@tangle-network/agent-eval 0.172.1 → 0.173.0

package/dist/index.js

@@ -1,27 +1,27 @@
 import { t as __exportAll } from "./rolldown-runtime-8H4AJuhK.js";
 import { i as JudgeError, o as NotFoundError, r as ConfigError, s as ValidationError, t as AgentEvalError } from "./errors-Dngq5h35.js";
 import { a as hashCanonical, r as canonicalString } from "./canonical-DPyQ_rpt.js";
-import { a as CODING_HARNESSES, c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, n as completionVerdict, o as HARNESS_NATIVE_MODEL, r as createLlmCorrectnessChecker, s as agentProfileHash, t as extractProducedState, u as expandProfileAxes } from "./produced-state-CxmbFxFd.js";
+import { a as CODING_HARNESSES, c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, n as completionVerdict, o as HARNESS_NATIVE_MODEL, r as createLlmCorrectnessChecker, s as agentProfileHash, t as extractProducedState, u as expandProfileAxes } from "./produced-state-7VYDwtkk.js";
 import { n as hashJson, r as manifestContentDigest } from "./pre-registration-D94b7Of5.js";
 import { AGENT_PROFILE_KINDS, agentProfileCellHashMaterial, agentProfileCellKey, buildAgentProfileCell, groupRunsByAgentProfileCell, toAgentProfileJson, validateAgentProfileCell, verifyAgentProfileCell } from "./profile-cell.js";
 import { t as mulberry32 } from "./random-Dn5fPWkt.js";
 import { a as spearmanR, c as weightedMean, i as ranks, n as partialCredit, o as summarizeNumberSeries, r as pearsonR, s as weightedComposite, t as confidenceInterval } from "./descriptive-1V17A-qa.js";
 import { i as verbosityBias, n as calibrateJudgeContinuous, r as continuousAgreement, t as calibrateJudge } from "./judge-calibration-BnpVKtnb.js";
-import { a as budgetBreachView, c as corpusInterRaterAgreementFromJudgeScores, i as failureClusterView, l as interRaterReliability, n as computeToolUseMetrics, o as classifyFailure, r as judgeAgreementView, s as corpusInterRaterAgreement, t as toolWasteView } from "./tool-waste-CwGHzBzX.js";
+import { a as budgetBreachView, c as INFRA_FAILURE_BLAMES, d as failureBlame, f as corpusInterRaterAgreement, i as failureClusterView, l as classifyFailure, m as interRaterReliability, n as computeToolUseMetrics, o as DEFAULT_FAILURE_REASON_RULES, p as corpusInterRaterAgreementFromJudgeScores, r as judgeAgreementView, s as FAILURE_BLAME, t as toolWasteView, u as classifyFailureReason } from "./tool-waste-B9tdWV6g.js";
 import { i as pairedCohensDz, n as cohensD, r as interpretCliffs, t as cliffsDelta } from "./effect-sizes-DiH8MGOH.js";
 import { a as requiredSampleSize, c as holm, i as requiredPairedSampleSize, n as mcnemarRequiredN, o as benjaminiHochberg, r as pairedMde, s as bonferroni, t as mcnemarPower } from "./power-and-mde-B8F2RdcD.js";
 import { _ as passAtK, a as MANN_WHITNEY_EXACT_MAX_STATES, c as mannWhitneyU, d as isBinaryOutcomeVector, f as mcnemar, g as pairedRiskDifferenceScore, h as pairedRiskDifferenceExact, i as DEFAULT_PERMUTATIONS, l as wilcoxonSignedRank, m as pairedRiskDifference, n as pairArms, o as MANN_WHITNEY_EXACT_MAX_WORK, p as pairedBinaryScale, r as pairRunRecords, s as WILCOXON_EXACT_MAX_N, t as comparePairedArms, v as wilson } from "./paired-arms-D4aeIHUy.js";
 import { a as pairedSignTest, i as pairedDeltaTieFraction, n as DECISION_PAIRED_DELTA_STATISTIC, o as pairedTTest, r as pairedBootstrap, t as BOOTSTRAP_GATE_MIN_N } from "./paired-tests-C8iCsioC.js";
 import { t as eProcess } from "./sequential-eprocess-D1jKoihe.js";
 import { n as iqr, r as welchsTTest } from "./baseline-BC-eBZ7U.js";
