@tangle-network/agent-eval 0.221.8 → 0.222.1

package/dist/{index-CdEVr8-O.d.ts → index-Bdcpx2s8.d.ts}

@@ -1,28 +1,25 @@
-import { d as CostReceipt, l as CostLedgerSummary, o as CostLedgerHandle, p as CostUsage, r as CostBreakdown, u as CostProvenance } from "./cost-ledger-C1W3ICmP.js";
-import { c as ValidationError, t as AgentEvalError } from "./errors-DEE6u6ot.js";
-import { a as RunRecord, c as RunSplitTag } from "./run-record-C3awiXPf.js";
-import { J as TraceAnalysisStore, i as AnalystFinding, p as AnalystUsageReceipt, u as AnalystRunResult, v as ProposalFinding } from "./types-6Lfdlshp.js";
-import { Z as RawProviderSink, j as LlmCallMetadata, rt as ServedModelPolicy, x as ChatClient } from "./types-B_IFcKp5.js";
-import { Dt as FinalEvidenceRecord, M as PairedArmsComparison, at as CampaignEvidenceContext, dt as EvidenceReceipt, wt as FinalEvidenceLedger } from "./statistical-heldout-BjzSOmLH.js";
-import { a as PairedPromotionDecision, u as PairedBootstrapResult } from "./paired-promotion-decision-CQMJ5xGN.js";
-import { E as SearchExecutionIdentity, G as SearchSplit, V as SearchProposer, a as SearchAttemptAccounting, i as SearchArtifactRef, j as SearchModelIdentity, nt as EvaluationClaim, rt as EvaluationUnitSummary, tt as SearchUnknown, z as SearchOperationRecordedEvent } from "./search-ledger-types-p2xKfiDJ.js";
-import { r as LedgerHash } from "./canonical-CFpojCN5.js";
-import { C as JudgeDimension, H as SurfaceProposer, N as ParetoParent, O as LabeledScenarioSource, R as Scenario, S as JudgeConfig, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, i as CampaignCostMeter, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, o as CampaignScenarioIdentity, p as Gate, r as CampaignCellResult, u as ComponentSurface, v as GateResult, y as GenerationCandidate } from "./types-M6eF9m-r.js";
-import { At as SearchHistoryReceipt, Dt as SearchHistoryCoverage, I as SearchPolicy, Tt as SearchHistoryAdmissionOptions, Wt as CampaignStorage, gt as SearchRecorder, mt as SearchLedgerBinding, o as SearchCellResult, pt as RegisterSearchNodeInput, vt as SearchRunIdentity } from "./search-kernel-CctpiDd2.js";
-import { E as AnalystIssueExpectation, G as ExternalOptimizerRunnerCommand, I as ExternalOptimizerEvaluationObservation, J as ExternalTextEvaluationRequest, K as ExternalOptimizerWireCounts, N as ExternalOptimizerCallbackLimits, R as ExternalOptimizerModelBudget, W as ExternalOptimizerResumeMode, m as AnalystBenchmarkLabelState, q as ExternalTextCandidate, t as AgentProfile$1, u as AnalystBenchmarkCase, z as ExternalOptimizerModelCall } from "./agent-profile-CNmwfaqq.js";
-import { r as DatasetScenario, t as Dataset } from "./dataset-DQqhOCPt.js";
-import { a as CheckerIdentity, n as VerdictCertification, t as DefaultVerdict, x as VerificationStrategySource } from "./verdict-C24NELtN.js";
-import { t as DetectRewardHackingInput } from "./reward-hacking-WTxUbtbb.js";
-import { f as TraceSpanEvent, t as HostedClient } from "./client-hwo3cJjR.js";
+import { d as CostReceipt, l as CostLedgerSummary, o as CostLedgerHandle, p as CostUsage, r as CostBreakdown, u as CostProvenance } from "./cost-ledger-Czro9wmo.js";
+import { c as ValidationError, t as AgentEvalError } from "./errors-E9I2hldP.js";
+import { a as RunRecord, c as RunSplitTag } from "./run-record-D2BFMPGl.js";
+import { J as TraceAnalysisStore, i as AnalystFinding, p as AnalystUsageReceipt, u as AnalystRunResult, v as ProposalFinding } from "./types-BEvjDziM.js";
+import { j as LlmCallMetadata, rt as ServedModelPolicy, x as ChatClient } from "./types-0KglnQ9o.js";
+import { Dt as FinalEvidenceRecord, M as PairedArmsComparison, at as CampaignEvidenceContext, dt as EvidenceReceipt, wt as FinalEvidenceLedger } from "./statistical-heldout-BYxHe3KO.js";
+import { a as PairedPromotionDecision, u as PairedBootstrapResult } from "./paired-promotion-decision-BDnqepZ1.js";
+import { E as SearchExecutionIdentity, G as SearchSplit, V as SearchProposer, a as SearchAttemptAccounting, i as SearchArtifactRef, j as SearchModelIdentity, nt as EvaluationClaim, rt as EvaluationUnitSummary, tt as SearchUnknown, z as SearchOperationRecordedEvent } from "./search-ledger-types-qiYRUo_N.js";
+import { r as LedgerHash } from "./canonical-gLeHiqzd.js";
+import { C as JudgeDimension, H as SurfaceProposer, N as ParetoParent, O as LabeledScenarioSource, R as Scenario, S as JudgeConfig, _ as GateDecision, a as CampaignResult, b as GenerationRecord, c as CampaignTraceWriter, d as DispatchContext, f as DispatchFn, g as GateContribution, i as CampaignCostMeter, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, o as CampaignScenarioIdentity, p as Gate, r as CampaignCellResult, u as ComponentSurface, v as GateResult, y as GenerationCandidate } from "./types-e3HGaI7P.js";
+import { At as SearchHistoryReceipt, Dt as SearchHistoryCoverage, I as SearchPolicy, Tt as SearchHistoryAdmissionOptions, Wt as CampaignStorage, gt as SearchRecorder, mt as SearchLedgerBinding, o as SearchCellResult, pt as RegisterSearchNodeInput, vt as SearchRunIdentity } from "./search-kernel-B77eE7Ac.js";
+import { E as AnalystIssueExpectation, G as ExternalOptimizerRunnerCommand, I as ExternalOptimizerEvaluationObservation, J as ExternalTextEvaluationRequest, K as ExternalOptimizerWireCounts, N as ExternalOptimizerCallbackLimits, R as ExternalOptimizerModelBudget, W as ExternalOptimizerResumeMode, m as AnalystBenchmarkLabelState, q as ExternalTextCandidate, t as AgentProfile$1, u as AnalystBenchmarkCase, z as ExternalOptimizerModelCall } from "./agent-profile-CSEQRC9l.js";
+import { r as DatasetScenario, t as Dataset } from "./dataset-DThC1S6G.js";
+import { t as DetectRewardHackingInput } from "./reward-hacking-C9hPVT0Q.js";
+import { f as TraceSpanEvent, t as HostedClient } from "./client-BIFwUg7I.js";
+import { n as CompletionVerdict, o as ProducedState, r as CorrectnessChecker, t as CompletionRequirement } from "./completion-verifier-7KEfroqx.js";
 import { AgentProfile } from "@tangle-network/agent-interface";
 import { z } from "zod";
-//#region src/campaign/search-ledger-errors.d.ts
 /** Base error for invalid search-ledger input or operations. */
 declare class SearchLedgerError extends ValidationError {}
 /** Error raised when durable search-ledger data fails an integrity check. */
 declare class SearchLedgerIntegrityError extends SearchLedgerError {}
-//#endregion
-//#region src/campaign/coverage.d.ts
 interface CampaignCoverage {
   complete: boolean;
   expectedCellIds: string[];
@@ -42,8 +39,6 @@ declare function campaignSplitDigestFromIdentities(scenarios: readonly CampaignS
 declare function campaignSplitDigest<TScenario extends Scenario>(scenarios: readonly TScenario[], reps: number): `sha256:${string}`;
 /** Exact designed-denominator receipt for one campaign. */
 declare function campaignCoverage<TArtifact, TScenario extends Scenario>(cells: readonly CampaignCellResult<TArtifact>[], scenarios: readonly TScenario[], reps: number, requireJudgeScore: boolean): CampaignCoverage;
-//#endregion
-//#region src/campaign/final-evidence.d.ts
 /** Durable final-data policy. Keep its ledger shared across related campaigns. */
 interface FinalEvidencePolicy {
   ledger: FinalEvidenceLedger;
@@ -54,8 +49,6 @@ interface FinalEvidenceUse {
   claim: EvaluationClaim;
   record: FinalEvidenceRecord;
 }
-//#endregion
-//#region src/campaign/cell-schedule.d.ts
 declare function buildCellSchedule<TScenario extends Scenario>(scenarios: TScenario[], seed: number, reps: number): Array<{
   scenario: TScenario;
   rep: number;
@@ -64,8 +57,6 @@ declare function buildCellSchedule<TScenario extends Scenario>(scenarios: TScena
 }>;
 type CellScheduleSlot<TScenario extends Scenario> = ReturnType<typeof buildCellSchedule<TScenario>>[number];
 declare function cellCachePath(runDir: string, cellId: string): string;
-//#endregion
-//#region src/campaign/cell-cache.d.ts
 type CacheIssueReason = 'missing' | 'manifest-mismatch' | 'cell-mismatch' | 'missing-cost-provenance' | 'invalid-cost-provenance' | 'invalid-cost-receipts' | 'corrupt';
 type CacheRead<TArtifact> = {
   status: 'hit';
@@ -74,8 +65,6 @@ type CacheRead<TArtifact> = {
   status: 'miss';
   reason: CacheIssueReason;
 };
-//#endregion
-//#region src/campaign/plan-campaign-run.d.ts
 interface CampaignRunPlanCell {
   cellId: string;
   scenarioId: string;
@@ -118,8 +107,6 @@ interface PlanCampaignRunOptions<TScenario extends Scenario, TArtifact> {
  * run-vs-cached schedule so callers can preview cost and resumability before spending.
