@tangle-network/agent-eval 0.123.5 → 0.123.7

package/dist/contract/index.js

@@ -37,7 +37,7 @@ import {
   runReferenceEquivalenceJudge,
   surfaceContentHash,
   surfaceHash
-} from "../chunk-SUN7QLPB.js";
+} from "../chunk-KKPPFIDS.js";
 import {
   campaignSplitDigest,
   createRunCostLedger,

package/dist/index.d.ts

@@ -6144,6 +6144,120 @@ declare function formatDriverReport(results: DriverResult[]): string;
 /** Print a compact summary to console */
 declare function printDriverSummary(results: DriverResult[]): void;
 
+/**
+ * Rollout export — `tangle.rollout.v1` training rows minted from the
+ * records the substrate ALREADY keeps. There is no separate rollout
+ * store: a rollout is the JOIN of a RunRecord (identity, provenance,
+ * cost, outcome) with its trace (spans share `runId`), optionally
+ * annotated with feedback labels. This module performs that join and
+ * serializes it to fine-tune-ready formats.
+ *
+ * Composition, not duplication:
+ *   - identity/provenance     → `RunRecord` (candidateId, splitTag, agentProfile, hashes)
+ *   - step structure          → `buildTrajectory` over the shared TraceStore
+ *   - preference-pair export  → `feedbackTrajectoryToOptimizerRow` (feedback-trajectory.ts)
+ *   - PRM / reward-model      → `reward-model-export.ts`
+ *   This module adds only the joined row + SFT/reward serialization,
+ *   which none of the above produce.
+ *
+ * Anti-Goodhart invariant: a run whose `outcome.realness.gated` is true
+ * is never exported with a positive reward — the gate travels into the
+ * training data, so a fine-tune cannot learn from gamed successes.
+ */
+
+declare const ROLLOUT_FORMAT: "tangle.rollout.v1";
+/** Compact, serialization-safe projection of one span for training rows. */
+interface RolloutStep {
+    kind: Span['kind'];
+    name: string;
+    /** llm: last-message summary · tool: stringified args. Scrubbed. */
+    input?: string;
+    /** llm: output text · tool: stringified result. Scrubbed. */
+    output?: string;
+    status?: 'ok' | 'error';
+    durationMs?: number;
+}
+interface RolloutRow {
+    format: typeof ROLLOUT_FORMAT;
+    runId: string;
+    experimentId: string;
+    candidateId: string;
+    scenarioId?: string;
+    splitTag: RunRecord['splitTag'];
+    model: string;
+    /** Canonical profile-cell identity when the record carries one. */
+    agentProfileCellId?: string;
+    /** holdoutScore ?? searchScore, forced to 0 when realness-gated. */
+    reward: number;
+    /** True when `outcome.realness.gated` — excluded from SFT positives. */
+    realnessGated: boolean;
+    costUsd: number;
+    totalTokens: number;
+    steps: RolloutStep[];
+    /** Full message history of the final llm span — the SFT conversation. */
+    conversation: Message[];
+}
+/** Redactor applied to every exported string (secrets, PII). Identity by default. */
+type RolloutScrubber = (text: string) => string;
+interface MintRolloutOptions {
+    scrub?: RolloutScrubber;
+    /** Cap steps per row (longest runs first drop middle steps). Default: no cap. */
+    maxSteps?: number;
+}
+interface MintRolloutResult {
+    rows: RolloutRow[];
+    /** runIds that had a RunRecord but no spans — surfaced, never silently dropped. */
+    missingTraces: string[];
+}
+declare function rolloutReward(record: RunRecord): {
+    reward: number;
+    gated: boolean;
+};
+/**
+ * Join RunRecords with their traces into rollout rows. Records without
+ * spans are reported in `missingTraces` — a capture gap is a finding,
+ * not a silent omission.
+ */
+declare function mintRolloutRows(records: RunRecord[], store: TraceStore, options?: MintRolloutOptions): Promise<MintRolloutResult>;
+interface SftExportOptions {
+    /** Export only rows with reward ≥ this (default 1 = clean successes only). */
+    minReward?: number;
+}
+interface SftRow {
+    messages: Message[];
+    metadata: {
+        runId: string;
+        candidateId: string;
+        scenarioId?: string;
+        reward: number;
+    };
+}
+/**
+ * Supervised fine-tune rows: the completed conversation of each
+ * qualifying rollout. Realness-gated rows never qualify (reward forced
+ * to 0 at mint time).
+ */
+declare function toSftRows(rows: RolloutRow[], options?: SftExportOptions): SftRow[];
+interface RewardRow {
+    prompt: string;
+    steps: RolloutStep[];
+    reward: number;
+    metadata: {
+        runId: string;
+        candidateId: string;
+        scenarioId?: string;
+        splitTag: RunRecord['splitTag'];
+    };
+}
+/**
+ * Reward-labeled rows (RFT / verifiers-style): every rollout, success or
+ * failure, with its scalar reward. Failures are signal here — only the
+ * realness-gate zeroing is applied, never filtering.
+ */
+declare function toRewardRows(rows: RolloutRow[]): RewardRow[];
+/** One JSON object per line — the interchange format for every export. */
+declare function toJsonl(rows: ReadonlyArray<unknown>): string;
+
 interface RunEvidenceMetadata {
     experimentId: string;
     candidateId: string;
@@ -16814,4 +16928,4 @@ type CachedJudge<TArtifact, TScenario extends Scenario = Scenario> = JudgeConfig
  */
 declare function cachedJudge<TArtifact, TScenario extends Scenario = Scenario>(judge: JudgeConfig<TArtifact, TScenario>, store: VerdictCacheStore, options: CachedJudgeOptions): CachedJudge<TArtifact, TScenario>;
 
-export { AGENT_PROFILE_KINDS, ATTESTATION_ALGORITHM, type ActionExecutionPolicy, type ActionPolicyDecision, type ActionableSideInfo, type ActiveLearningOptions, type AdapterRun, AgentDriver, type AgentDriverConfig, AgentEvalError, type AgentEvalErrorCode, type AgentInterfaceProfileLike, type AgentProfileCell, type AgentProfileCellInput, type AgentProfileCellSchemaVersion, AgentProfileCellValidationError, type AgentProfileDimensionValue, type AgentProfileHarness, type AgentProfileJson, type AgentProfileJsonObject, type AgentProfileKind, type AgentProfileRuntimeReceipt, type AgentProfileSource, type AgentProfileSourceInput, type AgreementResult, type AlignmentOp, type Analyst, type AnalystContext, type AnalystCost, type AnalystFinding, type AnalystHooks, type AnalystInputKind, AnalystRegistry, type AnalystRegistryOptions, type AnalystRequirements, type AnalystRunEvent, type AnalystRunInputs, type AnalystRunResult, type AnalystRunSummary, type AnalystSeverity, type AnalystUsageReceipt, type AnalyzeTracesInput, type AnalyzeTracesOptions, type AnalyzeTracesResult, type AnalyzeTracesTurnSnapshot, type AntiSlopConfig, type AntiSlopIssue, type AntiSlopReport, type Artifact$1 as Artifact, type ArtifactCheck, type Artifact as ArtifactCheckArtifact, type ArtifactEventLike, type ArtifactResult, type ArtifactValidator, type AsiSeverity, type AssertCapabilityHeadroomOptions, type AssertCrossFamilyOptions, type AssertSingleBackendOptions, type AttestationProvenance, type AttestationVerification, type AttestedReport, type AutoPrClient, AxGepaSteeringOptimizer, type AxSteeringOptimizerConfig, BENCHMARK_SPLIT_SEED, type BackendDescriptor, BackendIntegrityError, type BackendIntegrityReport, type BaselineOptions, type BaselineReport, BehaviorAssertion, type BehavioralMetrics, type BehavioralTokenSequence, type BenchmarkAdapter, type BenchmarkDatasetItem, type BenchmarkEvaluation, type BenchmarkFamily, type BenchmarkReport$1 as BenchmarkReport, type BenchmarkResponder, BenchmarkRunner, type BenchmarkRunnerConfig, type BenchmarkScenario, type BenchmarkSource, type BenchmarkTaskKind, type BisectOptions, type BisectResult, type BisectStep, type BlendWeights, type BootstrapOptions, type BootstrapResult, BudgetBreachError, BudgetGuard, type BudgetLedgerEntry, type BudgetPolicy, type BudgetSpec, type BuildAgreementJudgeOptions, CODING_HARNESSES, type CachedJudge, type CachedJudgeOptions, type CalibrationResult, CallExpectation, CallbackResearcher, type CallbackResearcherOptions, type CampaignFactoryParams, type CampaignIntegrityPolicy, type CampaignRunContext, type CampaignRunOutcome, type CampaignRunner, type CampaignScenario, type CampaignVariant, type CanaryAlert, type CanaryKind, type CanaryLeak, type CanaryOptions, type CanaryReport, type CanarySeverity, type CandidateComparison, type CandidateScenario, type CandidateScore, type CanonicalRawAnalystFinding, type CapabilityHeadroomOptions, type CapabilityHeadroomResult, type CaptureFetchContext, type CaptureFetchOptions, CaptureIntegrityError, type CausalAttributionReport, type CellVerdict, type ChannelRollup, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTransport, type CheckResult, type CliBridgeTransportOpts, type CliffsMagnitude, type ClusterBootstrapInterval, type ClusterSignFlipAlternative, type ClusterSignFlipResult, type ClusteredBinaryCluster, type ClusteredMatchedPair, type ClusteredPairedBinaryOptions, type ClusteredPairedBinaryResult, type ClusteredPairedBinaryStatistics, type CollectedArtifacts, type CommandRunner, type CompareLabels, type ComparePairedArmsOptions, type CompletionCriterion, type CompletionRequirement, type CompletionVerdict, type ConceptComplexity, type ConceptFinding, type ConceptSpec, type ConceptWeightStrategy, ConfigError, type ContinuityCheck, type ContinuityCheckResult, type ContinuityReport, type ContinuitySnapshotPair, type ContinuousAgreement, type ContinuousAgreementOptions, type ContinuousCalibrationResult, type ContractCheckResult, type ContractJudgeOptions, type ContractMetric, type ContractReport, type ContractRule, type ContractRuleKind, type ContractSpan, type ContractVerdict, type ContractViolation, type ControlActionFailureMode, type ControlActionOutcome, type ControlBudget, type ControlContext, type