@tangle-network/agent-eval 0.200.1 → 0.201.0

package/dist/index.d.ts

@@ -44,7 +44,7 @@ import { C as inferDomainKeywords, H as captureFetchToRawSink, I as OtlpExport,
 import { a as RunIntegrityReport, o as assertRunCaptured, t as RunIntegrityError } from "./integrity-CaeNeusS.js";
 import { c as judgeSpans, f as runsForScenario, i as argHash } from "./query-DBUQhehH.js";
 import { a as SteeringChange, c as CampaignRunContext, d as CampaignVariant, f as EvalCampaignOptions, h as runEvalCampaign, i as Researcher, l as CampaignRunOutcome, m as FailedRun, n as ExperimentResult, o as CampaignFactoryParams, p as EvalCampaignResult, r as FailureMode, s as CampaignIntegrityPolicy, t as ExperimentPlan, u as CampaignScenario } from "./researcher-DojEMb7E.js";
-import "./index-D4pC7Tvd2.js";
+import "./index-D4pC7Tvd.js";
 import { _ as iqr, a as ToolStats, c as computeToolUseMetrics, d as judgeAgreementView, i as toolWasteView, m as budgetBreachView, o as ToolUseMetrics, s as ToolUseOptions } from "./tool-waste-CP5DO383.js";
 import { SpanKind } from "@tangle-network/agent-trace-contract";
 //#region src/statistics/effect-sizes.d.ts
@@ -64,7 +64,7 @@ import { SpanKind } from "@tangle-network/agent-trace-contract";
  * means with zero spread is a genuine 0. Same contract as
  * {@link pairedCohensDz}.
  */
-declare function cohensD(a: number[], b: number[]): number | null;
+export declare function cohensD(a: number[], b: number[]): number | null;
 /**
  * Cohen's dz for paired observations: mean(after - before) divided by the
  * sample standard deviation of those within-pair deltas.
@@ -73,7 +73,7 @@ declare function cohensD(a: number[], b: number[]): number | null;
  * has zero observed variance. In that case the standardized effect is
  * undefined, not an arbitrarily large finite number.
  */
-declare function pairedCohensDz(before: number[], after: number[]): number | null;
+export declare function pairedCohensDz(before: number[], after: number[]): number | null;
 type CliffsMagnitude = 'negligible' | 'small' | 'medium' | 'large';
 /**
  * Cliff's delta — a non-parametric effect size for two independent samples.
@@ -85,13 +85,13 @@ type CliffsMagnitude = 'negligible' | 'small' | 'medium' | 'large';
  * `pairedBootstrap` / `wilcoxonSignedRank` for the non-parametric reporting
  * path. Returns 0 when either sample is empty.
  */
-declare function cliffsDelta(before: number[], after: number[]): number;
+export declare function cliffsDelta(before: number[], after: number[]): number;
 /**
  * Map a Cliff's delta to a qualitative magnitude using the standard
  * Romano et al. thresholds (|δ|): <0.147 negligible, <0.33 small,
  * <0.474 medium, else large.
  */
-declare function interpretCliffs(delta: number): CliffsMagnitude;
+export declare function interpretCliffs(delta: number): CliffsMagnitude;
 //#endregion
 //#region src/profile/baselines.d.ts
 /**
@@ -219,13 +219,13 @@ declare function applyDomainPatch(p: PromptProfile, sectionId: string, newBody:
 declare function sectionHash(section: AgentProfileSection): string;
 //#endregion
 //#region src/budget-guard.d.ts
-declare class BudgetBreachError extends AgentEvalError {
+export declare class BudgetBreachError extends AgentEvalError {
   dimension: keyof BudgetSpec;
   limit: number;
   attempted: number;
   constructor(dimension: keyof BudgetSpec, limit: number, attempted: number);
 }
-declare class BudgetGuard {
+export declare class BudgetGuard {
   private consumed;
   private emitter;
   private budget;
@@ -266,8 +266,8 @@ interface RunScoreWeights {
   costUsd: number;
   wallSeconds: number;
 }
-declare function aggregateRunScore(score: RunScore, weights?: Partial<RunScoreWeights>): number;
-declare function clamp01(value: number): number;
+export declare function aggregateRunScore(score: RunScore, weights?: Partial<RunScoreWeights>): number;
+export declare function clamp01(value: number): number;
 //#endregion
 //#region src/scorecard.d.ts
 /** One commit's measurement of one (scenario, profile) cell. */
@@ -319,15 +319,15 @@ interface RecordRunsOptions {
  * `profile` in `opts` (the harness ran them under it); the cell key is
  * `(scenarioId, agentProfileHash(profile))`.
  */
-declare function recordRuns(runs: RunRecord[], opts: RecordRunsOptions): ScorecardLogLine[];
+export declare function recordRuns(runs: RunRecord[], opts: RecordRunsOptions): ScorecardLogLine[];
 /** Record runs and append them to the log in one call. Returns the lines. */
-declare function recordRunsToScorecard(logPath: string, runs: RunRecord[], opts: RecordRunsOptions): ScorecardLogLine[];
+export declare function recordRunsToScorecard(logPath: string, runs: RunRecord[], opts: RecordRunsOptions): ScorecardLogLine[];
 /**
  * Fold the JSONL log into a queryable `Scorecard`. A missing file yields an
  * empty scorecard; a malformed line is skipped — a corrupt append never
  * breaks the read. Each cell's timeline is sorted chronologically.
  */
-declare function loadScorecard(logPath: string): Scorecard;
+export declare function loadScorecard(logPath: string): Scorecard;
 type CellVerdict = 'improved' | 'regressed' | 'flat' | 'new';
 interface ScorecardCellDiff {
   scenarioId: string;
@@ -375,13 +375,13 @@ interface DiffScorecardOptions {
  * only when it clears both the effect-size and significance gates; otherwise
  * `flat`. Cells with no prior entry are `new`.
  */
-declare function diffScorecard(scorecard: Scorecard, opts?: DiffScorecardOptions): ScorecardDiff;
+export declare function diffScorecard(scorecard: Scorecard, opts?: DiffScorecardOptions): ScorecardDiff;
 /**
  * Render a scorecard diff as a human-readable report — the block a feature
  * PR prints. Regressions are listed first; flat cells are summarised, not
  * enumerated.
  */
-declare function formatScorecardDiff(diff: ScorecardDiff): string;
+export declare function formatScorecardDiff(diff: ScorecardDiff): string;
 //#endregion
 //#region src/trace-contracts/types.d.ts
 /**
@@ -667,7 +667,7 @@ declare class TraceContractBuilder {
   private where;
   private add;
 }
-declare function traceContract(name: string): TraceContractBuilder;
+export declare function traceContract(name: string): TraceContractBuilder;
 //#endregion
 //#region src/trace-contracts/evaluate.d.ts
 /**
@@ -677,16 +677,16 @@ declare function traceContract(name: string): TraceContractBuilder;
  * `ValidationError` for a malformed contract; a rule that cannot be evaluated
  * is reported as `error` in the verdict instead.
  */
-declare function evaluateTraceContract(contract: TraceContract, spans: readonly ContractSpan[], options?: EvaluateTraceContractOptions): ContractVerdict;
+export declare function evaluateTraceContract(contract: TraceContract, spans: readonly ContractSpan[], options?: EvaluateTraceContractOptions): ContractVerdict;
 /**
  * Evaluate many contracts over one span sequence. Throws on an empty
  * contract list — `allValid: true` over zero contracts is a silent pass.
  */
-declare function checkTraceContracts(spans: readonly ContractSpan[], contracts: readonly TraceContract[], options?: EvaluateTraceContractOptions): ContractCheckResult;
+export declare function checkTraceContracts(spans: readonly ContractSpan[], contracts: readonly TraceContract[], options?: EvaluateTraceContractOptions): ContractCheckResult;
 //#endregion
 //#region src/trace-contracts/explain.d.ts
 /** Plain-language statement of everything a contract checks, one rule per line. */
-declare function explainTraceContract(contract: TraceContract): string;
+export declare function explainTraceContract(contract: TraceContract): string;
 //#endregion
 //#region src/trace-contracts/spec.d.ts
 interface ToolArgumentSpec {
@@ -767,7 +767,7 @@ interface TraceContractSpec extends PathSpec {
  * the path of the first problem: an unknown key, a wrong type, an unknown
  * status, or a contract that checks nothing.
  */
-declare function compileTraceContractSpec(input: unknown): TraceContract;
+export declare function compileTraceContractSpec(input: unknown): TraceContract;
 interface ContractLintFinding {
   /** `error`: the contract can never pass, or contradicts itself.
    *  `warning`: the contract works but likely does not say what was meant. */
@@ -780,7 +780,7 @@ interface ContractLintFinding {
  * Find contradictions and likely mistakes in a declarative contract. Parses
  * strictly first, so a malformed spec throws instead of linting.
  */
-declare function lintTraceContractSpec(input: unknown): ContractLintFinding[];
+export declare function lintTraceContractSpec(input: unknown): ContractLintFinding[];
 //#endregion
 //#region src/active-learning.d.ts
 type SynthesisReason = 'high-variance' | 'undersampled' | 'failure-cluster' | 'difficulty-gap';
@@ -803,7 +803,7 @@ interface ActiveLearningOptions {
   /** Max synthesis targets returned. */
   topK?: number;
 }
-declare function proposeSynthesisTargets(dataset: Dataset, traceStore: TraceStore, options?: ActiveLearningOptions): Promise<SynthesisTarget[]>;
+export declare function proposeSynthesisTargets(dataset: Dataset, traceStore: TraceStore, options?: ActiveLearningOptions): Promise<SynthesisTarget[]>;
 //#endregion
 //#region src/bounded-process.d.ts
 /**
@@ -833,7 +833,7 @@ declare function proposeSynthesisTargets(dataset: Dataset, traceStore: TraceStor
  * bytes are discarded and `outputTruncated` is set.
  */
 /** Wall-clock bound applied when the caller names none. */
-declare const DEFAULT_BOUNDED_PROCESS_TIMEOUT_MS: number;
+export declare const DEFAULT_BOUNDED_PROCESS_TIMEOUT_MS: number;
 interface BoundedProcessInput {
   /**
    * The command line, interpreted by a shell unless `shell` is `false`. When
@@ -1001,7 +1001,7 @@ interface BoundedProcessResult {
    */
   runnerError?: string;
 }
-declare function runBoundedProcess(input: BoundedProcessInput): Promise<BoundedProcessResult>;
+export declare function runBoundedProcess(input: BoundedProcessInput): Promise<BoundedProcessResult>;
 //#endregion
 //#region src/client.d.ts
 /**
@@ -1013,7 +1013,7 @@ declare function runBoundedProcess(input: BoundedProcessInput): Promise<BoundedP
  * substrate. Moves to the product repo in the next major
  * (tangle-network/agent-runtime#694); vendor it if you construct it directly.
  */
-declare class ProductClient {
+export declare class ProductClient {
   private baseUrl;
   private routes;
   private cookies;
@@ -1184,8 +1184,8 @@ interface CommandRunner {
  * runner used to be unbounded: a consumer whose command runs longer names its
  * own `capMs`, and a run the bound stops says so with `timedOut: true`.
  */
-declare const LOCAL_COMMAND_RUNNER_DEFAULT_CAP_MS: number;
-declare const localCommandRunner: CommandRunner;
+export declare const LOCAL_COMMAND_RUNNER_DEFAULT_CAP_MS: number;
+export declare const localCommandRunner: CommandRunner;
 //#endregion
 //#region src/discover-personas.d.ts
 /**
@@ -1221,7 +1221,7 @@ interface DiscoveredPersona {
   /** Filename without extension — the conventional persona id. */
   id: string;
 }
-declare function discoverPersonas(dir: string, opts?: DiscoverPersonasOptions): Promise<DiscoveredPersona[]>;
+export declare function discoverPersonas(dir: string, opts?: DiscoverPersonasOptions): Promise<DiscoveredPersona[]>;
 //#endregion
 //#region src/improvement-verdict.d.ts
 /** Verdict for one candidate relative to its parent. ITERATE is the neutral
@@ -1287,14 +1287,14 @@ interface ImprovementVerdictResult {
  * flag is the trust gate the verdict depends on: a sample whose spread exceeds
  * the configured bounds can't distinguish a real delta from run-to-run noise.