  */
 declare function planCampaignRun<TScenario extends Scenario, TArtifact>(opts: PlanCampaignRunOptions<TScenario, TArtifact>): CampaignRunPlan;
-//#endregion
-//#region src/campaign/run-campaign.d.ts
 interface RunCampaignOptions<TScenario extends Scenario, TArtifact> {
   scenarios: TScenario[];
   dispatch: DispatchFn<TScenario, TArtifact>;
@@ -194,7 +181,9 @@ interface RunCampaignOptions<TScenario extends Scenario, TArtifact> {
    * manifest, and pairing are unchanged. An attempt that failed because the
    * campaign was cancelled is never retried.
    */
-  cellRetry?: CampaignCellRetryPolicy;
+  cellRetry?: CampaignCellRetryPolicy<TArtifact>;
+  /** Once per final persisted cell, including cache hits, in completion order. */
+  onCellSettled?: (cell: CampaignCellResult<TArtifact>, scenario: TScenario) => void | Promise<void>;
   /**
    * Per-cell dispatch deadline in ms. A `dispatch` that neither resolves nor
    * rejects within this window is a hang (a stalled model request, an
@@ -287,7 +276,7 @@ interface CampaignCellFailureReceipt<TArtifact = unknown> {
  * execution directory, including attempts resumed in a later invocation.
  * The final cell records the retry count as `retryAttempts`.
  */
-interface CampaignCellRetryPolicy {
+interface CampaignCellRetryPolicy<TArtifact = unknown> {
   /** Total attempts per cell, including the first. A positive safe integer. */
   attempts: number;
   /** Decides whether a failed attempt is dispatched again. Receives the
@@ -295,13 +284,13 @@ interface CampaignCellRetryPolicy {
    *  ready-made predicate for infrastructure hiccups (502/503/504, dropped
    *  streams, admission rejections). */
   retryable: (failure: CampaignCellFailureReceipt['failure']) => boolean;
+  /** Retry a measured artifact without classifying it as a dispatch failure. */
+  retryResult?: (artifact: TArtifact) => boolean;
 }
 /**
  * Core campaign orchestrator: fan scenarios through dispatch, score with judges, aggregate bootstrap CIs, and persist reproducible `CampaignResult` records.
  */
 declare function runCampaign<TScenario extends Scenario, TArtifact>(opts: RunCampaignOptions<TScenario, TArtifact>): Promise<CampaignResult<TArtifact, TScenario>>;
-//#endregion
-//#region src/campaign/presets/run-eval.d.ts
 interface RunEvalOptions<TScenario extends Scenario, TArtifact> extends Omit<RunCampaignOptions<TScenario, TArtifact>, 'runDir'> {
   runDir: string;
 }
@@ -309,8 +298,6 @@ interface RunEvalOptions<TScenario extends Scenario, TArtifact> extends Omit<Run
  * Simplest evaluation preset: run scenarios through dispatch, score with judges, and return a `CampaignResult` — no optimizer, no gate, no PR.
  */
 declare function runEval<TScenario extends Scenario, TArtifact>(opts: RunEvalOptions<TScenario, TArtifact>): Promise<CampaignResult<TArtifact, TScenario>>;
-//#endregion
-//#region src/campaign/auto-pr.d.ts
 interface OpenAutoPrOptions<TArtifact, TScenario extends Scenario> {
   /** Campaign result to attach to the PR. */
   result: CampaignResult<TArtifact, TScenario>;
@@ -347,8 +334,6 @@ interface OpenAutoPrResult {
  * Open a GitHub PR for a gate-approved surface promotion, attaching the manifest hash, gate verdict, and diff as the PR body.
  */
 declare function openAutoPr<TArtifact, TScenario extends Scenario>(options: OpenAutoPrOptions<TArtifact, TScenario>): OpenAutoPrResult;
-//#endregion
-//#region src/campaign/presets/run-optimization.d.ts
 /** Improve a surface as a search on the kernel; see the module comment. */
 declare function runOptimization<TScenario extends Scenario, TArtifact>(opts: RunOptimizationOptions<TScenario, TArtifact>): Promise<RunOptimizationResult<TArtifact, TScenario>>;
 /** Optimization inputs, immutable proposal snapshots, and imported baseline validation. */
@@ -446,8 +431,6 @@ interface RunOptimizationResult<TArtifact, TScenario extends Scenario> {
   /** The non-dominated set of measured surfaces by per-scenario composite. */
   paretoFrontier: ParetoParent[];
 }
-//#endregion
-//#region src/campaign/presets/run-improvement-loop.d.ts
 type RunImprovementLoopOptions<TScenario extends Scenario, TArtifact> = RunOptimizationOptions<TScenario, TArtifact> & {
   /** Holdout scenarios kept OUT of the training optimization pool — used
    *  ONLY to score baseline vs winner for the gate. */
@@ -507,8 +490,6 @@ interface RunImprovementLoopResult<TArtifact, TScenario extends Scenario> extend
  * Gated-promotion shell over `runOptimization`: scores the winner against the baseline on a holdout set, runs the release gate, and optionally opens a PR.
  */
 declare function runImprovementLoop<TScenario extends Scenario, TArtifact>(opts: RunImprovementLoopOptions<TScenario, TArtifact>): Promise<RunImprovementLoopResult<TArtifact, TScenario>>;
-//#endregion
-//#region src/campaign/transient-failure.d.ts
 interface TransientFailureOptions {
   /**
    * Treat full-duration timeouts ("timeout after 180000ms") as transient.
@@ -542,8 +523,6 @@ declare function isTransientTransportFailure(message: string | null | undefined,
  * (`quotaExhaustedUntil`).
  */
 declare function transientDispatchFailure(opts?: TransientFailureOptions): (failure: CampaignCellFailureReceipt['failure']) => boolean;
-//#endregion
-//#region src/llm-judge.d.ts
 /** A rubric dimension as a bare key or the full `{ key, description }` shape. A
  *  bare string uses the key as its own description. */
 type LlmJudgeDimension = string | JudgeDimension;
@@ -616,8 +595,6 @@ interface LlmJudgeOptions<TArtifact, TScenario extends Scenario = Scenario> {
  * `weightedComposite`.
