@tangle-network/agent-eval 0.216.1 → 0.218.1

package/dist/index.d.ts

@@ -1,85 +1,58 @@
 import { _ as PendingCostCallView, a as CostLedgerHandle, c as CostLedgerSummary, d as CostReceiptInput, f as CostUsage, g as PendingCostCall, h as PaidCallResult, i as CostLedgerFilter, l as CostProvenance, m as MaximumCharge, n as ChannelRollup, o as CostLedgerOptions, p as CustomTokenPricing, r as CostChannel, s as CostLedgerPersistence, t as CostLedger, u as CostReceipt, v as RunPaidCallInput } from "./cost-ledger-Jk62OBfF.js";
 import { a as JudgeError, c as ValidationError, i as ConfigError, n as AgentEvalErrorCode, s as NotFoundError, t as AgentEvalError } from "./errors-DEE6u6ot.js";
-import { S as verifyAgentProfileCell, _ as buildAgentProfileCell, b as toAgentProfileJson, i as AgentProfileCellInput, l as AgentProfileJson, m as agentProfileCellKey, p as AgentProfileSourceInput, r as AgentProfileCell, s as AgentProfileDimensionValue, t as AGENT_PROFILE_KINDS, v as groupRunsByAgentProfileCell } from "./agent-profile-cell-BNp-Js9F.js";
-import { C as ToolSpan, D as isToolSpan, E as isLlmSpan, T as isJudgeSpan, a as FAILURE_CLASSES, c as JudgeSpan, f as Run, g as RunStatus, l as LlmSpan, n as BudgetLedgerEntry, o as FailureClass, p as RunCaptureReport, r as BudgetSpec, s as GenericSpan, t as Artifact, v as Span, w as TraceEvent } from "./schema-2g7trUwH.js";
-import { _ as runTaskScore, a as RunRecord, c as RunSplitTag, d as RunTokenUsage, f as RunTraceRef, g as parseRunRecordSafe, h as modelHasSnapshot, i as RunOutcome, l as RunTaskFailure, m as isRunRecord, n as RunCostProvenance, o as RunRecordValidationError, p as UNKNOWN_MODEL, r as RunJudgeMetadata, s as RunSearchCoordinates, t as JudgeScoresRecord, u as RunTerminalOutcome, v as searchCellRunId, y as validateRunRecord } from "./run-record-C1E24G2A.js";
-import { $ as DatasetOverview, C as makeFinding, Q as DEFAULT_TRACE_ANALYST_BUDGETS, S as computeFindingId, W as createBoundedTraceAnalysisStore, X as TraceAnalysisStore, a as AnalystFinding, at as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, b as ProposalFinding, ct as TraceAnalystSpan, d as AnalystRunResult, dt as TraceAnalystTraceSummary, et as ErrorCluster, f as AnalystRunSummary, ft as ViewSpansResult, h as EvidenceRef, it as SpanMatchRecord, l as AnalystRunEvent, lt as TraceAnalystSpanKind, m as AnalystUsageReceipt, mt as ViewTraceResult, n as Analyst, nt as SearchSpanResult, ot as TraceAnalystByteBudgets, p as AnalystSeverity, pt as ViewTraceOversized, r as AnalystContext, rt as SearchTraceResult, st as TraceAnalystFilters, tt as QueryTracesPage, u as AnalystRunInputs, ut as TraceAnalystSpanStatus, w as makeProposalFinding } from "./types-C1tnkEgM.js";
-import { A as createChatClient, B as LlmToolDefinition, C as ChatResponse, D as MockTransportOpts, E as CustomTransportOpts, F as LlmMessage, H as FileSystemRawProviderSink, K as NoopRawProviderSink, L as LlmTokenLogprob, M as LlmCallRequest, N as LlmCallResult, O as OpenAiCompatibleTransportOpts, P as LlmChargeBounds, R as LlmToolCall, S as ChatRequest, T as CreateChatClientOpts, V as LlmUsage, W as InMemoryRawProviderSink, X as RawProviderSink, Y as RawProviderEvent, _ as ScenarioResult, a as CompletionCriterion, at as ServedModelPolicy, b as ChatCallOpts, c as JudgeInput, ct as assertServedModel, d as PersonaConfig, dt as servedModelAcceptable, et as AssertCrossFamilyServedOptions, f as ProductClientConfig, ft as AssertCrossFamilyOptions, g as ScenarioFile, gt as judgeFamily, h as Scenario, ht as assertCrossFamily, i as CollectedArtifacts, it as ServedModelCheck, j as LlmCallMetadata, k as SandboxSdkTransportOpts, l as JudgeRubric, lt as assertServedModels, m as RubricDimension, mt as JudgeFamily, n as ArtifactResult, nt as ModelSubstitutionError, o as DriverState, ot as ServedModelVerdict, p as RouteMap, pt as CrossFamilyError, r as CheckResult, rt as ServedCrossFamilyError, s as JudgeFn, st as assertCrossFamilyServed, t as ArtifactCheck, tt as AssertServedModelOptions, u as JudgeScore, ut as checkServedModel, v as Turn, w as ChatTransport, x as ChatClient, y as TurnResult, z as LlmToolChoice } from "./types-C02qtnG8.js";
+import { _ as toAgentProfileJson, g as groupRunsByAgentProfileCell, h as buildAgentProfileCell, i as AgentProfileCellInput, l as AgentProfileJson, m as agentProfileCellKey, p as AgentProfileSourceInput, r as AgentProfileCell, s as AgentProfileDimensionValue, t as AGENT_PROFILE_KINDS, y as verifyAgentProfileCell } from "./agent-profile-cell-BcpFbNd5.js";
+import { E as isToolSpan, S as TraceEvent, T as isLlmSpan, _ as Span, a as FAILURE_CLASSES, c as JudgeSpan, f as Run, h as RunStatus, l as LlmSpan, n as BudgetLedgerEntry, o as FailureClass, p as RunCaptureReport, r as BudgetSpec, s as GenericSpan, t as Artifact, w as isJudgeSpan, x as ToolSpan } from "./schema-D6EVKzTB.js";
+import { _ as runTaskScore, a as RunRecord, c as RunSplitTag, d as RunTokenUsage, f as RunTraceRef, g as parseRunRecordSafe, h as modelHasSnapshot, i as RunOutcome, l as RunTaskFailure, m as isRunRecord, n as RunCostProvenance, o as RunRecordValidationError, p as UNKNOWN_MODEL, r as RunJudgeMetadata, s as RunSearchCoordinates, t as JudgeScoresRecord, u as RunTerminalOutcome, v as validateRunRecord } from "./run-record-CaxbLq6Y.js";
+import { $ as TraceAnalystByteBudgets, J as ErrorCluster, K as DEFAULT_TRACE_ANALYST_BUDGETS, Q as SpanMatchRecord, R as createBoundedTraceAnalysisStore, S as makeProposalFinding, W as TraceAnalysisStore, X as SearchSpanResult, Y as QueryTracesPage, Z as SearchTraceResult, at as ViewSpansResult, b as computeFindingId, c as AnalystRunEvent, d as AnalystRunSummary, et as TraceAnalystFilters, f as AnalystSeverity, i as AnalystFinding, it as TraceAnalystTraceSummary, l as AnalystRunInputs, m as EvidenceRef, n as AnalystContext, nt as TraceAnalystSpanKind, ot as ViewTraceOversized, p as AnalystUsageReceipt, q as DatasetOverview, rt as TraceAnalystSpanStatus, st as ViewTraceResult, t as Analyst, tt as TraceAnalystSpan, u as AnalystRunResult, v as ProposalFinding, x as makeFinding } from "./types-CW8ewATZ.js";
+import { $ as AssertServedModelOptions, A as createChatClient, B as LlmToolDefinition, C as ChatResponse, D as MockTransportOpts, E as CustomTransportOpts, F as LlmMessage, H as FileSystemRawProviderSink, K as NoopRawProviderSink, L as LlmTokenLogprob, M as LlmCallRequest, N as LlmCallResult, O as OpenAiCompatibleTransportOpts, P as LlmChargeBounds, Q as AssertCrossFamilyServedOptions, R as LlmToolCall, S as ChatRequest, T as CreateChatClientOpts, V as LlmUsage, W as InMemoryRawProviderSink, X as RawProviderSink, Y as RawProviderEvent, _ as ScenarioResult, a as CompletionCriterion, at as assertServedModel, b as ChatCallOpts, c as JudgeFn, ct as AssertCrossFamilyOptions, d as PersonaConfig, dt as assertCrossFamily, et as ModelSubstitutionError, f as ProductClientConfig, ft as judgeFamily, g as ScenarioFile, h as RubricDimension, i as CollectedArtifacts, it as assertCrossFamilyServed, j as LlmCallMetadata, k as SandboxSdkTransportOpts, l as JudgeInput, lt as CrossFamilyError, m as RouteMap, n as ArtifactResult, nt as ServedModelPolicy, o as DimensionJudgeScore, ot as assertServedModels, p as ProductScenario, r as CheckResult, rt as ServedModelVerdict, s as DriverState, st as checkServedModel, t as ArtifactCheck, tt as ServedModelCheck, u as JudgeRubric, ut as JudgeFamily, v as Turn, w as ChatTransport, x as ChatClient, y as TurnResult, z as LlmToolChoice } from "./types-D-mgRzlh.js";
 import { a as ContinuousCalibrationResult, d as calibrateJudgeContinuous, f as continuousAgreement, h as verbosityBias, i as ContinuousAgreementOptions, l as VerbosityBiasResult, n as CandidateScore, o as GoldenItem, r as ContinuousAgreement, t as CalibrationResult, u as calibrateJudge } from "./judge-calibration-DFtEMlde.js";
-import { G as CorpusAgreementPerDimension, H as SeriesConvergenceResult, J as corpusInterRaterAgreement, K as CorpusAgreementReport, U as analyzeSeries, V as SeriesConvergenceOptions, W as CorpusAgreementOptions, X as interRaterReliability, Y as corpusInterRaterAgreementFromJudgeScores, _ as assertJudgeMayGate, a as JudgeCalibrationSet, b as registerCalibrationSet, c as JudgeGateDecision, d as JudgeGateRefusedError, f as JudgeGateStatus, g as addCalibrationVerdicts, h as OwnerVerdict, i as JudgeAgreement, l as JudgeGateInput, m as JudgeRunMeta, n as CalibrationMerge, o as JudgeConfusion, p as JudgeRun, q as CorpusScoreRecord, r as DEFAULT_JUDGE_GATE_POLICY, s as JudgeExampleAgreement, t as CalibrationExample, u as JudgeGatePolicy, v as judgeGateDecision, x as snapshotFromJudgeAgreement, y as measureJudgeAgreement } from "./judge-gate-CE7CQAgt.js";
+import { B as SeriesConvergenceOptions, G as CorpusAgreementReport, H as analyzeSeries, J as corpusInterRaterAgreementFromJudgeScores, K as CorpusScoreRecord, U as CorpusAgreementOptions, V as SeriesConvergenceResult, W as CorpusAgreementPerDimension, Y as interRaterReliability, _ as assertJudgeMayGate, a as JudgeCalibrationSet, b as registerCalibrationSet, c as JudgeGateDecision, d as JudgeGateRefusedError, f as JudgeGateStatus, g as addCalibrationVerdicts, h as OwnerVerdict, i as JudgeAgreement, l as JudgeGateInput, m as JudgeRunMeta, n as CalibrationMerge, o as JudgeConfusion, p as JudgeRun, q as corpusInterRaterAgreement, r as DEFAULT_JUDGE_GATE_POLICY, s as JudgeExampleAgreement, t as CalibrationExample, u as JudgeGatePolicy, v as judgeGateDecision, x as snapshotFromJudgeAgreement, y as measureJudgeAgreement } from "./judge-gate-DsL0yFcb.js";
 import { a as partialCredit, c as spearmanR, d as weightedMean, i as confidenceInterval, l as summarizeNumberSeries, n as WeightedCompositeInput, o as pearsonR, r as WeightedCompositeResult, s as ranks, t as SeriesDistribution, u as weightedComposite } from "./descriptive-B2iPaT9J.js";
 import { n as bonferroni, r as holm, t as benjaminiHochberg } from "./multiplicity-DIWHvysC.js";
-import { A as PairArmsOptions, At as EProcess, Bt as isBinaryOutcomeVector, C as hashJson, D as ComparePairedArmsOptions, F as PairedCorrectness, Ft as ExactRiskDifferenceResult, Gt as pairedRiskDifferenceScore, Ht as pairedBinaryScale, I as PairedMetricDelta, It as McNemarResult, Kt as passAtK, L as comparePairedArms, Lt as ProportionInterval, M as PairRunRecordsResult, Mt as EProcessState, N as PairedArmRow, Nt as EProcessStep, O as MatchedPair, P as PairedArmsComparison, Pt as eProcess, R as pairArms, Rt as RiskDifferenceResult, Ut as pairedRiskDifference, Vt as mcnemar, Wt as pairedRiskDifferenceExact, _t as AttestationProvenance, bt as attest, et as Objective, gt as ATTESTATION_ALGORITHM, j as PairArmsResult, jt as EProcessOptions, k as MatchedRunRecordPair, nt as dominates, qt as wilson, rt as paretoFrontier, tt as ParetoResult, vt as AttestationVerification, w as manifestContentDigest, xt as verifyAttestation, yt as AttestedReport, z as pairRunRecords, zt as ScoreRiskDifferenceResult } from "./statistical-heldout-CYpGaoDH.js";
-import { _ as pairedSignTest, a as PairedPromotionDecision, c as BOOTSTRAP_GATE_MIN_N, d as PairedBootstrapResult, f as PairedSignTestResult, g as pairedDeltaTieFraction, h as pairedBootstrap, i as PairedMcNemarEvidence, l as DECISION_PAIRED_DELTA_STATISTIC, m as SignTestAlternative, n as PairedDecisionShape, o as PairedPromotionDecisionOptions, p as PairedTTestResult, r as PairedDecisionStatistic, s as decidePairedPromotion, t as PairedDecisionMethod, u as PairedBootstrapOptions, v as pairedTTest } from "./paired-promotion-decision-DPKz8ZnF.js";
+import { $ as ParetoResult, A as PairRunRecordsResult, At as EProcessStep, Bt as pairedRiskDifferenceScore, D as MatchedRunRecordPair, Dt as EProcess, E as MatchedPair, F as comparePairedArms, Ft as RiskDifferenceResult, Ht as wilson, I as pairArms, It as ScoreRiskDifferenceResult, L as pairRunRecords, Lt as mcnemar, M as PairedArmsComparison, Mt as ExactRiskDifferenceResult, N as PairedCorrectness, Nt as McNemarResult, O as PairArmsOptions, Ot as EProcessOptions, P as PairedMetricDelta, Pt as ProportionInterval, Q as Objective, Rt as pairedRiskDifference, S as manifestContentDigest, T as ComparePairedArmsOptions, Vt as passAtK, _t as attest, et as dominates, gt as AttestedReport, ht as AttestationVerification, j as PairedArmRow, jt as eProcess, k as PairArmsResult, kt as EProcessState, mt as AttestationProvenance, pt as ATTESTATION_ALGORITHM, tt as paretoFrontier, vt as verifyAttestation, x as hashJson, zt as pairedRiskDifferenceExact } from "./statistical-heldout-BHfLZHrm.js";