  */
-declare function computeExperimentStats(reps: ExperimentRep[], thresholds?: ImprovementThresholds): ExperimentStats;
+export declare function computeExperimentStats(reps: ExperimentRep[], thresholds?: ImprovementThresholds): ExperimentStats;
 /**
  * Verdict for a candidate against its parent. Pure — operates on already-computed
  * stats. KEEP/REGRESSION require both sides to have `>= minRepsForVerdict` reps
  * AND the candidate to be `stable`; otherwise the result is NOISE (unstable) or
  * ITERATE (not enough reps / no parent).
  */
-declare function improvementVerdict(candidate: ExperimentStats, parent: ExperimentStats | null, thresholds?: ImprovementThresholds): ImprovementVerdictResult;
+export declare function improvementVerdict(candidate: ExperimentStats, parent: ExperimentStats | null, thresholds?: ImprovementThresholds): ImprovementVerdictResult;
 //#endregion
 //#region src/propose-review.d.ts
 interface Verification {
@@ -1396,9 +1396,9 @@ interface ProposeReviewReport<State, Summary = unknown> {
   wallMs: number;
   score: number;
 }
-declare function inMemoryReviewStore(initial?: ReviewMemoryEntry[]): ReviewMemoryStore;
-declare function jsonlReviewStore(path: string): ReviewMemoryStore;
-declare function runProposeReview<State, Summary = unknown>(config: ProposeReviewConfig<State, Summary>): Promise<ProposeReviewReport<State, Summary>>;
+export declare function inMemoryReviewStore(initial?: ReviewMemoryEntry[]): ReviewMemoryStore;
+export declare function jsonlReviewStore(path: string): ReviewMemoryStore;
+export declare function runProposeReview<State, Summary = unknown>(config: ProposeReviewConfig<State, Summary>): Promise<ProposeReviewReport<State, Summary>>;
 type LlmJsonCall = (req: {
   system: string;
   user: string;
@@ -1410,7 +1410,7 @@ interface LlmReviewerConfig<State, Summary = unknown> {
   /** Appended to the default system prompt. */
   systemPromptAddendum?: string;
 }
-declare function createLlmReviewer<State, Summary = unknown>(cfg: LlmReviewerConfig<State, Summary>): ReviewFn<State, Summary>;
+export declare function createLlmReviewer<State, Summary = unknown>(cfg: LlmReviewerConfig<State, Summary>): ReviewFn<State, Summary>;
 //#endregion
 //#region src/propose-review-control.d.ts
 interface ProposeReviewControlState<State, Summary = unknown> {
@@ -1455,7 +1455,7 @@ interface ProposeReviewControlConfig<State, Summary = unknown> {
   failureClassFromVerification?: (verification: Verification) => FailureClass | undefined;
   actionFailure?: ControlRuntimeConfig<ProposeReviewControlState<State, Summary>, ProposeReviewControlAction, ProposeReviewControlResult<State, Summary>>['actionFailure'];
 }
-declare function runProposeReviewAsControlLoop<State, Summary = unknown>(config: ProposeReviewControlConfig<State, Summary>): Promise<ControlRunResult<ProposeReviewControlState<State, Summary>, ProposeReviewControlAction, ProposeReviewControlResult<State, Summary>>>;
+export declare function runProposeReviewAsControlLoop<State, Summary = unknown>(config: ProposeReviewControlConfig<State, Summary>): Promise<ControlRunResult<ProposeReviewControlState<State, Summary>, ProposeReviewControlAction, ProposeReviewControlResult<State, Summary>>>;
 //#endregion
 //#region src/reflective-mutation.d.ts
 interface TrialTrace {
@@ -1503,7 +1503,7 @@ interface ReflectionContext {
  * the user message. The system message should be small and stable (e.g.
  * "Output ONLY a JSON object matching the schema below.").
  */
-declare function buildReflectionPrompt(ctx: ReflectionContext): string;
+export declare function buildReflectionPrompt(ctx: ReflectionContext): string;
 interface ReflectionProposal {
   label: string;
   rationale: string;
@@ -1513,13 +1513,13 @@ interface ReflectionProposal {
  * Parse the model's JSON response back into proposals. Tolerates markdown
  * fences and surrounding prose. Returns at most `maxProposals`.
  */
-declare function parseReflectionResponse(raw: string, maxProposals?: number): ReflectionProposal[];
+export declare function parseReflectionResponse(raw: string, maxProposals?: number): ReflectionProposal[];
 //#endregion
 //#region src/analyst/policy-edit.d.ts
 type PolicyEditSchemaVersion = 'policy-edit/v1';
-declare const POLICY_EDIT_AXES: readonly ['carrier', 'representation', 'budget', 'sampling', 'output_contract', 'tool_contract', 'routing', 'memory', 'agent_profile', 'deployment_target'];
+export declare const POLICY_EDIT_AXES: readonly ['carrier', 'representation', 'budget', 'sampling', 'output_contract', 'tool_contract', 'routing', 'memory', 'agent_profile', 'deployment_target'];
 type PolicyEditAxis = (typeof POLICY_EDIT_AXES)[number];
-declare const POLICY_EDIT_TARGET_SURFACES: readonly ['prompt', 'tool-contract', 'runtime-config', 'memory', 'agent-profile', 'code', 'deployment'];
+export declare const POLICY_EDIT_TARGET_SURFACES: readonly ['prompt', 'tool-contract', 'runtime-config', 'memory', 'agent-profile', 'code', 'deployment'];
 type PolicyEditTargetSurface = (typeof POLICY_EDIT_TARGET_SURFACES)[number];
 type PolicyEditRisk = 'low' | 'medium' | 'high' | 'unknown';
 type PolicyEditGainDirection = 'increase' | 'decrease';
@@ -1578,7 +1578,7 @@ interface PolicyEdit {
   validationPlan?: string;
   metadata?: Record<string, unknown>;
 }
-declare const POLICY_EDIT_CANDIDATE_RECORD_SCHEMA: 'tangle.policy-edit-candidate.v1';
+export declare const POLICY_EDIT_CANDIDATE_RECORD_SCHEMA: 'tangle.policy-edit-candidate.v1';
 /** JSON-safe attribution carried with a measured candidate and its scores. */
 interface PolicyEditCandidateRecord {
   schema: typeof POLICY_EDIT_CANDIDATE_RECORD_SCHEMA;
@@ -1588,7 +1588,7 @@ type PolicyEditInit = Omit<PolicyEdit, 'schemaVersion' | 'editId'> & {
   schemaVersion?: PolicyEditSchemaVersion;
   editId?: string;
 };
-declare class PolicyEditValidationError extends ValidationError {
+export declare class PolicyEditValidationError extends ValidationError {
   readonly path: string;
   constructor(message: string, path?: string);
 }
@@ -1610,17 +1610,17 @@ interface PolicyEditAdmission {
   score: number;
   reasons: string[];
 }
-declare function makePolicyEdit(init: PolicyEditInit): PolicyEdit;
-declare function computePolicyEditId(edit: Omit<PolicyEdit, 'editId'> | PolicyEdit): string;
-declare function validatePolicyEdit(input: unknown): PolicyEdit;
-declare function makePolicyEditCandidateRecord(edit: PolicyEdit): PolicyEditCandidateRecord;
-declare function validatePolicyEditCandidateRecord(input: unknown): PolicyEditCandidateRecord;
-declare function isPolicyEdit(input: unknown): input is PolicyEdit;
-declare function policyEditsFromFindings(findings: ReadonlyArray<AnalystFinding>, opts?: FindingToPolicyEditOptions): PolicyEdit[];
-declare function policyEditFromFinding(finding: AnalystFinding, opts?: FindingToPolicyEditOptions): PolicyEdit | null;
-declare function scorePolicyEditReadiness(edit: PolicyEdit, opts?: PolicyEditAdmissionOptions): number;
-declare function admitPolicyEdit(edit: PolicyEdit, opts?: PolicyEditAdmissionOptions): PolicyEditAdmission;
-declare function applyPolicyEditToSurface(surface: unknown, edit: PolicyEdit): unknown;
+export declare function makePolicyEdit(init: PolicyEditInit): PolicyEdit;
+export declare function computePolicyEditId(edit: Omit<PolicyEdit, 'editId'> | PolicyEdit): string;
+export declare function validatePolicyEdit(input: unknown): PolicyEdit;
+export declare function makePolicyEditCandidateRecord(edit: PolicyEdit): PolicyEditCandidateRecord;
+export declare function validatePolicyEditCandidateRecord(input: unknown): PolicyEditCandidateRecord;
+export declare function isPolicyEdit(input: unknown): input is PolicyEdit;
+export declare function policyEditsFromFindings(findings: ReadonlyArray<AnalystFinding>, opts?: FindingToPolicyEditOptions): PolicyEdit[];
+export declare function policyEditFromFinding(finding: AnalystFinding, opts?: FindingToPolicyEditOptions): PolicyEdit | null;
+export declare function scorePolicyEditReadiness(edit: PolicyEdit, opts?: PolicyEditAdmissionOptions): number;
+export declare function admitPolicyEdit(edit: PolicyEdit, opts?: PolicyEditAdmissionOptions): PolicyEditAdmission;
+export declare function applyPolicyEditToSurface(surface: unknown, edit: PolicyEdit): unknown;
 //#endregion
 //#region src/analyst/steer-firewall.d.ts
 /**
@@ -1642,7 +1642,7 @@ declare function applyPolicyEditToSurface(surface: unknown, edit: PolicyEdit): u
  * lift site, and why ProposeContext.judgeScores?: never is the complementary
  * compile-time tripwire on the obvious direct channel.
  */
-declare function assertNoJudgeVerdict(findings: ReadonlyArray<AnalystFinding>, context?: string): ReadonlyArray<AnalystFinding>;
+export declare function assertNoJudgeVerdict(findings: ReadonlyArray<AnalystFinding>, context?: string): ReadonlyArray<AnalystFinding>;
 //#endregion
 //#region src/capability-headroom.d.ts
 /**
@@ -1724,7 +1724,7 @@ interface CapabilityHeadroomResult {
  * Throws on empty input, an out-of-range `keepThreshold`, or an
  * unrecognized outcome value (rows are often projected from untyped records).
  */
-declare function capabilityHeadroom(rows: readonly HeadroomInput[], opts?: CapabilityHeadroomOptions): CapabilityHeadroomResult;
+export declare function capabilityHeadroom(rows: readonly HeadroomInput[], opts?: CapabilityHeadroomOptions): CapabilityHeadroomResult;
 interface AssertCapabilityHeadroomOptions {
   /** Minimum tasks classified 'gap' for the comparison to proceed. Default 1. */
   minTasksWithGap?: number;
@@ -1735,7 +1735,7 @@ interface AssertCapabilityHeadroomOptions {
  * comparison run on it would measure noise. Call between the baseline
  * calibration run and the treatment run.
  */
-declare function assertCapabilityHeadroom(result: CapabilityHeadroomResult, opts?: AssertCapabilityHeadroomOptions): void;
+export declare function assertCapabilityHeadroom(result: CapabilityHeadroomResult, opts?: AssertCapabilityHeadroomOptions): void;
 //#endregion
 //#region src/cost-tracker.d.ts
 /**
@@ -1773,7 +1773,7 @@ interface ScenarioCost {
   /** Pass flag — set by consumer via markOutcome; used for cost-per-completed-task. */
   completed?: boolean;
 }
-declare class CostTracker {
+export declare class CostTracker {
   private byScenario;
   record(entry: Omit<CostEntry, 'timestamp'> & {
     timestamp?: number;
@@ -1858,7 +1858,7 @@ interface RepeatedActionOptions {
   readonly failureClass?: FailureClass;
 }
 /** Same action fingerprint N consecutive steps = a stuck loop (the #1 long-horizon failure mode). */
-declare function repeatedActionDetector(opts?: RepeatedActionOptions): StreamingDetector;
+export declare function repeatedActionDetector(opts?: RepeatedActionOptions): StreamingDetector;
 interface ErrorStreakOptions {
   /** Signal once this many CONSECUTIVE steps errored (default 3). `<= 0` disables. */
   readonly maxErrors?: number;
@@ -1866,10 +1866,10 @@ interface ErrorStreakOptions {
   readonly failureClass?: FailureClass;
 }
 /** N consecutive tool errors = the worker is hammering a broken approach. */
-declare function errorStreakDetector(opts?: ErrorStreakOptions): StreamingDetector;
+export declare function errorStreakDetector(opts?: ErrorStreakOptions): StreamingDetector;
 /** Fold one event through many detectors at once; returns every signal that fired this step. The
  *  natural shape for an online pipe watching with a whole panel of detectors. */
-declare function observeAll(detectors: ReadonlyArray<StreamingDetector>, event: DetectorEvent): DetectorSignal[];
+export declare function observeAll(detectors: ReadonlyArray<StreamingDetector>, event: DetectorEvent): DetectorSignal[];
 //#endregion
 //#region src/error-count-extractor.d.ts
 /**
@@ -1895,7 +1895,7 @@ interface ErrorCountPattern {
   /** Optional post-processing to extract a count from a single captured match. */
   transform?: (match: RegExpMatchArray) => number;
 }
-declare const ERROR_COUNT_PATTERNS: ErrorCountPattern[];
+export declare const ERROR_COUNT_PATTERNS: ErrorCountPattern[];
 interface ExtractOptions {
   /** Restrict to named patterns — default: all patterns. */
   only?: string[];
@@ -1917,7 +1917,7 @@ interface ExtractResult {
  * callsite that greps for "typescript errors" on cargo output should
  * NOT treat that as "zero TS errors" because the toolchain is wrong.
