@tangle-network/agent-eval 0.31.1 → 0.32.0

package/dist/index.d.ts

@@ -392,6 +392,15 @@ interface FeedbackPattern {
     trigger: string;
     response: string;
 }
+/**
+ * How hard the simulated user pushes back. The driver LLM scales its tone
+ * and follow-up aggression to this:
+ *   cooperative — forgiving early adopter; accepts reasonable answers.
+ *   demanding   — experienced professional; rejects vague or hedged answers.
+ *   relentless  — senior partner reviewing for a client who will litigate;
+ *                 interrogates every claim, accepts nothing undefended.
+ */
+type PersonaRigor = 'cooperative' | 'demanding' | 'relentless';
 interface PersonaConfig {
     id: string;
     role: string;
@@ -400,6 +409,28 @@ interface PersonaConfig {
     feedbackPatterns?: FeedbackPattern[];
     maxTurns: number;
     driverModel?: string;
+    /** How adversarial the simulated user is. Defaults to 'demanding'. */
+    rigor?: PersonaRigor;
+    /**
+     * Domain expertise the simulated user holds — quoted into the driver
+     * prompt so it challenges the agent with authority instead of vague
+     * dissatisfaction. e.g. "a 15-year M&A partner who knows GAAP
+     * working-capital mechanics cold".
+     */
+    expertise?: string;
+    /**
+     * Substantive issues a senior professional in this role would
+     * interrogate — traps the scenario hides, claims that must be defended.
+     * The driver probes these without revealing them verbatim; the agent
+     * must surface them on its own.
+     */
+    pressurePoints?: string[];
+    /**
+     * Curveballs the driver may inject once the agent is coasting — changed
+     * facts, a hostile counterparty position, a new constraint. Forces the
+     * agent to re-derive rather than recite.
+     */
+    curveballs?: string[];
 }
 interface DriverState {
     tasks: number;
@@ -437,8 +468,16 @@ interface TurnMetrics {
 }
 interface DriverResult {
     personaId: string;
+    /** True when the simulated user professionally signed off (driver said DONE). */
     completed: boolean;
+    /** Turn at which the simulated user signed off, or null if it never did. */
     turnsToCompletion: number | null;
+    /**
+     * Turn at which nominal completionCriteria were first all met, or null.
+     * Distinct from turnsToCompletion: criteria can be met while the
+     * simulated professional is still unsatisfied with the work's rigor.
+     */
+    criteriaMetAtTurn: number | null;
     totalTurns: number;
     metrics: TurnMetrics[];
     finalState: DriverState;
@@ -1761,9 +1800,17 @@ declare class AgentDriver {
     private decideNextMessage;
     /** Handle pending approvals based on persona feedback patterns */
     private handleApprovals;
-    /** Describe which completion criteria are met */
-    private describeCompletion;
 }
+/**
+ * Build the driver LLM's system prompt. The simulated user is an
+ * adversarial senior professional: it judges the agent's last response by a
+ * professional standard, refuses vague answers, challenges undefended
+ * claims, probes the persona's pressure points without revealing them, and
+ * signs off (DONE) only when a real practitioner would act on the work
+ * unmodified. Pure function of persona, product state, and product context
+ * — exported so harness authors can inspect and regression-test it.
+ */
+declare function buildDriverSystemPrompt(persona: PersonaConfig, state: DriverState, productContext?: string): string;
 
 interface ExecutorConfig {
     /** System prompt for the agent under test */
@@ -2508,6 +2555,119 @@ declare function containsAll(name: string, required: string[], options?: {
     caseSensitive?: boolean;
 }): ArtifactValidator;
 
+/**
+ * Completion verifier — the task-completion oracle.
+ *
+ * Answers the only eval question that is not a proxy: did the agent actually
+ * COMPLETE the task — produce every required deliverable, persisted and
+ * correct — rather than describe what should be done. A fluent transcript
+ * that never produces the artifact scores zero here.
+ *
+ * Per requirement, a two-stage check:
+ *   1. Structural — a produced item (vault artifact / approved proposal /
+ *      tool call) of the right kind is matched against the requirement and
+ *      carries non-empty content. Deterministic; no LLM.
+ *   2. Correctness — only if structurally present AND the matched item
+ *      carries content, one targeted check decides whether that item
+ *      actually fulfils the requirement. A hallucinated artifact fails here;
+ *      an absent one already failed stage 1.
+ *
+ * `completionRate` is satisfied / total. Quality dimensions are meaningless
+ * on an incomplete task — callers gate on `fullyComplete` / `completionRate`
+ * before scoring quality.