+import { a as PairedPromotionDecision, c as BOOTSTRAP_GATE_MIN_N, d as PairedSignTestResult, f as PairedTTestResult, g as pairedTTest, h as pairedSignTest, i as PairedMcNemarEvidence, l as PairedBootstrapOptions, m as pairedBootstrap, n as PairedDecisionShape, o as PairedPromotionDecisionOptions, p as SignTestAlternative, r as PairedDecisionStatistic, s as decidePairedPromotion, t as PairedDecisionMethod, u as PairedBootstrapResult } from "./paired-promotion-decision-CQMJ5xGN.js";
 import { a as requiredPairedSampleSize, i as pairedMde, n as mcnemarPower, o as requiredSampleSize, r as mcnemarRequiredN, t as mulberry32 } from "./random-DGEyr3zG.js";
-import { C as WilcoxonSignedRankResult, S as RankTestOptions, T as wilcoxonSignedRank, a as bootstrapCi, b as RankTestMethod, d as ReleaseConfidenceIssue, f as ReleaseConfidenceMetrics, g as ReleaseTraceEvidence, i as Verdict, n as BootstrapResult, p as ReleaseConfidenceScorecard, s as ActionableSideInfo, t as BootstrapOptions, u as ReleaseConfidenceInput, v as evaluateReleaseConfidence, w as mannWhitneyU, x as RankTestMethodRequest, y as MannWhitneyResult } from "./promotion-gate-BawrdPYr.js";
-import { Ar as DEFAULT_RED_TEAM_CORPUS, At as RequirementCheck, Br as CanaryKind, Ct as extractProducedState, Dt as LlmCorrectnessCheckerOpts, Et as CorrectnessChecker, Fr as redTeamDataset, Ft as verifyCompletion, Hr as CanaryReport, Ir as redTeamReport, Ji as llmJudge, Lr as scoreRedTeamOutput, Mr as RedTeamCategory, Mt as TaskGold, Nr as RedTeamFinding, Nt as completionVerdict, Ot as ProducedProposal, Pr as RedTeamReport, Pt as createLlmCorrectnessChecker, Qi as transientDispatchFailure, Rr as CanaryAlert, St as ToolCallEventLike, Tt as CompletionVerdict, Ur as runCanaries, Vr as CanaryOptions, Zi as quotaExhaustedUntil, _t as assertRealBackend, bt as ProposalEventLike, da as CampaignCellFailureReceipt, fa as CampaignCellRetryPolicy, gt as BackendIntegrityReport, ht as BackendIntegrityError, jr as RedTeamCase, jt as SatisfiedBy, kt as ProducedState, ma as runCampaign, pa as RunCampaignOptions, qi as LlmJudgeOptions, vt as summarizeBackendIntegrity, wt as CompletionRequirement, xt as RuntimeEventLike, yt as ArtifactEventLike, zr as CanaryEvaluation } from "./index-CyszSg4O.js";
-import { R as Scenario$1, S as JudgeConfig, _ as GateDecision, a as CampaignResult, w as JudgeScore$1 } from "./types-Cq8mSbq8.js";
-import { a as RunFilter, i as InMemoryTraceStore, n as FileSystemTraceStore, o as SpanFilter, s as TraceStore, t as EventFilter } from "./store-C4avFjIi.js";
+import { C as mannWhitneyU, S as WilcoxonSignedRankResult, _ as evaluateReleaseConfidence, a as bootstrapCi, b as RankTestMethodRequest, d as ReleaseConfidenceIssue, f as ReleaseConfidenceMetrics, g as ReleaseTraceEvidence, i as Verdict, n as BootstrapResult, p as ReleaseConfidenceScorecard, s as ActionableSideInfo, t as BootstrapOptions, u as ReleaseConfidenceInput, v as MannWhitneyResult, w as wilcoxonSignedRank, x as RankTestOptions, y as RankTestMethod } from "./promotion-gate-CxEHRsZW.js";
+import { $ as summarizeBackendIntegrity, Ci as transientDispatchFailure, Ct as RequirementCheck, Dt as createLlmCorrectnessChecker, Et as completionVerdict, Ii as CampaignCellFailureReceipt, Li as CampaignCellRetryPolicy, Ot as verifyCompletion, Q as assertRealBackend, Ri as RunCampaignOptions, St as ProducedState, Tt as TaskGold, X as BackendIntegrityError, Z as BackendIntegrityReport, _r as CanaryEvaluation, _t as CompletionRequirement, bi as llmJudge, br as CanaryReport, bt as LlmCorrectnessCheckerOpts, cr as DEFAULT_RED_TEAM_CORPUS, dr as RedTeamFinding, fr as RedTeamReport, ft as ArtifactEventLike, gr as CanaryAlert, gt as extractProducedState, hr as scoreRedTeamOutput, ht as ToolCallEventLike, lr as RedTeamCase, mr as redTeamReport, mt as RuntimeEventLike, pr as redTeamDataset, pt as ProposalEventLike, ur as RedTeamCategory, vr as CanaryKind, vt as CompletionVerdict, wt as SatisfiedBy, xr as runCanaries, xt as ProducedProposal, yi as LlmJudgeOptions, yr as CanaryOptions, yt as CorrectnessChecker, zi as runCampaign } from "./index-B16FNdW5.js";
+import { R as Scenario, S as JudgeConfig, _ as GateDecision, a as CampaignResult, w as JudgeScore } from "./types-DtkO1G9L.js";
+import { a as RunFilter, i as InMemoryTraceStore, n as FileSystemTraceStore, o as SpanFilter, s as TraceStore, t as EventFilter } from "./store-B3nWRy3B.js";
 import { a as MintedRolloutLine } from "./schema-C-ziyJD7.js";
-import { n as ScorePreference } from "./reward-CpCmAM8L.js";
-import { a as mintRolloutRows, i as RolloutSearchLineage, n as MintRolloutResult, t as MintRolloutOptions } from "./mint-DNU8QPG4.js";
-import { a as ProfileAxisSpec, c as expandProfileAxes, i as HarnessType, l as harnessAxisOf, n as CODING_HARNESSES, o as agentProfileHash, r as HARNESS_NATIVE_MODEL, s as agentProfileId, t as AgentProfile } from "./agent-profile-VFcQncbW.js";
+import { n as ScorePreference } from "./reward-DrZ71vnf.js";
+import { a as mintRolloutRows, i as RolloutSearchLineage, n as MintRolloutResult, t as MintRolloutOptions } from "./mint-BRZXuwRi.js";
+import { a as ProfileAxisSpec, c as expandProfileAxes, i as HarnessType, l as harnessAxisOf, n as CODING_HARNESSES, o as agentProfileHash, r as HARNESS_NATIVE_MODEL, s as agentProfileId, t as AgentProfile } from "./agent-profile-DTt8uGP9.js";
 import { i as DatasetSplit, n as DatasetManifest, r as DatasetScenario, t as Dataset } from "./dataset-DQqhOCPt.js";
-import { C as defineEquivalenceCheck, S as buildEquivalenceRecord, _ as StrategyChecker, a as CheckerIdentity, b as VerificationStrategyProfile, c as EquivalenceCheckDefinition, d as EquivalenceCheckerInput, f as EquivalenceCheckerResult, g as EquivalenceRecord, h as EquivalenceProtocolError, i as equivalenceVerdict, l as EquivalenceCheckSpec, m as EquivalenceObligationStatus, n as VerdictCertification, o as CheckerOutcome, p as EquivalenceObligation, r as certificationEvidenceDigest, s as EquivalenceArm, t as DefaultVerdict, u as EquivalenceChecker, v as VERIFICATION_STRATEGIES, w as runEquivalenceCheck, x as VerificationStrategySource, y as VERIFICATION_STRATEGY_SOURCES } from "./verdict-E4eRNf7-.js";
-import { a as MultiLayerVerifier, c as VerifyOptions, i as LayerStatus, l as gradeSemanticStatus, n as Layer, o as Severity, r as LayerResult, s as VerificationReport, t as Finding } from "./multi-layer-verifier-BUaQ4C17.js";
-import { C as HeldOutGateConfig, S as HeldOutGate, T as SplitCoverage, _ as gainHistogram, a as ResearchReportOptions, b as GateDecision$1, d as GainDistributionOptions, g as SummaryTableRow, h as SummaryTableOptions, l as GainDistributionBin, m as SummaryTable, p as ParetoPoint, t as ResearchReport, u as GainDistributionFigureSpec, v as paretoChart, w as HeldOutGateRejectionCode, x as GateEvidence, y as summaryTable } from "./summary-report-bdFDXq0b.js";
-import { a as FAILURE_BLAME, c as FailureContext, d as INFRA_FAILURE_BLAMES, f as classifyFailure, i as DEFAULT_FAILURE_REASON_RULES, l as FailureReasonRule, m as failureBlame, o as FailureBlame, p as classifyFailureReason, r as failureClusterView, s as FailureClassification, u as FailureRule } from "./failure-cluster-CbgEDcFJ.js";
-import { d as InsightReport, n as analyzeRuns, t as AnalyzeRunsOptions } from "./analyze-runs-D68fd-su.js";
-import { i as ExactAnalystRunEvent, l as ExactCapableAnalyst, o as ExactAnalystRunResult } from "./exact-types-ClugMJfI.js";
-import { a as BudgetPolicy, c as RegistryRunOpts, i as AnalystRegistryOptions, s as ExactRegistryRunOpts, t as AnalystRegistry } from "./registry-D7_hm2fK.js";
-import { _ as SelfImproveMethodResult, a as DefinedAgentEval, c as SelfImproveMethodOptions, d as SelfImproveProposerOptions, f as SelfImproveProposerResult, g as SelfImproveMethodProvenance, h as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveOptions, n as AgentEvalEvaluateOptions, o as defineAgentEval, p as SelfImproveResult, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveProgressEvent } from "./define-agent-eval-WSck0CZJ.js";
-import { a as createTraceAnalyst, i as TraceAnalystDefinition, n as buildDefaultAnalystRegistry, r as CreateTraceAnalystOptions, t as DefaultAnalystRegistryOptions } from "./default-registry-DzHiGruW.js";
-import { i as TraceAnalysisEngineResult, l as RawAnalystFinding, n as TraceAnalysisEngine, t as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS } from "./engine-3A_4jOlI.js";
-import { A as SemanticConceptJudgeInput, C as DspyRlmTraceEngineOptions, D as ConceptFinding, E as createChatTraceEngine, M as SemanticConceptJudgeResult, N as runSemanticConceptJudge, O as ConceptSpec, T as ChatTraceEngineOptions, _ as FindingSubject, a as FAILURE_MODE_KIND_SPEC, d as FindingsStore, f as PersistedFinding, j as SemanticConceptJudgeOptions, k as SEMANTIC_CONCEPT_JUDGE_VERSION, l as DiffPolicy, m as diffFindings, t as DEFAULT_TRACE_ANALYST_KINDS, u as FindingsDiff, v as FindingSubjectKind, w as createDspyRlmTraceEngine } from "./index-DV-jSzxF.js";
-import { i as BenchmarkEvaluation } from "./types-Cbe2fWBa.js";
-import { a as SpanHandle, n as CaptureWrite, o as TraceEmitter, s as TraceEmitterOptions, t as CaptureErrorHandler } from "./emitter-HcCmFQl_.js";
+import { C as runEquivalenceCheck, S as defineEquivalenceCheck, _ as StrategyChecker, a as CheckerIdentity, b as VerificationStrategyProfile, c as EquivalenceCheckDefinition, d as EquivalenceCheckerInput, f as EquivalenceCheckerResult, g as EquivalenceRecord, h as EquivalenceProtocolError, i as equivalenceVerdict, l as EquivalenceCheckSpec, m as EquivalenceObligationStatus, n as VerdictCertification, o as CheckerOutcome, p as EquivalenceObligation, r as certificationEvidenceDigest, s as EquivalenceArm, t as DefaultVerdict, u as EquivalenceChecker, v as VERIFICATION_STRATEGIES, x as VerificationStrategySource, y as VERIFICATION_STRATEGY_SOURCES } from "./verdict-C24NELtN.js";
+import { a as MultiLayerVerifier, c as VerifyOptions, i as LayerStatus, l as gradeSemanticStatus, n as Layer, o as Severity, r as LayerResult, s as VerificationReport, t as Finding } from "./multi-layer-verifier-DEgMUk6v.js";
+import { C as HeldOutGateRejectionCode, S as HeldOutGateDecision, _ as gainHistogram, a as ResearchReportOptions, b as HeldOutGate, d as GainDistributionOptions, g as SummaryTableRow, h as SummaryTableOptions, l as GainDistributionBin, m as SummaryTable, p as ParetoPoint, t as ResearchReport, u as GainDistributionFigureSpec, v as summaryTable, w as SplitCoverage, x as HeldOutGateConfig, y as GateEvidence } from "./summary-report-CkbRcSuu.js";
+import { a as FailureClassification, c as FailureRule, d as failureBlame, i as FailureBlame, l as classifyFailure, o as FailureContext, r as failureClusterView, s as FailureReasonRule, u as classifyFailureReason } from "./failure-cluster-DN0t63Z0.js";
+import { d as InsightReport, n as analyzeRuns, t as AnalyzeRunsOptions } from "./analyze-runs-DjlZOVXx.js";
+import { _ as SelfImproveMethodResult, a as DefinedAgentEval, c as SelfImproveMethodOptions, d as SelfImproveProposerOptions, f as SelfImproveProposerResult, g as SelfImproveMethodProvenance, h as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveOptions, n as AgentEvalEvaluateOptions, o as defineAgentEval, p as SelfImproveResult, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveProgressEvent } from "./define-agent-eval-BlH10zEw.js";
+import { i as ExactAnalystRunEvent, l as ExactCapableAnalyst, o as ExactAnalystRunResult } from "./exact-types-CflbJ9f3.js";
+import { a as createTraceAnalyst, i as TraceAnalystDefinition, n as buildDefaultAnalystRegistry, r as CreateTraceAnalystOptions, t as DefaultAnalystRegistryOptions } from "./default-registry-DXgLmoDI.js";
+import { r as TraceAnalysisEngineResult, s as RawAnalystFinding, t as TraceAnalysisEngine } from "./engine-BvZ0R-36.js";
