@tangle-network/agent-eval 0.138.0 → 0.139.0

package/dist/index.js

@@ -2,22 +2,22 @@ import { t as __exportAll } from "./rolldown-runtime-8H4AJuhK.js";
 import { _ as observeAll, a as jsonlReviewStore, c as scoreFromEvals, d as runAgentControlLoop, f as stopOnNoProgress, g as noProgressDetector, h as errorStreakDetector, i as inMemoryReviewStore, l as allCriticalPassed, m as subjectiveEval, n as runProposeReviewAsControlLoop, o as runProposeReview, p as stopOnRepeatedAction, r as createLlmReviewer, s as controlRunToRunRecord, t as controlFailureClassFromVerification, u as objectiveEval, v as repeatedActionDetector, y as evaluateActionPolicy } from "./propose-review-control-DLXz4FCX.js";
 import { a as LimitExceededError, c as ValidationError, i as JudgeError, l as VerificationError, n as CaptureIntegrityError, o as NotFoundError, r as ConfigError, s as ReplayError, t as AgentEvalError } from "./errors-D-LKuDhb.js";
 import { _ as verifyManifest, a as assertRunAgentProfileCell, c as groupRunsByAgentProfileCell, d as validateAgentProfileCell, f as verifyAgentProfileCell, g as signManifest, h as hashJson, i as agentProfileCellKey, l as requireAgentProfileCell, m as evaluateHypothesis, n as AgentProfileCellValidationError, o as buildAgentInterfaceProfileCell, p as canonicalize, r as agentProfileCellHashMaterial, s as buildAgentProfileCell, t as AGENT_PROFILE_KINDS, u as toAgentProfileJson } from "./agent-profile-cell-CbfBm2g6.js";
-import { C as renderUpstreamFindings, G as createChatClient, K as createAnalystAi, S as renderPriorFindings, W as computeTraceMetrics, a as analystFindingDigest, b as emitControlIntegrityFindings, c as completedAnalystReviewQuality, d as validateAnalystReviewDecisions, f as DEFAULT_TRACE_ANALYST_KINDS, g as FAILURE_MODE_KIND_SPEC, h as IMPROVEMENT_KIND_SPEC, i as assertExactRegistryRunOpts, l as readAnalystReview, m as KNOWLEDGE_GAP_KIND_SPEC, n as AnalystRegistry, o as analystRunDigest, p as KNOWLEDGE_POISONING_KIND_SPEC, r as ExactAnalystRunExecutionError, s as assertUniqueFindingIds, t as buildDefaultAnalystRegistry, u as snapshotAnalystRun, v as CONTROL_INTEGRITY_ANALYST, x as createTraceAnalystKind, y as ControlIntegrityAnalyst } from "./default-registry-lp5R0lve.js";
 import { a as estimateTokens, i as estimateCost, n as MetricsCollector, o as isModelPriced, r as TokenCounter, s as resolveModelPricing, t as MODEL_PRICING } from "./metrics-C9YY1OcL.js";
-import { a as CostLedgerPersistenceError, c as costForTokenPricing, i as CostLedger, l as costForUsage, n as CostCallConflictError, o as CostReceiptCaptureError, r as CostCeilingReachedError, s as CostReservationExceededError, t as CostAccountingIncompleteError, u as modelPriceKey } from "./cost-ledger-CHDLA0Ss.js";
+import { a as CostLedgerPersistenceError, c as costForTokenPricing, i as CostLedger, l as costForUsage, n as CostCallConflictError, o as CostReceiptCaptureError, r as CostCeilingReachedError, s as CostReservationExceededError, t as CostAccountingIncompleteError, u as modelPriceKey } from "./cost-ledger-CZ9diLxY.js";
 import { a as providerFromBaseUrl, i as defaultProviderRedactor, n as InMemoryRawProviderSink, r as NoopRawProviderSink, t as FileSystemRawProviderSink } from "./raw-provider-sink-BQd7mzyT.js";
-import { a as assertLlmRoute, c as callLlmJson, d as isTransientLlmError, f as maximumChargeForLlmRequest, i as LlmRouteAssertionError, l as costReceiptFromLlm, m as stripFencedJson, n as LlmClient, o as backoffMs, p as probeLlm, r as LlmResponseError, s as callLlm, t as LlmCallError, u as costReceiptFromLlmError } from "./llm-client-Cj3c7PEm.js";
+import { a as assertLlmRoute, c as callLlmJson, d as isTransientLlmError, f as maximumChargeForLlmRequest, i as LlmRouteAssertionError, l as costReceiptFromLlm, m as stripFencedJson, n as LlmClient, o as backoffMs, p as probeLlm, r as LlmResponseError, s as callLlm, t as LlmCallError, u as costReceiptFromLlmError } from "./llm-client-bkztEfIx.js";