+ */
+
+/** What kind of produced state can satisfy a requirement structurally. */
+type SatisfiedBy = 'artifact' | 'proposal' | 'tool-call' | 'any';
+interface CompletionRequirement {
+    /** Stable id from the task gold (e.g. a persona's `expected_requirements[].req_id`). */
+    reqId: string;
+    /** Human-readable description of the required deliverable. */
+    title: string;
+    /** Optional kind/category hint, matched against a produced item's kind. */
+    category?: string;
+    /** What produced state satisfies this requirement. Defaults to 'any'. */
+    satisfiedBy?: SatisfiedBy;
+}
+interface TaskGold {
+    taskId: string;
+    requirements: CompletionRequirement[];
+}
+interface ProducedProposal {
+    id: string;
+    title: string;
+    status: 'pending' | 'approved' | 'rejected';
+    /** Optional persisted body — when present, enables a correctness check. */
+    content?: string;
+}
+/** Everything observable about what a run actually produced. */
+interface ProducedState {
+    /** Persisted vault artifacts. Reuses the shared `Artifact` shape. */
+    artifacts: Artifact[];
+    /** Proposals / filings the agent created. */
+    proposals: ProducedProposal[];
+    /** Names of tools the agent invoked. */
+    toolCalls: string[];
+}
+interface RequirementCheck {
+    reqId: string;
+    title: string;
+    /** A produced item of the right kind matched the requirement, non-empty. */
+    structurallyPresent: boolean;
+    /**
+     * Whether the matched item actually fulfils the requirement. `null` when
+     * not structurally present, or when the matched item carries no content
+     * to assess.
+     */
+    correct: boolean | null;
+    /** structurallyPresent && correct !== false. */
+    satisfied: boolean;
+    /** Human-readable evidence for the verdict. */
+    evidence: string[];
+}
+interface CompletionVerdict {
+    taskId: string;
+    requirements: RequirementCheck[];
+    /** satisfied / total requirements. */
+    completionRate: number;
+    /** Every requirement satisfied. */
+    fullyComplete: boolean;
+}
+/**
+ * Decides whether a produced item's content actually fulfils a requirement.
+ * Injected so the structural verifier stays pure and unit-testable; the
+ * production implementation is `createLlmCorrectnessChecker`.
+ */
+type CorrectnessChecker = (requirement: CompletionRequirement, content: string) => Promise<{
+    correct: boolean;
+    reason: string;
+}>;
+/**
+ * Verify whether a run completed the task. `checkCorrectness` is injected —
+ * `createLlmCorrectnessChecker` for production, a deterministic stub in tests.
+ *
+ * Throws on a gold spec with no requirements: an eval task that requires
+ * nothing is a misconfiguration, not a vacuously-complete task.
+ */
+declare function verifyCompletion(gold: TaskGold, state: ProducedState, checkCorrectness: CorrectnessChecker): Promise<CompletionVerdict>;
+interface LlmCorrectnessCheckerOpts {
+    model?: string;
+    /** Max chars of artifact content sent to the checker. */
+    maxContentChars?: number;
+}
+/** Parse the correctness checker's model response. Fails loud on a bad shape. */
+declare function parseCorrectnessResponse(raw: string): {
+    correct: boolean;
+    reason: string;
+};
+/**
+ * Production `CorrectnessChecker` — one LLM call per matched artifact,
+ * deterministic (temperature 0), structured JSON out. Judges fulfilment
+ * only: a plan, a gesture, or a description of what should be done does not
+ * fulfil a requirement — the artifact must BE the deliverable.
+ */
+declare function createLlmCorrectnessChecker(tc: TCloud, opts?: LlmCorrectnessCheckerOpts): CorrectnessChecker;
+
 /**
  * ConvergenceTracker — tracks completion percentage over turns.
  *
@@ -3030,6 +3190,68 @@ declare function distillPlaybook(entries: PlaybookEntry[], options?: {
 }): Playbook;
 declare function renderPlaybookMarkdown(playbook: Playbook): string;
 
+/**
+ * Produced-state extraction — normalize a run's runtime event stream into the
+ * typed `ProducedState` the completion oracle consumes.
+ *
+ * `ProducedState` answers "what did the agent actually produce" — vault
+ * artifacts, proposals, tool calls. The runtime emits these as a stream of
+ * events; this module is the single normalization point from that stream to
+ * the shape `verifyCompletion` expects.
+ *
+ * Input is structurally typed (`RuntimeEventLike`) so this module does not
+ * depend on agent-runtime — agent-runtime's `RuntimeStreamEvent` satisfies it
+ * structurally. The `content` on `ArtifactEventLike` and the whole
+ * `proposal_created` variant are the runtime-side enrichments this contract
+ * requires; the runtime emits them, this module consumes them.
+ */
+
+/** A tool the agent invoked. */
+interface ToolCallEventLike {
+    type: 'tool_call';
+    toolName: string;
+}
+/**
+ * An artifact the agent produced. `content` is the enriched field — the
+ * runtime's base `artifact` event carries only metadata; the completion
+ * oracle needs the body to verify the deliverable, so the runtime emits it.
+ */
+interface ArtifactEventLike {
+    type: 'artifact';
+    artifactId: string;
+    name?: string;
+    mimeType?: string;
+    uri?: string;
+    content?: string;
+}
+/** A proposal / filing the agent created. */
+interface ProposalEventLike {
+    type: 'proposal_created';
+    proposalId: string;
+    title: string;
+    status?: 'pending' | 'approved' | 'rejected';
+}
+/**
+ * The subset of runtime stream events `extractProducedState` consumes.
