@tangle-network/agent-eval 0.221.8 → 0.222.0

package/dist/index.js

@@ -1,70 +1,72 @@
-import { t as __exportAll } from "./rolldown-runtime-8H4AJuhK.js";
-import { i as JudgeError, n as CaptureIntegrityError, o as NotFoundError, r as ConfigError, s as ValidationError, t as AgentEvalError } from "./errors-BcOs5MKW.js";
-import { a as hashCanonical, r as canonicalString } from "./canonical-Dq2hOQ8L.js";
-import { a as HARNESS_NATIVE_MODEL, c as agentProfileModelId, i as CODING_HARNESSES, l as expandProfileAxes, n as createLlmCorrectnessChecker, o as agentProfileHash, r as verifyCompletion, s as agentProfileId, t as completionVerdict, u as harnessAxisOf } from "./completion-verifier-asaV1nSj.js";
-import { a as groupRunsByAgentProfileCell, c as verifyAgentProfileCell, i as buildAgentProfileCell, l as hashJson, o as toAgentProfileJson, r as agentProfileCellKey, s as validateAgentProfileCell, t as AGENT_PROFILE_KINDS, u as manifestContentDigest } from "./agent-profile-cell-B1V1gUZs.js";
-import { r as observedSplitScore, t as isRealnessGated } from "./reward-nw2xZGZG.js";
-import { i as isToolSpan, n as isJudgeSpan, r as isLlmSpan, t as FAILURE_CLASSES } from "./schema-Bt2vLFm3.js";
-import { a as parseRunRecordSafe, c as validateRunRecord, i as modelHasSnapshot, n as UNKNOWN_MODEL, o as runTaskScore, r as isRunRecord, t as RunRecordValidationError } from "./run-record-DhEJgoPz.js";
-import { a as cliffsDelta, c as pairedCohensDz, i as summaryTable, n as gainHistogram, o as cohensD, s as interpretCliffs } from "./summary-report-DHDWxe6p.js";
-import { a as wilcoxonSignedRank, i as mannWhitneyU, n as pairArms, r as pairRunRecords, t as comparePairedArms } from "./paired-arms-B2ttLc0_.js";
-import { t as interpolatedQuantile } from "./quantile--0yb07pD.js";
-import { a as medianInPlace, i as makeRng, t as assertFiniteSample } from "./internal-BTK91Bqt.js";
-import { t as mulberry32 } from "./random-DDDcC7YU.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-CVNy48rR.js";
-import { n as calibrateJudgeContinuous, o as verbosityBias, r as continuousAgreement, t as calibrateJudge } from "./judge-calibration-BkXMIWA4.js";
-import { n as corpusInterRaterAgreementFromJudgeScores, r as interRaterReliability, t as corpusInterRaterAgreement } from "./agreement-irr-BhAqjluA.js";
-import { a as welchsTTest, c as fisherExactTwoSided, d as studentTTwoSidedPValue, f as twoByTwoPValue, i as iqr, l as linearRegressionSlope, n as checkCanaries, o as chiSquared, s as chiSquaredPValue, t as analyzeRuns, u as normalTwoSidedPValue } from "./analyze-runs-CHmsykNJ.js";
-import { n as bonferroni, r as holm, t as benjaminiHochberg } from "./multiplicity-Sg0U7h8j.js";
-import { a as pairedRiskDifferenceScore, i as pairedRiskDifferenceExact, o as passAtK, r as pairedRiskDifference, s as wilson, t as mcnemar } from "./paired-binary-B_nejA44.js";
-import { i as pairedBootstrap, n as BOOTSTRAP_GATE_MIN_N, o as pairedSignTest, s as pairedTTest, t as mean$1 } from "./mean-BbA_JCzW.js";
-import { a as requiredSampleSize, i as requiredPairedSampleSize, n as mcnemarRequiredN, r as pairedMde, t as mcnemarPower } from "./power-and-mde-CQX-0weG.js";
-import { t as eProcess } from "./sequential-eprocess-DodI6RoF.js";
-import { i as pairedDeltaTest, n as pairedDecisionShape, r as minimumPairsForPairedDeltaTest, t as decidePairedPromotion } from "./paired-promotion-decision-D0SHb-ZL.js";
-import { c as runsForScenario, i as judgeSpans, t as argHash } from "./query-B_RZ36an.js";
-import { n as contentHash, r as fileVerdictCache, t as canonicalJson } from "./verdict-cache-CL_7tePD.js";
-import { G as runCanaries, H as redTeamDataset, Q as runCampaign, U as redTeamReport, V as DEFAULT_RED_TEAM_CORPUS, W as scoreRedTeamOutput, a as aggregateRunScore, i as transientDispatchFailure, o as clamp01, t as llmJudge } from "./llm-judge-Cybq29bL.js";
-import { a as resolveModelPricing, i as isModelPriced, n as estimateCost, r as estimateTokens, t as MODEL_PRICING } from "./metrics-BKaPBtIG.js";
-import { c as costForTokenPricing, i as CostCeilingReachedError, l as costForUsage, n as CostAccountingIncompleteError, t as CostLedger } from "./cost-ledger-BGZB6yY2.js";
-import { D as surfaceContentHash, M as assertRealBackend, N as summarizeBackendIntegrity, _ as parseReflectionResponse, a as ATTESTATION_ALGORITHM, g as buildReflectionPrompt, j as BackendIntegrityError, o as attest, s as verifyAttestation } from "./campaign-evidence-DunzcJGv.js";
-import { a as redact, c as shareAllowed, n as assessShareSafety, o as redactForShare, s as redactText, t as REDACTION_VERSION } from "./redact-mtkTiJVk.js";
-import { a as makeProposalFinding, i as makeFinding, r as computeFindingId } from "./types-CnWmj8NW.js";
-import { S as requiredPairsForPairedPromotion, n as paretoFrontier, t as dominates, x as pairedPromotionPower } from "./pareto-BiBnQXAE.js";
-import { a as LlmCallError, i as stripFencedJson, n as isRecord, o as LlmResponseError } from "./json-BhdsJaOQ.js";
-import { r as selfImprove, t as defineAgentEval } from "./define-agent-eval-CCbT7Kfr.js";
-import { n as equivalenceVerdict, t as certificationEvidenceDigest } from "./verdict-BGQmxQ8A.js";
-import { c as createBoundedTraceAnalysisStore, p as DEFAULT_TRACE_ANALYST_BUDGETS, t as createTraceAnalyst } from "./kind-factory-Dl0PmNQL.js";