ControlDecision, type ControlEvalResult, type ControlRunResult, type ControlRunToRunRecordOptions, type ControlRuntimeConfig, type ControlRuntimeError, type ControlSeverity, type ControlStep, type ControlStopPolicies, ConvergenceTracker, type CorpusAgreementOptions, type CorpusAgreementPerDimension, type CorpusAgreementReport, type CorpusScoreRecord, type CorrectnessChecker, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, type CostChannel, type CostEntry, CostLedger, type CostLedgerEntry, type CostLedgerFilter, type CostLedgerHandle, type CostLedgerOptions, type CostLedgerPersistence, CostLedgerPersistenceError, type CostLedgerSummary, type CostReceipt, CostReceiptCaptureError, type CostReceiptInput, type CostReport, CostReservationExceededError, type CostResult, type CostSummary, CostTracker, type CostUsage, type CounterfactualContext, type CounterfactualMutation, type CounterfactualResult, type CounterfactualRunner, type CreateAnalystAiConfig, type CreateChatClientOpts, type CreateDefaultReviewerOptions, type CreateExperimentInput, type CreateSandboxPoolOpts, type CreateTraceAnalystKindOpts, CrossFamilyError, type CrossTraceDiff, type CrossTraceDiffOptions, type CustomTokenPricing, DEFAULT_AGENT_SLOS, DEFAULT_COMPLEXITY_WEIGHTS, DEFAULT_RULES as DEFAULT_FAILURE_RULES, DEFAULT_FINDERS, DEFAULT_HARNESS_OBJECTIVES, DEFAULT_MUTATION_PRIMITIVES, DEFAULT_MUTATORS, DEFAULT_PR_REVIEW_SCORE_WEIGHTS, DEFAULT_REDACTION_RULES, DEFAULT_RED_TEAM_CORPUS, DEFAULT_RUN_SCORE_WEIGHTS, DEFAULT_SEVERITY_WEIGHTS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, type DataAcquisitionPlan, Dataset, type DatasetDifficulty, type DatasetManifest, type DatasetOverview, type DatasetProvenance, type DatasetScenario, type DatasetSplit, type DecideNextUserTurnOpts, type DefaultAnalystRegistryOptions, type DefaultVerdict, type DeployFamily, type DeployGateLayerInput, type DeployRunResult, type DeployRunner, type DescriptionLengthCandidate, type DescriptionLengthConfig, type DescriptionLengthDecision, type DescriptionLengthEvidence, DescriptionLengthGate, type DescriptionLengthRejectionCode, type DetectorEvent, type DetectorSeverity, type DetectorSignal, type DiffPolicy, type DiffScorecardOptions, type DirEntry, type DirectProviderTransportOpts, type Direction, type DiscoverPersonasOptions, type DiscoveredPersona, DockerSandboxDriver, type DriverResult, type DriverState, DualAgentBench, type DualAgentBenchConfig, type DualAgentReport, type DualAgentRound, type DualAgentScenario, type DualAgentScenarioResult, type EProcess, type EProcessOptions, type EProcessState, type EProcessStep, ERROR_COUNT_PATTERNS, type EnsembleAggregate, type EnsembleJudgeOptions, type ErrorCluster, type ErrorCountPattern, type ErrorStreakOptions, type EvalCampaignOptions, type EvalCampaignResult, type EvalResult, type EvalToolDef, EvalTraceStore, type EventFilter, type EventKind, type EvidenceRef, type EvolutionRound, type ExecutorConfig, type Expectation, type Experiment, type ExperimentPlan, type ExperimentProvenance, type ExperimentRep, type ExperimentResult, type ExperimentStats, type ExperimentStore, ExperimentTracker, type ExperimentTrackerOptions, type ExperimentVerdict, type ExportableSpan, type ExportedRewardModel, type ExtractOptions, type ExtractResult, type ExtractUsageFromSseOptions, type ExtractedUsage, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, type FactorContribution, type FactorialCell, type FailedRun, type FailureClass, type FailureClassification, type FailureContext, type FailureMode, type FailureRule, type FeedbackArtifactType, type FeedbackAttempt, type FeedbackLabel, type FeedbackLabelKind, type FeedbackLabelSource, type FeedbackOptimizerRow, type FeedbackOutcome, type FeedbackPattern, type FeedbackReplayAdapter, type FeedbackReplayResult, type FeedbackSeverity, type FeedbackSplitPolicy, type FeedbackTask, type FeedbackTrajectory, type FeedbackTrajectoryFilter, type FeedbackTrajectoryStore, type FieldAgreementSpec, type FieldDestination, type FileChange, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, type FileSystemRawProviderSinkOptions, FileSystemTraceStore, type FileSystemTraceStoreOptions, type Finding, type FindingSubject, type FindingSubjectKind, type FindingToPolicyEditOptions, type FindingsDiff, FindingsStore, type FlattenOtlpOptions, type FlowAction, type FlowLayerEnv, type FlowLayerFactoryInput, type FlowRunner, type FlowRunnerStepResult, type FlowSpec, type FlowStep, type GainDistributionBin, type GainDistributionFigureSpec, type GainDistributionOptions, type GateDecision$1 as GateDecision, type GateEvidence, type GenericSpan, type GhCliClientOptions, type GoldScenario, type GoldSplit, type GoldenItem, type GoldenSeverity, type GoldenSpec, HARNESS_NATIVE_MODEL, type HarnessAdapter, type HarnessConfig, type HarnessExperimentConfig, type HarnessExperimentResult, type HarnessIntervention, type HarnessRunRequest, type HarnessRunResult, type HarnessScenario, type HarnessSelection, type HarnessVariant, type HarnessVariantReport, type HeadroomClass, type HeadroomInput, HeldOutGate, type HeldOutGateConfig, type HeldOutGateRejectionCode, type HeldOutPartition, type HiddenCriteriaGrader, type HiddenGradeResult, type HiddenLeak, HoldoutAuditor, HoldoutLockedError, type HttpGithubClientOptions, type HypothesisManifest, type HypothesisResult, IMPROVEMENT_KIND_SPEC, INPUT_VALUE, INTENT_MATCH_JUDGE_VERSION, type ImageData, type ImprovementThresholds, type ImprovementVerdictResult, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, type InMemoryRawProviderSinkOptions, InMemoryTraceStore, InMemoryWorkspaceInspector, type InferenceScorer, type InspectorContext, type IntentMatchInput, type IntentMatchOptions, type IntentMatchResult, type InteractionContribution, type InterimReleaseConfidence, type InterimReleaseConfidenceInput, type JudgeConfig$1 as JudgeConfig, JudgeError, type JudgeFamily, type JudgeFleetOptions, type JudgeFn, type JudgeInput, JudgeParseError, type JudgeReplayGateArgs, type JudgeReplayResult, type JudgeRetryOutcome, type JudgeRetryPolicy, type JudgeRubric, JudgeRunner, type JudgeScore$1 as JudgeScore, type JudgeScoreInput, type JudgeScoresRecord, type JudgeSpan, type JudgeVerdict, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, type KeywordConceptSpec, type KeywordCoverageFinding, type KeywordCoverageOptions, type KeywordCoverageResult, type KnowledgeAcquisitionMode, type KnowledgeBundle, type KnowledgeFallbackPolicy, type KnowledgeFreshness, type KnowledgeImportance, type KnowledgeReadinessReport, type KnowledgeRecommendedAction, type KnowledgeRequirement, type KnowledgeRequirementCategory, type KnowledgeResponsibleSurface, type KnowledgeSensitivity, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_CACHE_WRITE_TOKENS, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, LLM_CONTEXT_TOKENS, LLM_COST_ATTR_KEYS, LLM_COST_USD, LLM_INPUT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_MODEL_ATTR_KEYS, LLM_MODEL_NAME, LLM_OUTPUT_TOKENS, LLM_OUTPUT_TOKEN_ATTR_KEYS, LLM_REASONING_TOKENS, LLM_REASONING_TOKEN_ATTR_KEYS, type LangfuseEnvelope, type LangfuseGeneration, type LangfuseScore, type Layer, type LayerResult, type LayerStatus, type LeaderboardOptions, type LeaderboardRow, type LiveProofArtifact, type LiveProofConfig, type LiveProofContext, type LiveProofResult, LlmCallError, type LlmCallMetadata, type LlmCallRequest, type LlmCallResult, LlmClient, type LlmClientOptions, type LlmCorrectnessCheckerOpts, type LlmJsonCall, type LlmJudgeDimension, type LlmJudgeOptions, type LlmMessage, LlmResponseError, type LlmReviewerConfig, LlmRouteAssertionError, type LlmRouteRequirements, type LlmSpan, type LlmSpanOtlpInput, type LlmUsage, LockedJsonlAppender, MODEL_PRICING, type MakeEvalToolsConfig, type MatchResult, type MatchedPair, type MatcherResult, type MaximumCharge, type McNemarResult, type MeasurementPolicy, type MergeOptions, type Message, type MetricSamples, type MetricVerdict, MetricsCollector, type MockTransportOpts, type ModelCostRollup, type ModelPreflight, type ModelSeats, ModelsUnreachableError, type MuffledFinder, type MuffledFinding, MultiLayerVerifier, type MultiToolchainLayerConfig, type Mutator, Mutex, type NoLeakOptions, type NoProgressOptions, NoopRawProviderSink, NoopResearcher, NotFoundError, OPENINFERENCE_SPAN_KIND, OTEL_AGENT_EVAL_SCOPE, OUTPUT_VALUE, type Objective, type Oracle, type OracleObservation, type OracleReport, type OracleResult, type OrthogonalityInput, type OrthogonalityResult, type OtelExportConfig, type OtelExporter, type OtelPipelineHandle, type OtelPipelineOptions, type OtlpExport, OtlpFileTraceStore, type OtlpFileTraceStoreOptions, type OtlpFlatLine, type OtlpResourceSpans, type OtlpSpan, type OtlpToRunRecordsOptions, type OtlpTraceRunRecord, POLICY_EDIT_AXES, POLICY_EDIT_CANDIDATE_RECORD_SCHEMA, POLICY_EDIT_TARGET_SURFACES, type PaidCallResult, type PairArmsOptions, type PairArmsResult, type PairedArmRow, type PairedArmsComparison, type PairedBootstrapOptions, type PairedBootstrapResult, type PairedCorrectness, type PairedEvalueOptions, type PairedEvalueSequence, type PairedEvalueStep, type PairedMetricDelta, type PairedSignTestResult, PairwiseSteeringOptimizer, type ParaphraseRobustnessScenarioInput, type ParaphraseRobustnessScenarioResult, type ParetoFigureSpec, type ParetoPoint, type ParetoResult, type ParseStudentLabel, type PartitionHeldOutOptions, type PendingCostCall, type PendingCostCallView, type PersistedFinding, type PersonaConfig, type PersonaRigor, type Playbook, type PlaybookEntry, 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, PolicyEditValidationError, type