  */
 declare function llmJudge<TArtifact = unknown, TScenario extends Scenario = Scenario>(name: string, prompt: string, opts: LlmJudgeOptions<TArtifact, TScenario>): JudgeConfig<TArtifact, TScenario>;
-//#endregion
-//#region src/reference-equivalence-judge.d.ts
 interface ReferenceEquivalenceScenario extends Scenario {
   userRequest: string;
   expectedAnswer: string;
@@ -643,8 +620,6 @@ interface ReferenceEquivalenceJudgeResult extends LlmCallMetadata {
   score: number;
   rationale: string;
 }
-//#endregion
-//#region src/campaign/external-optimizer-observations.d.ts
 interface ExternalOptimizerObservationSummary {
   scope: 'callback-submitted-candidates';
   path: string;
@@ -692,8 +667,6 @@ declare function readExternalOptimizerObservationArtifact(input: {
   summary: ExternalOptimizerObservationSummary;
   storage?: CampaignStorage;
 }): ExternalOptimizerObservationArtifact;
-//#endregion
-//#region src/campaign/gepa-candidate-population.d.ts
 interface GepaCandidatePopulationSummary {
   readonly scope: 'gepa-candidate-population';
   readonly path: string;
@@ -742,8 +715,6 @@ declare function readGepaCandidatePopulationArtifact(input: {
   summary: GepaCandidatePopulationSummary;
   storage?: CampaignStorage;
 }): GepaCandidatePopulationArtifact;
-//#endregion
-//#region src/campaign/optimization-cost.d.ts
 /** Cost reported by a method or by final test scoring. */
 interface ComparisonCost {
   /** Known subtotal. Consult `costProvenance` before treating this as total spend. */
@@ -768,8 +739,6 @@ interface OptimizationTokenUsage {
   totalTokens: number;
   calls: number;
 }
-//#endregion
-//#region src/campaign/optimization-method.d.ts
 /** Shared campaign settings applied to every optimization method. */
 type OptimizationMethodRunOptions<TScenario extends Scenario, TArtifact> = Omit<RunCampaignOptions<TScenario, TArtifact>, 'costCeiling' | 'costLedger' | 'dispatch' | 'judges' | 'runDir' | 'scenarios' | 'seed'>;
 interface OptimizationPackageSource {
@@ -882,8 +851,6 @@ interface OptimizationMethod<TScenario extends Scenario = Scenario, TArtifact =
   name: string;
   optimize: (input: OptimizationMethodInput<TScenario, TArtifact>) => Promise<OptimizationMethodResult>;
 }
-//#endregion
-//#region src/campaign/external-text-evaluation.d.ts
 interface ExternalOptimizationExample {
   id: string;
   data: unknown;
@@ -907,8 +874,6 @@ interface ExternalTextEvaluationResponse {
   };
 }
 declare function decodeExternalTextCandidate(candidate: ExternalTextCandidate): MutableSurface;
-//#endregion
-//#region src/campaign/external-text-optimization-contract.d.ts
 interface ExternalTextOptimizerContext {
   readonly runId: string;
   readonly name: string;
@@ -966,8 +931,6 @@ interface ExternalTextOptimizationMethodConfig<TScenario extends Scenario, TArti
   describeArtifact?: (artifact: TArtifact, scenario: TScenario) => unknown;
   run: (context: ExternalTextOptimizerContext) => Promise<ExternalTextOptimizerResult>;
 }
-//#endregion
-//#region src/campaign/external-text-optimization.d.ts
 /**
  * Adapt a third-party text optimizer without reimplementing its search.
  *
@@ -975,8 +938,6 @@ interface ExternalTextOptimizationMethodConfig<TScenario extends Scenario, TArti
  * counted before execution and stop at `maxEvaluations`.
  */
 declare function externalTextOptimizationMethod<TScenario extends Scenario, TArtifact>(config: ExternalTextOptimizationMethodConfig<TScenario, TArtifact>): OptimizationMethod<TScenario, TArtifact>;
-//#endregion
-//#region src/campaign/optimizer-model.d.ts
 type OptimizerModelBudget = ExternalOptimizerModelBudget;
 /** One metered model path supplied by the package that owns execution. */
 interface OpenAICompatibleOptimizerModel {
@@ -1001,8 +962,6 @@ interface OpenAICompatibleOptimizerModel {
    */
   anthropicEndpoint?: boolean;
 }
-//#endregion
-//#region src/campaign/gepa-optimization-method.d.ts
 /** Shared settings for one bounded GEPA engine invocation. */
 interface GepaEngineOptions {
   /** GEPA engine name. GEPA validates names available in its Python runtime. */
@@ -1148,8 +1107,6 @@ interface GepaOptimizationMethodConfig<TScenario extends Scenario, TArtifact = u
  * stripped like any other credential.
  */
 declare function gepaOptimizationMethod<TScenario extends Scenario, TArtifact>(config: GepaOptimizationMethodConfig<TScenario, TArtifact>): OptimizationMethod<TScenario, TArtifact>;
-//#endregion
-//#region src/campaign/presets/compare-optimization-methods.d.ts
 interface OptimizationMethodScore {
   evidence?: {
     baseline: EvidenceReceipt;
@@ -1273,8 +1230,6 @@ interface CompareOptimizationMethodsOptions<TScenario extends Scenario, TArtifac
  * Compare complete optimization methods on disjoint train, selection, and final test data.
  */
 declare function compareOptimizationMethods<TScenario extends Scenario, TArtifact>(rawOpts: CompareOptimizationMethodsOptions<TScenario, TArtifact>): Promise<OptimizationMethodComparison>;
-//#endregion
-//#region src/campaign/skillopt-optimization-method.d.ts
 interface SkillOptTrainerConfig {
   epochs: number;
   batchSize: number;
@@ -1344,14 +1299,10 @@ interface SkillOptOptimizationMethodConfig<TScenario extends Scenario, TArtifact
 }
 /** Run Microsoft's SkillOpt trainer as a complete optimization method. */
 declare function skillOptOptimizationMethod<TScenario extends Scenario, TArtifact>(config: SkillOptOptimizationMethodConfig<TScenario, TArtifact>): OptimizationMethod<TScenario, TArtifact>;
-//#endregion
-//#region src/campaign/gates/compose.d.ts
 /** Compose gates — all must `ship` for the composite to `ship`. First
  *  non-ship verdict short-circuits the composite verdict, but ALL gates run
  *  (so the result records every gate's reason — useful for diagnostics). */
 declare function composeGate<TArtifact = unknown, TScenario extends Scenario = Scenario>(...gates: Array<Gate<TArtifact, TScenario>>): Gate<TArtifact, TScenario>;
-//#endregion
-//#region src/canary.d.ts
 type CanaryKind = 'silent_judge_fallback' | 'judge_calibration_drift' | 'distribution_shift';
 type CanarySeverity = 'info' | 'warn' | 'error';
 interface CanaryAlert {
@@ -1430,8 +1381,6 @@ interface CanaryOptions {
  * the input is used to define "recent" vs "historical" windows.
