@tangle-network/agent-eval 0.136.0 → 0.137.0

package/dist/index.js

@@ -1,46 +1,47 @@
 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 { L as computeTraceMetrics, R as createChatClient, a as KNOWLEDGE_GAP_KIND_SPEC, d as emitControlIntegrityFindings, f as createTraceAnalystKind, i as KNOWLEDGE_POISONING_KIND_SPEC, l as CONTROL_INTEGRITY_ANALYST, m as renderUpstreamFindings, n as AnalystRegistry, o as IMPROVEMENT_KIND_SPEC, p as renderPriorFindings, r as DEFAULT_TRACE_ANALYST_KINDS, s as FAILURE_MODE_KIND_SPEC, t as buildDefaultAnalystRegistry, u as ControlIntegrityAnalyst, z as createAnalystAi } from "./default-registry-CLXbRt0f.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-D-5_-dhi.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-DKB25jV8.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 applyToolSpanOtlpAttributes, C as extractOtlpAttributes, D as readOtlpStatus, E as projectOtlpFlatLine, M as isOtlpModelCall, N as traceSpanKindToOpenInferenceKind, T as inferOtlpKind, _ as TraceFileMalformedError, b as TraceNotFoundError, c as otlpTextToTraceAnalysisStore, d as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, f as TRACE_ANALYSIS_LIMITS, g as TraceAnalysisValidationError, h as TraceAnalysisStoreContractError, i as traceAnalystFunctionGroup, j as classifyOtlpSpanRole, k as stringField, l as createBoundedTraceAnalysisStore, m as TraceAnalysisLimitError, n as buildTraceAnalysisToolDescriptors, o as OtlpFileTraceStore, p as SpanNotFoundError, r as buildTraceAnalystTools, t as TRACE_ANALYST_TOOL_NAMESPACE, u as DEFAULT_TRACE_ANALYST_BUDGETS, v as TraceFileMissingError, w as firstStringAttr, x as asString, y as TraceFileTooLargeError } from "./tools-DZk2Jn64.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 { 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 { 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 { 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-Bz64IckK.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-f4o9sUT4.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 scoreOrigin, n as observedScore, o as trainingReward, r as observedSplitScore, s as trainingScore, t as isRealnessGated } from "./reward-nw2xZGZG.js";
+import { h as hashCanonical, m as canonicalString } from "./ledger-core-t6sItivm.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-Bju0T8Ls.js";
+import { n as extractUsageFromResponse, r as extractUsageFromSse, t as extractUsage } from "./extract-usage-p-56bh8q.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-lgObcHFC.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-DCLkQOmc.js";
+import { t as runEvalCampaign } from "./eval-campaign-CHqfLnff.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";
@@ -1602,6 +1603,343 @@ async function decideNextUserTurn(chat, opts) {
 	return paid.value.content.trim();
 }
 //#endregion
+//#region src/feedback-trajectory-review.ts
+/** Bind an analyst finding's complete canonical JSON content to a stable digest. */
+function analystFindingDigest(finding) {
+	return hashCanonical(snapshotAnalystFinding(finding, "analyst finding"));
+}
+/** Bind the complete analyst result to one immutable review target. */
+function analystRunDigest(run) {
+	return hashCanonical(snapshotAnalystRun(run, "analyst run"));
+}
+function snapshotAnalystRun(value, context = "analyst run") {
+	let snapshot;
+	try {
+		snapshot = JSON.parse(canonicalString(value));
+	} catch (cause) {
+		throw new TypeError(`${context} must have a canonical JSON representation`, { cause });
+	}
+	if (!isRecord$1(snapshot)) throw new TypeError(`${context} must be an object`);
+	assertOnlyKeys(snapshot, [
+		"run_id",
+		"correlation_id",
+		"started_at",
+		"ended_at",
+		"findings",
+		"per_analyst",
