMatch, regexMatches, relabelImportedSplit, renderMarkdownReport, renderPlaybookMarkdown, renderPreferenceMemoryMarkdown, renderPriorFindings, renderReleaseReport, renderSteeringText, renderSupervisorRunHeadline, renderSupervisorRunMarkdown, renderUpstreamFindings, repeatedActionDetector, replayFeedbackTrajectories, replayFeedbackTrajectory, replayScorerOverCorpus, replayTraceThroughJudge, requireAgentProfileCell, requiredPairedSampleSize, requiredSampleSize, researchReport, resolveModelPricing, resolveSeat, rollupSupervisorRuns, roundTripRunRecord, routeFields, rowCount, rowWhere, runAgentControlLoop, runAssertions, runBehavioralCanaries, runCanaries, runCounterfactual, runE2EWorkflow, runEvalCampaign, runExpectations, runFailureClass, runHarnessExperiment, runIntentMatchJudge, runJudgeFleet, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runLiveProof, runProposeReview, runProposeReviewAsControlLoop, runRecordToProductBenchmarkRecord, runReferenceEquivalenceJudge, runReferenceReplay, runScore, runSelfPlay, runSemanticConceptJudge, runTaskScore, runTestGradedScenario, runsForScenario, scalarScore, scanForMuffledGates, scoreContinuity, scoreFromEvals, scoreKnowledgeReadiness, scoreOrigin, scorePrReviewComments, scorePrReviewSource, scoreRedTeamOutput, scoreReferenceReplay, scoreTraceInsightReadiness, seatPresets, securityJudge, selectHarnessVariant, selfPreference, sentenceReorderMutator, serializeFeedbackTrajectoriesJsonl, showMeasured, signManifest, spearmanR, statusAdvanced, stopOnNoProgress, stopOnRepeatedAction, stringField, stripFencedJson, studentTCdf, subjectiveEval, summarizeAgentReceiptIntegrity, summarizeBackendIntegrity, summarizeHarnessResults, summarizePrReviewBenchmark, summarizePreferenceMemory, summaryTable, supervisorRunRolloutLines, testJudge, textInSnapshot, throwIfRunIncomplete, toAgentProfileJson, toHarborTrajectories, toHarborTrajectory, toJsonl, toLangfuseEnvelope, toOpenAiTool, toPrometheusText, toRewardRows, toSftRows, tokenizeDomainWords, toolNamesForRun, toolSpans, toolSpansToTraceAnalysisStore, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceContract, traceJudge, traceJudgeEnsemble, traceSpanKindToOpenInferenceKind, tracedAnalyzeTraces, trainingReward, trainingScore, typoMutator, unmintableReasons, urlContains, userQuestionsForKnowledgeGaps, validateAgentProfileCell, validateProductBenchmarkManifest, validateProductBenchmarkRecord, validateProductBenchmarkRun, validateRolloutLine, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyAttestation, verifyCompletion, verifyManifest, visualDiff, viteDeployRunner, vitestTestParser, weightedComposite, weightedMean, weightedRecall, welchsTTest, whitespaceCollapseMutator, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit, withHeldoutBlend, withJudgeRetry, withOtelPipeline, wranglerDeployRunner, writeSupervisorRunReport };