  */
 declare function runCanaries(runs: RunRecord[], opts?: CanaryOptions): CanaryReport;
-//#endregion
-//#region src/red-team.d.ts
 type RedTeamCategory = 'prompt_injection_direct' | 'prompt_injection_indirect' | 'jailbreak_persona' | 'jailbreak_dan' | 'pii_leak' | 'permission_escalation' | 'data_exfil' | 'policy_override';
 interface RedTeamPayload {
   category: RedTeamCategory;
@@ -1472,8 +1421,6 @@ declare function redTeamDataset(extraCases?: RedTeamCase[]): Dataset;
 declare function scoreRedTeamOutput(output: string, toolCalls: string[], rtCase: RedTeamCase): RedTeamFinding;
 /** Aggregate red-team findings into per-category pass rates. */
 declare function redTeamReport(findings: RedTeamFinding[]): RedTeamReport;
-//#endregion
-//#region src/campaign/gates/default-production-gate.d.ts
 type DefaultProductionGateCheck = 'dimension-regression' | 'budget' | 'red-team' | 'reward-hacking' | 'canary';
 type DefaultProductionRewardHackingOptions = Omit<DetectRewardHackingInput, 'runs' | 'truthOf'> & {
   truthOf: NonNullable<DetectRewardHackingInput['truthOf']>;
@@ -1538,8 +1485,6 @@ interface DefaultProductionGateOptions {
  * Opinionated production gate composing held-out significance, red-team, reward-hacking, and canary checks into a single `Gate.decide` decision.
  */
 declare function defaultProductionGate<TArtifact, TScenario extends Scenario>(options: DefaultProductionGateOptions): Gate<TArtifact, TScenario>;
-//#endregion
-//#region src/campaign/gates/heldout-gate.d.ts
 interface HeldOutGateOptions<TScenario extends Scenario = Scenario> {
   scenarios: TScenario[];
   /** Effect-size threshold the CI lower bound must clear, in the judge's native
@@ -1563,8 +1508,6 @@ interface HeldOutGateOptions<TScenario extends Scenario = Scenario> {
  * bootstrap otherwise. See {@link decidePairedPromotion}.
  */
 declare function heldOutGate<TArtifact, TScenario extends Scenario>(options: HeldOutGateOptions<TScenario>): Gate<TArtifact, TScenario>;
-//#endregion
-//#region src/campaign/score-utils.d.ts
 interface CampaignComparisonUnits {
   pairedCellN: number;
   unitIds: string[];
@@ -1594,8 +1537,6 @@ interface CampaignBreakdown {
 /** Per-candidate evidence a reflective/patch proposer grounds its next proposal
  *  on: mean score per judge dimension + per-scenario composite. */
 declare function campaignBreakdown<TArtifact, TScenario extends Scenario>(campaign: CampaignResult<TArtifact, TScenario>): CampaignBreakdown;
-//#endregion
-//#region src/campaign/provenance.d.ts
 interface LoopProvenanceCandidate {
   /** Generation index this candidate was proposed in. */
   generation: number;
@@ -1791,8 +1732,6 @@ interface EmitLoopProvenanceArgs<TArtifact, TScenario extends Scenario> extends
  * collector never fails the loop (the durable artifact is the source of truth).
  */
 declare function emitLoopProvenance<TArtifact, TScenario extends Scenario>(args: EmitLoopProvenanceArgs<TArtifact, TScenario>): Promise<EmitLoopProvenanceResult>;
-//#endregion
-//#region src/campaign/analyst-surface.d.ts
 interface TraceAnalystScenario extends Scenario {
   kind: 'trace-analyst';
   traceStore: TraceAnalysisStore;
@@ -1813,8 +1752,6 @@ interface BuildTraceAnalystSurfaceDispatchOptions {
     signal: AbortSignal;
   }): Promise<TraceAnalystArtifact>;
 }
-//#endregion
-//#region src/campaign/cost-breakdown.d.ts
 /** Read the source rates for this execution, never a copied model price table.
  * inputTokens includes cachedTokens. Seat credits are standard model credits;
  * discounts and MCP charges require provider observations, not token guesses.
@@ -1837,8 +1774,6 @@ declare function measureCampaignCost(input: {
   };
   signal?: AbortSignal;
 }): Promise<CostBreakdown>;
-//#endregion
-//#region src/campaign/cross-surface-types.d.ts
 /** Whether one candidate attempt produced a usable executable outcome. */
 type CrossSurfaceAttemptCompleteness = 'complete' | 'missing' | 'invalid';
 /** One independently proposed change on one caller-defined surface. */
@@ -2098,16 +2033,12 @@ interface CrossSurfaceInteractionReport<TRow extends CrossSurfaceTaskRow = Cross
   pairwise: CrossSurfacePairwiseEntry[];
   selections: CrossSurfaceSelections;
 }
-//#endregion
-//#region src/campaign/cross-surface-interaction.d.ts
 /**
  * Build the complete cross-surface evidence matrix and derive all three frozen
  * candidates. The task/candidate/component orders are part of the input so
  * neither insertion order nor an after-the-fact tie-break can change a result.
  */
 declare function analyzeCrossSurfaceInteractions<TRow extends CrossSurfaceTaskRow>(input: AnalyzeCrossSurfaceInteractionsInput<TRow>): CrossSurfaceInteractionReport<TRow>;
-//#endregion
-//#region src/campaign/fixtures.d.ts
 type EvalFixtureValidationMode = 'vitest' | 'none';
 interface EvalFixtureFile {
   path: string;
@@ -2167,8 +2098,6 @@ declare function loadEvalFixtureScenarios(evalsDir: string, options?: LoadEvalFi
  * and returns the plan plus each fixture's name/path/fingerprint.
  */
 declare function planEvalFixtureRun<TArtifact = unknown>(options: PlanEvalFixtureRunOptions<TArtifact>): EvalFixtureRunPlan;
-//#endregion
-//#region src/campaign/gates/neutralization-gate.d.ts
 interface NeutralizationGateOptions<TScenario extends Scenario = Scenario> {
   scenarios: TScenario[];
   /** Reject when the neutralized (content-blanked, footprint-matched) variant
@@ -2182,8 +2111,6 @@ interface NeutralizationGateOptions<TScenario extends Scenario = Scenario> {
  * mostly reproduced by a footprint-matched neutralized variant.
  */
 declare function neutralizationGate<TArtifact, TScenario extends Scenario>(options: NeutralizationGateOptions<TScenario>): Gate<TArtifact, TScenario>;
-//#endregion
-//#region src/campaign/gepa-search-import.d.ts
 /** The external optimizers whose searches this module records. */
 type ExternalSearchOptimizer = 'GEPA' | 'SkillOpt';
 interface GepaPopulationImport {
@@ -2234,8 +2161,6 @@ declare function importExternalEvaluations(input: {
 }): Promise<GepaEvaluationImport>;
 /** An optimizer candidate as the mutable surface it represents. */
 declare function externalSurface(candidate: ExternalTextCandidate): MutableSurface;
-//#endregion
-//#region src/campaign/grounded-reflection.d.ts
 /**
  * Evidence grounding for reflective optimizers (GEPA-style revise loops).
  *
@@ -2294,8 +2219,6 @@ interface UngroundedLiteralReport {
   /** The subset failing rollouts actually used - prescribing these is proven harmful. */
   readonly harmful: readonly string[];
 }
-//#endregion
-//#region src/campaign/labeled-store/fs-adapter.d.ts
 interface FsLabeledScenarioStoreOptions {
   /** Root directory for JSONL files. Created if missing. */
   root: string;
@@ -2305,8 +2228,6 @@ interface FsLabeledScenarioStoreOptions {
   /** Test seam — override `Date.now()` for deterministic tests. */
   now?: () => number;
 }
-//#endregion
-//#region src/campaign/neutralize.d.ts
 /**
  * @module
  * Footprint-matched neutralization — the placebo control for content-vs-footprint
@@ -2334,8 +2255,6 @@ interface FsLabeledScenarioStoreOptions {
  * variant from being a true footprint match.