+import { C as createChatClient, S as computeTraceMetrics, _ as CONTROL_INTEGRITY_ANALYST, a as analystFindingDigest, c as completedAnalystReviewQuality, d as validateAnalystReviewDecisions, f as DEFAULT_TRACE_ANALYST_KINDS, g as FAILURE_MODE_KIND_SPEC, h as IMPROVEMENT_KIND_SPEC, i as assertExactRegistryRunOpts, l as readAnalystReview, m as KNOWLEDGE_GAP_KIND_SPEC, n as AnalystRegistry, o as analystRunDigest, p as KNOWLEDGE_POISONING_KIND_SPEC, r as ExactAnalystRunExecutionError, s as assertUniqueFindingIds, t as buildDefaultAnalystRegistry, u as snapshotAnalystRun, v as ControlIntegrityAnalyst, y as emitControlIntegrityFindings } from "./default-registry-BgJJItGr.js";
 import { INPUT_VALUE, 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, OPENINFERENCE_SPAN_KIND, OUTPUT_VALUE, RUN_COST_ATTR_KEYS, SPAN_KIND_ATTR_KEYS, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, applyLlmSpanOtlpAttributes, asNumber, contextInputTokens, firstNumberAttr } from "./trace-attributes.js";
-import { A as classifyOtlpSpanRole, C as firstStringAttr, E as readOtlpStatus, M as traceSpanKindToOpenInferenceKind, O as stringField, S as extractOtlpAttributes, T as projectOtlpFlatLine, _ as TraceFileMissingError, a as createBoundedTraceAnalysisStore, b as asString, d as TRACE_ANALYSIS_LIMITS, f as SpanNotFoundError, g as TraceFileMalformedError, h as TraceAnalysisValidationError, i as otlpTextToTraceAnalysisStore, j as isOtlpModelCall, k as applyToolSpanOtlpAttributes, l as DEFAULT_TRACE_ANALYST_BUDGETS, m as TraceAnalysisStoreContractError, n as OtlpFileTraceStore, p as TraceAnalysisLimitError, u as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, v as TraceFileTooLargeError, w as inferOtlpKind, y as TraceNotFoundError } from "./store-otlp-BenKynPE.js";
-import { h as makeProposalFinding, m as makeFinding, p as computeFindingId } from "./single-run-lock-D_bS5xhj.js";
-import { f as Mutex, p as mapConcurrent } from "./ledger-core-C0Yx1I14.js";
-import { i as traceAnalystFunctionGroup, n as buildTraceAnalysisToolDescriptors, r as buildTraceAnalystTools, t as TRACE_ANALYST_TOOL_NAMESPACE } from "./tools-DZGdROtG.js";
+import { C as TraceNotFoundError, G as resolveTraceAnalystLimits, J as extractOtlpAttributes, K as asString, Q as readOtlpStatus, S as TraceFileTooLargeError, W as DEFAULT_TRACE_ANALYST_LIMITS, X as inferOtlpKind, Y as firstStringAttr, Z as projectOtlpFlatLine, _ as TraceAnalysisLimitError, b as TraceFileMalformedError, c as traceAnalystFunctionGroup, et as stringField, g as SpanNotFoundError, h as TRACE_ANALYSIS_LIMITS, i as runTraceAnalyst, it as traceSpanKindToOpenInferenceKind, l as createBoundedTraceAnalysisStore, m as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, n as renderPriorFindings, nt as classifyOtlpSpanRole, o as TRACE_ANALYST_TOOL_NAMESPACE, p as DEFAULT_TRACE_ANALYST_BUDGETS, r as renderUpstreamFindings, rt as isOtlpModelCall, s as buildTraceAnalysisToolDescriptors, t as createTraceAnalyst, tt as applyToolSpanOtlpAttributes, v as TraceAnalysisStoreContractError, x as TraceFileMissingError, y as TraceAnalysisValidationError } from "./kind-factory-CFxA0JQX.js";
+import { a as computeFindingId, o as makeFinding, s as makeProposalFinding } from "./usage-receipt-CgxMEBZq.js";
 import { c as SUPERVISOR_RUN_INTEGRITY_SCHEMA, n as supervisorRunRolloutLines, t as analyzeSupervisorRunIntegrity } from "./integrity-CCXTftiL.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 { 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 clamp01, i as aggregateRunScore, r as DEFAULT_RUN_SCORE_WEIGHTS } from "./proposal-findings-2GIUo1et.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-DYXDPZW0.js";
