@tangle-network/agent-eval 0.21.0 → 0.22.0

package/dist/index.d.ts

@@ -1,22 +1,24 @@
 import { TCloud } from '@tangle-network/tcloud';
-import { R as ReleaseConfidenceThresholds, a as ReleaseConfidenceScorecard } from './reporting-Da2ihlcM.js';
-export { B as BootstrapOptions, b as BootstrapResult, D as DEFAULT_FAILURE_RULES, F as FailureClassification, c as FailureCluster, d as FailureClusterReport, e as FailureContext, f as FailureRule, G as GainDistributionBin, g as GainDistributionFigureSpec, h as GainDistributionOptions, J as JudgeReplayGateArgs, P as PairedBootstrapOptions, i as PairedBootstrapResult, j as ParetoFigureSpec, k as ParetoPoint, l as RESEARCH_REPORT_HARD_PAIR_FLOOR, m as ReleaseConfidenceAxis, n as ReleaseConfidenceAxisName, o as ReleaseConfidenceInput, p as ReleaseConfidenceIssue, q as ReleaseConfidenceMetrics, r as ReleaseConfidenceStatus, s as ReleaseTraceEvidence, t as RenderReleaseReportOptions, u as ResearchReport, v as ResearchReportCandidate, w as ResearchReportDecision, x as ResearchReportMethodology, y as ResearchReportOptions, z as ResearchReportRecommendation, S as SummaryTable, A as SummaryTableOptions, C as SummaryTableRow, V as Verdict, E as assertReleaseConfidence, H as bhAdjust, I as bootstrapCi, K as classifyFailure, L as evaluateReleaseConfidence, M as failureClusterView, N as gainHistogram, O as judgeReplayGate, Q as pairedBootstrap, T as pairedWilcoxon, U as paretoChart, W as releaseTraceEvidenceFromMultiShotTrials, X as renderReleaseReport, Y as researchReport, Z as summaryTable } from './reporting-Da2ihlcM.js';
+import { R as ReleaseConfidenceThresholds, a as ReleaseConfidenceScorecard, O as OutcomeFilter, b as OutcomeStore } from './reporting-B82RSv9C.js';
+export { B as BootstrapOptions, c as BootstrapResult, D as DeploymentOutcome, F as FileSystemOutcomeStore, d as FileSystemOutcomeStoreOptions, I as InMemoryOutcomeStore, e as InterimReleaseConfidence, f as InterimReleaseConfidenceInput, J as JudgeReplayGateArgs, P as PairedBootstrapOptions, g as PairedBootstrapResult, h as PairedEvalueOptions, i as PairedEvalueSequence, j as PairedEvalueStep, k as ReleaseConfidenceAxis, l as ReleaseConfidenceAxisName, m as ReleaseConfidenceInput, n as ReleaseConfidenceIssue, o as ReleaseConfidenceMetrics, p as ReleaseConfidenceStatus, q as ReleaseTraceEvidence, r as RenderReleaseReportOptions, s as RubricOutcomePair, t as RubricPredictiveValidityInput, u as RubricPredictiveValidityReport, v as RubricRanking, S as SequentialDecision, V as Verdict, w as assertReleaseConfidence, x as bhAdjust, y as bootstrapCi, z as evaluateInterimReleaseConfidence, A as evaluateReleaseConfidence, C as judgeReplayGate, E as pairedBootstrap, G as pairedEvalueSequence, H as pairedWilcoxon, K as releaseTraceEvidenceFromMultiShotTrials, L as renderReleaseReport, M as rubricPredictiveValidity } from './reporting-B82RSv9C.js';
 import { F as FeedbackLabel, a as FeedbackTrajectoryStore, b as FeedbackTrajectory, C as ControlSeverity, c as ControlEvalResult } from './feedback-trajectory-CB0A32o3.js';
 export { d as ControlActionFailureMode, e as ControlActionOutcome, f as ControlBudget, g as ControlContext, h as ControlDecision, i as ControlRunResult, j as ControlRuntimeConfig, k as ControlRuntimeError, l as ControlStep, m as ControlStopPolicies, n as FeedbackArtifactType, o as FeedbackAttempt, p as FeedbackLabelKind, q as FeedbackLabelSource, r as FeedbackOptimizerRow, s as FeedbackOutcome, t as FeedbackReplayAdapter, u as FeedbackReplayResult, v as FeedbackSeverity, w as FeedbackSplitPolicy, x as FeedbackTask, y as FeedbackTrajectoryFilter, z as FileSystemFeedbackTrajectoryStore, I as InMemoryFeedbackTrajectoryStore, P as PreferenceMemoryEntry, A as ProposedSideEffect, S as StopDecision, B as allCriticalPassed, D as assignFeedbackSplit, E as controlRunToFeedbackTrajectory, G as createFeedbackTrajectory, H as feedbackTrajectoriesToDatasetScenarios, J as feedbackTrajectoriesToOptimizerRows, K as feedbackTrajectoryToDatasetScenario, L as feedbackTrajectoryToOptimizerRow, M as objectiveEval, N as parseFeedbackTrajectoriesJsonl, O as renderPreferenceMemoryMarkdown, Q as replayFeedbackTrajectories, R as replayFeedbackTrajectory, T as runAgentControlLoop, U as serializeFeedbackTrajectoriesJsonl, V as stopOnNoProgress, W as stopOnRepeatedAction, X as subjectiveEval, Y as summarizePreferenceMemory, Z as withAssignedFeedbackSplit } from './feedback-trajectory-CB0A32o3.js';
 export { A as ActionExecutionPolicy, a as ActionPolicyDecision, C as ControlRunToRunRecordOptions, L as LlmJsonCall, b as LlmReviewerConfig, P as ProposeFn, c as ProposeInput, d as ProposeOutput, e as ProposeReviewConfig, f as ProposeReviewControlAction, g as ProposeReviewControlConfig, h as ProposeReviewControlResult, i as ProposeReviewControlState, j as ProposeReviewReport, k as ProposeReviewShot, R as Review, l as ReviewFn, m as ReviewInput, n as ReviewMemoryEntry, o as ReviewMemoryStore, p as RunEvidenceMetadata, V as Verification, q as VerifyFn, r as controlFailureClassFromVerification, s as controlRunToRunRecord, t as createLlmReviewer, u as evaluateActionPolicy, v as inMemoryReviewStore, w as jsonlReviewStore, x as runProposeReview, y as runProposeReviewAsControlLoop, z as scoreFromEvals } from './control-cxwMOAsy.js';
 import { T as TraceEmitter } from './emitter-B2XqDKFU.js';
 export { R as RunCompleteHook, a as RunCompleteHookContext, S as SpanHandle, b as TraceEmitterOptions, l as llmSpanFromProvider } from './emitter-B2XqDKFU.js';
-import { A as ActionableSideInfo, O as Objective, P as ParetoResult, T as TrialCache, a as TrialResult, E as EvolvableVariant, M as MutateAdapter, V as VariantAggregate } from './multi-shot-optimization-Bvtz294B.js';
-export { b as AsiSeverity, D as DEFAULT_MUTATION_PRIMITIVES, c as Direction, G as GateDecision, d as GateEvidence, e as GenerationReport, H as HeldOutGate, f as HeldOutGateConfig, g as HeldOutGateRejectionCode, I as InMemoryTrialCache, h as MultiShotGateConfig, i as MultiShotGateResult, j as MultiShotMutateAdapter, k as MultiShotOptimizationConfig, l as MultiShotOptimizationResult, m as MultiShotRun, n as MultiShotRunInput, o as MultiShotRunner, p as MultiShotScore, q as MultiShotScorer, r as MultiShotSplit, s as MultiShotTrace, t as MultiShotTrialResult, u as MultiShotVariant, v as PromptEvolutionConfig, w as PromptEvolutionEvent, x as PromptEvolutionResult, R as ReflectionContext, y as ReflectionProposal, S as ScenarioAggregate, z as ScoreAdapter, B as TrialTrace, C as buildReflectionPrompt, F as crowdingDistance, J as defaultMultiShotObjectives, K as dominates, L as paretoFrontier, N as paretoFrontierWithCrowding, Q as parseReflectionResponse, U as runMultiShotOptimization, W as runPromptEvolution, X as scalarScore, Y as trialTraceFromMultiShotTrial } from './multi-shot-optimization-Bvtz294B.js';