  */
-declare function extractErrorCount(text: string, opts?: ExtractOptions): ExtractResult;
+export declare function extractErrorCount(text: string, opts?: ExtractOptions): ExtractResult;
 //#endregion
 //#region src/knowledge/types.d.ts
 type KnowledgeRequirementCategory = 'user_specific' | 'company_specific' | 'domain_specific' | 'codebase_specific' | 'market_specific' | 'regulatory' | 'tool_api' | 'credential_or_secret' | 'runtime_environment' | 'preference' | 'historical_context';
@@ -2004,18 +2004,18 @@ interface ScoreKnowledgeReadinessOptions {
   metadata?: Record<string, unknown>;
   now?: Date;
 }
-declare function scoreKnowledgeReadiness(options: ScoreKnowledgeReadinessOptions): KnowledgeReadinessReport;
-declare function blockingKnowledgeEval(report: KnowledgeReadinessReport, options?: {
+export declare function scoreKnowledgeReadiness(options: ScoreKnowledgeReadinessOptions): KnowledgeReadinessReport;
+export declare function blockingKnowledgeEval(report: KnowledgeReadinessReport, options?: {
   id?: string;
   minimumScore?: number;
   emitter?: TraceEmitter;
 }): ControlEvalResult;
-declare function knowledgeReadinessTracePayload(report: KnowledgeReadinessReport, options?: {
+export declare function knowledgeReadinessTracePayload(report: KnowledgeReadinessReport, options?: {
   passed?: boolean;
   minimumScore?: number;
 }): Record<string, unknown>;
-declare function userQuestionsForKnowledgeGaps(gaps: KnowledgeRequirement[]): UserQuestion[];
-declare function acquisitionPlansForKnowledgeGaps(gaps: KnowledgeRequirement[]): DataAcquisitionPlan[];
+export declare function userQuestionsForKnowledgeGaps(gaps: KnowledgeRequirement[]): UserQuestion[];
+export declare function acquisitionPlansForKnowledgeGaps(gaps: KnowledgeRequirement[]): DataAcquisitionPlan[];
 //#endregion
 //#region src/leaderboard.d.ts
 interface LeaderboardRow {
@@ -2049,7 +2049,7 @@ interface LeaderboardOptions {
  * group, and rank by pass-rate (cost as the tiebreaker — cheaper wins a tie).
  * Pure projection: no I/O, deterministic, safe to call on any RunRecord[].
  */
-declare function leaderboard(records: RunRecord[], opts: LeaderboardOptions): LeaderboardRow[];
+export declare function leaderboard(records: RunRecord[], opts: LeaderboardOptions): LeaderboardRow[];
 //#endregion
 //#region src/metrics.d.ts
 interface TokenPrice {
@@ -2057,20 +2057,20 @@ interface TokenPrice {
   output: number;
 }
 /** Per-1K token pricing for exact model ids. */
-declare const MODEL_PRICING: Record<string, TokenPrice>;
+export declare const MODEL_PRICING: Record<string, TokenPrice>;
 /** Resolve pricing for a model id: exact table, then family fallback.
  *  Returns null when the id matches nothing (caller decides — never a
  *  silent-zero masquerading as a real $0 cost). */
-declare function resolveModelPricing(model: string): TokenPrice | null;
+export declare function resolveModelPricing(model: string): TokenPrice | null;
 /** True when `model` has known pricing (exact or family). Lets cost-aware
  *  callers distinguish a real $0 from an unpriced model. */
-declare function isModelPriced(model: string): boolean;
+export declare function isModelPriced(model: string): boolean;
 /** Estimate token count from string length (chars / 4 approximation) */
-declare function estimateTokens(text: string): number;
+export declare function estimateTokens(text: string): number;
 /** Calculate cost in USD from token counts and model. Unknown models warn
  *  once (not a silent zero) and return 0 so callers that ignore pricing keep
  *  working; cost-sensitive callers should gate on {@link isModelPriced}. */
-declare function estimateCost(inputTokens: number, outputTokens: number, model: string): number;
+export declare function estimateCost(inputTokens: number, outputTokens: number, model: string): number;
 //#endregion
 //#region src/paired-delta-test.d.ts
 interface PairedDeltaTestOptions extends PairedBootstrapOptions {
@@ -2096,7 +2096,7 @@ interface PairedDeltaTestResult {
   significant: boolean;
 }
 /** Smallest all-positive sample that can clear a one-sided exact sign test. */
-declare function minimumPairsForPairedDeltaTest(confidence?: number): number;
+export declare function minimumPairsForPairedDeltaTest(confidence?: number): number;
 /**
  * Tests whether a paired candidate-minus-baseline delta clears a threshold.
  *
@@ -2132,7 +2132,7 @@ declare function minimumPairsForPairedDeltaTest(confidence?: number): number;
  * use {@link decidePairedPromotion}, which routes those to Tango's score
  * interval, rather than thresholding this function's bootstrap directly.
  */
-declare function pairedDeltaTest(before: number[], after: number[], options?: PairedDeltaTestOptions): PairedDeltaTestResult;
+export declare function pairedDeltaTest(before: number[], after: number[], options?: PairedDeltaTestOptions): PairedDeltaTestResult;
 //#endregion
 //#region src/partition-held-out.d.ts
 interface PartitionHeldOutOptions {
@@ -2170,7 +2170,7 @@ interface HeldOutPartition {
  * Order within each bucket follows the input order, so the result is stable and
  * reproducible.
  */
-declare function partitionHeldOut(ids: string[], options?: PartitionHeldOutOptions): HeldOutPartition;
+export declare function partitionHeldOut(ids: string[], options?: PartitionHeldOutOptions): HeldOutPartition;
 interface TrainSelectionTestOptions<T> {
   /** Partition seed. Bumping it reshuffles every unit. Default 'train-selection-test-v1'. */
   seed?: string;
@@ -2212,7 +2212,7 @@ interface TrainSelectionTestPartition<T> {
  * (0, 1) or summing to 1 or more, and a partition below its floor all throw.
  * Order within each partition follows the input order.
  */
-declare function partitionTrainSelectionTest<T extends {
+export declare function partitionTrainSelectionTest<T extends {
   id: string;
 }>(items: readonly T[], options?: TrainSelectionTestOptions<T>): TrainSelectionTestPartition<T>;
 //#endregion
@@ -2269,15 +2269,15 @@ interface ProductBenchmarkExportResult {
   readonly records: number;
 }
 /** Repo identity from the exporting process's cwd. `'unknown'` values are flagged by `validateProductBenchmarkRun`. */
-declare function productBenchmarkRepoIdentity(): ProductBenchmarkManifest['repo'];
+export declare function productBenchmarkRepoIdentity(): ProductBenchmarkManifest['repo'];
 /** Single-run convenience wrapper over `exportProductBenchmarkRuns`. */
-declare function exportProductBenchmark(options: ProductBenchmarkSingleRunExportOptions): ProductBenchmarkExportResult;
+export declare function exportProductBenchmark(options: ProductBenchmarkSingleRunExportOptions): ProductBenchmarkExportResult;
 /**
  * Export one or more product eval run dirs into a validated product
  * benchmark bundle at `outDir`. Both the manifest and every record are
  * run through the contract validators before anything is written.
  */
-declare function exportProductBenchmarkRuns(options: ProductBenchmarkExportOptions): ProductBenchmarkExportResult;
+export declare function exportProductBenchmarkRuns(options: ProductBenchmarkExportOptions): ProductBenchmarkExportResult;
 //#endregion
 //#region src/product-benchmark/index.d.ts
 declare const productBenchmarkSplits: readonly ['practice', 'dev', 'holdout', 'safety', 'sentinel'];
@@ -2350,13 +2350,13 @@ interface ProductBenchmarkValidationReport {
   readonly integrityFailures: readonly string[];
   readonly missingArtifacts: readonly string[];
 }
-declare function readProductBenchmarkManifest(path: string): ProductBenchmarkManifest;
+export declare function readProductBenchmarkManifest(path: string): ProductBenchmarkManifest;
 /**
  * Fail-loud gate over a bundle directory: locates the manifest + records,
  * runs `validateProductBenchmarkRun`, and throws with every repo,
  * integrity, and artifact failure listed. Returns the report when clean.
  */
-declare function assertProductBenchmarkRun(runDir: string): ProductBenchmarkValidationReport;
+export declare function assertProductBenchmarkRun(runDir: string): ProductBenchmarkValidationReport;
 //#endregion
 //#region src/runtime-trajectory.d.ts
 interface RuntimeTrajectoryHookEvent {
@@ -2402,7 +2402,7 @@ interface ProjectRuntimeTrajectoryEvidenceOptions<TRecord extends RuntimeTraject
   recordIdOf?: (record: TRecord, index: number) => string | undefined;
   scenarioIdOf?: (record: TRecord, index: number) => string | undefined;
 }
-declare function projectRuntimeTrajectoryEvidence<TRecord extends RuntimeTrajectoryRecord>(options: ProjectRuntimeTrajectoryEvidenceOptions<TRecord>): RuntimeTrajectoryEvidenceProjection;
+export declare function projectRuntimeTrajectoryEvidence<TRecord extends RuntimeTrajectoryRecord>(options: ProjectRuntimeTrajectoryEvidenceOptions<TRecord>): RuntimeTrajectoryEvidenceProjection;
 //#endregion
 //#region src/action-policy.d.ts
 interface ActionExecutionPolicy {
@@ -2424,7 +2424,7 @@ interface ActionPolicyDecision {
   reasons: string[];
   label?: FeedbackLabel;
 }
-declare function evaluateActionPolicy(action: ProposedSideEffect, policy?: ActionExecutionPolicy, options?: {
+export declare function evaluateActionPolicy(action: ProposedSideEffect, policy?: ActionExecutionPolicy, options?: {
   createdAt?: string;
 }): ActionPolicyDecision;
 //#endregion
@@ -2435,7 +2435,7 @@ interface CanaryLeak {
   runId?: string;
   evidence: string;
 }
-declare function checkCanaries(output: string, scenarios: DatasetScenario[]): CanaryLeak[];
+export declare function checkCanaries(output: string, scenarios: DatasetScenario[]): CanaryLeak[];
 //#endregion
 //#region src/deploy-gate-layer.d.ts
 type DeployFamily = 'frontend-static' | 'nextjs' | 'remix' | 'fullstack-ts';
@@ -2478,7 +2478,7 @@ interface DeployGateLayerInput {
  * server-rendered build that prints a manifest), set `requireArtifact:
  * false` and rely on the runner's own ok signal.
  */
-declare function deployGateLayer<Env = unknown>(input: DeployGateLayerInput): Layer<Env>;
+export declare function deployGateLayer<Env = unknown>(input: DeployGateLayerInput): Layer<Env>;
 interface ViteDeployRunnerInput {
   /** Workdir to build. The runner cd's here. */
   workdir: string;
@@ -2513,7 +2513,7 @@ interface ViteDeployRunnerInput {
  * validates `<artifactDir>/<artifactEntry>` exists. Use as the `runner:`
  * factory for {@link deployGateLayer}.
  */
-declare function viteDeployRunner(input: ViteDeployRunnerInput): DeployRunner;
+export declare function viteDeployRunner(input: ViteDeployRunnerInput): DeployRunner;
 interface WranglerDeployRunnerInput {
   workdir: string;
   exec: (cmd: string, opts?: {
@@ -2542,7 +2542,7 @@ interface WranglerDeployRunnerInput {
  * No wrangler config = skip with "no wrangler" evidence (not a failure
  * — the gate caller decides whether to require deploy validation).
  */
-declare function wranglerDeployRunner(input: WranglerDeployRunnerInput): DeployRunner;
+export declare function wranglerDeployRunner(input: WranglerDeployRunnerInput): DeployRunner;
 //#endregion
 //#region src/hidden-criteria-grading.d.ts
 /**
@@ -2587,7 +2587,7 @@ interface RoutedField {
  * `values` is an authoring error (fail loud) — every routed field must have a
  * value the firewall can check.