-import { At as pairedDeltaTest, Ct as paretoFrontier, D as runCanaries, Dt as decidePairedPromotion, Ft as summarizeBackendIntegrity, G as buildReflectionPrompt, It as JudgeParseError, K as parseReflectionResponse, Mt as assertRealAgentReceipts, Nt as assertRealBackend, Ot as pairedDecisionShape, Pt as summarizeAgentReceiptIntegrity, St as dominates, Tt as scalarScore, W as DEFAULT_MUTATION_PRIMITIVES, _t as canonicalJson, at as createReferenceEquivalenceJudge, bt as inMemoryVerdictCache, ct as redTeamDataset, dt as toolNamesForRun, ft as Dataset, gt as cachedJudge, ht as llmJudge, it as REFERENCE_EQUIVALENCE_JUDGE_VERSION, jt as BackendIntegrityError, kt as minimumPairsForPairedDeltaTest, lt as redTeamReport, mt as hashScenarios, ot as runReferenceEquivalenceJudge, pt as HoldoutLockedError, rt as REFERENCE_EQUIVALENCE_INPUT_LIMITS, st as DEFAULT_RED_TEAM_CORPUS, ut as scoreRedTeamOutput, vt as contentHash, wt as paretoFrontierWithCrowding, xt as crowdingDistance, yt as fileVerdictCache, z as surfaceContentHash } from "./skillopt-optimization-method-CjKMZy0d.js";
+import { a as RunCritic, d as defaultIsMaterial, f as diffFindings, h as defineTraceAnalyst, 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-BuIJ9IfB.js";
+import { t as createDspyRlmTraceEngine } from "./dspy-rlm-engine-DTkVyDX-.js";
+import { f as Mutex, p as mapConcurrent } from "./ledger-core-Dxz0Rkwa.js";
+import { At as pairedDeltaTest, Ct as paretoFrontier, D as runCanaries, Dt as decidePairedPromotion, Ft as summarizeBackendIntegrity, G as buildReflectionPrompt, It as JudgeParseError, K as parseReflectionResponse, Mt as assertRealAgentReceipts, Nt as assertRealBackend, Ot as pairedDecisionShape, Pt as summarizeAgentReceiptIntegrity, St as dominates, Tt as scalarScore, W as DEFAULT_MUTATION_PRIMITIVES, _t as canonicalJson, at as createReferenceEquivalenceJudge, bt as inMemoryVerdictCache, ct as redTeamDataset, dt as toolNamesForRun, ft as Dataset, gt as cachedJudge, ht as llmJudge, it as REFERENCE_EQUIVALENCE_JUDGE_VERSION, jt as BackendIntegrityError, kt as minimumPairsForPairedDeltaTest, lt as redTeamReport, mt as hashScenarios, ot as runReferenceEquivalenceJudge, pt as HoldoutLockedError, rt as REFERENCE_EQUIVALENCE_INPUT_LIMITS, st as DEFAULT_RED_TEAM_CORPUS, ut as scoreRedTeamOutput, vt as contentHash, wt as paretoFrontierWithCrowding, xt as crowdingDistance, yt as fileVerdictCache, z as surfaceContentHash } from "./skillopt-optimization-method-BbGnCC53.js";
 import { $ as selfPreference, A as pairedRiskDifference, B as requiredSampleSize, C as mulberry32, D as pairedCohensDz, E as pairedBootstrap, F as partialCredit, G as wilson, H as weightedComposite, I as passAtK, J as normalCdf, K as studentTCdf, L as pearsonR, M as pairedRiskDifferenceScore, N as pairedSignTest, O as pairedDeltaTieFraction, P as pairedTTest, Q as positionalBias, R as ranks, S as mcnemarRequiredN, T as pairedBinaryScale, U as weightedMean, V as spearmanR, W as wilcoxonSignedRank, X as calibrateJudgeContinuous, Y as calibrateJudge, Z as continuousAgreement, _ as interpretCliffs, a as MANN_WHITNEY_EXACT_MAX_WORK, b as mcnemar, c as bonferroni, d as confidenceInterval, et as verbosityBias, f as corpusInterRaterAgreement, g as interRaterReliability, h as holm, i as MANN_WHITNEY_EXACT_MAX_STATES, j as pairedRiskDifferenceExact, k as pairedMde, l as cliffsDelta, m as eProcess, n as DECISION_PAIRED_DELTA_STATISTIC, o as WILCOXON_EXACT_MAX_N, p as corpusInterRaterAgreementFromJudgeScores, r as DEFAULT_PERMUTATIONS, s as benjaminiHochberg, t as BOOTSTRAP_GATE_MIN_N, u as cohensD, v as isBinaryOutcomeVector, w as normalizeScores, x as mcnemarPower, y as mannWhitneyU, z as requiredPairedSampleSize } from "./statistics-ByxzSiOM.js";