PoolSlot, type PositionalBiasResult, type PrReviewAuditCase, type PrReviewBenchmarkSummary, type PrReviewComment, type PrReviewMatchedFinding, type PrReviewOutcome, type PrReviewReferenceFinding, type PrReviewScore, type PrReviewScoreWeights, type PrReviewSeverity, type PrReviewSource, type PreferenceMemoryEntry, type PreflightModelsOptions, type PreflightOutcome, type ProducedProposal, type ProducedState, type ProductBenchmarkArm, type ProductBenchmarkArtifactPaths, type ProductBenchmarkBudgets, type ProductBenchmarkExportOptions, type ProductBenchmarkExportResult, type ProductBenchmarkManifest, type ProductBenchmarkProfileRef, type ProductBenchmarkRecord, type ProductBenchmarkRepoRef, type ProductBenchmarkRunInput, type ProductBenchmarkScenario, type ProductBenchmarkSingleRunExportOptions, type ProductBenchmarkSplit, type ProductBenchmarkSubstrateVersions, type ProductBenchmarkValidationReport, ProductClient, type ProductClientConfig, type ProfileAxisSpec, type ProjectRuntimeTrajectoryEvidenceOptions, type ProjectedOtlpSpan, type PromptHandle, PromptRegistry, type ProportionInterval, type ProposalEventLike, type ProposeAutomatedPullRequestInput, type ProposeAutomatedPullRequestResult, type ProposeFn, type ProposeInput, type ProposeOutput, type ProposeReviewConfig, type ProposeReviewControlAction, type ProposeReviewControlConfig, type ProposeReviewControlResult, type ProposeReviewControlState, type ProposeReviewReport, type ProposeReviewShot, type ProposedSideEffect, type ProvenanceReader, type ProviderRedactor, type QueryTracesPage, REDACTION_VERSION, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, RESEARCH_REPORT_HARD_PAIR_FLOOR, RUN_COST_ATTR_KEYS, type RawAnalystEvidence, type RawAnalystFinding, type RawProviderDirection, type RawProviderEvent, type RawProviderSink, type RawProviderSinkFilter, type RecordRunsOptions, type RedTeamCase, type RedTeamCategory, type RedTeamFinding, type RedTeamPayload, type RedTeamReport, type RedactionReport, type RedactionRule, type ReferenceEquivalenceJudgeInput, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceJudgeResult, type ReferenceEquivalenceScenario, type ReferenceMatchResult, type ReferenceReplayAdapter, type ReferenceReplayAdapterFn, type ReferenceReplayAdapterLike, type ReferenceReplayAggregate, type ReferenceReplayCandidate, type ReferenceReplayCase, type ReferenceReplayCaseRun, type ReferenceReplayExecutionScenario, type ReferenceReplayItem, type ReferenceReplayMatch, type ReferenceReplayMatchStrategy, type ReferenceReplayMatcher, type ReferenceReplayPromotionDecision, type ReferenceReplayPromotionPolicy, type ReferenceReplayRun, type ReferenceReplayRunContext, type ReferenceReplayRunOptions, type ReferenceReplayRunStore, type ReferenceReplayScenario, type ReferenceReplayScenarioScore, type ReferenceReplayScore, type ReferenceReplayScoreOptions, type ReferenceReplaySplit, type ReferenceReplaySplitComparison, type ReferenceReplaySteeringRowsOptions, type ReflectionContext, type ReflectionProposal, type RegistryRunOpts, type ReleaseConfidenceAxis, type ReleaseConfidenceAxisName, type ReleaseConfidenceInput, type ReleaseConfidenceIssue, type ReleaseConfidenceMetrics, type ReleaseConfidenceScorecard, type ReleaseConfidenceStatus, type ReleaseConfidenceThresholds, type ReleaseTraceEvidence, type RenderReleaseReportOptions, type RenderStudentPrompt, type RepeatedActionOptions, ReplayCache, type ReplayCacheEntry, ReplayCacheMissError, type ReplayCacheStats, ReplayError, type ReplayFetchOptions, type RepoRef, type RequirementCheck, type ResearchReport, type ResearchReportCandidate, type ResearchReportDecision, type ResearchReportMethodology, type ResearchReportOptions, type ResearchReportRecommendation, type Researcher, type RetrievalSpan, type Review, type ReviewFn, type ReviewInput, type ReviewMemoryEntry, type ReviewMemoryStore, type ReviewerMemoryEntry, type ReviewerOutput, type ReviewerPromptInput, type ReviewerSoftFailDefaults, type ReviewerVerificationSummary, type RiskDifferenceResult, type RobustnessResult, type RouteMap, type RoutedField, type RouterTransportOpts, type RubricDimension, type Run, type RunCommandInput, type RunCommandResult, type RunCompleteHook, type RunCompleteHookContext, type RunCostProvenance, RunCritic, type RunCriticOptions, type RunDistillationOptions, type RunDistillationResult, type RunEvidenceMetadata, type RunFilter, RunIntegrityError, type RunIntegrityExpectations, type RunIntegrityIssue, type RunIntegrityIssueCode, type RunIntegrityReport, type RunJudgeMetadata, type RunLayer, type RunOutcome, type RunPaidCallInput, type RunRecord, type RunRecordBackend, type RunRecordFilter, RunRecordValidationError, type RunScore, type RunScoreWeights, type RunSplitTag, type RunStatus, type RunTokenUsage, type RunTrace, type RuntimeEventLike, type RuntimeResolution, type RuntimeTrajectoryEvidenceProjection, type RuntimeTrajectoryEvidenceSummary, type RuntimeTrajectoryHookEvent, type RuntimeTrajectoryRecord, type RuntimeTrajectoryRunRecord, SEMANTIC_CONCEPT_JUDGE_VERSION, SKILL_USAGE_ANALYST, SPAN_KIND_ATTR_KEYS, type SandboxDriver, SandboxHarness, type SandboxHarnessResult, type SandboxJudgeKind, type SandboxJudgeResult, type SandboxJudgeSpec, type SandboxPool, type SandboxResult, type SandboxSdkTransportOpts, type SandboxSpan, type SatisfiedBy, type ScanOptions, type Scenario$1 as Scenario, type ScenarioCost, type ScenarioFile, ScenarioRegistry, type ScenarioResult, type ScoreKnowledgeReadinessOptions, type Scorecard, type ScorecardCell, type ScorecardCellDiff, type ScorecardDiff, type ScorecardEntry, type ScorecardLogLine, type ScoredTarget, type SearchSpanResult, type SearchTraceResult, type SeatName, type SeatPresetName, SeatUnsetError, type SelfPlayOptions, type SelfPlayProposer, type SelfPlayScorer, type SelfPreferenceResult, type SemanticConceptJudgeInput, type SemanticConceptJudgeOptions, type SemanticConceptJudgeResult, type SequentialDecision, type SerializedRegex, type SeriesConvergenceOptions, type SeriesConvergenceResult, type Severity, type SignTestAlternative, type SignedManifest, type SignedManifestAlgo, type SingleBackendDivergence, SingleBackendError, type SingleBackendField, type SingleBackendReport, SkillUsageAnalyst, type SliceOptions, type Slo, type SloCheckResult, type SloComparator, type SloReport, type SloSeverity, type SlopCategory, type SlotFactory, type Span, type SpanBase, type SpanFilter, type SpanHandle, type SpanKind, type SpanMatchRecord, SpanNotFoundError, type SpanPredicate, type SpanStatus, type SplitGoldOptions, type SseUsageMode, type SteeringBundle, type SteeringChange, type SteeringDelta, type SteeringOptimizationResult, type SteeringOptimizationRow, type SteeringOptimizationSelector, type SteeringOptimizerBackend, type SteeringOptimizerConfig, type SteeringRolePrompt, type StepAttribution, type StopDecision, type StreamingDetector, type SuboptimalCode, type SuboptimalSignal, SubprocessSandboxDriver, type SubprocessSandboxDriverOptions, type SummaryTable, type SummaryTableOptions, type SummaryTableRow, type SynthesisReason, type SynthesisTarget, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TRACE_SCHEMA_VERSION, type TaskGold, type TaskHeadroom, type TestGradedRunOptions, type TestGradedRunResult, type TestGradedScenario, type TestOutputParser, type TestResult, type TextMatcher, type ThresholdContract, TokenCounter, type TokenSpec, type ToolCallEventLike, type ToolMatcher, type ToolSpan, type ToolSpanOtlpInput, type ToolStats, type ToolUseMetrics, type ToolUseOptions, type TraceAggregate, type TraceAnalysisStore, type TraceAnalystByteBudgets, type TraceAnalystFilters, type TraceAnalystGolden, type TraceAnalystHookOptions, type TraceAnalystKindSpec, type TraceAnalystSpan, type TraceAnalystSpanKind, type TraceAnalystSpanStatus, type TraceAnalystTraceSummary, type TraceContract, TraceContractBuilder, TraceEmitter, type TraceEmitterOptions, type TraceEvent, TraceFileMissingError, type TraceInsightContext, type TraceInsightFinding, type TraceInsightPanelRole, type TraceInsightPromptInput, type TraceInsightQualityGate, type TraceInsightQuestion, type TraceInsightReadiness, type TraceInsightSuite, type TraceInsightTask, TraceNotFoundError, type TraceStore, type TraceStoreSource, type TraceStoreToOtlpOptions, type TracedAnalystOptions, type TracedJudgeOptions, type TracesToOtlpResult, type Trajectory, type TrajectoryStep, type TreatmentClass, type TreatmentGate, type TreatmentGateInput, type TreatmentGateOptions, type TrialTrace, type Turn, type TurnMetrics, type TurnResult, UI_FINDING_SEVERITIES, UI_LENSES, UNIVERSAL_FINDERS, type UiFinding, type UiFindingScreenshot, type UiFindingSeverity, type UiLens, type UserQuestion, type ValidationContext, ValidationError, type ValidationIssue, type ValidationResult, type VerbosityBiasResult, type Verdict, type VerdictCacheStats, type VerdictCacheStore, type Verification, VerificationError, type VerificationReport, type VerifyContext, type VerifyFn, type VerifyOptions, type ViewSpansResult, type ViewTraceOversized, type ViewTraceResult, type VisualDiffOptions, type VisualDiffResult, type ViteDeployRunnerInput, type WeightedCompositeInput, type WeightedCompositeResult, type WorkerDriverContext, type WorkflowTopology, type WorkspaceAssertion, type WorkspaceAssertionResult, type WorkspaceInspector, type WorkspaceSnapshot, type