-import { C as snapshotAnalystRun, S as readAnalystReview, b as assertUniqueFindingIds, c as FAILURE_MODE_KIND_SPEC, i as DEFAULT_TRACE_ANALYST_KINDS, n as AnalystRegistry, t as buildDefaultAnalystRegistry, v as analystFindingDigest, w as validateAnalystReviewDecisions, x as completedAnalystReviewQuality, y as analystRunDigest } from "./default-registry-BBTDgkDM.js";
+import { t as __exportAll } from "./rolldown-runtime-BJohB4LM.js";
+import { i as JudgeError, n as CaptureIntegrityError, o as NotFoundError, r as ConfigError, s as ValidationError, t as AgentEvalError } from "./errors-DeEEV9Vx.js";
+import { a as hashCanonical, r as canonicalString } from "./canonical-CicJlINA.js";
+import { a as agentProfileModelId, i as agentProfileId, n as HARNESS_NATIVE_MODEL, o as expandProfileAxes, r as agentProfileHash, s as harnessAxisOf, t as CODING_HARNESSES } from "./agent-profile-Dngk3LZ0.js";
+import { a as groupRunsByAgentProfileCell, c as verifyAgentProfileCell, i as buildAgentProfileCell, l as hashJson, o as toAgentProfileJson, r as agentProfileCellKey, s as validateAgentProfileCell, t as AGENT_PROFILE_KINDS, u as manifestContentDigest } from "./agent-profile-cell-CT5JYueM.js";
+import { r as observedSplitScore, t as isRealnessGated } from "./reward-P0VD_Aft.js";
+import { i as isToolSpan, n as isJudgeSpan, r as isLlmSpan, t as FAILURE_CLASSES } from "./schema-D-xu3MU_.js";
+import { a as parseRunRecordSafe, c as validateRunRecord, i as modelHasSnapshot, n as UNKNOWN_MODEL, o as runTaskScore, r as isRunRecord, t as RunRecordValidationError } from "./run-record-DEjx43cI.js";
+import { a as cliffsDelta, c as pairedCohensDz, i as summaryTable, n as gainHistogram, o as cohensD, s as interpretCliffs } from "./summary-report-CPfsrLac.js";
+import { t as interpolatedQuantile } from "./quantile-CncXP-Ss.js";
+import { a as wilcoxonSignedRank, i as mannWhitneyU, n as pairArms, r as pairRunRecords, t as comparePairedArms } from "./paired-arms-D-54hO-k.js";
+import { a as medianInPlace, i as makeRng, t as assertFiniteSample } from "./internal-v6XfOeJW.js";
+import { t as mulberry32 } from "./random-DrzcgC7F.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-cPwITVfy.js";
+import { n as calibrateJudgeContinuous, o as verbosityBias, r as continuousAgreement, t as calibrateJudge } from "./judge-calibration-DGqlfWW8.js";
+import { n as corpusInterRaterAgreementFromJudgeScores, r as interRaterReliability, t as corpusInterRaterAgreement } from "./agreement-irr-BTjDN0i6.js";
+import { a as welchsTTest, c as fisherExactTwoSided, d as studentTTwoSidedPValue, f as twoByTwoPValue, i as iqr, l as linearRegressionSlope, n as checkCanaries, o as chiSquared, s as chiSquaredPValue, t as analyzeRuns, u as normalTwoSidedPValue } from "./analyze-runs-CTheASYV.js";
+import { n as bonferroni, r as holm, t as benjaminiHochberg } from "./multiplicity-DAd4xaDC.js";
+import { a as pairedRiskDifferenceScore, i as pairedRiskDifferenceExact, o as passAtK, r as pairedRiskDifference, s as wilson, t as mcnemar } from "./paired-binary-BR0nwHOA.js";
+import { i as pairedBootstrap, n as BOOTSTRAP_GATE_MIN_N, o as pairedSignTest, s as pairedTTest, t as mean$1 } from "./mean-CdDYG-us.js";
+import { a as requiredSampleSize, i as requiredPairedSampleSize, n as mcnemarRequiredN, r as pairedMde, t as mcnemarPower } from "./power-and-mde-ClI6RWw4.js";
+import { t as eProcess } from "./sequential-eprocess-C1YLwFaz.js";
+import { i as pairedDeltaTest, n as pairedDecisionShape, r as minimumPairsForPairedDeltaTest, t as decidePairedPromotion } from "./paired-promotion-decision-FGvHA4p_.js";
+import { c as runsForScenario, i as judgeSpans, t as argHash } from "./query-CGXzO9d4.js";
+import { n as contentHash, r as fileVerdictCache, t as canonicalJson } from "./verdict-cache-BVH-fl18.js";
+import { B as DEFAULT_RED_TEAM_CORPUS, H as redTeamReport, U as scoreRedTeamOutput, V as redTeamDataset, W as runCanaries, Z as runCampaign, a as clamp01, i as aggregateRunScore, r as transientDispatchFailure, t as llmJudge } from "./llm-judge-cgbhTKQp.js";
+import { a as resolveModelPricing, i as isModelPriced, n as estimateCost, r as estimateTokens, t as MODEL_PRICING } from "./metrics-DNCAUtuB.js";
+import { c as costForTokenPricing, i as CostCeilingReachedError, l as costForUsage, n as CostAccountingIncompleteError, t as CostLedger } from "./cost-ledger-CS8R3IZj.js";
+import { C as assertRealBackend, S as BackendIntegrityError, _ as parseReflectionResponse, a as ATTESTATION_ALGORITHM, g as buildReflectionPrompt, o as attest, s as verifyAttestation, w as summarizeBackendIntegrity } from "./campaign-evidence-DN8rWuBj.js";
+import { s as surfaceContentHash } from "./json-recovery-DK4MDm5r.js";
+import { a as redact, c as shareAllowed, n as assessShareSafety, o as redactForShare, s as redactText, t as REDACTION_VERSION } from "./redact-D6APJS1d.js";
+import { a as makeProposalFinding, i as makeFinding, r as computeFindingId } from "./types-BK0blWhe.js";
+import { S as requiredPairsForPairedPromotion, n as paretoFrontier, t as dominates, x as pairedPromotionPower } from "./pareto-B9udAt2i.js";
+import { a as LlmCallError, i as stripFencedJson, n as isRecord, o as LlmResponseError } from "./json-DtvZNAU9.js";
+import { r as selfImprove, t as defineAgentEval } from "./define-agent-eval-BLoP4Lf9.js";