 import { n as pairArms, r as pairRunRecords, t as comparePairedArms } from "./paired-arms-iZ08VFMN.js";
 import { n as llmSpanFromProvider, t as TraceEmitter } from "./emitter-CPBAhxum.js";
@@ -25,13 +25,14 @@ import { a as isRetrievalSpan, i as isLlmSpan, n as TRACE_SCHEMA_VERSION, o as i
 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-vRgqWmJw.js";
 import { a as evaluateReleaseConfidence, i as assertReleaseConfidence, n as bootstrapCi, r as judgeReplayGate, t as renderReleaseReport } from "./release-report-B5XPBvAU.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-DQFl0UXA.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-8nj3mYvx.js";
 import { t as buildTrajectory } from "./trajectory-D_7rLrvE.js";
 import { n as unmintableReasons, t as mintRolloutRows } from "./mint-Ctwk079K.js";
 import { C as SUPERVISOR_RUN_SCHEMA, T as showMeasured, _ as readClaudeCodeSupervisorRun, b as rollupSupervisorRuns, c as writeSupervisorRunReport, d as renderSupervisorRunHeadline, f as renderSupervisorRunMarkdown, g as claudeCodeSupervisorRunReader, t as analyzeSupervisorRun, v as analyzeSupervisorRunSources, w as isUnavailable } from "./supervisor-run-B2EWUmQY.js";
-import { A as TRACE_ANALYST_ACTOR_DESCRIPTION, C as domainEvidencePattern, D as tokenizeDomainWords, E as scoreTraceInsightReadiness, O as traceAnalystOnRunComplete, S as describeTraceInsightScope, T as planTraceInsightQuestions, _ as otlpToTraceRunRecords, a as convertTraceStoresToOtlp, b as buildTraceInsightPrompt, c as otelRunCompleteHook, d as captureFetchToRawSink, f as ToolTraceMissingError, g as otlpToRunRecords, h as otlpRowsToTraceRunRecords, i as iterateRawCalls, j as TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, k as analyzeTraces, l as OTEL_AGENT_EVAL_SCOPE, m as otlpRowsToRunRecords, n as ReplayCacheMissError, o as createOtelExporter, p as toolSpansToTraceAnalysisStore, r as createReplayFetch, s as createOtelTracingStore, t as ReplayCache, u as exportRunAsOtlp, v as flattenOtlpExportToNdjson, w as inferDomainKeywords, x as defaultTraceInsightPanel, y as buildTraceInsightContext } from "./replay-Cb-4Vf0k.js";
-import { n as extractUsageFromResponse, r as extractUsageFromSse, t as extractUsage } from "./extract-usage-CS391dOE.js";
-import { B as agentProfileModelId, G as createLlmCorrectnessChecker, H as harnessAxisOf, I as CODING_HARNESSES, J as verifyCompletion, K as createTokenRecallChecker, L as HARNESS_NATIVE_MODEL, R as agentProfileHash, U as extractProducedState, V as expandProfileAxes, W as completionVerdict, q as parseCorrectnessResponse, z as agentProfileId } from "./campaign-BIBS-NHV.js";
+import { A as TRACE_ANALYST_ACTOR_DESCRIPTION, C as domainEvidencePattern, D as tokenizeDomainWords, E as scoreTraceInsightReadiness, O as traceAnalystOnRunComplete, S as describeTraceInsightScope, T as planTraceInsightQuestions, _ as otlpToTraceRunRecords, a as convertTraceStoresToOtlp, b as buildTraceInsightPrompt, c as otelRunCompleteHook, d as captureFetchToRawSink, f as ToolTraceMissingError, g as otlpToRunRecords, h as otlpRowsToTraceRunRecords, i as iterateRawCalls, j as TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, k as analyzeTraces, l as OTEL_AGENT_EVAL_SCOPE, m as otlpRowsToRunRecords, n as ReplayCacheMissError, o as createOtelExporter, p as toolSpansToTraceAnalysisStore, r as createReplayFetch, s as createOtelTracingStore, t as ReplayCache, u as exportRunAsOtlp, v as flattenOtlpExportToNdjson, w as inferDomainKeywords, x as defaultTraceInsightPanel, y as buildTraceInsightContext } from "./replay-SA4OB7O7.js";
+import { i as otlpTextToTraceAnalysisStore, n as OtlpFileTraceStore } from "./store-otlp-DX4fGIcf.js";
+import { n as extractUsageFromResponse, r as extractUsageFromSse, t as extractUsage } from "./extract-usage-DZs601Va.js";
