@tangle-network/agent-eval 0.137.0 → 0.139.0

package/dist/analyst/index.js

@@ -1,9 +1,12 @@
-import { A as FindingSubjectStringSchema, C as parseRawFinding, D as FINDING_SUBJECT_GRAMMAR_PROMPT, E as stripCodeFences, F as behavioralAnalyst, I as deriveEfficiencyFindings, M as findingSubjectGrammarPromptFor, N as parseFindingSubject, O as FINDING_SUBJECT_KINDS, P as renderFindingSubject, R as createChatClient, S as evidenceRefsFromRawFinding, T as coerceToFindingRows, _ as structureFindings, a as KNOWLEDGE_GAP_KIND_SPEC, b as RawAnalystEvidenceSchema, c as buildTraceToolsForGroup, d as emitControlIntegrityFindings, f as createTraceAnalystKind, g as validateUsageSettlementTimeout, h as settleUsageReceiptFromCostLedger, i as KNOWLEDGE_POISONING_KIND_SPEC, j as KIND_EXPECTED_SUBJECTS, k as FINDING_SUBJECT_SYNTAX, l as CONTROL_INTEGRITY_ANALYST, m as renderUpstreamFindings, n as AnalystRegistry, o as IMPROVEMENT_KIND_SPEC, p as renderPriorFindings, r as DEFAULT_TRACE_ANALYST_KINDS, s as FAILURE_MODE_KIND_SPEC, t as buildDefaultAnalystRegistry, u as ControlIntegrityAnalyst, v as ANALYST_SEVERITIES, w as coerceJson, x as RawAnalystFindingSchema, y as RAW_FINDING_SCHEMA_PROMPT, z as createAnalystAi } from "../default-registry-CLXbRt0f.js";
-import { i as CostLedger } from "../cost-ledger-D-5_-dhi.js";
-import { c as makeFinding, l as makeProposalFinding, n as isProposalFinding, s as computeFindingId, t as assertProposalFindings } from "../proposal-findings-DCawte-y.js";
-import { a as RunCritic, c as buildSkillUsageReport, d as defaultIsMaterial, f as diffFindings, i as runSemanticConceptJudge, l as emitSkillUsageFindings, n as SEMANTIC_CONCEPT_JUDGE_VERSION, o as SKILL_USAGE_ANALYST, s as SkillUsageAnalyst, u as FindingsStore } from "../semantic-concept-judge-Bz64IckK.js";
-import { E as pairedBootstrap } from "../statistics-ByxzSiOM.js";
-import { i as traceStoreEvidenceResolver, n as runAnalystBenchmark, r as scoreAnalystFindings, t as registryBenchmarkRunner } from "../benchmark-YDrpumqB.js";
+import { i as CostLedger } from "../cost-ledger-CZ9diLxY.js";
+import { C as createChatClient, _ as CONTROL_INTEGRITY_ANALYST, b as behavioralAnalyst, f as DEFAULT_TRACE_ANALYST_KINDS, g as FAILURE_MODE_KIND_SPEC, h as IMPROVEMENT_KIND_SPEC, i as assertExactRegistryRunOpts, m as KNOWLEDGE_GAP_KIND_SPEC, n as AnalystRegistry, p as KNOWLEDGE_POISONING_KIND_SPEC, r as ExactAnalystRunExecutionError, t as buildDefaultAnalystRegistry, v as ControlIntegrityAnalyst, x as deriveEfficiencyFindings, y as emitControlIntegrityFindings } from "../default-registry-BgJJItGr.js";
+import { A as coerceJson, B as renderFindingSubject, D as RawAnalystFindingSchema, E as RawAnalystEvidenceSchema, F as FINDING_SUBJECT_SYNTAX, G as resolveTraceAnalystLimits, I as FindingSubjectStringSchema, L as KIND_EXPECTED_SUBJECTS, M as stripCodeFences, N as FINDING_SUBJECT_GRAMMAR_PROMPT, O as evidenceRefsFromRawFinding, P as FINDING_SUBJECT_KINDS, R as findingSubjectGrammarPromptFor, T as RAW_FINDING_SCHEMA_PROMPT, W as DEFAULT_TRACE_ANALYST_LIMITS, a as buildTraceToolsForGroup, i as runTraceAnalyst, j as coerceToFindingRows, k as parseRawFinding, n as renderPriorFindings, r as renderUpstreamFindings, t as createTraceAnalyst, w as ANALYST_SEVERITIES, z as parseFindingSubject } from "../kind-factory-CFxA0JQX.js";
+import { a as computeFindingId, i as validateUsageSettlementTimeout, n as settleUsageReceiptFromCostLedger, o as makeFinding, s as makeProposalFinding } from "../usage-receipt-CgxMEBZq.js";
+import { n as isProposalFinding, t as assertProposalFindings } from "../proposal-findings-2GIUo1et.js";
+import { a as RunCritic, c as buildSkillUsageReport, d as defaultIsMaterial, f as diffFindings, h as defineTraceAnalyst, i as runSemanticConceptJudge, l as emitSkillUsageFindings, m as defineCustomAnalyst, n as SEMANTIC_CONCEPT_JUDGE_VERSION, o as SKILL_USAGE_ANALYST, s as SkillUsageAnalyst, u as FindingsStore } from "../semantic-concept-judge-BuIJ9IfB.js";
+import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-DTkVyDX-.js";
+import { a as scoreAnalystFindings, n as runAnalystBenchmark, r as traceStoreEvidenceResolver, t as registryBenchmarkRunner } from "../benchmark-CYtcIF2V.js";
+import { A as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, B as summarizeAgentRxCalibration, C as ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM, D as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, E as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, F as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, G as normalizeAgentRxCategory, H as codeTracerPredictionsToFindings, I as ANALYST_BENCHMARK_MANIFEST_FILE, K as roundAgentRxStep, L as ANALYST_BENCHMARK_OBSERVATIONS_FILE, M as analystBenchmarkDependencyLockDigest, N as analystBenchmarkImplementationDigest, O as ANALYST_BENCHMARK_EVIDENCE_PACKAGE_VERSION, P as ANALYST_BENCHMARK_COST_LEDGER_FILE, R as AGENT_RX_UPSTREAM_REVISION, S as emptyPublicBenchmarkRunner, T as ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256, U as agentRxBenchmarkCase, V as codeTraceBenchCase, W as agentRxPredictionsToFindings, _ as DEFAULT_MAX_VERIFICATION_ARTIFACT_BYTES, a as renderCodeTraceCalibrationMarkdown, b as parseVerificationOutcome, c as createPublicBenchmarkRlmRunner, d as publicBenchmarkProtocolSha256, f as loadPublicBenchmarkRows, g as selectPublicBenchmarkRows, h as publicBenchmarkSelectionReport, i as readAnalystBenchmarkArtifact, j as ANALYST_BENCHMARK_IMPLEMENTATION_SHA256, k as ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM, l as CODE_TRACE_BENCH_ANALYST_PROMPT, m as publicBenchmarkDistributions, n as runAnalystBenchmarkCommand, o as summarizeCodeTraceCalibration, p as preparePublicAnalystBenchmark, q as normalizeBenchmarkLabel, r as renderAnalystBenchmarkMarkdown, s as compareAnalystRunners, t as ANALYST_BENCHMARK_HELP, u as createPublicBenchmarkDirectRunner, v as appendVerificationArtifactsToOtlp, w as ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES, x as adaptPublicBenchmarkFindings, y as loadCodeTraceVerificationArtifacts, z as renderAgentRxCalibrationMarkdown } from "../benchmark-command-BKfjOBJ5.js";