WranglerDeployRunnerInput, acquisitionPlansForKnowledgeGaps, admitPolicyEdit, adversarialJudge, agentProfileCellHashMaterial, agentProfileCellKey, agentProfileHash, agentProfileId, agentProfileModelId, agentVisibleFields, aggregateJudgeVerdicts, aggregateLlm, aggregatePrReviewScore, aggregateRunScore, allCriticalPassed, analyzeAntiSlop, analyzeSeries, analyzeTraces, appendScorecard, applyLlmSpanOtlpAttributes, applyPolicyEditToSurface, applyToolSpanOtlpAttributes, argHash, asNumber, asString, assertCapabilityHeadroom, assertCrossFamily, assertLlmRoute, assertModelsServed, assertNoHiddenLeak, assertProductBenchmarkRun, assertRealAgentReceipts, assertRealBackend, assertReleaseConfidence, assertRunAgentProfileCell, assertRunCaptured, assertSingleBackend, assignFeedbackSplit, assignHeldOutTag, attachCostToReport, attest, attributeCounterfactuals, backoffMs, deterministicSplit as benchmarkDeterministicSplit, index$1 as benchmarks, benjaminiHochberg, bisect, blendHeldout, blockingKnowledgeEval, bonferroni, bootstrapCi, buildAgentInterfaceProfileCell, buildAgentProfileCell, buildAgreementJudge, buildDefaultAnalystRegistry, buildDriverSystemPrompt, buildProductBenchmarkManifest, buildReflectionPrompt, buildReviewerPrompt, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, buildWorkerDriverSystemPrompt, byteLengthRange, cachedJudge, calibrateJudge, calibrateJudgeContinuous, callLlm, callLlmJson, canaryLeakView, canonicalJson, canonicalize, capabilityHeadroom, captureFetchToRawSink, causalAttribution, checkBehavioralCanary, checkCanaries, checkSlos, checkTraceContracts, clamp01, classifyFailure, classifyTreatment, cliffsDelta, clusteredPairedBinary, codeExecutionJudge, cohensD, coherenceJudge, collectionPreserved, commentsForSource, commitBisect, comparePairedArms, compareReferenceReplay, compareToBaseline, compilerJudge, completionVerdict, composeParsers, composeValidators, computeExperimentStats, computeFindingId, computePolicyEditId, computeToolUseMetrics, computeTraceMetrics, confidenceInterval, containsAll, contentHash, contextInputTokens, continuousAgreement, contractJudge, controlFailureClassFromVerification, controlRunToFeedbackTrajectory, controlRunToRunRecord, convertTraceStoresToOtlp, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, costForTokenPricing, costForUsage, costReceiptFromLlm, costReceiptFromLlmError, costReport, createAnalystAi, createAntiSlopJudge, createChatClient, createCustomJudge, createDefaultReviewer, createDomainExpertJudge, createFeedbackTrajectory, createIntentMatchJudge, createLlmCorrectnessChecker, createLlmReviewer, createOtelExporter, createOtelTracingStore, createReferenceEquivalenceJudge, createReplayFetch, createSandboxPool, createSemanticConceptJudge, createTokenRecallChecker, createTraceAnalystKind, crossTraceDiff, crowdingDistance, dataDescriptionBits, decideNextUserTurn, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultBlendWeights, defaultIsMaterial, defaultJudges, defaultParseStudentLabel, defaultProviderRedactor, defaultReferenceReplayMatcher, defaultRenderStudentPrompt, defaultTraceInsightPanel, deployGateLayer, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, distillPlaybook, domainEvidencePattern, dominates, eProcess, ensembleJudge, errorStreakDetector, estimateCost, estimateTokens, evaluateActionPolicy, evaluateContract, evaluateHypothesis, evaluateInterimReleaseConfidence, evaluateOracles, evaluateReleaseConfidence, evaluateTraceContract, executeScenario, expandProfileAxes, expectAgent, exportProductBenchmark, exportProductBenchmarkRuns, exportRewardModel, exportRunAsOtlp, extractAssetUrls, extractErrorCount, extractOtlpAttributes, extractProducedState, extractUsage, extractUsageFromResponse, extractUsageFromSse, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToDatasetScenario, feedbackTrajectoryToOptimizerRow, fieldAgreement, fileContains, fileExists, fileExperimentStore, fileVerdictCache, findAutoMatchNoExpectation, findConstructorCwdDropped, findFallbackToPass, findLiteralTruePass, findProductBenchmarkArtifacts, findSkipCountsAsPass, firstNumberAttr, firstStringAttr, flattenOtlpExportToNdjson, flowLayer, fnv1a32, formatBenchmarkReport, formatDriverReport, formatFindings, formatScorecardDiff, gainHistogram, gateTreatmentApplied, gateTreatmentFromMetrics, gateTreatmentFromSpans, gateTreatmentFromToolSpans, ghCliClient, gitProvenanceReader, precision as goldenPrecision, gradeOnHidden, gradeSemanticStatus, groupBy, groupRunsByAgentProfileCell, harnessAxisOf, hasCapturedToolArgs, hashContent, hashJson, hashScenarios, hashToUnit, hiddenGrade, holm, htmlContainsElement, httpGithubClient, improvementVerdict, inMemoryExperimentStore, inMemoryReferenceReplayStore, inMemoryReviewStore, inMemoryRunRecordBackend, inMemoryVerdictCache, inferDomainKeywords, inferOtlpKind, interRaterReliability, interpretCliffs, iqr, isHiddenDestination, isJudgeSpan, isLlmSpan, isModelPriced, isOtelConfigured, isPolicyEdit, isRetrievalSpan, isRunRecord, isSandboxSpan, isToolSpan, isTransientLlmError, iterateRawCalls, jestTestParser, jsonHasKeys, jsonShape, jsonlReferenceReplayStore, jsonlReviewStore, jsonlRunRecordBackend, judgeFamily, judgeReplayGate, judgeSpans, keyPreserved, knowledgeReadinessTracePayload, leaderboard, linterJudge, llmJudge, llmSpanFromProvider, llmSpans, loadGoldScenarios, loadScorecard, loadScorerFromGrader, localCommandRunner, lowercaseMutator, makeEvalTools, makeFinding, makePolicyEdit, makePolicyEditCandidateRecord, mannWhitneyU, mapConcurrent, matchGoldens, matchSpan, maximumChargeForLlmRequest, mcnemar, mcnemarPower, mcnemarRequiredN, mergeLayerResults, mergeSteeringBundle, modelDescriptionBits, modelHasSnapshot, modelPriceKey, mulberry32, multiToolchainLayer, noProgressDetector, normalizeScores, notBlocked, objectiveEval, observeAll, otelRunCompleteHook, otlpRowsToRunRecords, otlpRowsToTraceRunRecords, otlpToRunRecords, otlpToTraceRunRecords, pairArms, pairedBootstrap, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedSignTest, pairedTTest, paraphraseRobustness, paraphraseRobustnessScenarios, paretoChart, paretoFrontier, paretoFrontierWithCrowding, parseCorrectnessResponse, parseFeedbackTrajectoriesJsonl, parseGoldJsonl, parseReflectionResponse, parseRunRecordSafe, parseRuntimeTrajectoryHookEvent, partialCredit, partitionHeldOut, passAtK, passOrthogonality, pearsonR, pixelDeltaRatio, planTraceInsightQuestions, policyEditFromFinding, policyEditsFromFindings, politenessPrefixMutator, positionalBias, preflightModels, printDriverSummary, probeLlm, productBenchmarkIntegrityFailures, productBenchmarkMutableSurfaces, productBenchmarkRepoIdentity, productBenchmarkSplits, index as profile, projectOtlpFlatLine, projectRuntimeTrajectoryEvidence, promptBisect, proposeSynthesisTargets, providerFromBaseUrl, pytestTestParser, ranks, readOtlpStatus, readProductBenchmarkManifest, readProductBenchmarkRecords, recordRuns, recordRunsToScorecard, redTeamDataset, redTeamReport, redactString, redactValue, referenceReplayRunsToSteeringRows, referenceReplayScenarioToRunScore, regexMatch, regexMatches, renderMarkdownReport, renderPlaybookMarkdown, renderPreferenceMemoryMarkdown, renderPriorFindings, renderReleaseReport, renderSteeringText, renderUpstreamFindings, repeatedActionDetector, replayFeedbackTrajectories, replayFeedbackTrajectory, replayScorerOverCorpus, replayTraceThroughJudge, requireAgentProfileCell, requiredSampleSize, researchReport, resolveModelPricing, resolveRunCostProvenance, resolveSeat, roundTripRunRecord, routeFields, rowCount, rowWhere, runAgentControlLoop, runAssertions, runBehavioralCanaries, runCanaries, runCounterfactual, runDistillation, runE2EWorkflow, runEvalCampaign, runExpectations, runFailureClass, runHarnessExperiment, runIntentMatchJudge, runJudgeFleet, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runLiveProof, runProposeReview, runProposeReviewAsControlLoop, runRecordToProductBenchmarkRecord, runReferenceEquivalenceJudge, runReferenceReplay, runScore, runSelfPlay, runSemanticConceptJudge, runTestGradedScenario, runsForScenario, scalarScore, scanForMuffledGates, scoreContinuity, scoreFromEvals, scoreKnowledgeReadiness, scorePolicyEditReadiness, scorePrReviewComments, scorePrReviewSource, scoreRedTeamOutput, scoreReferenceReplay, scoreTraceInsightReadiness, seatPresets, securityJudge, selectHarnessVariant, selfPreference, sentenceReorderMutator, serializeFeedbackTrajectoriesJsonl, signManifest, spearmanR, splitGold, statusAdvanced, stopOnNoProgress, stopOnRepeatedAction, stringField, stripFencedJson, subjectiveEval, summarizeAgentReceiptIntegrity, summarizeBackendIntegrity, summarizeHarnessResults, summarizePrReviewBenchmark, summarizePreferenceMemory, summaryTable, testJudge, textInSnapshot, throwIfRunIncomplete, toAgentProfileJson, toLangfuseEnvelope, toOpenAiTool, toPrometheusText, tokenizeDomainWords, toolNamesForRun, toolSpans, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceContract, traceJudge, traceJudgeEnsemble, traceSpanKindToOpenInferenceKind, tracedAnalyzeTraces, typoMutator, urlContains, userQuestionsForKnowledgeGaps, validateAgentProfileCell, validatePolicyEdit, validatePolicyEditCandidateRecord, validateProductBenchmarkManifest, validateProductBenchmarkRecord, validateProductBenchmarkRun, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyAttestation, verifyCompletion, verifyManifest, visualDiff, viteDeployRunner, vitestTestParser, weightedComposite, weightedMean, weightedRecall, welchsTTest, whitespaceCollapseMutator, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit, withHeldoutBlend, withJudgeRetry, withOtelPipeline, wranglerDeployRunner };