+import { n as equivalenceVerdict, t as certificationEvidenceDigest } from "./verdict-C8TcdwfD.js";
+import { c as createBoundedTraceAnalysisStore, p as DEFAULT_TRACE_ANALYST_BUDGETS, t as createTraceAnalyst } from "./kind-factory-VKysp1ds.js";
+import { C as snapshotAnalystRun, S as readAnalystReview, b as assertUniqueFindingIds, c as FAILURE_MODE_KIND_SPEC, i as DEFAULT_TRACE_ANALYST_KINDS, n as AnalystRegistry, t as buildDefaultAnalystRegistry, v as analystFindingDigest, w as validateAnalystReviewDecisions, x as completedAnalystReviewQuality, y as analystRunDigest } from "./default-registry-C1tPEjZ4.js";
 import { OUTPUT_VALUE } from "./trace-attributes.js";
-import { n as extractUsageFromSse, t as extractUsage } from "./extract-usage-BZLU01kR.js";
-import { n as InMemoryRawProviderSink, r as NoopRawProviderSink, t as FileSystemRawProviderSink } from "./raw-provider-sink-75w4FWBj.js";
-import { _ as exportRunAsOtlp, b as captureFetchToRawSink, c as describeTraceInsightScope, g as analyzeTraces, h as tokenizeDomainWords, l as domainEvidencePattern, m as scoreTraceInsightReadiness, n as toolSpansToTraceAnalysisStore, o as buildTraceInsightContext, s as buildTraceInsightPrompt, u as inferDomainKeywords } from "./store-tool-spans-BK_8FjGG.js";
-import { t as TraceEmitter } from "./emitter-DbMXUIVP.js";
-import { n as assertRunCaptured, t as RunIntegrityError } from "./integrity-W4CADcFc.js";
-import { n as InMemoryTraceStore, t as FileSystemTraceStore } from "./store-BhNuLdJw.js";
-import { t as packageVersion$1 } from "./package-version-D7lQHt_-.js";
-import { t as buildTrajectory } from "./trajectory-D_7rLrvE.js";
-import { i as failureBlame, n as classifyFailure, r as classifyFailureReason } from "./failure-taxonomy-Bx5oWqaW.js";
-import { n as runBoundedProcess, t as DEFAULT_BOUNDED_PROCESS_TIMEOUT_MS } from "./bounded-process-Cj9YjXbE.js";
-import { t as runEvalCampaign } from "./eval-campaign-DoX5vz3E.js";
-import "./rollout-Dd1ucT_M.js";
-import { t as mintRolloutRows } from "./mint-CbdwjTPW.js";
-import { n as pairedEvalueSequence, t as evaluateInterimReleaseConfidence } from "./sequential-BVv1sjsw.js";
-import { _ as judgeFamily, a as maximumChargeForLlmRequest, c as ModelSubstitutionError, f as assertServedModels, g as assertCrossFamily, h as CrossFamilyError, i as costReceiptFromLlmError, l as assertCrossFamilyServed, m as describeServedModelCheck, p as checkServedModel, r as costReceiptFromLlm, s as isTransientLlmError, u as assertServedModel } from "./llm-client-b1ysZVIU.js";
-import { n as paidJsonChat } from "./chat-json-call-BzWYp18J.js";
-import { t as createDspyRlmTraceEngine } from "./dspy-rlm-engine-DN1p_5jx.js";
-import { n as diffFindings, r as createChatTraceEngine, t as FindingsStore } from "./findings-store-Bg4sdS-e.js";
-import { a as budgetBreachView, i as failureClusterView, n as computeToolUseMetrics, r as judgeAgreementView, t as toolWasteView } from "./tool-waste-DLsjS_p-.js";
-import { _ as analyzeSeries, a as judgeGateDecision, c as snapshotFromJudgeAgreement, g as withJudgeRetry, i as assertJudgeMayGate, n as JudgeGateRefusedError, o as measureJudgeAgreement, r as addCalibrationVerdicts, s as registerCalibrationSet, t as DEFAULT_JUDGE_GATE_POLICY } from "./judge-gate-BVvmP00P.js";
-import { a as spanRecordsToTraceAnalysisStore, i as otlpTextToTraceAnalysisStore, r as OtlpFileTraceStore } from "./store-otlp-CFf27E8_.js";
-import { n as bootstrapCi, t as evaluateReleaseConfidence } from "./release-confidence-CYIFH-fp.js";
-import { t as createChatClient } from "./chat-client-BNATB3rJ.js";
+import { n as extractUsageFromSse, t as extractUsage } from "./extract-usage-ISFqhfDz.js";
+import { n as InMemoryRawProviderSink, r as NoopRawProviderSink, t as FileSystemRawProviderSink } from "./raw-provider-sink-DDLkP_eF.js";
+import { _ as exportRunAsOtlp, b as captureFetchToRawSink, c as describeTraceInsightScope, g as analyzeTraces, h as tokenizeDomainWords, l as domainEvidencePattern, m as scoreTraceInsightReadiness, n as toolSpansToTraceAnalysisStore, o as buildTraceInsightContext, s as buildTraceInsightPrompt, u as inferDomainKeywords } from "./store-tool-spans-C2g8UYm-.js";
+import { t as TraceEmitter } from "./emitter-DhQ65mza.js";
+import { n as assertRunCaptured, t as RunIntegrityError } from "./integrity-DOHt1mwU.js";
+import { n as InMemoryTraceStore, t as FileSystemTraceStore } from "./store-ZykIyQmy.js";
+import { t as packageVersion$1 } from "./package-version-CJl_pY9W.js";
+import { t as buildTrajectory } from "./trajectory-O3GDHmev.js";
+import { i as failureBlame, n as classifyFailure, r as classifyFailureReason } from "./failure-taxonomy-9eQfXKoW.js";
+import { n as runBoundedProcess, t as DEFAULT_BOUNDED_PROCESS_TIMEOUT_MS } from "./bounded-process-BZrWXIKW.js";
+import { t as runEvalCampaign } from "./eval-campaign-Cgixrvf7.js";
+import { _ as assertCrossFamily, a as maximumChargeForLlmRequest, d as assertServedModel, g as CrossFamilyError, h as describeServedModelCheck, i as costReceiptFromLlmError, l as ModelSubstitutionError, m as checkServedModel, p as assertServedModels, r as costReceiptFromLlm, s as isTransientLlmError, u as assertCrossFamilyServed, v as judgeFamily } from "./llm-client-kJ9EOqjt.js";
+import { n as createLlmCorrectnessChecker, r as verifyCompletion, t as completionVerdict } from "./completion-verifier-rqbMh5-4.js";