+import { a as BudgetPolicy, c as RegistryRunOpts, i as AnalystRegistryOptions, s as ExactRegistryRunOpts, t as AnalystRegistry } from "./registry-ssNoHzwD.js";
+import { A as SemanticConceptJudgeOptions, C as createDspyRlmTraceEngine, D as ConceptSpec, E as ConceptFinding, M as runSemanticConceptJudge, O as SEMANTIC_CONCEPT_JUDGE_VERSION, S as DspyRlmTraceEngineOptions, T as createChatTraceEngine, _ as FindingSubjectKind, a as FAILURE_MODE_KIND_SPEC, d as FindingsStore, f as PersistedFinding, g as FindingSubject, j as SemanticConceptJudgeResult, k as SemanticConceptJudgeInput, l as DiffPolicy, p as diffFindings, t as DEFAULT_TRACE_ANALYST_KINDS, u as FindingsDiff, w as ChatTraceEngineOptions } from "./index-BLGqZ81N.js";
+import { i as BenchmarkEvaluation } from "./types-BKCzzmit.js";
+import { a as SpanHandle, n as CaptureWrite, o as TraceEmitter, s as TraceEmitterOptions, t as CaptureErrorHandler } from "./emitter-DABYYTcH.js";
 import { n as SandboxDriver } from "./sandbox-harness-BlSOu4LX.js";
-import { _ as redactText, d as ShareSafetyVerdict, f as assessShareSafety, g as redactForShare, h as redact, i as RedactOptions, o as RedactionProfile, r as REDACTION_VERSION, s as RedactionReport, u as ShareSafetyStatus, v as shareAllowed } from "./redact-mtCCcT8z.js";
-import { a as PairedEvalueStep, c as SequentialDecision, i as PairedEvalueSequence, l as evaluateInterimReleaseConfidence, n as InterimReleaseConfidenceInput, r as PairedEvalueOptions, t as InterimReleaseConfidence, u as pairedEvalueSequence } from "./sequential-BhsrMupG.js";
-import { _ as WelchTestResult, b as welchsTTest, d as ToolStats, f as ToolUseMetrics, i as toolWasteView, m as computeToolUseMetrics, p as ToolUseOptions, s as judgeAgreementView, u as budgetBreachView, v as WelchTestStatus, y as iqr } from "./tool-waste-BKgvd-TP.js";
-import { A as AnalystFindingDigest, B as ControlRunResult, C as createFeedbackTrajectory, D as renderPreferenceMemoryMarkdown, E as feedbackTrajectoryToOptimizerRow, F as ControlActionOutcome, G as StopDecision, H as ControlRuntimeError, I as ControlBudget, L as ControlContext, M as AnalystRunDigest, N as analystFindingDigest, O as summarizePreferenceMemory, P as analystRunDigest, R as ControlDecision, S as controlRunToFeedbackTrajectory, T as feedbackTrajectoriesToOptimizerRows, U as ControlSeverity, V as ControlRuntimeConfig, W as ControlStep, _ as PreferenceMemoryEntry, a as FeedbackLabel, b as analystRunToReviewRequests, c as FeedbackOptimizerRow, d as FeedbackTask, f as FeedbackTrajectory, g as InMemoryFeedbackTrajectoryStore, h as FileSystemFeedbackTrajectoryStore, i as FeedbackAttempt, j as AnalystReviewDecision, k as withAssignedFeedbackSplit, l as FeedbackOutcome, m as FeedbackTrajectoryStore, n as AnalystReviewRequest, o as FeedbackLabelKind, p as FeedbackTrajectoryFilter, r as FeedbackArtifactType, s as FeedbackLabelSource, t as AnalystFeedbackTrajectoryOptions, u as FeedbackSplitPolicy, v as ProposedSideEffect, w as feedbackTrajectoriesToDatasetScenarios, x as assignFeedbackSplit, y as analystRunToFeedbackTrajectory, z as ControlEvalResult } from "./feedback-trajectory-C4EG0nNH.js";
-import { B as inferDomainKeywords, C as exportRunAsOtlp, F as buildTraceInsightContext, G as scoreTraceInsightReadiness, I as buildTraceInsightPrompt, K as tokenizeDomainWords, M as TraceInsightReadiness, N as TraceInsightSuite, R as describeTraceInsightScope, S as OtlpSpan, X as analyzeTraces, Y as AnalyzeTracesResult, _ as spanRecordsToTraceAnalysisStore, a as captureFetchToRawSink, b as OtlpExport, c as ExtractedUsage, f as extractUsageFromSse, g as otlpTextToTraceAnalysisStore, h as SpanRecordsToTraceAnalysisStoreOptions, n as toolSpansToTraceAnalysisStore, o as OtlpFlatLine, p as OtlpFileTraceStore, u as extractUsage, z as domainEvidencePattern } from "./store-tool-spans-DdCSPa7C.js";
-import { a as RunIntegrityReport, o as assertRunCaptured, t as RunIntegrityError } from "./integrity-DAMEDFaC.js";
-import { a as SteeringChange, c as CampaignRunContext, d as CampaignVariant, f as EvalCampaignOptions, h as runEvalCampaign, i as Researcher, l as CampaignRunOutcome, m as FailedRun, n as ExperimentResult, o as CampaignFactoryParams, p as EvalCampaignResult, r as FailureMode, s as CampaignIntegrityPolicy, t as ExperimentPlan, u as CampaignScenario } from "./researcher-CEcfku-T.js";
-import "./index-BFRO6chs.js";
-import { n as TrajectoryStep, r as buildTrajectory, t as Trajectory } from "./trajectory-Y1fmdlw7.js";
-import { a as runCounterfactual, i as CounterfactualRunner, n as CounterfactualMutation, r as CounterfactualResult, t as CounterfactualContext } from "./counterfactual-CPaWN99-.js";
+import { _ as shareAllowed, a as RedactionProfile, d as assessShareSafety, g as redactText, h as redactForShare, l as ShareSafetyStatus, m as redact, n as REDACTION_VERSION, o as RedactionReport, r as RedactOptions, u as ShareSafetyVerdict } from "./redact-DXdlFb_U.js";
+import { a as PairedEvalueStep, c as SequentialDecision, i as PairedEvalueSequence, l as evaluateInterimReleaseConfidence, n as InterimReleaseConfidenceInput, r as PairedEvalueOptions, t as InterimReleaseConfidence, u as pairedEvalueSequence } from "./sequential-KwGspfeR.js";
+import { S as welchsTTest, _ as computeToolUseMetrics, b as WelchTestStatus, d as Trajectory, f as TrajectoryStep, g as ToolUseOptions, h as ToolUseMetrics, i as toolWasteView, m as ToolStats, p as buildTrajectory, s as judgeAgreementView, u as budgetBreachView, x as iqr, y as WelchTestResult } from "./tool-waste-CkjzWmeY.js";
+import { A as AnalystFindingDigest, B as ControlRuntimeConfig, C as createFeedbackTrajectory, D as renderPreferenceMemoryMarkdown, E as feedbackTrajectoryToOptimizerRow, F as ControlBudget, H as ControlSeverity, I as ControlContext, L as ControlDecision, M as AnalystRunDigest, N as analystRunDigest, O as summarizePreferenceMemory, P as ControlActionOutcome, R as ControlEvalResult, S as controlRunToFeedbackTrajectory, T as feedbackTrajectoriesToOptimizerRows, U as ControlStep, V as ControlRuntimeError, W as StopDecision, _ as PreferenceMemoryEntry, a as FeedbackLabel, b as analystRunToReviewRequests, c as FeedbackOptimizerRow, d as FeedbackTask, f as FeedbackTrajectory, g as InMemoryFeedbackTrajectoryStore, h as FileSystemFeedbackTrajectoryStore, i as FeedbackAttempt, j as AnalystReviewDecision, k as withAssignedFeedbackSplit, l as FeedbackOutcome, m as FeedbackTrajectoryStore, n as AnalystReviewRequest, o as FeedbackLabelKind, p as FeedbackTrajectoryFilter, r as FeedbackArtifactType, s as FeedbackLabelSource, t as AnalystFeedbackTrajectoryOptions, u as FeedbackSplitPolicy, v as ProposedSideEffect, w as feedbackTrajectoriesToDatasetScenarios, x as assignFeedbackSplit, y as analystRunToFeedbackTrajectory, z as ControlRunResult } from "./feedback-trajectory-BWmhRBGu.js";
+import { A as TraceInsightReadiness, F as describeTraceInsightScope, G as AnalyzeTracesResult, H as tokenizeDomainWords, I as domainEvidencePattern, K as analyzeTraces, L as inferDomainKeywords, N as buildTraceInsightContext, P as buildTraceInsightPrompt, V as scoreTraceInsightReadiness, a as captureFetchToRawSink, b as OtlpSpan, c as ExtractedUsage, d as extractUsageFromSse, f as OtlpFileTraceStore, g as spanRecordsToTraceAnalysisStore, h as otlpTextToTraceAnalysisStore, j as TraceInsightSuite, m as SpanRecordsToTraceAnalysisStoreOptions, n as toolSpansToTraceAnalysisStore, o as OtlpFlatLine, u as extractUsage, v as OtlpExport, x as exportRunAsOtlp } from "./store-tool-spans-DZS0HWVf.js";
+import { a as RunIntegrityReport, o as assertRunCaptured, t as RunIntegrityError } from "./integrity-_qgZLA5p.js";
+import { a as SteeringChange, c as CampaignRunContext, d as CampaignVariant, f as EvalCampaignOptions, h as runEvalCampaign, i as Researcher, l as CampaignRunOutcome, m as FailedRun, n as ExperimentResult, o as CampaignFactoryParams, p as EvalCampaignResult, r as FailureMode, s as CampaignIntegrityPolicy, t as ExperimentPlan, u as CampaignScenario } from "./researcher-GU1tV4dn.js";
+import "./index-CiRwLv2R.js";
 import { y as OUTPUT_VALUE } from "./attribute-vocabulary-CcXqHODT.js";
-import { d as runsForScenario, i as argHash, s as judgeSpans } from "./query-DK1nWcmp.js";
+import { i as judgeSpans, n as argHash, s as runsForScenario } from "./query-Cmv_U8vf.js";
 import { SpanKind } from "@tangle-network/agent-trace-contract";
 //#region src/cost-ledger/errors.d.ts
 export declare class CostCeilingReachedError extends ValidationError {
   constructor(ceilingUsd: number, committedUsd: number, requestedUsd: number, phase: string, actor: string);
 }
 export declare class CostAccountingIncompleteError extends ValidationError {}
-export declare class CostReservationExceededError extends ValidationError {
-  constructor(actor: string, actualUsd: number, maximumUsd: number);
-}
-export declare class CostCallConflictError extends ValidationError {
-  readonly callId?: string;
-  readonly receipt?: CostReceipt;
-  constructor(message: string, options?: {
-    callId?: string;
-    receipt?: CostReceipt;
-    cause?: unknown;
-  });
-}
-export declare class CostLedgerPersistenceError extends ValidationError {
-  readonly callId?: string;
-  readonly receipt?: CostReceipt;
-  constructor(cause: unknown, callId?: string, receipt?: CostReceipt);
-}
-export declare class CostReceiptCaptureError extends ValidationError {
-  readonly callId: string;
-  readonly receipt?: CostReceipt;
-  readonly receiptError: Error;
-  constructor(callId: string, cause: unknown, receiptError: unknown, receipt?: CostReceipt);
-}
 //#endregion
 //#region src/cost-ledger/pricing.d.ts
-/** Return the canonical pricing-table key, or null when the model is unpriced. */
-export declare function modelPriceKey(model: string): string | null;
 interface CostResult {
   costUsd: number;
   costUnknown: boolean;
@@ -305,7 +278,7 @@ interface PolicyEdit {
   validationPlan?: string;
   metadata?: Record<string, unknown>;
 }
-export declare const POLICY_EDIT_CANDIDATE_RECORD_SCHEMA: 'tangle.policy-edit-candidate.v1';
+declare const POLICY_EDIT_CANDIDATE_RECORD_SCHEMA: 'tangle.policy-edit-candidate.v1';
 /** JSON-safe attribution carried with a measured candidate and its scores. */
 interface PolicyEditCandidateRecord {
   schema: typeof POLICY_EDIT_CANDIDATE_RECORD_SCHEMA;
@@ -315,10 +288,6 @@ type PolicyEditInit = Omit<PolicyEdit, 'schemaVersion' | 'editId'> & {
   schemaVersion?: PolicyEditSchemaVersion;
   editId?: string;
 };
-export declare class PolicyEditValidationError extends ValidationError {
-  readonly path: string;
-  constructor(message: string, path?: string);
-}
 interface FindingToPolicyEditOptions {
   expectedGain?: PolicyEditExpectedGain | ((finding: AnalystFinding) => PolicyEditExpectedGain | null | undefined);
   risk?: PolicyEditRisk | ((finding: AnalystFinding) => PolicyEditRisk);
@@ -337,14 +306,7 @@ interface PolicyEditAdmission {
   score: number;
   reasons: string[];
 }
-export declare function makePolicyEdit(init: PolicyEditInit): PolicyEdit;
-export declare function computePolicyEditId(edit: Omit<PolicyEdit, 'editId'> | PolicyEdit): string;
-export declare function validatePolicyEdit(input: unknown): PolicyEdit;
 export declare function makePolicyEditCandidateRecord(edit: PolicyEdit): PolicyEditCandidateRecord;
-export declare function validatePolicyEditCandidateRecord(input: unknown): PolicyEditCandidateRecord;
-export declare function policyEditsFromFindings(findings: ReadonlyArray<AnalystFinding>, opts?: FindingToPolicyEditOptions): PolicyEdit[];
-export declare function admitPolicyEdit(edit: PolicyEdit, opts?: PolicyEditAdmissionOptions): PolicyEditAdmission;
-export declare function applyPolicyEditToSurface(surface: unknown, edit: PolicyEdit): unknown;
 //#endregion
 //#region src/capability-headroom.d.ts
 /**
@@ -792,6 +754,23 @@ export declare function minimumPairsForPairedDeltaTest(confidence?: number): num
 export declare function pairedDeltaTest(before: number[], after: number[], options?: PairedDeltaTestOptions): PairedDeltaTestResult;
 //#endregion
 //#region src/partition-held-out.d.ts
+/**
+ * Deterministic held-out partitioning over stable ids.