-import { a as isToolSpan, i as isLlmSpan, r as isJudgeSpan, t as FAILURE_CLASSES } from "./schema-CdIX2aHu.js";
-import { d as runsForScenario, r as argHash, s as judgeSpans } from "./query-BPGMVlbM.js";
-import { i as analyzeRuns, o as checkCanaries, r as selfImprove, t as defineAgentEval } from "./define-agent-eval-jS8xj_Q_.js";
+import { a as isToolSpan, i as isLlmSpan, r as isJudgeSpan, t as FAILURE_CLASSES } from "./schema-CSf6qWgZ.js";
+import { d as runsForScenario, r as argHash, s as judgeSpans } from "./query-D1nLIKt7.js";
+import { i as analyzeRuns, o as checkCanaries, r as selfImprove, t as defineAgentEval } from "./define-agent-eval-D_i_s69h.js";
 import { r as observedSplitScore, t as isRealnessGated } from "./reward-nw2xZGZG.js";
-import { a as parseRunRecordSafe, c as validateRunRecord, i as modelHasSnapshot, n as UNKNOWN_MODEL, o as roundTripRunRecord, r as isRunRecord, s as runTaskScore, t as RunRecordValidationError } from "./run-record-DLORoL7t.js";
+import { a as parseRunRecordSafe, c as validateRunRecord, i as modelHasSnapshot, n as UNKNOWN_MODEL, o as roundTripRunRecord, r as isRunRecord, s as runTaskScore, t as RunRecordValidationError } from "./run-record-ZIsR9Fif.js";
 import { a as summaryTable, n as gainHistogram, r as paretoChart } from "./summary-report-Bgh8CpNK.js";
 import { n as contentHash, r as fileVerdictCache, t as canonicalJson } from "./verdict-cache-B3eCVQtY.js";
-import { $ as redTeamDataset, C as paretoFrontier, F as surfaceContentHash, K as buildReflectionPrompt, Q as DEFAULT_RED_TEAM_CORPUS, S as dominates, a as aggregateRunScore, ct as BackendIntegrityError, et as redTeamReport, i as transientDispatchFailure, it as runCampaign, lt as assertRealBackend, nt as runCanaries, o as clamp01, q as parseReflectionResponse, t as llmJudge, tt as scoreRedTeamOutput, ut as summarizeBackendIntegrity } from "./llm-judge-aQHIk5_-.js";
+import { $ as DEFAULT_RED_TEAM_CORPUS, C as dominates, I as surfaceContentHash, J as parseReflectionResponse, a as transientDispatchFailure, at as runCampaign, dt as summarizeBackendIntegrity, et as redTeamDataset, i as quotaExhaustedUntil, lt as BackendIntegrityError, nt as scoreRedTeamOutput, o as aggregateRunScore, q as buildReflectionPrompt, rt as runCanaries, s as clamp01, t as llmJudge, tt as redTeamReport, ut as assertRealBackend, w as paretoFrontier } from "./llm-judge-B2YxbAJb.js";
 import { a as resolveModelPricing, i as isModelPriced, n as estimateCost, r as estimateTokens, t as MODEL_PRICING } from "./metrics-Qv-cpptD.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-B1qx30B4.js";
 import { n as REDACTION_VERSION, r as redactString, t as DEFAULT_REDACTION_RULES } from "./redact-7Aq1ukl-.js";
@@ -31,28 +31,29 @@ import { n as equivalenceVerdict, t as certificationEvidenceDigest } from "./ver
 import { t as TraceEmitter } from "./emitter-DeQHiDMm.js";
 import { t as buildTrajectory } from "./trajectory-D_7rLrvE.js";
 import { t as runCounterfactual } from "./counterfactual-Bjq1mlUu.js";
-import { N as parseFindingSubject, c as createBoundedTraceAnalysisStore, f as DEFAULT_TRACE_ANALYST_BUDGETS, p as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, t as createTraceAnalyst } from "./kind-factory-DMeEoMQZ.js";
-import { _ as assertUniqueFindingIds, a as DEFAULT_TRACE_ANALYST_KINDS, b as snapshotAnalystRun, g as analystRunDigest, h as analystFindingDigest, l as FAILURE_MODE_KIND_SPEC, n as buildDefaultAnalystRegistry, r as AnalystRegistry, t as createChatClient, v as completedAnalystReviewQuality, x as validateAnalystReviewDecisions, y as readAnalystReview } from "./chat-client-DI79OPye.js";