+export { AGENT_PROFILE_KINDS, ATIF_SCHEMA_VERSION, ATTESTATION_ALGORITHM, AgentDriver, AgentEvalError, AgentProfileCellValidationError, AnalystRegistry, AxGepaSteeringOptimizer, BENCHMARK_SPLIT_SEED, BOOTSTRAP_GATE_MIN_N, BackendIntegrityError, BenchmarkRunner, BudgetBreachError, BudgetGuard, CODING_HARNESSES, CONTROL_INTEGRITY_ANALYST, CallbackResearcher, CaptureIntegrityError, ConfigError, ControlIntegrityAnalyst, ConvergenceTracker, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, CostLedger, CostLedgerPersistenceError, CostReceiptCaptureError, CostReservationExceededError, CostTracker, CrossFamilyError, DECISION_PAIRED_DELTA_STATISTIC, DEFAULT_AGENT_SLOS, DEFAULT_COMPLEXITY_WEIGHTS, DEFAULT_RULES as DEFAULT_FAILURE_RULES, DEFAULT_FINDERS, DEFAULT_HARNESS_OBJECTIVES, DEFAULT_MUTATION_PRIMITIVES, DEFAULT_MUTATORS, DEFAULT_PERMUTATIONS, DEFAULT_PR_REVIEW_SCORE_WEIGHTS, DEFAULT_REDACTION_RULES, DEFAULT_RED_TEAM_CORPUS, DEFAULT_RUN_SCORE_WEIGHTS, DEFAULT_SEVERITY_WEIGHTS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, Dataset, DescriptionLengthGate, DockerSandboxDriver, DualAgentBench, ERROR_COUNT_PATTERNS, EvalTraceStore, ExactAnalystRunExecutionError, ExperimentTracker, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, FileSystemTraceStore, FindingsStore, HARBOR_IMPORT_GAP, HARNESS_NATIVE_MODEL, HeldOutGate, HoldoutAuditor, HoldoutLockedError, IMPROVEMENT_KIND_SPEC, INPUT_VALUE, INTENT_MATCH_JUDGE_VERSION, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, InMemoryTraceStore, InMemoryWorkspaceInspector, JudgeError, JudgeParseError, JudgeRunner, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_CACHE_WRITE_TOKENS, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, LLM_CONTEXT_TOKENS, 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, LLM_REASONING_TOKENS, LLM_REASONING_TOKEN_ATTR_KEYS, LimitExceededError, LlmCallError, LlmClient, LlmResponseError, LlmRouteAssertionError, LockedJsonlAppender, MANN_WHITNEY_EXACT_MAX_STATES, MANN_WHITNEY_EXACT_MAX_WORK, MODEL_PRICING, MetricsCollector, ModelsUnreachableError, MultiLayerVerifier, Mutex, NoopRawProviderSink, NoopResearcher, NotFoundError, OPENINFERENCE_SPAN_KIND, OTEL_AGENT_EVAL_SCOPE, OUTPUT_VALUE, OtlpFileTraceStore, PairwiseSteeringOptimizer, ProductClient, PromptRegistry, REDACTION_VERSION, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, RESEARCH_REPORT_HARD_PAIR_FLOOR, ROLLOUT_SCHEMA, RUN_COST_ATTR_KEYS, ReplayCache, ReplayCacheMissError, ReplayError, RunCritic, RunIntegrityError, RunRecordValidationError, SEMANTIC_CONCEPT_JUDGE_VERSION, SKILL_USAGE_ANALYST, SPAN_KIND_ATTR_KEYS, SUPERVISOR_RUN_INTEGRITY_SCHEMA, SUPERVISOR_RUN_SCHEMA, SandboxHarness, ScenarioRegistry, SeatUnsetError, SingleBackendError, SkillUsageAnalyst, SpanNotFoundError, SubprocessSandboxDriver, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYSIS_LIMITS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_TOOL_NAMESPACE, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TRACE_SCHEMA_VERSION, TokenCounter, ToolTraceMissingError, TraceAnalysisLimitError, TraceAnalysisStoreContractError, TraceAnalysisValidationError, TraceContractBuilder, TraceEmitter, TraceFileMalformedError, TraceFileMissingError, TraceFileTooLargeError, TraceNotFoundError, UI_FINDING_SEVERITIES, UI_LENSES, UNIVERSAL_FINDERS, ValidationError, VerificationError, WILCOXON_EXACT_MAX_N, acquisitionPlansForKnowledgeGaps, agentProfileCellHashMaterial, agentProfileCellKey, agentProfileHash, agentProfileId, agentProfileModelId, agentVisibleFields, aggregateJudgeVerdicts, aggregateLlm, aggregatePrReviewScore, aggregateRunScore, allCriticalPassed, analystFindingDigest, analystRunDigest, analystRunToFeedbackTrajectory, analystRunToReviewRequests, analyzeAntiSlop, analyzeSeries, analyzeSupervisorRun, analyzeSupervisorRunIntegrity, analyzeSupervisorRunSources, analyzeTraces, appendScorecard, applyLlmSpanOtlpAttributes, applyToolSpanOtlpAttributes, argHash, asNumber, asString, assertCapabilityHeadroom, assertCrossFamily, assertExactRegistryRunOpts, assertLlmRoute, assertMinted, assertMintedLines, assertModelsServed, assertNoHiddenLeak, assertProductBenchmarkRun, assertRealAgentReceipts, assertRealBackend, assertReleaseConfidence, assertRolloutLine, assertRunAgentProfileCell, assertRunCaptured, assertSingleBackend, assignFeedbackSplit, assignHeldOutTag, attachCostToReport, attest, attributeCounterfactuals, backoffMs, deterministicSplit as benchmarkDeterministicSplit, benchmarks_exports as benchmarks, benjaminiHochberg, bisect, blendHeldout, blockingKnowledgeEval, bonferroni, bootstrapCi, buildAgentInterfaceProfileCell, buildAgentProfileCell, buildDefaultAnalystRegistry, buildDriverSystemPrompt, buildProductBenchmarkManifest, buildReflectionPrompt, buildReviewerPrompt, buildTraceAnalysisToolDescriptors, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, buildWorkerDriverSystemPrompt, byteLengthRange, cachedJudge, calibrateJudge, calibrateJudgeContinuous, callLlm, callLlmJson, canaryLeakView, canonicalJson, canonicalize, capabilityHeadroom, captureFetchToRawSink, causalAttribution, checkBehavioralCanary, checkCanaries, checkSlos, checkTraceContracts, clamp01, classifyFailure, classifyOtlpSpanRole, classifyTreatment, claudeCodeSupervisorRunReader, cliffsDelta, clusteredPairedBinary, cohensD, collectionPreserved, commentsForSource, commitBisect, comparePairedArms, compareReferenceReplay, compareToBaseline, compilerJudge, completionVerdict, composeParsers, composeValidators, computeExperimentStats, computeFindingId, computeToolUseMetrics, computeTraceMetrics, confidenceInterval, containsAll, contentHash, contextInputTokens, continuousAgreement, contractJudge, controlFailureClassFromVerification, controlRunToFeedbackTrajectory, controlRunToRunRecord, convertTraceStoresToOtlp, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, costForTokenPricing, costForUsage, costReceiptFromLlm, costReceiptFromLlmError, costReport, createAnalystAi, createAntiSlopJudge, createBoundedTraceAnalysisStore, createChatClient, createDefaultReviewer, createFeedbackTrajectory, createIntentMatchJudge, createLlmCorrectnessChecker, createLlmReviewer, createOtelExporter, createOtelTracingStore, createReferenceEquivalenceJudge, createReplayFetch, createSandboxPool, createSemanticConceptJudge, createTokenRecallChecker, createTraceAnalystKind, crossTraceDiff, crowdingDistance, dataDescriptionBits, decideNextUserTurn, decidePairedPromotion, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultBlendWeights, defaultIsMaterial, defaultProviderRedactor, defaultReferenceReplayMatcher, defaultTraceInsightPanel, deployGateLayer, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, distillPlaybook, domainEvidencePattern, dominates, eProcess, emitControlIntegrityFindings, ensembleJudge, errorStreakDetector, estimateCost, estimateTokens, evaluateActionPolicy, evaluateContract, evaluateHypothesis, evaluateInterimReleaseConfidence, evaluateOracles, evaluateReleaseConfidence, evaluateTraceContract, executeScenario, expandProfileAxes, expectAgent, exportProductBenchmark, exportProductBenchmarkRuns, exportRewardModel, exportRunAsOtlp, extractAssetUrls, extractErrorCount, extractOtlpAttributes, extractProducedState, extractUsage, extractUsageFromResponse, extractUsageFromSse, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToDatasetScenario, feedbackTrajectoryToOptimizerRow, fileContains, fileExists, fileExperimentStore, fileVerdictCache, findAutoMatchNoExpectation, findConstructorCwdDropped, findFallbackToPass, findLiteralTruePass, findProductBenchmarkArtifacts, findSkipCountsAsPass, firstNumberAttr, firstStringAttr, flattenOtlpExportToNdjson, flowLayer, fnv1a32, formatBenchmarkReport, formatDriverReport, formatFindings, formatScorecardDiff, fromHarborTrajectory, gainHistogram, gateTreatmentApplied, gateTreatmentFromMetrics, gateTreatmentFromSpans, gateTreatmentFromToolSpans, ghCliClient, gitProvenanceReader, precision as goldenPrecision, gradeOnHidden, gradeSemanticStatus, groupBy, groupRunsByAgentProfileCell, harnessAxisOf, hasCapturedToolArgs, hashContent, hashJson, hashScenarios, hashToUnit, hiddenGrade, holm, htmlContainsElement, httpGithubClient, improvementVerdict, inMemoryExperimentStore, inMemoryReferenceReplayStore, inMemoryReviewStore, inMemoryRunRecordBackend, inMemoryVerdictCache, inferDomainKeywords, inferOtlpKind, interRaterReliability, interpretCliffs, iqr, isBinaryOutcomeVector, isHiddenDestination, isJudgeSpan, isLlmSpan, isModelPriced, isOtelConfigured, isOtlpModelCall, isRealnessGated, isRetrievalSpan, isRolloutLine, isRunRecord, isSandboxSpan, isToolSpan, isTrainableSplit, isTransientLlmError, isUnavailable, iterateRawCalls, jestTestParser, jsonHasKeys, jsonShape, jsonlReferenceReplayStore, jsonlReviewStore, jsonlRunRecordBackend, judgeFamily, judgeReplayGate, judgeSpans, keyPreserved, knowledgeReadinessTracePayload, leaderboard, linterJudge, llmJudge, llmSpanFromProvider, llmSpans, loadScorecard, loadScorerFromGrader, localCommandRunner, lowercaseMutator, makeEvalTools, makeFinding, makeProposalFinding, mannWhitneyU, mapConcurrent, matchGoldens, matchSpan, maximumChargeForLlmRequest, mcnemar, mcnemarPower, mcnemarRequiredN, mergeLayerResults, mergeSteeringBundle, minimumPairsForPairedDeltaTest, mintRolloutRows, modelDescriptionBits, modelHasSnapshot, modelPriceKey, mulberry32, multiToolchainLayer, noProgressDetector, normalCdf, normalizeScores, notBlocked, objectiveEval, observeAll, observedScore, observedSplitScore, otelRunCompleteHook, otlpRowsToRunRecords, otlpRowsToTraceRunRecords, otlpTextToTraceAnalysisStore, otlpToRunRecords, otlpToTraceRunRecords, pairArms, pairRunRecords, pairedBinaryScale, pairedBootstrap, pairedCohensDz, pairedDecisionShape, pairedDeltaTest, pairedDeltaTieFraction, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedRiskDifferenceExact, pairedRiskDifferenceScore, pairedSignTest, pairedTTest, paraphraseRobustness, paraphraseRobustnessScenarios, paretoChart, paretoFrontier, paretoFrontierWithCrowding, parseCorrectnessResponse, parseFeedbackTrajectoriesJsonl, parseReflectionResponse, parseRunRecordSafe, parseRuntimeTrajectoryHookEvent, partialCredit, partitionHeldOut, passAtK, passOrthogonality, pearsonR, pixelDeltaRatio, planTraceInsightQuestions, politenessPrefixMutator, positionalBias, preflightModels, printDriverSummary, probeLlm, productBenchmarkIntegrityFailures, productBenchmarkMutableSurfaces, productBenchmarkRepoIdentity, productBenchmarkSplits, profile_exports