  */
-declare function routeFields<TName extends string>(routing: Readonly<Record<TName, FieldDestination>>, values: Readonly<Record<TName, string>>): RoutedField[];
+export declare function routeFields<TName extends string>(routing: Readonly<Record<TName, FieldDestination>>, values: Readonly<Record<TName, string>>): RoutedField[];
 /** A single detected leak: a hidden field whose value appears in the agent context. */
 interface HiddenLeak {
   field: string;
@@ -2610,7 +2610,7 @@ interface NoLeakOptions {
  * that matters — a hidden answer key, held-out case, or rubric anchor pasted
  * into the prompt. Returns the routed fields on success so a caller can chain.
  */
-declare function assertNoHiddenLeak(fields: readonly RoutedField[], agentContext: string, opts?: NoLeakOptions): readonly RoutedField[];
+export declare function assertNoHiddenLeak(fields: readonly RoutedField[], agentContext: string, opts?: NoLeakOptions): readonly RoutedField[];
 /** What a hidden-criteria grader reports. `passRate = passed / total` over the
  *  hidden checks; `total === 0` means the criteria never ran (e.g. the artifact
  *  did not even load) — an honest zero, never a spurious pass. */
@@ -2639,7 +2639,7 @@ type HiddenCriteriaGrader<TArtifact, THidden> = (artifact: TArtifact, hiddenCrit
 /** Normalize a grader's raw `{passed, total}` into a `HiddenGradeResult` with a
  *  consistent, fail-loud `passRate` — the canonical "honest zero on no-run"
  *  rule, single-sourced so every domain grader gets it. */
-declare function hiddenGrade(passed: number, total: number, notes?: string): HiddenGradeResult;
+export declare function hiddenGrade(passed: number, total: number, notes?: string): HiddenGradeResult;
 /**
  * Run a domain's hidden-criteria grader behind the firewall. Before grading, it
  * re-asserts the firewall against the agent context the run actually used —
@@ -2649,7 +2649,7 @@ declare function hiddenGrade(passed: number, total: number, notes?: string): Hid
  * this; a domain that already asserted the firewall at dispatch time can call
  * its grader directly and feed the result to `blendHeldout`.
  */
-declare function gradeOnHidden<TArtifact, THidden>(args: {
+export declare function gradeOnHidden<TArtifact, THidden>(args: {
   artifact: TArtifact;
   hiddenCriteria: THidden;
   grader: HiddenCriteriaGrader<TArtifact, THidden>;
@@ -2671,7 +2671,7 @@ interface BlendWeights {
 }
 /** Default blend: 0.7 hidden correctness, 0.3 judge quality. The coding bench's
  *  long-standing split — execution truth dominates, style refines. */
-declare const defaultBlendWeights: BlendWeights;
+export declare const defaultBlendWeights: BlendWeights;
 /** The input shape a judge's `score` receives — exactly `JudgeConfig.score`'s
  *  argument: the artifact, plus any scenario/signal fields the judge carries.
  *  `withHeldoutBlend` only reads `artifact`; the rest rides through. */
@@ -2688,7 +2688,7 @@ interface JudgeScoreInput<TArtifact> {
  * CORRECT solution still earns the bulk of the points. Both inputs are clamped
  * to [0,1] — a judge on a non-unit scale must be normalized by the caller first.
  */
-declare function blendHeldout(heldoutPassRate: number, judgeScore: number, weights?: BlendWeights): number;
+export declare function blendHeldout(heldoutPassRate: number, judgeScore: number, weights?: BlendWeights): number;
 /**
  * Wrap a judge's `score` so the `composite` it REPORTS is the held-out-weighted
  * blend. The judge still scores its quality dimensions (recorded, secondary),
@@ -2702,7 +2702,7 @@ declare function blendHeldout(heldoutPassRate: number, judgeScore: number, weigh
  * function. The input is the judge's `{ artifact, ... }` — any extra fields
  * (`scenario`, `signal`) ride through untouched via the index signature.
  */
-declare function withHeldoutBlend<TArtifact>(score: (input: JudgeScoreInput<TArtifact>) => JudgeScore$1 | Promise<JudgeScore$1>, heldoutPassRate: (artifact: TArtifact) => number, weights?: BlendWeights): (input: JudgeScoreInput<TArtifact>) => Promise<JudgeScore$1>;
+export declare function withHeldoutBlend<TArtifact>(score: (input: JudgeScoreInput<TArtifact>) => JudgeScore$1 | Promise<JudgeScore$1>, heldoutPassRate: (artifact: TArtifact) => number, weights?: BlendWeights): (input: JudgeScoreInput<TArtifact>) => Promise<JudgeScore$1>;
 //#endregion
 //#region src/integrity/single-backend.d.ts
 /**
@@ -2750,7 +2750,7 @@ interface SingleBackendReport {
  * way that would re-route the judge through a different backend than the
  * agent. Returns the report so callers can log it in either case.
  */
-declare function assertSingleBackend(agent: BackendDescriptor, judge: BackendDescriptor, opts?: AssertSingleBackendOptions): SingleBackendReport;
+export declare function assertSingleBackend(agent: BackendDescriptor, judge: BackendDescriptor, opts?: AssertSingleBackendOptions): SingleBackendReport;
 //#endregion
 //#region src/intent-match-judge.d.ts
 interface IntentMatchInput {
@@ -2798,7 +2798,7 @@ interface IntentMatchOptions {
 /**
  * Run the intent-match judge. Soft-fails to available=false on error.
  */
-declare function runIntentMatchJudge(input: IntentMatchInput, options: IntentMatchOptions): Promise<IntentMatchResult>;
+export declare function runIntentMatchJudge(input: IntentMatchInput, options: IntentMatchOptions): Promise<IntentMatchResult>;
 //#endregion
 //#region src/judge-ensemble.d.ts
 /** One judge's verdict. `perDimension: null` ⇒ that judge call failed (threw or
@@ -2850,7 +2850,7 @@ interface EnsembleAggregate<D extends string = string> {
  *   sum-normalized weighting, not re-implemented here.
  * - Throws if `verdicts` or `dimensionKeys` is empty, or if every judge failed.
  */
-declare function aggregateJudgeVerdicts<D extends string>(verdicts: readonly JudgeVerdict<D>[], dimensionKeys: readonly D[], weights?: Partial<Record<D, number>>): EnsembleAggregate<D>;
+export declare function aggregateJudgeVerdicts<D extends string>(verdicts: readonly JudgeVerdict<D>[], dimensionKeys: readonly D[], weights?: Partial<Record<D, number>>): EnsembleAggregate<D>;
 //#endregion
 //#region src/judge-retry.d.ts
 /**
@@ -2921,7 +2921,7 @@ interface JudgeRetryOutcome<T> {
  * `value`. The library refuses to default to a silent zero score because a
  * synthetic zero is indistinguishable from a real low score downstream.
  */
-declare function withJudgeRetry<T>(judgeFn: (model: string, signal: AbortSignal) => Promise<T>, policy?: JudgeRetryPolicy): Promise<JudgeRetryOutcome<T>>;
+export declare function withJudgeRetry<T>(judgeFn: (model: string, signal: AbortSignal) => Promise<T>, policy?: JudgeRetryPolicy): Promise<JudgeRetryOutcome<T>>;
 //#endregion
 //#region src/judge-panel.d.ts
 interface EnsembleJudgeOptions<D extends string> {
@@ -2979,7 +2979,7 @@ interface EnsembleJudgeOptions<D extends string> {
  * fails, `aggregateJudgeVerdicts` throws — the campaign engine records a
  * failed cell instead of averaging a fabricated zero.
  */
-declare function ensembleJudge<D extends string>(opts: EnsembleJudgeOptions<D>): JudgeConfig<unknown>;
+export declare function ensembleJudge<D extends string>(opts: EnsembleJudgeOptions<D>): JudgeConfig<unknown>;
 //#endregion
 //#region src/keyword-coverage-judge.d.ts
 /**
@@ -3050,13 +3050,13 @@ interface KeywordCoverageOptions {
  * and doesn't want a fresh HTTP round-trip — e.g. sandbox runtime where
  * the preview content was fetched via curl from inside the container.
  */
-declare function runKeywordCoverageJudge(html: string, expectedConcepts: ReadonlyArray<KeywordConceptSpec>, assets?: ReadonlyArray<string>): KeywordCoverageResult;
+export declare function runKeywordCoverageJudge(html: string, expectedConcepts: ReadonlyArray<KeywordConceptSpec>, assets?: ReadonlyArray<string>): KeywordCoverageResult;
 /**
  * URL-fetch flavor — GET the preview, parallel-fetch every linked
  * stylesheet + script (with bounded timeouts, soft-fail individually),
  * then score via {@link runKeywordCoverageJudge}.
  */
-declare function runKeywordCoverageJudgeUrl(previewUrl: string, expectedConcepts: ReadonlyArray<KeywordConceptSpec>, options?: KeywordCoverageOptions): Promise<KeywordCoverageResult>;
+export declare function runKeywordCoverageJudgeUrl(previewUrl: string, expectedConcepts: ReadonlyArray<KeywordConceptSpec>, options?: KeywordCoverageOptions): Promise<KeywordCoverageResult>;
 //#endregion
 //#region src/oracle.d.ts
 interface OracleObservation {
@@ -3079,19 +3079,19 @@ interface Oracle {
   id: string;
   check(obs: OracleObservation): OracleResult;
 }
-declare function textInSnapshot(needle: string, opts?: {
+export declare function textInSnapshot(needle: string, opts?: {
   caseSensitive?: boolean;
 }): Oracle;
-declare function urlContains(fragment: string): Oracle;
-declare function jsonShape(expected: Record<string, unknown>): Oracle;
-declare function regexMatches(pattern: RegExp): Oracle;
+export declare function urlContains(fragment: string): Oracle;
+export declare function jsonShape(expected: Record<string, unknown>): Oracle;
+export declare function regexMatches(pattern: RegExp): Oracle;
 /**
  * Anti-bot detector — distinguishes genuine failures from blocked navigation
  * (cloudflare, recaptcha, etc). Returns an Oracle that PASSES when no block
  * marker is present; on block, detail names the blocker so runners can tag
  * results as "blocked" rather than "failed". Lifted from browser-agent-driver.
  */
-declare function notBlocked(): Oracle;
+export declare function notBlocked(): Oracle;
 /** Extends the substrate verdict spine: `valid` = every oracle passed (and
  *  at least one ran); `score` = ratio of oracles passed; `scores` = per-
  *  oracle 0|1 keyed by oracle id. */
@@ -3101,7 +3101,7 @@ interface OracleReport extends DefaultVerdict {
   failCount: number;
 }
 /** Run all oracles against one observation and aggregate. */
-declare function evaluateOracles(obs: OracleObservation, oracles: Oracle[]): OracleReport;
+export declare function evaluateOracles(obs: OracleObservation, oracles: Oracle[]): OracleReport;
 //#endregion
 //#region src/paired-promotion-power.d.ts
 /** A serializable law for one scalar draw. */
@@ -3248,7 +3248,7 @@ interface RequiredPairsForPairedPromotionResult {
  * Joint power of the configured calls at `n` under the registered alternative:
  * the fraction of simulated data sets on which every call promotes.
  */
-declare function pairedPromotionPower(options: PairedPromotionPowerOptions): PairedPromotionPowerResult;
+export declare function pairedPromotionPower(options: PairedPromotionPowerOptions): PairedPromotionPowerResult;
 /**
  * First n at which {@link pairedPromotionPower} reaches the target, scanning
  * every n from `minPairs` upward. Both the point-estimate answer and the
@@ -3258,7 +3258,7 @@ declare function pairedPromotionPower(options: PairedPromotionPowerOptions): Pai
  * simulation at that n on a different seed; a candidate that fails it is
  * recorded and the scan continues.
  */
-declare function requiredPairsForPairedPromotion(options: RequiredPairsForPairedPromotionOptions): RequiredPairsForPairedPromotionResult;
+export declare function requiredPairsForPairedPromotion(options: RequiredPairsForPairedPromotionOptions): RequiredPairsForPairedPromotionResult;
 //#endregion
 //#region src/treatment-gate.d.ts
 /** A tool-name matcher: does this tool name belong to the treatment under test?
@@ -3296,7 +3296,7 @@ interface TreatmentGate {
  * Core predicate. Gated (treatment-not-applied) ONLY when telemetry was
  * captured AND zero matching calls appear; an empty histogram fails open.