+ * agent-runtime's full `RuntimeStreamEvent` union satisfies this structurally;
+ * the `{ type: string }` catch-all keeps the input permissive so callers can
+ * pass the whole unfiltered telemetry stream — unrecognized events are skipped.
+ */
+type RuntimeEventLike = ToolCallEventLike | ArtifactEventLike | ProposalEventLike | {
+    type: string;
+};
+/**
+ * Normalize a run's runtime event stream into `ProducedState`.
+ *
+ * Pure and total — unrecognized event types are skipped. `toolCalls` is
+ * deduplicated by name in first-seen order (completion cares about a tool's
+ * presence, not its call count). An artifact with neither a name nor a uri
+ * still yields an entry keyed by its `artifactId` so it is never silently
+ * dropped; an artifact with no `content` yields empty content, which the
+ * completion oracle's structural check then rejects on its own.
+ */
+declare function extractProducedState(events: readonly RuntimeEventLike[]): ProducedState;
+
 /**
  * Versioned prompt registry.
  *
@@ -6243,4 +6465,4 @@ declare function aggregateTrialsByMode(trials: TrialResult[], opts: {
     mode: AggregatorMode;
 }): TrialAggregate;
 
-export { ANALYST_SEVERITIES, ActionableSideInfo, type ActiveLearningOptions, type AdapterRun, AgentDriver, type AgentDriverConfig, AgentEvalError, type AggregatorMode, 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, AnalyzeTracesOptions, type AntiSlopConfig, type AntiSlopIssue, type AntiSlopReport, Artifact$1 as Artifact, type ArtifactCheck, type Artifact as ArtifactCheckArtifact, type ArtifactResult, type ArtifactValidator, type AutoPrClient, AxGepaSteeringOptimizer, type AxSteeringOptimizerConfig, BackendIntegrityError, type BackendIntegrityReport, BaselineReport, BehaviorAssertion, type BenchmarkReport, BenchmarkRunner, type BenchmarkRunnerConfig, type BisectOptions, type BisectResult, type BisectStep, BudgetBreachError, BudgetGuard, BudgetLedgerEntry, type BudgetPolicy, BudgetSpec, CallExpectation, type CanaryAlert, type CanaryKind, type CanaryLeak, type CanaryOptions, type CanaryReport, type CanarySeverity, type CandidateScenario, type CausalAttributionReport, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTransport, type CheckResult, type CliBridgeTransportOpts, type CodeMutationOutcome, type CodeMutationRunner, type CollectedArtifacts, type CommandRunner, type CompletionCriterion, type CompositePolicy, type ConceptComplexity, type ConceptFinding, type ConceptSpec, type ConceptWeightStrategy, type ContinuityCheck, type ContinuityCheckResult, type ContinuityReport, type ContinuitySnapshotPair, ContinuousAgreement, ContinuousAgreementOptions, type ContractMetric, type ContractReport, ControlEvalResult, ConvergenceTracker, type CorpusAgreementOptions, type CorpusAgreementPerDimension, type CorpusAgreementReport, type CorpusScoreRecord, type CostEntry, CostLedger, type CostLedgerGeneration, type CostLedgerSnapshot, type CostSummary, CostTracker, type CounterfactualContext, type CounterfactualMutation, type CounterfactualResult, type CounterfactualRunner, type CreateChatClientOpts, type CreateCompositeMutatorOpts, type CreateDefaultReviewerOptions, type CreateSandboxCodeMutatorOpts, type CreateSandboxPoolOpts, type CreateTraceAnalystKindOpts, type CrossTraceDiff, type CrossTraceDiffOptions, D1ExperimentStore, type D1ExperimentStoreOptions, type D1Like, type D1PreparedStatementLike, DEFAULT_AGENT_SLOS, DEFAULT_COMPLEXITY_WEIGHTS, DEFAULT_FINDERS, DEFAULT_HARNESS_OBJECTIVES, DEFAULT_MUTATORS, DEFAULT_RUN_SCORE_WEIGHTS, DEFAULT_SEVERITY_WEIGHTS, DEFAULT_TRACE_ANALYST_KINDS, Dataset, DatasetScenario, type DeployFamily, type DeployGateLayerInput, type DeployRunResult, type DeployRunner, type DiffPolicy, type DirEntry, type DirectProviderTransportOpts, type DiscoverPersonasOptions, type DiscoveredPersona, type DriverResult, type DriverState, DualAgentBench, type DualAgentBenchConfig, type DualAgentReport, type DualAgentRound, type DualAgentScenario, type DualAgentScenarioResult, ERROR_COUNT_PATTERNS, type ErrorCountPattern, type EvalResult, type EvidenceRef, type EvolutionRound, EvolvableVariant, type ExecutorConfig, type Expectation, type Experiment, type Run as ExperimentRun, type ExperimentStore, ExperimentTracker, type ExportedRewardModel, type ExtractOptions, type ExtractResult, FAILURE_MODE_KIND_SPEC, FINDING_SUBJECT_GRAMMAR_PROMPT, FINDING_SUBJECT_KINDS, type FactorContribution, type FactorialCell, type FailureClusterConfig, FeedbackLabel, type