  */
 declare function neutralizeText(content: string): string;
-//#endregion
-//#region src/campaign/optimization-method-composition.d.ts
 /** Optimize a projection while evaluating every proposal in the complete candidate. */
 declare function scopedOptimizationMethod<S extends Scenario, A>(options: {
   name: string;
@@ -2348,175 +2267,6 @@ declare function sequentialOptimizationMethod<S extends Scenario, A>(options: {
   name: string;
   methods: readonly OptimizationMethod<S, A>[];
 }): OptimizationMethod<S, A>;
-//#endregion
-//#region src/artifact-validator.d.ts
-/** An artifact an agent produced, as completion and produced-state checks read it. */
-interface ValidatedArtifact {
-  /** Logical kind — validators type-guard on this */
-  kind: 'file' | 'json' | 'text' | 'binary' | string;
-  /** Filesystem-style path, optional */
-  path?: string;
-  /** String content for text/json/file kinds */
-  content?: string;
-  /** Binary content (if kind === 'binary') */
-  bytes?: Uint8Array;
-  /** Caller-supplied metadata (mimeType, sha256, size, etc.) */
-  metadata?: Record<string, unknown>;
-}
-//#endregion
-//#region src/completion-verifier.d.ts
-/** 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 `ValidatedArtifact` shape. */
-  artifacts: ValidatedArtifact[];
-  /** 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, when the matched item carries no content
-   * to assess, or when the correctness check itself failed (`unmeasured`).
-   */
-  correct: boolean | null;
-  /** structurallyPresent && !unmeasured && correct !== false. */
-  satisfied: boolean;
-  /**
-   * Set when the correctness check itself errored (LLM call failure or an
-   * unparseable response after retry). The requirement's fulfilment is
-   * UNKNOWN — `correct` stays null, `satisfied` is false, and
-   * `completionVerdict` excludes the row from `completionRate`'s
-   * denominator. Never folded into a zero: a synthetic zero is
-   * indistinguishable from a real failure (see `JudgeParseError`).
-   */
-  unmeasured?: true;
-  /** Why the correctness check could not be measured (present iff `unmeasured`). */
-  unmeasuredReason?: string;
-  /** Human-readable evidence for the verdict. */
-  evidence: string[];
-}
-/** Extends the substrate verdict spine: `valid` = `fullyComplete` and
- *  `score` = `completionRate` — derived in `completionVerdict()`, the one
- *  place those equalities hold by construction. */
-interface CompletionVerdict extends DefaultVerdict {
-  taskId: string;
-  requirements: RequirementCheck[];
-  /** satisfied / MEASURABLE requirements (unmeasured rows leave the denominator). */
-  completionRate: number;
-  /** Every measurable requirement satisfied (false when anything is unmeasured). */
-  fullyComplete: boolean;
-  /** Requirements whose correctness check errored — reported, never scored as zero. */
-  unmeasuredCount: number;
-}
-/**
- * Construct a `CompletionVerdict` from the per-requirement checks, deriving
- * `completionRate` / `fullyComplete` and the spine fields (`valid` =
- * `fullyComplete`, `score` = `completionRate`) in one place. Throws on zero
- * requirements — a verdict over nothing is a misconfiguration, mirroring
- * `verifyCompletion`'s gold-spec guard.
- */
-declare function completionVerdict(input: {
-  taskId: string;
-  requirements: RequirementCheck[];
-  /** What certified the correctness stage, when anything did. Omitted =
-   *  an uncertified verdict — the honest default for a bare checker. */
-  certification?: VerdictCertification;
-}): CompletionVerdict;
-/**
- * What a correctness checker declares about itself so the completion
- * verdict can carry a certification: which strategy member it discharges,
- * its exact identity, and the steps its answers rest on unverified.
- */
-interface CorrectnessCheckerAttestation {
-  strategy: VerificationStrategySource;
-  checker: CheckerIdentity;
-  assumptions: string[];
-}
-/**
- * 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`.
- *
- * `attestation` is optional metadata on the function value: a checker that
- * carries one yields certified completion verdicts; a bare function yields
- * the same verdict uncertified. A plain arrow function remains a valid
- * checker.
- */
-interface CorrectnessChecker {
-  (requirement: CompletionRequirement, content: string): Promise<{
-    correct: boolean;
-    reason: string;
-  }>;
-  attestation?: CorrectnessCheckerAttestation;
-}
-/**
- * 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;
-  /** Optional ledger for direct use. */
-  costLedger?: CostLedgerHandle;
-  costPhase?: string;
-  costTags?: Record<string, string>;
-  signal?: AbortSignal;
-  /** Max chars of artifact content sent to the checker. */
-  maxContentChars?: number;
-  /**
-   * Checker LLM calls per requirement before giving up (parse failures and
-   * call errors both consume attempts). The failure then surfaces as an
-   * `unmeasured` requirement, never a zero.
-   */
-  maxAttempts?: number;
-  /**
-   * Forensic capture of every checker request/response/error — without it a
-   * checker failure is unauditable (the agent-turn raws never contain the
-   * checker's own calls). Same sink contract as `LlmClient`.
-   */
-  rawSink?: RawProviderSink;
-}
-/**
- * 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(chat: ChatClient, opts?: LlmCorrectnessCheckerOpts): CorrectnessChecker;
-//#endregion
-//#region src/produced-state.d.ts
 /** A tool the agent invoked. */
 interface ToolCallEventLike {
   type: 'tool_call';
@@ -2566,8 +2316,6 @@ type RuntimeEventLike = ToolCallEventLike | ArtifactEventLike | ProposalEventLik
  * recover dropped events, or establish freshness beyond the supplied stream.
  */
 declare function extractProducedState(events: readonly RuntimeEventLike[]): ProducedState;
-//#endregion
-//#region src/campaign/presets/playback.d.ts
 /** One step of a user story — what the user does. The driver interprets
  *  `payload` (a Playwright selector + action, or a sandbox chat turn). */
 interface PlaybackStep {
@@ -2664,8 +2412,6 @@ interface ScoreboardRenderOptions {
  * rows in, same bytes out (no clock/random), so it is safe to snapshot.
  */
 declare function renderScoreboardMarkdown(rows: readonly ScoreboardRow[], opts?: ScoreboardRenderOptions): string;
-//#endregion
-//#region src/integrity/backend-integrity.d.ts
 interface BackendIntegrityReport {
   /** Total records inspected. */
   totalRecords: number;
@@ -2711,8 +2457,6 @@ declare function summarizeBackendIntegrity(records: ReadonlyArray<RunRecord>): B
 declare function assertRealBackend(records: ReadonlyArray<RunRecord>, opts?: {
   allowMixed?: boolean;
 }): BackendIntegrityReport;
-//#endregion
-//#region src/campaign/presets/run-profile-matrix.d.ts
 /** Dispatch for one cell: render `profile` against `scenario`, returning the
  *  artifact the judges score. Run LLM work through `ctx.cost.runPaidCall` —
  *  the integrity check depends on its receipt. */
@@ -2842,8 +2586,6 @@ interface RunProfileMatrixResult<TArtifact, TScenario extends Scenario> {
  * Profile × scenario matrix runner: fan N agent profiles across M scenarios, project each cell to a validated `RunRecord` with real token usage, and enforce the backend-integrity guard before returning.
  */
 declare function runProfileMatrix<TScenario extends Scenario, TArtifact>(opts: RunProfileMatrixOptions<TScenario, TArtifact>): Promise<RunProfileMatrixResult<TArtifact, TScenario>>;
-//#endregion
-//#region src/campaign/presets/segmented-profile-matrix.d.ts
 interface ProfileMatrixRow {
   rowId: string;
   ordinal: number;
@@ -2922,8 +2664,6 @@ interface FinalizedProfileMatrixResult<TArtifact, TScenario extends Scenario> ex
 declare function createProfileMatrixPlan<TScenario extends Scenario, TArtifact>(opts: CreateProfileMatrixPlanOptions<TScenario, TArtifact>): ProfileMatrixPlan<TScenario, TArtifact>;
 declare function runProfileMatrixSegment<TScenario extends Scenario, TArtifact>(opts: RunProfileMatrixSegmentOptions<TScenario, TArtifact>): Promise<ProfileMatrixSegmentResult<TArtifact, TScenario>>;
 declare function finalizeProfileMatrix<TScenario extends Scenario, TArtifact>(opts: FinalizeProfileMatrixOptions<TScenario, TArtifact>): Promise<FinalizedProfileMatrixResult<TArtifact, TScenario>>;
-//#endregion
-//#region src/campaign/retained-dispatch.d.ts
 type RetainedDispatchOutcome<T> = {
   succeeded: true;
   value: T;
@@ -2972,15 +2712,11 @@ declare function createRetainedDispatch<Lane extends string, T>(options: Retaine
   /** Counts include uncertain dispatches, which keep their reservation. No pricing is inferred. */
   committed(): ReadonlyMap<Lane, number>;
 };
-//#endregion
-//#region src/campaign/run-dir.d.ts
 /** Resolve a campaign `runDir`. An absolute path is honored as-is (the caller
  *  chose an explicit location). A bare name is placed under the shared home root
  *  so bundles never pollute a repo working tree — the default the harness should
  *  compute so callers pass a *name*, not a path. */
 declare function resolveRunDir(runDir: string, repo?: string): string;
-//#endregion
-//#region src/campaign/scenario-selection.d.ts
 /**
  * Discriminative scenario selection (research claim E2).