 //#region src/analyst/adapters.ts
 /**
 * Adapter factories — lift each existing agent-eval primitive into the
@@ -291,606 +294,6 @@ function createSemanticConceptJudgeAdapter(opts = {}) {
 	};
 }
 //#endregion
-//#region src/analyst/benchmark-comparison.ts
-function compareAnalystRunners(result, options) {
-	const confidence = options.confidence ?? .95;
-	const resamples = options.resamples ?? 2e3;
-	assertComparisonControls(confidence, resamples);
-	const runnerIds = new Set(result.summaries.map((summary) => summary.runnerId));
-	if (!runnerIds.has(options.baselineRunnerId)) throw new TypeError(`unknown baseline analyst runner '${options.baselineRunnerId}'`);
-	if (!runnerIds.has(options.candidateRunnerId)) throw new TypeError(`unknown candidate analyst runner '${options.candidateRunnerId}'`);
-	if (options.baselineRunnerId === options.candidateRunnerId) throw new TypeError("baseline and candidate analyst runners must be different");
-	const baseline = observationsByCase(result.observations, options.baselineRunnerId);
-	const candidate = observationsByCase(result.observations, options.candidateRunnerId);
-	const metrics = [];
-	for (const metric of METRICS) {
-		const before = [];
-		const after = [];
-		let pairedObservations = 0;
-		for (const [caseId, baselineObservations] of baseline) {
-			const candidateObservations = candidate.get(caseId);
-			if (!candidateObservations) continue;
-			const candidateByRepetition = new Map(candidateObservations.map((observation) => [observation.repetition, observation]));
-			const caseBefore = [];
-			const caseAfter = [];
-			for (const baselineObservation of baselineObservations) {
-				const candidateObservation = candidateByRepetition.get(baselineObservation.repetition);
-				if (!candidateObservation) continue;
-				const baselineValue = metricValue(baselineObservation, metric);
-				const candidateValue = metricValue(candidateObservation, metric);
-				if (baselineValue === null || candidateValue === null) continue;
-				caseBefore.push(baselineValue);
-				caseAfter.push(candidateValue);
-			}
-			if (caseBefore.length === 0) continue;
-			before.push(mean(caseBefore));
-			after.push(mean(caseAfter));
-			pairedObservations += caseBefore.length;
-		}
-		if (before.length === 0) continue;
-		const interval = pairedBootstrap(before, after, {
-			confidence,
-			resamples,
-			statistic: "mean",
-			seed: options.seed
-		});
-		const comparison = {
-			metric,
-			direction: LOWER_IS_BETTER.has(metric) ? "lower" : "higher",
-			pairedCases: interval.n,
-			pairedObservations,
-			baselineMean: mean(before),
-			candidateMean: mean(after),
-			meanDelta: interval.mean,
-			intervalLow: interval.low,
-			intervalHigh: interval.high,
-			confidence: interval.confidence,
-			resamples: interval.resamples,
-			enoughCasesForInference: interval.gateEligible
-		};
-		assertValidComparison(comparison);
-		metrics.push(comparison);
-	}
-	return {
-		baselineRunnerId: options.baselineRunnerId,
-		candidateRunnerId: options.candidateRunnerId,
-		metrics
-	};
-}
-const METRICS = [
-	"completion",
-	"issueRecall",
-	"findingPrecision",
-	"f1",
-	"criticalStepAccuracy",
-	"citationCoverage",
-	"citationLabelAgreement",
-	"citationResolution",
-	"cleanAccuracy",
-	"latencyMs",
-	"calls",
-	"inputTokens",
-	"outputTokens",
-	"reasoningTokens",
-	"cachedTokens",
-	"cacheWriteTokens",
-	"costUsd"
-];
-const LOWER_IS_BETTER = /* @__PURE__ */ new Set([
-	"latencyMs",
-	"calls",
-	"inputTokens",
-	"outputTokens",
-	"reasoningTokens",
-	"cachedTokens",
-	"cacheWriteTokens",
-	"costUsd"
-]);
-function observationsByCase(observations, runnerId) {
-	const byCase = /* @__PURE__ */ new Map();
-	for (const observation of observations) {
-		if (observation.runnerId !== runnerId) continue;
-		const rows = byCase.get(observation.caseId) ?? [];
-		rows.push(observation);
-		byCase.set(observation.caseId, rows);
-	}
-	return byCase;
-}
-function metricValue(observation, metric) {
-	if (metric === "completion") return observation.error ? 0 : 1;
-	if (metric === "latencyMs") return observation.latencyMs;
-	if (metric === "cleanAccuracy") {
-		if (observation.score.expectedIssueCount !== 0) return null;
-		if (observation.error) return 0;
-		return observation.score.cleanFalsePositive ? 0 : 1;
-	}
-	if ((metric === "issueRecall" || metric === "findingPrecision" || metric === "f1") && observation.score.expectedIssueCount === 0) return null;
-	if (observation.error && (metric === "issueRecall" || metric === "findingPrecision" || metric === "f1" || metric === "criticalStepAccuracy")) {
-		if (metric === "criticalStepAccuracy" && observation.score.criticalStepAccuracy === null) return null;
-		return 0;
-	}
-	if (observation.error && (metric === "citationCoverage" || metric === "citationLabelAgreement" || metric === "citationResolution")) return null;
-	if (metric === "issueRecall") return observation.score.issueRecall;
-	if (metric === "findingPrecision") return observation.score.findingPrecision;
-	if (metric === "f1") return observation.score.f1;
-	if (metric === "criticalStepAccuracy") return observation.score.criticalStepAccuracy;
