@tangle-network/agent-eval 0.133.2 → 0.133.3

package/dist/index.js

@@ -11,14 +11,14 @@ import { S as traceSpanKindToOpenInferenceKind, a as SpanNotFoundError, b as cla
 import { n as aggregateRunScore, r as clamp01, t as DEFAULT_RUN_SCORE_WEIGHTS } from "./run-score-iEEAWiBY.js";
 import { n as mapConcurrent, t as Mutex } from "./concurrency-MUjT7VjM.js";
 import { a as RunCritic, d as defaultIsMaterial, f as diffFindings, i as runSemanticConceptJudge, n as SEMANTIC_CONCEPT_JUDGE_VERSION, o as SKILL_USAGE_ANALYST, p as LockedJsonlAppender, r as createSemanticConceptJudge, s as SkillUsageAnalyst, t as DEFAULT_COMPLEXITY_WEIGHTS, u as FindingsStore } from "./semantic-concept-judge-BypLt6Fw.js";
-import { At as dominates, B as surfaceContentHash, Bt as JudgeParseError, Ct as llmJudge, D as runCanaries, Dt as fileVerdictCache, Et as contentHash, Ft as BackendIntegrityError, G as DEFAULT_MUTATION_PRIMITIVES, It as assertRealAgentReceipts, K as buildReflectionPrompt, Lt as assertRealBackend, Mt as paretoFrontierWithCrowding, Nt as scalarScore, Ot as inMemoryVerdictCache, Rt as summarizeAgentReceiptIntegrity, St as hashScenarios, Tt as canonicalJson, _t as redTeamReport, bt as Dataset, dt as REFERENCE_EQUIVALENCE_INPUT_LIMITS, ft as REFERENCE_EQUIVALENCE_JUDGE_VERSION, gt as redTeamDataset, ht as DEFAULT_RED_TEAM_CORPUS, jt as paretoFrontier, kt as crowdingDistance, mt as runReferenceEquivalenceJudge, pt as createReferenceEquivalenceJudge, q as parseReflectionResponse, vt as scoreRedTeamOutput, wt as cachedJudge, xt as HoldoutLockedError, yt as toolNamesForRun, zt as summarizeBackendIntegrity } from "./skillopt-optimization-method-CF6a327Q.js";
-import { A as spearmanR, B as positionalBias, C as pairedTTest, D as ranks, E as pearsonR, H as verbosityBias, L as calibrateJudge, M as weightedMean, N as wilcoxonSignedRank, O as requiredPairedSampleSize, P as wilson, R as calibrateJudgeContinuous, S as pairedSignTest, T as passAtK, V as selfPreference, _ as normalizeScores, a as confidenceInterval, b as pairedMde, c as eProcess, d as interpretCliffs, f as mannWhitneyU, g as mulberry32, h as mcnemarRequiredN, i as cohensD, j as weightedComposite, k as requiredSampleSize, l as holm, m as mcnemarPower, n as bonferroni, o as corpusInterRaterAgreement, p as mcnemar, r as cliffsDelta, s as corpusInterRaterAgreementFromJudgeScores, t as benjaminiHochberg, u as interRaterReliability, v as pairedBootstrap, w as partialCredit, x as pairedRiskDifference, y as pairedCohensDz, z as continuousAgreement } from "./statistics-DWM_AyLe.js";
-import { n as pairArms, r as pairRunRecords, t as comparePairedArms } from "./paired-arms-6XItKzd1.js";
+import { At as dominates, B as surfaceContentHash, Bt as summarizeAgentReceiptIntegrity, Ct as llmJudge, D as runCanaries, Dt as fileVerdictCache, Et as contentHash, Ft as minimumPairsForPairedDeltaTest, G as DEFAULT_MUTATION_PRIMITIVES, Ht as JudgeParseError, It as pairedDeltaTest, K as buildReflectionPrompt, Lt as BackendIntegrityError, Mt as paretoFrontierWithCrowding, Nt as scalarScore, Ot as inMemoryVerdictCache, Rt as assertRealAgentReceipts, St as hashScenarios, Tt as canonicalJson, Vt as summarizeBackendIntegrity, _t as redTeamReport, bt as Dataset, dt as REFERENCE_EQUIVALENCE_INPUT_LIMITS, ft as REFERENCE_EQUIVALENCE_JUDGE_VERSION, gt as redTeamDataset, ht as DEFAULT_RED_TEAM_CORPUS, jt as paretoFrontier, kt as crowdingDistance, mt as runReferenceEquivalenceJudge, pt as createReferenceEquivalenceJudge, q as parseReflectionResponse, vt as scoreRedTeamOutput, wt as cachedJudge, xt as HoldoutLockedError, yt as toolNamesForRun, zt as assertRealBackend } from "./skillopt-optimization-method-vvJ4bMNI.js";
+import { A as passAtK, B as studentTCdf, C as pairedBootstrap, D as pairedSignTest, E as pairedRiskDifference, F as spearmanR, G as continuousAgreement, H as normalCdf, I as weightedComposite, J as verbosityBias, K as positionalBias, L as weightedMean, M as ranks, N as requiredPairedSampleSize, O as pairedTTest, P as requiredSampleSize, R as wilcoxonSignedRank, S as normalizeScores, T as pairedMde, U as calibrateJudge, W as calibrateJudgeContinuous, _ as mannWhitneyU, a as WILCOXON_EXACT_MAX_N, b as mcnemarRequiredN, c as cliffsDelta, d as corpusInterRaterAgreement, f as corpusInterRaterAgreementFromJudgeScores, g as interpretCliffs, h as interRaterReliability, i as MANN_WHITNEY_EXACT_MAX_WORK, j as pearsonR, k as partialCredit, l as cohensD, m as holm, n as DEFAULT_PERMUTATIONS, o as benjaminiHochberg, p as eProcess, q as selfPreference, r as MANN_WHITNEY_EXACT_MAX_STATES, s as bonferroni, t as BOOTSTRAP_GATE_MIN_N, u as confidenceInterval, v as mcnemar, w as pairedCohensDz, x as mulberry32, y as mcnemarPower, z as wilson } from "./statistics-RwRNu2__.js";
+import { n as pairArms, r as pairRunRecords, t as comparePairedArms } from "./paired-arms-CA_8pN01.js";
 import { n as llmSpanFromProvider, t as TraceEmitter } from "./emitter-CPBAhxum.js";
 import { a as scoreOrigin, n as observedScore, o as trainingReward, r as observedSplitScore, s as trainingScore, t as isRealnessGated } from "./reward-nw2xZGZG.js";
 import { a as isRetrievalSpan, i as isLlmSpan, n as TRACE_SCHEMA_VERSION, o as isSandboxSpan, r as isJudgeSpan, s as isToolSpan, t as FAILURE_CLASSES } from "./schema-CRhEY1SO.js";