+ *
+ * A promotion gate's paired-delta math needs a fixed partition: the same
+ * scenario id lands in the same bucket forever, regardless of when it ran or
+ * which loop emitted the record — otherwise the gate has nothing paired to
+ * compare against. Picking one global split per run is wrong: every run then
+ * stamps either ALL search or ALL holdout, and no id is ever observed on both
+ * sides. Per-id deterministic hashing fixes that.
+ *
+ * Two consumers hand-roll this (agent-builder's `deterministicSplit` and the
+ * frontier persona-splitter). The substrate already has a 3-way benchmark
+ * `deterministicSplit` in `./benchmarks`; this is the generic, validated batch
+ * partitioner: it takes a list of ids, splits them by a stable FNV-1a hash, and
+ * fails loud when the inputs can't support a trustworthy held-out comparison
+ * (duplicate ids, or a holdout set below the significance floor).
+ */
 interface PartitionHeldOutOptions {
   /** Partition seed. Bumping it reshuffles every assignment — do that only when
    *  the corpus/policy changes meaningfully so old and new records can't pair.
@@ -857,21 +836,6 @@ interface TrainSelectionTestPartition<T> {
   selectionFraction: number;
   testFraction: number;
 }
-/**
- * Partition items into disjoint train, selection and test sets, the three
- * partitions `compareOptimizationMethods` and Runtime `improve()` take.
- *
- * Assignment is per unit: `hashToUnit(unit, seed)` below `testFraction` is
- * test, below `testFraction + selectionFraction` is selection, and the rest is
- * train. A unit therefore keeps its partition as the corpus grows, and items of
- * one unit never straddle two partitions. Fail-loud like `partitionHeldOut`:
- * empty input, a duplicate or empty id, an empty unit, fractions outside
- * (0, 1) or summing to 1 or more, and a partition below its floor all throw.
- * Order within each partition follows the input order.
- */
-export declare function partitionTrainSelectionTest<T extends {
-  id: string;
-}>(items: readonly T[], options?: TrainSelectionTestOptions<T>): TrainSelectionTestPartition<T>;
 //#endregion
 //#region src/product-benchmark/export.d.ts
 interface ProductBenchmarkExportOptions {
@@ -1354,37 +1318,6 @@ interface CommandRunner {
   /** List a directory. Returns `[]` if unreadable / missing. */
   readDir(path: string): Promise<DirEntry[]>;
 }
-/**
- * Host-process runner, mapped onto {@link runBoundedProcess} so the package
- * has one spawn site rather than two.
- *
- * It used `spawnSync` and therefore had none of the bounds that runner exists
- * to provide: a `capMs` timeout killed the direct child and left any
- * descendant it had started running, a killed run could still report exit
- * code 0, captured output was unbounded, and the whole run blocked the event
- * loop. This runner spawns build, test and lint commands, which are exactly
- * the commands most likely to start descendants.
- *
- * It is not the package's last private spawn: `nodeProcessRunner`
- * (`src/trace-repair/docker-environment.ts`) and `runProcess`
- * (`src/campaign/external-optimizer-subprocess.ts`) still spawn their own
- * detached children, and several one-shot `spawnSync` git and CLI calls
- * remain. What this removes is the second spawn behind the `CommandRunner`
- * contract, which is the one that graded build, test and lint commands.
- *
- * Two behaviours the mapping changes on purpose:
- *   - a run with no `capMs` is now bounded by
- *     {@link LOCAL_COMMAND_RUNNER_DEFAULT_CAP_MS} instead of running forever;
- *   - the run no longer blocks the event loop, so a minutes-long `capMs` no
- *     longer stalls everything else in the process.
- */
-/**
- * Wall clock {@link localCommandRunner} applies when the caller names no
- * `capMs`. It is `runBoundedProcess`'s own default, stated here because this
- * runner used to be unbounded: a consumer whose command runs longer names its
- * own `capMs`, and a run the bound stops says so with `timedOut: true`.
- */
-export declare const LOCAL_COMMAND_RUNNER_DEFAULT_CAP_MS: number;
 export declare const localCommandRunner: CommandRunner;
 //#endregion
 //#region src/discover-personas.d.ts
@@ -1482,13 +1415,6 @@ interface ImprovementVerdictResult {
   /** Human-readable reason for the verdict — for dashboards and logs. */
   reason: string;
 }
-/**
- * Verdict for a candidate against its parent. Pure — operates on already-computed
- * stats. KEEP/REGRESSION require both sides to have `>= minRepsForVerdict` reps
- * AND the candidate to be `stable`; otherwise the result is NOISE (unstable) or
- * ITERATE (not enough reps / no parent).
- */
-export declare function improvementVerdict(candidate: ExperimentStats, parent: ExperimentStats | null, thresholds?: ImprovementThresholds): ImprovementVerdictResult;
 //#endregion
 //#region src/propose-review.d.ts
 interface Verification {
@@ -1649,7 +1575,6 @@ interface ProposeReviewControlConfig<State, Summary = unknown> {
   failureClassFromVerification?: (verification: Verification) => FailureClass | undefined;
   actionFailure?: ControlRuntimeConfig<ProposeReviewControlState<State, Summary>, ProposeReviewControlAction, ProposeReviewControlResult<State, Summary>>['actionFailure'];
 }
-export declare function runProposeReviewAsControlLoop<State, Summary = unknown>(config: ProposeReviewControlConfig<State, Summary>): Promise<ControlRunResult<ProposeReviewControlState<State, Summary>, ProposeReviewControlAction, ProposeReviewControlResult<State, Summary>>>;
 //#endregion
 //#region src/reflective-mutation.d.ts
 interface TrialTrace {
@@ -1804,6 +1729,31 @@ interface PreflightOutcome {
 export declare function preflightModels(opts: PreflightModelsOptions): Promise<PreflightOutcome>;
 //#endregion
 //#region src/model-seats.d.ts
+/**
+ * ModelSeats — the program's model seating chart.
+ *
+ * One object names which model fills each role in an eval program: the worker
+ * under evaluation, the judge panel, the analyst, the reflection/driver model,
+ * and the verifier. Re-tiering an entire program (economy ↔ frontier) is one
+ * swapped object instead of a hunt through call sites.
+ *
+ * Wiring points — consumers thread seats; this module implements none of them
+ * (those files belong to other surfaces):
+ *  - `judges`     → `ensembleJudge({ models: seats.judges, … })` (src/judge-panel.ts)
+ *                   and the `JudgeConfig`s handed to `makeEvalTools({ judges })`
+ *                   (src/eval-tools.ts).
+ *  - `reflection` → the model configured by a custom `SurfaceProposer` or
+ *                   external optimization engine.
+ *  - `worker`     → the dispatch model the agent itself calls — the model an
+ *                   `AgentProfile` declares.
+ *  - `analyst`    → the LLM behind `analyzeRuns` / analyst-registry kinds.
+ *  - `verifier`   → completion-verifier / objective-checker model.
+ *  - campaign cells thread `judges` + driver models the same way; that wiring
+ *    lands with the campaign surface, not here.
+ *
+ * `resolveSeat` is the only read path: an unset seat with no explicit fallback
+ * throws — a model id is a budget decision, never a silent default.
+ */
 interface ModelSeats {
   /** The model under evaluation — what the agent itself dispatches with. */
   worker?: string;
@@ -2066,6 +2016,60 @@ export declare class PairwiseSteeringOptimizer {
   optimize(rows: SteeringOptimizationRow[], config?: SteeringOptimizerConfig): SteeringOptimizationResult;
 }
 //#endregion
+//#region src/counterfactual.d.ts
+type CounterfactualMutation = {
+  kind: 'swap-model';
+  at: number;
+  newModel: string;
+} | {
+  kind: 'swap-tool-result';
+  at: number;
+  newResult: unknown;
+} | {
+  kind: 'truncate-after';
+  at: number;
+} | {
+  kind: 'inject-system-message';
+  at: number;
+  content: string;
+} | {
+  kind: 'custom';
+  at: number;
+  describe: string;
+  apply: (step: TrajectoryStep) => TrajectoryStep;
+};
+interface CounterfactualContext {
+  originalRunId: string;
+  originalTrajectory: Trajectory;
+  /** Steps up to (but not including) the mutation point — the prefix the
+   *  replayed agent inherits as its prior conversation/tool history. */
+  prefix: TrajectoryStep[];
+  mutation: CounterfactualMutation;
+  /** Pre-applied mutation on the step at `mutation.at`. Consumers use this
+   *  as the FIRST step the replayed agent emits (they decide whether to
+   *  re-emit it or continue from there). */
+  mutatedStep: TrajectoryStep;
+}
+interface CounterfactualResult {
+  counterfactualRunId: string;
+  originalRunId: string;
+  mutation: CounterfactualMutation;
+  /** Structured delta summary — caller can extend via scoring. */
+  delta: {
+    originalOutcomeScore: number | null;
+    counterfactualOutcomeScore: number | null;
+    deltaScore: number | null;
+  };
+}
+interface CounterfactualRunner {
+  /**
+   * Execute the agent from `ctx.prefix` with the mutation applied.
+   * MUST emit spans into the provided emitter so they become part of
+   * the counterfactual run. MUST call emitter.endRun() with a verdict.
+   */
+  executeFrom: (ctx: CounterfactualContext, emitter: TraceEmitter) => Promise<void>;
+}
+//#endregion
 //#region src/scorecard.d.ts
 /** One commit's measurement of one (scenario, profile) cell. */
 interface ScorecardEntry {
@@ -2720,6 +2724,40 @@ interface WranglerDeployRunnerInput {
 export declare function wranglerDeployRunner(input: WranglerDeployRunnerInput): DeployRunner;
 //#endregion
 //#region src/hidden-criteria-grading.d.ts
+/**
+ * Hidden-criteria grading firewall — grade an agent on criteria it never saw.