  *
@@ -3021,8 +2757,6 @@ interface DiscriminationScore {
 declare function scoreDiscrimination(signals: ScenarioSignal[], opts?: {
   saturationCeiling?: number;
 }): DiscriminationScore[];
-//#endregion
-//#region src/campaign/search-campaign-cell.d.ts
 /**
  * What `cell-settled` records for one campaign cell: `passed` with the
  * composite (and each judge's) when every judge scored it, else `errored`,
@@ -3042,8 +2776,6 @@ declare function searchReceiptAccounting(receipts: ReadonlyArray<CostReceipt>):
 /** What ran a proposal: a model when it made a paid call other than a judge's,
  * else deterministic code. */
 declare function searchProposalExecution(receipts: ReadonlyArray<CostReceipt>, provider: string, source: SearchOperationRecordedEvent['execution']['source']): SearchOperationRecordedEvent['execution'];
-//#endregion
-//#region src/campaign/single-run-lock.d.ts
 /**
  * Single-run lock for evaluations that share one mutable environment.
  *
@@ -3079,8 +2811,6 @@ interface SingleRunLock {
  * leaves a marker that fails closed instead of admitting overlapping runs.
  */
 declare function acquireSingleRunLock(opts: SingleRunLockOptions): SingleRunLock;
-//#endregion
-//#region src/campaign/surface-identity.d.ts
 /** Validate the immutable identity shape; the owning executor verifies the Git objects and patch. */
 declare function assertCodeSurfaceIdentity(surface: unknown): asserts surface is CodeSurface;
 /**
@@ -3105,8 +2835,6 @@ declare function surfaceHash(surface: MutableSurface): string;
 declare function renderSurfaceDiff(winnerSurface: MutableSurface, baselineSurface: MutableSurface): string;
 /** Bind a campaign cache entry to the exact surface and caller-owned execution revision. */
 declare function surfaceDispatchRef(surface: MutableSurface, executionRef?: string): string;
-//#endregion
-//#region src/campaign/upstream-evaluators.d.ts
 interface PhoenixEvaluationResultLike {
   score?: number;
   label?: string;
@@ -3128,8 +2856,6 @@ interface UpstreamEvaluationContext {
   readonly signal: AbortSignal;
   readonly callId?: string;
 }
-//#endregion
-//#region src/campaign/worktree/index.d.ts
 type GitOutput = string | Uint8Array;
 type GitEnvironment = Readonly<Record<string, string>>;
 type GitRunner = (args: string[], cwd: string, env?: GitEnvironment) => GitOutput;
@@ -3191,6 +2917,5 @@ declare function verifyCodeSurface(surface: CodeSurface, worktreeDir?: string):
 /** Resolve a code candidate for evaluation only after verifying its immutable
  *  identity against the checkout at `worktreeRef`. */
 declare function resolveWorktreePath(surface: CodeSurface, worktreeDir?: string): string;
-//#endregion
-export { summarizeBackendIntegrity as $, assertCampaignDesign as $i, CampaignBreakdown as $n, OptimizationMethodResult as $r, PlanEvalFixtureRunOptions as $t, RetainedDispatchOptions as A, RunOptimizationResult as Ai, CrossSurfacePairEvidence as An, OptimizationMethodPairwise as Ar, sequentialOptimizationMethod as At, RunProfileMatrixSegmentOptions as B, runCampaign as Bi, TraceAnalystArtifact as Bn, OpenAICompatibleOptimizerModel as Br, GepaPopulationImport as Bt, searchModelIdentity as C, isTransientTransportFailure as Ci, CrossSurfaceInteractionAwareSelection as Cn, composeGate as Cr, RequirementCheck as Ct, ScenarioSignal as D, runImprovementLoop as Di, CrossSurfaceInteractionTask as Dn, skillOptOptimizationMethod as Dr, createLlmCorrectnessChecker as Dt, DiscriminationScore as E, RunImprovementLoopResult as Ei, CrossSurfaceInteractionReport as En, SkillOptTrainerConfig as Er, completionVerdict as Et, FinalizedProfileMatrixResult as F, RunEvalOptions as Fi, CrossSurfaceSelectionPolicy as Fn, GepaEngineRun as Fr, RolloutCall as Ft, ProfileSummary as G, CacheIssueReason as Gi, LoopProvenanceArgsFromResult as Gn, ExternalTextOptimizerResult as Gr, neutralizationGate as Gt, finalizeProfileMatrix as H, CampaignRunPlanCell as Hi, BuildLoopProvenanceArgs as Hn, externalTextOptimizationMethod as Hr, importExternalEvaluations as Ht, ProfileMatrixCoverage as I, runEval as Ii, CrossSurfaceSelections as In, GepaOptimizationMethodConfig as Ir, ScoredRollout as It, ScenarioRollup as J, buildCellSchedule as Ji, LoopProvenanceEvidence as Jn, decodeExternalTextCandidate as Jr, EvalFixtureLoadOptions as Jt, RunProfileMatrixOptions as K, CacheRead as Ki, LoopProvenanceBackend as Kn, ExternalOptimizationExample as Kr, EvalFixture as Kt, ProfileMatrixPlan as L, CampaignCellFailureReceipt as Li, CrossSurfaceTaskRow as Ln, GepaOptimizationRecipe as Lr, UngroundedLiteralReport as Lt, createRetainedDispatch as M, OpenAutoPrOptions as Mi, CrossSurfacePairwiseEntry as Mn, compareOptimizationMethods as Mr, FsLabeledScenarioStoreOptions as Mt, CreateProfileMatrixPlanOptions as N, OpenAutoPrResult as Ni, CrossSurfaceRankedSingle as Nn, GepaAdaptiveEngineRun as Nr, RolloutArgumentDiff as Nt, scoreDiscrimination as O, PremeasuredOptimizationBaseline as Oi, CrossSurfaceNaiveStackSelection as On, CompareOptimizationMethodsOptions as Or, verifyCompletion as Ot, FinalizeProfileMatrixOptions as P, openAutoPr as Pi, CrossSurfaceRelativeCost as Pn, GepaEngineOptions as Pr, RolloutArgumentDiffOptions as Pt, assertRealBackend as Q, CampaignCoverage as Qi, emitLoopProvenance as Qn, OptimizationMethodProvenance as Qr, LoadEvalFixtureScenariosOptions as Qt, ProfileMatrixRow as R, CampaignCellRetryPolicy as Ri, measureCampaignCost as Rn, GepaRunnerCommand as Rr, ExternalSearchOptimizer as Rt, campaignCellSearchResult as S, TransientFailureOptions as Si, CrossSurfaceIneligibilityReason as Sn, runCanaries as Sr, ProducedState as St, searchReceiptAccounting as T, RunImprovementLoopOptions as Ti, CrossSurfaceInteractionPath as Tn, SkillOptRunnerCommand as Tr, TaskGold as Tt, runProfileMatrixSegment as U, PlanCampaignRunOptions as Ui, EmitLoopProvenanceArgs as Un, ExternalTextOptimizationMethodConfig as Ur, importGepaPopulation as Ut, createProfileMatrixPlan as V, CampaignRunPlan as Vi, TraceAnalystScenario as Vn, OptimizerModelBudget as Vr, externalSurface as Vt, ProfileDispatchFn as W, planCampaignRun as Wi, EmitLoopProvenanceResult as Wn, ExternalTextOptimizerContext as Wr, NeutralizationGateOptions as Wt, BackendIntegrityError as