 import { a as roundTripRunRecord, i as parseRunRecordSafe, n as isRunRecord, o as runTaskScore, r as modelHasSnapshot, s as validateRunRecord, t as RunRecordValidationError } from "./run-record-BIwU2wdV.js";
-import { a as evaluateReleaseConfidence, i as assertReleaseConfidence, n as bootstrapCi, r as judgeReplayGate, t as renderReleaseReport } from "./release-report-wuilQkvK.js";
+import { a as evaluateReleaseConfidence, i as assertReleaseConfidence, n as bootstrapCi, r as judgeReplayGate, t as renderReleaseReport } from "./release-report-BVZBmRZp.js";
 import { c as assertMintedLines, f as isRolloutLine, i as ROLLOUT_SCHEMA, m as validateRolloutLine, p as isTrainableSplit, s as assertMinted, u as assertRolloutLine } from "./schema-C6DW4ZHR.js";
 import { i as toRewardRows, r as toJsonl, s as toSftRows } from "./exporters-q9iL-2Jf.js";
 import { a as toHarborTrajectories, i as relabelImportedSplit, n as HARBOR_IMPORT_GAP, o as toHarborTrajectory, r as fromHarborTrajectory, t as ATIF_SCHEMA_VERSION } from "./rollout-CreDz__7.js";
@@ -28,17 +28,18 @@ import { D as showMeasured, E as isUnavailable, S as rollupSupervisorRuns, T as
 import { n as TRACE_ANALYST_ACTOR_DESCRIPTION, r as TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, t as analyzeTraces } from "./analyst-LsnNpSkm.js";
 import { C as planTraceInsightQuestions, E as traceAnalystOnRunComplete, S as inferDomainKeywords, T as tokenizeDomainWords, _ as buildTraceInsightContext, a as convertTraceStoresToOtlp, b as describeTraceInsightScope, c as otelRunCompleteHook, d as captureFetchToRawSink, f as otlpRowsToRunRecords, g as flattenOtlpExportToNdjson, h as otlpToTraceRunRecords, i as iterateRawCalls, l as OTEL_AGENT_EVAL_SCOPE, m as otlpToRunRecords, n as ReplayCacheMissError, o as createOtelExporter, p as otlpRowsToTraceRunRecords, r as createReplayFetch, s as createOtelTracingStore, t as ReplayCache, u as exportRunAsOtlp, v as buildTraceInsightPrompt, w as scoreTraceInsightReadiness, x as domainEvidencePattern, y as defaultTraceInsightPanel } from "./replay-CJfGLdx4.js";
 import { n as extractUsageFromResponse, r as extractUsageFromSse, t as extractUsage } from "./extract-usage-2j25whHw.js";
-import { B as harnessAxisOf, F as HARNESS_NATIVE_MODEL, G as parseCorrectnessResponse, H as completionVerdict, I as agentProfileHash, K as verifyCompletion, L as agentProfileId, P as CODING_HARNESSES, R as agentProfileModelId, U as createLlmCorrectnessChecker, V as extractProducedState, W as createTokenRecallChecker, z as expandProfileAxes } from "./campaign-BJjn1rhw.js";
+import { B as harnessAxisOf, F as HARNESS_NATIVE_MODEL, G as parseCorrectnessResponse, H as completionVerdict, I as agentProfileHash, K as verifyCompletion, L as agentProfileId, P as CODING_HARNESSES, R as agentProfileModelId, U as createLlmCorrectnessChecker, V as extractProducedState, W as createTokenRecallChecker, z as expandProfileAxes } from "./campaign--V4ffEKR.js";
+import { n as iqr, r as welchsTTest, t as compareToBaseline } from "./baseline-BaPxoROc.js";
 import { a as judgeSpans, c as runsForScenario, i as hasCapturedToolArgs, l as toolSpans, n as argHash, o as llmSpans, r as groupBy, s as runFailureClass, t as aggregateLlm } from "./query-Di7eEQ79.js";
-import { a as checkBehavioralCanary, i as canaryLeakView, o as checkCanaries, r as HoldoutAuditor, s as runBehavioralCanaries, t as analyzeRuns } from "./analyze-runs-B-afTpCv.js";
-import { a as summaryTable, i as researchReport, n as gainHistogram, r as paretoChart, t as RESEARCH_REPORT_HARD_PAIR_FLOOR } from "./summary-report-Ci17nIdU.js";
+import { a as checkBehavioralCanary, i as canaryLeakView, o as checkCanaries, r as HoldoutAuditor, s as runBehavioralCanaries, t as analyzeRuns } from "./analyze-runs-qk8op0tN.js";
+import { a as summaryTable, i as researchReport, n as gainHistogram, r as paretoChart, t as RESEARCH_REPORT_HARD_PAIR_FLOOR } from "./summary-report-BxtossFi.js";
 import { a as composeParsers, c as vitestTestParser, i as SubprocessSandboxDriver, n as DockerSandboxDriver, o as jestTestParser, r as SandboxHarness, s as pytestTestParser, t as runTestGradedScenario } from "./test-graded-scenario-BsqWLmPt.js";
 import { a as InMemoryTraceStore, i as FileSystemTraceStore, n as assertRunCaptured, r as throwIfRunIncomplete, t as RunIntegrityError } from "./integrity-BzRbCHzi.js";
 import { i as redactValue, n as REDACTION_VERSION, r as redactString, t as DEFAULT_REDACTION_RULES } from "./redact-7Aq1ukl-.js";
-import { a as DEFAULT_RULES, i as computeToolUseMetrics, n as iqr, o as classifyFailure, r as welchsTTest, t as compareToBaseline } from "./baseline-DcX5hQDv.js";
+import { n as DEFAULT_RULES, r as classifyFailure, t as computeToolUseMetrics } from "./tool-use-metrics-DEGMKycK.js";
 import { t as analyzeSeries } from "./series-convergence-CjO2QdRW.js";
-import { _ as deterministicSplit, g as BENCHMARK_SPLIT_SEED, t as benchmarks_exports } from "./benchmarks-CJr1H1_a.js";
-import { t as runEvalCampaign } from "./eval-campaign-DXhpZghy.js";
+import { _ as deterministicSplit, g as BENCHMARK_SPLIT_SEED, t as benchmarks_exports } from "./benchmarks-BP9sgMia.js";