+import { B as agentProfileModelId, G as createLlmCorrectnessChecker, H as harnessAxisOf, I as CODING_HARNESSES, J as verifyCompletion, K as createTokenRecallChecker, L as HARNESS_NATIVE_MODEL, R as agentProfileHash, U as extractProducedState, V as expandProfileAxes, W as completionVerdict, q as parseCorrectnessResponse, z as agentProfileId } from "./campaign-DrS6_hLd.js";
 import { n as iqr, r as welchsTTest, t as compareToBaseline } from "./baseline-C-GocmIW.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-BScZvqMV.js";
@@ -41,16 +42,15 @@ import { a as InMemoryTraceStore, i as FileSystemTraceStore, n as assertRunCaptu
 import { i as redactValue, n as REDACTION_VERSION, r as redactString, t as DEFAULT_REDACTION_RULES } from "./redact-7Aq1ukl-.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-BJ_xK5rQ.js";
-import { t as runEvalCampaign } from "./eval-campaign-9MozgKL7.js";
+import { _ as deterministicSplit, g as BENCHMARK_SPLIT_SEED, t as benchmarks_exports } from "./benchmarks-zxhy1QV3.js";
+import { t as runEvalCampaign } from "./eval-campaign-BmptJj50.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";
 import { basename, delimiter, dirname, extname, isAbsolute, join, relative, resolve } from "node:path";
 import { createHash } from "node:crypto";
+import { execSync, spawnSync } from "node:child_process";
 import { readFile } from "node:fs/promises";
 import { cpus } from "node:os";
-import { execSync, spawnSync } from "node:child_process";
 import { gzipSync } from "node:zlib";
 //#region src/auto-pr.ts
 /**
@@ -4265,80 +4265,21 @@ function renderSteeringText(bundle) {
 }
 //#endregion
 //#region src/steering-optimizer.ts
+/**
+* Rank already-evaluated steering variants.
+*
+* Model-backed candidate generation belongs to the official GEPA campaign
+* integration. This class only performs the deterministic selection it owns.
+*/
 var PairwiseSteeringOptimizer = class {
 	optimize(rows, config = {}) {
-		const ranked = rankRows(rows, config.weights);
-		if (!ranked.length) throw new Error("no steering optimization rows");
+		const rankings = rankRows(rows, config.weights);
+		if (rankings.length === 0) throw new Error("no steering optimization rows");
 		return {
 			backend: "pairwise",
-			recommendedVariantId: ranked[0].variantId,
+			recommendedVariantId: rankings[0].variantId,
 			rationale: `Highest observed mean aggregate across ${rows.length} scored run(s).`,
-			rankings: ranked
-		};
-	}
-};
-var AxGepaSteeringOptimizer = class {
-	config;
-	constructor(config) {
-		this.config = config;
-	}
-	async optimize(rows) {
-		const fallback = new PairwiseSteeringOptimizer().optimize(rows, this.config);
-		const minScenarioWinners = this.config.minScenarioWinners ?? 6;
-		const variantIds = [...new Set(rows.map((row) => row.variantId))];
-		const byScenario = collapseScenarioWinners(rows, this.config.weights);
-		if (variantIds.length < 2 || byScenario.length < minScenarioWinners) return {
-			...fallback,
-			backend: "ax-gepa",
-			skipped: true,
-			rationale: `AxGEPA skipped: need >=2 variants and >=${minScenarioWinners} scenario winners, got ${variantIds.length} variant(s) and ${byScenario.length} scenario winner(s).`
-		};
-		const signature = `task:string, split:string, seedPreview:string -> variantId:class "${variantIds.join(", ")}", rationale:string`;
-		const selector = ax(AxSignature.from(signature), { description: "Choose the best steering bundle variant for an autopilot task." });
-		const shuffled = seededShuffle(byScenario, signature);
-		const splitIndex = Math.max(1, Math.floor(shuffled.length * .8));
-		const train = shuffled.slice(0, splitIndex);
-		const validation = shuffled.slice(splitIndex);
-		if (!validation.length) return {
-			...fallback,
-			backend: "ax-gepa",
-			skipped: true,
-			rationale: "AxGEPA skipped: no validation examples after split."