+import { A as ActionableSideInfo, O as Objective, P as ParetoResult, T as TrialCache, a as TrialResult, E as EvolvableVariant, M as MutateAdapter, V as VariantAggregate } from './summary-report-D4p7RlDu.js';
+export { b as AsiSeverity, D as DEFAULT_FAILURE_RULES, c as DEFAULT_MUTATION_PRIMITIVES, d as Direction, F as FailureClassification, e as FailureCluster, f as FailureClusterReport, g as FailureContext, h as FailureRule, G as GainDistributionBin, i as GainDistributionFigureSpec, j as GainDistributionOptions, k as GateDecision, l as GateEvidence, m as GenerationReport, H as HeldOutGate, n as HeldOutGateConfig, o as HeldOutGateRejectionCode, I as InMemoryTrialCache, p as MultiShotGateConfig, q as MultiShotGateResult, r as MultiShotMutateAdapter, s as MultiShotOptimizationConfig, t as MultiShotOptimizationResult, u as MultiShotRun, v as MultiShotRunInput, w as MultiShotRunner, x as MultiShotScore, y as MultiShotScorer, z as MultiShotSplit, B as MultiShotTrace, C as MultiShotTrialResult, J as MultiShotVariant, K as ParetoFigureSpec, L as ParetoPoint, N as PromptEvolutionConfig, Q as PromptEvolutionEvent, R as PromptEvolutionResult, S as RESEARCH_REPORT_HARD_PAIR_FLOOR, U as ReflectionContext, W as ReflectionProposal, X as ResearchReport, Y as ResearchReportCandidate, Z as ResearchReportDecision, _ as ResearchReportMethodology, $ as ResearchReportOptions, a0 as ResearchReportRecommendation, a1 as ScenarioAggregate, a2 as ScoreAdapter, a3 as SummaryTable, a4 as SummaryTableOptions, a5 as SummaryTableRow, a6 as TrialTrace, a7 as buildReflectionPrompt, a8 as classifyFailure, a9 as crowdingDistance, aa as defaultMultiShotObjectives, ab as dominates, ac as failureClusterView, ad as gainHistogram, ae as paretoChart, af as paretoFrontier, ag as paretoFrontierWithCrowding, ah as parseReflectionResponse, ai as researchReport, aj as runMultiShotOptimization, ak as runPromptEvolution, al as scalarScore, am as summaryTable, an as trialTraceFromMultiShotTrial } from './summary-report-D4p7RlDu.js';
 import { a as Run$1, S as Span, f as TraceEvent, A as Artifact$1, B as BudgetLedgerEntry, T as TraceStore, F as FailureClass, g as BudgetSpec, c as ToolSpan, h as RunFilter, L as LlmSpan, J as JudgeSpan } from './store-u47QaJ9G.js';
 export { i as EventFilter, E as EventKind, j as FAILURE_CLASSES, k as FileSystemTraceStore, l as FileSystemTraceStoreOptions, G as GenericSpan, I as InMemoryTraceStore, M as Message, d as RetrievalSpan, m as RunLayer, n as RunStatus, e as SandboxSpan, o as SpanBase, p as SpanFilter, b as SpanKind, q as SpanStatus, r as TRACE_SCHEMA_VERSION, s as isJudgeSpan, t as isLlmSpan, u as isRetrievalSpan, v as isSandboxSpan, w as isToolSpan } from './store-u47QaJ9G.js';
-import { llmSpans, RawProviderSink, ProviderRedactor } from './traces.js';
-export { AnalyzeTracesInput, AnalyzeTracesOptions, AnalyzeTracesResult, AnalyzeTracesTurnSnapshot, DEFAULT_REDACTION_RULES, DEFAULT_TRACE_ANALYST_BUDGETS, DatasetOverview, FileSystemRawProviderSink, FileSystemRawProviderSinkOptions, InMemoryRawProviderSink, InMemoryRawProviderSinkOptions, NoopRawProviderSink, OTEL_AGENT_EVAL_SCOPE, OtlpExport, OtlpFileTraceStore, OtlpFileTraceStoreOptions, OtlpResourceSpans, OtlpSpan, QueryTracesPage, REDACTION_VERSION, RawProviderDirection, RawProviderEvent, RawProviderSinkFilter, RedactionReport, RedactionRule, RunIntegrityError, RunIntegrityExpectations, RunIntegrityIssue, RunIntegrityIssueCode, RunIntegrityReport, SearchSpanResult, SearchTraceResult, SpanMatchRecord, SpanNotFoundError, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_SUBAGENT_DESCRIPTION, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TraceAnalysisStore, TraceAnalystByteBudgets, TraceAnalystFilters, TraceAnalystHookOptions, TraceAnalystSpan, TraceAnalystSpanKind, TraceAnalystSpanStatus, TraceAnalystTraceSummary, TraceFileMissingError, TraceInsightContext, TraceInsightFinding, TraceInsightPanelRole, TraceInsightPromptInput, TraceInsightQualityGate, TraceInsightQuestion, TraceInsightReadiness, TraceInsightSuite, TraceInsightTask, TraceNotFoundError, ViewSpansResult, ViewTraceOversized, ViewTraceResult, aggregateLlm, analyzeTraces, argHash, assertRunCaptured, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, defaultProviderRedactor, defaultTraceInsightPanel, describeTraceInsightScope, domainEvidencePattern, exportRunAsOtlp, groupBy, inferDomainKeywords, judgeSpans, planTraceInsightQuestions, providerFromBaseUrl, redactString, redactValue, runFailureClass, runsForScenario, scoreTraceInsightReadiness, throwIfRunIncomplete, tokenizeDomainWords, toolSpans, traceAnalystFunctionGroup, traceAnalystOnRunComplete } from './traces.js';
+import { llmSpans } from './traces.js';
+export { AnalyzeTracesInput, AnalyzeTracesOptions, AnalyzeTracesResult, AnalyzeTracesTurnSnapshot, DEFAULT_REDACTION_RULES, DEFAULT_TRACE_ANALYST_BUDGETS, DatasetOverview, OTEL_AGENT_EVAL_SCOPE, OtlpExport, OtlpFileTraceStore, OtlpFileTraceStoreOptions, OtlpResourceSpans, OtlpSpan, QueryTracesPage, REDACTION_VERSION, RedactionReport, RedactionRule, ReplayCache, ReplayCacheEntry, ReplayCacheMissError, ReplayCacheStats, ReplayFetchOptions, SearchSpanResult, SearchTraceResult, SpanMatchRecord, SpanNotFoundError, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_SUBAGENT_DESCRIPTION, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TraceAnalysisStore, TraceAnalystByteBudgets, TraceAnalystFilters, TraceAnalystHookOptions, TraceAnalystSpan, TraceAnalystSpanKind, TraceAnalystSpanStatus, TraceAnalystTraceSummary, TraceFileMissingError, TraceInsightContext, TraceInsightFinding, TraceInsightPanelRole, TraceInsightPromptInput, TraceInsightQualityGate, TraceInsightQuestion, TraceInsightReadiness, TraceInsightSuite, TraceInsightTask, TraceNotFoundError, ViewSpansResult, ViewTraceOversized, ViewTraceResult, aggregateLlm, analyzeTraces, argHash, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, createReplayFetch, defaultTraceInsightPanel, describeTraceInsightScope, domainEvidencePattern, exportRunAsOtlp, groupBy, inferDomainKeywords, iterateRawCalls, judgeSpans, planTraceInsightQuestions, redactString, redactValue, runFailureClass, runsForScenario, scoreTraceInsightReadiness, tokenizeDomainWords, toolSpans, traceAnalystFunctionGroup, traceAnalystOnRunComplete } from './traces.js';
+export { F as FileSystemRawProviderSink, a as FileSystemRawProviderSinkOptions, I as InMemoryRawProviderSink, b as InMemoryRawProviderSinkOptions, N as NoopRawProviderSink, P as ProviderRedactor, R as RawProviderDirection, c as RawProviderEvent, d as RawProviderSink, e as RawProviderSinkFilter, f as RunIntegrityError, g as RunIntegrityExpectations, h as RunIntegrityIssue, i as RunIntegrityIssueCode, j as RunIntegrityReport, k as assertRunCaptured, l as defaultProviderRedactor, p as providerFromBaseUrl, t as throwIfRunIncomplete } from './integrity-K2oVlF57.js';
 import { a as DatasetScenario, b as Dataset, c as DatasetManifest } from './dataset-B9qvlm_o.js';