+import "./rollout-BhkMpAJT.js";
+import { t as mintRolloutRows } from "./mint-D9ZysYtY.js";
+import { n as pairedEvalueSequence, t as evaluateInterimReleaseConfidence } from "./sequential-Bjs5roiK.js";
+import { n as paidJsonChat } from "./chat-json-call-8yUv4LFO.js";
+import { t as createDspyRlmTraceEngine } from "./dspy-rlm-engine-jDiWkul6.js";
+import { n as diffFindings, r as createChatTraceEngine, t as FindingsStore } from "./findings-store-B75Dspaq.js";
+import { a as budgetBreachView, i as failureClusterView, n as computeToolUseMetrics, r as judgeAgreementView, t as toolWasteView } from "./tool-waste-CLJ46AbT.js";
+import { _ as analyzeSeries, a as judgeGateDecision, c as snapshotFromJudgeAgreement, g as withJudgeRetry, i as assertJudgeMayGate, n as JudgeGateRefusedError, o as measureJudgeAgreement, r as addCalibrationVerdicts, s as registerCalibrationSet, t as DEFAULT_JUDGE_GATE_POLICY } from "./judge-gate-C2BCOt-M.js";
+import { a as spanRecordsToTraceAnalysisStore, i as otlpTextToTraceAnalysisStore, r as OtlpFileTraceStore } from "./store-otlp-DmLm65XW.js";
+import { n as bootstrapCi, t as evaluateReleaseConfidence } from "./release-confidence-CTUJz5mM.js";
+import { t as createChatClient } from "./chat-client-CO3uB33i.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";
 import { spawnSync } from "node:child_process";
 import { INPUT_TOKEN_ATTR_KEYS, MODEL_ATTR_KEYS, OUTPUT_TOKEN_ATTR_KEYS, SPAN_KINDS, TOOL_NAME_ATTR_KEYS, firstNumberAttr, firstStringAttr, resolveSpanKind } from "@tangle-network/agent-trace-contract";
-//#region src/statistics/samples.ts
+// src/statistics/samples.ts
 /** Interpolated sample median; an empty population has no measured median. */
 function sampleMedian(values) {
 	assertFiniteSample("sampleMedian", "values", values);
@@ -138,8 +140,7 @@ function summarizeSample(values, options = {}) {
 	].some((value) => value !== null && !Number.isFinite(value))) throw new ValidationError("summarizeSample: sample arithmetic exceeded finite range");
 	return summary;
 }
-//#endregion
-//#region src/profile/baselines.ts
+// src/profile/baselines.ts
 /**
 * Strong, generically-useful baseline ROLES — the top zone of an `AgentProfile`
 * before any domain layer. A product composes one of these with its own
@@ -202,8 +203,7 @@ const BASELINE_ROLES = {
 	researcher: researcherRole,
 	generalist: generalistRole
 };
-//#endregion
-//#region src/profile/index.ts
+// src/profile/index.ts
 /**
 * Structured prompt profile — the system prompt as named, addressable sections
 * instead of one opaque blob. The self-improvement loop targets ONE evolvable
@@ -348,8 +348,7 @@ function sectionHash(section) {
 		body: section.body
 	}));
 }
-//#endregion
-//#region src/verification-strategy.ts
+// src/verification-strategy.ts
 /**
 * The family registry. `Record` over the union keeps it exhaustive: adding
 * a member to `VerificationStrategySource` without a profile here fails to
@@ -502,8 +501,7 @@ function assertObligation(obligation) {
 	if (separatingWitness !== void 0) throw new EquivalenceProtocolError("witness-on-unresolved", "an unresolved obligation cannot carry a separating witness — a witness in hand is a refutation");
 	if (typeof unresolvedReason !== "string" || unresolvedReason.trim() === "") throw new EquivalenceProtocolError("reason-missing", "an unresolved obligation must say why — 'unresolved' with no reason erases the diagnostic");
 }
-//#endregion
-//#region src/budget-guard.ts
+// src/budget-guard.ts
 /**
 * BudgetGuard — enforces token / wall-clock / call / $ caps, records
 * a ledger entry on every decrement, emits `budget_breach` + throws
@@ -570,8 +568,7 @@ var BudgetGuard = class {
 		return { ...this.consumed };
 	}
 };
-//#endregion
-//#region src/feedback-trajectory.ts
+// src/feedback-trajectory.ts
 const DEFAULT_SPLIT_POLICY = {
 	trainPct: 70,
 	devPct: 15,
@@ -1029,8 +1026,7 @@ function stableHash(input) {
 	}
 	return hash >>> 0;
 }
-//#endregion
-//#region src/scorecard.ts
+// src/scorecard.ts
 /**
 * @stable
 *
@@ -1047,12 +1043,6 @@ function stableHash(input) {
 * malformed line never breaks the read. `diffScorecard` compares the latest
 * entry of each cell against its predecessor with Cohen's d + Welch's t-test.
 */
-function median(xs) {
-	if (xs.length === 0) return 0;
-	const sorted = [...xs].sort((a, b) => a - b);
-	const mid = Math.floor(sorted.length / 2);
-	return sorted.length % 2 === 0 ? (sorted[mid - 1] + sorted[mid]) / 2 : sorted[mid];
-}
 /** Mean of each judge dimension across the runs that reported one. */
 function aggregatePerDimension(runs) {
 	const sums = /* @__PURE__ */ new Map();
@@ -1105,7 +1095,7 @@ function recordRuns(runs, opts) {
 			commitSha: opts.commitSha,
 			timestamp,
 			scores,
-			composite: median(scores),
+			composite: medianInPlace([...scores]),
 			runIds: scored.map((s) => s.run.runId)
 		};
 		const perDimension = aggregatePerDimension(scenarioRuns);
@@ -1264,8 +1254,7 @@ function formatScorecardDiff(diff) {
 	}
 	return lines.join("\n");
 }
-//#endregion
-//#region src/trace-contracts/spans.ts
+// src/trace-contracts/spans.ts
 const EVAL_KIND_TO_CONTRACT_KIND = /* @__PURE__ */ new Map([["judge", "EVALUATOR"], ["retrieval", "RETRIEVER"]]);
 /**
 * The span's contract kind: a declared kind when recognised, else inferred.