-		};
-		const studentAI = createAxService(this.config.provider, this.config.apiKey, this.config.model);
-		const optimizer = new AxGEPA({
-			studentAI,
-			teacherAI: createAxService(this.config.provider, this.config.apiKey, this.config.teacherModel ?? this.config.model),
-			numTrials: 8,
-			minibatch: true,
-			minibatchSize: 4,
-			earlyStoppingTrials: 3,
-			sampleCount: 1
-		});
-		const metric = ({ prediction, example }) => stringField$1(prediction, "variantId") === stringField$1(example, "variantId") ? 1 : 0;
-		const compiled = await optimizer.compile(selector, train, metric, {
-			validationExamples: validation,
-			maxMetricCalls: 64
-		});
-		if (compiled.optimizedProgram !== void 0) selector.applyOptimization(compiled.optimizedProgram);
-		const selectVariant = async (row) => {
-			const prediction = await selector.forward(studentAI, row);
-			return {
-				variantId: String(prediction.variantId),
-				rationale: String(prediction.rationale ?? "")
-			};
-		};
-		return {
-			...fallback,
-			backend: "ax-gepa",
-			rationale: `AxGEPA trained a variant selector from ${byScenario.length} scored scenario winner(s); default winner remains ${fallback.recommendedVariantId}.`,
-			selector: {
-				backend: "ax-gepa",
-				signature,
-				labels: variantIds,
-				rationale: compiled.bestScore !== void 0 ? `bestScore=${compiled.bestScore}` : void 0
-			},
-			selectVariant
+			rankings
 		};
 	}
 };
@@ -4355,64 +4296,6 @@ function rankRows(rows, weights) {
 		runs: values.length
 	})).sort((a, b) => b.mean - a.mean);
 }
-function collapseScenarioWinners(rows, weights) {
-	const byScenario = /* @__PURE__ */ new Map();
-	for (const row of rows) {
-		const bucket = byScenario.get(row.scenarioId) ?? [];
-		bucket.push(row);
-		byScenario.set(row.scenarioId, bucket);
-	}
-	return [...byScenario.entries()].map(([scenarioId, scenarioRows]) => {
-		const best = scenarioRows.map((row) => ({
-			row,
-			aggregate: aggregateRunScore(row.score, weights)
-		})).sort((a, b) => b.aggregate - a.aggregate)[0];
-		return {
-			task: String(best.row.metadata?.task ?? best.row.metadata?.seed_preview ?? scenarioId),
-			split: String(best.row.metadata?.split ?? "train"),
-			seedPreview: String(best.row.metadata?.seed_preview ?? ""),
-			variantId: best.row.variantId
-		};
-	});
-}
-function createAxService(provider, apiKey, model) {
-	return createAnalystAi({
-		provider,
-		apiKey,
-		model
-	});
-}
-function stringField$1(value, field) {
-	if (!value || typeof value !== "object") return void 0;
-	const candidate = value[field];
-	return typeof candidate === "string" ? candidate : void 0;
-}
-function seededShuffle(items, seed) {
-	const rng = mulberry32$1(hashString(seed));
-	const out = [...items];
-	for (let i = out.length - 1; i > 0; i--) {
-		const j = Math.floor(rng() * (i + 1));
-		[out[i], out[j]] = [out[j], out[i]];
-	}
-	return out;
-}
-function hashString(value) {
-	let h = 2166136261;
-	for (let i = 0; i < value.length; i++) {
-		h ^= value.charCodeAt(i);
-		h = Math.imul(h, 16777619);
-	}
-	return h >>> 0;
-}
-function mulberry32$1(seed) {
-	let a = seed >>> 0;
-	return () => {
-		a = a + 1831565813 | 0;
-		let t = Math.imul(a ^ a >>> 15, 1 | a);
-		t = t + Math.imul(t ^ t >>> 7, 61 | t) ^ t;
-		return ((t ^ t >>> 14) >>> 0) / 4294967296;
-	};
-}
 //#endregion
 //#region src/workspace-inspector.ts
 var InMemoryWorkspaceInspector = class {
@@ -11316,56 +11199,48 @@ function sectionHash(section) {
 }
 //#endregion
 //#region src/traced-analyst.ts
-/**
-* Run `analyzeTraces` wrapped in a parent span with per-turn child spans.