X, FinalEvidencePolicy as Xi, LoopProvenanceRecord as Xn, OptimizationMethodComposition as Xr, EvalFixtureScenario as Xt, runProfileMatrix as Y, cellCachePath as Yi, LoopProvenanceOptimizationMethod as Yn, OptimizationMethod as Yr, EvalFixtureRunPlan as Yt, BackendIntegrityReport as Z, FinalEvidenceUse as Zi, canonicalDigest as Zn, OptimizationMethodInput as Zr, EvalFixtureValidationMode as Zt, surfaceDispatchRef as _, ReferenceEquivalenceJudgeResult as _i, CrossSurfaceComponentEvidence as _n, CanaryAlert as _r, CompletionRequirement as _t, gitWorktreeAdapter as a, SearchLedgerIntegrityError as aa, GepaCandidatePopulationArtifact as ai, AnalyzeCrossSurfaceInteractionsInput as an, DefaultProductionGateCheck as ar, ScoreboardSummary as at, SingleRunLockOptions as b, LlmJudgeOptions as bi, CrossSurfaceEligibility as bn, CanaryOptions as br, LlmCorrectnessCheckerOpts as bt, AutoevalsScoreLike as c, GepaCandidateSelectionScore as ci, CrossSurfaceAttemptCompleteness as cn, defaultProductionGate as cr, renderScoreboardMarkdown as ct, PhoenixEvaluatorLike as d, ExternalOptimizerObservationArtifact as di, CrossSurfaceCandidate as dn, RedTeamCategory as dr, userStoryScoreboard as dt, campaignCoverage as ea, OptimizationMethodRunOptions as ei, discoverEvalFixtures as en, CampaignComparisonUnits as er, PlaybackContext as et, assertCodeSurfaceIdentity as f, ExternalOptimizerObservationSummary as fi, CrossSurfaceCandidateComparison as fn, RedTeamFinding as fr, ArtifactEventLike as ft, surfaceContentHash as g, ReferenceEquivalenceJudgeOptions as gi, CrossSurfaceComponent as gn, scoreRedTeamOutput as gr, extractProducedState as gt, renderSurfaceDiff as h, ReferenceEquivalenceJudgeInput as hi, CrossSurfaceCandidateSummary as hn, redTeamReport as hr, ToolCallEventLike as ht, WorktreeAdapter as i, SearchLedgerError as ia, costFromLedgerSummary as ii, analyzeCrossSurfaceInteractions as in, heldOutGate as ir, ScoreboardRow as it, RetainedDispatchOutcome as j, runOptimization as ji, CrossSurfacePairIncompatibilityReason as jn, OptimizationMethodScore as jr, neutralizeText as jt, resolveRunDir as k, RunOptimizationOptions as ki, CrossSurfacePairCompatibility as kn, OptimizationMethodComparison as kr, scopedOptimizationMethod as kt, AutoevalsScorerLike as l, readGepaCandidatePopulationArtifact as li, CrossSurfaceBestSingleSelection as ln, DEFAULT_RED_TEAM_CORPUS as lr, scoreUserStory as lt, componentSurfaceIdentityMaterial as m, readExternalOptimizerObservationArtifact as mi, CrossSurfaceCandidateOutcome as mn, redTeamDataset as mr, RuntimeEventLike as mt, GitWorktreeAdapterOptions as n, campaignSplitDigest as na, ComparisonCost as ni, loadEvalFixtureScenarios as nn, campaignMeanComposite as nr, PlaybackStep as nt, resolveWorktreePath as o, GepaCandidatePopulationCandidate as oi, CrossSurfaceAdditionDecision as on, DefaultProductionGateOptions as or, UserStory as ot, codeSurfaceIdentityMaterial as p, ExternalOptimizerSubmittedCandidate as pi, CrossSurfaceCandidateEvidence as pn, RedTeamReport as pr, ProposalEventLike as pt, RunProfileMatrixResult as q, CellScheduleSlot as qi, LoopProvenanceCandidate as qn, ExternalTextEvaluationResponse as qr, EvalFixtureFile as qt, Worktree as r, campaignSplitDigestFromIdentities as ra, OptimizationTokenUsage as ri, planEvalFixtureRun as rn, HeldOutGateOptions as rr, ScoreboardRenderOptions as rt, verifyCodeSurface as s, GepaCandidatePopulationSummary as si, CrossSurfaceAdditionRejectionReason as sn, DefaultProductionRewardHackingOptions as sr, UserStoryVerdict as st, CodeSurfaceVerification as t, campaignScenarioIdentity as ta, OptimizationPackageSource as ti, loadEvalFixture as tn, campaignBreakdown as tr, PlaybackDriver as tt, PhoenixEvaluationResultLike as u, ExternalOptimizerExecutionSummary as ui, CrossSurfaceBootstrapPolicy as un, RedTeamCase as ur, scoreboardSummary as ut, surfaceHash as v, ReferenceEquivalenceScenario as vi, CrossSurfaceCompositionStep as vn, CanaryEvaluation as vr, CompletionVerdict as vt, searchProposalExecution as w, transientDispatchFailure as wi, CrossSurfaceInteractionEffect as wn, SkillOptOptimizationMethodConfig as wr, SatisfiedBy as wt, acquireSingleRunLock as x, llmJudge as xi, CrossSurfaceEvidenceBreakdown as xn, CanaryReport as xr, ProducedProposal as xt, SingleRunLock as y, LlmJudgeDimension as yi, CrossSurfaceDistribution as yn, CanaryKind as yr, CorrectnessChecker as yt, ProfileMatrixSegmentResult as z, RunCampaignOptions as zi, BuildTraceAnalystSurfaceDispatchOptions as zn, gepaOptimizationMethod as zr, GepaEvaluationImport as zt };
-//# sourceMappingURL=index-CdEVr8-O.d.ts.map
+export { summarizeBackendIntegrity as $, RedTeamCategory as $n, ExternalOptimizerObservationArtifact as $r, CrossSurfaceCandidate as $t, RetainedDispatchOptions as A, CampaignRunPlanCell as Ai, BuildLoopProvenanceArgs as An, externalTextOptimizationMethod as Ar, importExternalEvaluations as At, RunProfileMatrixSegmentOptions as B, CampaignCoverage as Bi, emitLoopProvenance as Bn, OptimizationMethodProvenance as Br, LoadEvalFixtureScenariosOptions as Bt, searchModelIdentity as C, RunEvalOptions as Ci, CrossSurfaceSelectionPolicy as Cn, GepaEngineRun as Cr, RolloutCall as Ct, ScenarioSignal as D, RunCampaignOptions as Di, BuildTraceAnalystSurfaceDispatchOptions as Dn, gepaOptimizationMethod as Dr, GepaEvaluationImport as Dt, DiscriminationScore as E, CampaignCellRetryPolicy as Ei, measureCampaignCost as En, GepaRunnerCommand as Er, ExternalSearchOptimizer as Et, FinalizedProfileMatrixResult as F, CellScheduleSlot as Fi, LoopProvenanceCandidate as Fn, ExternalTextEvaluationResponse as Fr, EvalFixtureFile as Ft, ProfileSummary as G, campaignSplitDigestFromIdentities as Gi, HeldOutGateOptions as Gn, OptimizationTokenUsage as Gr, planEvalFixtureRun as Gt, finalizeProfileMatrix as H, campaignCoverage as Hi, CampaignComparisonUnits as Hn, OptimizationMethodRunOptions as Hr, discoverEvalFixtures as Ht, ProfileMatrixCoverage as I, buildCellSchedule as Ii, LoopProvenanceEvidence as In, decodeExternalTextCandidate as Ir, EvalFixtureLoadOptions as It, ScenarioRollup as J, DefaultProductionGateOptions as Jn, GepaCandidatePopulationCandidate as Jr, CrossSurfaceAdditionDecision as Jt, RunProfileMatrixOptions as