+		"total_cost_usd",
+		"total_cost_provenance"
+	], context);
+	requiredString(snapshot.run_id, `${context} run_id`);
+	requiredString(snapshot.correlation_id, `${context} correlation_id`);
+	canonicalTimestamp(snapshot.started_at, `${context} started_at`);
+	canonicalTimestamp(snapshot.ended_at, `${context} ended_at`);
+	snapshot.findings = snapshotAnalystFindings(snapshot.findings, `${context} findings`);
+	if (!Array.isArray(snapshot.per_analyst)) throw new TypeError(`${context} per_analyst must be an array`);
+	for (const [index, summary] of snapshot.per_analyst.entries()) {
+		if (!isRecord$1(summary)) throw new TypeError(`${context} per_analyst ${index} must be an object`);
+		requiredString(summary.analyst_id, `${context} per_analyst ${index} analyst_id`);
+	}
+	if (typeof snapshot.total_cost_usd !== "number" || !Number.isFinite(snapshot.total_cost_usd) || snapshot.total_cost_usd < 0) throw new TypeError(`${context} total_cost_usd must be a finite non-negative number`);
+	if (snapshot.total_cost_provenance !== void 0 && !isRecord$1(snapshot.total_cost_provenance)) throw new TypeError(`${context} total_cost_provenance must be an object`);
+	return snapshot;
+}
+function snapshotAnalystFindings(value, context = "analyst run findings") {
+	if (!Array.isArray(value)) throw new TypeError(`${context} must be an array`);
+	const findings = value.map((finding, index) => snapshotAnalystFinding(finding, `${context} finding ${index}`));
+	assertUniqueFindingIds(findings.map((finding) => finding.finding_id));
+	return findings;
+}
+function readAnalystReview(trajectory) {
+	const analystAttempts = trajectory.attempts.filter((attempt) => isRecord$1(attempt.artifact) && attempt.artifact.type === "analyst-run");
+	const analysis = isRecord$1(trajectory.metadata?.analysis) ? trajectory.metadata.analysis : void 0;
+	if (analystAttempts.length === 0) {
+		if (analysis?.kind === "analyst-run") throw new TypeError(`feedbackTrajectoryToOptimizerRow: analyst trajectory "${trajectory.id}" is missing its archived run`);
+		return;
+	}
+	if (analystAttempts.length !== 1) throw new TypeError(`feedbackTrajectoryToOptimizerRow: analyst trajectory "${trajectory.id}" must contain exactly one archived run`);
+	if (analysis?.kind !== "analyst-run") throw new TypeError(`feedbackTrajectoryToOptimizerRow: analyst trajectory "${trajectory.id}" is missing review state`);
+	const artifact = analystAttempts[0].artifact;
+	if (!isRecord$1(artifact)) throw new TypeError(`feedbackTrajectoryToOptimizerRow: analyst trajectory "${trajectory.id}" has an invalid archived run`);
+	const runId = requiredString(artifact.analystRunId, `analyst trajectory "${trajectory.id}" run id`);
+	if (analysis.runId !== runId) throw new TypeError(`feedbackTrajectoryToOptimizerRow: analyst trajectory "${trajectory.id}" run identity does not match its review state`);
+	const artifactRunDigest = requiredDigest(artifact.runDigest, `analyst trajectory "${trajectory.id}" archived run digest`);
+	const storedRunDigest = requiredDigest(analysis.runDigest, `analyst trajectory "${trajectory.id}" review run digest`);
+	if (artifactRunDigest !== storedRunDigest) throw new TypeError(`feedbackTrajectoryToOptimizerRow: analyst trajectory "${trajectory.id}" run digest does not match its review state`);
+	const findings = snapshotAnalystFindings(artifact.findings, `analyst trajectory "${trajectory.id}"`);
+	const findingIds = findings.map((finding) => finding.finding_id);
+	const analystIds = stringArray(artifact.analystIds, `analyst trajectory "${trajectory.id}" analyst ids`);
+	const attemptMetadata = analystAttempts[0].metadata;
+	if (!isRecord$1(attemptMetadata)) throw new TypeError(`feedbackTrajectoryToOptimizerRow: analyst trajectory "${trajectory.id}" is missing archived run metadata`);