FeedbackPattern, FeedbackTrajectory, FeedbackTrajectoryStore, type FileChange, FileSystemExperimentStore, type FileSystemExperimentStoreOptions, type FindingSubject, type FindingSubjectKind, FindingSubjectStringSchema, type FindingsDiff, FindingsStore, type FlowAction, type FlowLayerEnv, type FlowLayerFactoryInput, type FlowRunner, type FlowRunnerStepResult, type FlowSpec, type FlowStep, GateDecision, type GhCliClientOptions, type GoldenSeverity, type GoldenSpec, type HarnessAdapter, HarnessConfig, type HarnessExperimentConfig, type HarnessExperimentResult, type HarnessIntervention, type HarnessRunRequest, type HarnessRunResult, type HarnessScenario, type HarnessSelection, type HarnessVariant, type HarnessVariantReport, HeldOutGateConfig, HoldoutAuditor, type HostedJudgeConfig, type HostedJudgeDimension, type HostedJudgeRequest, type HostedJudgeResponse, type HostedRunCriticConfig, type HostedRunScoreRequest, type HostedRunScoreResponse, type HttpGithubClientOptions, type HypothesisManifest, type HypothesisResult, IMPROVEMENT_KIND_SPEC, INTENT_MATCH_JUDGE_VERSION, type ImageData, InMemoryExperimentStore, InMemoryWorkspaceInspector, type InferenceScorer, type InspectorContext, type IntegrationGateSurface, type IntegrationInvokeFailureInput, type IntegrationManifestGateInput, type IntentMatchInput, type IntentMatchOptions, type IntentMatchResult, type InteractionContribution, JsonlTrialCache, type JudgeAdapterOpts, type JudgeConfig, type JudgeFleetOptions, type JudgeFn, type JudgeInput, type JudgeReplayResult, type JudgeRetryOutcome, type JudgeRetryPolicy, type JudgeRubric, JudgeRunner, type JudgeScore, KIND_EXPECTED_SUBJECTS, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, type KeywordConceptSpec, type KeywordCoverageFinding, type KeywordCoverageOptions, type KeywordCoverageResult, type LangfuseEnvelope, type LangfuseGeneration, type LangfuseScore, Layer, LayerResult, type LineageKind, type LineageKindResolver, type LineageNode, LineageRecorder, type LiveProofArtifact, type LiveProofConfig, type LiveProofContext, type LiveProofResult, LlmCallRequest, LlmCallResult, LlmClientOptions, LlmSpan, LockedJsonlAppender, MODEL_PRICING, type MatchResult, type MatcherResult, type MeasurementPolicy, type MergeOptions, MetricsCollector, type MockTransportOpts, type MuffledFinder, type MuffledFinding, MultiLayerVerifier, MultiShotMutateAdapter, MultiShotOptimizationResult, MultiShotRunner, MultiShotScorer, MultiShotTrialResult, type MultiToolchainLayerConfig, MutateAdapter, type MutationAttempt, type MutationChannel, MutationTelemetry, type Mutator, Mutex, Objective, type Oracle, type OracleObservation, type OracleReport, type OracleResult, type OrthogonalityInput, type OrthogonalityResult, PairwiseSteeringOptimizer, type ParaphraseRobustnessScenarioInput, type ParaphraseRobustnessScenarioResult, ParetoResult, type PersistedFinding, type PersonaConfig, type Playbook, type PlaybookEntry, type PoolSlot, ProductClient, type ProductClientConfig, type ProductionEvolveConfig, type ProductionLoopCronConfig, type ProductionLoopDecision, type ProductionLoopRenderContext, type ProductionLoopResult, type ProductionShipConfig, type PromptHandle, PromptRegistry, TrialResult as PromptTrialResult, type ProposeAutomatedPullRequestInput, type ProposeAutomatedPullRequestResult, RAW_FINDING_SCHEMA_PROMPT, type RawAnalystFinding, RawAnalystFindingSchema, 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 RegistryRunOpts, ReleaseConfidenceScorecard, ReleaseConfidenceThresholds, type RepoRef, type ReviewerMemoryEntry, type ReviewerOutput, type ReviewerPromptInput, type ReviewerSoftFailDefaults, type ReviewerVerificationSummary, type RobustnessResult, type RouteMap, type RouterTransportOpts, type RubricDimension, Run$1 as Run, type RunCommandInput, type RunCommandResult, type RunConfig, RunCritic, type RunCriticAdapterOpts, type RunCriticOptions, type RunDiff, RunFilter, type RunProductionLoopOptions, RunRecord, type RunScore, type RunScoreWeights, RunSplitTag, type RunTrace, SEMANTIC_CONCEPT_JUDGE_VERSION, SandboxDriver, SandboxHarnessResult, type SandboxJudgeKind, type SandboxJudgeResult, type SandboxJudgeSpec, type SandboxPool, type SandboxSdkTransportOpts, type ScanOptions, type Scenario, type ScenarioCost, type ScenarioFile, ScenarioRegistry, type ScenarioResult, type ScoredTarget, type SelfPlayOptions, type SelfPlayProposer, type SelfPlayScorer, type SemanticConceptJudgeAdapterOpts, type SemanticConceptJudgeInput, type