-*/
-async function tracedAnalyzeTraces(input, options, traceOpts) {
-	const parentSpan = await traceOpts.emitter.span({
+/** Run a recursive trace investigation and record its engine steps. */
+async function tracedAnalyzeTraces(input, options, traceOptions) {
+	const parentSpan = await traceOptions.emitter.span({
 		kind: "custom",
 		name: "analyst:analyze-traces",
-		parentSpanId: traceOpts.parentSpanId,
+		parentSpanId: traceOptions.parentSpanId,
 		attributes: {
+			"analyst.engine": options.engine.id,
 			"analyst.question_length": input.question.length,
-			"analyst.max_turns": options.maxTurns ?? 12,
-			"analyst.max_subqueries": options.maxSubqueries ?? 4,
+			"analyst.max_iterations": options.limits?.maxIterations ?? 12,
+			"analyst.max_llm_calls": options.limits?.maxLlmCalls ?? 8,
 			"eval.phase": "analyst"
 		}
 	});
-	const originalOnTurn = options.onTurn;
-	const wrappedOptions = {
-		...options,
-		onTurn: async (turn) => {
-			const turnSpan = await traceOpts.emitter.span({
+	try {
+		const result = await analyzeTraces(input, options);
+		for (const [index, step] of result.trajectory.entries()) {
+			const encoded = JSON.stringify(step);
+			await (await traceOptions.emitter.span({
 				kind: "custom",
-				name: `analyst:turn-${turn.turn}`,
+				name: `analyst:step-${index + 1}`,
 				parentSpanId: parentSpan.span.spanId,
 				attributes: {
-					"analyst.stage": turn.stage,
-					"analyst.turn": turn.turn,
-					"analyst.is_error": turn.isError,
-					"analyst.code_length": turn.code.length,
-					"analyst.output_length": turn.output.length,
+					"analyst.step": index + 1,
+					"analyst.step_bytes": Buffer.byteLength(encoded),
 					"eval.phase": "analyst"
 				}
-			});
-			if (turn.isError) await turnSpan.fail("Turn produced an error");
-			else await turnSpan.end();
-			if (originalOnTurn) await originalOnTurn(turn);
+			})).end();
 		}
-	};
-	try {
-		const result = await analyzeTraces(input, wrappedOptions);
 		await parentSpan.end({ attributes: {
-			"analyst.question_length": input.question.length,
-			"analyst.turn_count": result.turnCount,
+			"analyst.engine": options.engine.id,
+			"analyst.model_calls": result.modelCalls,
+			"analyst.tool_calls": result.toolCalls,
+			"analyst.step_count": result.trajectory.length,
 			"analyst.finding_count": result.findings.length,
 			"analyst.answer_length": result.answer.length,
 			"eval.phase": "analyst"
 		} });
 		return result;
-	} catch (err) {
-		await parentSpan.fail(err instanceof Error ? err : String(err));
-		throw err;
+	} catch (error) {
+		await parentSpan.fail(error instanceof Error ? error : String(error));
+		throw error;
 	}
 }
 //#endregion
@@ -12512,6 +12387,6 @@ function assertProductBenchmarkRun(runDir) {
 	return report;
 }
 //#endregion
-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, CONTROL_INTEGRITY_ANALYST, CallbackResearcher, CaptureIntegrityError, ConfigError, ControlIntegrityAnalyst, ConvergenceTracker, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, CostLedger, CostLedgerPersistenceError, CostReceiptCaptureError, CostReservationExceededError, CostTracker, CrossFamilyError, DECISION_PAIRED_DELTA_STATISTIC, 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, ExactAnalystRunExecutionError, 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, LimitExceededError, 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_INTEGRITY_SCHEMA, SUPERVISOR_RUN_SCHEMA, SandboxHarness, ScenarioRegistry, SeatUnsetError, SingleBackendError, SkillUsageAnalyst, SpanNotFoundError, SubprocessSandboxDriver, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYSIS_LIMITS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_TOOL_NAMESPACE, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TRACE_SCHEMA_VERSION, TokenCounter, ToolTraceMissingError, TraceAnalysisLimitError, TraceAnalysisStoreContractError, TraceAnalysisValidationError, TraceContractBuilder, TraceEmitter, TraceFileMalformedError, TraceFileMissingError, TraceFileTooLargeError, 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, analystFindingDigest, analystRunDigest, analystRunToFeedbackTrajectory, analystRunToReviewRequests, analyzeAntiSlop, analyzeSeries, analyzeSupervisorRun, analyzeSupervisorRunIntegrity, analyzeSupervisorRunSources, analyzeTraces, appendScorecard, applyLlmSpanOtlpAttributes, applyToolSpanOtlpAttributes, argHash, asNumber, asString, assertCapabilityHeadroom, assertCrossFamily, assertExactRegistryRunOpts, 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, buildTraceAnalysisToolDescriptors, 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, createBoundedTraceAnalysisStore, createChatClient, createDefaultReviewer, createFeedbackTrajectory, createIntentMatchJudge, createLlmCorrectnessChecker, createLlmReviewer, createOtelExporter, createOtelTracingStore, createReferenceEquivalenceJudge, createReplayFetch, createSandboxPool, createSemanticConceptJudge, createTokenRecallChecker, createTraceAnalystKind, crossTraceDiff, crowdingDistance, dataDescriptionBits, decideNextUserTurn, decidePairedPromotion, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultBlendWeights, defaultIsMaterial, defaultProviderRedactor, defaultReferenceReplayMatcher, defaultTraceInsightPanel, deployGateLayer, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, distillPlaybook, domainEvidencePattern, dominates, eProcess, emitControlIntegrityFindings, 