as profile, projectOtlpFlatLine, projectRuntimeTrajectoryEvidence, promptBisect, proposeSynthesisTargets, providerFromBaseUrl, pytestTestParser, ranks, readClaudeCodeSupervisorRun, readOtlpStatus, readProductBenchmarkManifest, readProductBenchmarkRecords, recordRuns, recordRunsToScorecard, redTeamDataset, redTeamReport, redactString, redactValue, referenceReplayRunsToSteeringRows, referenceReplayScenarioToRunScore, regexMatch, regexMatches, relabelImportedSplit, renderMarkdownReport, renderPlaybookMarkdown, renderPreferenceMemoryMarkdown, renderPriorFindings, renderReleaseReport, renderSteeringText, renderSupervisorRunHeadline, renderSupervisorRunMarkdown, renderUpstreamFindings, repeatedActionDetector, replayFeedbackTrajectories, replayFeedbackTrajectory, replayScorerOverCorpus, replayTraceThroughJudge, requireAgentProfileCell, requiredPairedSampleSize, requiredSampleSize, researchReport, resolveModelPricing, resolveSeat, rollupSupervisorRuns, roundTripRunRecord, routeFields, rowCount, rowWhere, runAgentControlLoop, runAssertions, runBehavioralCanaries, runCanaries, runCounterfactual, runE2EWorkflow, runEvalCampaign, runExpectations, runFailureClass, runHarnessExperiment, runIntentMatchJudge, runJudgeFleet, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runLiveProof, runProposeReview, runProposeReviewAsControlLoop, runRecordToProductBenchmarkRecord, runReferenceEquivalenceJudge, runReferenceReplay, runScore, runSelfPlay, runSemanticConceptJudge, runTaskScore, runTestGradedScenario, runsForScenario, scalarScore, scanForMuffledGates, scoreContinuity, scoreFromEvals, scoreKnowledgeReadiness, scoreOrigin, scorePrReviewComments, scorePrReviewSource, scoreRedTeamOutput, scoreReferenceReplay, scoreTraceInsightReadiness, seatPresets, securityJudge, selectHarnessVariant, selfPreference, sentenceReorderMutator, serializeFeedbackTrajectoriesJsonl, showMeasured, signManifest, spearmanR, statusAdvanced, stopOnNoProgress, stopOnRepeatedAction, stringField, stripFencedJson, studentTCdf, subjectiveEval, summarizeAgentReceiptIntegrity, summarizeBackendIntegrity, summarizeHarnessResults, summarizePrReviewBenchmark, summarizePreferenceMemory, summaryTable, supervisorRunRolloutLines, testJudge, textInSnapshot, throwIfRunIncomplete, toAgentProfileJson, toHarborTrajectories, toHarborTrajectory, toJsonl, toLangfuseEnvelope, toOpenAiTool, toPrometheusText, toRewardRows, toSftRows, tokenizeDomainWords, toolNamesForRun, toolSpans, toolSpansToTraceAnalysisStore, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceContract, traceJudge, traceJudgeEnsemble, traceSpanKindToOpenInferenceKind, tracedAnalyzeTraces, trainingReward, trainingScore, typoMutator, unmintableReasons, urlContains, userQuestionsForKnowledgeGaps, validateAgentProfileCell, validateProductBenchmarkManifest, validateProductBenchmarkRecord, validateProductBenchmarkRun, validateRolloutLine, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyAttestation, verifyCompletion, verifyManifest, visualDiff, viteDeployRunner, vitestTestParser, weightedComposite, weightedMean, weightedRecall, welchsTTest, whitespaceCollapseMutator, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit, withHeldoutBlend, withJudgeRetry, withOtelPipeline, wranglerDeployRunner, writeSupervisorRunReport };
 
 //# sourceMappingURL=index.js.map