SemanticConceptJudgeOptions, type SemanticConceptJudgeResult, type SeriesConvergenceOptions, type SeriesConvergenceResult, Severity, type SignedManifest, type SignedManifestAlgo, type Slo, type SloCheckResult, type SloComparator, type SloReport, type SloSeverity, type SlopCategory, type SlotFactory, Span, type SteeringBundle, type SteeringDelta, type SteeringOptimizationResult, type SteeringOptimizationRow, type SteeringOptimizationSelector, type SteeringOptimizerBackend, type SteeringOptimizerConfig, type SteeringRolePrompt, type StepAttribution, type SynthesisReason, type SynthesisTarget, type TestResult, type ThresholdContract, TokenCounter, type TokenSpec, TraceAnalysisStore, type TraceAnalystAdapterOpts, type TraceAnalystGolden, type TraceAnalystKindSpec, TraceEmitter, TraceEvent, TraceStore, type TraceToolGroupName, Trajectory, TrajectoryStep, type TrialAggregate, type TrialAttempt, TrialCache, TrialTelemetry, type Turn, type TurnMetrics, type TurnResult, UNIVERSAL_FINDERS, type ValidationContext, type ValidationIssue, type ValidationResult, VariantAggregate, type VerifierAdapterOpts, VerifyContext, VerifyOptions, type VisualDiffOptions, type VisualDiffResult, type ViteDeployRunnerInput, type WorkflowTopology, type WorkspaceAssertion, type WorkspaceAssertionResult, type WorkspaceInspector, type WorkspaceSnapshot, type WranglerDeployRunnerInput, adversarialJudge, aggregateRunScore, aggregateTrialsByMode, analyzeAntiSlop, analyzeSeries, assertRealBackend, attributeCounterfactuals, benjaminiHochberg, bisect, bonferroni, buildReviewerPrompt, buildTraceToolsForGroup, byteLengthRange, canaryLeakView, canonicalize, causalAttribution, checkBehavioralCanary, checkCanaries, checkSlos, clamp01, codeExecutionJudge, cohensD, coherenceJudge, collectionPreserved, commitBisect, compareReferenceReplay, compilerJudge, composeValidators, computeFindingId, confidenceInterval, containsAll, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, createAntiSlopJudge, createChatClient, createCompositeMutator, createCustomJudge, createDefaultReviewer, createDomainExpertJudge, createIntentMatchJudge, createJudgeAdapter, createRunCriticAdapter, createSandboxCodeMutator, createSandboxPool, createSemanticConceptJudge, createSemanticConceptJudgeAdapter, createTraceAnalystAdapter, createTraceAnalystKind, createVerifierAdapter, crossTraceDiff, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultIsMaterial, defaultJudges, defaultReferenceReplayMatcher, deployGateLayer, diffFindings, discoverPersonas, distillPlaybook, estimateCost, estimateTokens, evaluateContract, evaluateHypothesis, evaluateOracles, executeScenario, expectAgent, exportRewardModel, extractAssetUrls, extractErrorCount, fileContains, fileExists, findAutoMatchNoExpectation, findConstructorCwdDropped, findFallbackToPass, findLiteralTruePass, findSkipCountsAsPass, flowLayer, formatBenchmarkReport, formatDriverReport, formatFindings, ghCliClient, precision as goldenPrecision, hashContent, hashJson, htmlContainsElement, httpGithubClient, inMemoryReferenceReplayStore, integrationAsi, integrationGateEvals, integrationInvokeFailedPayload, integrationManifestResolvedPayload, integrationManifestValidatedPayload, interRaterReliability, jsonHasKeys, jsonShape, jsonlReferenceReplayStore, keyPreserved, liftSeverity, linterJudge, loadScorerFromGrader, localCommandRunner, lowercaseMutator, makeFinding, mannWhitneyU, matchGoldens, mergeLayerResults, mergeSteeringBundle, multiToolchainLayer, normalizeScores, notBlocked, pairedTTest, paraphraseRobustness, paraphraseRobustnessScenarios, parseFindingSubject, parseRawFinding, partialCredit, passOrthogonality, pixelDeltaRatio, politenessPrefixMutator, printDriverSummary, promptBisect, proposeAutomatedPullRequest, proposeSynthesisTargets, referenceReplayRunsToSteeringRows, referenceReplayScenarioToRunScore, regexMatch, regexMatches, renderFindingSubject, renderMarkdownReport, renderPlaybookMarkdown, renderPriorFindings, renderSteeringText, replayScorerOverCorpus, replayTraceThroughJudge, requiredSampleSize, resetLockedAppendersForTesting, rowCount, rowWhere, runAssertions, runBehavioralCanaries, runCanaries, runCounterfactual, runE2EWorkflow, runExpectations, runHarnessExperiment, runIntentMatchJudge, runJudgeFleet, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runLiveProof, runProductionLoop, runReferenceReplay, runSelfPlay, runSemanticConceptJudge, scanForMuffledGates, scoreContinuity, scoreReferenceReplay, securityJudge, selectHarnessVariant, sentenceReorderMutator, signManifest, statusAdvanced, summarizeBackendIntegrity, summarizeHarnessResults, testJudge, textInSnapshot, toLangfuseEnvelope, toPrometheusText, typoMutator, urlContains, verifyManifest, visualDiff, viteDeployRunner, weightedMean, weightedRecall, whitespaceCollapseMutator, wilcoxonSignedRank, withJudgeRetry, wranglerDeployRunner };