+	const archivedRun = snapshotAnalystRun({
+		run_id: runId,
+		correlation_id: artifact.correlationId,
+		started_at: analysis.startedAt,
+		ended_at: analysis.endedAt,
+		findings,
+		per_analyst: attemptMetadata.perAnalyst,
+		total_cost_usd: analysis.knownCostUsd,
+		...analysis.costProvenance === void 0 ? {} : { total_cost_provenance: analysis.costProvenance }
+	}, `analyst trajectory "${trajectory.id}" archived run`);
+	const knownAnalystIds = new Set(analystIds);
+	for (const [index, finding] of findings.entries()) if (!knownAnalystIds.has(finding.analyst_id)) throw new TypeError(`feedbackTrajectoryToOptimizerRow: analyst trajectory "${trajectory.id}" omits generating analyst "${finding.analyst_id}" at finding ${index}`);
+	const reviewDecisions = validateAnalystReviewDecisions({
+		runId,
+		runDigest: storedRunDigest,
+		findings,
+		analystIds,
+		decisions: analysis.reviewDecisions,
+		requireComplete: true
+	});
+	const expectedRunDigest = analystRunDigest(archivedRun);
+	if (storedRunDigest !== expectedRunDigest) throw new TypeError(`feedbackTrajectoryToOptimizerRow: analyst trajectory "${trajectory.id}" archived run digest mismatch`);
+	return {
+		runId,
+		runDigest: expectedRunDigest,
+		findings,
+		findingIds,
+		analystIds,
+		reviewDecisions
+	};
+}
+function completedAnalystReviewQuality(review) {
+	const findingDecisions = review.reviewDecisions.filter((decision) => decision.verdict !== "completeness_assessed");
+	const completeness = review.reviewDecisions.filter((decision) => decision.verdict === "completeness_assessed");
+	if (completeness.length !== 1) throw new TypeError("feedbackTrajectoryToOptimizerRow: analyst run requires exactly one independent completeness_assessed decision");
+	const confirmed = findingDecisions.filter((decision) => decision.verdict === "confirmed").length;
+	const rejected = findingDecisions.length - confirmed;
+	const emitted = review.findingIds.length;
+	const missed = completeness[0].missedIssues.length;
+	const precision = emitted === 0 ? 1 : confirmed / emitted;
+	const recallDenominator = confirmed + missed;
+	const recall = recallDenominator === 0 ? 1 : confirmed / recallDenominator;
+	return {
+		precision,
+		recall,
+		f1: precision + recall === 0 ? 0 : 2 * precision * recall / (precision + recall),
+		counts: {
+			emitted,
+			confirmed,
+			rejected,
+			missed
+		}
+	};
+}
+function validateAnalystReviewDecisions(input) {
+	if (!Array.isArray(input.decisions)) throw new TypeError("analyst review decisions must be an array");
+	const findings = snapshotAnalystFindings(input.findings);
+	const expectedRunDigest = requiredDigest(input.runDigest, "analyst review run digest");
+	const findingsById = new Map(findings.map((finding) => [finding.finding_id, finding]));
+	const generatingAnalystIds = new Set(input.analystIds);
+	const seenFindingIds = /* @__PURE__ */ new Set();
+	let completenessCount = 0;
+	const decisions = input.decisions.map((value, index) => {
+		if (!isRecord$1(value)) throw new TypeError(`analyst review decision ${index} must be an object`);
+		const source = requiredString(value.source, `analyst review decision ${index} source`);
+		if (!isAnalystReviewSource(source)) throw new TypeError(`analyst review decision ${index} source must be user, judge, environment, metric, or policy`);
+		const reviewerId = requiredString(value.reviewerId, `analyst review decision ${index} reviewerId`);
+		if (generatingAnalystIds.has(reviewerId)) throw new TypeError(`analyst review decision ${index} reviewerId must differ from the generating analyst`);
+		const reviewId = requiredString(value.reviewId, `analyst review decision ${index} reviewId`);