+export { AGENT_PROFILE_KINDS, ATTESTATION_ALGORITHM, type ActionExecutionPolicy, type ActionPolicyDecision, type ActionableSideInfo, type ActiveLearningOptions, type AdapterRun, AgentDriver, type AgentDriverConfig, AgentEvalError, type AgentEvalErrorCode, type AgentInterfaceProfileLike, type AgentProfileCell, type AgentProfileCellInput, type AgentProfileCellSchemaVersion, AgentProfileCellValidationError, type AgentProfileDimensionValue, type AgentProfileHarness, type AgentProfileJson, type AgentProfileJsonObject, type AgentProfileKind, type AgentProfileRuntimeReceipt, type AgentProfileSource, type AgentProfileSourceInput, type AgreementResult, type AlignmentOp, type Analyst, type AnalystContext, type AnalystCost, type AnalystFinding, type AnalystHooks, type AnalystInputKind, AnalystRegistry, type AnalystRegistryOptions, type AnalystRequirements, type AnalystRunEvent, type AnalystRunInputs, type AnalystRunResult, type AnalystRunSummary, type AnalystSeverity, type AnalystUsageReceipt, type AnalyzeTracesInput, type AnalyzeTracesOptions, type AnalyzeTracesResult, type AnalyzeTracesTurnSnapshot, type AntiSlopConfig, type AntiSlopIssue, type AntiSlopReport, type Artifact$1 as Artifact, type ArtifactCheck, type Artifact as ArtifactCheckArtifact, type ArtifactEventLike, type ArtifactResult, type ArtifactValidator, type AsiSeverity, type AssertCapabilityHeadroomOptions, type AssertCrossFamilyOptions, type AssertSingleBackendOptions, type AttestationProvenance, type AttestationVerification, type AttestedReport, type AutoPrClient, AxGepaSteeringOptimizer, type AxSteeringOptimizerConfig, BENCHMARK_SPLIT_SEED, type BackendDescriptor, BackendIntegrityError, type BackendIntegrityReport, type BaselineOptions, type BaselineReport, BehaviorAssertion, type BehavioralMetrics, type BehavioralTokenSequence, type BenchmarkAdapter, type BenchmarkDatasetItem, type BenchmarkEvaluation, type BenchmarkFamily, type BenchmarkReport$1 as BenchmarkReport, type BenchmarkResponder, BenchmarkRunner, type BenchmarkRunnerConfig, type BenchmarkScenario, type BenchmarkSource, type BenchmarkTaskKind, type BisectOptions, type BisectResult, type BisectStep, type BlendWeights, type BootstrapOptions, type BootstrapResult, BudgetBreachError, BudgetGuard, type BudgetLedgerEntry, type BudgetPolicy, type BudgetSpec, type BuildAgreementJudgeOptions, CODING_HARNESSES, type CachedJudge, type CachedJudgeOptions, type CalibrationResult, CallExpectation, CallbackResearcher, type CallbackResearcherOptions, type CampaignFactoryParams, type CampaignIntegrityPolicy, type CampaignRunContext, type CampaignRunOutcome, type CampaignRunner, type CampaignScenario, type CampaignVariant, type CanaryAlert, type CanaryKind, type CanaryLeak, type CanaryOptions, type CanaryReport, type CanarySeverity, type CandidateComparison, type CandidateScenario, type CandidateScore, type CanonicalRawAnalystFinding, type CapabilityHeadroomOptions, type CapabilityHeadroomResult, type CaptureFetchContext, type CaptureFetchOptions, CaptureIntegrityError, type CausalAttributionReport, type CellVerdict, type ChannelRollup, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTransport, type CheckResult, type CliBridgeTransportOpts, type CliffsMagnitude, type ClusterBootstrapInterval, type ClusterSignFlipAlternative, type ClusterSignFlipResult, type ClusteredBinaryCluster, type ClusteredMatchedPair, type ClusteredPairedBinaryOptions, type ClusteredPairedBinaryResult, type ClusteredPairedBinaryStatistics, type CollectedArtifacts, type CommandRunner, type CompareLabels, type ComparePairedArmsOptions, type CompletionCriterion, type CompletionRequirement, type CompletionVerdict, type ConceptComplexity, type ConceptFinding, type ConceptSpec, type ConceptWeightStrategy, ConfigError, type ContinuityCheck, type ContinuityCheckResult, type ContinuityReport, type ContinuitySnapshotPair, type ContinuousAgreement, type ContinuousAgreementOptions, type ContinuousCalibrationResult, type ContractCheckResult, type ContractJudgeOptions, type ContractMetric, type ContractReport, type ContractRule, type ContractRuleKind, type ContractSpan, type ContractVerdict, type ContractViolation, type ControlActionFailureMode, type ControlActionOutcome, type ControlBudget, type ControlContext, type ControlDecision, type ControlEvalResult, type ControlRunResult, type ControlRunToRunRecordOptions, type ControlRuntimeConfig, type ControlRuntimeError, type ControlSeverity, type ControlStep, type ControlStopPolicies, ConvergenceTracker, type CorpusAgreementOptions, type CorpusAgreementPerDimension, type CorpusAgreementReport, type CorpusScoreRecord, type CorrectnessChecker, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, type CostChannel, type CostEntry, CostLedger, type CostLedgerEntry, type CostLedgerFilter, type CostLedgerHandle, type CostLedgerOptions, type CostLedgerPersistence, CostLedgerPersistenceError, type CostLedgerSummary, type CostReceipt, CostReceiptCaptureError, type CostReceiptInput, type CostReport, CostReservationExceededError, type CostResult, type CostSummary, CostTracker, type CostUsage, type CounterfactualContext, type CounterfactualMutation, type CounterfactualResult, type CounterfactualRunner, type CreateAnalystAiConfig, type CreateChatClientOpts, type CreateDefaultReviewerOptions, type CreateExperimentInput, type CreateSandboxPoolOpts, type CreateTraceAnalystKindOpts, CrossFamilyError, type CrossTraceDiff, type CrossTraceDiffOptions, type CustomTokenPricing, DEFAULT_AGENT_SLOS, DEFAULT_COMPLEXITY_WEIGHTS, DEFAULT_RULES as DEFAULT_FAILURE_RULES, DEFAULT_FINDERS, DEFAULT_HARNESS_OBJECTIVES, DEFAULT_MUTATION_PRIMITIVES, DEFAULT_MUTATORS, DEFAULT_PR_REVIEW_SCORE_WEIGHTS, DEFAULT_REDACTION_RULES, DEFAULT_RED_TEAM_CORPUS, DEFAULT_RUN_SCORE_WEIGHTS, DEFAULT_SEVERITY_WEIGHTS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, type DataAcquisitionPlan, Dataset, type DatasetDifficulty, type DatasetManifest, type DatasetOverview, type DatasetProvenance, type DatasetScenario, type DatasetSplit, type DecideNextUserTurnOpts, type DefaultAnalystRegistryOptions, type DefaultVerdict, type DeployFamily, type DeployGateLayerInput, type DeployRunResult, type DeployRunner, type DescriptionLengthCandidate, type DescriptionLengthConfig, type DescriptionLengthDecision, type DescriptionLengthEvidence, DescriptionLengthGate, type DescriptionLengthRejectionCode, type DetectorEvent, type DetectorSeverity, type DetectorSignal, type DiffPolicy, type DiffScorecardOptions, type DirEntry, type DirectProviderTransportOpts, type Direction, type DiscoverPersonasOptions, type DiscoveredPersona, DockerSandboxDriver, type DriverResult, type DriverState, DualAgentBench, type DualAgentBenchConfig, type DualAgentReport, type DualAgentRound, type DualAgentScenario, type DualAgentScenarioResult, type EProcess, type EProcessOptions, type EProcessState, type EProcessStep, ERROR_COUNT_PATTERNS, type EnsembleAggregate, type EnsembleJudgeOptions, type ErrorCluster, type ErrorCountPattern, type ErrorStreakOptions, type EvalCampaignOptions, type EvalCampaignResult, type EvalResult, type EvalToolDef, EvalTraceStore, type EventFilter, type EventKind, type EvidenceRef, type EvolutionRound, type ExecutorConfig, type Expectation, type Experiment, type ExperimentPlan, type ExperimentProvenance, type ExperimentRep, type ExperimentResult, type ExperimentStats, type ExperimentStore, ExperimentTracker, type ExperimentTrackerOptions, type ExperimentVerdict, type ExportableSpan, type ExportedRewardModel, type ExtractOptions, type ExtractResult, type ExtractUsageFromSseOptions, type ExtractedUsage, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, type FactorContribution, type FactorialCell, type FailedRun, type FailureClass, type FailureClassification, type FailureContext, type FailureMode, type FailureRule, type FeedbackArtifactType, type FeedbackAttempt, type FeedbackLabel, type FeedbackLabelKind, type FeedbackLabelSource, type FeedbackOptimizerRow, type FeedbackOutcome, type FeedbackPattern, type FeedbackReplayAdapter, type FeedbackReplayResult, type FeedbackSeverity, type FeedbackSplitPolicy, type FeedbackTask, type FeedbackTrajectory, type FeedbackTrajectoryFilter, type FeedbackTrajectoryStore, type FieldAgreementSpec, type FieldDestination, type FileChange, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, type FileSystemRawProviderSinkOptions, FileSystemTraceStore, type FileSystemTraceStoreOptions, type Finding, type FindingSubject, type FindingSubjectKind, type FindingToPolicyEditOptions, type FindingsDiff, FindingsStore, type FlattenOtlpOptions, type FlowAction, type FlowLayerEnv, type FlowLayerFactoryInput, type FlowRunner, type FlowRunnerStepResult, type FlowSpec, type FlowStep, type GainDistributionBin, type GainDistributionFigureSpec, type GainDistributionOptions, type GateDecision$1 as GateDecision, type GateEvidence, type GenericSpan, type GhCliClientOptions, type GoldScenario, type GoldSplit, type GoldenItem, type GoldenSeverity, type GoldenSpec, HARNESS_NATIVE_MODEL, type HarnessAdapter, type HarnessConfig, type HarnessExperimentConfig, type HarnessExperimentResult, type HarnessIntervention, type HarnessRunRequest, type HarnessRunResult, type HarnessScenario, type HarnessSelection, type HarnessVariant, type HarnessVariantReport, type HeadroomClass, type HeadroomInput, HeldOutGate, type HeldOutGateConfig, type HeldOutGateRejectionCode, type HeldOutPartition, type HiddenCriteriaGrader, type HiddenGradeResult, type HiddenLeak, HoldoutAuditor, HoldoutLockedError, type HttpGithubClientOptions, type HypothesisManifest, type HypothesisResult, IMPROVEMENT_KIND_SPEC, INPUT_VALUE, INTENT_MATCH_JUDGE_VERSION, type ImageData, type ImprovementThresholds, type ImprovementVerdictResult, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, type