+export { ANALYST_SEVERITIES, ActionableSideInfo, type ActiveLearningOptions, type AdapterRun, AgentDriver, type AgentDriverConfig, AgentEvalError, type AggregatorMode, 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, AnalyzeTracesOptions, type AntiSlopConfig, type AntiSlopIssue, type AntiSlopReport, Artifact$1 as Artifact, type ArtifactCheck, type Artifact as ArtifactCheckArtifact, type ArtifactEventLike, type ArtifactResult, type ArtifactValidator, type AutoPrClient, AxGepaSteeringOptimizer, type AxSteeringOptimizerConfig, BackendIntegrityError, type BackendIntegrityReport, BaselineReport, BehaviorAssertion, type BenchmarkReport, BenchmarkRunner, type BenchmarkRunnerConfig, type BisectOptions, type BisectResult, type BisectStep, BudgetBreachError, BudgetGuard, BudgetLedgerEntry, type BudgetPolicy, BudgetSpec, CallExpectation, type CanaryAlert, type CanaryKind, type CanaryLeak, type CanaryOptions, type CanaryReport, type CanarySeverity, type CandidateScenario, type CausalAttributionReport, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTransport, type CheckResult, type CliBridgeTransportOpts, type CodeMutationOutcome, type CodeMutationRunner, type CollectedArtifacts, type CommandRunner, type CompletionCriterion, type CompletionRequirement, type CompletionVerdict, type CompositePolicy, type ConceptComplexity, type ConceptFinding, type ConceptSpec, type ConceptWeightStrategy, type ContinuityCheck, type ContinuityCheckResult, type ContinuityReport, type ContinuitySnapshotPair, ContinuousAgreement, ContinuousAgreementOptions, type ContractMetric, type ContractReport, ControlEvalResult, ConvergenceTracker, type CorpusAgreementOptions, type CorpusAgreementPerDimension, type CorpusAgreementReport, type CorpusScoreRecord, type CorrectnessChecker, type CostEntry, CostLedger, type CostLedgerGeneration, type CostLedgerSnapshot, type CostSummary, CostTracker, type CounterfactualContext, type CounterfactualMutation, type CounterfactualResult, type CounterfactualRunner, type CreateChatClientOpts, type CreateCompositeMutatorOpts, type CreateDefaultReviewerOptions, type CreateSandboxCodeMutatorOpts, type CreateSandboxPoolOpts, type CreateTraceAnalystKindOpts, type CrossTraceDiff, type CrossTraceDiffOptions, D1ExperimentStore, type D1ExperimentStoreOptions, type D1Like, type D1PreparedStatementLike, DEFAULT_AGENT_SLOS, DEFAULT_COMPLEXITY_WEIGHTS, DEFAULT_FINDERS, DEFAULT_HARNESS_OBJECTIVES, DEFAULT_MUTATORS, DEFAULT_RUN_SCORE_WEIGHTS, DEFAULT_SEVERITY_WEIGHTS, DEFAULT_TRACE_ANALYST_KINDS, Dataset, DatasetScenario, type DeployFamily, type DeployGateLayerInput, type DeployRunResult, type DeployRunner, type DiffPolicy, type DirEntry, type DirectProviderTransportOpts, type DiscoverPersonasOptions, type DiscoveredPersona, type DriverResult, type DriverState, DualAgentBench, type DualAgentBenchConfig, type DualAgentReport, type DualAgentRound, type DualAgentScenario, type DualAgentScenarioResult, ERROR_COUNT_PATTERNS, type ErrorCountPattern, type EvalResult, type EvidenceRef, type EvolutionRound, EvolvableVariant, type ExecutorConfig, type Expectation, type Experiment, type Run as ExperimentRun, type ExperimentStore, ExperimentTracker, type ExportedRewardModel, type ExtractOptions, type ExtractResult, FAILURE_MODE_KIND_SPEC, FINDING_SUBJECT_GRAMMAR_PROMPT, FINDING_SUBJECT_KINDS, type FactorContribution, type FactorialCell, type FailureClusterConfig, FeedbackLabel, type FeedbackPattern, FeedbackTrajectory, FeedbackTrajectoryStore, type FileChange, FileSystemExperimentStore, type FileSystemExperimentStoreOptions, type FindingSubject, type FindingSubjectKind, FindingSubjectStringSchema, type FindingsDiff, FindingsStore, type FlowAction, type FlowLayerEnv, type FlowLayerFactoryInput, type FlowRunner, type FlowRunnerStepResult, type FlowSpec, type FlowStep, GateDecision, type GhCliClientOptions, type GoldenSeverity, type GoldenSpec, type HarnessAdapter, HarnessConfig, type HarnessExperimentConfig, type HarnessExperimentResult, type HarnessIntervention, type HarnessRunRequest, type HarnessRunResult, type HarnessScenario, type HarnessSelection, type HarnessVariant, type