+import { t as runEvalCampaign } from "./eval-campaign-CvPcvqXC.js";
 import { n as pairedEvalueSequence, t as evaluateInterimReleaseConfidence } from "./sequential-Br0mAPHA.js";
 import { AxGEPA, AxSignature, ax } from "@ax-llm/ax";
 import { accessSync, appendFileSync, constants, cpSync, existsSync, mkdirSync, promises, readFileSync, readdirSync, statSync, writeFileSync } from "node:fs";
@@ -6004,7 +6005,7 @@ function diffScorecard(scorecard, opts = {}) {
 			d = cohensD(baseline.scores, current.scores);
 			const t = welchsTTest(baseline.scores, current.scores);
 			p = Number.isFinite(t.p) ? t.p : null;
-			verdict = Math.abs(d) >= minEffect && p !== null && p <= maxP ? delta > 0 ? "improved" : "regressed" : "flat";
+			verdict = (d === null || Math.abs(d) >= minEffect) && p !== null && p <= maxP ? delta > 0 ? "improved" : "regressed" : "flat";
 		} else verdict = Math.abs(delta) >= minDelta ? delta > 0 ? "improved" : "regressed" : "flat";
 		cells.push({
 			...base,
@@ -6816,7 +6817,7 @@ async function evaluateContract(store, contract) {
 		if (!decl) continue;
 		if (metric.verdict === "regressed") {
 			const magnitude = Math.abs(metric.delta);
-			if (decl.maxRegression === void 0 || magnitude > decl.maxRegression) breaches.push(`metric "${metric.metric}" regressed by ${metric.delta.toFixed(4)} (d=${metric.cohensD.toFixed(2)}, p=${metric.welchP.toExponential(2)})`);
+			if (decl.maxRegression === void 0 || magnitude > decl.maxRegression) breaches.push(`metric "${metric.metric}" regressed by ${metric.delta.toFixed(4)} (d=${formatEffectSize(metric.cohensD)}, p=${metric.welchP.toExponential(2)})`);
 		}
 	}
 	if (sloReport) for (const r of sloReport.criticalBreaches) breaches.push(`SLO "${r.slo.id}" breached: ${r.detail}`);
@@ -6844,7 +6845,7 @@ function renderMarkdownReport(reports) {
 			lines.push("");
 			lines.push("| metric | baseline | candidate | Δ | Cohen d | p | verdict |");
 			lines.push("|---|---|---|---|---|---|---|");
-			for (const m of r.baselineReport.metrics) lines.push(`| ${m.metric} | ${m.baselineMean.toFixed(4)} | ${m.candidateMean.toFixed(4)} | ${m.delta.toFixed(4)} | ${m.cohensD.toFixed(2)} | ${m.welchP.toExponential(2)} | ${m.verdict} |`);
+			for (const m of r.baselineReport.metrics) lines.push(`| ${m.metric} | ${m.baselineMean.toFixed(4)} | ${m.candidateMean.toFixed(4)} | ${m.delta.toFixed(4)} | ${formatEffectSize(m.cohensD)} | ${m.welchP.toExponential(2)} | ${m.verdict} |`);
 		}
 		if (r.sloReport && r.sloReport.results.length > 0) {
 			lines.push("");
@@ -6882,6 +6883,11 @@ function average(xs) {
 	if (xs.length === 0) return 0;
 	return xs.reduce((a, b) => a + b, 0) / xs.length;
 }
+/** Cohen's d is null when both samples are constant with unequal means — an
+*  unbounded effect, which must not be rendered as a number. */
+function formatEffectSize(d) {
+	return d === null ? "unbounded (zero pooled spread)" : d.toFixed(2);
+}
 async function extractAll(runs, extract, store) {
 	const out = [];
 	for (const r of runs) {
@@ -10124,11 +10130,11 @@ var HeldOutGate = class {
 	costPerTaskCeiling;
 	constructor(config) {
 		if (!config.baselineKey) throw new Error("HeldOutGate: baselineKey is required");
-		this.minProductiveRuns = config.minProductiveRuns ?? 3;
+		this.confidence = config.confidence ?? .95;
+		this.minProductiveRuns = Math.max(config.minProductiveRuns ?? 3, minimumPairsForPairedDeltaTest(this.confidence));
 		this.pairedDeltaThreshold = config.pairedDeltaThreshold ?? 0;
 		this.overfitGapThreshold = config.overfitGapThreshold ?? .15;
 		this.baselineKey = config.baselineKey;
-		this.confidence = config.confidence ?? .95;
 		this.resamples = config.bootstrapResamples ?? 2e3;
 		this.seed = config.seed;
 		if (config.costPerTaskCeiling !== void 0 && !(Number.isFinite(config.costPerTaskCeiling) && config.costPerTaskCeiling > 0)) throw new Error("HeldOutGate: costPerTaskCeiling must be a positive finite number");
@@ -10209,12 +10215,15 @@ var HeldOutGate = class {
 			rejectionCode: "few_runs"
 		};
 		if (overfitGap === null || baselineOverfitGap === null) throw new Error("HeldOutGate: complete split scores did not produce overfit gaps");
-		const ci = pairedBootstrap(beforeHoldout, afterHoldout, {
+		const deltaTest = pairedDeltaTest(beforeHoldout, afterHoldout, {
 			confidence: this.confidence,
 			resamples: this.resamples,
 			statistic: "median",
-			seed: this.seed
+			seed: this.seed,
+			threshold: this.pairedDeltaThreshold,
+			minPairs: this.minProductiveRuns
 		});
+		const ci = deltaTest.bootstrap;
 		const wilcoxon = wilcoxonSignedRank(beforeHoldout, afterHoldout);
 		const evidence = {
 			...commonEvidence,
@@ -10233,12 +10242,12 @@ var HeldOutGate = class {
 			reason: `indeterminate_delta: paired holdout median CI=[${fmt(ci.low)}, ${fmt(ci.high)}] carries no direction`,
 			rejectionCode: "indeterminate_delta"
 		};
-		if (!(ci.low > this.pairedDeltaThreshold)) return {
+		if (!deltaTest.significant) return {
 			promote: false,
 			candidateId,
 			baselineId,
 			evidence,
-			reason: `negative_delta: paired holdout median Δ=${fmt(ci.median)} CI=[${fmt(ci.low)}, ${fmt(ci.high)}] does not clear threshold ${fmt(this.pairedDeltaThreshold)}`,