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, isBinaryOutcomeVector, 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, makeProposalFinding, 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, otlpTextToTraceAnalysisStore, otlpToRunRecords, otlpToTraceRunRecords, pairArms, pairRunRecords, pairedBinaryScale, pairedBootstrap, pairedCohensDz, pairedDecisionShape, pairedDeltaTest, pairedDeltaTieFraction, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedRiskDifferenceExact, pairedRiskDifferenceScore, 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, toolSpansToTraceAnalysisStore, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceContract, traceJudge, traceJudgeEnsemble, traceSpanKindToOpenInferenceKind, tracedAnalyzeTraces, trainingReward, trainingScore, typoMutator, unmintableReasons, 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, BENCHMARK_SPLIT_SEED, BOOTSTRAP_GATE_MIN_N, BackendIntegrityError, BenchmarkRunner, BudgetBreachError, BudgetGuard, CODING_HARNESSES, CONTROL_INTEGRITY_ANALYST, CallbackResearcher, CaptureIntegrityError, ConfigError, ControlIntegrityAnalyst, ConvergenceTracker, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, CostLedger, CostLedgerPersistenceError, CostReceiptCaptureError, CostReservationExceededError, CostTracker, CrossFamilyError, DECISION_PAIRED_DELTA_STATISTIC, 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, DEFAULT_TRACE_ANALYST_LIMITS, Dataset, DescriptionLengthGate, DockerSandboxDriver, DualAgentBench, ERROR_COUNT_PATTERNS, EvalTraceStore, ExactAnalystRunExecutionError, 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, LimitExceededError, 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_INTEGRITY_SCHEMA, SUPERVISOR_RUN_SCHEMA, SandboxHarness, ScenarioRegistry, SeatUnsetError, SingleBackendError, SkillUsageAnalyst, SpanNotFoundError, SubprocessSandboxDriver, TOOL_ARGS_CAPTURED, TOOL_LATENCY_MS, TOOL_NAME, TOOL_NAME_ATTR_KEYS, TRACE_ANALYSIS_LIMITS, TRACE_ANALYST_ACTOR_DESCRIPTION, TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION, TRACE_ANALYST_TOOL_NAMESPACE, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TRACE_SCHEMA_VERSION, TokenCounter, ToolTraceMissingError, TraceAnalysisLimitError, TraceAnalysisStoreContractError, TraceAnalysisValidationError, TraceContractBuilder, TraceEmitter, TraceFileMalformedError, TraceFileMissingError, TraceFileTooLargeError, 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, analystFindingDigest, analystRunDigest, analystRunToFeedbackTrajectory, analystRunToReviewRequests, analyzeAntiSlop, analyzeSeries, analyzeSupervisorRun, analyzeSupervisorRunIntegrity, analyzeSupervisorRunSources, analyzeTraces, appendScorecard, applyLlmSpanOtlpAttributes, applyToolSpanOtlpAttributes, argHash, asNumber, asString, assertCapabilityHeadroom, assertCrossFamily, assertExactRegistryRunOpts, 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, buildTraceAnalysisToolDescriptors, 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, createAntiSlopJudge, createBoundedTraceAnalysisStore, createChatClient, createDefaultReviewer, createDspyRlmTraceEngine, createFeedbackTrajectory, createIntentMatchJudge, createLlmCorrectnessChecker, createLlmReviewer, createOtelExporter, createOtelTracingStore, createReferenceEquivalenceJudge, createReplayFetch, createSandboxPool, createSemanticConceptJudge, createTokenRecallChecker, createTraceAnalyst, crossTraceDiff, crowdingDistance, dataDescriptionBits, decideNextUserTurn, decidePairedPromotion, decideReferenceReplayPromotion, decideReferenceReplayRunPromotion, defaultBlendWeights, defaultIsMaterial, defaultProviderRedactor, defaultReferenceReplayMatcher, defaultTraceInsightPanel, defineTraceAnalyst, deployGateLayer, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, distillPlaybook, domainEvidencePattern, dominates, eProcess, emitControlIntegrityFindings, 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, isBinaryOutcomeVector, 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, makeProposalFinding, 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, otlpTextToTraceAnalysisStore, otlpToRunRecords, otlpToTraceRunRecords, pairArms, pairRunRecords, pairedBinaryScale, pairedBootstrap, pairedCohensDz, pairedDecisionShape, pairedDeltaTest, pairedDeltaTieFraction, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedRiskDifferenceExact, pairedRiskDifferenceScore, 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, resolveTraceAnalystLimits, 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, runTraceAnalyst, 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, toolSpansToTraceAnalysisStore, traceAnalystFunctionGroup, traceAnalystOnRunComplete, traceContract, traceJudge, traceJudgeEnsemble, traceSpanKindToOpenInferenceKind, tracedAnalyzeTraces, trainingReward, trainingScore, typoMutator, unmintableReasons, 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