InMemoryRawProviderSinkOptions, InMemoryTraceStore, InMemoryWorkspaceInspector, type InferenceScorer, type InspectorContext, type IntentMatchInput, type IntentMatchOptions, type IntentMatchResult, type InteractionContribution, type InterimReleaseConfidence, type InterimReleaseConfidenceInput, type JudgeConfig$1 as JudgeConfig, JudgeError, type JudgeFamily, type JudgeFleetOptions, type JudgeFn, type JudgeInput, JudgeParseError, type JudgeReplayGateArgs, type JudgeReplayResult, type JudgeRetryOutcome, type JudgeRetryPolicy, type JudgeRubric, JudgeRunner, type JudgeScore$1 as JudgeScore, type JudgeScoreInput, type JudgeScoresRecord, type JudgeSpan, type JudgeVerdict, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, type KeywordConceptSpec, type KeywordCoverageFinding, type KeywordCoverageOptions, type KeywordCoverageResult, type KnowledgeAcquisitionMode, type KnowledgeBundle, type KnowledgeFallbackPolicy, type KnowledgeFreshness, type KnowledgeImportance, type KnowledgeReadinessReport, type KnowledgeRecommendedAction, type KnowledgeRequirement, type KnowledgeRequirementCategory, type KnowledgeResponsibleSurface, type KnowledgeSensitivity, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_CACHE_WRITE_TOKENS, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, LLM_CONTEXT_TOKENS, LLM_COST_ATTR_KEYS, LLM_COST_USD, LLM_INPUT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_MODEL_ATTR_KEYS, LLM_MODEL_NAME, LLM_OUTPUT_TOKENS, LLM_OUTPUT_TOKEN_ATTR_KEYS, LLM_REASONING_TOKENS, LLM_REASONING_TOKEN_ATTR_KEYS, type LangfuseEnvelope, type LangfuseGeneration, type LangfuseScore, type Layer, type LayerResult, type LayerStatus, type LeaderboardOptions, type LeaderboardRow, type LiveProofArtifact, type LiveProofConfig, type LiveProofContext, type LiveProofResult, LlmCallError, type LlmCallMetadata, type LlmCallRequest, type LlmCallResult, LlmClient, type LlmClientOptions, type LlmCorrectnessCheckerOpts, type LlmJsonCall, type LlmJudgeDimension, type LlmJudgeOptions, type LlmMessage, LlmResponseError, type LlmReviewerConfig, LlmRouteAssertionError, type LlmRouteRequirements, type LlmSpan, type LlmSpanOtlpInput, type LlmUsage, LockedJsonlAppender, MODEL_PRICING, type MakeEvalToolsConfig, type MatchResult, type MatchedPair, type MatcherResult, type MaximumCharge, type McNemarResult, type MeasurementPolicy, type MergeOptions, type Message, type MetricSamples, type MetricVerdict, MetricsCollector, type MintRolloutOptions, type MintRolloutResult, type MockTransportOpts, type ModelCostRollup, type ModelPreflight, type ModelSeats, ModelsUnreachableError, type MuffledFinder, type MuffledFinding, MultiLayerVerifier, type MultiToolchainLayerConfig, type Mutator, Mutex, type NoLeakOptions, type NoProgressOptions, NoopRawProviderSink, NoopResearcher, NotFoundError, OPENINFERENCE_SPAN_KIND, OTEL_AGENT_EVAL_SCOPE, OUTPUT_VALUE, type Objective, type Oracle, type OracleObservation, type OracleReport, type OracleResult, type OrthogonalityInput, type OrthogonalityResult, type OtelExportConfig, type OtelExporter, type OtelPipelineHandle, type OtelPipelineOptions, type OtlpExport, OtlpFileTraceStore, type OtlpFileTraceStoreOptions, type OtlpFlatLine, type OtlpResourceSpans, type OtlpSpan, type OtlpToRunRecordsOptions, type OtlpTraceRunRecord, POLICY_EDIT_AXES, POLICY_EDIT_CANDIDATE_RECORD_SCHEMA, POLICY_EDIT_TARGET_SURFACES, type PaidCallResult, type PairArmsOptions, type PairArmsResult, type PairedArmRow, type PairedArmsComparison, type PairedBootstrapOptions, type PairedBootstrapResult, type PairedCorrectness, type PairedEvalueOptions, type PairedEvalueSequence, type PairedEvalueStep, type PairedMetricDelta, type PairedSignTestResult, PairwiseSteeringOptimizer, type ParaphraseRobustnessScenarioInput, type ParaphraseRobustnessScenarioResult, type ParetoFigureSpec, type ParetoPoint, type ParetoResult, type ParseStudentLabel, type PartitionHeldOutOptions, type PendingCostCall, type PendingCostCallView, type PersistedFinding, type PersonaConfig, type PersonaRigor, type Playbook, type PlaybookEntry, 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, PolicyEditValidationError, type PoolSlot, type PositionalBiasResult, type PrReviewAuditCase, type PrReviewBenchmarkSummary, type PrReviewComment, type PrReviewMatchedFinding, type PrReviewOutcome, type PrReviewReferenceFinding, type PrReviewScore, type PrReviewScoreWeights, type PrReviewSeverity, type PrReviewSource, type PreferenceMemoryEntry, type PreflightModelsOptions, type PreflightOutcome, type ProducedProposal, type ProducedState, type ProductBenchmarkArm, type ProductBenchmarkArtifactPaths, type ProductBenchmarkBudgets, type ProductBenchmarkExportOptions, type ProductBenchmarkExportResult, type ProductBenchmarkManifest, type ProductBenchmarkProfileRef, type ProductBenchmarkRecord, type ProductBenchmarkRepoRef, type ProductBenchmarkRunInput, type ProductBenchmarkScenario, type ProductBenchmarkSingleRunExportOptions, type ProductBenchmarkSplit, type ProductBenchmarkSubstrateVersions, type ProductBenchmarkValidationReport, ProductClient, type ProductClientConfig, type ProfileAxisSpec, type ProjectRuntimeTrajectoryEvidenceOptions, type ProjectedOtlpSpan, type PromptHandle, PromptRegistry, type ProportionInterval, type ProposalEventLike, type ProposeAutomatedPullRequestInput, type ProposeAutomatedPullRequestResult, type ProposeFn, type ProposeInput, type ProposeOutput, type ProposeReviewConfig, type ProposeReviewControlAction, type ProposeReviewControlConfig, type ProposeReviewControlResult, type ProposeReviewControlState, type ProposeReviewReport, type ProposeReviewShot, type ProposedSideEffect, type ProvenanceReader, type ProviderRedactor, type QueryTracesPage, REDACTION_VERSION, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, RESEARCH_REPORT_HARD_PAIR_FLOOR, ROLLOUT_FORMAT, RUN_COST_ATTR_KEYS, type RawAnalystEvidence, type RawAnalystFinding, type RawProviderDirection, type RawProviderEvent, type RawProviderSink, type RawProviderSinkFilter, type RecordRunsOptions, type RedTeamCase, type RedTeamCategory, type RedTeamFinding, type RedTeamPayload, type RedTeamReport, type RedactionReport, type RedactionRule, type ReferenceEquivalenceJudgeInput, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceJudgeResult, type ReferenceEquivalenceScenario, type ReferenceMatchResult, type ReferenceReplayAdapter, type ReferenceReplayAdapterFn, type ReferenceReplayAdapterLike, type ReferenceReplayAggregate, type ReferenceReplayCandidate, type ReferenceReplayCase, type ReferenceReplayCaseRun, type ReferenceReplayExecutionScenario, type ReferenceReplayItem, type ReferenceReplayMatch, type ReferenceReplayMatchStrategy, type ReferenceReplayMatcher, type ReferenceReplayPromotionDecision, type ReferenceReplayPromotionPolicy, type ReferenceReplayRun, type ReferenceReplayRunContext, type ReferenceReplayRunOptions, type ReferenceReplayRunStore, type ReferenceReplayScenario, type ReferenceReplayScenarioScore, type ReferenceReplayScore, type ReferenceReplayScoreOptions, type ReferenceReplaySplit, type ReferenceReplaySplitComparison, type ReferenceReplaySteeringRowsOptions, type ReflectionContext, type ReflectionProposal, type RegistryRunOpts, type ReleaseConfidenceAxis, type ReleaseConfidenceAxisName, type ReleaseConfidenceInput, type ReleaseConfidenceIssue, type ReleaseConfidenceMetrics, type ReleaseConfidenceScorecard, type ReleaseConfidenceStatus, type ReleaseConfidenceThresholds, type ReleaseTraceEvidence, type RenderReleaseReportOptions, type RenderStudentPrompt, type RepeatedActionOptions, ReplayCache, type ReplayCacheEntry, ReplayCacheMissError, type ReplayCacheStats, ReplayError, type ReplayFetchOptions, type RepoRef, type RequirementCheck, type ResearchReport, type ResearchReportCandidate, type ResearchReportDecision, type ResearchReportMethodology, type ResearchReportOptions, type ResearchReportRecommendation, type Researcher, type RetrievalSpan, type Review, type ReviewFn, type ReviewInput, type ReviewMemoryEntry, type ReviewMemoryStore, type ReviewerMemoryEntry, type ReviewerOutput, type ReviewerPromptInput, type ReviewerSoftFailDefaults, type ReviewerVerificationSummary, type RewardRow, type RiskDifferenceResult, type RobustnessResult, type RolloutRow, type RolloutScrubber, type RolloutStep, type RouteMap, type RoutedField, type RouterTransportOpts, type RubricDimension, type Run, type RunCommandInput, type RunCommandResult, type RunCompleteHook, type RunCompleteHookContext, type RunCostProvenance, RunCritic, type RunCriticOptions, type RunDistillationOptions, type RunDistillationResult, type RunEvidenceMetadata, type RunFilter, RunIntegrityError, type RunIntegrityExpectations, type RunIntegrityIssue, type RunIntegrityIssueCode, type RunIntegrityReport, type RunJudgeMetadata, type RunLayer, type RunOutcome, type RunPaidCallInput, type RunRecord, type RunRecordBackend, type RunRecordFilter, RunRecordValidationError, type RunScore, type RunScoreWeights, type RunSplitTag, type RunStatus, type RunTokenUsage, type RunTrace, type RuntimeEventLike, type RuntimeResolution, type RuntimeTrajectoryEvidenceProjection, type RuntimeTrajectoryEvidenceSummary, type RuntimeTrajectoryHookEvent, type RuntimeTrajectoryRecord, type RuntimeTrajectoryRunRecord, SEMANTIC_CONCEPT_JUDGE_VERSION, SKILL_USAGE_ANALYST, SPAN_KIND_ATTR_KEYS, type SandboxDriver, SandboxHarness, type SandboxHarnessResult, type SandboxJudgeKind, type SandboxJudgeResult, type SandboxJudgeSpec, type SandboxPool, type SandboxResult, type SandboxSdkTransportOpts, type SandboxSpan, type SatisfiedBy, type ScanOptions, type Scenario$1 as Scenario, type ScenarioCost, type ScenarioFile, ScenarioRegistry, type