 export { d as DatasetDifficulty, e as DatasetProvenance, D as DatasetSplit, H as HoldoutLockedError, S as SliceOptions, h as hashScenarios } from './dataset-B9qvlm_o.js';
+import { L as LlmClientOptions } from './optimization-UVDNKaO6.js';
+export { C as CallbackResearcher, a as CallbackResearcherOptions, b as CampaignFactoryParams, c as CampaignIntegrityPolicy, d as CampaignRunContext, e as CampaignRunOutcome, f as CampaignRunner, g as CampaignScenario, h as CampaignVariant, E as EvalCampaignOptions, i as EvalCampaignResult, j as ExperimentPlan, k as ExperimentResult, F as FailedRun, l as FailureMode, m as LlmCallError, n as LlmCallRequest, o as LlmCallResult, p as LlmClient, q as LlmMessage, r as LlmRouteAssertionError, s as LlmRouteRequirements, t as LlmUsage, N as NoopResearcher, R as Researcher, S as SteeringChange, u as assertLlmRoute, v as callLlm, w as callLlmJson, x as probeLlm, y as runEvalCampaign, z as stripFencedJson } from './optimization-UVDNKaO6.js';
 import { a as RunRecord } from './run-record-CX_jcAyr.js';
 export { b as RunJudgeMetadata, c as RunOutcome, d as RunRecordValidationError, R as RunSplitTag, e as RunTokenUsage, i as isRunRecord, p as parseRunRecordSafe, r as roundTripRunRecord, v as validateRunRecord } from './run-record-CX_jcAyr.js';
-export { CallbackResearcher, CallbackResearcherOptions, ExperimentPlan, ExperimentResult, FailureMode, NoopResearcher, Researcher, SteeringChange } from './optimization.js';
 export { B as BENCHMARK_SPLIT_SEED, a as BenchmarkAdapter, b as BenchmarkDatasetItem, c as BenchmarkEvaluation, d as benchmarkDeterministicSplit, i as benchmarks } from './index-c5saLbKD.js';
 import '@ax-llm/ax';
 
@@ -3289,68 +3291,6 @@ declare class ProjectRegistry {
     projectChats(projectId: string): Promise<ChatSummary[]>;
 }
 
-/**
- * OutcomeStore — deployment outcomes attached to Run IDs.
- *
- * Outcomes arrive asynchronously from production telemetry after the
- * eval run completed: user ratings, retention flags, conversion events,
- * revenue, support-ticket rate, anything a product team can measure.
- * The store is a peer to TraceStore — separate lifecycle, same runId
- * foreign key.
- *
- * The whole point of this module is to make the meta-eval correlation
- * question computable: `correlate(evalMetric, outcomeMetric) → r, ρ, n, CI`.
- */
-interface DeploymentOutcome {
-    runId: string;
-    capturedAt: number;
-    /** Numeric outcomes keyed by name — retention_7d, csat, revenue_usd, etc. */
-    metrics: Record<string, number>;
-    /** Dimensions for stratified analysis — cohort, region, user_segment. */
-    labels?: Record<string, string>;
-    /** Free-form provenance (source system, pipeline version). */
-    source?: string;
-}
-interface OutcomeFilter {
-    runIds?: string[];
-    since?: number;
-    until?: number;
-    label?: {
-        key: string;
-        value: string;
-    };
-    source?: string;
-}
-interface OutcomeStore {
-    append(outcome: DeploymentOutcome): Promise<void>;
-    /** All outcomes attached to this run (a single run can have many — multiple
-     *  capture windows over deployment time). */
-    forRun(runId: string): Promise<DeploymentOutcome[]>;
-    list(filter?: OutcomeFilter): Promise<DeploymentOutcome[]>;
-}
-declare class InMemoryOutcomeStore implements OutcomeStore {
-    private items;
-    append(outcome: DeploymentOutcome): Promise<void>;
-    forRun(runId: string): Promise<DeploymentOutcome[]>;
-    list(filter?: OutcomeFilter): Promise<DeploymentOutcome[]>;
-}
-interface FileSystemOutcomeStoreOptions {
-    dir: string;
-    maxBytes?: number;
-}
-declare class FileSystemOutcomeStore implements OutcomeStore {
-    private dir;
-    private maxBytes;
-    private memo?;
-    private loaded;
-    constructor(options: FileSystemOutcomeStoreOptions);
-    private ensureDir;
-    append(outcome: DeploymentOutcome): Promise<void>;
-    private load;
-    forRun(runId: string): Promise<DeploymentOutcome[]>;
-    list(filter?: OutcomeFilter): Promise<DeploymentOutcome[]>;
-}
-
 /**
  * Correlation study — "does our eval score predict real-world outcomes?"
  *
@@ -4286,220 +4226,6 @@ interface UseCaseSignals {
 declare function classifyEuAiRisk(signals: UseCaseSignals): EuRiskClass;
 declare function euAiActReport(ctx: GovernanceContext, signals: UseCaseSignals): Promise<GovernanceReport>;
 
-/**
- * LLM client with graceful degrade.
- *
- * OpenAI-compatible `/v1/chat/completions` client with:
- *   - Exponential-backoff retry on 429 + 5xx gateway errors (502/503/504).
- *   - Retry on transient network errors (fetch failed, AbortError, ECONNRESET).
- *   - Graceful json_schema → json_object degrade on 400 with schema-reject body.
- *   - Fenced-JSON stripping (```json ... ```) for models that wrap structured output.
- *   - Configurable base URL + api key / bearer, works with LiteLLM proxies, OpenAI
- *     directly, cli-bridge subscriptions, and any router that speaks the spec.
- *
- * Usage:
- *   const { value, result } = await callLlmJson<MyType>(
- *     { model: 'gpt-4o', messages: [...], jsonSchema: { name: 'x', schema: {...} } },
- *     { baseUrl: 'https://router.tangle.tools/v1', apiKey: process.env.KEY },
- *   )
- *
- * This is THE llm-calling seam for agent-eval primitives that need structured
- * output (semantic concept judge, reviewer directives, critic scores). Primitives
- * that need free-form text use `callLlm` and parse output themselves.
- */
-
-interface LlmMessage {
-    role: 'system' | 'user' | 'assistant';
-    /**
-     * Either a plain text content string OR a multimodal content array
-     * (text + image_url parts) for vision-capable models.
-     */
-    content: string | Array<{
-        type: 'text';
-        text: string;
-    } | {
-        type: 'image_url';
-        image_url: {
-            url: string;
-            detail?: 'auto' | 'low' | 'high';
-        };
-    }>;
-}
-interface LlmCallRequest {
-    model: string;
-    messages: LlmMessage[];
-    /** Optional JSON-mode response format (response_format: json_object). */
-    jsonMode?: boolean;
-    /** Optional structured output via JSON Schema. Falls back to json_object on 400. */
-    jsonSchema?: {
-        name: string;
-        schema: Record<string, unknown>;
-    };
-    temperature?: number;
-    maxTokens?: number;
-    /** Per-call timeout, default 60s. */
-    timeoutMs?: number;
-}
-interface LlmUsage {
-    promptTokens: number;
-    completionTokens: number;
-    totalTokens: number;
-    /** Proxies populate this when prompt caching is on. */
-    cachedPromptTokens?: number;
-}
-interface LlmCallResult {
-    /** The text content of the first choice. Empty string if none. */
-    content: string;
-    usage: LlmUsage;
-    /**
-     * Cost in USD. Pulled from proxy's `_response_cost` field when present;
-     * `null` when neither the proxy nor the caller can derive it.