+ *
+ * A trustworthy benchmark splits every scenario's data by WHERE each field is
+ * allowed to flow, then proves the held-out / judge-only fields never reach the
+ * agent during the run. The coding bench expresses this with four destinations
+ * (prompt / develop-against / held-out suite / rubric); this module lifts the
+ * domain-FREE core out of it so research, legal, tax, content — any domain —
+ * can declare the same routing and get the same firewall enforcement and the
+ * same held-out-weighted composite, plugging in its OWN grader.
+ *
+ * Two reusable pieces, both domain-agnostic:
+ *
+ *   1. FIELD ROUTING BY DESTINATION. A scenario declares each field's
+ *      `FieldDestination`; `assertNoHiddenLeak` is a pure checker that throws if
+ *      a grading-only or judge-only field's value appears in what reaches the
+ *      agent. The domain decides which fields exist and where they go — the
+ *      substrate only enforces "hidden stays hidden".
+ *
+ *   2. HIDDEN-CRITERIA GRADING. The domain supplies its own grader
+ *      `(artifact, hiddenCriteria) => { passRate, total }` — the coding
+ *      node-test executor is ONE such grader a consumer plugs in; the substrate
+ *      bakes in NO node/test/TS/exec/regex. `gradeOnHidden` runs that grader
+ *      behind the firewall and `blendHeldout` composes its pass rate with a
+ *      judge score into the final number the leaderboard ranks on.
+ *
+ * Shape mirrors `treatment-gate`/`authenticity`: pure predicates and pure
+ * composition over already-computed values, fail-loud, with the
+ * "which field / which weight / which grader" decisions left as parameters and
+ * no domain literal anywhere in the module.
+ *
+ * Lives next to `test-graded-scenario` and `partition-held-out` — it is a
+ * scorecard/grading concept that makes sense without a running agent loop.
+ */
 /**
  * Where one scenario field is allowed to flow. The firewall guarantee is keyed
  * on this tag, not on a field name — a domain can have any number of fields per
@@ -2755,14 +2793,6 @@ interface RoutedField {
   value: string;
   destination: FieldDestination;
 }
-/**
- * Route a domain's named fields by destination into the firewall's input shape.
- * The `routing` declares each field's destination; the `values` carry each
- * field's renderable string. A field present in `routing` but missing from
- * `values` is an authoring error (fail loud) — every routed field must have a
- * value the firewall can check.
- */
-export declare function routeFields<TName extends string>(routing: Readonly<Record<TName, FieldDestination>>, values: Readonly<Record<TName, string>>): RoutedField[];
 /** A single detected leak: a hidden field whose value appears in the agent context. */
 interface HiddenLeak {
   field: string;
@@ -2811,31 +2841,6 @@ interface HiddenGradeResult {
  * answer key, the hidden requirements) — opaque to the substrate.
  */
 type HiddenCriteriaGrader<TArtifact, THidden> = (artifact: TArtifact, hiddenCriteria: THidden, signal?: AbortSignal) => Promise<HiddenGradeResult> | HiddenGradeResult;
-/** Normalize a grader's raw `{passed, total}` into a `HiddenGradeResult` with a
- *  consistent, fail-loud `passRate` — the canonical "honest zero on no-run"
- *  rule, single-sourced so every domain grader gets it. */
-export declare function hiddenGrade(passed: number, total: number, notes?: string): HiddenGradeResult;
-/**
- * Run a domain's hidden-criteria grader behind the firewall. Before grading, it
- * re-asserts the firewall against the agent context the run actually used —
- * proving (at grading time, on real data) that the hidden criteria never
- * reached the agent — then invokes the grader and returns its pass rate. A
- * domain that wants the firewall and the grader wired together in one call uses
- * this; a domain that already asserted the firewall at dispatch time can call
- * its grader directly and feed the result to `blendHeldout`.
- */
-export declare function gradeOnHidden<TArtifact, THidden>(args: {
-  artifact: TArtifact;
-  hiddenCriteria: THidden;
-  grader: HiddenCriteriaGrader<TArtifact, THidden>;
-  /** The routed fields + the exact agent context, re-checked before grading. */
-  firewall: {
-    fields: readonly RoutedField[];
-    agentContext: string;
-    options?: NoLeakOptions;
-  };
-  signal?: AbortSignal;
-}): Promise<HiddenGradeResult>;
 /** Weights for the held-out / judge blend. Must be finite and non-negative;
  *  they are renormalized to sum to 1 so a caller can pass any positive ratio. */
 interface BlendWeights {
@@ -2844,9 +2849,6 @@ interface BlendWeights {
   /** Weight on the judge's quality composite (the secondary style/quality score). */
   judge: number;
 }
-/** Default blend: 0.7 hidden correctness, 0.3 judge quality. The coding bench's
- *  long-standing split — execution truth dominates, style refines. */
-export declare const defaultBlendWeights: BlendWeights;
 /** The input shape a judge's `score` receives — exactly `JudgeConfig.score`'s
  *  argument: the artifact, plus any scenario/signal fields the judge carries.
  *  `withHeldoutBlend` only reads `artifact`; the rest rides through. */
@@ -2864,20 +2866,6 @@ interface JudgeScoreInput<TArtifact> {
  * to [0,1] — a judge on a non-unit scale must be normalized by the caller first.
  */
 export declare function blendHeldout(heldoutPassRate: number, judgeScore: number, weights?: BlendWeights): number;
-/**
- * Wrap a judge's `score` so the `composite` it REPORTS is the held-out-weighted
- * blend. The judge still scores its quality dimensions (recorded, secondary),
- * but the composite that downstream selection/scorecard reads becomes
- * `blendHeldout(heldoutPassRate(artifact), judgeComposite, weights)`. The held-
- * out pass rate is read off the artifact via `heldoutPassRate` — already
- * computed before the judge runs — so no second grading pass is needed.
- *
- * Generic over the artifact type, inferred from `heldoutPassRate`, so it
- * composes with both a `campaign` `JudgeConfig.score` and a bare scoring
- * function. The input is the judge's `{ artifact, ... }` — any extra fields
- * (`scenario`, `signal`) ride through untouched via the index signature.
- */
-export declare function withHeldoutBlend<TArtifact>(score: (input: JudgeScoreInput<TArtifact>) => JudgeScore$1 | Promise<JudgeScore$1>, heldoutPassRate: (artifact: TArtifact) => number, weights?: BlendWeights): (input: JudgeScoreInput<TArtifact>) => Promise<JudgeScore$1>;
 //#endregion
 //#region src/integrity/single-backend.d.ts
 /**
@@ -3254,19 +3242,7 @@ interface Oracle {
   id: string;
   check(obs: OracleObservation): OracleResult;
 }
-export declare function textInSnapshot(needle: string, opts?: {
-  caseSensitive?: boolean;
-}): Oracle;
-export declare function urlContains(fragment: string): Oracle;
 export declare function jsonShape(expected: Record<string, unknown>): Oracle;
-export declare function regexMatches(pattern: RegExp): Oracle;
-/**
- * Anti-bot detector — distinguishes genuine failures from blocked navigation
- * (cloudflare, recaptcha, etc). Returns an Oracle that PASSES when no block
- * marker is present; on block, detail names the blocker so runners can tag
- * results as "blocked" rather than "failed". Lifted from browser-agent-driver.
- */
-export declare function notBlocked(): Oracle;
 /** Extends the substrate verdict spine: `valid` = every oracle passed (and
  *  at least one ran); `score` = ratio of oracles passed; `scores` = per-
  *  oracle 0|1 keyed by oracle id. */
@@ -3436,6 +3412,45 @@ export declare function pairedPromotionPower(options: PairedPromotionPowerOption
 export declare function requiredPairsForPairedPromotion(options: RequiredPairsForPairedPromotionOptions): RequiredPairsForPairedPromotionResult;
 //#endregion
 //#region src/treatment-gate.d.ts
+/**
+ * Treatment-applied gate — "did the treatment's tool actually fire this run?"
+ *
+ * A tool-treatment A/B (mount a search provider, a browser, a code-exec
+ * sandbox, a retrieval MCP — anything one arm gets and the other doesn't) only
+ * means something on the treatment arm if that arm actually EXERCISED the tool.
+ * A treatment run that never called the mounted tool is not a weak data point
+ * — it is a NON-data-point: the manipulation never happened, so the run cannot
+ * speak to the treatment and must be excluded from the objective, the same way
+ * an infra-tainted run is excluded.
+ *
+ * This is the manipulation/validity precondition that protects the paired A/B
+ * tests downstream (`mcnemar`, `pairedRiskDifference`): they must only ever see
+ * runs where the treatment was applied, or they measure noise and report it as
+ * an effect.
+ *
+ * Shape mirrors `authenticity`'s `gateRealness`: a pure predicate over already-
+ * computed signals, with explicit fail-open/fail-closed discipline. It is not a
+ * trace parser — the tool-call telemetry it reads is the deterministic
+ * `toolHistogram` that `computeTraceMetrics(spans)` already produces (OTLP
+ * tool-name extraction via `TOOL_NAME_ATTR_KEYS`), or the `ToolSpan[]` a
+ * `toolSpans(store, runId)` query returns. The gate never re-derives either.
+ *
+ * General by construction:
+ *   - The "which tool counts as the treatment" decision is a `matches` PARAMETER
+ *     — a `(toolName: string) => boolean`. There is no `search`/`web`/`fetch`
+ *     literal anywhere in this module; a search A/B passes its own matcher.
+ *   - The "is THIS run a treatment arm (vs a control)" decision is the caller's
+ *     policy and stays at the call site. The gate only answers the narrower,
+ *     domain-free question: "given this run's telemetry, did a matching tool
+ *     fire?" — so it generalizes to any tool-treatment, not search.
+ *
+ * Fail-open on telemetry absence: when NO tool calls were captured at all
+ * (`sum(toolHistogram) === 0`), the gate cannot distinguish "the agent used no
+ * tools" from "this harness's tool calls weren't recorded". Quarantining on a
+ * telemetry gap would silently delete real runs, so the default is
+ * applied=true / gated=false. Pass `failOpenWhenNoTelemetry: false` to flip to
+ * fail-closed for harnesses where tool capture is guaranteed.
+ */
 /** A tool-name matcher: does this tool name belong to the treatment under test?
  *  The caller supplies it — the substrate ships no `search` (or any) literal. */
 type ToolMatcher = (toolName: string) => boolean;
@@ -3488,8 +3503,8 @@ export declare function contentHash(value: unknown): string;
 /** Pluggable verdict store. Sync or async on both legs — `cachedJudge`
  *  awaits the results either way. */
 interface VerdictCacheStore {
-  get(key: string): Promise<JudgeScore$1 | undefined> | JudgeScore$1 | undefined;
-  set(key: string, score: JudgeScore$1): Promise<void> | void;
+  get(key: string): Promise<JudgeScore | undefined> | JudgeScore | undefined;
+  set(key: string, score: JudgeScore): Promise<void> | void;
 }
 /**
  * JSONL-file-backed store: the full file is loaded into an in-memory index at
@@ -3731,28 +3746,6 @@ export declare function runAgentControlLoop<TState, TAction, TActionResult, TEva
 export declare function objectiveEval(input: Omit<ControlEvalResult, 'objective'>): ControlEvalResult;
 export declare function subjectiveEval(input: Omit<ControlEvalResult, 'objective'>): ControlEvalResult;
 //#endregion
-//#region src/analyst/steer-firewall.d.ts
-/**
- * THE steer firewall. Fail-loud guard for any path that admits analyst findings
- * as STEERING input (the `f(trace)` role): rejects — naming the offenders — any
- * finding whose provenance is a judge verdict, rather than let `J` leak into the
- * loop. Returns the findings unchanged for chaining.
- *
- * Call this at the chokepoint where a detector that ALSO scores/gates has its
- * findings turned into a steer (the judge-and-steer dual-role case). It keys on
- * provenance, so it correctly admits evidence-less trace-analyst observations and
- * correctly rejects an artifact-citing judge verdict — the cases an evidence
- * check gets backwards.
- *
- * It is necessary, not sufficient: it stops PROVENANCE-tagged verdicts. A judge
- * whose output is laundered through a hand-built finding with no provenance flag
- * is out of its reach — provenance must be honestly set at every judge→finding
- * lift (today: createJudgeAdapter). That is why the integrity rule lives at the
- * lift site, and why ProposeContext.judgeScores?: never is the complementary
- * compile-time tripwire on the obvious direct channel.