  */
-declare function gateTreatmentApplied(input: TreatmentGateInput, opts?: TreatmentGateOptions): TreatmentGate;
+export declare function gateTreatmentApplied(input: TreatmentGateInput, opts?: TreatmentGateOptions): TreatmentGate;
 //#endregion
 //#region src/verdict-cache.d.ts
 /**
@@ -3305,11 +3305,11 @@ declare function gateTreatmentApplied(input: TreatmentGateInput, opts?: Treatmen
  * insertion order. Throws `LedgerCanonicalizationError` on any value with no
  * faithful JSON form — see `ledger-core/canonical`.
  */
-declare function canonicalJson(value: unknown): string;
+export declare function canonicalJson(value: unknown): string;
 /** Hex sha-256 over `canonicalJson(value)`. The content address used by the
  *  verdict cache and report attestation. Equals `hashCanonical(value)` with
  *  the `sha256:` prefix removed. */
-declare function contentHash(value: unknown): string;
+export declare function contentHash(value: unknown): string;
 /** Pluggable verdict store. Sync or async on both legs — `cachedJudge`
  *  awaits the results either way. */
 interface VerdictCacheStore {
@@ -3323,7 +3323,7 @@ interface VerdictCacheStore {
  * file:line — a skipped line would silently re-judge (cost) or, worse, mask
  * a half-written file that needs operator attention.
  */
-declare function fileVerdictCache(path: string): VerdictCacheStore;
+export declare function fileVerdictCache(path: string): VerdictCacheStore;
 //#endregion
 //#region src/anti-slop.d.ts
 interface AntiSlopConfig {
@@ -3348,7 +3348,7 @@ interface AntiSlopConfig {
 }
 type SlopCategory = 'banned_phrase' | 'banned_opening' | 'hedging' | 'apology' | 'repetition' | 'length';
 /** Create a reusable Judge function from an anti-slop config. */
-declare function createAntiSlopJudge(config?: AntiSlopConfig): JudgeFn;
+export declare function createAntiSlopJudge(config?: AntiSlopConfig): JudgeFn;
 interface AntiSlopIssue {
   category: SlopCategory;
   detail: string;
@@ -3365,7 +3365,7 @@ interface AntiSlopReport {
  * Pure function — analyze one or more outputs against the config. Exposed
  * separately so consumers can build their own reporters on top.
  */
-declare function analyzeAntiSlop(outputs: string[], config: Omit<Required<AntiSlopConfig>, 'domain'> & {
+export declare function analyzeAntiSlop(outputs: string[], config: Omit<Required<AntiSlopConfig>, 'domain'> & {
   penaltyWeights: Record<SlopCategory, number>;
 }): AntiSlopReport;
 //#endregion
@@ -3384,7 +3384,7 @@ interface RunRecordBackend {
  * loud on load (a corrupt corpus must not silently shrink the analysis set);
  * pass `skipInvalid` only for forensics on a known-bad file.
  */
-declare function jsonlRunRecordBackend(path: string, opts?: {
+export declare function jsonlRunRecordBackend(path: string, opts?: {
   skipInvalid?: boolean;
 }): RunRecordBackend;
 //#endregion
@@ -3461,7 +3461,7 @@ interface PreflightOutcome {
  * additionally gets a small chat probe so a model that is listed but
  * unconfigured (a 401 `model_not_found` from the router) is caught.
  */
-declare function preflightModels(opts: PreflightModelsOptions): Promise<PreflightOutcome>;
+export declare function preflightModels(opts: PreflightModelsOptions): Promise<PreflightOutcome>;
 //#endregion
 //#region src/model-seats.d.ts
 interface ModelSeats {
@@ -3500,7 +3500,7 @@ type SeatName = keyof ModelSeats;
  * your own: `{ ...seatPresets.frontier, worker: '<your-frontier-id>', … }` —
  * `resolveSeat` throws on every seat you haven't filled.
  */
-declare const seatPresets: Record<'economy' | 'frontier', ModelSeats>;
+export declare const seatPresets: Record<'economy' | 'frontier', ModelSeats>;
 /**
  * Read one seat. Blank strings and empty arrays count as unset (env-var
  * plumbing produces them); malformed values (non-string seat, non-array or
@@ -3508,9 +3508,9 @@ declare const seatPresets: Record<'economy' | 'frontier', ModelSeats>;
  * explicit `fallback` is returned (`[fallback]` for `judges` — a one-model
  * panel); without one, `SeatUnsetError`.
  */
-declare function resolveSeat(seats: ModelSeats, seat: 'judges', fallback?: string): string[];
-declare function resolveSeat(seats: ModelSeats, seat: Exclude<SeatName, 'judges'>, fallback?: string): string;
-declare function resolveSeat(seats: ModelSeats, seat: SeatName, fallback?: string): string | string[];
+export declare function resolveSeat(seats: ModelSeats, seat: 'judges', fallback?: string): string[];
+export declare function resolveSeat(seats: ModelSeats, seat: Exclude<SeatName, 'judges'>, fallback?: string): string;
+export declare function resolveSeat(seats: ModelSeats, seat: SeatName, fallback?: string): string | string[];
 //#endregion
 //#region src/prompt-registry.d.ts
 /**
@@ -3534,7 +3534,7 @@ interface PromptHandle {
   /** Full prompt body */
   content: string;
 }
-declare class PromptRegistry {
+export declare class PromptRegistry {
   private readonly entries;
   /**
    * Register a prompt. Re-registering the same id+version with DIFFERENT
@@ -3552,7 +3552,7 @@ declare class PromptRegistry {
   verifyHash(id: string, version: string, expectedHash: string): boolean | null;
 }
 /** SHA-256(content) → first 12 hex chars. Stable across runtimes. */
-declare function hashContent(content: string): Promise<string>;
+export declare function hashContent(content: string): Promise<string>;
 //#endregion
 //#region src/reference-replay.d.ts
 /**
@@ -3691,9 +3691,9 @@ interface SteeringOptimizerConfig {
  * Model-backed candidate generation belongs to the official GEPA campaign
  * integration. This class only performs the deterministic selection it owns.
  */
-declare class PairwiseSteeringOptimizer {
+export declare class PairwiseSteeringOptimizer {
   optimize(rows: SteeringOptimizationRow[], config?: SteeringOptimizerConfig): SteeringOptimizationResult;
 }
 //#endregion
-export { AGENT_PROFILE_KINDS, ATTESTATION_ALGORITHM, type ActionExecutionPolicy, type ActionPolicyDecision, type ActionableSideInfo, type ActiveLearningOptions, type AgentEvalAgent, AgentEvalError, type AgentEvalErrorCode, type AgentEvalEvaluateOptions, type AgentEvalImproveOptions, type AgentProfile, type AgentProfileCell, type AgentProfileCellInput, type AgentProfileDimensionValue, type AgentProfileJson, type AgentProfileSourceInput, type Analyst, type AnalystContext, type AnalystFeedbackTrajectoryOptions, type AnalystFinding, type AnalystFindingDigest, AnalystRegistry, type AnalystRegistryOptions, type AnalystReviewDecision, type AnalystReviewRequest, type AnalystRunDigest, type AnalystRunEvent, type AnalystRunInputs, type AnalystRunResult, type AnalystRunSummary, type AnalystSeverity, type AnalystUsageReceipt, type AnalyzeRunsOptions, type AnalyzeTracesResult, type ArgumentCheck, type Artifact, type ArtifactCheck, type ArtifactEventLike, type ArtifactResult, type AssertCapabilityHeadroomOptions, type AssertCrossFamilyOptions, type AssertCrossFamilyServedOptions, type AssertServedModelOptions, type AssertSingleBackendOptions, type AttestationProvenance, type AttestationVerification, type AttestedReport, BOOTSTRAP_GATE_MIN_N, type BackendDescriptor, BackendIntegrityError, type BackendIntegrityReport, type BenchmarkEvaluation, type BlendWeights, type BootstrapOptions, type BootstrapResult, type BoundedProcessInput, type BoundedProcessResult, BudgetBreachError, BudgetGuard, type BudgetLedgerEntry, type BudgetPolicy, type BudgetSpec, CODING_HARNESSES, type CalibrationResult, type CampaignCellFailureReceipt, type CampaignCellRetryPolicy, type CampaignFactoryParams, type CampaignIntegrityPolicy, type CampaignResult, type CampaignRunContext, type CampaignRunOutcome, type CampaignScenario, type CampaignVariant, type CanaryAlert, type CanaryEvaluation, type CanaryKind, type CanaryLeak, type CanaryOptions, type CanaryReport, type CandidateScore, type CapabilityHeadroomOptions, type CapabilityHeadroomResult, type CaptureErrorHandler, type CaptureWrite, type CellVerdict, type ChannelRollup, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTraceEngineOptions, type ChatTransport, type CheckResult, type CheckerIdentity, type CheckerOutcome, type CliffsMagnitude, type CollectedArtifacts, type CommandRunner, type ComparePairedArmsOptions, type CompletionCriterion, type CompletionRequirement, type CompletionVerdict, type ConceptFinding, type ConceptSpec, ConfigError, type ContinuousAgreement, type ContinuousAgreementOptions, type ContinuousCalibrationResult, type ContractCheckResult, type ContractLintFinding, type ContractRule, type ContractRuleExecution, type ContractRun, type ContractSpan, type ContractStatus, type ContractVerdict, type ControlActionOutcome, type ControlBudget, type ControlContext, type ControlDecision, type ControlEvalResult, type ControlRunResult, type ControlRuntimeConfig, type ControlRuntimeError, type ControlStep, type CorpusAgreementOptions, type CorpusAgreementPerDimension, type CorpusAgreementReport, type CorpusScoreRecord, type CorrectnessChecker, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, type CostChannel, type CostEntry, CostLedger, type CostLedgerFilter, type CostLedgerHandle, type CostLedgerOptions, type CostLedgerPersistence, CostLedgerPersistenceError, type CostLedgerSummary, type CostProvenance, type CostReceipt, CostReceiptCaptureError, type CostReceiptInput, CostReservationExceededError, type CostResult, type CostSummary, CostTracker, type CostUsage, type CounterfactualContext, type CounterfactualMutation, type CounterfactualResult, type CounterfactualRunner, type CreateChatClientOpts, type CreateTraceAnalystOptions, CrossFamilyError, type CustomTokenPricing, type CustomTransportOpts, DECISION_PAIRED_DELTA_STATISTIC, DEFAULT_BOUNDED_PROCESS_TIMEOUT_MS, DEFAULT_FAILURE_REASON_RULES, DEFAULT_PERMUTATIONS, DEFAULT_RED_TEAM_CORPUS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, type DataAcquisitionPlan, type DatasetManifest, type DatasetOverview, type DatasetScenario, type DatasetSplit, type DefaultAnalystRegistryOptions, type DefaultVerdict, type DefineAgentEvalOptions, type DefinedAgentEval, type DeployGateLayerInput, type DeployRunResult, type DeployRunner, type DetectorEvent, type DetectorSignal, type DiffPolicy, type DiffScorecardOptions, type JudgeScore as DimensionJudgeScore, type DirEntry, type DiscoverPersonasOptions, type DiscoveredPersona, type DriverState, type DspyRlmTraceEngineOptions, type EProcess, type EProcessOptions, type EProcessState, type EProcessStep, ERROR_COUNT_PATTERNS, type EnsembleAggregate, type EnsembleJudgeOptions, type EquivalenceArm, type EquivalenceCheckDefinition, type EquivalenceCheckSpec, type EquivalenceChecker, type EquivalenceCheckerInput, type EquivalenceCheckerResult, type EquivalenceObligation, type EquivalenceObligationStatus, EquivalenceProtocolError, type EquivalenceRecord, type ErrorCluster, type ErrorCountPattern, type ErrorStreakOptions, type EvalCampaignOptions, type EvalCampaignResult, type EventFilter, type EvidenceRef, type ExactAnalystRunEvent, type ExactAnalystRunResult, type ExactCapableAnalyst, type ExactRegistryRunOpts, type ExactRiskDifferenceResult, type ExperimentPlan, type ExperimentRep, type ExperimentResult, type ExperimentStats, type ExtractOptions, type ExtractResult, type ExtractedUsage, FAILURE_BLAME, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, type FailedRun, type FailureBlame, type FailureClass, type FailureClassification, type FailureContext, type FailureMode, type