ScenarioResult, type ScoreKnowledgeReadinessOptions, type Scorecard, type ScorecardCell, type ScorecardCellDiff, type ScorecardDiff, type ScorecardEntry, type ScorecardLogLine, type ScoredTarget, type SearchSpanResult, type SearchTraceResult, type SeatName, type SeatPresetName, SeatUnsetError, type SelfPlayOptions, type SelfPlayProposer, type SelfPlayScorer, type SelfPreferenceResult, type SemanticConceptJudgeInput, type SemanticConceptJudgeOptions, type SemanticConceptJudgeResult, type SequentialDecision, type SerializedRegex, type SeriesConvergenceOptions, type SeriesConvergenceResult, type Severity, type SftExportOptions, type SftRow, type SignTestAlternative, type SignedManifest, type SignedManifestAlgo, type SingleBackendDivergence, SingleBackendError, type SingleBackendField, type SingleBackendReport, SkillUsageAnalyst, type SliceOptions, type Slo, type SloCheckResult, type SloComparator, type SloReport, type SloSeverity, type SlopCategory, type SlotFactory, type Span, type SpanBase, type SpanFilter, type SpanHandle, type SpanKind, type SpanMatchRecord, SpanNotFoundError, type SpanPredicate, type SpanStatus, type SplitGoldOptions, type SseUsageMode, type SteeringBundle, type SteeringChange, type SteeringDelta, type SteeringOptimizationResult, type SteeringOptimizationRow, type SteeringOptimizationSelector, type SteeringOptimizerBackend, type SteeringOptimizerConfig, type SteeringRolePrompt, type StepAttribution, type StopDecision, type StreamingDetector, type SuboptimalCode, type SuboptimalSignal, SubprocessSandboxDriver, type SubprocessSandboxDriverOptions, type SummaryTable, type SummaryTableOptions, type SummaryTableRow, type SynthesisReason, type SynthesisTarget, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TRACE_SCHEMA_VERSION, type TaskGold, type TaskHeadroom, type TestGradedRunOptions, type TestGradedRunResult, type TestGradedScenario, type TestOutputParser, type TestResult, type TextMatcher, type ThresholdContract, TokenCounter, type TokenSpec, type ToolCallEventLike, type ToolMatcher, type ToolSpan, type ToolSpanOtlpInput, type ToolStats, type ToolUseMetrics, type ToolUseOptions, type TraceAggregate, type TraceAnalysisStore, type TraceAnalystByteBudgets, type TraceAnalystFilters, type TraceAnalystGolden, type TraceAnalystHookOptions, type TraceAnalystKindSpec, type TraceAnalystSpan, type TraceAnalystSpanKind, type TraceAnalystSpanStatus, type TraceAnalystTraceSummary, type TraceContract, TraceContractBuilder, TraceEmitter, type TraceEmitterOptions, type TraceEvent, TraceFileMissingError, type TraceInsightContext, type TraceInsightFinding, type TraceInsightPanelRole, type TraceInsightPromptInput, type TraceInsightQualityGate, type TraceInsightQuestion, type TraceInsightReadiness, type TraceInsightSuite, type TraceInsightTask, TraceNotFoundError, type TraceStore, type TraceStoreSource, type TraceStoreToOtlpOptions, type TracedAnalystOptions, type TracedJudgeOptions, type TracesToOtlpResult, type Trajectory, type TrajectoryStep, type TreatmentClass, type TreatmentGate, type TreatmentGateInput, type TreatmentGateOptions, type TrialTrace, type Turn, type TurnMetrics, type TurnResult, UI_FINDING_SEVERITIES, UI_LENSES, UNIVERSAL_FINDERS, type UiFinding, type UiFindingScreenshot, type UiFindingSeverity, type UiLens, type UserQuestion, type ValidationContext, ValidationError, type ValidationIssue, type ValidationResult, type VerbosityBiasResult, type Verdict, type VerdictCacheStats, type VerdictCacheStore, type Verification, VerificationError, type VerificationReport, type VerifyContext, type VerifyFn, type VerifyOptions, type ViewSpansResult, type ViewTraceOversized, type ViewTraceResult, type VisualDiffOptions, type VisualDiffResult, type ViteDeployRunnerInput, type WeightedCompositeInput, type WeightedCompositeResult, type WorkerDriverContext, type WorkflowTopology, type WorkspaceAssertion, type WorkspaceAssertionResult, type WorkspaceInspector, type WorkspaceSnapshot, type WranglerDeployRunnerInput, acquisitionPlansForKnowledgeGaps, admitPolicyEdit, adversarialJudge, agentProfileCellHashMaterial, agentProfileCellKey, agentProfileHash, agentProfileId, agentProfileModelId, agentVisibleFields, aggregateJudgeVerdicts, aggregateLlm, aggregatePrReviewScore, aggregateRunScore, allCriticalPassed, analyzeAntiSlop, analyzeSeries, analyzeTraces, appendScorecard, applyLlmSpanOtlpAttributes, applyPolicyEditToSurface, applyToolSpanOtlpAttributes, argHash, asNumber, asString, assertCapabilityHeadroom, assertCrossFamily, assertLlmRoute, assertModelsServed, assertNoHiddenLeak, assertProductBenchmarkRun, assertRealAgentReceipts, assertRealBackend, assertReleaseConfidence, assertRunAgentProfileCell, assertRunCaptured, assertSingleBackend, assignFeedbackSplit, assignHeldOutTag, attachCostToReport, attest, attributeCounterfactuals, backoffMs, deterministicSplit as benchmarkDeterministicSplit, index$1 as benchmarks, benjaminiHochberg, bisect, blendHeldout, blockingKnowledgeEval, bonferroni, bootstrapCi, buildAgentInterfaceProfileCell, buildAgentProfileCell, buildAgreementJudge, buildDefaultAnalystRegistry, buildDriverSystemPrompt, buildProductBenchmarkManifest, buildReflectionPrompt, buildReviewerPrompt, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, buildWorkerDriverSystemPrompt, byteLengthRange, cachedJudge, calibrateJudge, calibrateJudgeContinuous, callLlm, callLlmJson, canaryLeakView, canonicalJson, canonicalize, capabilityHeadroom, captureFetchToRawSink, causalAttribution, checkBehavioralCanary, checkCanaries, checkSlos, checkTraceContracts, clamp01, classifyFailure, classifyTreatment, cliffsDelta, clusteredPairedBinary, codeExecutionJudge, cohensD, coherenceJudge, collectionPreserved, commentsForSource, commitBisect, comparePairedArms, compareReferenceReplay, compareToBaseline, compilerJudge, completionVerdict, composeParsers, composeValidators, computeExperimentStats, computeFindingId, computePolicyEditId, computeToolUseMetrics, computeTraceMetrics, confidenceInterval, containsAll, contentHash, contextInputTokens, continuousAgreement, contractJudge, controlFailureClassFromVerification, controlRunToFeedbackTrajectory, controlRunToRunRecord, convertTraceStoresToOtlp, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, costForTokenPricing, costForUsage, costReceiptFromLlm, costReceiptFromLlmError, costReport, createAnalystAi, createAntiSlopJudge, createChatClient, createCustomJudge, createDefaultReviewer, createDomainExpertJudge, createFeedbackTrajectory, createIntentMatchJudge, createLlmCorrectnessChecker, createLlmReviewer, createOtelExporter, createOtelTracingStore, createReferenceEquivalenceJudge, createReplayFetch, createSandboxPool, createSemanticConceptJudge, createTokenRecallChecker, createTraceAnalystKind, crossTraceDiff, crowdingDistance, dataDescriptionBits, decideNextUserTurn, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultBlendWeights, defaultIsMaterial, defaultJudges, defaultParseStudentLabel, defaultProviderRedactor, defaultReferenceReplayMatcher, defaultRenderStudentPrompt, defaultTraceInsightPanel, deployGateLayer, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, distillPlaybook, domainEvidencePattern, dominates, eProcess, ensembleJudge, errorStreakDetector, estimateCost, estimateTokens, evaluateActionPolicy, evaluateContract, evaluateHypothesis, evaluateInterimReleaseConfidence, evaluateOracles, evaluateReleaseConfidence, evaluateTraceContract, executeScenario, expandProfileAxes, expectAgent, exportProductBenchmark, exportProductBenchmarkRuns, exportRewardModel, exportRunAsOtlp, extractAssetUrls, extractErrorCount, extractOtlpAttributes, extractProducedState, extractUsage, extractUsageFromResponse, extractUsageFromSse, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToDatasetScenario, feedbackTrajectoryToOptimizerRow, fieldAgreement, fileContains, fileExists, fileExperimentStore, fileVerdictCache, findAutoMatchNoExpectation, findConstructorCwdDropped, findFallbackToPass, findLiteralTruePass, findProductBenchmarkArtifacts, findSkipCountsAsPass, firstNumberAttr, firstStringAttr, flattenOtlpExportToNdjson, flowLayer, fnv1a32, formatBenchmarkReport, formatDriverReport, formatFindings, formatScorecardDiff, gainHistogram, gateTreatmentApplied, gateTreatmentFromMetrics, gateTreatmentFromSpans, gateTreatmentFromToolSpans, ghCliClient, gitProvenanceReader, precision as goldenPrecision, gradeOnHidden, gradeSemanticStatus, groupBy, groupRunsByAgentProfileCell, harnessAxisOf, hasCapturedToolArgs, hashContent, hashJson, hashScenarios, hashToUnit, hiddenGrade, holm, htmlContainsElement, httpGithubClient, improvementVerdict, inMemoryExperimentStore, inMemoryReferenceReplayStore, inMemoryReviewStore, inMemoryRunRecordBackend, inMemoryVerdictCache, inferDomainKeywords, inferOtlpKind, interRaterReliability, interpretCliffs, iqr, isHiddenDestination, isJudgeSpan, isLlmSpan, isModelPriced, isOtelConfigured, isPolicyEdit, isRetrievalSpan, isRunRecord, isSandboxSpan, isToolSpan, isTransientLlmError, iterateRawCalls, jestTestParser, jsonHasKeys, jsonShape, jsonlReferenceReplayStore, jsonlReviewStore, jsonlRunRecordBackend, judgeFamily, judgeReplayGate, judgeSpans, keyPreserved, knowledgeReadinessTracePayload, leaderboard, linterJudge, llmJudge, llmSpanFromProvider, llmSpans, loadGoldScenarios, loadScorecard, loadScorerFromGrader, localCommandRunner, lowercaseMutator, makeEvalTools, makeFinding, makePolicyEdit, makePolicyEditCandidateRecord, mannWhitneyU, mapConcurrent, matchGoldens, matchSpan, maximumChargeForLlmRequest, mcnemar, mcnemarPower, mcnemarRequiredN, mergeLayerResults, mergeSteeringBundle, mintRolloutRows, modelDescriptionBits, modelHasSnapshot, modelPriceKey, mulberry32, multiToolchainLayer, noProgressDetector, normalizeScores, notBlocked, objectiveEval, observeAll, otelRunCompleteHook, otlpRowsToRunRecords, otlpRowsToTraceRunRecords, otlpToRunRecords, otlpToTraceRunRecords, pairArms, pairedBootstrap, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedSignTest, pairedTTest, paraphraseRobustness, paraphraseRobustnessScenarios, paretoChart, paretoFrontier, paretoFrontierWithCrowding, parseCorrectnessResponse, parseFeedbackTrajectoriesJsonl, parseGoldJsonl, parseReflectionResponse, parseRunRecordSafe, parseRuntimeTrajectoryHookEvent, partialCredit, partitionHeldOut, passAtK, passOrthogonality, pearsonR, pixelDeltaRatio, planTraceInsightQuestions, policyEditFromFinding, policyEditsFromFindings, politenessPrefixMutator, positionalBias, preflightModels, printDriverSummary, probeLlm, productBenchmarkIntegrityFailures, productBenchmarkMutableSurfaces, productBenchmarkRepoIdentity, productBenchmarkSplits, index