-     */
-    costUsd: number | null;
-    /** Model name actually used (echoed from response). */
-    model: string;
-    /** Wall-clock duration of the HTTP call (last attempt, if retried). */
-    durationMs: number;
-    /** Raw response body. */
-    raw: Record<string, unknown>;
-}
-declare class LlmCallError extends Error {
-    readonly status: number;
-    readonly body: string;
-    readonly model: string;
-    constructor(message: string, status: number, body: string, model: string);
-}
-interface LlmClientOptions {
-    /** Base URL (without trailing slash). Must end at the `/v1` prefix. */
-    baseUrl?: string;
-    /** Bearer token — either `apiKey` or `bearer` populates `Authorization: Bearer ...`. */
-    apiKey?: string;
-    bearer?: string;
-    /** Override for the `Authorization` header (e.g. `X-Auth: ...`). Takes precedence over apiKey/bearer. */
-    authHeader?: {
-        name: string;
-        value: string;
-    };
-    /** Default timeout in ms. Per-call can override. */
-    defaultTimeoutMs?: number;
-    /** Max retry attempts on retriable errors. Default 3 (1 initial + 2 retries). */
-    maxRetries?: number;
-    /** Fetch implementation — defaults to global `fetch`. Override for custom transport (e.g. tests). */
-    fetch?: typeof fetch;
-    /**
-     * Optional raw HTTP capture sink. When provided, every request, response,
-     * and error (across all retry attempts) is recorded to the sink, with auth
-     * headers and credential-shaped body fields redacted by default. This is
-     * the layer-1 forensics primitive: structured `LlmSpan`s record intent,
-     * raw events record what actually crossed the wire.
-     */
-    rawSink?: RawProviderSink;
-    /**
-     * Logical provider id attached to raw events. When omitted, derived from
-     * `baseUrl` via `providerFromBaseUrl`.
-     */
-    provider?: string;
-    /** Trace context attached to raw events; populated by emitter-aware callers. */
-    traceContext?: {
-        runId?: string;
-        spanId?: string;
-    };
-    /** Override the redaction strategy for this call. Defaults to `defaultProviderRedactor`. */
-    redactor?: ProviderRedactor;
-}
-/**
- * Strip a ```json / ``` code fence if the model emitted one.
- * Idempotent for naked JSON. Some models (claude-code via router, certain
- * deepseek models) wrap output even under json_object.
- */
-declare function stripFencedJson(raw: string): string;
-/**
- * Low-level call. Returns raw content + usage + cost. Retries on transient
- * failures; does NOT degrade schema here — callers that want graceful
- * degrade use `callLlmJson`.
- */
-declare function callLlm(req: LlmCallRequest, opts?: LlmClientOptions): Promise<LlmCallResult>;
-/**
- * Structured-output call. Returns parsed JSON plus the raw result envelope.
- * Degrades `jsonSchema` → `jsonMode` on a 400 that names the schema param —
- * critical for deepseek-v3/v4, kimi-k2.6, and other models that don't accept
- * the `response_format.json_schema` shape but DO accept `json_object`.
- */
-declare function callLlmJson<T = unknown>(req: LlmCallRequest, opts?: LlmClientOptions): Promise<{
-    value: T;
-    result: LlmCallResult;
-}>;
-declare class LlmRouteAssertionError extends Error {
-    readonly code: 'no_explicit_base_url' | 'base_url_blocked' | 'base_url_not_allowed' | 'no_auth' | 'wrong_provider';
-    readonly baseUrl: string;
-    constructor(message: string, code: 'no_explicit_base_url' | 'base_url_blocked' | 'base_url_not_allowed' | 'no_auth' | 'wrong_provider', baseUrl: string);
-}
-interface LlmRouteRequirements {
-    /**
-     * Throw if `opts.baseUrl` is undefined, i.e. the call would fall back to
-     * `DEFAULT_BASE_URL`. Set this for evaluation runs where silently using
-     * the public/free-tier router is a defect — the launch reviewer needs to
-     * know exactly which provider answered.
-     */
-    requireExplicitBaseUrl?: boolean;
-    /**
-     * Allowlist of acceptable base URLs. Strings match by prefix
-     * (case-insensitive); RegExps test against the full base URL.
-     */
-    allowedBaseUrls?: Array<string | RegExp>;
-    /** Blocklist that takes precedence over `allowedBaseUrls`. */
-    blockedBaseUrls?: Array<string | RegExp>;
-    /** Throw if no auth header / api key is configured. */
-    requireAuth?: boolean;
-    /**
-     * Logical provider id the configured `baseUrl` is expected to match (via
-     * `providerFromBaseUrl`). Mainly useful when paired with `requireExplicitBaseUrl`.
-     */
-    expectedProvider?: string;
-}
-/**
- * Fail-loud assertion that the configured LLM client points at the route
- * the caller intends. Designed for the matrix-runner preflight: invoke
- * once before any LLM call to catch misconfiguration before a sweep burns
- * dollars on the wrong provider.
- *
- * Throws `LlmRouteAssertionError`. Pure — no I/O — so it's safe to call
- * from constructors and CI gates.
- */
-declare function assertLlmRoute(opts: LlmClientOptions, req?: LlmRouteRequirements): void;
-/**
- * Probe whether a model is reachable. Returns latency + null error on
- * success; `ok=false` + error message on any failure (HTTP, timeout,
- * network, parse). Designed for sweep preflights — fail loud at the
- * boundary before burning a 30-leaf run on a misconfigured router.
- *
- * Sends a tiny `ping` message with `maxTokens=64`. Reasoning models
- * (glm-5.1, deepseek-v4) can burn the entire budget on internal reasoning
- * for short prompts, so don't tighten this further. We don't validate
- * content; HTTP 200 means reachable.
- */
-declare function probeLlm(model: string, opts?: LlmClientOptions & {
-    timeoutMs?: number;
-}): Promise<{
-    ok: boolean;
-    latencyMs: number;
-    error: string | null;
-}>;
-/**
- * Stateful client — construct once with defaults, call many times.
- * Thin wrapper around the free functions; exists for callers that want
- * to inject a single configured instance into multiple primitives.
- */
-declare class LlmClient {
-    private readonly opts;
-    constructor(opts?: LlmClientOptions);
-    call(req: LlmCallRequest, per?: LlmClientOptions): Promise<LlmCallResult>;
-    callJson<T = unknown>(req: LlmCallRequest, per?: LlmClientOptions): Promise<{
-        value: T;
-        result: LlmCallResult;
-    }>;
-}
-
 /**
  * Multi-layer verifier — ordered pipeline of verification layers.