-	if (metric === "citationCoverage") return observation.score.citationCoverage;
-	if (metric === "citationLabelAgreement") return observation.score.citationLabelAgreement;
-	if (metric === "citationResolution") return observation.evidenceResolution?.validity ?? null;
-	if (metric === "calls") return observation.usage?.calls ?? null;
-	if (metric === "inputTokens") return observation.usage?.tokens?.input ?? null;
-	if (metric === "outputTokens") return observation.usage?.tokens?.output ?? null;
-	if (metric === "reasoningTokens") return observation.usage?.tokens?.reasoning ?? null;
-	if (metric === "cachedTokens") return observation.usage?.tokens?.cached ?? null;
-	if (metric === "cacheWriteTokens") return observation.usage?.tokens?.cacheWrite ?? null;
-	if (observation.usage?.cost.kind === "uncaptured") return null;
-	return observation.usage?.cost.usd ?? null;
-}
-function mean(values) {
-	return values.reduce((sum, value) => sum + value, 0) / values.length;
-}
-function assertComparisonControls(confidence, resamples) {
-	if (!Number.isSafeInteger(resamples) || resamples <= 0 || resamples > 1e6) throw new Error(`compareAnalystRunners: resamples must be a positive safe integer no greater than 1000000, got ${String(resamples)}`);
-	if (!Number.isFinite(confidence) || confidence <= 0 || confidence >= 1) throw new Error(`compareAnalystRunners: confidence must be a finite number in (0,1), got ${String(confidence)}`);
-}
-function assertValidComparison(comparison) {
-	if ([
-		"pairedCases",
-		"pairedObservations",
-		"baselineMean",
-		"candidateMean",
-		"meanDelta",
-		"intervalLow",
-		"intervalHigh",
-		"confidence",
-		"resamples"
-	].some((field) => !Number.isFinite(comparison[field]))) throw new Error(`compareAnalystRunners: ${comparison.metric} produced non-finite comparison output`);
-	if (comparison.intervalLow > comparison.intervalHigh) throw new Error(`compareAnalystRunners: ${comparison.metric} produced an invalid confidence interval`);
-}
-//#endregion
-//#region src/analyst/benchmark-datasets.ts
-function agentRxBenchmarkCase(row, input, options = {}) {
-	const trajectoryId = externalId(row.trajectory_id, "AgentRx trajectory_id");
-	if (!Array.isArray(row.failures) || row.failures.length === 0) throw new TypeError(`AgentRx trajectory '${trajectoryId}' must contain failures`);
-	if (row.num_failures !== void 0 && row.num_failures !== row.failures.length) throw new TypeError(`AgentRx trajectory '${trajectoryId}' declares ${row.num_failures} failures but contains ${row.failures.length}`);
-	const rootCauseId = externalId(row.root_cause_failure_id ?? row.root_cause?.failure_id, `AgentRx trajectory '${trajectoryId}' root cause failure id`);
-	const failureIds = /* @__PURE__ */ new Set();
-	const evidenceKind = options.evidenceKind ?? "span";
-	const uri = options.stepUri ?? defaultStepUri;
-	const allIssues = row.failures.map((failure) => {
-		const failureId = externalId(failure.failure_id, `AgentRx trajectory '${trajectoryId}' failure id`);
-		if (failureIds.has(failureId)) throw new TypeError(`AgentRx trajectory '${trajectoryId}' repeats failure id '${failureId}'`);
-		failureIds.add(failureId);
-		const step = positiveStep(failure.step_number, `AgentRx trajectory '${trajectoryId}'`);
-		const evidence = [{
-			kind: evidenceKind,
-			uri: uri(trajectoryId, step)
-		}];
-		return {
-			id: failureId,
-			areas: [normalizeAgentRxCategory(failure.failure_category)],
-			...failureId === rootCauseId && (options.target ?? "root-cause") === "root-cause" ? {} : { evidence },
-			criticalEvidence: failureId === rootCauseId ? evidence : void 0
-		};
-	});
-	if (!failureIds.has(rootCauseId)) throw new TypeError(`AgentRx trajectory '${trajectoryId}' root cause '${rootCauseId}' is not in failures`);
-	if (options.stepCount !== void 0 && row.failures.some((failure) => failure.step_number > options.stepCount)) throw new RangeError(`AgentRx trajectory '${trajectoryId}' contains a failure beyond stepCount ${options.stepCount}`);
-	const expectedIssues = (options.target ?? "root-cause") === "root-cause" ? allIssues.filter((issue) => issue.id === rootCauseId) : allIssues;
-	return {
-		id: `agentrx:${trajectoryId}`,
-		input,
-		expectedIssues,
-		labeledEvidence: expectedIssues.flatMap((issue) => issue.evidence ?? issue.criticalEvidence ?? []),
-		tags: ["agentrx"],
-		metadata: {
-			benchmark: "AgentRx",
-			trajectoryId,
-			failureSummary: row.failure_summary,
-			rootCauseReason: row.root_cause_reason ?? row.root_cause?.reason_for_root_cause,
-			annotatedFailures: row.failures.length,
-			target: options.target ?? "root-cause"
-		}
-	};
-}
-function codeTraceBenchCase(row, input, options = {}) {
-	const trajectoryId = nonEmpty(row.traj_id, "CodeTraceBench traj_id");
-	const stepCount = positiveStep(row.step_count, `CodeTraceBench '${trajectoryId}' step_count`);
-	const stages = parseCodeTraceStages(row.incorrect_stages, trajectoryId);
-	const evidenceKind = options.evidenceKind ?? "span";
-	const uri = options.stepUri ?? defaultStepUri;
-	const labelSet = codeTraceLabelSet(options.labelSet);
-	const labels = /* @__PURE__ */ new Set();
-	const expectedIssues = stages.flatMap((stage) => {
-		const incorrect = stepIssues("incorrect", stage.incorrect_step_ids ?? []);
-		const unuseful = stepIssues("unuseful", stage.unuseful_step_ids ?? []);
-		return labelSet === "incorrect-only" ? incorrect : [...incorrect, ...unuseful];
-	});
-	return {
-		id: `codetrace:${trajectoryId}`,
-		input,
-		expectedIssues,
-		labeledEvidence: expectedIssues.flatMap((issue) => issue.evidence ?? []),
-		tags: [
-			"codetracebench",