K, SearchLedgerError as Ki, heldOutGate as Kn, costFromLedgerSummary as Kr, analyzeCrossSurfaceInteractions as Kt, ProfileMatrixPlan as L, cellCachePath as Li, LoopProvenanceOptimizationMethod as Ln, OptimizationMethod as Lr, EvalFixtureRunPlan as Lt, createRetainedDispatch as M, planCampaignRun as Mi, EmitLoopProvenanceResult as Mn, ExternalTextOptimizerContext as Mr, NeutralizationGateOptions as Mt, CreateProfileMatrixPlanOptions as N, CacheIssueReason as Ni, LoopProvenanceArgsFromResult as Nn, ExternalTextOptimizerResult as Nr, neutralizationGate as Nt, scoreDiscrimination as O, runCampaign as Oi, TraceAnalystArtifact as On, OpenAICompatibleOptimizerModel as Or, GepaPopulationImport as Ot, FinalizeProfileMatrixOptions as P, CacheRead as Pi, LoopProvenanceBackend as Pn, ExternalOptimizationExample as Pr, EvalFixture as Pt, assertRealBackend as Q, RedTeamCase as Qn, ExternalOptimizerExecutionSummary as Qr, CrossSurfaceBootstrapPolicy as Qt, ProfileMatrixRow as R, FinalEvidencePolicy as Ri, LoopProvenanceRecord as Rn, OptimizationMethodComposition as Rr, EvalFixtureScenario as Rt, campaignCellSearchResult as S, openAutoPr as Si, CrossSurfaceRelativeCost as Sn, GepaEngineOptions as Sr, RolloutArgumentDiffOptions as St, searchReceiptAccounting as T, CampaignCellFailureReceipt as Ti, CrossSurfaceTaskRow as Tn, GepaOptimizationRecipe as Tr, UngroundedLiteralReport as Tt, runProfileMatrixSegment as U, campaignScenarioIdentity as Ui, campaignBreakdown as Un, OptimizationPackageSource as Ur, loadEvalFixture as Ut, createProfileMatrixPlan as V, assertCampaignDesign as Vi, CampaignBreakdown as Vn, OptimizationMethodResult as Vr, PlanEvalFixtureRunOptions as Vt, ProfileDispatchFn as W, campaignSplitDigest as Wi, campaignMeanComposite as Wn, ComparisonCost as Wr, loadEvalFixtureScenarios as Wt, BackendIntegrityError as X, defaultProductionGate as Xn, GepaCandidateSelectionScore as Xr, CrossSurfaceAttemptCompleteness as Xt, runProfileMatrix as Y, DefaultProductionRewardHackingOptions as Yn, GepaCandidatePopulationSummary as Yr, CrossSurfaceAdditionRejectionReason as Yt, BackendIntegrityReport as Z, DEFAULT_RED_TEAM_CORPUS as Zn, readGepaCandidatePopulationArtifact as Zr, CrossSurfaceBestSingleSelection as Zt, surfaceDispatchRef as _, RunOptimizationOptions as _i, CrossSurfacePairCompatibility as _n, OptimizationMethodComparison as _r, scopedOptimizationMethod as _t, gitWorktreeAdapter as a, ReferenceEquivalenceJudgeResult as ai, CrossSurfaceComponentEvidence as an, CanaryAlert as ar, ScoreboardSummary as at, SingleRunLockOptions as b, OpenAutoPrOptions as bi, CrossSurfacePairwiseEntry as bn, compareOptimizationMethods as br, FsLabeledScenarioStoreOptions as bt, AutoevalsScoreLike as c, LlmJudgeOptions as ci, CrossSurfaceEligibility as cn, CanaryOptions as cr, renderScoreboardMarkdown as ct, PhoenixEvaluatorLike as d, isTransientTransportFailure as di, CrossSurfaceInteractionAwareSelection as dn, composeGate as dr, userStoryScoreboard as dt, ExternalOptimizerObservationSummary as ei, CrossSurfaceCandidateComparison as en, RedTeamFinding as er, PlaybackContext as et, assertCodeSurfaceIdentity as f, transientDispatchFailure as fi, CrossSurfaceInteractionEffect as fn, SkillOptOptimizationMethodConfig as fr, ArtifactEventLike as ft, surfaceContentHash as g, PremeasuredOptimizationBaseline as gi, CrossSurfaceNaiveStackSelection as gn, CompareOptimizationMethodsOptions as gr, extractProducedState as gt, renderSurfaceDiff as h, runImprovementLoop as hi, CrossSurfaceInteractionTask as hn, skillOptOptimizationMethod as hr, ToolCallEventLike as ht, WorktreeAdapter as i, ReferenceEquivalenceJudgeOptions as ii, CrossSurfaceComponent as in, scoreRedTeamOutput as ir, ScoreboardRow as it, RetainedDispatchOutcome as j, PlanCampaignRunOptions as ji, EmitLoopProvenanceArgs as jn, ExternalTextOptimizationMethodConfig as jr, importGepaPopulation as jt, resolveRunDir as k, CampaignRunPlan as ki, TraceAnalystScenario as kn, OptimizerModelBudget as kr, externalSurface as kt, AutoevalsScorerLike as l, llmJudge as li, CrossSurfaceEvidenceBreakdown as ln, CanaryReport as lr, scoreUserStory as lt, componentSurfaceIdentityMaterial as m, RunImprovementLoopResult as mi, CrossSurfaceInteractionReport as mn, SkillOptTrainerConfig as mr, RuntimeEventLike as mt, GitWorktreeAdapterOptions as n, readExternalOptimizerObservationArtifact as ni, CrossSurfaceCandidateOutcome as nn, redTeamDataset as nr, PlaybackStep as nt, resolveWorktreePath as o, ReferenceEquivalenceScenario as oi, CrossSurfaceCompositionStep as on, CanaryEvaluation as or, UserStory as ot, codeSurfaceIdentityMaterial as p, RunImprovementLoopOptions as pi, CrossSurfaceInteractionPath as pn, SkillOptRunnerCommand as pr, ProposalEventLike as pt, RunProfileMatrixResult as q, SearchLedgerIntegrityError as qi, DefaultProductionGateCheck as qn, GepaCandidatePopulationArtifact as qr, AnalyzeCrossSurfaceInteractionsInput as qt, Worktree as r, ReferenceEquivalenceJudgeInput as ri, CrossSurfaceCandidateSummary as rn, redTeamReport as rr, ScoreboardRenderOptions as rt, verifyCodeSurface as s, LlmJudgeDimension as si, CrossSurfaceDistribution as sn, CanaryKind as sr, UserStoryVerdict as st, CodeSurfaceVerification as t, ExternalOptimizerSubmittedCandidate as ti, CrossSurfaceCandidateEvidence as tn, RedTeamReport as tr, PlaybackDriver as tt, PhoenixEvaluationResultLike as u, TransientFailureOptions as ui, CrossSurfaceIneligibilityReason as un, runCanaries as ur, scoreboardSummary as ut, surfaceHash as v, RunOptimizationResult as vi, CrossSurfacePairEvidence as vn, OptimizationMethodPairwise as vr, sequentialOptimizationMethod as vt, searchProposalExecution as w, runEval as wi, CrossSurfaceSelections as wn, GepaOptimizationMethodConfig as wr, ScoredRollout as wt, acquireSingleRunLock as x, OpenAutoPrResult as xi, CrossSurfaceRankedSingle as xn, GepaAdaptiveEngineRun as xr, RolloutArgumentDiff as xt, SingleRunLock as y, runOptimization as yi, CrossSurfacePairIncompatibilityReason as yn, OptimizationMethodScore as yr, neutralizeText as yt, ProfileMatrixSegmentResult as z, FinalEvidenceUse as zi, canonicalDigest as zn, OptimizationMethodInput as zr, EvalFixtureValidationMode as zt };
+//# sourceMappingURL=index-Bdcpx2s8.d.ts.map