HarnessVariantReport, HeldOutGateConfig, HoldoutAuditor, type HostedJudgeConfig, type HostedJudgeDimension, type HostedJudgeRequest, type HostedJudgeResponse, type HostedRunCriticConfig, type HostedRunScoreRequest, type HostedRunScoreResponse, type HttpGithubClientOptions, type HypothesisManifest, type HypothesisResult, IMPROVEMENT_KIND_SPEC, INTENT_MATCH_JUDGE_VERSION, type ImageData, InMemoryExperimentStore, InMemoryWorkspaceInspector, type InferenceScorer, type InspectorContext, type IntegrationGateSurface, type IntegrationInvokeFailureInput, type IntegrationManifestGateInput, type IntentMatchInput, type IntentMatchOptions, type IntentMatchResult, type InteractionContribution, JsonlTrialCache, type JudgeAdapterOpts, type JudgeConfig, type JudgeFleetOptions, type JudgeFn, type JudgeInput, type JudgeReplayResult, type JudgeRetryOutcome, type JudgeRetryPolicy, type JudgeRubric, JudgeRunner, type JudgeScore, KIND_EXPECTED_SUBJECTS, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, type KeywordConceptSpec, type KeywordCoverageFinding, type KeywordCoverageOptions, type KeywordCoverageResult, type LangfuseEnvelope, type LangfuseGeneration, type LangfuseScore, Layer, LayerResult, type LineageKind, type LineageKindResolver, type LineageNode, LineageRecorder, type LiveProofArtifact, type LiveProofConfig, type LiveProofContext, type LiveProofResult, LlmCallRequest, LlmCallResult, LlmClientOptions, type LlmCorrectnessCheckerOpts, LlmSpan, LockedJsonlAppender, MODEL_PRICING, type MatchResult, type MatcherResult, type MeasurementPolicy, type MergeOptions, MetricsCollector, type MockTransportOpts, type MuffledFinder, type MuffledFinding, MultiLayerVerifier, MultiShotMutateAdapter, MultiShotOptimizationResult, MultiShotRunner, MultiShotScorer, MultiShotTrialResult, type MultiToolchainLayerConfig, MutateAdapter, type MutationAttempt, type MutationChannel, MutationTelemetry, type Mutator, Mutex, Objective, type Oracle, type OracleObservation, type OracleReport, type OracleResult, type OrthogonalityInput, type OrthogonalityResult, PairwiseSteeringOptimizer, type ParaphraseRobustnessScenarioInput, type ParaphraseRobustnessScenarioResult, ParetoResult, type PersistedFinding, type PersonaConfig, type PersonaRigor, type Playbook, type PlaybookEntry, type PoolSlot, type ProducedProposal, type ProducedState, ProductClient, type ProductClientConfig, type ProductionEvolveConfig, type ProductionLoopCronConfig, type ProductionLoopDecision, type ProductionLoopRenderContext, type ProductionLoopResult, type ProductionShipConfig, type PromptHandle, PromptRegistry, TrialResult as PromptTrialResult, type ProposalEventLike, type ProposeAutomatedPullRequestInput, type ProposeAutomatedPullRequestResult, RAW_FINDING_SCHEMA_PROMPT, type RawAnalystFinding, RawAnalystFindingSchema, 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 RegistryRunOpts, ReleaseConfidenceScorecard, ReleaseConfidenceThresholds, type RepoRef, type RequirementCheck, type ReviewerMemoryEntry, type ReviewerOutput, type ReviewerPromptInput, type ReviewerSoftFailDefaults, type ReviewerVerificationSummary, type RobustnessResult, type RouteMap, type RouterTransportOpts, type RubricDimension, Run$1 as Run, type RunCommandInput, type RunCommandResult, type RunConfig, RunCritic, type RunCriticAdapterOpts, type RunCriticOptions, type RunDiff, RunFilter, type RunProductionLoopOptions, RunRecord, type RunScore, type RunScoreWeights, RunSplitTag, type RunTrace, type RuntimeEventLike, SEMANTIC_CONCEPT_JUDGE_VERSION, SandboxDriver, SandboxHarnessResult, type SandboxJudgeKind, type SandboxJudgeResult, type SandboxJudgeSpec, type SandboxPool, type SandboxSdkTransportOpts, type SatisfiedBy, type ScanOptions, type Scenario, type ScenarioCost, type ScenarioFile, ScenarioRegistry, type ScenarioResult, type ScoredTarget, type SelfPlayOptions, type SelfPlayProposer, type SelfPlayScorer, type SemanticConceptJudgeAdapterOpts, type SemanticConceptJudgeInput, type SemanticConceptJudgeOptions, type SemanticConceptJudgeResult, type SeriesConvergenceOptions, type SeriesConvergenceResult, Severity, type SignedManifest, type SignedManifestAlgo, type Slo, type SloCheckResult, type SloComparator, type SloReport, type SloSeverity, type SlopCategory, type SlotFactory, Span, type SteeringBundle, type SteeringDelta, type SteeringOptimizationResult, type SteeringOptimizationRow, type SteeringOptimizationSelector, type SteeringOptimizerBackend, type SteeringOptimizerConfig, type SteeringRolePrompt, type