-			row.agent,
-			row.model,
-			...row.difficulty ? [row.difficulty] : [],
-			...row.category ? [row.category] : [],
-			...parseTags(row.tags, trajectoryId)
-		],
-		metadata: {
-			benchmark: "CodeTraceBench",
-			trajectoryId,
-			taskName: row.task_name,
-			agent: row.agent,
-			model: row.model,
-			solved: row.solved,
-			stepCount,
-			labelSet
-		}
-	};
-	function stepIssues(label, steps) {
-		return steps.map((rawStep) => {
-			const step = positiveStep(rawStep, `CodeTraceBench '${trajectoryId}' ${label} step`);
-			if (step > stepCount) throw new RangeError(`CodeTraceBench '${trajectoryId}' ${label} step ${step} exceeds step_count ${stepCount}`);
-			const id = `${label}:${step}`;
-			if (labels.has(id)) throw new TypeError(`CodeTraceBench '${trajectoryId}' repeats label '${id}'`);
-			labels.add(id);
-			return {
-				id,
-				areas: [label],
-				evidence: [{
-					kind: evidenceKind,
-					uri: uri(trajectoryId, step)
-				}]
-			};
-		});
-	}
-}
-/** Translate AgentRx `Report.to_dict()` output or its `failures` array into findings. */
-function agentRxPredictionsToFindings(trajectoryIdValue, output, options = {}) {
-	const trajectoryId = externalId(trajectoryIdValue, "AgentRx prediction trajectory id");
-	const parsed = parseAgentRxPredictions(output, trajectoryId);
-	for (const prediction of parsed.predictions) {
-		assertStepWithinRange(prediction.step_number, parsed.report?.trajectory_length, `AgentRx prediction '${trajectoryId}' report`);
-		assertStepWithinRange(prediction.step_number, options.stepCount, `AgentRx prediction '${trajectoryId}'`);
-	}
-	const consensus = agentRxConsensus(parsed, trajectoryId);
-	if (consensus.failureCase === 0) return [];
-	const confidence = predictionConfidence(options.confidence);
-	const uri = options.stepUri ?? defaultStepUri;
-	assertStepWithinRange(consensus.step, options.stepCount, `AgentRx prediction '${trajectoryId}'`);
-	const area = AGENT_RX_TAXONOMY.get(consensus.failureCase);
-	return [makeFinding({
-		analyst_id: options.analystId ?? "agentrx",
-		produced_at: options.producedAt,
-		area,
-		subject: "root-cause",
-		claim: `AgentRx classified step ${consensus.step} as ${area}.`,
-		id_basis: `${area}:${consensus.step}`,
-		rationale: consensus.representative.description,
-		severity: "high",
-		confidence,
-		evidence_refs: [{
-			kind: options.evidenceKind ?? "span",
-			uri: uri(trajectoryId, consensus.step)
-		}],
-		metadata: {
-			upstream: "AgentRx",
-			failure_case: consensus.failureCase,
-			step: consensus.step,
-			step_mean: consensus.stepMean,
-			judge_votes: parsed.predictions.length,
-			consensus_votes: consensus.votes,
-			category_agreement: consensus.votes / parsed.predictions.length,
-			...consensus.representative.checklist_reasoning === void 0 || consensus.representative.checklist_reasoning === null ? {} : { checklist_reasoning: consensus.representative.checklist_reasoning }
-		}
-	})];
-}
-/** Translate CodeTracer's `codetracer_labels.json` into shared findings. */
-function codeTracerPredictionsToFindings(trajectoryIdValue, predictions, options = {}) {
-	const trajectoryId = nonEmpty(trajectoryIdValue, "CodeTracer prediction trajectory id");
-	const stages = parseCodeTraceStages(predictions, trajectoryId);
-	const confidence = predictionConfidence(options.confidence);
-	const uri = options.stepUri ?? defaultStepUri;
-	const labelSet = codeTraceLabelSet(options.labelSet);
-	const seen = /* @__PURE__ */ new Set();
-	const findings = [];
-	for (const stage of stages) for (const [area, steps] of [["incorrect", stage.incorrect_step_ids ?? []], ["unuseful", stage.unuseful_step_ids ?? []]]) for (const rawStep of steps) {
-		const step = positiveStep(rawStep, `CodeTracer prediction '${trajectoryId}' ${area} step`);
-		assertStepWithinRange(step, options.stepCount, `CodeTracer prediction '${trajectoryId}'`);
-		const key = `${area}:${step}`;
-		if (seen.has(key)) throw new TypeError(`CodeTracer prediction '${trajectoryId}' repeats label '${key}'`);
-		seen.add(key);
-		if (area === "unuseful" && labelSet === "incorrect-only") continue;
-		findings.push(makeFinding({
-			analyst_id: options.analystId ?? "codetracer",
-			produced_at: options.producedAt,
-			area,
-			subject: `step-${step}`,
-			claim: `CodeTracer labeled step ${step} as ${area}.`,
-			id_basis: key,
-			rationale: stage.reasoning,
-			severity: "medium",
-			confidence,
-			evidence_refs: [{
-				kind: options.evidenceKind ?? "span",
-				uri: uri(trajectoryId, step)
-			}],
-			metadata: {
-				upstream: "CodeTracer",
-				stage_id: stage.stage_id,
-				step
-			}
-		}));
-	}
-	return findings;
-}
-function normalizeBenchmarkLabel(value) {
-	const normalized = nonEmpty(value, "benchmark label").toLowerCase().replace(/[^a-z0-9]+/g, "-").replace(/^-|-$/g, "");
-	if (!normalized) throw new TypeError("benchmark label must contain letters or digits");
-	return normalized;
-}
-function parseCodeTraceStages(value, trajectoryId) {
-	let parsed = value;
-	if (typeof value === "string") try {
-		parsed = JSON.parse(value);
-	} catch (error) {
-		throw new TypeError(`CodeTraceBench '${trajectoryId}' incorrect_stages is invalid JSON: ${error instanceof Error ? error.message : String(error)}`);
-	}
-	if (!Array.isArray(parsed)) throw new TypeError(`CodeTraceBench '${trajectoryId}' incorrect_stages must be an array`);
-	for (const stage of parsed) if (!stage || typeof stage !== "object" || !Number.isSafeInteger(stage.stage_id) || stage.stage_id < 1 || !optionalStepArray(stage.incorrect_step_ids) || !optionalStepArray(stage.unuseful_step_ids) || stage.reasoning !== void 0 && typeof stage.reasoning !== "string") throw new TypeError(`CodeTraceBench '${trajectoryId}' contains an invalid stage annotation`);