FailureReasonRule, type FailureRule, type FeedbackArtifactType, type FeedbackAttempt, type FeedbackLabel, type FeedbackLabelKind, type FeedbackLabelSource, type FeedbackOptimizerRow, type FeedbackOutcome, type FeedbackSplitPolicy, type FeedbackTask, type FeedbackTrajectory, type FeedbackTrajectoryFilter, type FeedbackTrajectoryStore, type FieldDestination, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, FileSystemTraceStore, type Finding, type FindingSubject, type FindingSubjectKind, type FindingToPolicyEditOptions, type FindingsDiff, FindingsStore, type GainDistributionBin, type GainDistributionFigureSpec, type GainDistributionOptions, type GateDecision, type GateEvidence, type GenericSpan, type GoldenItem, HARNESS_NATIVE_MODEL, type HarnessType, type HeadroomInput, HeldOutGate, type HeldOutGateConfig, type GateDecision$1 as HeldOutGateDecision, type HeldOutGateRejectionCode, type HeldOutPartition, type HiddenCriteriaGrader, type HiddenGradeResult, type HiddenLeak, INFRA_FAILURE_BLAMES, type ImprovementThresholds, type ImprovementVerdictResult, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, InMemoryTraceStore, type InsightReport, type IntentMatchInput, type IntentMatchOptions, type IntentMatchResult, type InterimReleaseConfidence, type InterimReleaseConfidenceInput, JudgeError, type JudgeFamily, type JudgeFn, type JudgeInput, type JudgeRetryOutcome, type JudgeRetryPolicy, type JudgeRubric, type JudgeScore$1 as JudgeScore, type JudgeScoreInput, type JudgeScoresRecord, type JudgeSpan, type JudgeVerdict, type KeywordConceptSpec, type KeywordCoverageFinding, type KeywordCoverageOptions, type KeywordCoverageResult, type KnowledgeAcquisitionMode, type KnowledgeBundle, type KnowledgeFreshness, type KnowledgeImportance, type KnowledgeReadinessReport, type KnowledgeRequirement, type KnowledgeRequirementCategory, type KnowledgeSensitivity, LOCAL_COMMAND_RUNNER_DEFAULT_CAP_MS, type Layer, type LayerResult, type LayerStatus, type LeaderboardOptions, type LeaderboardRow, LlmCallError, type LlmCallMetadata, type LlmCallRequest, type LlmCallResult, type LlmChargeBounds, type LlmCorrectnessCheckerOpts, type LlmJudgeOptions, type LlmMessage, LlmResponseError, type LlmReviewerConfig, type LlmSpan, type LlmTokenLogprob, type LlmToolCall, type LlmToolChoice, type LlmToolDefinition, type LlmUsage, MANN_WHITNEY_EXACT_MAX_STATES, MANN_WHITNEY_EXACT_MAX_WORK, MODEL_PRICING, type MannWhitneyResult, type MatchedPair, type MatchedRunRecordPair, type MaximumCharge, type McNemarResult, type MintRolloutOptions, type MintRolloutResult, type MintedRolloutLine, type MockTransportOpts, type ModelEndpointCheck, type ModelEndpointRequest, type ModelSeats, ModelSubstitutionError, MultiLayerVerifier, type NoLeakOptions, NoopRawProviderSink, NotFoundError, OUTPUT_VALUE, type Objective, type OpenAiCompatibleTransportOpts, type Oracle, type OracleObservation, type OracleReport, type OracleResult, type OrderMode, type OtlpExport, OtlpFileTraceStore, type OtlpFlatLine, type OtlpSpan, POLICY_EDIT_AXES, POLICY_EDIT_CANDIDATE_RECORD_SCHEMA, POLICY_EDIT_TARGET_SURFACES, type PaidCallResult, type PairArmsOptions, type PairArmsResult, type PairRunRecordsResult, type PairedArmRow, type PairedArmsComparison, type PairedBootstrapOptions, type PairedBootstrapResult, type PairedCorrectness, type PairedDecisionMethod, type PairedDecisionShape, type PairedDecisionStatistic, type PairedDeltaTestOptions, type PairedDeltaTestResult, type PairedEvalueOptions, type PairedEvalueSequence, type PairedEvalueStep, type PairedMcNemarEvidence, type PairedMetricDelta, type PairedPromotionAlternative, type PairedPromotionAlternativeCell, type PairedPromotionCallRates, type PairedPromotionDecision, type PairedPromotionDecisionOptions, type PairedPromotionPowerCall, type PairedPromotionPowerOptions, type PairedPromotionPowerResult, type PairedSignTestResult, type PairedTTestResult, type PairedValueLaw, PairwiseSteeringOptimizer, type ParetoPoint, type ParetoResult, type PartitionHeldOutOptions, type PendingCostCall, type PendingCostCallView, type PersistedFinding, type PersonaConfig, type PolicyEdit, type PolicyEditAdmission, type PolicyEditAdmissionOptions, type PolicyEditAxis, type PolicyEditCandidateRecord, type PolicyEditChange, type PolicyEditExpectedGain, type PolicyEditGainDirection, type PolicyEditGainUnit, type PolicyEditInit, type PolicyEditRisk, type PolicyEditSchemaVersion, type PolicyEditSource, type PolicyEditTarget, type PolicyEditTargetSurface, PolicyEditValidationError, type PreferenceMemoryEntry, type ProducedProposal, type ProducedState, type ProductBenchmarkExportOptions, type ProductBenchmarkExportResult, type ProductBenchmarkManifest, type ProductBenchmarkSingleRunExportOptions, type ProductBenchmarkSplit, type ProductBenchmarkValidationReport, ProductClient, type ProductClientConfig, type Scenario as ProductScenario, type ProfileAxisSpec, type ProjectRuntimeTrajectoryEvidenceOptions, type PromptHandle, PromptRegistry, type ProportionInterval, type ProposalEventLike, type ProposalFinding, type ProposeFn, type ProposeInput, type ProposeOutput, type ProposeReviewConfig, type ProposeReviewControlAction, type ProposeReviewControlConfig, type ProposeReviewControlResult, type ProposeReviewControlState, type ProposeReviewReport, type ProposedSideEffect, type QueryTracesPage, REDACTION_VERSION, type RankTestMethod, type RankTestMethodRequest, type RankTestOptions, type RawAnalystFinding, type RawProviderEvent, type RawProviderSink, type RecordRunsOptions, type RedTeamCase, type RedTeamCategory, type RedTeamFinding, type RedTeamReport, type RedactOptions, type RedactionProfile, type RedactionReport, type ReferenceReplayCaseRun, type ReferenceReplayRun, type ReferenceReplaySplit, type ReflectionContext, type ReflectionProposal, type RegistryRunOpts, type ReleaseConfidenceInput, type ReleaseConfidenceIssue, type ReleaseConfidenceMetrics, type ReleaseConfidenceScorecard, type ReleaseTraceEvidence, type RepeatedActionOptions, type RequiredPairsForPairedPromotionOptions, type RequiredPairsForPairedPromotionResult, type RequirementCheck, type ResearchReport, type ResearchReportOptions, type Researcher, type RetryWrite, type Review, type ReviewFn, type ReviewInput, type ReviewMemoryEntry, type ReviewMemoryStore, type RiskDifferenceResult, type RolloutSearchLineage, type RouteMap, type RoutedField, type RubricDimension, type Run, type RunCampaignOptions, type RunCaptureReport, type RunCommandInput, type RunCommandResult, type RunCostProvenance, type RunFilter, RunIntegrityError, type RunIntegrityReport, type RunJudgeMetadata, type RunOutcome, type RunPaidCallInput, type RunRecord, type RunRecordBackend, RunRecordValidationError, type RunScore, type RunScoreWeights, type RunSearchCoordinates, type RunSplitTag, type RunStatus, type RunTaskFailure, type RunTerminalOutcome, type RunTokenUsage, type RunTraceRef, type RuntimeEventLike, type RuntimeTrajectoryEvidenceProjection, type RuntimeTrajectoryEvidenceSummary, type RuntimeTrajectoryHookEvent, type RuntimeTrajectoryRecord, type RuntimeTrajectoryRunRecord, SEMANTIC_CONCEPT_JUDGE_VERSION, type SandboxDriver, type SandboxSdkTransportOpts, type SatisfiedBy, type Scenario$1 as Scenario, type ScenarioCost, type ScenarioFile, type ScenarioResult, type ScorePreference, type ScoreRiskDifferenceResult, type Scorecard, type ScorecardCell, type ScorecardCellDiff, type ScorecardDiff, type ScorecardEntry, type ScorecardLogLine, type SearchSpanResult, type SearchTraceResult, type SelfImproveBudget, type SelfImproveMethodOptions, type SelfImproveMethodProvenance, type SelfImproveMethodResult, type SelfImproveOptions, type SelfImproveProgressEvent, type SelfImproveProposerOptions, type SelfImproveProposerResult, type SelfImproveResult, type SemanticConceptJudgeInput, type SemanticConceptJudgeOptions, type SemanticConceptJudgeResult, type SequentialDecision, type SeriesConvergenceOptions, type SeriesConvergenceResult, type SeriesDistribution, ServedCrossFamilyError, type ServedModelCheck, type ServedModelPolicy, type ServedModelVerdict, type Severity, type ShareSafetyStatus, type ShareSafetyVerdict, type SignTestAlternative, type SingleBackendDivergence, type SingleBackendReport, type Span, type SpanFilter, type SpanHandle, type SpanMatchRecord, type SpanPredicate, type SplitCoverage, type SteeringBundle, type SteeringChange, type SteeringOptimizationResult, type SteeringOptimizationRow, type StopDecision, type StrategyChecker, type StreamingDetector, type SummaryTable, type SummaryTableOptions, type SummaryTableRow, type SynthesisTarget, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, type TaskGold, type TaskHeadroom, type ToolCallEventLike, type ToolSpan, type ToolStats, type ToolUseMetrics, type ToolUseOptions, type TraceAnalysisEngine, type TraceAnalysisEngineResult, type TraceAnalysisStore, type TraceAnalystByteBudgets, type TraceAnalystDefinition, type TraceAnalystFilters, type TraceAnalystSpan, type TraceAnalystSpanKind, type TraceAnalystSpanStatus, type TraceAnalystTraceSummary, type TraceContract, type TraceContractSpec, TraceEmitter, type TraceEmitterOptions, type TraceEvent, type TraceInsightReadiness, type TraceInsightSuite, type TraceStore, type TrainSelectionTestOptions, type TrainSelectionTestPartition, type Trajectory, type TrajectoryStep, type TreatmentGate, type TreatmentGateInput, type TreatmentGateOptions, type TrialTrace, type Turn, type TurnResult, UNKNOWN_MODEL, type UserQuestion, VERIFICATION_STRATEGIES, VERIFICATION_STRATEGY_SOURCES, ValidationError, type VerbosityBiasResult, type Verdict, type VerdictCacheStore, type VerdictCertification, type Verification, type VerificationReport, type VerificationStrategyProfile, type VerificationStrategySource, type VerifyFn, type VerifyOptions, type ViewSpansResult, type ViewTraceOversized, type ViewTraceResult, type ViteDeployRunnerInput, WILCOXON_EXACT_MAX_N, type WeightedCompositeInput, type WeightedCompositeResult, type WilcoxonSignedRankResult, type