  *
@@ -6312,4 +6038,4 @@ interface OrthogonalityResult {
 }
 declare function passOrthogonality<T>(input: OrthogonalityInput<T>): OrthogonalityResult;
 
-export { ActionableSideInfo, type ActiveLearningOptions, type AdapterRun, AgentDriver, type AgentDriverConfig, type AlignmentOp, type AntiSlopConfig, type AntiSlopIssue, type AntiSlopReport, Artifact$1 as Artifact, type ArtifactCheck, type Artifact as ArtifactCheckArtifact, type ArtifactResult, type ArtifactValidator, AxGepaSteeringOptimizer, type AxSteeringOptimizerConfig, type BaselineOptions, type BaselineReport, BehaviorAssertion, type BenchmarkReport, BenchmarkRunner, type BenchmarkRunnerConfig, type BestOfNResult, type BisectOptions, type BisectResult, type BisectStep, BudgetBreachError, type BudgetBreachFinding, type BudgetBreachReport, BudgetGuard, BudgetLedgerEntry, BudgetSpec, BuilderSession, type BuilderSessionInit, type CalibrationBin, type CalibrationOptions, type CalibrationReport, type CalibrationResult, CallExpectation, type CanaryAlert, type CanaryKind, type CanaryLeak, type CanaryOptions, type CanaryReport, type CanarySeverity, type CandidateScenario, type CandidateScore, type CausalAttributionReport, type ChatSummary, type CheckResult, 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, type ContractMetric, type ContractReport, ControlEvalResult, ControlSeverity, ConvergenceTracker, type CorrelationReport, type CorrelationResult, type CorrelationStudyOptions, type CorrelationStudyResult, type CostEntry, CostLedger, type CostLedgerGeneration, type CostLedgerSnapshot, type CostSummary, CostTracker, type CounterfactualContext, type CounterfactualMutation, type CounterfactualResult, type CounterfactualRunner, type CreateCompositeMutatorOpts, type CreateDefaultReviewerOptions, type CreateSandboxCodeMutatorOpts, type CreateSandboxPoolOpts, 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_RED_TEAM_CORPUS, DEFAULT_RUN_SCORE_WEIGHTS, DEFAULT_SEVERITY_WEIGHTS, type DataAcquisitionPlan, Dataset, DatasetManifest, DatasetScenario, type DeployFamily, type DeployGateLayerInput, type DeployRunResult, type DeployRunner, type DeploymentOutcome, type DirEntry, type DivergenceOptions, type DivergenceReport, DockerSandboxDriver, type DriverResult, type DriverState, DualAgentBench, type DualAgentBenchConfig, type DualAgentReport, type DualAgentRound, type DualAgentScenario, type DualAgentScenarioResult, ERROR_COUNT_PATTERNS, type ErrorCountPattern, type EuRiskClass, type EvalMetricSpec, type EvalResult, type EvolutionRound, EvolvableVariant, type ExecutorConfig, type Expectation, type Experiment, type Run as ExperimentRun, type ExperimentStore, ExperimentTracker, type ExportedRewardModel, type ExtractOptions, type ExtractResult, type FactorContribution, type FactorialCell, FailureClass, FeedbackLabel, type FeedbackPattern, FeedbackTrajectory, FeedbackTrajectoryStore, FileSystemExperimentStore, type FileSystemExperimentStoreOptions, FileSystemOutcomeStore, type FileSystemOutcomeStoreOptions, type Finding, type FlowAction, type FlowLayerEnv, type FlowLayerFactoryInput, type FlowRunner, type FlowRunnerStepResult, type FlowSpec, type FlowStep, type GoldenItem, type GoldenSeverity, type GoldenSpec, type GovernanceContext, type GovernanceFinding, type GovernanceReport, type GradedStep, type HarnessAdapter, type HarnessConfig, type HarnessExperimentConfig, type HarnessExperimentResult, type HarnessIntervention, type HarnessRunRequest, type HarnessRunResult, type HarnessScenario, type HarnessSelection, type HarnessVariant, type HarnessVariantReport, HoldoutAuditor, type HostedJudgeConfig, type HostedJudgeDimension, type HostedJudgeRequest, type HostedJudgeResponse, type HostedRunCriticConfig, type HostedRunScoreRequest, type HostedRunScoreResponse, type HypothesisManifest, type HypothesisResult, INTENT_MATCH_JUDGE_VERSION, type ImageData, InMemoryExperimentStore, InMemoryOutcomeStore, InMemoryWorkspaceInspector, type InferenceScorer, type InspectorContext, type IntegrationGateSurface, type IntegrationInvokeFailureInput, type IntegrationManifestGateInput, type IntentMatchInput, type IntentMatchOptions, type IntentMatchResult, type InteractionContribution, JsonlTrialCache, type JudgeAgreementReport, type JudgeConfig, type JudgeFleetOptions, type JudgeFn, type JudgeInput, type JudgePair, type JudgeReplayResult, type JudgeRubric, JudgeRunner, type JudgeScore, JudgeSpan, 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, type LangfuseEnvelope, type LangfuseGeneration, type LangfuseScore, type Layer, type LayerCorrelation, type LayerResult, type LayerStatus, type LineageKind, type LineageKindResolver, type LineageNode, LineageRecorder, type LiveProofArtifact, type LiveProofConfig, type LiveProofContext, type LiveProofResult, LlmCallError, type LlmCallRequest, type LlmCallResult, LlmClient, type LlmClientOptions, type LlmMessage, LlmRouteAssertionError, type LlmRouteRequirements, LlmSpan, type LlmUsage, LockedJsonlAppender, MODEL_PRICING, type MatchResult, type MatcherResult, type MeasurementPolicy, type MergeOptions, type MetricSamples, type MetricVerdict, MetricsCollector, type MuffledFinder, type MuffledFinding, MultiLayerVerifier, type MultiToolchainLayerConfig, MutateAdapter, type MutationAttempt, type MutationChannel, MutationTelemetry, type Mutator, Mutex, Objective, type Oracle, type OracleObservation, type OracleReport, type OracleResult, type OrthogonalityInput, type OrthogonalityResult, type OutcomeFilter, type OutcomePair, type OutcomeStore, PairwiseSteeringOptimizer, type ParaphraseRobustnessScenarioInput, type ParaphraseRobustnessScenarioResult, ParetoResult, type PersonaConfig, type Playbook, type PlaybookEntry, type PoolSlot, type PositionalBiasResult, type PrmGradedTrace, PrmGrader, type PrmTrainingSample, ProductClient, type ProductClientConfig, type ProjectKind, ProjectRegistry, type ProjectSummary, type ProjectTimelineEntry, type PromptHandle, PromptRegistry, TrialResult as PromptTrialResult, ProviderRedactor, RawProviderSink, type RedTeamCase, type RedTeamCategory, type RedTeamFinding, type RedTeamPayload, type RedTeamReport, 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 RegressionOptions, type RegressionSpec, ReleaseConfidenceScorecard, ReleaseConfidenceThresholds, type ReviewerMemoryEntry, type ReviewerOutput, type ReviewerPromptInput, type ReviewerSoftFailDefaults, type ReviewerVerificationSummary, type RobustnessResult, type RouteMap, type RubricDimension, Run$1 as Run, type RunAppScenarioOptions, type RunCommandInput, type RunCommandResult, type RunConfig, RunCritic, type RunCriticOptions, type RunDiff, RunFilter, RunRecord, type RunScore, type RunScoreWeights, type RunTrace, SEMANTIC_CONCEPT_JUDGE_VERSION, type SandboxDriver, SandboxHarness, type SandboxHarnessResult, type SandboxJudgeKind, type SandboxJudgeResult, type SandboxJudgeSpec, type SandboxPool, type SandboxResult, type ScanOptions, type Scenario, type ScenarioCost, type ScenarioFile, ScenarioRegistry, type ScenarioResult, type ScoreKnowledgeReadinessOptions, type ScoredTarget, type SelfPlayOptions, type SelfPlayProposer, type SelfPlayScorer, type SelfPreferenceResult, type SemanticConceptJudgeInput, type SemanticConceptJudgeOptions, type