@@ -1393,8 +1382,7 @@ function finite(value) {
 function isSpanError(span) {
 	return typeof span.status === "string" && /^(?:error|status_code_error)$/i.test(span.status);
 }
-//#endregion
-//#region src/trace-contracts/predicates.ts
+// src/trace-contracts/predicates.ts
 function isSerializedRegex(v) {
 	return typeof v === "object" && v !== null && typeof v.$regex === "string" && typeof v.flags === "string";
 }
@@ -1486,8 +1474,7 @@ function normalizePredicate(p, where) {
 	}
 	return out;
 }
-//#endregion
-//#region src/trace-contracts/types.ts
+// src/trace-contracts/types.ts
 const ORDER_MODES = [
 	"start-order",
 	"finish-before-start",
@@ -1513,8 +1500,7 @@ const ARGUMENT_TYPES = [
 	"array",
 	"null"
 ];
-//#endregion
-//#region src/trace-contracts/validate.ts
+// src/trace-contracts/validate.ts
 function assertOrder(order, where) {
 	if (order === void 0) return void 0;
 	if (!ORDER_MODES.includes(order)) throw new ValidationError(`${where}: order must be one of ${ORDER_MODES.join(", ")}`);
@@ -1638,8 +1624,7 @@ function assertContract(contract, where) {
 		}
 	}
 }
-//#endregion
-//#region src/trace-contracts/explain.ts
+// src/trace-contracts/explain.ts
 function describeMatcher(m) {
 	if (typeof m === "string") return m;
 	if (m instanceof RegExp) return `/${m.source}/${m.flags}`;
@@ -1708,8 +1693,7 @@ function explainTraceContract(contract) {
 	}
 	return lines.join("\n");
 }
-//#endregion
-//#region src/trace-contracts/builder.ts
+// src/trace-contracts/builder.ts
 var TraceContractBuilder = class TraceContractBuilder {
 	name;
 	rules = [];
@@ -1887,8 +1871,7 @@ function traceContract(name) {
 	if (typeof name !== "string" || name.length === 0) throw new ValidationError("traceContract: name must be a non-empty string");
 	return new TraceContractBuilder(name);
 }
-//#endregion
-//#region src/trace-contracts/rules.ts
+// src/trace-contracts/rules.ts
 function spanRef(span, index) {
 	return span.spanId ?? `#${index}`;
 }
@@ -2365,8 +2348,7 @@ function executeRule(rule, ctx) {
 		};
 	}
 }
-//#endregion
-//#region src/trace-contracts/evaluate.ts
+// src/trace-contracts/evaluate.ts
 /**
 * A run derived from one span: the `run.status` attribute a run-anchor span
 * carries (store-to-otlp writes it), else failed on an error status, else
@@ -2536,8 +2518,7 @@ function checkTraceContracts(spans, contracts, options = {}) {
 		allValid: status === "pass"
 	};
 }
-//#endregion
-//#region src/trace-contracts/spec.ts
+// src/trace-contracts/spec.ts
 /**
 * The declarative contract form: a JSON document that names tools, model
 * calls, and run limits, compiled onto the temporal operators.
@@ -3113,8 +3094,7 @@ function lintPath(path, prefix, out) {
 	else if (run?.requireCompleted && run.allowedStatuses?.some((s) => s !== "completed")) warn("run.statuses-beside-completed", "run.allowedStatuses", "requireCompleted accepts only completed, so the other allowed statuses never apply");
 	if (path.llm?.maxCalls === 0 && (path.llm.maxTotalTokens !== void 0 || path.llm.allowedModels)) warn("llm.no-calls", "llm", "maxCalls is 0, so the token and model limits never apply");
 }
-//#endregion
-//#region src/active-learning.ts
+// src/active-learning.ts
 async function proposeSynthesisTargets(dataset, traceStore, options = {}) {
 	const minPerBand = options.minPerBand ?? 5;
 	const varianceThreshold = options.varianceThreshold ?? .05;
@@ -3142,8 +3122,8 @@ async function proposeSynthesisTargets(dataset, traceStore, options = {}) {
 	const runs = await traceStore.listRuns();
 	const runCountByScenario = /* @__PURE__ */ new Map();
 	for (const r of runs) runCountByScenario.set(r.scenarioId, (runCountByScenario.get(r.scenarioId) ?? 0) + 1);
-	const runCounts = [...runCountByScenario.values()];
-	const p25 = runCounts.length > 0 ? quantile(runCounts, .25) : 0;
+	const runCounts = [...runCountByScenario.values()].sort((a, b) => a - b);
+	const p25 = runCounts.length > 0 ? interpolatedQuantile(runCounts, .25) : 0;
 	for (const s of scenarios) {
 		const count = runCountByScenario.get(s.id) ?? 0;
 		if (count <= p25 && count < 3) targets.push({
@@ -3196,15 +3176,7 @@ async function proposeSynthesisTargets(dataset, traceStore, options = {}) {
 	}
 	return targets.sort((a, b) => b.priority - a.priority).slice(0, topK);
 }
-function quantile(xs, p) {
-	const sorted = [...xs].sort((a, b) => a - b);
-	const idx = p * (sorted.length - 1);
-	const lo = Math.floor(idx);
-	const hi = Math.ceil(idx);
-	return sorted[lo] + (sorted[hi] - sorted[lo]) * (idx - lo);
-}
-//#endregion
-//#region src/deprecation.ts
+// src/deprecation.ts
 /**
 * One-shot deprecation warnings. A deprecated surface warns the FIRST time it
 * is used in a process — loud enough to be seen, quiet enough not to flood a
@@ -3217,8 +3189,7 @@ function warnDeprecatedOnce(key, message) {
 	warned.add(key);
 	console.warn(`[agent-eval deprecation] ${message}`);
 }
-//#endregion
-//#region src/client.ts
+// src/client.ts
 const DEFAULT_REQUEST_TIMEOUT_MS = 3e4;
 /**
 * ProductClient — configurable HTTP client for exercising any agent's APIs.
@@ -3439,8 +3410,7 @@ var ProductClient = class {
 		return this.request("PATCH", path, body);
 	}
 };
-//#endregion
-//#region src/command-runner.ts
+// src/command-runner.ts
 /**
 * CommandRunner — abstract subprocess execution surface.
 *
@@ -3551,8 +3521,7 @@ function isExecutable(path) {
 		throw err;
 	}
 }
-//#endregion
-//#region src/control-runtime/policy.ts
+// src/control-runtime/policy.ts
 const DEFAULT_BUDGET = {
 	maxSteps: 8,
 	maxWallMs: 3e5
@@ -3622,8 +3591,7 @@ function budgetStopDecision(budget, spentCostUsd) {
 		reason: ""
 	};
 }
-//#endregion
-//#region src/detectors/index.ts
+// src/detectors/index.ts
 /** Same action fingerprint N consecutive steps = a stuck loop (the #1 long-horizon failure mode). */
 function repeatedActionDetector(opts = {}) {
 	const max = opts.maxRepeated ?? 3;
@@ -3734,8 +3702,7 @@ function observeAll(detectors, event) {
 	}
 	return signals;
 }
-//#endregion
-//#region src/control-runtime.ts
+// src/control-runtime.ts
 async function runAgentControlLoop(config) {
 	const budget = normalizeBudget(config.budget);
 	const controller = new AbortController();
@@ -4261,8 +4228,7 @@ function normalizeActionCostUsd(costUsd, runtimeErrors, stepIndex) {
 	}
 	return costUsd;
 }
-//#endregion
-//#region src/discover-personas.ts
+// src/discover-personas.ts
 /**
 * Walk a personas directory and return every file matching the convention
 * `NN-slug.{yaml,yml,json,md}`. Sorted by filename so the numeric prefix
@@ -4310,8 +4276,7 @@ async function discoverPersonas(dir, opts = {}) {
 	results.sort((a, b) => a.filename.localeCompare(b.filename));
 	return results;
 }
-//#endregion
-//#region src/propose-review.ts
+// src/propose-review.ts
 /**
 * Propose / Verify / Review — the core multi-shot primitive.