+import { B as parseFindingSubject, H as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, c as createBoundedTraceAnalysisStore, f as DEFAULT_TRACE_ANALYST_BUDGETS, p as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, t as createTraceAnalyst } from "./kind-factory-gP6lDySe.js";
+import { _ as completedAnalystReviewQuality, b as validateAnalystReviewDecisions, c as FAILURE_MODE_KIND_SPEC, g as assertUniqueFindingIds, h as analystRunDigest, i as DEFAULT_TRACE_ANALYST_KINDS, m as analystFindingDigest, n as AnalystRegistry, t as buildDefaultAnalystRegistry, v as readAnalystReview, y as snapshotAnalystRun } from "./default-registry-B0bKikCb.js";
 import { OUTPUT_VALUE } from "./trace-attributes.js";
 import { r as extractUsageFromSse, t as extractUsage } from "./extract-usage-BrQ8mCLX.js";
 import { n as InMemoryRawProviderSink, r as NoopRawProviderSink, t as FileSystemRawProviderSink } from "./raw-provider-sink-BQd7mzyT.js";
-import { d as inferDomainKeywords, h as analyzeTraces, l as describeTraceInsightScope, m as tokenizeDomainWords, n as toolSpansToTraceAnalysisStore, o as buildTraceInsightContext, p as scoreTraceInsightReadiness, s as buildTraceInsightPrompt, u as domainEvidencePattern, w as captureFetchToRawSink, y as exportRunAsOtlp } from "./store-tool-spans-B9o6tU8f.js";
+import { d as inferDomainKeywords, h as analyzeTraces, l as describeTraceInsightScope, m as tokenizeDomainWords, n as toolSpansToTraceAnalysisStore, o as buildTraceInsightContext, p as scoreTraceInsightReadiness, s as buildTraceInsightPrompt, u as domainEvidencePattern, w as captureFetchToRawSink, y as exportRunAsOtlp } from "./store-tool-spans-CeNj_m2L.js";
 import { n as assertRunCaptured, t as RunIntegrityError } from "./integrity-Cy9WHAtb.js";
 import { n as InMemoryTraceStore, t as FileSystemTraceStore } from "./store-DNe_Uv1Q.js";
 import { t as packageVersion$1 } from "./package-version-D7lQHt_-.js";
-import { t as runBoundedProcess } from "./bounded-process-VIi0KSL2.js";
-import { t as runEvalCampaign } from "./eval-campaign-JDTeE6Pl.js";
+import { n as runBoundedProcess, t as DEFAULT_BOUNDED_PROCESS_TIMEOUT_MS } from "./bounded-process-BBZob7vl.js";
+import { t as runEvalCampaign } from "./eval-campaign-BeAjdhzC.js";
 import { a as computeExperimentStats, n as attest, r as verifyAttestation, s as improvementVerdict, t as ATTESTATION_ALGORITHM } from "./attestation-XSUpbc4o.js";
-import "./rollout-Crypdx8s.js";
-import { t as mintRolloutRows } from "./mint-DjfDUMHr.js";
+import "./rollout-C-znbbYg.js";
+import { t as mintRolloutRows } from "./mint-vWOdD8Ae.js";
 import { n as pairedEvalueSequence, t as evaluateInterimReleaseConfidence } from "./sequential-CzK5DarL.js";
-import { S as judgeFamily, _ as assertServedModels, b as CrossFamilyError, d as stripFencedJson, f as ModelSubstitutionError, h as assertServedModel, i as backoffMs, l as isTransientLlmError, m as assertCrossFamilyServed, o as costReceiptFromLlm, p as ServedCrossFamilyError, r as LlmResponseError, s as costReceiptFromLlmError, t as LlmCallError, u as maximumChargeForLlmRequest, v as checkServedModel, x as assertCrossFamily, y as servedModelAcceptable } from "./llm-client-BFMRpmqb.js";
-import { t as createDspyRlmTraceEngine } from "./dspy-rlm-engine-CS3qcCEk.js";
-import { a as diffFindings, n as runSemanticConceptJudge, r as FindingsStore, t as SEMANTIC_CONCEPT_JUDGE_VERSION } from "./semantic-concept-judge-I36eejJx.js";