-	return parsed;
-}
-function parseTags(value, trajectoryId) {
-	if (value === void 0) return [];
-	let parsed = value;
-	if (typeof value === "string") try {
-		parsed = JSON.parse(value);
-	} catch (error) {
-		throw new TypeError(`CodeTraceBench '${trajectoryId}' tags is invalid JSON: ${error instanceof Error ? error.message : String(error)}`);
-	}
-	if (!Array.isArray(parsed) || !parsed.every((tag) => typeof tag === "string")) throw new TypeError(`CodeTraceBench '${trajectoryId}' tags must be strings`);
-	return [...parsed];
-}
-const AGENT_RX_TAXONOMY = /* @__PURE__ */ new Map([
-	[1, "instruction-plan-adherence-failure"],
-	[2, "invention-of-new-information"],
-	[3, "invalid-invocation"],
-	[4, "misinterpretation-of-tool-output-handoff-failure"],
-	[5, "intent-plan-misalignment"],
-	[6, "underspecified-user-intent"],
-	[7, "intent-not-supported"],
-	[8, "guardrails-triggered"],
-	[9, "system-failure"],
-	[10, "inconclusive"]
-]);
-const AGENT_RX_CATEGORY_ALIASES = /* @__PURE__ */ new Map([["instruction-adherence-failure", "instruction-plan-adherence-failure"], ["misinterpretation-of-tool-output", "misinterpretation-of-tool-output-handoff-failure"]]);
-const AGENT_RX_TAXONOMY_BY_LABEL = new Map([...AGENT_RX_TAXONOMY].map(([failureCase, label]) => [label, failureCase]));
-function normalizeAgentRxCategory(value) {
-	const normalized = normalizeBenchmarkLabel(value);
-	return AGENT_RX_CATEGORY_ALIASES.get(normalized) ?? normalized;
-}
-function parseAgentRxFailureCase(value, field) {
-	if (typeof value !== "number" && typeof value !== "string") throw new TypeError(`${field} must be a taxonomy number or label`);
-	if (typeof value === "string" && !/^\d+$/.test(value.trim())) {
-		const normalized = normalizeAgentRxCategory(value);
-		const failureCase = AGENT_RX_TAXONOMY_BY_LABEL.get(normalized);
-		if (failureCase === void 0) throw new RangeError(`${field} '${value}' is not an AgentRx taxonomy label`);
-		return failureCase;
-	}
-	const numeric = typeof value === "number" ? value : Number(value);
-	if (!Number.isSafeInteger(numeric)) throw new TypeError(`${field} must be a taxonomy number or label`);
-	if (numeric < 0 || numeric > 10) throw new RangeError(`${field} ${numeric} is outside 0-10`);
-	return numeric;
-}
-function predictionConfidence(value) {
-	const confidence = value ?? .5;
-	if (!Number.isFinite(confidence) || confidence < 0 || confidence > 1) throw new RangeError("upstream prediction confidence must be between 0 and 1");
-	return confidence;
-}
-function codeTraceLabelSet(value) {
-	if (value === void 0 || value === "incorrect-only") return "incorrect-only";
-	if (value === "incorrect-and-unuseful") return value;
-	throw new TypeError("CodeTraceBench labelSet must be 'incorrect-only' or 'incorrect-and-unuseful'");
-}
-function assertStepWithinRange(step, stepCount, field) {
-	if (stepCount === void 0) return;
-	const count = positiveStep(stepCount, `${field} stepCount`);
-	if (step > count) throw new RangeError(`${field} step ${step} exceeds stepCount ${count}`);
-}
-function parseAgentRxPredictions(output, trajectoryId) {
-	let failures;
-	let report;
-	if (Array.isArray(output)) failures = output;
-	else if (isRecord(output)) {
-		assertMatchingAgentRxTaskId(output.task_id, trajectoryId, "report.task_id");
-		if (!Object.hasOwn(output, "failures")) throw new TypeError(`AgentRx prediction '${trajectoryId}' report must contain failures`);
-		failures = output.failures;
-		if (output.num_judges !== void 0) {
-			if (!Number.isSafeInteger(output.num_judges) || output.num_judges < 0) throw new RangeError(`AgentRx prediction '${trajectoryId}' report.num_judges must be a non-negative safe integer`);
-		}
-		if (output.trajectory_length !== void 0) positiveStep(output.trajectory_length, `AgentRx prediction '${trajectoryId}' report.trajectory_length`);
-		if (output.step_mean !== void 0) {
-			if (typeof output.step_mean !== "number" || !Number.isFinite(output.step_mean)) throw new TypeError(`AgentRx prediction '${trajectoryId}' report.step_mean must be finite`);
-		}
-		if (output.modes !== void 0 && !Array.isArray(output.modes)) throw new TypeError(`AgentRx prediction '${trajectoryId}' report.modes must be an array`);
-		report = output;
-	} else throw new TypeError(`AgentRx prediction '${trajectoryId}' must be a report or failures array`);
-	if (!Array.isArray(failures)) throw new TypeError(`AgentRx prediction '${trajectoryId}' failures must be an array`);
-	if (failures.length === 0) throw new TypeError(`AgentRx prediction '${trajectoryId}' failures must contain a judge prediction`);
-	if (isRecord(output) && output.num_judges !== void 0 && output.num_judges !== failures.length) throw new TypeError(`AgentRx prediction '${trajectoryId}' declares ${output.num_judges} judges but contains ${failures.length} failures`);
-	return {
-		predictions: failures.map((value, index) => {
-			const field = `AgentRx prediction '${trajectoryId}' failures[${index}]`;
-			if (!isRecord(value)) throw new TypeError(`${field} must be an object`);
-			assertMatchingAgentRxTaskId(value.task_id, trajectoryId, `${field}.task_id`);
-			const failureCase = parseAgentRxFailureCase(value.failure_case, `${field}.failure_case`);
-			if (!Number.isSafeInteger(value.step_number)) throw new TypeError(`${field}.step_number must be a safe integer`);
-			const stepNumber = value.step_number;
-			if (failureCase === 0 ? stepNumber !== 0 : stepNumber < 1) throw new RangeError(failureCase === 0 ? `${field}.step_number must be 0 when failure_case is 0` : `${field}.step_number must be positive when failure_case is 1-10`);