as profile, projectOtlpFlatLine, projectRuntimeTrajectoryEvidence, promptBisect, proposeSynthesisTargets, providerFromBaseUrl, pytestTestParser, ranks, readOtlpStatus, readProductBenchmarkManifest, readProductBenchmarkRecords, recordRuns, recordRunsToScorecard, redTeamDataset, redTeamReport, redactString, redactValue, referenceReplayRunsToSteeringRows, referenceReplayScenarioToRunScore, regexMatch, regexMatches, renderMarkdownReport, renderPlaybookMarkdown, renderPreferenceMemoryMarkdown, renderPriorFindings, renderReleaseReport, renderSteeringText, renderUpstreamFindings, repeatedActionDetector, replayFeedbackTrajectories, replayFeedbackTrajectory, replayScorerOverCorpus, replayTraceThroughJudge, requireAgentProfileCell, requiredSampleSize, researchReport, resolveModelPricing, resolveRunCostProvenance, resolveSeat, rolloutReward, roundTripRunRecord, routeFields, rowCount, rowWhere, runAgentControlLoop, runAssertions, runBehavioralCanaries, runCanaries, runCounterfactual, runDistillation, runE2EWorkflow, runEvalCampaign, runExpectations, runFailureClass, runHarnessExperiment, runIntentMatchJudge, runJudgeFleet, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runLiveProof, runProposeReview, runProposeReviewAsControlLoop, runRecordToProductBenchmarkRecord, runReferenceEquivalenceJudge, runReferenceReplay, runScore, runSelfPlay, runSemanticConceptJudge, runTestGradedScenario, runsForScenario, scalarScore, scanForMuffledGates, scoreContinuity, scoreFromEvals, scoreKnowledgeReadiness, scorePolicyEditReadiness, scorePrReviewComments, scorePrReviewSource, scoreRedTeamOutput, scoreReferenceReplay, scoreTraceInsightReadiness, seatPresets, securityJudge, selectHarnessVariant, selfPreference, sentenceReorderMutator, serializeFeedbackTrajectoriesJsonl, signManifest, spearmanR, splitGold, statusAdvanced, stopOnNoProgress, stopOnRepeatedAction, stringField, stripFencedJson, subjectiveEval, summarizeAgentReceiptIntegrity, summarizeBackendIntegrity, summarizeHarnessResults, summarizePrReviewBenchmark, summarizePreferenceMemory, summaryTable, testJudge, textInSnapshot, throwIfRunIncomplete, toAgentProfileJson, toJsonl, toLangfuseEnvelope, toOpenAiTool, toPrometheusText, toRewardRows, toSftRows, tokenizeDomainWords, toolNamesForRun, toolSpans, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceContract, traceJudge, traceJudgeEnsemble, traceSpanKindToOpenInferenceKind, tracedAnalyzeTraces, typoMutator, urlContains, userQuestionsForKnowledgeGaps, validateAgentProfileCell, validatePolicyEdit, validatePolicyEditCandidateRecord, validateProductBenchmarkManifest, validateProductBenchmarkRecord, validateProductBenchmarkRun, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyAttestation, verifyCompletion, verifyManifest, visualDiff, viteDeployRunner, vitestTestParser, weightedComposite, weightedMean, weightedRecall, welchsTTest, whitespaceCollapseMutator, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit, withHeldoutBlend, withJudgeRetry, withOtelPipeline, wranglerDeployRunner };

package/dist/index.js

@@ -23,7 +23,7 @@ import {
   computeToolUseMetrics,
   iqr,
   welchsTTest
-} from "./chunk-QBRSJK47.js";
+} from "./chunk-J3LHTAAB.js";
 import {
   analyzeSeries
 } from "./chunk-BOD4O7OF.js";
@@ -53,7 +53,7 @@ import {
   pairArms,
   parseCorrectnessResponse,
   verifyCompletion
-} from "./chunk-LT4J7ULK.js";
+} from "./chunk-VPDOSN3L.js";
 import {
   DEFAULT_MUTATION_PRIMITIVES,
   DEFAULT_RED_TEAM_CORPUS,
@@ -91,7 +91,7 @@ import {
   scoreRedTeamOutput,
   surfaceContentHash,
   toolNamesForRun
-} from "./chunk-SUN7QLPB.js";
+} from "./chunk-KKPPFIDS.js";
 import {
   BackendIntegrityError,
   assertRealAgentReceipts,
@@ -3204,6 +3204,117 @@ function printDriverSummary(results) {
   console.log(`${completedCount}/${results.length} personas completed`);
 }
 
+// src/rollout-export.ts
+var ROLLOUT_FORMAT = "tangle.rollout.v1";
+var asText = (v, scrub) => {
+  const s = typeof v === "string" ? v : JSON.stringify(v);
+  return scrub(s ?? "");
+};
+function projectStep(span, scrub) {
+  const base = {
+    kind: span.kind,
+    name: scrub(span.name),
+    status: span.status,
+    durationMs: span.endedAt !== void 0 ? span.endedAt - span.startedAt : void 0
+  };
+  if (span.kind === "llm") {
+    const llm = span;
+    const last = llm.messages[llm.messages.length - 1];
+    if (last) base.input = scrub(last.content);
+    if (llm.output !== void 0) base.output = scrub(llm.output);
+  } else if (span.kind === "tool") {
+    const tool = span;
+    base.input = asText(tool.args, scrub);
+    if (tool.result !== void 0) base.output = asText(tool.result, scrub);
+  }
+  return base;
+}
+function scrubMessages(messages, scrub) {
+  return messages.map((m) => ({ ...m, content: scrub(m.content) }));
+}
+function finalConversation(spans, scrub) {
+  const llms = spans.filter((s) => s.kind === "llm");
+  const last = llms[llms.length - 1];
+  if (!last) return [];
+  const messages = scrubMessages(last.messages, scrub);
+  if (last.output !== void 0 && last.output !== "") {
+    messages.push({ role: "assistant", content: scrub(last.output) });
+  }
+  return messages;
+}
+function rolloutReward(record) {
+  const gated = record.outcome.realness?.gated === true;
+  const raw = record.outcome.holdoutScore ?? record.outcome.searchScore ?? 0;
+  return { reward: gated ? 0 : raw, gated };
+}
+async function mintRolloutRows(records, store, options = {}) {
+  const scrub = options.scrub ?? ((t) => t);
+  const rows = [];
+  const missingTraces = [];
+  for (const record of records) {
+    const trajectory = await buildTrajectory(store, record.runId);
+    if (trajectory.steps.length === 0) {
+      missingTraces.push(record.runId);
+      continue;
+    }
+    let steps = trajectory.steps.map((s) => projectStep(s.span, scrub));
+    if (options.maxSteps !== void 0 && steps.length > options.maxSteps) {
+      const head = Math.ceil(options.maxSteps / 2);
+      const tail = options.maxSteps - head;
+      steps = [...steps.slice(0, head), ...steps.slice(steps.length - tail)];
+    }
+    const { reward, gated } = rolloutReward(record);
+    rows.push({
+      format: ROLLOUT_FORMAT,
+      runId: record.runId,
+      experimentId: record.experimentId,
+      candidateId: record.candidateId,
+      scenarioId: record.scenarioId,
+      splitTag: record.splitTag,
+      model: record.model,
+      agentProfileCellId: record.agentProfile?.cellId,
+      reward,
+      realnessGated: gated,
+      costUsd: record.costUsd,
+      totalTokens: record.tokenUsage.input + record.tokenUsage.output,
+      steps,
+      conversation: finalConversation(
+        trajectory.steps.map((s) => s.span),
+        scrub
+      )
+    });
+  }
+  return { rows, missingTraces };
+}
+function toSftRows(rows, options = {}) {
+  const minReward = options.minReward ?? 1;
+  return rows.filter((r) => r.reward >= minReward && !r.realnessGated && r.conversation.length > 0).map((r) => ({
+    messages: r.conversation,
+    metadata: {
+      runId: r.runId,
+      candidateId: r.candidateId,
+      scenarioId: r.scenarioId,
+      reward: r.reward
+    }
+  }));
+}
+function toRewardRows(rows) {
+  return rows.map((r) => ({
+    prompt: r.conversation.find((m) => m.role === "user")?.content ?? "",
+    steps: r.steps,
+    reward: r.reward,
+    metadata: {
+      runId: r.runId,
+      candidateId: r.candidateId,
+      scenarioId: r.scenarioId,
+      splitTag: r.splitTag
+    }
+  }));
+}
+function toJsonl(rows) {
+  return rows.map((r) => JSON.stringify(r)).join("\n") + (rows.length ? "\n" : "");
+}
+
 // src/treatment-gate.ts
 function gateTreatmentApplied(input, opts = {}) {
   const failOpen = opts.failOpenWhenNoTelemetry ?? true;
@@ -11645,6 +11756,7 @@ export {
   REFERENCE_EQUIVALENCE_INPUT_LIMITS,
   REFERENCE_EQUIVALENCE_JUDGE_VERSION,
   RESEARCH_REPORT_HARD_PAIR_FLOOR,
+  ROLLOUT_FORMAT,
   RUN_COST_ATTR_KEYS,
   ReplayCache,
   ReplayCacheMissError,
@@ -11969,6 +12081,7 @@ export {
   mcnemarRequiredN,
   mergeLayerResults,
   mergeSteeringBundle,
+  mintRolloutRows,
   modelDescriptionBits,
   modelHasSnapshot,
   modelPriceKey,
@@ -12059,6 +12172,7 @@ export {
   resolveModelPricing,
   resolveRunCostProvenance,
   resolveSeat,
+  rolloutReward,
   roundTripRunRecord,
   routeFields,
   rowCount,
@@ -12125,9 +12239,12 @@ export {
   textInSnapshot,
   throwIfRunIncomplete,
   toAgentProfileJson,
+  toJsonl,
   toLangfuseEnvelope,
   toOpenAiTool,
   toPrometheusText,
+  toRewardRows,
+  toSftRows,
   tokenizeDomainWords,
   toolNamesForRun,
   toolSpans,