+import { S as judgeFamily, _ as assertServedModels, b as CrossFamilyError, d as stripFencedJson, f as ModelSubstitutionError, h as assertServedModel, i as backoffMs, l as isTransientLlmError, m as assertCrossFamilyServed, o as costReceiptFromLlm, p as ServedCrossFamilyError, r as LlmResponseError, s as costReceiptFromLlmError, t as LlmCallError, u as maximumChargeForLlmRequest, v as checkServedModel, x as assertCrossFamily, y as servedModelAcceptable } from "./llm-client-CxQtdtd6.js";
+import { n as paidJsonChat } from "./chat-json-call-C26igCih.js";
+import { t as createDspyRlmTraceEngine } from "./dspy-rlm-engine-D5byiHn9.js";
+import { a as diffFindings, n as runSemanticConceptJudge, o as createChatTraceEngine, r as FindingsStore, t as SEMANTIC_CONCEPT_JUDGE_VERSION } from "./semantic-concept-judge-Ct3QU7t5.js";
 import { t as analyzeSeries } from "./series-convergence-CjO2QdRW.js";
-import { i as otlpTextToTraceAnalysisStore, n as OtlpFileTraceStore } from "./store-otlp-CHjBvWQY.js";
-import { t as paidJsonChat } from "./chat-json-call-5Jxna-aV.js";
-import { n as evaluateReleaseConfidence, r as bootstrapCi } from "./release-confidence-CzUHc4z4.js";
+import { i as otlpTextToTraceAnalysisStore, n as OtlpFileTraceStore } from "./store-otlp-Dow0pk_5.js";
+import { n as evaluateReleaseConfidence, r as bootstrapCi } from "./release-confidence-BsGEg_xg.js";
+import { t as createChatClient } from "./chat-client-Db4bqYfA.js";
 import { createHash } from "node:crypto";
 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";
@@ -1844,36 +1845,58 @@ var ProductClient = class {
 * SDK-specific Box / Sandbox types that don't belong in this package.
 */
 /**
-* Host-process runner. Uses node:child_process spawnSync (synchronous
-* under the hood — wrapped in a Promise to satisfy the interface). For
-* very long-running commands consider an async-spawn variant; this
-* shape matches VB's existing behavior and is fine for build/test/lint
-* subprocesses that finish in seconds-to-minutes.
+* Host-process runner, mapped onto {@link runBoundedProcess} so the package
+* has one spawn site rather than two.
+*
+* It used `spawnSync` and therefore had none of the bounds that runner exists
+* to provide: a `capMs` timeout killed the direct child and left any
+* descendant it had started running, a killed run could still report exit
+* code 0, captured output was unbounded, and the whole run blocked the event
+* loop. This runner spawns build, test and lint commands, which are exactly
+* the commands most likely to start descendants.
+*
+* It is not the package's last private spawn: `nodeProcessRunner`
+* (`src/trace-repair/docker-environment.ts`) and `runProcess`
+* (`src/campaign/external-optimizer-subprocess.ts`) still spawn their own
+* detached children, and several one-shot `spawnSync` git and CLI calls
+* remain. What this removes is the second spawn behind the `CommandRunner`
+* contract, which is the one that graded build, test and lint commands.
+*
+* Two behaviours the mapping changes on purpose:
+*   - a run with no `capMs` is now bounded by
+*     {@link LOCAL_COMMAND_RUNNER_DEFAULT_CAP_MS} instead of running forever;
+*   - the run no longer blocks the event loop, so a minutes-long `capMs` no
+*     longer stalls everything else in the process.
 */
+/**
+* Wall clock {@link localCommandRunner} applies when the caller names no
+* `capMs`. It is `runBoundedProcess`'s own default, stated here because this
+* runner used to be unbounded: a consumer whose command runs longer names its
+* own `capMs`, and a run the bound stops says so with `timedOut: true`.