- */
-export declare function assertNoJudgeVerdict(findings: ReadonlyArray<AnalystFinding>, context?: string): ReadonlyArray<AnalystFinding>;
-//#endregion
 //#region src/knowledge/readiness.d.ts
 interface ScoreKnowledgeReadinessOptions {
   taskId: string;
@@ -3800,7 +3793,7 @@ export declare function estimateCost(inputTokens: number, outputTokens: number,
 //#endregion
 //#region src/anti-slop.d.ts
 interface AntiSlopConfig {
-  /** Domain label — appears in the JudgeScore output */
+  /** Domain label — appears in the DimensionJudgeScore output */
   domain?: string;
   /** Case-insensitive substrings that must not appear. Each occurrence = penalty. */
   bannedPhrases?: string[];
@@ -3842,5 +3835,5 @@ export declare function analyzeAntiSlop(outputs: string[], config: Omit<Required
   penaltyWeights: Record<SlopCategory, number>;
 }): AntiSlopReport;
 //#endregion
-export { AGENT_PROFILE_KINDS, ATTESTATION_ALGORITHM, type ActionExecutionPolicy, type ActionPolicyDecision, type ActionableSideInfo, type ActiveLearningOptions, type AgentEvalAgent, AgentEvalError, type AgentEvalErrorCode, type AgentEvalEvaluateOptions, type AgentEvalImproveOptions, type AgentProfile, type AgentProfileCell, type AgentProfileCellInput, type AgentProfileDimensionValue, type AgentProfileJson, type AgentProfileSourceInput, type Analyst, type AnalystContext, type AnalystFeedbackTrajectoryOptions, type AnalystFinding, type AnalystFindingDigest, AnalystRegistry, type AnalystRegistryOptions, type AnalystReviewDecision, type AnalystReviewRequest, type AnalystRunDigest, type AnalystRunEvent, type AnalystRunInputs, type AnalystRunResult, type AnalystRunSummary, type AnalystSeverity, type AnalystUsageReceipt, type AnalyzeRunsOptions, type AnalyzeTracesResult, type ArgumentCheck, type Artifact, type ArtifactCheck, type ArtifactEventLike, type ArtifactResult, type AssertCapabilityHeadroomOptions, type AssertCrossFamilyOptions, type AssertCrossFamilyServedOptions, type AssertServedModelOptions, type AssertSingleBackendOptions, type AttestationProvenance, type AttestationVerification, type AttestedReport, BOOTSTRAP_GATE_MIN_N, type BackendDescriptor, BackendIntegrityError, type BackendIntegrityReport, type BenchmarkEvaluation, type BlendWeights, type BootstrapOptions, type BootstrapResult, type BoundedProcessInput, type BoundedProcessResult, type BudgetLedgerEntry, type BudgetPolicy, type BudgetSpec, CODING_HARNESSES, type CalibrationExample, type CalibrationMerge, type CalibrationResult, type CampaignCellFailureReceipt, type CampaignCellRetryPolicy, type CampaignFactoryParams, type CampaignIntegrityPolicy, type CampaignResult, type CampaignRunContext, type CampaignRunOutcome, type CampaignScenario, type CampaignVariant, type CanaryAlert, type CanaryEvaluation, type CanaryKind, type CanaryLeak, type CanaryOptions, type CanaryReport, type CandidateScore, type CapabilityHeadroomOptions, type CapabilityHeadroomResult, type CaptureErrorHandler, type CaptureWrite, type CellVerdict, type ChannelRollup, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTraceEngineOptions, type ChatTransport, type CheckResult, type CheckerIdentity, type CheckerOutcome, type CliffsMagnitude, type CollectedArtifacts, type CommandRunner, type ComparePairedArmsOptions, type CompletionCriterion, type CompletionRequirement, type CompletionVerdict, type ConceptFinding, type ConceptSpec, ConfigError, type ContinuousAgreement, type ContinuousAgreementOptions, type ContinuousCalibrationResult, type ContractCheckResult, type ContractLintFinding, type ContractRule, type ContractRuleExecution, type ContractRun, type ContractSpan, type ContractStatus, type ContractVerdict, type ControlActionOutcome, type ControlBudget, type ControlContext, type ControlDecision, type ControlEvalResult, type ControlRunResult, type ControlRuntimeConfig, type ControlRuntimeError, type ControlStep, type CorpusAgreementOptions, type CorpusAgreementPerDimension, type CorpusAgreementReport, type CorpusScoreRecord, type CorrectnessChecker, type CostChannel, type CostEntry, CostLedger, type CostLedgerFilter, type CostLedgerHandle, type CostLedgerOptions, type CostLedgerPersistence, type CostLedgerSummary, type CostProvenance, type CostReceipt, type CostReceiptInput, type CostResult, type CostSummary, type CostUsage, type CounterfactualContext, type CounterfactualMutation, type CounterfactualResult, type CounterfactualRunner, type CreateChatClientOpts, type CreateTraceAnalystOptions, CrossFamilyError, type CustomTokenPricing, type CustomTransportOpts, DECISION_PAIRED_DELTA_STATISTIC, DEFAULT_FAILURE_REASON_RULES, DEFAULT_JUDGE_GATE_POLICY, DEFAULT_RED_TEAM_CORPUS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, type DataAcquisitionPlan, type DatasetManifest, type DatasetOverview, type DatasetScenario, type DatasetSplit, type DefaultAnalystRegistryOptions, type DefaultVerdict, type DefineAgentEvalOptions, type DefinedAgentEval, type DeployGateLayerInput, type DeployRunResult, type DeployRunner, type DetectorEvent, type DetectorSignal, type DiffPolicy, type DiffScorecardOptions, type JudgeScore as DimensionJudgeScore, type DirEntry, type DiscoverPersonasOptions, type DiscoveredPersona, type DriverState, type DspyRlmTraceEngineOptions, type EProcess, type EProcessOptions, type EProcessState, type EProcessStep, type EnsembleAggregate, type EnsembleJudgeOptions, type EquivalenceArm, type EquivalenceCheckDefinition, type EquivalenceCheckSpec, type EquivalenceChecker, type EquivalenceCheckerInput, type EquivalenceCheckerResult, type EquivalenceObligation, type EquivalenceObligationStatus, EquivalenceProtocolError, type EquivalenceRecord, type ErrorCluster, type ErrorCountPattern, type ErrorStreakOptions, type EvalCampaignOptions, type EvalCampaignResult, type EventFilter, type EvidenceRef, type ExactAnalystRunEvent, type ExactAnalystRunResult, type ExactCapableAnalyst, type ExactRegistryRunOpts, type ExactRiskDifferenceResult, type ExperimentPlan, type ExperimentRep, type ExperimentResult, type ExperimentStats, type ExtractOptions, type ExtractResult, type ExtractedUsage, FAILURE_BLAME, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, type FailedRun, type FailureBlame, type FailureClass, type FailureClassification, type FailureContext, type FailureMode, type FailureReasonRule, type FailureRule, type FeedbackArtifactType, type FeedbackAttempt, type FeedbackLabel, type FeedbackLabelKind, type FeedbackLabelSource, type FeedbackOptimizerRow, type FeedbackOutcome, type FeedbackSplitPolicy, type FeedbackTask, type FeedbackTrajectory, type FeedbackTrajectoryFilter, type FeedbackTrajectoryStore, type FieldDestination, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, FileSystemTraceStore, type Finding, type FindingSubject, type FindingSubjectKind, type FindingToPolicyEditOptions, type FindingsDiff, FindingsStore, type GainDistributionBin, type GainDistributionFigureSpec, type GainDistributionOptions, type GateDecision, type GateEvidence, type GenericSpan, type GoldenItem, HARNESS_NATIVE_MODEL, type HarnessType, type HeadroomInput, HeldOutGate, type HeldOutGateConfig, type GateDecision$1 as HeldOutGateDecision, type HeldOutGateRejectionCode, type HeldOutPartition, type HiddenCriteriaGrader, type HiddenGradeResult, type HiddenLeak, INFRA_FAILURE_BLAMES, type ImprovementThresholds, type ImprovementVerdictResult, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, InMemoryTraceStore, type InsightReport, type IntentMatchInput, type IntentMatchOptions, type IntentMatchResult, type InterimReleaseConfidence, type InterimReleaseConfidenceInput, type JudgeAgreement, type JudgeCalibrationSet, type JudgeConfusion, JudgeError, type JudgeExampleAgreement, type JudgeFamily, type JudgeFn, type JudgeGateDecision, type JudgeGateInput, type JudgeGatePolicy, JudgeGateRefusedError, type JudgeGateStatus, type JudgeInput, type JudgeRetryOutcome, type JudgeRetryPolicy, type JudgeRubric, type JudgeRun, type JudgeRunMeta, type JudgeScore$1 as JudgeScore, type JudgeScoreInput, type JudgeScoresRecord, type JudgeSpan, type JudgeVerdict, type KeywordConceptSpec, type KeywordCoverageFinding, type KeywordCoverageOptions, type KeywordCoverageResult, type KnowledgeAcquisitionMode, type KnowledgeBundle, type KnowledgeFreshness, type KnowledgeImportance, type KnowledgeReadinessReport, type KnowledgeRequirement, type KnowledgeRequirementCategory, type KnowledgeSensitivity, type Layer, type LayerResult, type LayerStatus, type LeaderboardOptions, type LeaderboardRow, type LinearRegressionResult, type LlmCallMetadata, type LlmCallRequest, type LlmCallResult, type LlmChargeBounds, type LlmCorrectnessCheckerOpts, type LlmJudgeOptions, type LlmMessage, type LlmReviewerConfig, type LlmSpan, type LlmTokenLogprob, type LlmToolCall, type LlmToolChoice, type LlmToolDefinition, type LlmUsage, type MannWhitneyResult, type MatchedPair, type MatchedRunRecordPair, type MaximumCharge, type McNemarResult, type MintRolloutOptions, type MintRolloutResult, type MintedRolloutLine, type MockTransportOpts, type ModelEndpointCheck, type ModelEndpointRequest, type ModelSeats, ModelSubstitutionError, MultiLayerVerifier, type NoLeakOptions, NoopRawProviderSink, NotFoundError, OUTPUT_VALUE, type Objective, type OpenAiCompatibleTransportOpts, type Oracle, type OracleObservation, type OracleReport, type OracleResult, type OrderMode, type OtlpExport, OtlpFileTraceStore, type OtlpFlatLine, type OtlpSpan, type OwnerVerdict, type PaidCallResult, type PairArmsOptions, type PairArmsResult, type PairRunRecordsResult, type PairedArmRow, type PairedArmsComparison, type PairedBootstrapOptions, type PairedBootstrapResult, type PairedCorrectness, type PairedDecisionMethod, type PairedDecisionShape, type PairedDecisionStatistic, type PairedDeltaTestOptions, type PairedDeltaTestResult, type PairedEvalueOptions, type PairedEvalueSequence, type PairedEvalueStep, type PairedMcNemarEvidence, type PairedMetricDelta, type PairedPromotionAlternative, type PairedPromotionAlternativeCell, type PairedPromotionCallRates, type PairedPromotionDecision, type PairedPromotionDecisionOptions, type PairedPromotionPowerCall, type PairedPromotionPowerOptions, type PairedPromotionPowerResult, type PairedSignTestResult, type PairedTTestResult, type PairedValueLaw, type ParetoPoint, type ParetoResult, type PartitionHeldOutOptions, type PendingCostCall, type PendingCostCallView, type PersistedFinding, type PersonaConfig, type PolicyEdit, type PolicyEditAdmission, type PolicyEditAdmissionOptions, type PolicyEditAxis, type PolicyEditCandidateRecord, type PolicyEditChange, type PolicyEditExpectedGain, type PolicyEditGainDirection, type PolicyEditGainUnit, type PolicyEditInit, type PolicyEditRisk, type PolicyEditSchemaVersion, type PolicyEditSource, type PolicyEditTarget, type PolicyEditTargetSurface, type PreferenceMemoryEntry, type ProducedProposal, type ProducedState, type ProductBenchmarkExportOptions, type ProductBenchmarkExportResult, type ProductBenchmarkManifest, type ProductBenchmarkSingleRunExportOptions, type ProductBenchmarkSplit, type ProductBenchmarkValidationReport, type ProductClientConfig, type Scenario as ProductScenario, type ProfileAxisSpec, type ProjectRuntimeTrajectoryEvidenceOptions, type PromptHandle, type ProportionInterval, type ProposalEventLike, type ProposalFinding, type ProposeFn, type ProposeInput, type ProposeOutput, type ProposeReviewConfig, type ProposeReviewControlAction, type ProposeReviewControlConfig, type ProposeReviewControlResult, type ProposeReviewControlState, type ProposeReviewReport, type ProposedSideEffect, type QueryTracesPage, REDACTION_VERSION, type RankTestMethod, type RankTestMethodRequest, type RankTestOptions, type RawAnalystFinding, type RawProviderEvent, type RawProviderSink, type RecordRunsOptions, type RedTeamCase, type RedTeamCategory, type RedTeamFinding, type RedTeamReport, type RedactOptions, type RedactionProfile, type RedactionReport, type ReferenceReplayCaseRun, type ReferenceReplayRun, type ReferenceReplaySplit, type ReflectionContext, type ReflectionProposal, type RegistryRunOpts, type ReleaseConfidenceInput, type ReleaseConfidenceIssue, type ReleaseConfidenceMetrics, type ReleaseConfidenceScorecard, type ReleaseTraceEvidence, type RepeatedActionOptions, type RequiredPairsForPairedPromotionOptions, type RequiredPairsForPairedPromotionResult, type RequirementCheck, type ResearchReport, type ResearchReportOptions, type Researcher, type RetryWrite, type Review, type ReviewFn, type ReviewInput, type ReviewMemoryEntry, type ReviewMemoryStore, type RiskDifferenceResult, type RolloutSearchLineage, type RouteMap, type RoutedField, type RubricDimension, type Run, type RunCampaignOptions, type