 *
@@ -4620,8 +4585,7 @@ function safeJson(x) {
 		return String(x);
 	}
 }
-//#endregion
-//#region src/analyst/policy-edit.ts
+// src/analyst/policy-edit.ts
 const POLICY_EDIT_AXES = [
 	"carrier",
 	"representation",
@@ -4839,8 +4803,7 @@ function expectNonEmptyStringArray(value, path) {
 	expectStringArray$1(value, path);
 	if (value.length === 0) throw new PolicyEditValidationError("expected non-empty array", path);
 }
-//#endregion
-//#region src/capability-headroom.ts
+// src/capability-headroom.ts
 /**
 * Capability-headroom gate — "can this task set even SEE the capability
 * you're about to A/B?"
@@ -4936,8 +4899,7 @@ function assertCapabilityHeadroom(result, opts = {}) {
 		throw new ValidationError(`assertCapabilityHeadroom: only ${tasksWithGap} of ${total} task(s) have baseline headroom (need ≥ ${minTasksWithGap}). ${tasksSaturated} saturated (baseline already passes), ${tasksUnknown} unknown (no determinable baseline outcome). A capability A/B on this task set cannot detect the capability — add tasks the baseline fails, or fix baseline outcome capture, before measuring.`);
 	}
 }
-//#endregion
-//#region src/cost-tracker.ts
+// src/cost-tracker.ts
 /**
 * Cost tracker — token + USD accounting per scenario and per run.
 *
@@ -5036,8 +4998,7 @@ function costFor(entry) {
 function assertNonNegative(n, name) {
 	if (!Number.isFinite(n) || n < 0) throw new Error(`CostTracker: ${name} must be a non-negative finite number, got ${n}`);
 }
-//#endregion
-//#region src/error-count-extractor.ts
+// src/error-count-extractor.ts
 const ERROR_COUNT_PATTERNS = [
 	{
 		name: "typescript-tsc",
@@ -5094,8 +5055,7 @@ function extractErrorCount(text, opts = {}) {
 		samples: []
 	};
 }
-//#endregion
-//#region src/knowledge/readiness.ts
+// src/knowledge/readiness.ts
 function scoreKnowledgeReadiness(options) {
 	const now = options.now ?? /* @__PURE__ */ new Date();
 	const requirements = options.requirements.map(normalizeRequirement);
@@ -5274,8 +5234,7 @@ function clamp01$2(value) {
 function unique(items) {
 	return [...new Set(items)];
 }
-//#endregion
-//#region src/leaderboard.ts
+// src/leaderboard.ts
 function mean(xs) {
 	return xs.length ? xs.reduce((a, b) => a + b, 0) / xs.length : null;
 }
@@ -5324,8 +5283,7 @@ function leaderboard(records, opts) {
 function isFiniteNumber(value) {
 	return typeof value === "number" && Number.isFinite(value);
 }
-//#endregion
-//#region src/partition-held-out.ts
+// src/partition-held-out.ts
 /**
 * Deterministic held-out partitioning over stable ids.
 *
@@ -5403,8 +5361,7 @@ function partitionHeldOut(ids, options = {}) {
 function assertFraction(f) {
 	if (!Number.isFinite(f) || f <= 0 || f >= 1) throw new ValidationError(`partitionHeldOut: holdoutFraction must be in (0, 1), got ${f}`);
 }
-//#endregion
-//#region src/product-benchmark/export.ts
+// src/product-benchmark/export.ts
 /**
 * Export side of the product benchmark bundle contract: convert product
 * eval run directories (`records.jsonl` of `RunRecord` rows + trace/raw
@@ -5937,8 +5894,7 @@ function resolveOptions(options) {
 		backendVersion: options.backendVersion ?? packageVersion("@tangle-network/sandbox")
 	};
 }
-//#endregion
-//#region src/product-benchmark/index.ts
+// src/product-benchmark/index.ts
 const productBenchmarkSplits = [
 	"practice",
 	"dev",
@@ -6286,8 +6242,7 @@ function assertProductBenchmarkRun(runDir) {
 	if (failures.length > 0) fail(runDir, `product benchmark validation failed:\n${failures.join("\n")}`);
 	return report;
 }
-//#endregion
-//#region src/runtime-trajectory.ts
+// src/runtime-trajectory.ts
 const DEFAULT_SPLIT_TAG = "search";
 function projectRuntimeTrajectoryEvidence(options) {
 	const diagnostics = [];
@@ -6373,8 +6328,7 @@ function stringOrUndefined(value) {
 function finiteNumberOrUndefined(value) {
 	return typeof value === "number" && Number.isFinite(value) ? value : void 0;
 }
-//#endregion
-//#region src/action-policy.ts
+// src/action-policy.ts
 function evaluateActionPolicy(action, policy = {}, options = {}) {
 	const reasons = [];
 	let blocked = false;
@@ -6441,8 +6395,7 @@ function evaluateActionPolicy(action, policy = {}, options = {}) {
 		label
 	};
 }
-//#endregion
-//#region src/deploy-gate-layer.ts
+// src/deploy-gate-layer.ts
 /**
 * Build a deploy gate layer that runs the production build and verifies
 * the artifact. Pass: ok && artifactValid. Score: 1.0 (pass) or 0 (fail).
@@ -6614,8 +6567,7 @@ function wranglerDeployRunner(input) {
 		};
 	} };
 }
-//#endregion
-//#region src/held-out-gate-inputs.ts
+// src/held-out-gate-inputs.ts
 /** Drop gamed runs, pair both splits and measure coverage, scores, gaps and cost. */
 function collectGateInputs(candidate, baseline, baselineKey) {
 	const candidateId = inferCandidateId(candidate, baselineKey);
@@ -6744,12 +6696,6 @@ function diffOrNull(a, b) {
 	if (a === null || b === null) return null;
 	return a - b;
 }
-function medianFinite(xs) {
-	const ys = xs.filter((x) => Number.isFinite(x)).sort((x, y) => x - y);
-	if (ys.length === 0) return null;
-	const mid = Math.floor(ys.length / 2);
-	return ys.length % 2 === 0 ? (ys[mid - 1] + ys[mid]) / 2 : ys[mid];
-}
 function completeCostMedian(runs) {
 	if (runs.length === 0) return null;
 	const costs = [];
@@ -6758,10 +6704,9 @@ function completeCostMedian(runs) {
 		if (usd === null) return null;
 		costs.push(usd);
 	}
-	return medianFinite(costs);
+	return sampleMedian(costs.filter(Number.isFinite));
 }
-//#endregion
-//#region src/held-out-gate.ts
+// src/held-out-gate.ts
 /**
 * HeldOutGate — first-class held-out paired-delta promotion gate.