-			if (value.description !== void 0 && typeof value.description !== "string") throw new TypeError(`${field}.description must be a string`);
-			if (value.checklist_reasoning !== void 0 && value.checklist_reasoning !== null && typeof value.checklist_reasoning !== "string") throw new TypeError(`${field}.checklist_reasoning must be a string or null`);
-			return {
-				...value.task_id === void 0 ? {} : { task_id: value.task_id },
-				failure_case: failureCase,
-				step_number: stepNumber,
-				...value.description === void 0 ? {} : { description: value.description },
-				...value.checklist_reasoning === void 0 ? {} : { checklist_reasoning: value.checklist_reasoning }
-			};
-		}),
-		report
-	};
-}
-function agentRxConsensus(parsed, trajectoryId) {
-	const counts = /* @__PURE__ */ new Map();
-	for (const prediction of parsed.predictions) counts.set(prediction.failure_case, (counts.get(prediction.failure_case) ?? 0) + 1);
-	const maxVotes = Math.max(...counts.values());
-	let failureCase = [...counts].find(([, count]) => count === maxVotes)[0];
-	if (parsed.report?.most_common_failure !== void 0) {
-		const declared = parseAgentRxFailureCase(parsed.report.most_common_failure, `AgentRx prediction '${trajectoryId}' report.most_common_failure`);
-		if ((counts.get(declared) ?? 0) !== maxVotes) throw new TypeError(`AgentRx prediction '${trajectoryId}' report.most_common_failure disagrees with failures`);
-		failureCase = declared;
-	}
-	if (parsed.report?.modes !== void 0) {
-		const declaredModes = parsed.report.modes.map((value, index) => parseAgentRxFailureCase(value, `AgentRx prediction '${trajectoryId}' report.modes[${index}]`));
-		if (new Set(declaredModes).size !== declaredModes.length) throw new TypeError(`AgentRx prediction '${trajectoryId}' report.modes contains duplicates`);
-		const expectedModes = [...counts].filter(([, count]) => count === maxVotes).map(([value]) => value).sort((left, right) => left - right);
-		if ([...new Set(declaredModes)].sort((left, right) => left - right).join(",") !== expectedModes.join(",")) throw new TypeError(`AgentRx prediction '${trajectoryId}' report.modes disagrees with failures`);
-	}
-	const computedStepMean = parsed.predictions.reduce((sum, prediction) => sum + prediction.step_number, 0) / parsed.predictions.length;
-	if (parsed.report?.step_mean !== void 0 && Math.abs(parsed.report.step_mean - computedStepMean) > 1e-12) throw new TypeError(`AgentRx prediction '${trajectoryId}' report.step_mean disagrees with failures`);
-	const stepMean = parsed.report?.step_mean ?? computedStepMean;
-	const step = failureCase === 0 ? 0 : positiveStep(roundHalfToEven(stepMean), `AgentRx prediction '${trajectoryId}' consensus step`);
-	const representative = parsed.predictions.filter((prediction) => prediction.failure_case === failureCase).sort((left, right) => Math.abs(left.step_number - step) - Math.abs(right.step_number - step))[0] ?? parsed.predictions[0];
-	return {
-		failureCase,
-		step,
-		stepMean,
-		votes: counts.get(failureCase),
-		representative
-	};
-}
-function roundHalfToEven(value) {
-	const lower = Math.floor(value);
-	const fraction = value - lower;
-	if (Math.abs(fraction - .5) <= Number.EPSILON * Math.max(1, Math.abs(value))) return lower % 2 === 0 ? lower : lower + 1;
-	return Math.round(value);
-}
-function assertMatchingAgentRxTaskId(value, trajectoryId, field) {
-	if (value === void 0) return;
-	const taskId = externalId(value, `AgentRx prediction '${trajectoryId}' ${field}`);
-	if (taskId !== trajectoryId) throw new TypeError(`AgentRx prediction '${trajectoryId}' ${field} '${taskId}' does not match trajectory id`);
-}
-function defaultStepUri(trajectoryId, step) {
-	return `trace://${encodeURIComponent(trajectoryId)}/span/step-${step}`;
-}
-function optionalStepArray(value) {
-	return value === void 0 || Array.isArray(value) && value.every(Number.isSafeInteger);
-}
-function isRecord(value) {
-	return typeof value === "object" && value !== null && !Array.isArray(value);
-}
-function positiveStep(value, field) {
-	if (!Number.isSafeInteger(value) || value < 1) throw new RangeError(`${field} must be a positive safe integer`);
-	return value;
-}
-function externalId(value, field) {
-	if (typeof value !== "string" && typeof value !== "number") throw new TypeError(`${field} must be a string or number`);
-	if (typeof value === "number" && !Number.isSafeInteger(value)) throw new TypeError(`${field} must be a safe integer when numeric`);
-	return nonEmpty(String(value), field);
-}
-function nonEmpty(value, field) {
-	if (!value.trim()) throw new TypeError(`${field} must not be empty`);
-	return value;
-}
-//#endregion
-//#region src/analyst/benchmark-report.ts
-function renderAnalystBenchmarkMarkdown(result, comparisons = []) {
-	const { provenance } = result;
-	const lines = [
-		"# Trace analyst benchmark",
-		"",
-		"## Run",
-		"",
-		"| Field | Value |",
-		"| --- | --- |",
-		`| Benchmark | ${escapeCell(provenance.id ?? "unspecified")} |`,
-		`| Dataset | ${escapeCell(provenance.dataset?.id ?? "unspecified")} |`,
-		`| Dataset revision | ${escapeCell(provenance.dataset?.revision ?? "unspecified")} |`,
-		`| Dataset split | ${escapeCell(provenance.dataset?.split ?? "unspecified")} |`,
-		`| Started | ${escapeCell(provenance.startedAt)} |`,
-		`| Ended | ${escapeCell(provenance.endedAt)} |`,
-		`| Cases | ${provenance.caseCount} |`,
-		`| Runners | ${escapeCell(provenance.runnerIds.join(", "))} |`,
-		`| Repetitions | ${provenance.repetitions} |`,
-		`| Maximum concurrency | ${provenance.maxConcurrency} |`,