SemanticConceptJudgeResult, type SeriesConvergenceOptions, type SeriesConvergenceResult, type Severity, type ShipOptions, 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 StepContext, type StepRubric, type StuckLoopFinding, type StuckLoopOptions, type StuckLoopReport, SubprocessSandboxDriver, type SubprocessSandboxDriverOptions, type SynthesisReason, type SynthesisTarget, type TestGradedRunOptions, type TestGradedRunResult, type TestGradedScenario, type TestOutputParser, type TestResult, type ThreeLayerProjectReport, type ThresholdContract, TokenCounter, type TokenSpec, ToolSpan, type ToolStats, type ToolUseMetrics, type ToolUseOptions, type ToolWasteFinding, type ToolWasteOptions, type ToolWasteReport, TraceEmitter, TraceEvent, TraceStore, type Trajectory, type TrajectoryStep, type TrialAttempt, TrialCache, TrialTelemetry, type Turn, type TurnMetrics, type TurnResult, UNIVERSAL_FINDERS, type UseCaseSignals, type UserQuestion, type ValidationContext, type ValidationIssue, type ValidationResult, VariantAggregate, type VerbosityBiasResult, type VerificationReport, type VerifyContext, type VerifyOptions, type VisualDiffOptions, type VisualDiffResult, type ViteDeployRunnerInput, type WorkflowTopology, type WorkspaceAssertion, type WorkspaceAssertionResult, type WorkspaceInspector, type WorkspaceSnapshot, type WranglerDeployRunnerInput, acquisitionPlansForKnowledgeGaps, adversarialJudge, aggregateRunScore, analyzeAntiSlop, analyzeSeries, assertLlmRoute, attributeCounterfactuals, benjaminiHochberg, bisect, blockingKnowledgeEval, bonferroni, budgetBreachView, buildReviewerPrompt, buildTrajectory, byteLengthRange, calibrateJudge, calibrationCurve, callLlm, callLlmJson, canaryLeakView, canonicalize, causalAttribution, checkBehavioralCanary, checkCanaries, checkSlos, clamp01, classifyEuAiRisk, codeExecutionJudge, cohensD, coherenceJudge, collectionPreserved, commitBisect, compareReferenceReplay, compareToBaseline, compilerJudge, composeParsers, composeValidators, computeToolUseMetrics, confidenceInterval, containsAll, correlateLayers, correlationStudy, createAntiSlopJudge, createCompositeMutator, createCustomJudge, createDefaultReviewer, createDomainExpertJudge, createIntentMatchJudge, createSandboxCodeMutator, createSandboxPool, createSemanticConceptJudge, crossTraceDiff, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultJudges, defaultReferenceReplayMatcher, deployGateLayer, distillPlaybook, estimateCost, estimateTokens, euAiActReport, evaluateContract, evaluateHypothesis, evaluateOracles, executeScenario, expectAgent, exportRewardModel, exportTrainingData, extractAssetUrls, extractErrorCount, fileContains, fileExists, findAutoMatchNoExpectation, findConstructorCwdDropped, findFallbackToPass, findLiteralTruePass, findSkipCountsAsPass, firstDivergenceView, flowLayer, formatBenchmarkReport, formatDriverReport, formatFindings, precision as goldenPrecision, gradeSemanticStatus, hashContent, hashJson, htmlContainsElement, inMemoryReferenceReplayStore, integrationAsi, integrationGateEvals, integrationInvokeFailedPayload, integrationManifestResolvedPayload, integrationManifestValidatedPayload, interRaterReliability, iqr, isPrmVerdict, jestTestParser, jsonHasKeys, jsonShape, jsonlReferenceReplayStore, judgeAgreementView, keyPreserved, knowledgeReadinessTracePayload, linterJudge, llmSpans, loadScorerFromGrader, localCommandRunner, lowercaseMutator, mannWhitneyU, matchGoldens, mergeLayerResults, mergeSteeringBundle, multiToolchainLayer, nistAiRmfReport, nonRefusalRubric, normalizeScores, notBlocked, outputLengthRubric, pairedTTest, paraphraseRobustness, paraphraseRobustnessScenarios, partialCredit, passOrthogonality, pixelDeltaRatio, politenessPrefixMutator, positionalBias, printDriverSummary, prmBestOfN, prmEnsembleBestOfN, probeLlm, promptBisect, proposeSynthesisTargets, pytestTestParser, redTeamDataset, redTeamReport, referenceReplayRunsToSteeringRows, referenceReplayScenarioToRunScore, regexMatch, regexMatches, regressionView, renderMarkdown, renderMarkdownReport, renderPlaybookMarkdown, renderSteeringText, replayScorerOverCorpus, replayTraceThroughJudge, requiredSampleSize, resetLockedAppendersForTesting, resumeBuilderSession, rowCount, rowWhere, runAssertions, runBehavioralCanaries, runCanaries, runCounterfactual, runE2EWorkflow, runExpectations, runHarnessExperiment, runIntentMatchJudge, runJudgeFleet, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runLiveProof, runReferenceReplay, runSelfPlay, runSemanticConceptJudge, runTestGradedScenario, scanForMuffledGates, scoreAllProjects, scoreContinuity, scoreKnowledgeReadiness, scoreProject, scoreRedTeamOutput, scoreReferenceReplay, securityJudge, selectHarnessVariant, selfPreference, sentenceReorderMutator, signManifest, soc2Report, statusAdvanced, stripFencedJson, stuckLoopView, summarize, summarizeHarnessResults, testJudge, textInSnapshot, toLangfuseEnvelope, toNdjson, toPrometheusText, toolIntentAlignmentRubric, toolNamesForRun, toolNonRedundantRubric, toolSuccessRubric, toolWasteView, typoMutator, urlContains, userQuestionsForKnowledgeGaps, verbosityBias, verifyManifest, visualDiff, viteDeployRunner, vitestTestParser, weightedMean, weightedRecall, welchsTTest, whitespaceCollapseMutator, wilcoxonSignedRank, wranglerDeployRunner };
+export { ActionableSideInfo, type ActiveLearningOptions, type AdapterRun, AgentDriver, type AgentDriverConfig, type AlignmentOp, type AntiSlopConfig, type AntiSlopIssue, type AntiSlopReport, Artifact$1 as Artifact, type ArtifactCheck, type Artifact as ArtifactCheckArtifact, type ArtifactResult, type ArtifactValidator, AxGepaSteeringOptimizer, type AxSteeringOptimizerConfig, type BaselineOptions, type BaselineReport, BehaviorAssertion, type BenchmarkReport, BenchmarkRunner, type BenchmarkRunnerConfig, type BestOfNResult, type BisectOptions, type BisectResult, type BisectStep, BudgetBreachError, type BudgetBreachFinding, type BudgetBreachReport, BudgetGuard, BudgetLedgerEntry, BudgetSpec, BuilderSession, type BuilderSessionInit, type CalibrationBin, type CalibrationOptions, type CalibrationReport, type CalibrationResult, CallExpectation, type CanaryAlert, type CanaryKind, type CanaryLeak, type CanaryOptions, type CanaryReport, type CanarySeverity, type CandidateScenario, type CandidateScore, type CausalAttributionReport, type ChatSummary, type CheckResult, 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, type ContractMetric, type ContractReport, ControlEvalResult, ControlSeverity, ConvergenceTracker, type CorrelationReport, type CorrelationResult, type CorrelationStudyOptions, type CorrelationStudyResult, type CostEntry, CostLedger, type CostLedgerGeneration, type CostLedgerSnapshot, type CostSummary, CostTracker, type CounterfactualContext, type CounterfactualMutation, type CounterfactualResult, type CounterfactualRunner, type CreateCompositeMutatorOpts, type CreateDefaultReviewerOptions, type CreateSandboxCodeMutatorOpts, type CreateSandboxPoolOpts, 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_RED_TEAM_CORPUS, DEFAULT_RUN_SCORE_WEIGHTS, DEFAULT_SEVERITY_WEIGHTS, type DataAcquisitionPlan, Dataset, DatasetManifest, DatasetScenario, type DeployFamily, type