 *
@@ -7060,8 +7005,7 @@ function fmt(x) {
 	if (x === null) return "n/a";
 	return x.toFixed(4);
 }
-//#endregion
-//#region src/hidden-criteria-grading.ts
+// src/hidden-criteria-grading.ts
 /**
 * Hidden-criteria grading firewall — grade an agent on criteria it never saw.
 *
@@ -7165,8 +7109,7 @@ function clampUnit(value) {
 	if (!Number.isFinite(value)) return 0;
 	return Math.max(0, Math.min(1, value));
 }
-//#endregion
-//#region src/integrity/single-backend.ts
+// src/integrity/single-backend.ts
 /**
 * Single-backend guard: assert the agent and the rubric judge run through the
 * SAME backend config, so the judge can't silently re-route through a
@@ -7240,8 +7183,7 @@ function assertSingleBackend(agent, judge, opts = {}) {
 	}
 	return report;
 }
-//#endregion
-//#region src/intent-match-judge.ts
+// src/intent-match-judge.ts
 const INTENT_MATCH_JUDGE_VERSION = "intent-match-judge-v1-2026-04-24";
 const DEFAULT_MODEL$1 = "claude-sonnet-4-6";
 const DEFAULT_TIMEOUT$1 = 3e5;
@@ -7394,8 +7336,7 @@ async function runIntentMatchJudge(input, options) {
 		};
 	}
 }
-//#endregion
-//#region src/judge-ensemble.ts
+// src/judge-ensemble.ts
 /**
 * Reduce per-judge verdicts to one aggregate. Generic over the rubric: pass the
 * stable-ordered `dimensionKeys` the judges scored.
@@ -7463,8 +7404,7 @@ function aggregateJudgeVerdicts(verdicts, dimensionKeys, weights) {
 		verdicts: [...verdicts]
 	};
 }
-//#endregion
-//#region src/judge-panel.ts
+// src/judge-panel.ts
 /**
 * Build a campaign-shaped `JudgeConfig` whose `score()` runs every panel
 * model in parallel and reduces the surviving verdicts to one canonical
@@ -7573,8 +7513,7 @@ function ensembleJudge(opts) {
 		}
 	};
 }
-//#endregion
-//#region src/keyword-coverage-judge.ts
+// src/keyword-coverage-judge.ts
 /**
 * Element-presence check using a tiny CSS-selector subset. Returns
 * null when the selector isn't supported — caller treats that as
@@ -7715,8 +7654,7 @@ function softFail(expectedConcepts, start, error) {
 		error
 	};
 }
-//#endregion
-//#region src/multi-layer-verifier.ts
+// src/multi-layer-verifier.ts
 /**
 * Multi-layer verifier — ordered pipeline of verification layers.
 *
@@ -7927,8 +7865,7 @@ function mergeSignals(a, b) {
 	b.addEventListener("abort", onAbort(b), { once: true });
 	return c.signal;
 }
-//#endregion
-//#region src/oracle.ts
+// src/oracle.ts
 /**
 * Declarative oracles — ground-truth assertions without an LLM.
 *
@@ -8011,8 +7948,7 @@ function evaluateOracles(obs, oracles) {
 function isObject(v) {
 	return typeof v === "object" && v !== null && !Array.isArray(v);
 }
-//#endregion
-//#region src/produced-state.ts
+// src/produced-state.ts
 function artifactKind(mimeType) {
 	if (!mimeType) return "file";
 	if (mimeType.includes("json")) return "json";
@@ -8066,8 +8002,7 @@ function extractProducedState(events) {
 		toolCalls
 	};
 }
-//#endregion
-//#region src/semantic-concept-judge.ts
+// src/semantic-concept-judge.ts
 const DEFAULT_COMPLEXITY_WEIGHTS = {
 	render: 1,
 	integrate: 2,
@@ -8306,8 +8241,7 @@ async function runSemanticConceptJudge(input, options) {
 		};
 	}
 }
-//#endregion
-//#region src/treatment-gate.ts
+// src/treatment-gate.ts
 /**
 * Core predicate. Gated (treatment-not-applied) ONLY when telemetry was
 * captured AND zero matching calls appear; an empty histogram fails open.
@@ -8348,8 +8282,7 @@ function gateTreatmentApplied(input, opts = {}) {
 		observedTools
 	};
 }
-//#endregion
-//#region src/anti-slop.ts
+// src/anti-slop.ts
 const DEFAULT_HEDGES = [
 	/\bi\s+could\s+be\s+wrong\b/i,
 	/\bi\s+think\s+maybe\b/i,
@@ -8507,8 +8440,7 @@ function splitSentences(text) {
 function normalizeForDupe(s) {
 	return s.toLowerCase().replace(/\s+/g, " ").replace(/[^a-z0-9 ]/g, "").trim();
 }
-//#endregion
-//#region src/eval-trace-store.ts
+// src/eval-trace-store.ts
 /**
 * EvalTraceStore — JSONL save / query / compare over the analysis-time
 * `RunRecord` row.
@@ -8574,8 +8506,7 @@ function jsonlRunRecordBackend(path, opts = {}) {
 		}
 	};
 }
-//#endregion
-//#region src/integrity/preflight.ts
+// src/integrity/preflight.ts
 /**
 * Provider signature for "the budget ran out before an answer". The model is
 * alive — a provider took the request and consumed the budget — so this must
@@ -8684,8 +8615,7 @@ async function preflightModels(opts) {
 		error: null
 	};
 }
-//#endregion
-//#region src/model-seats.ts
+// src/model-seats.ts
 /**
 * ModelSeats — the program's model seating chart.
 *
@@ -8776,8 +8706,7 @@ function resolveSeat(seats, seat, fallback) {
 	}
 	throw new SeatUnsetError(seat);
 }
-//#endregion
-//#region src/prompt-registry.ts
+// src/prompt-registry.ts
 var PromptRegistry = class {
 	entries = /* @__PURE__ */ new Map();
 	/**
@@ -8842,8 +8771,7 @@ function validateId(id) {
 function validateVersion(version) {
 	if (!version || version.length > 64) throw new Error(`Invalid version "${version}": must be 1–64 chars`);
 }
-//#endregion
-//#region src/steering-optimizer.ts
+// src/steering-optimizer.ts
 /**
 * Rank already-evaluated steering variants.