-		`| Runner-order seed | ${provenance.runnerOrderSeed} |`,
-		`| Command | ${escapeCell(provenance.command ?? "uncaptured")} |`,
-		`| Environment | ${escapeCell(json(provenance.environment))} |`,
-		`| Metadata | ${escapeCell(json(provenance.metadata))} |`,
-		"",
-		"## Summary",
-		""
-	];
-	lines.push("| Runner | Runs | Failed | Recall | Precision | F1 | Critical step | Citation coverage | Label-location agreement | Citation resolution | Resolution unknown runs | Unresolved citations | Resolution errors | Clean false positives | Clean failures | Repeat agreement | Latency min/mean/p50/p95/max ms | Calls | Input tokens | Output tokens | Reasoning tokens | Cached tokens | Cache-write tokens | Known cost USD | Unknown calls | Unknown input/output | Unknown reasoning | Unknown cached | Unknown cache-write | Unknown cost |", "| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |");
-	for (const summary of result.summaries) lines.push(`| ${escapeCell(summary.runnerId)} | ${summary.completedRuns}/${summary.plannedRuns} | ${summary.failedRuns} | ${optionalRate(summary.issueRecall)} | ${optionalRate(summary.findingPrecision)} | ${optionalRate(summary.f1)} | ${optionalRate(summary.criticalStepAccuracy)} | ${optionalRate(summary.citationCoverage)} | ${optionalRate(summary.citationLabelAgreement)} | ${optionalRate(summary.citationResolution)} | ${summary.citationResolutionUnknownRuns} | ${summary.unresolvedCitations} | ${summary.citationResolutionErrors} | ${optionalRate(summary.cleanCaseFalsePositiveRate)} | ${optionalRate(summary.cleanCaseFailureRate)} | ${optionalRate(summary.runAgreement)} | ${latency(summary.latencyMs)} | ${summary.calls} | ${summary.inputTokens} | ${summary.outputTokens} | ${summary.reasoningTokens} | ${summary.cachedTokens} | ${summary.cacheWriteTokens} | ${summary.knownCostUsd.toFixed(6)} | ${summary.callsUnknownRuns} | ${summary.tokenUsageUnknownRuns} | ${summary.reasoningTokenUsageUnknownRuns} | ${summary.cachedTokenUsageUnknownRuns} | ${summary.cacheWriteTokenUsageUnknownRuns} | ${summary.costUnknownRuns} |`);
-	for (const comparison of comparisons) {
-		lines.push("", `## ${escapeCell(comparison.candidateRunnerId)} compared with ${escapeCell(comparison.baselineRunnerId)}`, "", "| Metric | Better direction | Paired cases | Paired observations | Baseline mean | Candidate mean | Delta | Interval | At least 20 independent cases |", "| --- | --- | ---: | ---: | ---: | ---: | ---: | ---: | --- |");
-		for (const metric of comparison.metrics) lines.push(`| ${metric.metric} | ${metric.direction} | ${metric.pairedCases} | ${metric.pairedObservations} | ${number(metric.baselineMean)} | ${number(metric.candidateMean)} | ${signed(metric.meanDelta)} | [${number(metric.intervalLow)}, ${number(metric.intervalHigh)}] | ${metric.enoughCasesForInference ? "yes" : "no"} |`);
-	}
-	lines.push("", "## Runs", "", "| Runner | Case | Tags | Case metadata | Runner metadata | Rep | Execution index | Completed | Recall | Precision | F1 | Critical step | Citation coverage | Label-location agreement | Citation resolution | Clean false positive | Findings | Unlabeled citations | Unresolved citations | Resolution errors | Latency ms | Calls | Input tokens | Output tokens | Reasoning tokens | Cached tokens | Cache-write tokens | Cost USD | Known cost USD | Cost source | Error class | Error |", "| --- | --- | --- | --- | --- | ---: | ---: | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | --- | --- | --- |");
-	for (const observation of result.observations) {
-		const usage = observation.usage;
-		const cost = usage?.cost.kind === "uncaptured" ? null : usage?.cost.usd;
-		lines.push(`| ${escapeCell(observation.runnerId)} | ${escapeCell(observation.caseId)} | ${escapeCell(observation.caseTags.join(", "))} | ${escapeCell(json(observation.caseMetadata))} | ${escapeCell(json(observation.runnerMetadata))} | ${observation.repetition} | ${observation.executionIndex} | ${observation.error ? "no" : "yes"} | ${rate(observation.score.issueRecall)} | ${rate(observation.score.findingPrecision)} | ${rate(observation.score.f1)} | ${optionalRate(observation.score.criticalStepAccuracy)} | ${optionalRate(observation.score.citationCoverage)} | ${optionalRate(observation.score.citationLabelAgreement)} | ${optionalRate(observation.evidenceResolution?.validity ?? null)} | ${observation.score.expectedIssueCount === 0 ? observation.score.cleanFalsePositive ? "yes" : "no" : "n/a"} | ${observation.score.supportedFindingIndexes.length}/${observation.findings.length} | ${observation.score.unlabeledEvidence.length} | ${observation.evidenceResolution?.unresolvedEvidence.length ?? "unknown"} | ${observation.evidenceResolution?.errors.length ?? "unknown"} | ${number(observation.latencyMs)} | ${usage?.calls ?? "unknown"} | ${usage?.tokens?.input ?? "unknown"} | ${usage?.tokens?.output ?? "unknown"} | ${usage?.tokens?.reasoning ?? "unknown"} | ${usage?.tokens?.cached ?? "unknown"} | ${usage?.tokens?.cacheWrite ?? "unknown"} | ${cost === null || cost === void 0 ? "unknown" : cost.toFixed(6)} | ${usage?.knownCostUsd?.toFixed(6) ?? (cost === null || cost === void 0 ? "unknown" : cost.toFixed(6))} | ${usage?.cost.kind ?? "unknown"} | ${escapeCell(observation.error?.class ?? "")} | ${escapeCell(observation.error?.message ?? "")} |`);
-	}
-	return `${lines.join("\n")}\n`;
-}
-function rate(value) {
-	return `${(value * 100).toFixed(1)}%`;
-}
-function optionalRate(value) {
-	return value === null ? "n/a" : rate(value);
-}
-function number(value) {
-	return Number.isInteger(value) ? String(value) : value.toFixed(3);
-}
-function signed(value) {
-	return `${value >= 0 ? "+" : ""}${number(value)}`;