+*/
+const LOCAL_COMMAND_RUNNER_DEFAULT_CAP_MS = DEFAULT_BOUNDED_PROCESS_TIMEOUT_MS;
 const localCommandRunner = {
 	name: "local",
 	async run(input) {
-		const start = Date.now();
-		const res = spawnSync(input.cmd, input.argv, {
-			cwd: input.cwd,
-			encoding: "utf8",
-			timeout: input.capMs,
+		const result = await runBoundedProcess({
+			command: input.cmd,
+			args: input.argv,
+			envMode: "merge",
 			env: {
-				...process.env,
 				CI: "1",
 				...input.env ?? {}
 			},
-			input: input.stdin
+			...input.cwd === void 0 ? {} : { cwd: input.cwd },
+			...input.capMs === void 0 ? {} : { timeoutMs: input.capMs },
+			...input.stdin === void 0 ? {} : { stdin: input.stdin }
 		});
-		const durationMs = Date.now() - start;
-		const timedOut = !!(res.error && "code" in res.error && res.error.code === "ETIMEDOUT");
 		return {
-			status: res.status ?? null,
-			stdout: (res.stdout ?? "").toString(),
-			stderr: (res.stderr ?? "").toString(),
-			durationMs,
-			timedOut,
-			runnerError: res.error && !timedOut ? String(res.error.message ?? res.error) : void 0
+			status: result.killedByTimeout || result.killedBySignal || !result.spawned ? null : result.exitCode,
+			stdout: result.stdout,
+			stderr: result.stderr,
+			durationMs: result.wallMs,
+			timedOut: result.killedByTimeout,
+			...result.runnerError === void 0 ? {} : { runnerError: result.runnerError }
 		};
 	},
 	async hasBin(name) {
@@ -7943,6 +7966,6 @@ function rankRows(rows, weights) {
 	})).sort((a, b) => b.mean - a.mean);
 }
 //#endregion
-export { AGENT_PROFILE_KINDS, ATTESTATION_ALGORITHM, AgentEvalError, AnalystRegistry, BOOTSTRAP_GATE_MIN_N, BackendIntegrityError, BudgetBreachError, BudgetGuard, CODING_HARNESSES, ConfigError, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, CostLedger, CostLedgerPersistenceError, CostReceiptCaptureError, CostReservationExceededError, CostTracker, CrossFamilyError, DECISION_PAIRED_DELTA_STATISTIC, DEFAULT_PERMUTATIONS, DEFAULT_REDACTION_RULES, DEFAULT_RED_TEAM_CORPUS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, ERROR_COUNT_PATTERNS, EquivalenceProtocolError, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, FileSystemTraceStore, FindingsStore, HARNESS_NATIVE_MODEL, HeldOutGate, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, InMemoryTraceStore, JudgeError, LlmCallError, LlmResponseError, MANN_WHITNEY_EXACT_MAX_STATES, MANN_WHITNEY_EXACT_MAX_WORK, MODEL_PRICING, ModelSubstitutionError, MultiLayerVerifier, NoopRawProviderSink, NotFoundError, OUTPUT_VALUE, OtlpFileTraceStore, POLICY_EDIT_AXES, POLICY_EDIT_CANDIDATE_RECORD_SCHEMA, POLICY_EDIT_TARGET_SURFACES, PairwiseSteeringOptimizer, PolicyEditValidationError, ProductClient, PromptRegistry, REDACTION_VERSION, RunIntegrityError, RunRecordValidationError, SEMANTIC_CONCEPT_JUDGE_VERSION, ServedCrossFamilyError, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TraceEmitter, UNKNOWN_MODEL, VERIFICATION_STRATEGIES, VERIFICATION_STRATEGY_SOURCES, ValidationError, WILCOXON_EXACT_MAX_N, acquisitionPlansForKnowledgeGaps, admitPolicyEdit, agentProfileCellHashMaterial, agentProfileCellKey, agentProfileHash, agentProfileId, aggregateJudgeVerdicts, aggregateRunScore, analystFindingDigest, analystRunDigest, analystRunToFeedbackTrajectory, analystRunToReviewRequests, analyzeAntiSlop, analyzeRuns, analyzeSeries, analyzeTraces, applyPolicyEditToSurface, argHash, assertCapabilityHeadroom, assertCrossFamily, assertCrossFamilyServed, assertNoHiddenLeak, assertNoJudgeVerdict, assertProductBenchmarkRun, assertRealBackend, assertRunCaptured, assertServedModel, assertServedModels, assertSingleBackend, assignFeedbackSplit, attest, benjaminiHochberg, blendHeldout, blockingKnowledgeEval, bonferroni, bootstrapCi, budgetBreachView, buildAgentProfileCell, buildDefaultAnalystRegistry, buildEquivalenceRecord, buildReflectionPrompt, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, calibrateJudge, calibrateJudgeContinuous, canonicalJson, capabilityHeadroom, captureFetchToRawSink, certificationEvidenceDigest, checkCanaries, checkServedModel, checkTraceContracts, clamp01, classifyFailure, cliffsDelta, cohensD, comparePairedArms, completionVerdict, computeExperimentStats, computeFindingId, computePolicyEditId, computeToolUseMetrics, confidenceInterval, contentHash, continuousAgreement, controlRunToFeedbackTrajectory, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, costForTokenPricing, costForUsage, costReceiptFromLlm, costReceiptFromLlmError, createAntiSlopJudge, createBoundedTraceAnalysisStore, createChatClient, createDspyRlmTraceEngine, createFeedbackTrajectory, createLlmCorrectnessChecker, createLlmReviewer, createTraceAnalyst, decidePairedPromotion, defaultBlendWeights, defineAgentEval, defineEquivalenceCheck, deployGateLayer, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, domainEvidencePattern, dominates, eProcess, ensembleJudge, equivalenceVerdict, errorStreakDetector, estimateCost, estimateTokens, evaluateActionPolicy, evaluateInterimReleaseConfidence, evaluateOracles, evaluateReleaseConfidence, expandProfileAxes, exportProductBenchmark, exportProductBenchmarkRuns, exportRunAsOtlp, extractErrorCount, extractProducedState, extractUsage, extractUsageFromSse, failureClusterView, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToOptimizerRow, fileVerdictCache, formatScorecardDiff, gainHistogram, gateTreatmentApplied, gradeOnHidden, gradeSemanticStatus, groupRunsByAgentProfileCell, harnessAxisOf, hashContent, hashJson, hiddenGrade, holm, improvementVerdict, inMemoryReviewStore, inferDomainKeywords, interRaterReliability, interpretCliffs, iqr, isBinaryOutcomeVector, isJudgeSpan, isLlmSpan, isModelPriced, isPolicyEdit, isRunRecord, isToolSpan, isTransientLlmError, jsonShape, jsonlReviewStore, jsonlRunRecordBackend, judgeAgreementView, judgeFamily, judgeSpans, knowledgeReadinessTracePayload, leaderboard, llmJudge, loadScorecard, localCommandRunner, makeFinding, makePolicyEdit, makePolicyEditCandidateRecord, makeProposalFinding, manifestContentDigest, mannWhitneyU, maximumChargeForLlmRequest, mcnemar, mcnemarPower, mcnemarRequiredN, minimumPairsForPairedDeltaTest, mintRolloutRows, modelHasSnapshot, modelPriceKey, mulberry32, notBlocked, objectiveEval, observeAll, otlpTextToTraceAnalysisStore, pairArms, pairRunRecords, pairedBinaryScale, pairedBootstrap, pairedCohensDz, pairedDeltaTest, pairedDeltaTieFraction, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedRiskDifferenceExact, pairedRiskDifferenceScore, pairedSignTest, pairedTTest, paretoChart, paretoFrontier, parseReflectionResponse, parseRunRecordSafe, partialCredit, partitionHeldOut, passAtK, pearsonR, policyEditFromFinding, policyEditsFromFindings, preflightModels, productBenchmarkRepoIdentity, profile_exports as profile, projectRuntimeTrajectoryEvidence, proposeSynthesisTargets, ranks, readProductBenchmarkManifest, recordRuns, recordRunsToScorecard, redTeamDataset, redTeamReport, redactString, regexMatches, renderPreferenceMemoryMarkdown, repeatedActionDetector, requiredPairedSampleSize, requiredSampleSize, resolveModelPricing, resolveSeat, roundTripRunRecord, routeFields, runAgentControlLoop, runBoundedProcess, runCampaign, runCanaries, runCounterfactual, runEquivalenceCheck, runEvalCampaign, runIntentMatchJudge, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runProposeReview, runProposeReviewAsControlLoop, runSemanticConceptJudge, runTaskScore, runsForScenario, scoreKnowledgeReadiness, scorePolicyEditReadiness, scoreRedTeamOutput, scoreTraceInsightReadiness, seatPresets, selfImprove, servedModelAcceptable, spearmanR, stripFencedJson, subjectiveEval, summarizeBackendIntegrity, summarizeNumberSeries, summarizePreferenceMemory, summaryTable, textInSnapshot, toAgentProfileJson, tokenizeDomainWords, toolSpansToTraceAnalysisStore, toolWasteView, traceContract, transientDispatchFailure, urlContains, userQuestionsForKnowledgeGaps, validatePolicyEdit, validatePolicyEditCandidateRecord, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyAttestation, verifyCompletion, viteDeployRunner, weightedComposite, weightedMean, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit, withHeldoutBlend, withJudgeRetry, wranglerDeployRunner };
+export { AGENT_PROFILE_KINDS, ATTESTATION_ALGORITHM, AgentEvalError, AnalystRegistry, BOOTSTRAP_GATE_MIN_N, BackendIntegrityError, BudgetBreachError, BudgetGuard, CODING_HARNESSES, ConfigError, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, CostLedger, CostLedgerPersistenceError, CostReceiptCaptureError, CostReservationExceededError, CostTracker, CrossFamilyError, DECISION_PAIRED_DELTA_STATISTIC, DEFAULT_BOUNDED_PROCESS_TIMEOUT_MS, DEFAULT_FAILURE_REASON_RULES, DEFAULT_PERMUTATIONS, DEFAULT_REDACTION_RULES, DEFAULT_RED_TEAM_CORPUS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, ERROR_COUNT_PATTERNS, EquivalenceProtocolError, FAILURE_BLAME, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, FileSystemTraceStore, FindingsStore, HARNESS_NATIVE_MODEL, HeldOutGate, INFRA_FAILURE_BLAMES, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, InMemoryTraceStore, JudgeError, LOCAL_COMMAND_RUNNER_DEFAULT_CAP_MS, LlmCallError, LlmResponseError, MANN_WHITNEY_EXACT_MAX_STATES, MANN_WHITNEY_EXACT_MAX_WORK, MODEL_PRICING, ModelSubstitutionError, MultiLayerVerifier, NoopRawProviderSink, NotFoundError, OUTPUT_VALUE, OtlpFileTraceStore, POLICY_EDIT_AXES, POLICY_EDIT_CANDIDATE_RECORD_SCHEMA, POLICY_EDIT_TARGET_SURFACES, PairwiseSteeringOptimizer, PolicyEditValidationError, ProductClient, PromptRegistry, REDACTION_VERSION, RunIntegrityError, RunRecordValidationError, SEMANTIC_CONCEPT_JUDGE_VERSION, ServedCrossFamilyError, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TraceEmitter, UNKNOWN_MODEL, VERIFICATION_STRATEGIES, VERIFICATION_STRATEGY_SOURCES, ValidationError, WILCOXON_EXACT_MAX_N, acquisitionPlansForKnowledgeGaps, admitPolicyEdit, agentProfileCellHashMaterial, agentProfileCellKey, agentProfileHash, agentProfileId, aggregateJudgeVerdicts, aggregateRunScore, analystFindingDigest, analystRunDigest, analystRunToFeedbackTrajectory, analystRunToReviewRequests, analyzeAntiSlop, analyzeRuns, analyzeSeries, analyzeTraces, applyPolicyEditToSurface, argHash, assertCapabilityHeadroom, assertCrossFamily, assertCrossFamilyServed, assertNoHiddenLeak, assertNoJudgeVerdict, assertProductBenchmarkRun, assertRealBackend, assertRunCaptured, assertServedModel, assertServedModels, assertSingleBackend, assignFeedbackSplit, attest, benjaminiHochberg, blendHeldout, blockingKnowledgeEval, bonferroni, bootstrapCi, budgetBreachView, buildAgentProfileCell, buildDefaultAnalystRegistry, buildEquivalenceRecord, buildReflectionPrompt, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, calibrateJudge, calibrateJudgeContinuous, canonicalJson, capabilityHeadroom, captureFetchToRawSink, certificationEvidenceDigest, checkCanaries, checkServedModel, checkTraceContracts, clamp01, classifyFailure, classifyFailureReason, cliffsDelta, cohensD, comparePairedArms, completionVerdict, computeExperimentStats, computeFindingId, computePolicyEditId, computeToolUseMetrics, confidenceInterval, contentHash, continuousAgreement, controlRunToFeedbackTrajectory, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, costForTokenPricing, costForUsage, costReceiptFromLlm, costReceiptFromLlmError, createAntiSlopJudge, createBoundedTraceAnalysisStore, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createFeedbackTrajectory, createLlmCorrectnessChecker, createLlmReviewer, createTraceAnalyst, decidePairedPromotion, defaultBlendWeights, defineAgentEval, defineEquivalenceCheck, deployGateLayer, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, domainEvidencePattern, dominates, eProcess, ensembleJudge, equivalenceVerdict, errorStreakDetector, estimateCost, estimateTokens, evaluateActionPolicy, evaluateInterimReleaseConfidence, evaluateOracles, evaluateReleaseConfidence, expandProfileAxes, exportProductBenchmark, exportProductBenchmarkRuns, exportRunAsOtlp, extractErrorCount, extractProducedState, extractUsage, extractUsageFromSse, failureBlame, failureClusterView, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToOptimizerRow, fileVerdictCache, formatScorecardDiff, gainHistogram, gateTreatmentApplied, gradeOnHidden, gradeSemanticStatus, groupRunsByAgentProfileCell, harnessAxisOf, hashContent, hashJson, hiddenGrade, holm, improvementVerdict, inMemoryReviewStore, inferDomainKeywords, interRaterReliability, interpretCliffs, iqr, isBinaryOutcomeVector, isJudgeSpan, isLlmSpan, isModelPriced, isPolicyEdit, isRunRecord, isToolSpan, isTransientLlmError, jsonShape, jsonlReviewStore, jsonlRunRecordBackend, judgeAgreementView, judgeFamily, judgeSpans, knowledgeReadinessTracePayload, leaderboard, llmJudge, loadScorecard, localCommandRunner, makeFinding, makePolicyEdit, makePolicyEditCandidateRecord, makeProposalFinding, manifestContentDigest, mannWhitneyU, maximumChargeForLlmRequest, mcnemar, mcnemarPower, mcnemarRequiredN, minimumPairsForPairedDeltaTest, mintRolloutRows, modelHasSnapshot, modelPriceKey, mulberry32, notBlocked, objectiveEval, observeAll, otlpTextToTraceAnalysisStore, pairArms, pairRunRecords, pairedBinaryScale, pairedBootstrap, pairedCohensDz, pairedDeltaTest, pairedDeltaTieFraction, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedRiskDifferenceExact, pairedRiskDifferenceScore, pairedSignTest, pairedTTest, paretoChart, paretoFrontier, parseReflectionResponse, parseRunRecordSafe, partialCredit, partitionHeldOut, passAtK, pearsonR, policyEditFromFinding, policyEditsFromFindings, preflightModels, productBenchmarkRepoIdentity, profile_exports as profile, projectRuntimeTrajectoryEvidence, proposeSynthesisTargets, quotaExhaustedUntil, ranks, readProductBenchmarkManifest, recordRuns, recordRunsToScorecard, redTeamDataset, redTeamReport, redactString, regexMatches, renderPreferenceMemoryMarkdown, repeatedActionDetector, requiredPairedSampleSize, requiredSampleSize, resolveModelPricing, resolveSeat, roundTripRunRecord, routeFields, runAgentControlLoop, runBoundedProcess, runCampaign, runCanaries, runCounterfactual, runEquivalenceCheck, runEvalCampaign, runIntentMatchJudge, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runProposeReview, runProposeReviewAsControlLoop, runSemanticConceptJudge, runTaskScore, runsForScenario, scoreKnowledgeReadiness, scorePolicyEditReadiness, scoreRedTeamOutput, scoreTraceInsightReadiness, seatPresets, selfImprove, servedModelAcceptable, spearmanR, stripFencedJson, subjectiveEval, summarizeBackendIntegrity, summarizeNumberSeries, summarizePreferenceMemory, summaryTable, textInSnapshot, toAgentProfileJson, tokenizeDomainWords, toolSpansToTraceAnalysisStore, toolWasteView, traceContract, transientDispatchFailure, urlContains, userQuestionsForKnowledgeGaps, validatePolicyEdit, validatePolicyEditCandidateRecord, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyAttestation, verifyCompletion, viteDeployRunner, weightedComposite, weightedMean, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit, withHeldoutBlend, withJudgeRetry, wranglerDeployRunner };
 
 //# sourceMappingURL=index.js.map