+			reason: `negative_delta: paired holdout median Δ=${fmt(ci.median)} ${deltaTest.method === "bootstrap-ci" ? `CI=[${fmt(ci.low)}, ${fmt(ci.high)}]` : `exact one-sided p=${fmt(deltaTest.pValue ?? 1)}`} does not clear threshold ${fmt(this.pairedDeltaThreshold)}`,
 			rejectionCode: "negative_delta"
 		};
 		if (overfitGap > baselineOverfitGap + this.overfitGapThreshold) return {
@@ -12114,6 +12123,6 @@ function assertProductBenchmarkRun(runDir) {
 	return report;
 }
 //#endregion
-export { AGENT_PROFILE_KINDS, ATIF_SCHEMA_VERSION, ATTESTATION_ALGORITHM, AgentDriver, AgentEvalError, AgentProfileCellValidationError, AnalystRegistry, AxGepaSteeringOptimizer, BENCHMARK_SPLIT_SEED, BackendIntegrityError, BenchmarkRunner, BudgetBreachError, BudgetGuard, CODING_HARNESSES, CallbackResearcher, CaptureIntegrityError, ConfigError, ConvergenceTracker, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, CostLedger, CostLedgerPersistenceError, CostReceiptCaptureError, CostReservationExceededError, CostTracker, CrossFamilyError, DEFAULT_AGENT_SLOS, DEFAULT_COMPLEXITY_WEIGHTS, DEFAULT_RULES as DEFAULT_FAILURE_RULES, DEFAULT_FINDERS, DEFAULT_HARNESS_OBJECTIVES, DEFAULT_MUTATION_PRIMITIVES, DEFAULT_MUTATORS, DEFAULT_PR_REVIEW_SCORE_WEIGHTS, DEFAULT_REDACTION_RULES, DEFAULT_RED_TEAM_CORPUS, DEFAULT_RUN_SCORE_WEIGHTS, DEFAULT_SEVERITY_WEIGHTS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, Dataset, DescriptionLengthGate, DockerSandboxDriver, DualAgentBench, ERROR_COUNT_PATTERNS, EvalTraceStore, ExperimentTracker, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, FileSystemTraceStore, FindingsStore, HARBOR_IMPORT_GAP, HARNESS_NATIVE_MODEL, HeldOutGate, HoldoutAuditor, HoldoutLockedError, IMPROVEMENT_KIND_SPEC, INPUT_VALUE, INTENT_MATCH_JUDGE_VERSION, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, InMemoryTraceStore, InMemoryWorkspaceInspector, JudgeError, JudgeParseError, JudgeRunner, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_CACHE_WRITE_TOKENS, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, LLM_CONTEXT_TOKENS, LLM_COST_ATTR_KEYS, LLM_COST_USD, LLM_INPUT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_MODEL_ATTR_KEYS, LLM_MODEL_NAME, LLM_OUTPUT_TOKENS, LLM_OUTPUT_TOKEN_ATTR_KEYS, LLM_REASONING_TOKENS, LLM_REASONING_TOKEN_ATTR_KEYS, LlmCallError, LlmClient, LlmResponseError, LlmRouteAssertionError, LockedJsonlAppender, MODEL_PRICING, MetricsCollector, ModelsUnreachableError, MultiLayerVerifier, Mutex, NoopRawProviderSink, NoopResearcher, NotFoundError, OPENINFERENCE_SPAN_KIND, OTEL_AGENT_EVAL_SCOPE, OUTPUT_VALUE, OtlpFileTraceStore, PairwiseSteeringOptimizer, ProductClient, PromptRegistry, REDACTION_VERSION, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, RESEARCH_REPORT_HARD_PAIR_FLOOR, ROLLOUT_SCHEMA, RUN_COST_ATTR_KEYS, ReplayCache, ReplayCacheMissError, ReplayError, RunCritic, RunIntegrityError, RunRecordValidationError, SEMANTIC_CONCEPT_JUDGE_VERSION, SKILL_USAGE_ANALYST, SPAN_KIND_ATTR_KEYS, SUPERVISOR_RUN_SCHEMA, SandboxHarness, ScenarioRegistry, SeatUnsetError, SingleBackendError, SkillUsageAnalyst, SpanNotFoundError, SubprocessSandboxDriver, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TRACE_SCHEMA_VERSION, TokenCounter, TraceContractBuilder, TraceEmitter, TraceFileMissingError, TraceNotFoundError, UI_FINDING_SEVERITIES, UI_LENSES, UNIVERSAL_FINDERS, ValidationError, VerificationError, acquisitionPlansForKnowledgeGaps, agentProfileCellHashMaterial, agentProfileCellKey, agentProfileHash, agentProfileId, agentProfileModelId, agentVisibleFields, aggregateJudgeVerdicts, aggregateLlm, aggregatePrReviewScore, aggregateRunScore, allCriticalPassed, analyzeAntiSlop, analyzeSeries, analyzeSupervisorRun, analyzeSupervisorRunSources, analyzeTraces, appendScorecard, applyLlmSpanOtlpAttributes, applyToolSpanOtlpAttributes, argHash, asNumber, asString, assertCapabilityHeadroom, assertCrossFamily, assertLlmRoute, assertMinted, assertMintedLines, assertModelsServed, assertNoHiddenLeak, assertProductBenchmarkRun, assertRealAgentReceipts, assertRealBackend, assertReleaseConfidence, assertRolloutLine, assertRunAgentProfileCell, assertRunCaptured, assertSingleBackend, assignFeedbackSplit, assignHeldOutTag, attachCostToReport, attest, attributeCounterfactuals, backoffMs, deterministicSplit as benchmarkDeterministicSplit, benchmarks_exports as benchmarks, benjaminiHochberg, bisect, blendHeldout, blockingKnowledgeEval, bonferroni, bootstrapCi, buildAgentInterfaceProfileCell, buildAgentProfileCell, buildDefaultAnalystRegistry, buildDriverSystemPrompt, buildProductBenchmarkManifest, buildReflectionPrompt, buildReviewerPrompt, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, buildWorkerDriverSystemPrompt, byteLengthRange, cachedJudge, calibrateJudge, calibrateJudgeContinuous, callLlm, callLlmJson, canaryLeakView, canonicalJson, canonicalize, capabilityHeadroom, captureFetchToRawSink, causalAttribution, checkBehavioralCanary, checkCanaries, checkSlos, checkTraceContracts, clamp01, classifyFailure, classifyOtlpSpanRole, classifyTreatment, claudeCodeSupervisorRunReader, cliffsDelta, clusteredPairedBinary, cohensD, collectionPreserved, commentsForSource, commitBisect, comparePairedArms, compareReferenceReplay, compareToBaseline, compilerJudge, completionVerdict, composeParsers, composeValidators, computeExperimentStats, computeFindingId, computeToolUseMetrics, computeTraceMetrics, confidenceInterval, containsAll, contentHash, contextInputTokens, continuousAgreement, contractJudge, controlFailureClassFromVerification, controlRunToFeedbackTrajectory, controlRunToRunRecord, convertTraceStoresToOtlp, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, costForTokenPricing, costForUsage, costReceiptFromLlm, costReceiptFromLlmError, costReport, createAnalystAi, createAntiSlopJudge, createChatClient, createDefaultReviewer, createFeedbackTrajectory, createIntentMatchJudge, createLlmCorrectnessChecker, createLlmReviewer, createOtelExporter, createOtelTracingStore, createReferenceEquivalenceJudge, createReplayFetch, createSandboxPool, createSemanticConceptJudge, createTokenRecallChecker, createTraceAnalystKind, crossTraceDiff, crowdingDistance, dataDescriptionBits, decideNextUserTurn, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultBlendWeights, defaultIsMaterial, defaultProviderRedactor, defaultReferenceReplayMatcher, defaultTraceInsightPanel, deployGateLayer, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, distillPlaybook, domainEvidencePattern, dominates, eProcess, ensembleJudge, errorStreakDetector, estimateCost, estimateTokens, evaluateActionPolicy, evaluateContract, evaluateHypothesis, evaluateInterimReleaseConfidence, evaluateOracles, evaluateReleaseConfidence, evaluateTraceContract, executeScenario, expandProfileAxes, expectAgent, exportProductBenchmark, exportProductBenchmarkRuns, exportRewardModel, exportRunAsOtlp, extractAssetUrls, extractErrorCount, extractOtlpAttributes, extractProducedState, extractUsage, extractUsageFromResponse, extractUsageFromSse, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToDatasetScenario, feedbackTrajectoryToOptimizerRow, fileContains, fileExists, fileExperimentStore, fileVerdictCache, findAutoMatchNoExpectation, findConstructorCwdDropped, findFallbackToPass, findLiteralTruePass, findProductBenchmarkArtifacts, findSkipCountsAsPass, firstNumberAttr, firstStringAttr, flattenOtlpExportToNdjson, flowLayer, fnv1a32, formatBenchmarkReport, formatDriverReport, formatFindings, formatScorecardDiff, fromHarborTrajectory, gainHistogram, gateTreatmentApplied, gateTreatmentFromMetrics, gateTreatmentFromSpans, gateTreatmentFromToolSpans, ghCliClient, gitProvenanceReader, precision as goldenPrecision, gradeOnHidden, gradeSemanticStatus, groupBy, groupRunsByAgentProfileCell, harnessAxisOf, hasCapturedToolArgs, hashContent, hashJson, hashScenarios, hashToUnit, hiddenGrade, holm, htmlContainsElement, httpGithubClient, improvementVerdict, inMemoryExperimentStore, inMemoryReferenceReplayStore, inMemoryReviewStore, inMemoryRunRecordBackend, inMemoryVerdictCache, inferDomainKeywords, inferOtlpKind, interRaterReliability, interpretCliffs, iqr, isHiddenDestination, isJudgeSpan, isLlmSpan, isModelPriced, isOtelConfigured, isOtlpModelCall, isRealnessGated, isRetrievalSpan, isRolloutLine, isRunRecord, isSandboxSpan, isToolSpan, isTrainableSplit, isTransientLlmError, isUnavailable, iterateRawCalls, jestTestParser, jsonHasKeys, jsonShape, jsonlReferenceReplayStore, jsonlReviewStore, jsonlRunRecordBackend, judgeFamily, judgeReplayGate, judgeSpans, keyPreserved, knowledgeReadinessTracePayload, leaderboard, linterJudge, llmJudge, llmSpanFromProvider, llmSpans, loadScorecard, loadScorerFromGrader, localCommandRunner, lowercaseMutator, makeEvalTools, makeFinding, mannWhitneyU, mapConcurrent, matchGoldens, matchSpan, maximumChargeForLlmRequest, mcnemar, mcnemarPower, mcnemarRequiredN, mergeLayerResults, mergeSteeringBundle, mintRolloutRows, modelDescriptionBits, modelHasSnapshot, modelPriceKey, mulberry32, multiToolchainLayer, noProgressDetector, normalizeScores, notBlocked, objectiveEval, observeAll, observedScore, observedSplitScore, otelRunCompleteHook, otlpRowsToRunRecords, otlpRowsToTraceRunRecords, otlpToRunRecords, otlpToTraceRunRecords, pairArms, pairRunRecords, pairedBootstrap, pairedCohensDz, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedSignTest, pairedTTest, paraphraseRobustness, paraphraseRobustnessScenarios, paretoChart, paretoFrontier, paretoFrontierWithCrowding, parseCorrectnessResponse, parseFeedbackTrajectoriesJsonl, parseReflectionResponse, parseRunRecordSafe, parseRuntimeTrajectoryHookEvent, partialCredit, partitionHeldOut, passAtK, passOrthogonality, pearsonR, pixelDeltaRatio, planTraceInsightQuestions, politenessPrefixMutator, positionalBias, preflightModels, printDriverSummary, probeLlm, productBenchmarkIntegrityFailures, productBenchmarkMutableSurfaces, productBenchmarkRepoIdentity, productBenchmarkSplits, profile_exports as