+		if (reviewId === input.runId) throw new TypeError(`analyst review decision ${index} reviewId must identify an independent review`);
+		const reason = requiredString(value.reason, `analyst review decision ${index} reason`);
+		const decidedAt = canonicalTimestamp(value.decidedAt, `analyst review decision ${index} decidedAt`);
+		const runDigest = requiredDigest(value.runDigest, `analyst review decision ${index} runDigest`);
+		if (runDigest !== expectedRunDigest) throw new TypeError(`analyst review decision ${index} run digest mismatch`);
+		if (value.verdict === "completeness_assessed") {
+			assertOnlyKeys(value, [
+				"runDigest",
+				"verdict",
+				"missedIssues",
+				"source",
+				"reviewerId",
+				"reviewId",
+				"reason",
+				"decidedAt"
+			], `analyst review decision ${index}`);
+			completenessCount += 1;
+			if (completenessCount > 1) throw new TypeError("duplicate completeness_assessed analyst review decision");
+			return {
+				runDigest,
+				verdict: "completeness_assessed",
+				missedIssues: validateMissedIssues(value.missedIssues, findingsById, `analyst review decision ${index}`),
+				source,
+				reviewerId,
+				reviewId,
+				reason,
+				decidedAt
+			};
+		}
+		if (value.verdict !== "confirmed" && value.verdict !== "rejected") throw new TypeError(`analyst review decision ${index} verdict must be confirmed, rejected, or completeness_assessed`);
+		assertOnlyKeys(value, [
+			"runDigest",
+			"findingId",
+			"findingDigest",
+			"verdict",
+			"source",
+			"reviewerId",
+			"reviewId",
+			"reason",
+			"decidedAt"
+		], `analyst review decision ${index}`);
+		const findingId = requiredString(value.findingId, `analyst review decision ${index} findingId`);
+		const finding = findingsById.get(findingId);
+		if (!finding) throw new TypeError(`analyst review decision references unknown finding id "${findingId}"`);
+		if (seenFindingIds.has(findingId)) throw new TypeError(`duplicate analyst review decision for finding id "${findingId}"`);
+		seenFindingIds.add(findingId);
+		const findingDigest = requiredString(value.findingDigest, `analyst review decision ${index} findingDigest`);
+		const expectedDigest = analystFindingDigest(finding);
+		if (findingDigest !== expectedDigest) throw new TypeError(`analyst review decision ${index} digest mismatch for finding id "${findingId}"`);
+		return {
+			runDigest,
+			findingId,
+			findingDigest: expectedDigest,
+			verdict: value.verdict,
+			source,
+			reviewerId,
+			reviewId,
+			reason,
+			decidedAt
+		};
+	});
+	if (input.requireComplete) {
+		const missing = findings.map((finding) => finding.finding_id).filter((findingId) => !seenFindingIds.has(findingId));
+		if (missing.length > 0) throw new TypeError(`feedbackTrajectoryToOptimizerRow: missing independent decisions for finding ids: ${missing.join(", ")}`);
+		if (completenessCount !== 1) throw new TypeError("feedbackTrajectoryToOptimizerRow: analyst run requires exactly one independent completeness_assessed decision");
+	}
+	return decisions;
+}
+function assertUniqueFindingIds(findingIds) {
+	const seen = /* @__PURE__ */ new Set();
+	for (const findingId of findingIds) {
+		if (findingId.trim().length === 0) throw new TypeError("analyst finding id must not be empty");
+		if (seen.has(findingId)) throw new TypeError(`analyst run contains duplicate finding id "${findingId}"`);
+		seen.add(findingId);
+	}
+}
+function snapshotAnalystFinding(value, context) {
+	let snapshot;
+	try {
+		snapshot = JSON.parse(canonicalString(value));
+	} catch (cause) {
+		throw new TypeError(`${context} must have a canonical JSON representation`, { cause });
+	}
+	assertAnalystFinding(snapshot, context);
+	return snapshot;
+}
+function assertAnalystFinding(value, context) {