StepAttribution, type SynthesisReason, type SynthesisTarget, type TaskGold, type TestResult, type ThresholdContract, TokenCounter, type TokenSpec, type ToolCallEventLike, TraceAnalysisStore, type TraceAnalystAdapterOpts, type TraceAnalystGolden, type TraceAnalystKindSpec, TraceEmitter, TraceEvent, TraceStore, type TraceToolGroupName, Trajectory, TrajectoryStep, type TrialAggregate, type TrialAttempt, TrialCache, TrialTelemetry, type Turn, type TurnMetrics, type TurnResult, UNIVERSAL_FINDERS, type ValidationContext, type ValidationIssue, type ValidationResult, VariantAggregate, type VerifierAdapterOpts, VerifyContext, VerifyOptions, type VisualDiffOptions, type VisualDiffResult, type ViteDeployRunnerInput, type WorkflowTopology, type WorkspaceAssertion, type WorkspaceAssertionResult, type WorkspaceInspector, type WorkspaceSnapshot, type WranglerDeployRunnerInput, adversarialJudge, aggregateRunScore, aggregateTrialsByMode, analyzeAntiSlop, analyzeSeries, assertRealBackend, attributeCounterfactuals, benjaminiHochberg, bisect, bonferroni, buildDriverSystemPrompt, buildReviewerPrompt, buildTraceToolsForGroup, byteLengthRange, canaryLeakView, canonicalize, causalAttribution, checkBehavioralCanary, checkCanaries, checkSlos, clamp01, codeExecutionJudge, cohensD, coherenceJudge, collectionPreserved, commitBisect, compareReferenceReplay, compilerJudge, composeValidators, computeFindingId, confidenceInterval, containsAll, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, createAntiSlopJudge, createChatClient, createCompositeMutator, createCustomJudge, createDefaultReviewer, createDomainExpertJudge, createIntentMatchJudge, createJudgeAdapter, createLlmCorrectnessChecker, createRunCriticAdapter, createSandboxCodeMutator, createSandboxPool, createSemanticConceptJudge, createSemanticConceptJudgeAdapter, createTraceAnalystAdapter, createTraceAnalystKind, createVerifierAdapter, crossTraceDiff, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultIsMaterial, defaultJudges, defaultReferenceReplayMatcher, deployGateLayer, diffFindings, discoverPersonas, distillPlaybook, estimateCost, estimateTokens, evaluateContract, evaluateHypothesis, evaluateOracles, executeScenario, expectAgent, exportRewardModel, extractAssetUrls, extractErrorCount, extractProducedState, fileContains, fileExists, findAutoMatchNoExpectation, findConstructorCwdDropped, findFallbackToPass, findLiteralTruePass, findSkipCountsAsPass, flowLayer, formatBenchmarkReport, formatDriverReport, formatFindings, ghCliClient, precision as goldenPrecision, hashContent, hashJson, htmlContainsElement, httpGithubClient, inMemoryReferenceReplayStore, integrationAsi, integrationGateEvals, integrationInvokeFailedPayload, integrationManifestResolvedPayload, integrationManifestValidatedPayload, interRaterReliability, jsonHasKeys, jsonShape, jsonlReferenceReplayStore, keyPreserved, liftSeverity, linterJudge, loadScorerFromGrader, localCommandRunner, lowercaseMutator, makeFinding, mannWhitneyU, matchGoldens, mergeLayerResults, mergeSteeringBundle, multiToolchainLayer, normalizeScores, notBlocked, pairedTTest, paraphraseRobustness, paraphraseRobustnessScenarios, parseCorrectnessResponse, parseFindingSubject, parseRawFinding, partialCredit, passOrthogonality, pixelDeltaRatio, politenessPrefixMutator, printDriverSummary, promptBisect, proposeAutomatedPullRequest, proposeSynthesisTargets, referenceReplayRunsToSteeringRows, referenceReplayScenarioToRunScore, regexMatch, regexMatches, renderFindingSubject, renderMarkdownReport, renderPlaybookMarkdown, renderPriorFindings, renderSteeringText, replayScorerOverCorpus, replayTraceThroughJudge, requiredSampleSize, resetLockedAppendersForTesting, rowCount, rowWhere, runAssertions, runBehavioralCanaries, runCanaries, runCounterfactual, runE2EWorkflow, runExpectations, runHarnessExperiment, runIntentMatchJudge, runJudgeFleet, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runLiveProof, runProductionLoop, runReferenceReplay, runSelfPlay, runSemanticConceptJudge, scanForMuffledGates, scoreContinuity, scoreReferenceReplay, securityJudge, selectHarnessVariant, sentenceReorderMutator, signManifest, statusAdvanced, summarizeBackendIntegrity, summarizeHarnessResults, testJudge, textInSnapshot, toLangfuseEnvelope, toPrometheusText, typoMutator, urlContains, verifyCompletion, verifyManifest, visualDiff, viteDeployRunner, weightedMean, weightedRecall, whitespaceCollapseMutator, wilcoxonSignedRank, withJudgeRetry, wranglerDeployRunner };