profile, projectOtlpFlatLine, projectRuntimeTrajectoryEvidence, promptBisect, proposeSynthesisTargets, providerFromBaseUrl, pytestTestParser, ranks, readClaudeCodeSupervisorRun, readOtlpStatus, readProductBenchmarkManifest, readProductBenchmarkRecords, recordRuns, recordRunsToScorecard, redTeamDataset, redTeamReport, redactString, redactValue, referenceReplayRunsToSteeringRows, referenceReplayScenarioToRunScore, regexMatch, regexMatches, relabelImportedSplit, renderMarkdownReport, renderPlaybookMarkdown, renderPreferenceMemoryMarkdown, renderPriorFindings, renderReleaseReport, renderSteeringText, renderSupervisorRunHeadline, renderSupervisorRunMarkdown, renderUpstreamFindings, repeatedActionDetector, replayFeedbackTrajectories, replayFeedbackTrajectory, replayScorerOverCorpus, replayTraceThroughJudge, requireAgentProfileCell, requiredPairedSampleSize, requiredSampleSize, researchReport, resolveModelPricing, resolveSeat, rollupSupervisorRuns, roundTripRunRecord, routeFields, rowCount, rowWhere, runAgentControlLoop, runAssertions, runBehavioralCanaries, runCanaries, runCounterfactual, runE2EWorkflow, runEvalCampaign, runExpectations, runFailureClass, runHarnessExperiment, runIntentMatchJudge, runJudgeFleet, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runLiveProof, runProposeReview, runProposeReviewAsControlLoop, runRecordToProductBenchmarkRecord, runReferenceEquivalenceJudge, runReferenceReplay, runScore, runSelfPlay, runSemanticConceptJudge, runTaskScore, runTestGradedScenario, runsForScenario, scalarScore, scanForMuffledGates, scoreContinuity, scoreFromEvals, scoreKnowledgeReadiness, scoreOrigin, scorePrReviewComments, scorePrReviewSource, scoreRedTeamOutput, scoreReferenceReplay, scoreTraceInsightReadiness, seatPresets, securityJudge, selectHarnessVariant, selfPreference, sentenceReorderMutator, serializeFeedbackTrajectoriesJsonl, showMeasured, signManifest, spearmanR, statusAdvanced, stopOnNoProgress, stopOnRepeatedAction, stringField, stripFencedJson, subjectiveEval, summarizeAgentReceiptIntegrity, summarizeBackendIntegrity, summarizeHarnessResults, summarizePrReviewBenchmark, summarizePreferenceMemory, summaryTable, supervisorRunRolloutLines, testJudge, textInSnapshot, throwIfRunIncomplete, toAgentProfileJson, toHarborTrajectories, toHarborTrajectory, toJsonl, toLangfuseEnvelope, toOpenAiTool, toPrometheusText, toRewardRows, toSftRows, tokenizeDomainWords, toolNamesForRun, toolSpans, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceContract, traceJudge, traceJudgeEnsemble, traceSpanKindToOpenInferenceKind, tracedAnalyzeTraces, trainingReward, trainingScore, typoMutator, urlContains, userQuestionsForKnowledgeGaps, validateAgentProfileCell, validateProductBenchmarkManifest, validateProductBenchmarkRecord, validateProductBenchmarkRun, validateRolloutLine, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyAttestation, verifyCompletion, verifyManifest, visualDiff, viteDeployRunner, vitestTestParser, weightedComposite, weightedMean, weightedRecall, welchsTTest, whitespaceCollapseMutator, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit, withHeldoutBlend, withJudgeRetry, withOtelPipeline, wranglerDeployRunner, writeSupervisorRunReport };
+export { AGENT_PROFILE_KINDS, ATIF_SCHEMA_VERSION, ATTESTATION_ALGORITHM, AgentDriver, AgentEvalError, AgentProfileCellValidationError, AnalystRegistry, AxGepaSteeringOptimizer, BENCHMARK_SPLIT_SEED, BOOTSTRAP_GATE_MIN_N, BackendIntegrityError, BenchmarkRunner, BudgetBreachError, BudgetGuard, CODING_HARNESSES, CallbackResearcher, CaptureIntegrityError, ConfigError, ConvergenceTracker, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, CostLedger, CostLedgerPersistenceError, CostReceiptCaptureError, CostReservationExceededError, CostTracker, CrossFamilyError, DEFAULT_AGENT_SLOS, DEFAULT_COMPLEXITY_WEIGHTS, DEFAULT_RULES as DEFAULT_FAILURE_RULES, DEFAULT_FINDERS, DEFAULT_HARNESS_OBJECTIVES, DEFAULT_MUTATION_PRIMITIVES, DEFAULT_MUTATORS, DEFAULT_PERMUTATIONS, DEFAULT_PR_REVIEW_SCORE_WEIGHTS, DEFAULT_REDACTION_RULES, DEFAULT_RED_TEAM_CORPUS, DEFAULT_RUN_SCORE_WEIGHTS, DEFAULT_SEVERITY_WEIGHTS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, Dataset, DescriptionLengthGate, DockerSandboxDriver, DualAgentBench, ERROR_COUNT_PATTERNS, EvalTraceStore, ExperimentTracker, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, FileSystemTraceStore, FindingsStore, HARBOR_IMPORT_GAP, HARNESS_NATIVE_MODEL, HeldOutGate, HoldoutAuditor, HoldoutLockedError, IMPROVEMENT_KIND_SPEC, INPUT_VALUE, INTENT_MATCH_JUDGE_VERSION, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, InMemoryTraceStore, InMemoryWorkspaceInspector, JudgeError, JudgeParseError, JudgeRunner, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, LLM_CACHED_TOKENS, LLM_CACHED_TOKEN_ATTR_KEYS, LLM_CACHE_WRITE_TOKENS, LLM_CACHE_WRITE_TOKEN_ATTR_KEYS, LLM_CONTEXT_TOKENS, LLM_COST_ATTR_KEYS, LLM_COST_USD, LLM_INPUT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_MODEL_ATTR_KEYS, LLM_MODEL_NAME, LLM_OUTPUT_TOKENS, LLM_OUTPUT_TOKEN_ATTR_KEYS, LLM_REASONING_TOKENS, LLM_REASONING_TOKEN_ATTR_KEYS, LlmCallError, LlmClient, LlmResponseError, LlmRouteAssertionError, LockedJsonlAppender, MANN_WHITNEY_EXACT_MAX_STATES, MANN_WHITNEY_EXACT_MAX_WORK, MODEL_PRICING, MetricsCollector, ModelsUnreachableError, MultiLayerVerifier, Mutex, NoopRawProviderSink, NoopResearcher, NotFoundError, OPENINFERENCE_SPAN_KIND, OTEL_AGENT_EVAL_SCOPE, OUTPUT_VALUE, OtlpFileTraceStore, PairwiseSteeringOptimizer, ProductClient, PromptRegistry, REDACTION_VERSION, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, RESEARCH_REPORT_HARD_PAIR_FLOOR, ROLLOUT_SCHEMA, RUN_COST_ATTR_KEYS, ReplayCache, ReplayCacheMissError, ReplayError, RunCritic, RunIntegrityError, RunRecordValidationError, SEMANTIC_CONCEPT_JUDGE_VERSION, SKILL_USAGE_ANALYST, SPAN_KIND_ATTR_KEYS, SUPERVISOR_RUN_SCHEMA, SandboxHarness, ScenarioRegistry, SeatUnsetError, SingleBackendError, SkillUsageAnalyst, SpanNotFoundError, SubprocessSandboxDriver, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TRACE_SCHEMA_VERSION, TokenCounter, TraceContractBuilder, TraceEmitter, TraceFileMissingError, TraceNotFoundError, UI_FINDING_SEVERITIES, UI_LENSES, UNIVERSAL_FINDERS, ValidationError, VerificationError, WILCOXON_EXACT_MAX_N, acquisitionPlansForKnowledgeGaps, agentProfileCellHashMaterial, agentProfileCellKey, agentProfileHash, agentProfileId, agentProfileModelId, agentVisibleFields, aggregateJudgeVerdicts, aggregateLlm, aggregatePrReviewScore, aggregateRunScore, allCriticalPassed, analyzeAntiSlop, analyzeSeries, analyzeSupervisorRun, analyzeSupervisorRunSources, analyzeTraces, appendScorecard, applyLlmSpanOtlpAttributes, applyToolSpanOtlpAttributes, argHash, asNumber, asString, assertCapabilityHeadroom, assertCrossFamily, assertLlmRoute, assertMinted, assertMintedLines, assertModelsServed, assertNoHiddenLeak, assertProductBenchmarkRun, assertRealAgentReceipts, assertRealBackend, assertReleaseConfidence, assertRolloutLine, assertRunAgentProfileCell, assertRunCaptured, assertSingleBackend, assignFeedbackSplit, assignHeldOutTag, attachCostToReport, attest, attributeCounterfactuals, backoffMs, deterministicSplit as benchmarkDeterministicSplit, benchmarks_exports as benchmarks, benjaminiHochberg, bisect, blendHeldout, blockingKnowledgeEval, bonferroni, bootstrapCi, buildAgentInterfaceProfileCell, buildAgentProfileCell, buildDefaultAnalystRegistry, buildDriverSystemPrompt, buildProductBenchmarkManifest, buildReflectionPrompt, buildReviewerPrompt, buildTraceAnalystTools, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, buildWorkerDriverSystemPrompt, byteLengthRange, cachedJudge, calibrateJudge, calibrateJudgeContinuous, callLlm, callLlmJson, canaryLeakView, canonicalJson, canonicalize, capabilityHeadroom, captureFetchToRawSink, causalAttribution, checkBehavioralCanary, checkCanaries, checkSlos, checkTraceContracts, clamp01, classifyFailure, classifyOtlpSpanRole, classifyTreatment, claudeCodeSupervisorRunReader, cliffsDelta, clusteredPairedBinary, cohensD, collectionPreserved, commentsForSource, commitBisect, comparePairedArms, compareReferenceReplay, compareToBaseline, compilerJudge, completionVerdict, composeParsers, composeValidators, computeExperimentStats, computeFindingId, computeToolUseMetrics, computeTraceMetrics, confidenceInterval, containsAll, contentHash, contextInputTokens, continuousAgreement, contractJudge, controlFailureClassFromVerification, controlRunToFeedbackTrajectory, controlRunToRunRecord, convertTraceStoresToOtlp, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, costForTokenPricing, costForUsage, costReceiptFromLlm, costReceiptFromLlmError, costReport, createAnalystAi, createAntiSlopJudge, createChatClient, createDefaultReviewer, createFeedbackTrajectory, createIntentMatchJudge, createLlmCorrectnessChecker, createLlmReviewer, createOtelExporter, createOtelTracingStore, createReferenceEquivalenceJudge, createReplayFetch, createSandboxPool, createSemanticConceptJudge, createTokenRecallChecker, createTraceAnalystKind, crossTraceDiff, crowdingDistance, dataDescriptionBits, decideNextUserTurn, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultBlendWeights, defaultIsMaterial, defaultProviderRedactor, defaultReferenceReplayMatcher, defaultTraceInsightPanel, deployGateLayer, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, distillPlaybook, domainEvidencePattern, dominates, eProcess, ensembleJudge, errorStreakDetector, estimateCost, estimateTokens, evaluateActionPolicy, evaluateContract, evaluateHypothesis, evaluateInterimReleaseConfidence, evaluateOracles, evaluateReleaseConfidence, evaluateTraceContract, executeScenario, expandProfileAxes, expectAgent, exportProductBenchmark, exportProductBenchmarkRuns, exportRewardModel, exportRunAsOtlp, extractAssetUrls, extractErrorCount, extractOtlpAttributes, extractProducedState, extractUsage, extractUsageFromResponse, extractUsageFromSse, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToDatasetScenario, feedbackTrajectoryToOptimizerRow, fileContains, fileExists, fileExperimentStore, fileVerdictCache, findAutoMatchNoExpectation, findConstructorCwdDropped, findFallbackToPass, findLiteralTruePass, findProductBenchmarkArtifacts, findSkipCountsAsPass, firstNumberAttr, firstStringAttr, flattenOtlpExportToNdjson, flowLayer, fnv1a32, formatBenchmarkReport, formatDriverReport, formatFindings, formatScorecardDiff, fromHarborTrajectory, gainHistogram, gateTreatmentApplied, gateTreatmentFromMetrics, gateTreatmentFromSpans, gateTreatmentFromToolSpans, ghCliClient, gitProvenanceReader, precision as