+	if (!isRecord$1(value)) throw new TypeError(`${context} must be an object`);
+	assertOnlyKeys(value, [
+		"schema_version",
+		"finding_id",
+		"analyst_id",
+		"produced_at",
+		"severity",
+		"area",
+		"claim",
+		"rationale",
+		"evidence_refs",
+		"recommended_action",
+		"validation_plan",
+		"confidence",
+		"subject",
+		"derived_from_judge",
+		"metadata"
+	], context);
+	if (value.schema_version !== "1.0.0") throw new TypeError(`${context} schema_version must be "1.0.0"`);
+	requiredString(value.finding_id, `${context} finding_id`);
+	requiredString(value.analyst_id, `${context} analyst_id`);
+	canonicalTimestamp(value.produced_at, `${context} produced_at`);
+	if (value.severity !== "critical" && value.severity !== "high" && value.severity !== "medium" && value.severity !== "low" && value.severity !== "info") throw new TypeError(`${context} severity is invalid`);
+	requiredString(value.area, `${context} area`);
+	requiredString(value.claim, `${context} claim`);
+	optionalString$1(value.rationale, `${context} rationale`);
+	value.evidence_refs = validateEvidenceRefs(value.evidence_refs, `${context} evidence_refs`);
+	optionalString$1(value.recommended_action, `${context} recommended_action`);
+	optionalString$1(value.validation_plan, `${context} validation_plan`);
+	if (typeof value.confidence !== "number" || !Number.isFinite(value.confidence) || value.confidence < 0 || value.confidence > 1) throw new TypeError(`${context} confidence must be a finite number from 0 through 1`);
+	optionalString$1(value.subject, `${context} subject`);
+	if (value.derived_from_judge !== void 0 && typeof value.derived_from_judge !== "boolean") throw new TypeError(`${context} derived_from_judge must be a boolean`);
+	if (value.metadata !== void 0 && !isRecord$1(value.metadata)) throw new TypeError(`${context} metadata must be an object`);
+}
+function validateMissedIssues(value, findingsById, context) {
+	if (!Array.isArray(value)) throw new TypeError(`${context} missedIssues must be an array`);
+	const seen = /* @__PURE__ */ new Set();
+	return value.map((issue, index) => {
+		const issueContext = `${context} missedIssues ${index}`;
+		if (!isRecord$1(issue)) throw new TypeError(`${issueContext} must be an object`);
+		assertOnlyKeys(issue, [
+			"id",
+			"reason",
+			"evidence"
+		], issueContext);
+		const id = requiredString(issue.id, `${issueContext} id`);
+		if (findingsById.has(id)) throw new TypeError(`${issueContext} id "${id}" is already an emitted finding id`);
+		if (seen.has(id)) throw new TypeError(`duplicate missed issue id "${id}"`);
+		seen.add(id);
+		return {
+			id,
+			reason: requiredString(issue.reason, `${issueContext} reason`),
+			...issue.evidence === void 0 ? {} : { evidence: validateEvidenceRefs(issue.evidence, `${issueContext} evidence`) }
+		};
+	});
+}
+function validateEvidenceRefs(value, context) {
+	if (!Array.isArray(value)) throw new TypeError(`${context} must be an array`);
+	return value.map((evidence, index) => {
+		const evidenceContext = `${context} ${index}`;
+		if (!isRecord$1(evidence)) throw new TypeError(`${evidenceContext} must be an object`);
+		assertOnlyKeys(evidence, [
+			"kind",
+			"uri",
+			"excerpt"
+		], evidenceContext);
+		if (evidence.kind !== "span" && evidence.kind !== "event" && evidence.kind !== "artifact" && evidence.kind !== "finding" && evidence.kind !== "metric") throw new TypeError(`${evidenceContext} kind is invalid`);
+		const uri = requiredString(evidence.uri, `${evidenceContext} uri`);
+		const excerpt = evidence.excerpt;
+		optionalString$1(excerpt, `${evidenceContext} excerpt`);
+		return {
+			kind: evidence.kind,
+			uri,
+			...excerpt === void 0 ? {} : { excerpt }
+		};
+	});
+}
+function assertOnlyKeys(value, allowed, name) {
+	const allowedKeys = new Set(allowed);