DeployGateLayerInput, type DeployRunResult, type DeployRunner, type DirEntry, type DivergenceOptions, type DivergenceReport, DockerSandboxDriver, type DriverResult, type DriverState, DualAgentBench, type DualAgentBenchConfig, type DualAgentReport, type DualAgentRound, type DualAgentScenario, type DualAgentScenarioResult, ERROR_COUNT_PATTERNS, type ErrorCountPattern, type EuRiskClass, type EvalMetricSpec, type EvalResult, type EvolutionRound, EvolvableVariant, type ExecutorConfig, type Expectation, type Experiment, type Run as ExperimentRun, type ExperimentStore, ExperimentTracker, type ExportedRewardModel, type ExtractOptions, type ExtractResult, type FactorContribution, type FactorialCell, FailureClass, FeedbackLabel, type FeedbackPattern, FeedbackTrajectory, FeedbackTrajectoryStore, FileSystemExperimentStore, type FileSystemExperimentStoreOptions, type Finding, type FlowAction, type FlowLayerEnv, type FlowLayerFactoryInput, type FlowRunner, type FlowRunnerStepResult, type FlowSpec, type FlowStep, type GoldenItem, type GoldenSeverity, type GoldenSpec, type GovernanceContext, type GovernanceFinding, type GovernanceReport, type GradedStep, type HarnessAdapter, type HarnessConfig, type HarnessExperimentConfig, type HarnessExperimentResult, type HarnessIntervention, type HarnessRunRequest, type HarnessRunResult, type HarnessScenario, type HarnessSelection, type HarnessVariant, type HarnessVariantReport, HoldoutAuditor, type HostedJudgeConfig, type HostedJudgeDimension, type HostedJudgeRequest, type HostedJudgeResponse, type HostedRunCriticConfig, type HostedRunScoreRequest, type HostedRunScoreResponse, type HypothesisManifest, type HypothesisResult, 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 JudgeAgreementReport, type JudgeConfig, type JudgeFleetOptions, type JudgeFn, type JudgeInput, type JudgePair, type JudgeReplayResult, type JudgeRubric, JudgeRunner, type JudgeScore, JudgeSpan, 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, type LangfuseEnvelope, type LangfuseGeneration, type LangfuseScore, type Layer, type LayerCorrelation, type LayerResult, type LayerStatus, type LineageKind, type LineageKindResolver, type LineageNode, LineageRecorder, type LiveProofArtifact, type LiveProofConfig, type LiveProofContext, type LiveProofResult, LlmClientOptions, LlmSpan, LockedJsonlAppender, MODEL_PRICING, type MatchResult, type MatcherResult, type MeasurementPolicy, type MergeOptions, type MetricSamples, type MetricVerdict, MetricsCollector, type MuffledFinder, type MuffledFinding, MultiLayerVerifier, type MultiToolchainLayerConfig, MutateAdapter, type MutationAttempt, type MutationChannel, MutationTelemetry, type Mutator, Mutex, Objective, type Oracle, type OracleObservation, type OracleReport, type OracleResult, type OrthogonalityInput, type OrthogonalityResult, OutcomeFilter, type OutcomePair, OutcomeStore, PairwiseSteeringOptimizer, type ParaphraseRobustnessScenarioInput, type ParaphraseRobustnessScenarioResult, ParetoResult, type PersonaConfig, type Playbook, type PlaybookEntry, type PoolSlot, type PositionalBiasResult, type PrmGradedTrace, PrmGrader, type PrmTrainingSample, ProductClient, type ProductClientConfig, type ProjectKind, ProjectRegistry, type ProjectSummary, type ProjectTimelineEntry, type PromptHandle, PromptRegistry, TrialResult as PromptTrialResult, type RedTeamCase, type RedTeamCategory, type RedTeamFinding, type RedTeamPayload, type RedTeamReport, 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 RegressionOptions, type RegressionSpec, ReleaseConfidenceScorecard, ReleaseConfidenceThresholds, type ReviewerMemoryEntry, type ReviewerOutput, type ReviewerPromptInput, type ReviewerSoftFailDefaults, type ReviewerVerificationSummary, type RobustnessResult, type RouteMap, type RubricDimension, Run$1 as Run, type RunAppScenarioOptions, type RunCommandInput, type RunCommandResult, type RunConfig, RunCritic, type RunCriticOptions, type RunDiff, RunFilter, RunRecord, type RunScore, type RunScoreWeights, type RunTrace, SEMANTIC_CONCEPT_JUDGE_VERSION, type SandboxDriver, SandboxHarness, type SandboxHarnessResult, type SandboxJudgeKind, type SandboxJudgeResult, type SandboxJudgeSpec, type SandboxPool, type SandboxResult, type ScanOptions, type Scenario, type ScenarioCost, type ScenarioFile, ScenarioRegistry, type ScenarioResult, type ScoreKnowledgeReadinessOptions, type ScoredTarget, type SelfPlayOptions, type SelfPlayProposer, type SelfPlayScorer, type SelfPreferenceResult, type SemanticConceptJudgeInput, type SemanticConceptJudgeOptions, type SemanticConceptJudgeResult, type SeriesConvergenceOptions, type SeriesConvergenceResult, type Severity, type ShipOptions, 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 StepContext, type StepRubric, type StuckLoopFinding, type StuckLoopOptions, type StuckLoopReport, SubprocessSandboxDriver, type SubprocessSandboxDriverOptions, type SynthesisReason, type SynthesisTarget, type TestGradedRunOptions, type TestGradedRunResult, type TestGradedScenario, type TestOutputParser, type TestResult, type ThreeLayerProjectReport, type ThresholdContract, TokenCounter, type TokenSpec, ToolSpan, type ToolStats, type ToolUseMetrics, type ToolUseOptions, type ToolWasteFinding, type ToolWasteOptions, type ToolWasteReport, TraceEmitter, TraceEvent, TraceStore, type Trajectory, type TrajectoryStep, type TrialAttempt, TrialCache, TrialTelemetry, type Turn, type TurnMetrics, type TurnResult, UNIVERSAL_FINDERS, type UseCaseSignals, type UserQuestion, type ValidationContext, type ValidationIssue, type ValidationResult, VariantAggregate, type VerbosityBiasResult, type VerificationReport, type VerifyContext, type VerifyOptions, type VisualDiffOptions, type VisualDiffResult, type ViteDeployRunnerInput, type WorkflowTopology, type WorkspaceAssertion, type WorkspaceAssertionResult, type WorkspaceInspector, type WorkspaceSnapshot, type WranglerDeployRunnerInput, acquisitionPlansForKnowledgeGaps, adversarialJudge, aggregateRunScore, analyzeAntiSlop, analyzeSeries, attributeCounterfactuals, benjaminiHochberg, bisect, blockingKnowledgeEval, bonferroni, budgetBreachView, buildReviewerPrompt, buildTrajectory, byteLengthRange, calibrateJudge, calibrationCurve, canaryLeakView, canonicalize, causalAttribution, checkBehavioralCanary, checkCanaries, checkSlos, clamp01, classifyEuAiRisk, codeExecutionJudge, cohensD, coherenceJudge, collectionPreserved, commitBisect, compareReferenceReplay, compareToBaseline, compilerJudge, composeParsers, composeValidators, computeToolUseMetrics, confidenceInterval, containsAll, correlateLayers, correlationStudy, createAntiSlopJudge, createCompositeMutator, createCustomJudge, createDefaultReviewer, createDomainExpertJudge, createIntentMatchJudge, createSandboxCodeMutator, createSandboxPool, createSemanticConceptJudge, crossTraceDiff, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultJudges, defaultReferenceReplayMatcher, deployGateLayer, distillPlaybook, estimateCost, estimateTokens, euAiActReport, evaluateContract, evaluateHypothesis, evaluateOracles, executeScenario, expectAgent, exportRewardModel, exportTrainingData, extractAssetUrls, extractErrorCount, fileContains, fileExists, findAutoMatchNoExpectation, findConstructorCwdDropped, findFallbackToPass, findLiteralTruePass, findSkipCountsAsPass, firstDivergenceView, flowLayer, formatBenchmarkReport, formatDriverReport, formatFindings, precision