 *
@@ -8875,7 +8803,6 @@ function rankRows(rows, weights) {
 		runs: values.length
 	})).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, CostCeilingReachedError, CostLedger, CostTracker, CrossFamilyError, DEFAULT_BOUNDED_PROCESS_TIMEOUT_MS, DEFAULT_JUDGE_GATE_POLICY, 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, JudgeGateRefusedError, LlmCallError, LlmResponseError, MODEL_PRICING, ModelSubstitutionError, MultiLayerVerifier, NoopRawProviderSink, NotFoundError, OUTPUT_VALUE, OtlpFileTraceStore, PairwiseSteeringOptimizer, ProductClient, PromptRegistry, REDACTION_VERSION, RunIntegrityError, RunRecordValidationError, SEMANTIC_CONCEPT_JUDGE_VERSION, TraceEmitter, UNKNOWN_MODEL, VERIFICATION_STRATEGIES, VERIFICATION_STRATEGY_SOURCES, ValidationError, acquisitionPlansForKnowledgeGaps, addCalibrationVerdicts, agentProfileCellKey, agentProfileHash, agentProfileId, aggregateJudgeVerdicts, aggregateRunScore, analystRunDigest, analystRunToFeedbackTrajectory, analystRunToReviewRequests, analyzeAntiSlop, analyzeRuns, analyzeSeries, analyzeTraces, argHash, assertCapabilityHeadroom, assertCrossFamily, assertCrossFamilyServed, assertJudgeMayGate, assertNoHiddenLeak, assertProductBenchmarkRun, assertRealBackend, assertRunCaptured, assertServedModel, assertServedModels, assertSingleBackend, assessShareSafety, assignFeedbackSplit, attest, benjaminiHochberg, blendHeldout, blockingKnowledgeEval, bonferroni, bootstrapCi, budgetBreachView, buildAgentProfileCell, buildDefaultAnalystRegistry, buildReflectionPrompt, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, calibrateJudge, calibrateJudgeContinuous, canonicalJson, capabilityHeadroom, captureFetchToRawSink, certificationEvidenceDigest, checkCanaries, checkServedModel, checkTraceContracts, chiSquared, chiSquaredPValue, clamp01, classifyFailure, classifyFailureReason, cliffsDelta, cohensD, comparePairedArms, compileTraceContractSpec, completionVerdict, computeFindingId, computeToolUseMetrics, confidenceInterval, contentHash, continuousAgreement, controlRunToFeedbackTrajectory, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, costForTokenPricing, costForUsage, costReceiptFromLlm, costReceiptFromLlmError, createAntiSlopJudge, createBoundedTraceAnalysisStore, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createFeedbackTrajectory, createLlmCorrectnessChecker, createLlmReviewer, createTraceAnalyst, decidePairedPromotion, defineAgentEval, defineEquivalenceCheck, deployGateLayer, describeServedModelCheck, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, domainEvidencePattern, dominates, eProcess, ensembleJudge, equivalenceVerdict, errorStreakDetector, estimateCost, estimateTokens, evaluateActionPolicy, evaluateInterimReleaseConfidence, evaluateOracles, evaluateReleaseConfidence, evaluateTraceContract, expandProfileAxes, explainTraceContract, exportProductBenchmark, exportProductBenchmarkRuns, exportRunAsOtlp, extractErrorCount, extractProducedState, extractUsage, extractUsageFromSse, failureBlame, failureClusterView, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToOptimizerRow, fileVerdictCache, fisherExactTwoSided, formatScorecardDiff, gainHistogram, gateTreatmentApplied, gradeSemanticStatus, groupRunsByAgentProfileCell, harnessAxisOf, hashContent, hashJson, holm, inMemoryReviewStore, inferDomainKeywords, interRaterReliability, interpretCliffs, iqr, isJudgeSpan, isLlmSpan, isModelPriced, isRunRecord, isToolSpan, isTransientLlmError, jsonShape, jsonlReviewStore, jsonlRunRecordBackend, judgeAgreementView, judgeFamily, judgeGateDecision, judgeSpans, knowledgeReadinessTracePayload, leaderboard, linearRegressionSlope, lintTraceContractSpec, llmJudge, loadScorecard, localCommandRunner, makeFinding, makePolicyEditCandidateRecord, makeProposalFinding, manifestContentDigest, mannWhitneyU, maximumChargeForLlmRequest, mcnemar, mcnemarPower, mcnemarRequiredN, measureJudgeAgreement, minimumPairsForPairedDeltaTest, mintRolloutRows, modelHasSnapshot, mulberry32, normalTwoSidedPValue, objectiveEval, observeAll, otlpTextToTraceAnalysisStore, pairArms, pairRunRecords, pairedBootstrap, pairedCohensDz, pairedDeltaTest, pairedEvalueSequence, pairedMde, pairedPromotionPower, pairedRiskDifference, pairedRiskDifferenceExact, pairedRiskDifferenceScore, pairedSignTest, pairedTTest, paretoFrontier, parseReflectionResponse, parseRunRecordSafe, partialCredit, partitionHeldOut, passAtK, pearsonR, preflightModels, productBenchmarkRepoIdentity, profile_exports as profile, projectRuntimeTrajectoryEvidence, proposeSynthesisTargets, ranks, readProductBenchmarkManifest, recordRuns, recordRunsToScorecard, redTeamDataset, redTeamReport, redact, redactForShare, redactText, registerCalibrationSet, renderPreferenceMemoryMarkdown, repeatedActionDetector, requiredPairedSampleSize, requiredPairsForPairedPromotion, requiredSampleSize, resolveModelPricing, resolveSeat, runAgentControlLoop, runBoundedProcess, runCampaign, runCanaries, runEquivalenceCheck, runEvalCampaign, runIntentMatchJudge, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runProposeReview, runSemanticConceptJudge, runTaskScore, runsForScenario, sampleMedian, scoreKnowledgeReadiness, scoreRedTeamOutput, scoreTraceInsightReadiness, seatPresets, selfImprove, shareAllowed, snapshotFromJudgeAgreement, spanRecordsToTraceAnalysisStore, spearmanR, stripFencedJson, studentTTwoSidedPValue, subjectiveEval, summarizeBackendIntegrity, summarizeNumberSeries, summarizePreferenceMemory, summarizeSample, summaryTable, toAgentProfileJson, tokenizeDomainWords, toolSpansToTraceAnalysisStore, toolWasteView, traceContract, transientDispatchFailure, twoByTwoPValue, userQuestionsForKnowledgeGaps, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyAttestation, verifyCompletion, viteDeployRunner, weightedComposite, weightedMean, welchsTTest, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit, withJudgeRetry, wranglerDeployRunner };
 
 //# sourceMappingURL=index.js.map