-}
-function latency(value) {
-	return [
-		value.min,
-		value.mean,
-		value.p50,
-		value.p95,
-		value.max
-	].map(number).join("/");
-}
-function json(value) {
-	return value === void 0 ? "uncaptured" : JSON.stringify(value);
-}
-function escapeCell(value) {
-	return value.replaceAll("|", "\\|").replaceAll("\n", " ");
-}
-//#endregion
-//#region src/analyst/define.ts
-/** Define a custom trace analyst without repeating fixed registry fields. */
-function defineTraceAnalyst(options) {
-	if (!options.id.trim()) throw new TypeError("defineTraceAnalyst: id must not be empty");
-	if (!options.description.trim()) throw new TypeError("defineTraceAnalyst: description must not be empty");
-	if (options.cost === void 0) throw new TypeError("defineTraceAnalyst: cost must be declared");
-	return {
-		id: options.id,
-		description: options.description,
-		version: options.version ?? "1.0.0",
-		inputKind: "trace-store",
-		cost: options.cost,
-		analyze: options.analyze
-	};
-}
-//#endregion
-export { ANALYST_SEVERITIES, AnalystRegistry, CONTROL_INTEGRITY_ANALYST, ControlIntegrityAnalyst, DEFAULT_TRACE_ANALYST_KINDS, FAILURE_MODE_KIND_SPEC, FINDING_SUBJECT_GRAMMAR_PROMPT, FINDING_SUBJECT_KINDS, FINDING_SUBJECT_SYNTAX, FindingSubjectStringSchema, FindingsStore, IMPROVEMENT_KIND_SPEC, KIND_EXPECTED_SUBJECTS, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, RAW_FINDING_SCHEMA_PROMPT, RawAnalystEvidenceSchema, RawAnalystFindingSchema, SKILL_USAGE_ANALYST, SkillUsageAnalyst, agentRxBenchmarkCase, agentRxPredictionsToFindings, assertProposalFindings, behavioralAnalyst, buildDefaultAnalystRegistry, buildSkillUsageReport, buildTraceToolsForGroup, codeTraceBenchCase, codeTracerPredictionsToFindings, coerceJson, coerceToFindingRows, compareAnalystRunners, computeFindingId, createAnalystAi, createChatClient, createJudgeAdapter, createRunCriticAdapter, createSemanticConceptJudgeAdapter, createTraceAnalystKind, createVerifierAdapter, defaultIsMaterial, defineTraceAnalyst, deriveEfficiencyFindings, diffFindings, emitControlIntegrityFindings, emitSkillUsageFindings, evidenceRefsFromRawFinding, findingSubjectGrammarPromptFor, isProposalFinding, liftSeverity, makeFinding, makeProposalFinding, normalizeAgentRxCategory, normalizeBenchmarkLabel, parseFindingSubject, parseRawFinding, registryBenchmarkRunner, renderAnalystBenchmarkMarkdown, renderFindingSubject, renderPriorFindings, renderUpstreamFindings, runAnalystBenchmark, scoreAnalystFindings, stripCodeFences, structureFindings, traceStoreEvidenceResolver };
+export { AGENT_RX_UPSTREAM_REVISION, ANALYST_BENCHMARK_COST_LEDGER_FILE, ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM, ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES, ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256, ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, ANALYST_BENCHMARK_EVIDENCE_PACKAGE_VERSION, ANALYST_BENCHMARK_HELP, ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM, ANALYST_BENCHMARK_IMPLEMENTATION_FILES, ANALYST_BENCHMARK_IMPLEMENTATION_SHA256, ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, ANALYST_BENCHMARK_MANIFEST_FILE, ANALYST_BENCHMARK_OBSERVATIONS_FILE, ANALYST_SEVERITIES, AnalystRegistry, CODE_TRACE_BENCH_ANALYST_PROMPT, CONTROL_INTEGRITY_ANALYST, ControlIntegrityAnalyst, DEFAULT_MAX_VERIFICATION_ARTIFACT_BYTES, DEFAULT_TRACE_ANALYST_KINDS, DEFAULT_TRACE_ANALYST_LIMITS, ExactAnalystRunExecutionError, FAILURE_MODE_KIND_SPEC, FINDING_SUBJECT_GRAMMAR_PROMPT, FINDING_SUBJECT_KINDS, FINDING_SUBJECT_SYNTAX, FindingSubjectStringSchema, FindingsStore, IMPROVEMENT_KIND_SPEC, KIND_EXPECTED_SUBJECTS, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, RAW_FINDING_SCHEMA_PROMPT, RawAnalystEvidenceSchema, RawAnalystFindingSchema, SKILL_USAGE_ANALYST, SkillUsageAnalyst, adaptPublicBenchmarkFindings, agentRxBenchmarkCase, agentRxPredictionsToFindings, analystBenchmarkDependencyLockDigest, analystBenchmarkImplementationDigest, appendVerificationArtifactsToOtlp, assertExactRegistryRunOpts, assertProposalFindings, behavioralAnalyst, buildDefaultAnalystRegistry, buildSkillUsageReport, buildTraceToolsForGroup, codeTraceBenchCase, codeTracerPredictionsToFindings, coerceJson, coerceToFindingRows, compareAnalystRunners, computeFindingId, createChatClient, createDspyRlmTraceEngine, createJudgeAdapter, createPublicBenchmarkDirectRunner, createPublicBenchmarkRlmRunner, createRunCriticAdapter, createSemanticConceptJudgeAdapter, createTraceAnalyst, createVerifierAdapter, defaultIsMaterial, defineCustomAnalyst, defineTraceAnalyst, deriveEfficiencyFindings, diffFindings, emitControlIntegrityFindings, emitSkillUsageFindings, emptyPublicBenchmarkRunner, evidenceRefsFromRawFinding, findingSubjectGrammarPromptFor, isProposalFinding, liftSeverity, loadCodeTraceVerificationArtifacts, loadPublicBenchmarkRows, makeFinding, makeProposalFinding, normalizeAgentRxCategory, normalizeBenchmarkLabel, parseFindingSubject, parseRawFinding, parseVerificationOutcome, preparePublicAnalystBenchmark, publicBenchmarkDistributions, publicBenchmarkProtocolSha256, publicBenchmarkSelectionReport, readAnalystBenchmarkArtifact, registryBenchmarkRunner, renderAgentRxCalibrationMarkdown, renderAnalystBenchmarkMarkdown, renderCodeTraceCalibrationMarkdown, renderFindingSubject, renderPriorFindings, renderUpstreamFindings, resolveTraceAnalystLimits, roundAgentRxStep, runAnalystBenchmark, runAnalystBenchmarkCommand, runTraceAnalyst, scoreAnalystFindings, selectPublicBenchmarkRows, stripCodeFences, summarizeAgentRxCalibration, summarizeCodeTraceCalibration, traceStoreEvidenceResolver };
 
 //# sourceMappingURL=index.js.map