goldenPrecision, gradeOnHidden, gradeSemanticStatus, groupBy, groupRunsByAgentProfileCell, harnessAxisOf, hasCapturedToolArgs, hashContent, hashJson, hashScenarios, hashToUnit, hiddenGrade, holm, htmlContainsElement, httpGithubClient, improvementVerdict, inMemoryExperimentStore, inMemoryReferenceReplayStore, inMemoryReviewStore, inMemoryRunRecordBackend, inMemoryVerdictCache, inferDomainKeywords, inferOtlpKind, interRaterReliability, interpretCliffs, iqr, isHiddenDestination, isJudgeSpan, isLlmSpan, isModelPriced, isOtelConfigured, isOtlpModelCall, isRealnessGated, isRetrievalSpan, isRolloutLine, isRunRecord, isSandboxSpan, isToolSpan, isTrainableSplit, isTransientLlmError, isUnavailable, iterateRawCalls, jestTestParser, jsonHasKeys, jsonShape, jsonlReferenceReplayStore, jsonlReviewStore, jsonlRunRecordBackend, judgeFamily, judgeReplayGate, judgeSpans, keyPreserved, knowledgeReadinessTracePayload, leaderboard, linterJudge, llmJudge, llmSpanFromProvider, llmSpans, loadScorecard, loadScorerFromGrader, localCommandRunner, lowercaseMutator, makeEvalTools, makeFinding, mannWhitneyU, mapConcurrent, matchGoldens, matchSpan, maximumChargeForLlmRequest, mcnemar, mcnemarPower, mcnemarRequiredN, mergeLayerResults, mergeSteeringBundle, minimumPairsForPairedDeltaTest, mintRolloutRows, modelDescriptionBits, modelHasSnapshot, modelPriceKey, mulberry32, multiToolchainLayer, noProgressDetector, normalCdf, normalizeScores, notBlocked, objectiveEval, observeAll, observedScore, observedSplitScore, otelRunCompleteHook, otlpRowsToRunRecords, otlpRowsToTraceRunRecords, otlpToRunRecords, otlpToTraceRunRecords, pairArms, pairRunRecords, pairedBootstrap, pairedCohensDz, pairedDeltaTest, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedSignTest, pairedTTest, paraphraseRobustness, paraphraseRobustnessScenarios, paretoChart, paretoFrontier, paretoFrontierWithCrowding, parseCorrectnessResponse, parseFeedbackTrajectoriesJsonl, parseReflectionResponse, parseRunRecordSafe, parseRuntimeTrajectoryHookEvent, partialCredit, partitionHeldOut, passAtK, passOrthogonality, pearsonR, pixelDeltaRatio, planTraceInsightQuestions, politenessPrefixMutator, positionalBias, preflightModels, printDriverSummary, probeLlm, productBenchmarkIntegrityFailures, productBenchmarkMutableSurfaces, productBenchmarkRepoIdentity, productBenchmarkSplits, profile_exports as profile, projectOtlpFlatLine, projectRuntimeTrajectoryEvidence, promptBisect, proposeSynthesisTargets, providerFromBaseUrl, pytestTestParser, ranks, readClaudeCodeSupervisorRun, readOtlpStatus, readProductBenchmarkManifest, readProductBenchmarkRecords, recordRuns, recordRunsToScorecard, redTeamDataset, redTeamReport, redactString, redactValue, referenceReplayRunsToSteeringRows, referenceReplayScenarioToRunScore, regexMatch, regexMatches, relabelImportedSplit, renderMarkdownReport, renderPlaybookMarkdown, renderPreferenceMemoryMarkdown, renderPriorFindings, renderReleaseReport, renderSteeringText, renderSupervisorRunHeadline, renderSupervisorRunMarkdown, renderUpstreamFindings, repeatedActionDetector, replayFeedbackTrajectories, replayFeedbackTrajectory, replayScorerOverCorpus, replayTraceThroughJudge, requireAgentProfileCell, requiredPairedSampleSize, requiredSampleSize, researchReport, resolveModelPricing, resolveSeat, rollupSupervisorRuns, roundTripRunRecord, routeFields, rowCount, rowWhere, runAgentControlLoop, runAssertions, runBehavioralCanaries, runCanaries, runCounterfactual, runE2EWorkflow, runEvalCampaign, runExpectations, runFailureClass, runHarnessExperiment, runIntentMatchJudge, runJudgeFleet, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runLiveProof, runProposeReview, runProposeReviewAsControlLoop, runRecordToProductBenchmarkRecord, runReferenceEquivalenceJudge, runReferenceReplay, runScore, runSelfPlay, runSemanticConceptJudge, runTaskScore, runTestGradedScenario, runsForScenario, scalarScore, scanForMuffledGates, scoreContinuity, scoreFromEvals, scoreKnowledgeReadiness, scoreOrigin, scorePrReviewComments, scorePrReviewSource, scoreRedTeamOutput, scoreReferenceReplay, scoreTraceInsightReadiness, seatPresets, securityJudge, selectHarnessVariant, selfPreference, sentenceReorderMutator, serializeFeedbackTrajectoriesJsonl, showMeasured, signManifest, spearmanR, statusAdvanced, stopOnNoProgress, stopOnRepeatedAction, stringField, stripFencedJson, studentTCdf, subjectiveEval, summarizeAgentReceiptIntegrity, summarizeBackendIntegrity, summarizeHarnessResults, summarizePrReviewBenchmark, summarizePreferenceMemory, summaryTable, supervisorRunRolloutLines, testJudge, textInSnapshot, throwIfRunIncomplete, toAgentProfileJson, toHarborTrajectories, toHarborTrajectory, toJsonl, toLangfuseEnvelope, toOpenAiTool, toPrometheusText, toRewardRows, toSftRows, tokenizeDomainWords, toolNamesForRun, toolSpans, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceContract, traceJudge, traceJudgeEnsemble, traceSpanKindToOpenInferenceKind, tracedAnalyzeTraces, trainingReward, trainingScore, typoMutator, urlContains, userQuestionsForKnowledgeGaps, validateAgentProfileCell, validateProductBenchmarkManifest, validateProductBenchmarkRecord, validateProductBenchmarkRun, validateRolloutLine, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyAttestation, verifyCompletion, verifyManifest, visualDiff, viteDeployRunner, vitestTestParser, weightedComposite, weightedMean, weightedRecall, welchsTTest, whitespaceCollapseMutator, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit, withHeldoutBlend, withJudgeRetry, withOtelPipeline, wranglerDeployRunner, writeSupervisorRunReport };
 
 //# sourceMappingURL=index.js.map