RunCaptureReport, type RunCommandInput, type RunCommandResult, type RunCostProvenance, type RunFilter, RunIntegrityError, type RunIntegrityReport, type RunJudgeMetadata, type RunOutcome, type RunPaidCallInput, type RunRecord, type RunRecordBackend, RunRecordValidationError, type RunScore, type RunScoreWeights, type RunSearchCoordinates, type RunSplitTag, type RunStatus, type RunTaskFailure, type RunTerminalOutcome, type RunTokenUsage, type RunTraceRef, type RuntimeEventLike, type RuntimeTrajectoryEvidenceProjection, type RuntimeTrajectoryEvidenceSummary, type RuntimeTrajectoryHookEvent, type RuntimeTrajectoryRecord, type RuntimeTrajectoryRunRecord, SEMANTIC_CONCEPT_JUDGE_VERSION, type SampleSummary, type SampleSummaryOptions, type SandboxDriver, type SandboxSdkTransportOpts, type SatisfiedBy, type Scenario$1 as Scenario, type ScenarioCost, type ScenarioFile, type ScenarioResult, type ScorePreference, type ScoreRiskDifferenceResult, type Scorecard, type ScorecardCell, type ScorecardCellDiff, type ScorecardDiff, type ScorecardEntry, type ScorecardLogLine, type SearchSpanResult, type SearchTraceResult, type SelfImproveBudget, type SelfImproveMethodOptions, type SelfImproveMethodProvenance, type SelfImproveMethodResult, type SelfImproveOptions, type SelfImproveProgressEvent, type SelfImproveProposerOptions, type SelfImproveProposerResult, type SelfImproveResult, type SemanticConceptJudgeInput, type SemanticConceptJudgeOptions, type SemanticConceptJudgeResult, type SequentialDecision, type SeriesConvergenceOptions, type SeriesConvergenceResult, type SeriesDistribution, ServedCrossFamilyError, type ServedModelCheck, type ServedModelPolicy, type ServedModelVerdict, type Severity, type ShareSafetyStatus, type ShareSafetyVerdict, type SignTestAlternative, type SingleBackendDivergence, type SingleBackendReport, type Span, type SpanFilter, type SpanHandle, type SpanMatchRecord, type SpanPredicate, type SpanRecordsToTraceAnalysisStoreOptions, type SplitCoverage, type SteeringBundle, type SteeringChange, type SteeringOptimizationResult, type SteeringOptimizationRow, type StopDecision, type StrategyChecker, type StreamingDetector, type SummaryTable, type SummaryTableOptions, type SummaryTableRow, type SynthesisTarget, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, type TaskGold, type TaskHeadroom, type ToolCallEventLike, type ToolSpan, type ToolStats, type ToolUseMetrics, type ToolUseOptions, type TraceAnalysisEngine, type TraceAnalysisEngineResult, type TraceAnalysisStore, type TraceAnalystByteBudgets, type TraceAnalystDefinition, type TraceAnalystFilters, type TraceAnalystSpan, type TraceAnalystSpanKind, type TraceAnalystSpanStatus, type TraceAnalystTraceSummary, type TraceContract, type TraceContractSpec, TraceEmitter, type TraceEmitterOptions, type TraceEvent, type TraceInsightReadiness, type TraceInsightSuite, type TraceStore, type TrainSelectionTestOptions, type TrainSelectionTestPartition, type Trajectory, type TrajectoryStep, type TreatmentGate, type TreatmentGateInput, type TreatmentGateOptions, type TrialTrace, type Turn, type TurnResult, UNKNOWN_MODEL, type UserQuestion, VERIFICATION_STRATEGIES, VERIFICATION_STRATEGY_SOURCES, ValidationError, type VerbosityBiasResult, type Verdict, type VerdictCacheStore, type VerdictCertification, type Verification, type VerificationReport, type VerificationStrategyProfile, type VerificationStrategySource, type VerifyFn, type VerifyOptions, type ViewSpansResult, type ViewTraceOversized, type ViewTraceResult, type ViteDeployRunnerInput, type WeightedCompositeInput, type WeightedCompositeResult, type WelchTestResult, type WelchTestStatus, type WilcoxonSignedRankResult, type WranglerDeployRunnerInput, addCalibrationVerdicts, agentProfileCellKey, agentProfileHash, agentProfileId, analystFindingDigest, analystRunDigest, analystRunToFeedbackTrajectory, analystRunToReviewRequests, analyzeRuns, analyzeSeries, analyzeTraces, argHash, assertCrossFamily, assertCrossFamilyServed, assertJudgeMayGate, assertRealBackend, assertRunCaptured, assertServedModel, assertServedModels, assessShareSafety, assignFeedbackSplit, attest, benjaminiHochberg, bonferroni, bootstrapCi, budgetBreachView, buildAgentProfileCell, buildDefaultAnalystRegistry, buildEquivalenceRecord, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, calibrateJudge, calibrateJudgeContinuous, captureFetchToRawSink, certificationEvidenceDigest, checkServedModel, classifyFailure, classifyFailureReason, comparePairedArms, completionVerdict, computeFindingId, computeToolUseMetrics, confidenceInterval, continuousAgreement, controlRunToFeedbackTrajectory, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, createBoundedTraceAnalysisStore, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createFeedbackTrajectory, createLlmCorrectnessChecker, createTraceAnalyst, decidePairedPromotion, defineAgentEval, defineEquivalenceCheck, describeTraceInsightScope, diffFindings, domainEvidencePattern, dominates, eProcess, equivalenceVerdict, evaluateInterimReleaseConfidence, evaluateReleaseConfidence, expandProfileAxes, exportRunAsOtlp, extractProducedState, extractUsage, extractUsageFromSse, failureBlame, failureClusterView, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToOptimizerRow, gainHistogram, gradeSemanticStatus, groupRunsByAgentProfileCell, harnessAxisOf, hashJson, holm, inferDomainKeywords, interRaterReliability, iqr, isBinaryOutcomeVector, isJudgeSpan, isLlmSpan, isRunRecord, isToolSpan, judgeAgreementView, judgeFamily, judgeGateDecision, judgeSpans, llmJudge, makeFinding, makeProposalFinding, manifestContentDigest, mannWhitneyU, mcnemar, mcnemarPower, mcnemarRequiredN, measureJudgeAgreement, mintRolloutRows, modelHasSnapshot, mulberry32, otlpTextToTraceAnalysisStore, pairArms, pairRunRecords, pairedBinaryScale, pairedBootstrap, pairedDeltaTieFraction, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedRiskDifferenceExact, pairedRiskDifferenceScore, pairedSignTest, pairedTTest, paretoChart, paretoFrontier, parseRunRecordSafe, partialCredit, passAtK, pearsonR, index_d_exports as profile, quotaExhaustedUntil, ranks, redTeamDataset, redTeamReport, redact, redactForShare, redactText, registerCalibrationSet, renderPreferenceMemoryMarkdown, requiredPairedSampleSize, requiredSampleSize, runCampaign, runCanaries, runCounterfactual, runEquivalenceCheck, runEvalCampaign, runSemanticConceptJudge, runTaskScore, runsForScenario, scoreRedTeamOutput, scoreTraceInsightReadiness, searchCellRunId, selfImprove, servedModelAcceptable, shareAllowed, snapshotFromJudgeAgreement, spanRecordsToTraceAnalysisStore, spearmanR, summarizeBackendIntegrity, summarizeNumberSeries, summarizePreferenceMemory, summaryTable, toAgentProfileJson, tokenizeDomainWords, toolSpansToTraceAnalysisStore, toolWasteView, transientDispatchFailure, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyAttestation, verifyCompletion, weightedComposite, weightedMean, welchsTTest, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit };
+export { AGENT_PROFILE_KINDS, ATTESTATION_ALGORITHM, type ActionExecutionPolicy, type ActionPolicyDecision, type ActionableSideInfo, type ActiveLearningOptions, type AgentEvalAgent, AgentEvalError, type AgentEvalErrorCode, type AgentEvalEvaluateOptions, type AgentEvalImproveOptions, type AgentProfile, type AgentProfileCell, type AgentProfileCellInput, type AgentProfileDimensionValue, type AgentProfileJson, type AgentProfileSourceInput, type Analyst, type AnalystContext, type AnalystFeedbackTrajectoryOptions, type AnalystFinding, type AnalystFindingDigest, AnalystRegistry, type AnalystRegistryOptions, type AnalystReviewDecision, type AnalystReviewRequest, type AnalystRunDigest, type AnalystRunEvent, type AnalystRunInputs, type AnalystRunResult, type AnalystRunSummary, type AnalystSeverity, type AnalystUsageReceipt, type AnalyzeRunsOptions, type AnalyzeTracesResult, type ArgumentCheck, type Artifact, type ArtifactCheck, type ArtifactEventLike, type ArtifactResult, type AssertCapabilityHeadroomOptions, type AssertCrossFamilyOptions, type AssertCrossFamilyServedOptions, type AssertServedModelOptions, type AssertSingleBackendOptions, type AttestationProvenance, type AttestationVerification, type AttestedReport, BOOTSTRAP_GATE_MIN_N, type BackendDescriptor, BackendIntegrityError, type BackendIntegrityReport, type BenchmarkEvaluation, type BlendWeights, type BootstrapOptions, type BootstrapResult, type BoundedProcessInput, type BoundedProcessResult, type BudgetLedgerEntry, type BudgetPolicy, type BudgetSpec, CODING_HARNESSES, type CalibrationExample, type CalibrationMerge, type CalibrationResult, type CampaignCellFailureReceipt, type CampaignCellRetryPolicy, type CampaignFactoryParams, type CampaignIntegrityPolicy, type CampaignResult, type CampaignRunContext, type CampaignRunOutcome, type CampaignScenario, type CampaignVariant, type CanaryAlert, type CanaryEvaluation, type CanaryKind, type CanaryLeak, type CanaryOptions, type CanaryReport, type CandidateScore, type CapabilityHeadroomOptions, type CapabilityHeadroomResult, type CaptureErrorHandler, type CaptureWrite, type CellVerdict, type ChannelRollup, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTraceEngineOptions, type ChatTransport, type CheckResult, type CheckerIdentity, type CheckerOutcome, type CliffsMagnitude, type CollectedArtifacts, type CommandRunner, type ComparePairedArmsOptions, type CompletionCriterion, type CompletionRequirement, type CompletionVerdict, type ConceptFinding, type ConceptSpec, ConfigError, type ContinuousAgreement, type ContinuousAgreementOptions, type ContinuousCalibrationResult, type ContractCheckResult, type ContractLintFinding, type ContractRule, type ContractRuleExecution, type ContractRun, type ContractSpan, type ContractStatus, type ContractVerdict, type ControlActionOutcome, type ControlBudget, type ControlContext, type ControlDecision, type ControlEvalResult, type ControlRunResult, type ControlRuntimeConfig, type ControlRuntimeError, type ControlStep, type CorpusAgreementOptions, type CorpusAgreementPerDimension, type CorpusAgreementReport, type CorpusScoreRecord, type CorrectnessChecker, type CostChannel, type CostEntry, CostLedger, type CostLedgerFilter, type CostLedgerHandle, type CostLedgerOptions, type CostLedgerPersistence, type CostLedgerSummary, type CostProvenance, type CostReceipt, type CostReceiptInput, type CostResult, type CostSummary, type CostUsage, type CounterfactualContext, type CounterfactualMutation, type CounterfactualResult, type CounterfactualRunner, type CreateChatClientOpts, type CreateTraceAnalystOptions, CrossFamilyError, type CustomTokenPricing, type CustomTransportOpts, DEFAULT_JUDGE_GATE_POLICY, DEFAULT_RED_TEAM_CORPUS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, type DataAcquisitionPlan, type DatasetManifest, type DatasetOverview, type DatasetScenario, type DatasetSplit, type DefaultAnalystRegistryOptions, type DefaultVerdict, type DefineAgentEvalOptions, type DefinedAgentEval, type DeployGateLayerInput, type DeployRunResult, type DeployRunner, type DetectorEvent, type DetectorSignal, type DiffPolicy, type DiffScorecardOptions, type DimensionJudgeScore, type DirEntry, type DiscoverPersonasOptions, type DiscoveredPersona, type DriverState, type DspyRlmTraceEngineOptions, type EProcess, type EProcessOptions, type EProcessState, type EProcessStep, type EnsembleAggregate, type EnsembleJudgeOptions, type EquivalenceArm, type EquivalenceCheckDefinition, type EquivalenceCheckSpec, type EquivalenceChecker, type EquivalenceCheckerInput, type EquivalenceCheckerResult, type EquivalenceObligation, type EquivalenceObligationStatus, EquivalenceProtocolError, type EquivalenceRecord, type ErrorCluster, type ErrorCountPattern, type ErrorStreakOptions, type EvalCampaignOptions, type EvalCampaignResult, type EventFilter, type EvidenceRef, type ExactAnalystRunEvent, type ExactAnalystRunResult, type ExactCapableAnalyst, type ExactRegistryRunOpts, type ExactRiskDifferenceResult, type ExperimentPlan, type ExperimentRep, type ExperimentResult, type ExperimentStats, type ExtractOptions, type ExtractResult, type ExtractedUsage, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, type FailedRun, type FailureBlame, type FailureClass, type FailureClassification, type FailureContext, type FailureMode, type FailureReasonRule, type FailureRule, type FeedbackArtifactType, type FeedbackAttempt, type FeedbackLabel, type FeedbackLabelKind, type FeedbackLabelSource, type FeedbackOptimizerRow, type FeedbackOutcome, type FeedbackSplitPolicy, type FeedbackTask, type FeedbackTrajectory, type FeedbackTrajectoryFilter, type FeedbackTrajectoryStore, type