WranglerDeployRunnerInput, acquisitionPlansForKnowledgeGaps, admitPolicyEdit, agentProfileCellHashMaterial, agentProfileCellKey, agentProfileHash, agentProfileId, aggregateJudgeVerdicts, aggregateRunScore, analystFindingDigest, analystRunDigest, analystRunToFeedbackTrajectory, analystRunToReviewRequests, analyzeAntiSlop, analyzeRuns, analyzeSeries, analyzeTraces, applyPolicyEditToSurface, argHash, assertCapabilityHeadroom, assertCrossFamily, assertCrossFamilyServed, assertNoHiddenLeak, assertNoJudgeVerdict, assertProductBenchmarkRun, assertRealBackend, assertRunCaptured, assertServedModel, assertServedModels, assertSingleBackend, assessShareSafety, assignFeedbackSplit, attest, benjaminiHochberg, blendHeldout, blockingKnowledgeEval, bonferroni, bootstrapCi, budgetBreachView, buildAgentProfileCell, buildDefaultAnalystRegistry, buildEquivalenceRecord, buildReflectionPrompt, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, calibrateJudge, calibrateJudgeContinuous, canonicalJson, capabilityHeadroom, captureFetchToRawSink, certificationEvidenceDigest, checkCanaries, checkServedModel, checkTraceContracts, clamp01, classifyFailure, classifyFailureReason, cliffsDelta, cohensD, comparePairedArms, compileTraceContractSpec, completionVerdict, computeExperimentStats, computeFindingId, computePolicyEditId, computeToolUseMetrics, confidenceInterval, contentHash, continuousAgreement, controlRunToFeedbackTrajectory, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, costForTokenPricing, costForUsage, costReceiptFromLlm, costReceiptFromLlmError, createAntiSlopJudge, createBoundedTraceAnalysisStore, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createFeedbackTrajectory, createLlmCorrectnessChecker, createLlmReviewer, createTraceAnalyst, decidePairedPromotion, defaultBlendWeights, defineAgentEval, defineEquivalenceCheck, deployGateLayer, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, domainEvidencePattern, dominates, eProcess, ensembleJudge, equivalenceVerdict, errorStreakDetector, estimateCost, estimateTokens, evaluateActionPolicy, evaluateInterimReleaseConfidence, evaluateOracles, evaluateReleaseConfidence, evaluateTraceContract, expandProfileAxes, explainTraceContract, exportProductBenchmark, exportProductBenchmarkRuns, exportRunAsOtlp, extractErrorCount, extractProducedState, extractUsage, extractUsageFromSse, failureBlame, failureClusterView, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToOptimizerRow, fileVerdictCache, formatScorecardDiff, gainHistogram, gateTreatmentApplied, gradeOnHidden, gradeSemanticStatus, groupRunsByAgentProfileCell, harnessAxisOf, hashContent, hashJson, hiddenGrade, holm, improvementVerdict, inMemoryReviewStore, inferDomainKeywords, interRaterReliability, interpretCliffs, iqr, isBinaryOutcomeVector, isJudgeSpan, isLlmSpan, isModelPriced, isPolicyEdit, isRunRecord, isToolSpan, isTransientLlmError, jsonShape, jsonlReviewStore, jsonlRunRecordBackend, judgeAgreementView, judgeFamily, judgeSpans, knowledgeReadinessTracePayload, leaderboard, lintTraceContractSpec, llmJudge, loadScorecard, localCommandRunner, makeFinding, makePolicyEdit, makePolicyEditCandidateRecord, makeProposalFinding, manifestContentDigest, mannWhitneyU, maximumChargeForLlmRequest, mcnemar, mcnemarPower, mcnemarRequiredN, minimumPairsForPairedDeltaTest, mintRolloutRows, modelHasSnapshot, modelPriceKey, mulberry32, notBlocked, objectiveEval, observeAll, otlpTextToTraceAnalysisStore, pairArms, pairRunRecords, pairedBinaryScale, pairedBootstrap, pairedCohensDz, pairedDeltaTest, pairedDeltaTieFraction, pairedEvalueSequence, pairedMde, pairedPromotionPower, pairedRiskDifference, pairedRiskDifferenceExact, pairedRiskDifferenceScore, pairedSignTest, pairedTTest, paretoChart, paretoFrontier, parseReflectionResponse, parseRunRecordSafe, partialCredit, partitionHeldOut, partitionTrainSelectionTest, passAtK, pearsonR, policyEditFromFinding, policyEditsFromFindings, preflightModels, productBenchmarkRepoIdentity, index_d_exports as profile, projectRuntimeTrajectoryEvidence, proposeSynthesisTargets, quotaExhaustedUntil, ranks, readProductBenchmarkManifest, recordRuns, recordRunsToScorecard, redTeamDataset, redTeamReport, redact, redactForShare, redactText, regexMatches, renderPreferenceMemoryMarkdown, repeatedActionDetector, requiredPairedSampleSize, requiredPairsForPairedPromotion, requiredSampleSize, resolveModelPricing, resolveSeat, roundTripRunRecord, routeFields, runAgentControlLoop, runBoundedProcess, runCampaign, runCanaries, runCostFloorUsd, runCounterfactual, runEquivalenceCheck, runEvalCampaign, runIntentMatchJudge, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runProposeReview, runProposeReviewAsControlLoop, runSemanticConceptJudge, runTaskScore, runsForScenario, scoreKnowledgeReadiness, scorePolicyEditReadiness, scoreRedTeamOutput, scoreTraceInsightReadiness, searchCellRunId, seatPresets, selfImprove, servedModelAcceptable, shareAllowed, spearmanR, stripFencedJson, subjectiveEval, summarizeBackendIntegrity, summarizeNumberSeries, summarizePreferenceMemory, summaryTable, textInSnapshot, toAgentProfileJson, tokenizeDomainWords, toolSpansToTraceAnalysisStore, toolWasteView, traceContract, transientDispatchFailure, urlContains, userQuestionsForKnowledgeGaps, validatePolicyEdit, validatePolicyEditCandidateRecord, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyAttestation, verifyCompletion, viteDeployRunner, weightedComposite, weightedMean, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit, withHeldoutBlend, withJudgeRetry, wranglerDeployRunner };
+export { AGENT_PROFILE_KINDS, ATTESTATION_ALGORITHM, type ActionExecutionPolicy, type ActionPolicyDecision, type ActionableSideInfo, type ActiveLearningOptions, type AgentEvalAgent, AgentEvalError, type AgentEvalErrorCode, type AgentEvalEvaluateOptions, type AgentEvalImproveOptions, type AgentProfile, type AgentProfileCell, type AgentProfileCellInput, type AgentProfileDimensionValue, type AgentProfileJson, type AgentProfileSourceInput, type Analyst, type AnalystContext, type AnalystFeedbackTrajectoryOptions, type AnalystFinding, type AnalystFindingDigest, AnalystRegistry, type AnalystRegistryOptions, type AnalystReviewDecision, type AnalystReviewRequest, type AnalystRunDigest, type AnalystRunEvent, type AnalystRunInputs, type AnalystRunResult, type AnalystRunSummary, type AnalystSeverity, type AnalystUsageReceipt, type AnalyzeRunsOptions, type AnalyzeTracesResult, type ArgumentCheck, type Artifact, type ArtifactCheck, type ArtifactEventLike, type ArtifactResult, type AssertCapabilityHeadroomOptions, type AssertCrossFamilyOptions, type AssertCrossFamilyServedOptions, type AssertServedModelOptions, type AssertSingleBackendOptions, type AttestationProvenance, type AttestationVerification, type AttestedReport, BOOTSTRAP_GATE_MIN_N, type BackendDescriptor, BackendIntegrityError, type BackendIntegrityReport, type BenchmarkEvaluation, type BlendWeights, type BootstrapOptions, type BootstrapResult, type BoundedProcessInput, type BoundedProcessResult, type BudgetLedgerEntry, type BudgetPolicy, type BudgetSpec, CODING_HARNESSES, type CalibrationResult, type CampaignCellFailureReceipt, type CampaignCellRetryPolicy, type CampaignFactoryParams, type CampaignIntegrityPolicy, type CampaignResult, type CampaignRunContext, type CampaignRunOutcome, type CampaignScenario, type CampaignVariant, type CanaryAlert, type CanaryEvaluation, type CanaryKind, type CanaryLeak, type CanaryOptions, type CanaryReport, type CandidateScore, type CapabilityHeadroomOptions, type CapabilityHeadroomResult, type CaptureErrorHandler, type CaptureWrite, type CellVerdict, type ChannelRollup, type ChatCallOpts, type ChatClient, type ChatRequest, type ChatResponse, type ChatTraceEngineOptions, type ChatTransport, type CheckResult, type CheckerIdentity, type CheckerOutcome, type CliffsMagnitude, type CollectedArtifacts, type CommandRunner, type ComparePairedArmsOptions, type CompletionCriterion, type CompletionRequirement, type CompletionVerdict, type ConceptFinding, type ConceptSpec, ConfigError, type ContinuousAgreement, type ContinuousAgreementOptions, type ContinuousCalibrationResult, type ContractCheckResult, type ContractLintFinding, type ContractRule, type ContractRuleExecution, type ContractRun, type ContractSpan, type ContractStatus, type ContractVerdict, type ControlActionOutcome, type ControlBudget, type ControlContext, type ControlDecision, type ControlEvalResult, type ControlRunResult, type ControlRuntimeConfig, type ControlRuntimeError, type ControlStep, type CorpusAgreementOptions, type CorpusAgreementPerDimension, type CorpusAgreementReport, type CorpusScoreRecord, type CorrectnessChecker, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, type CostChannel, type CostEntry, CostLedger, type CostLedgerFilter, type CostLedgerHandle, type CostLedgerOptions, type CostLedgerPersistence, CostLedgerPersistenceError, type CostLedgerSummary, type CostProvenance, type CostReceipt, CostReceiptCaptureError, type CostReceiptInput, CostReservationExceededError, type CostResult, type CostSummary, type CostUsage, type CounterfactualContext, type CounterfactualMutation, type CounterfactualResult, type CounterfactualRunner, type CreateChatClientOpts, type CreateTraceAnalystOptions, CrossFamilyError, type CustomTokenPricing, type CustomTransportOpts, DECISION_PAIRED_DELTA_STATISTIC, DEFAULT_FAILURE_REASON_RULES, DEFAULT_PERMUTATIONS, DEFAULT_RED_TEAM_CORPUS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, type DataAcquisitionPlan, type DatasetManifest, type DatasetOverview, type DatasetScenario, type DatasetSplit, type DefaultAnalystRegistryOptions, type DefaultVerdict, type DefineAgentEvalOptions, type DefinedAgentEval, type DeployGateLayerInput, type DeployRunResult, type DeployRunner, type DetectorEvent, type DetectorSignal, type DiffPolicy, type DiffScorecardOptions, type JudgeScore as DimensionJudgeScore, type DirEntry, type DiscoverPersonasOptions, type DiscoveredPersona, type DriverState, type DspyRlmTraceEngineOptions, type EProcess, type EProcessOptions, type EProcessState, type EProcessStep, type EnsembleAggregate, type EnsembleJudgeOptions, type EquivalenceArm, type EquivalenceCheckDefinition, type EquivalenceCheckSpec, type EquivalenceChecker, type EquivalenceCheckerInput, type EquivalenceCheckerResult, type EquivalenceObligation, type EquivalenceObligationStatus, EquivalenceProtocolError, type EquivalenceRecord, type ErrorCluster, type ErrorCountPattern, type ErrorStreakOptions, type EvalCampaignOptions, type EvalCampaignResult, type EventFilter, type EvidenceRef, type ExactAnalystRunEvent, type ExactAnalystRunResult, type ExactCapableAnalyst, type ExactRegistryRunOpts, type ExactRiskDifferenceResult, type ExperimentPlan, type ExperimentRep, type ExperimentResult, type ExperimentStats, type ExtractOptions, type ExtractResult, type ExtractedUsage, FAILURE_BLAME, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, type FailedRun, type FailureBlame, type FailureClass, type FailureClassification, type FailureContext, type FailureMode, type FailureReasonRule, type FailureRule, type FeedbackArtifactType, type FeedbackAttempt, type FeedbackLabel, type FeedbackLabelKind, type FeedbackLabelSource, type FeedbackOptimizerRow, type FeedbackOutcome, type FeedbackSplitPolicy, type FeedbackTask, type FeedbackTrajectory, type FeedbackTrajectoryFilter, type FeedbackTrajectoryStore, type FieldDestination, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, FileSystemTraceStore, type Finding, type FindingSubject, type FindingSubjectKind, type FindingToPolicyEditOptions, type FindingsDiff, FindingsStore, type GainDistributionBin, type GainDistributionFigureSpec, type GainDistributionOptions, type GateDecision, type GateEvidence, type GenericSpan, type GoldenItem, HARNESS_NATIVE_MODEL, type HarnessType, type HeadroomInput, HeldOutGate, type HeldOutGateConfig, type GateDecision$1 as HeldOutGateDecision, type HeldOutGateRejectionCode, type HeldOutPartition, type HiddenCriteriaGrader, type HiddenGradeResult, type HiddenLeak, INFRA_FAILURE_BLAMES, type ImprovementThresholds, type ImprovementVerdictResult, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, InMemoryTraceStore, type InsightReport, type IntentMatchInput, type IntentMatchOptions, type IntentMatchResult, type InterimReleaseConfidence, type InterimReleaseConfidenceInput, JudgeError, type JudgeFamily, type JudgeFn, type JudgeInput, type JudgeRetryOutcome, type JudgeRetryPolicy, type JudgeRubric, type JudgeScore$1 as JudgeScore, type JudgeScoreInput, type JudgeScoresRecord, type JudgeSpan, type JudgeVerdict, type KeywordConceptSpec, type KeywordCoverageFinding, type KeywordCoverageOptions, type KeywordCoverageResult, type KnowledgeAcquisitionMode, type KnowledgeBundle, type KnowledgeFreshness, type KnowledgeImportance, type KnowledgeReadinessReport, type