as goldenPrecision, gradeSemanticStatus, hashContent, hashJson, htmlContainsElement, inMemoryReferenceReplayStore, integrationAsi, integrationGateEvals, integrationInvokeFailedPayload, integrationManifestResolvedPayload, integrationManifestValidatedPayload, interRaterReliability, iqr, isPrmVerdict, jestTestParser, jsonHasKeys, jsonShape, jsonlReferenceReplayStore, judgeAgreementView, keyPreserved, knowledgeReadinessTracePayload, linterJudge, llmSpans, loadScorerFromGrader, localCommandRunner, lowercaseMutator, mannWhitneyU, matchGoldens, mergeLayerResults, mergeSteeringBundle, multiToolchainLayer, nistAiRmfReport, nonRefusalRubric, normalizeScores, notBlocked, outputLengthRubric, pairedTTest, paraphraseRobustness, paraphraseRobustnessScenarios, partialCredit, passOrthogonality, pixelDeltaRatio, politenessPrefixMutator, positionalBias, printDriverSummary, prmBestOfN, prmEnsembleBestOfN, promptBisect, proposeSynthesisTargets, pytestTestParser, redTeamDataset, redTeamReport, referenceReplayRunsToSteeringRows, referenceReplayScenarioToRunScore, regexMatch, regexMatches, regressionView, renderMarkdown, renderMarkdownReport, renderPlaybookMarkdown, renderSteeringText, replayScorerOverCorpus, replayTraceThroughJudge, requiredSampleSize, resetLockedAppendersForTesting, resumeBuilderSession, rowCount, rowWhere, runAssertions, runBehavioralCanaries, runCanaries, runCounterfactual, runE2EWorkflow, runExpectations, runHarnessExperiment, runIntentMatchJudge, runJudgeFleet, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runLiveProof, runReferenceReplay, runSelfPlay, runSemanticConceptJudge, runTestGradedScenario, scanForMuffledGates, scoreAllProjects, scoreContinuity, scoreKnowledgeReadiness, scoreProject, scoreRedTeamOutput, scoreReferenceReplay, securityJudge, selectHarnessVariant, selfPreference, sentenceReorderMutator, signManifest, soc2Report, statusAdvanced, stuckLoopView, summarize, summarizeHarnessResults, testJudge, textInSnapshot, toLangfuseEnvelope, toNdjson, toPrometheusText, toolIntentAlignmentRubric, toolNamesForRun, toolNonRedundantRubric, toolSuccessRubric, toolWasteView, typoMutator, urlContains, userQuestionsForKnowledgeGaps, verbosityBias, verifyManifest, visualDiff, viteDeployRunner, vitestTestParser, weightedMean, weightedRecall, welchsTTest, whitespaceCollapseMutator, wilcoxonSignedRank, wranglerDeployRunner };

package/dist/index.js

@@ -46,6 +46,7 @@ import {
   renderPreferenceMemoryMarkdown,
   replayFeedbackTrajectories,
   replayFeedbackTrajectory,
+  runEvalCampaign,
   runMultiShotOptimization,
   runPromptEvolution,
   scalarScore,
@@ -53,7 +54,7 @@ import {
   summarizePreferenceMemory,
   trialTraceFromMultiShotTrial,
   withAssignedFeedbackSplit
-} from "./chunk-HRZELXCR.js";
+} from "./chunk-USHQBPMH.js";
 import {
   RunRecordValidationError,
   isRunRecord,
@@ -62,40 +63,38 @@ import {
   validateRunRecord
 } from "./chunk-YUFXO3TU.js";
 import {
-  RESEARCH_REPORT_HARD_PAIR_FLOOR,
   assertReleaseConfidence,
   bootstrapCi,
-  canonicalize,
-  evaluateHypothesis,
+  evaluateInterimReleaseConfidence,
   evaluateReleaseConfidence,
-  gainHistogram,
-  hashJson,
   judgeReplayGate,
-  paretoChart,
+  pairedEvalueSequence,
   releaseTraceEvidenceFromMultiShotTrials,
   renderReleaseReport,
-  researchReport,
-  signManifest,
-  summaryTable,
-  verifyManifest
-} from "./chunk-3IX6QTB7.js";
+  rubricPredictiveValidity
+} from "./chunk-UAND2LOT.js";
 import {
+  RESEARCH_REPORT_HARD_PAIR_FLOOR,
   benjaminiHochberg,
   bhAdjust,
   bonferroni,
   cohensD,
   confidenceInterval,
+  gainHistogram,
   interRaterReliability,
   mannWhitneyU,
   normalizeScores,
   pairedBootstrap,
   pairedTTest,
   pairedWilcoxon,
+  paretoChart,
   partialCredit,
   requiredSampleSize,
+  researchReport,
+  summaryTable,
   weightedMean,
   wilcoxonSignedRank
-} from "./chunk-KRR4VMH7.js";
+} from "./chunk-IOXMGMHQ.js";
 import {
   DEFAULT_REDACTION_RULES,
   DEFAULT_TRACE_ANALYST_BUDGETS,
@@ -105,7 +104,8 @@ import {
   OTEL_AGENT_EVAL_SCOPE,
   OtlpFileTraceStore,
   REDACTION_VERSION,
-  RunIntegrityError,
+  ReplayCache,
+  ReplayCacheMissError,
   SpanNotFoundError,
   TRACE_ANALYST_ACTOR_DESCRIPTION,
   TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
@@ -117,10 +117,10 @@ import {
   aggregateLlm,
   analyzeTraces,
   argHash,
-  assertRunCaptured,
   buildTraceAnalystTools,
   buildTraceInsightContext,
   buildTraceInsightPrompt,
+  createReplayFetch,
   defaultTraceInsightPanel,
   describeTraceInsightScope,
   domainEvidencePattern,
@@ -132,6 +132,7 @@ import {
   isRetrievalSpan,
   isSandboxSpan,
   isToolSpan,
+  iterateRawCalls,
   judgeSpans,
   llmSpans,
   planTraceInsightQuestions,
@@ -140,16 +141,27 @@ import {
   runFailureClass,
   runsForScenario,
   scoreTraceInsightReadiness,
-  throwIfRunIncomplete,
   tokenizeDomainWords,
   toolSpans,
   traceAnalystFunctionGroup,
   traceAnalystOnRunComplete
-} from "./chunk-WOK2RTWG.js";
+} from "./chunk-4W4NCYM2.js";
+import {
+  RunIntegrityError,
+  assertRunCaptured,
+  throwIfRunIncomplete
+} from "./chunk-QUKKGHTZ.js";
 import {
   TraceEmitter,
   llmSpanFromProvider
 } from "./chunk-5IIQKMD5.js";
+import {
+  canonicalize,
+  evaluateHypothesis,
+  hashJson,
+  signManifest,
+  verifyManifest
+} from "./chunk-6M774GY6.js";
 import {
   LlmCallError,
   LlmClient,
@@ -159,14 +171,14 @@ import {
   callLlmJson,
   probeLlm,
   stripFencedJson
-} from "./chunk-3GN6U53I.js";
+} from "./chunk-KAO3Q65R.js";
 import {
   FileSystemRawProviderSink,
   InMemoryRawProviderSink,
   NoopRawProviderSink,
   defaultProviderRedactor,
   providerFromBaseUrl
-} from "./chunk-SNUHRBDL.js";
+} from "./chunk-SQQLHODJ.js";
 import "./chunk-PZ5AY32C.js";
 
 // src/client.ts
@@ -10492,6 +10504,8 @@ export {
   PromptRegistry,
   REDACTION_VERSION,
   RESEARCH_REPORT_HARD_PAIR_FLOOR,
+  ReplayCache,
+  ReplayCacheMissError,
   RunCritic,
   RunIntegrityError,
   RunRecordValidationError,
@@ -10580,6 +10594,7 @@ export {
   createFeedbackTrajectory,
   createIntentMatchJudge,
   createLlmReviewer,
+  createReplayFetch,
   createSandboxCodeMutator,
   createSandboxPool,
   createSemanticConceptJudge,
@@ -10603,6 +10618,7 @@ export {
   evaluateActionPolicy,
   evaluateContract,
   evaluateHypothesis,
+  evaluateInterimReleaseConfidence,
   evaluateOracles,
   evaluateReleaseConfidence,
   executeScenario,
@@ -10654,6 +10670,7 @@ export {
   isRunRecord,
   isSandboxSpan,
   isToolSpan,
+  iterateRawCalls,
   jestTestParser,
   jsonHasKeys,
   jsonShape,
@@ -10682,6 +10699,7 @@ export {
   objectiveEval,
   outputLengthRubric,
   pairedBootstrap,
+  pairedEvalueSequence,
   pairedTTest,
   pairedWilcoxon,
   paraphraseRobustness,
@@ -10733,12 +10751,14 @@ export {
   roundTripRunRecord,
   rowCount,
   rowWhere,
+  rubricPredictiveValidity,
   runAgentControlLoop,
   runAssertions,
   runBehavioralCanaries,
   runCanaries,
   runCounterfactual,
   runE2EWorkflow,
+  runEvalCampaign,
   runExpectations,
   runFailureClass,
   runHarnessExperiment,