+	const unexpected = Object.keys(value).filter((key) => !allowedKeys.has(key));
+	if (unexpected.length > 0) throw new TypeError(`${name} contains unknown fields: ${unexpected.sort().join(", ")}`);
+}
+function stringArray(value, name) {
+	if (!Array.isArray(value) || value.some((item) => typeof item !== "string")) throw new TypeError(`${name} must be an array of strings`);
+	const strings = value.map((item) => requiredString(item, name));
+	if (new Set(strings).size !== strings.length) throw new TypeError(`${name} must contain unique values`);
+	return strings;
+}
+function requiredString(value, name) {
+	if (typeof value !== "string" || value.trim().length === 0) throw new TypeError(`${name} must be a non-empty string`);
+	return value;
+}
+function requiredDigest(value, name) {
+	const digest = requiredString(value, name);
+	if (!/^sha256:[a-f0-9]{64}$/.test(digest)) throw new TypeError(`${name} must be a sha256 digest`);
+	return digest;
+}
+function optionalString$1(value, name) {
+	if (value !== void 0 && typeof value !== "string") throw new TypeError(`${name} must be a string`);
+}
+function canonicalTimestamp(value, name) {
+	const timestamp = requiredString(value, name);
+	const parsed = new Date(timestamp);
+	if (Number.isNaN(parsed.valueOf()) || parsed.toISOString() !== timestamp) throw new TypeError(`${name} must be a canonical ISO 8601 UTC timestamp`);
+	return timestamp;
+}
+function isAnalystReviewSource(value) {
+	return value === "user" || value === "judge" || value === "environment" || value === "metric" || value === "policy";
+}
+function isRecord$1(value) {
+	return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+//#endregion
 //#region src/feedback-trajectory.ts
 const DEFAULT_SPLIT_POLICY = {
 	trainPct: 70,
@@ -1735,6 +2073,115 @@ function createFeedbackTrajectory(input) {
 		metadata: input.metadata
 	};
 }
+/** Preserve a trace-analysis run for review, whether or not review is complete. */
+function analystRunToFeedbackTrajectory(run, options) {
+	const archivedRun = snapshotAnalystRun(run, `analyst run "${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:${archivedRun.run_id}`,
+			stepIndex: 0,
+			artifactType: "research",
+			artifact: {
+				type: "analyst-run",
+				analystRunId: archivedRun.run_id,
+				runDigest,
+				correlationId: archivedRun.correlation_id,
+				analystIds,
+				findings
+			},
+			createdAt,
+			metadata: {
+				perAnalyst: archivedRun.per_analyst,
+				trace: options.trace,
+				reviewRequests
+			}
+		}],
+		labels: [...options.labels ?? []],
+		outcome: options.outcome ? {
+			...options.outcome,
+			costUsd: options.outcome.costUsd ?? capturedCost
+		} : capturedCost === void 0 ? void 0 : {
+			costUsd: capturedCost,
+			observedAt: archivedRun.ended_at
+		},
+		split: options.split,
+		tags: options.tags,
+		createdAt,
+		metadata: {
+			...options.metadata,
+			analysis: {
+				kind: "analyst-run",
+				runId: archivedRun.run_id,
+				runDigest,
+				startedAt: archivedRun.started_at,
+				endedAt: archivedRun.ended_at,
+				reviewDecisions,
+				costProvenance: archivedRun.total_cost_provenance,
+				knownCostUsd
+			}
+		}
+	});
+}
+/** Convert one immutable analyst run into revision requests for an independent reviewer. */
+function analystRunToReviewRequests(run, options = {}) {
+	const archivedRun = snapshotAnalystRun(run, `analyst run "${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 +2219,37 @@ 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 +2389,14 @@ 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 +12846,6 @@ 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, 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, 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