KnowledgeRequirement, type KnowledgeRequirementCategory, type KnowledgeSensitivity, type Layer, type LayerResult, type LayerStatus, type LeaderboardOptions, type LeaderboardRow, LlmCallError, type LlmCallMetadata, type LlmCallRequest, type LlmCallResult, type LlmChargeBounds, type LlmCorrectnessCheckerOpts, type LlmJudgeOptions, type LlmMessage, LlmResponseError, type LlmReviewerConfig, type LlmSpan, type LlmTokenLogprob, type LlmToolCall, type LlmToolChoice, type LlmToolDefinition, type LlmUsage, MANN_WHITNEY_EXACT_MAX_STATES, MANN_WHITNEY_EXACT_MAX_WORK, type MannWhitneyResult, type MatchedPair, type MatchedRunRecordPair, type MaximumCharge, type McNemarResult, type MintRolloutOptions, type MintRolloutResult, type MintedRolloutLine, type MockTransportOpts, type ModelEndpointCheck, type ModelEndpointRequest, type ModelSeats, ModelSubstitutionError, MultiLayerVerifier, type NoLeakOptions, NoopRawProviderSink, NotFoundError, OUTPUT_VALUE, type Objective, type OpenAiCompatibleTransportOpts, type Oracle, type OracleObservation, type OracleReport, type OracleResult, type OrderMode, type OtlpExport, OtlpFileTraceStore, type OtlpFlatLine, type OtlpSpan, type PaidCallResult, type PairArmsOptions, type PairArmsResult, type PairRunRecordsResult, type PairedArmRow, type PairedArmsComparison, type PairedBootstrapOptions, type PairedBootstrapResult, type PairedCorrectness, type PairedDecisionMethod, type PairedDecisionShape, type PairedDecisionStatistic, type PairedDeltaTestOptions, type PairedDeltaTestResult, type PairedEvalueOptions, type PairedEvalueSequence, type PairedEvalueStep, type PairedMcNemarEvidence, type PairedMetricDelta, type PairedPromotionAlternative, type PairedPromotionAlternativeCell, type PairedPromotionCallRates, type PairedPromotionDecision, type PairedPromotionDecisionOptions, type PairedPromotionPowerCall, type PairedPromotionPowerOptions, type PairedPromotionPowerResult, type PairedSignTestResult, type PairedTTestResult, type PairedValueLaw, type ParetoPoint, type ParetoResult, type PartitionHeldOutOptions, type PendingCostCall, type PendingCostCallView, type PersistedFinding, type PersonaConfig, type PolicyEdit, type PolicyEditAdmission, type PolicyEditAdmissionOptions, type PolicyEditAxis, type PolicyEditCandidateRecord, type PolicyEditChange, type PolicyEditExpectedGain, type PolicyEditGainDirection, type PolicyEditGainUnit, type PolicyEditInit, type PolicyEditRisk, type PolicyEditSchemaVersion, type PolicyEditSource, type PolicyEditTarget, type PolicyEditTargetSurface, type PreferenceMemoryEntry, type ProducedProposal, type ProducedState, type ProductBenchmarkExportOptions, type ProductBenchmarkExportResult, type ProductBenchmarkManifest, type ProductBenchmarkSingleRunExportOptions, type ProductBenchmarkSplit, type ProductBenchmarkValidationReport, type ProductClientConfig, type Scenario as ProductScenario, type ProfileAxisSpec, type ProjectRuntimeTrajectoryEvidenceOptions, type PromptHandle, type ProportionInterval, type ProposalEventLike, type ProposalFinding, type ProposeFn, type ProposeInput, type ProposeOutput, type ProposeReviewConfig, type ProposeReviewControlAction, type ProposeReviewControlConfig, type ProposeReviewControlResult, type ProposeReviewControlState, type ProposeReviewReport, type ProposedSideEffect, type QueryTracesPage, REDACTION_VERSION, type RankTestMethod, type RankTestMethodRequest, type RankTestOptions, type RawAnalystFinding, type RawProviderEvent, type RawProviderSink, type RecordRunsOptions, type RedTeamCase, type RedTeamCategory, type RedTeamFinding, type RedTeamReport, type RedactOptions, type RedactionProfile, type RedactionReport, type ReferenceReplayCaseRun, type ReferenceReplayRun, type ReferenceReplaySplit, type ReflectionContext, type ReflectionProposal, type RegistryRunOpts, type ReleaseConfidenceInput, type ReleaseConfidenceIssue, type ReleaseConfidenceMetrics, type ReleaseConfidenceScorecard, type ReleaseTraceEvidence, type RepeatedActionOptions, type RequiredPairsForPairedPromotionOptions, type RequiredPairsForPairedPromotionResult, type RequirementCheck, type ResearchReport, type ResearchReportOptions, type Researcher, type RetryWrite, type Review, type ReviewFn, type ReviewInput, type ReviewMemoryEntry, type ReviewMemoryStore, type RiskDifferenceResult, type RolloutSearchLineage, type RouteMap, type RoutedField, type RubricDimension, type Run, type RunCampaignOptions, type RunCaptureReport, type RunCommandInput, type RunCommandResult, type RunCostProvenance, type RunFilter, RunIntegrityError, type RunIntegrityReport, type RunJudgeMetadata, type RunOutcome, type RunPaidCallInput, type RunRecord, type RunRecordBackend, RunRecordValidationError, type RunScore, type RunScoreWeights, type RunSearchCoordinates, type RunSplitTag, type RunStatus, type RunTaskFailure, type RunTerminalOutcome, type RunTokenUsage, type RunTraceRef, type RuntimeEventLike, type RuntimeTrajectoryEvidenceProjection, type RuntimeTrajectoryEvidenceSummary, type RuntimeTrajectoryHookEvent, type RuntimeTrajectoryRecord, type RuntimeTrajectoryRunRecord, SEMANTIC_CONCEPT_JUDGE_VERSION, type SandboxDriver, type SandboxSdkTransportOpts, type SatisfiedBy, type Scenario$1 as Scenario, type ScenarioCost, type ScenarioFile, type ScenarioResult, type ScorePreference, type ScoreRiskDifferenceResult, type Scorecard, type ScorecardCell, type ScorecardCellDiff, type ScorecardDiff, type ScorecardEntry, type ScorecardLogLine, type SearchSpanResult, type SearchTraceResult, type SelfImproveBudget, type SelfImproveMethodOptions, type SelfImproveMethodProvenance, type SelfImproveMethodResult, type SelfImproveOptions, type SelfImproveProgressEvent, type SelfImproveProposerOptions, type SelfImproveProposerResult, type SelfImproveResult, type SemanticConceptJudgeInput, type SemanticConceptJudgeOptions, type SemanticConceptJudgeResult, type SequentialDecision, type SeriesConvergenceOptions, type SeriesConvergenceResult, type SeriesDistribution, ServedCrossFamilyError, type ServedModelCheck, type ServedModelPolicy, type ServedModelVerdict, type Severity, type ShareSafetyStatus, type ShareSafetyVerdict, type SignTestAlternative, type SingleBackendDivergence, type SingleBackendReport, type Span, type SpanFilter, type SpanHandle, type SpanMatchRecord, type SpanPredicate, type SplitCoverage, type SteeringBundle, type SteeringChange, type SteeringOptimizationResult, type SteeringOptimizationRow, type StopDecision, type StrategyChecker, type StreamingDetector, type SummaryTable, type SummaryTableOptions, type SummaryTableRow, type SynthesisTarget, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, type TaskGold, type TaskHeadroom, type ToolCallEventLike, type ToolSpan, type ToolStats, type ToolUseMetrics, type ToolUseOptions, type TraceAnalysisEngine, type TraceAnalysisEngineResult, type TraceAnalysisStore, type TraceAnalystByteBudgets, type TraceAnalystDefinition, type TraceAnalystFilters, type TraceAnalystSpan, type TraceAnalystSpanKind, type TraceAnalystSpanStatus, type TraceAnalystTraceSummary, type TraceContract, type TraceContractSpec, TraceEmitter, type TraceEmitterOptions, type TraceEvent, type TraceInsightReadiness, type TraceInsightSuite, type TraceStore, type TrainSelectionTestOptions, type TrainSelectionTestPartition, type Trajectory, type TrajectoryStep, type TreatmentGate, type TreatmentGateInput, type TreatmentGateOptions, type TrialTrace, type Turn, type TurnResult, UNKNOWN_MODEL, type UserQuestion, VERIFICATION_STRATEGIES, VERIFICATION_STRATEGY_SOURCES, ValidationError, type VerbosityBiasResult, type Verdict, type VerdictCacheStore, type VerdictCertification, type Verification, type VerificationReport, type VerificationStrategyProfile, type VerificationStrategySource, type VerifyFn, type VerifyOptions, type ViewSpansResult, type ViewTraceOversized, type ViewTraceResult, type ViteDeployRunnerInput, WILCOXON_EXACT_MAX_N, type WeightedCompositeInput, type WeightedCompositeResult, type WilcoxonSignedRankResult, type WranglerDeployRunnerInput, agentProfileCellHashMaterial, agentProfileCellKey, agentProfileHash, agentProfileId, analystFindingDigest, analystRunDigest, analystRunToFeedbackTrajectory, analystRunToReviewRequests, analyzeRuns, analyzeSeries, analyzeTraces, argHash, assertCrossFamily, assertCrossFamilyServed, assertRealBackend, assertRunCaptured, assertServedModel, assertServedModels, assessShareSafety, assignFeedbackSplit, attest, benjaminiHochberg, bonferroni, bootstrapCi, budgetBreachView, buildAgentProfileCell, buildDefaultAnalystRegistry, buildEquivalenceRecord, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, calibrateJudge, calibrateJudgeContinuous, captureFetchToRawSink, certificationEvidenceDigest, checkServedModel, classifyFailure, classifyFailureReason, comparePairedArms, completionVerdict, computeFindingId, computeToolUseMetrics, confidenceInterval, continuousAgreement, controlRunToFeedbackTrajectory, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, costForTokenPricing, costForUsage, costReceiptFromLlm, costReceiptFromLlmError, createBoundedTraceAnalysisStore, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createFeedbackTrajectory, createLlmCorrectnessChecker, createTraceAnalyst, decidePairedPromotion, defineAgentEval, defineEquivalenceCheck, describeTraceInsightScope, diffFindings, domainEvidencePattern, dominates, eProcess, equivalenceVerdict, evaluateInterimReleaseConfidence, evaluateReleaseConfidence, expandProfileAxes, exportRunAsOtlp, extractProducedState, extractUsage, extractUsageFromSse, failureBlame, failureClusterView, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToOptimizerRow, gainHistogram, gradeSemanticStatus, groupRunsByAgentProfileCell, harnessAxisOf, hashJson, holm, inferDomainKeywords, interRaterReliability, iqr, isBinaryOutcomeVector, isJudgeSpan, isLlmSpan, isRunRecord, isToolSpan, isTransientLlmError, judgeAgreementView, judgeFamily, judgeSpans, llmJudge, makeFinding, makeProposalFinding, manifestContentDigest, mannWhitneyU, maximumChargeForLlmRequest, mcnemar, mcnemarPower, mcnemarRequiredN, mintRolloutRows, modelHasSnapshot, modelPriceKey, mulberry32, objectiveEval, otlpTextToTraceAnalysisStore, pairArms, pairRunRecords, pairedBinaryScale, pairedBootstrap, pairedDeltaTieFraction, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedRiskDifferenceExact, pairedRiskDifferenceScore, pairedSignTest, pairedTTest, paretoChart, paretoFrontier, parseRunRecordSafe, partialCredit, passAtK, pearsonR, index_d_exports as profile, quotaExhaustedUntil, ranks, redTeamDataset, redTeamReport, redact, redactForShare, redactText, renderPreferenceMemoryMarkdown, requiredPairedSampleSize, requiredSampleSize, roundTripRunRecord, runAgentControlLoop, runCampaign, runCanaries, runCostFloorUsd, runCounterfactual, runEquivalenceCheck, runEvalCampaign, runSemanticConceptJudge, runTaskScore, runsForScenario, scoreRedTeamOutput, scoreTraceInsightReadiness, searchCellRunId, selfImprove, servedModelAcceptable, shareAllowed, spearmanR, stripFencedJson, subjectiveEval, summarizeBackendIntegrity, summarizeNumberSeries, summarizePreferenceMemory, summaryTable, toAgentProfileJson, tokenizeDomainWords, toolSpansToTraceAnalysisStore, toolWasteView, transientDispatchFailure, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyAttestation, verifyCompletion, weightedComposite, weightedMean, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit };
 //# sourceMappingURL=index.d.ts.map