FieldDestination, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, FileSystemTraceStore, type Finding, type FindingSubject, type FindingSubjectKind, type FindingToPolicyEditOptions, type FindingsDiff, FindingsStore, type GainDistributionBin, type GainDistributionFigureSpec, type GainDistributionOptions, type GateDecision, type GateEvidence, type GenericSpan, type GoldenItem, HARNESS_NATIVE_MODEL, type HarnessType, type HeadroomInput, HeldOutGate, type HeldOutGateConfig, type HeldOutGateDecision, type HeldOutGateRejectionCode, type HeldOutPartition, type HiddenCriteriaGrader, type HiddenGradeResult, type HiddenLeak, type ImprovementThresholds, type ImprovementVerdictResult, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, InMemoryTraceStore, type InsightReport, type IntentMatchInput, type IntentMatchOptions, type IntentMatchResult, type InterimReleaseConfidence, type InterimReleaseConfidenceInput, type JudgeAgreement, type JudgeCalibrationSet, type JudgeConfusion, JudgeError, type JudgeExampleAgreement, type JudgeFamily, type JudgeFn, type JudgeGateDecision, type JudgeGateInput, type JudgeGatePolicy, JudgeGateRefusedError, type JudgeGateStatus, type JudgeInput, type JudgeRetryOutcome, type JudgeRetryPolicy, type JudgeRubric, type JudgeRun, type JudgeRunMeta, type JudgeScore, type JudgeScoreInput, type JudgeScoresRecord, type JudgeSpan, type JudgeVerdict, type KeywordConceptSpec, type KeywordCoverageFinding, type KeywordCoverageOptions, type KeywordCoverageResult, type KnowledgeAcquisitionMode, type KnowledgeBundle, type KnowledgeFreshness, type KnowledgeImportance, type KnowledgeReadinessReport, type KnowledgeRequirement, type KnowledgeRequirementCategory, type KnowledgeSensitivity, type Layer, type LayerResult, type LayerStatus, type LeaderboardOptions, type LeaderboardRow, type LinearRegressionResult, type LlmCallMetadata, type LlmCallRequest, type LlmCallResult, type LlmChargeBounds, type LlmCorrectnessCheckerOpts, type LlmJudgeOptions, type LlmMessage, type LlmReviewerConfig, type LlmSpan, type LlmTokenLogprob, type LlmToolCall, type LlmToolChoice, type LlmToolDefinition, type LlmUsage, type MannWhitneyResult, type MatchedPair, type MatchedRunRecordPair, type MaximumCharge, type McNemarResult, type MintRolloutOptions, type MintRolloutResult, type MintedRolloutLine, type MockTransportOpts, type ModelEndpointCheck, type ModelEndpointRequest, type ModelSeats, ModelSubstitutionError, MultiLayerVerifier, type NoLeakOptions, NoopRawProviderSink, NotFoundError, OUTPUT_VALUE, type Objective, type OpenAiCompatibleTransportOpts, type Oracle, type OracleObservation, type OracleReport, type OracleResult, type OrderMode, type OtlpExport, OtlpFileTraceStore, type OtlpFlatLine, type OtlpSpan, type OwnerVerdict, type PaidCallResult, type PairArmsOptions, type PairArmsResult, type PairRunRecordsResult, type PairedArmRow, type PairedArmsComparison, type PairedBootstrapOptions, type PairedBootstrapResult, type PairedCorrectness, type PairedDecisionMethod, type PairedDecisionShape, type PairedDecisionStatistic, type PairedDeltaTestOptions, type PairedDeltaTestResult, type PairedEvalueOptions, type PairedEvalueSequence, type PairedEvalueStep, type PairedMcNemarEvidence, type PairedMetricDelta, type PairedPromotionAlternative, type PairedPromotionAlternativeCell, type PairedPromotionCallRates, type PairedPromotionDecision, type PairedPromotionDecisionOptions, type PairedPromotionPowerCall, type PairedPromotionPowerOptions, type PairedPromotionPowerResult, type PairedSignTestResult, type PairedTTestResult, type PairedValueLaw, type ParetoPoint, type ParetoResult, type PartitionHeldOutOptions, type PendingCostCall, type PendingCostCallView, type PersistedFinding, type PersonaConfig, type PolicyEdit, type PolicyEditAdmission, type PolicyEditAdmissionOptions, type PolicyEditAxis, type PolicyEditCandidateRecord, type PolicyEditChange, type PolicyEditExpectedGain, type PolicyEditGainDirection, type PolicyEditGainUnit, type PolicyEditInit, type PolicyEditRisk, type PolicyEditSchemaVersion, type PolicyEditSource, type PolicyEditTarget, type PolicyEditTargetSurface, type PreferenceMemoryEntry, type ProducedProposal, type ProducedState, type ProductBenchmarkExportOptions, type ProductBenchmarkExportResult, type ProductBenchmarkManifest, type ProductBenchmarkSingleRunExportOptions, type ProductBenchmarkSplit, type ProductBenchmarkValidationReport, type ProductClientConfig, type ProductScenario, type ProfileAxisSpec, type ProjectRuntimeTrajectoryEvidenceOptions, type PromptHandle, type ProportionInterval, type ProposalEventLike, type ProposalFinding, type ProposeFn, type ProposeInput, type ProposeOutput, type ProposeReviewConfig, type ProposeReviewControlAction, type ProposeReviewControlConfig, type ProposeReviewControlResult, type ProposeReviewControlState, type ProposeReviewReport, type ProposedSideEffect, type QueryTracesPage, REDACTION_VERSION, type RankTestMethod, type RankTestMethodRequest, type RankTestOptions, type RawAnalystFinding, type RawProviderEvent, type RawProviderSink, type RecordRunsOptions, type RedTeamCase, type RedTeamCategory, type RedTeamFinding, type RedTeamReport, type RedactOptions, type RedactionProfile, type RedactionReport, type ReferenceReplayCaseRun, type ReferenceReplayRun, type ReferenceReplaySplit, type ReflectionContext, type ReflectionProposal, type RegistryRunOpts, type ReleaseConfidenceInput, type ReleaseConfidenceIssue, type ReleaseConfidenceMetrics, type ReleaseConfidenceScorecard, type ReleaseTraceEvidence, type RepeatedActionOptions, type RequiredPairsForPairedPromotionOptions, type RequiredPairsForPairedPromotionResult, type RequirementCheck, type ResearchReport, type ResearchReportOptions, type Researcher, type RetryWrite, type Review, type ReviewFn, type ReviewInput, type ReviewMemoryEntry, type ReviewMemoryStore, type RiskDifferenceResult, type RolloutSearchLineage, type RouteMap, type RoutedField, type RubricDimension, type Run, type RunCampaignOptions, type RunCaptureReport, type RunCommandInput, type RunCommandResult, type RunCostProvenance, type RunFilter, RunIntegrityError, type RunIntegrityReport, type RunJudgeMetadata, type RunOutcome, type RunPaidCallInput, type RunRecord, type RunRecordBackend, RunRecordValidationError, type RunScore, type RunScoreWeights, type RunSearchCoordinates, type RunSplitTag, type RunStatus, type RunTaskFailure, type RunTerminalOutcome, type RunTokenUsage, type RunTraceRef, type RuntimeEventLike, type RuntimeTrajectoryEvidenceProjection, type RuntimeTrajectoryEvidenceSummary, type RuntimeTrajectoryHookEvent, type RuntimeTrajectoryRecord, type RuntimeTrajectoryRunRecord, SEMANTIC_CONCEPT_JUDGE_VERSION, type SampleSummary, type SampleSummaryOptions, type SandboxDriver, type SandboxSdkTransportOpts, type SatisfiedBy, type Scenario, type ScenarioCost, type ScenarioFile, type ScenarioResult, type ScorePreference, type ScoreRiskDifferenceResult, type Scorecard, type ScorecardCell, type ScorecardCellDiff, type ScorecardDiff, type ScorecardEntry, type ScorecardLogLine, type SearchSpanResult, type SearchTraceResult, type SelfImproveBudget, type SelfImproveMethodOptions, type SelfImproveMethodProvenance, type SelfImproveMethodResult, type SelfImproveOptions, type SelfImproveProgressEvent, type SelfImproveProposerOptions, type SelfImproveProposerResult, type SelfImproveResult, type SemanticConceptJudgeInput, type SemanticConceptJudgeOptions, type SemanticConceptJudgeResult, type SequentialDecision, type SeriesConvergenceOptions, type SeriesConvergenceResult, type SeriesDistribution, type ServedModelCheck, type ServedModelPolicy, type ServedModelVerdict, type Severity, type ShareSafetyStatus, type ShareSafetyVerdict, type SignTestAlternative, type SingleBackendDivergence, type SingleBackendReport, type Span, type SpanFilter, type SpanHandle, type SpanMatchRecord, type SpanPredicate, type SpanRecordsToTraceAnalysisStoreOptions, type SplitCoverage, type SteeringBundle, type SteeringChange, type SteeringOptimizationResult, type SteeringOptimizationRow, type StopDecision, type StrategyChecker, type StreamingDetector, type SummaryTable, type SummaryTableOptions, type SummaryTableRow, type SynthesisTarget, type TaskGold, type TaskHeadroom, type ToolCallEventLike, type ToolSpan, type ToolStats, type ToolUseMetrics, type ToolUseOptions, type TraceAnalysisEngine, type TraceAnalysisEngineResult, type TraceAnalysisStore, type TraceAnalystByteBudgets, type TraceAnalystDefinition, type TraceAnalystFilters, type TraceAnalystSpan, type TraceAnalystSpanKind, type TraceAnalystSpanStatus, type TraceAnalystTraceSummary, type TraceContract, type TraceContractSpec, TraceEmitter, type TraceEmitterOptions, type TraceEvent, type TraceInsightReadiness, type TraceInsightSuite, type TraceStore, type TrainSelectionTestOptions, type TrainSelectionTestPartition, type Trajectory, type TrajectoryStep, type TreatmentGate, type TreatmentGateInput, type TreatmentGateOptions, type TrialTrace, type Turn, type TurnResult, UNKNOWN_MODEL, type UserQuestion, VERIFICATION_STRATEGIES, VERIFICATION_STRATEGY_SOURCES, ValidationError, type VerbosityBiasResult, type Verdict, type VerdictCacheStore, type VerdictCertification, type Verification, type VerificationReport, type VerificationStrategyProfile, type VerificationStrategySource, type VerifyFn, type VerifyOptions, type ViewSpansResult, type ViewTraceOversized, type ViewTraceResult, type ViteDeployRunnerInput, type WeightedCompositeInput, type WeightedCompositeResult, type WelchTestResult, type WelchTestStatus, type WilcoxonSignedRankResult, type WranglerDeployRunnerInput, addCalibrationVerdicts, agentProfileCellKey, agentProfileHash, agentProfileId, analystRunDigest, analystRunToFeedbackTrajectory, analystRunToReviewRequests, analyzeRuns, analyzeSeries, analyzeTraces, argHash, assertCrossFamily, assertCrossFamilyServed, assertJudgeMayGate, assertRealBackend, assertRunCaptured, assertServedModel, assertServedModels, assessShareSafety, assignFeedbackSplit, attest, benjaminiHochberg, bonferroni, bootstrapCi, budgetBreachView, buildAgentProfileCell, buildDefaultAnalystRegistry, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, calibrateJudge, calibrateJudgeContinuous, captureFetchToRawSink, certificationEvidenceDigest, checkServedModel, classifyFailure, classifyFailureReason, comparePairedArms, completionVerdict, computeFindingId, computeToolUseMetrics, confidenceInterval, continuousAgreement, controlRunToFeedbackTrajectory, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, createBoundedTraceAnalysisStore, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createFeedbackTrajectory, createLlmCorrectnessChecker, createTraceAnalyst, decidePairedPromotion, defineAgentEval, defineEquivalenceCheck, describeTraceInsightScope, diffFindings, domainEvidencePattern, dominates, eProcess, equivalenceVerdict, evaluateInterimReleaseConfidence, evaluateReleaseConfidence, expandProfileAxes, exportRunAsOtlp, extractProducedState, extractUsage, extractUsageFromSse, failureBlame, failureClusterView, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToOptimizerRow, gainHistogram, gradeSemanticStatus, groupRunsByAgentProfileCell, harnessAxisOf, hashJson, holm, inferDomainKeywords, interRaterReliability, iqr, isJudgeSpan, isLlmSpan, isRunRecord, isToolSpan, judgeAgreementView, judgeFamily, judgeGateDecision, judgeSpans, llmJudge, makeFinding, makeProposalFinding, manifestContentDigest, mannWhitneyU, mcnemar, mcnemarPower, mcnemarRequiredN, measureJudgeAgreement, mintRolloutRows, modelHasSnapshot, mulberry32, otlpTextToTraceAnalysisStore, pairArms, pairRunRecords, pairedBootstrap, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedRiskDifferenceExact, pairedRiskDifferenceScore, pairedSignTest, pairedTTest, paretoFrontier, parseRunRecordSafe, partialCredit, passAtK, pearsonR, index_d_exports as profile, ranks, redTeamDataset, redTeamReport, redact, redactForShare, redactText, registerCalibrationSet, renderPreferenceMemoryMarkdown, requiredPairedSampleSize, requiredSampleSize, runCampaign, runCanaries, runEquivalenceCheck, runEvalCampaign, runSemanticConceptJudge, runTaskScore, runsForScenario, scoreRedTeamOutput, scoreTraceInsightReadiness, selfImprove, shareAllowed, snapshotFromJudgeAgreement, spanRecordsToTraceAnalysisStore, spearmanR, summarizeBackendIntegrity, summarizeNumberSeries, summarizePreferenceMemory, summaryTable, toAgentProfileJson, tokenizeDomainWords, toolSpansToTraceAnalysisStore, toolWasteView, transientDispatchFailure, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyAttestation, verifyCompletion, weightedComposite, weightedMean, welchsTTest, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit };
 //# sourceMappingURL=index.d.ts.map