@tangle-network/agent-eval 0.222.1 → 0.223.0

package/dist/{benchmark-command-Cr7i7YSX.js → benchmark-command-CH86twZj.js}

@@ -1,7 +1,8 @@
 import { c as errorMessage, s as ValidationError, t as AgentEvalError } from "./errors-DeEEV9Vx.js";
 import { a as hashCanonical, i as compareCodeUnits, o as jsonDocument, r as canonicalString } from "./canonical-CicJlINA.js";
 import { o as summarizeNumberSeries } from "./descriptive-cPwITVfy.js";
-import { i as pairedBootstrap, t as mean$3 } from "./mean-CdDYG-us.js";
+import { r as pairedBootstrap } from "./paired-tests-C2zUqgUW.js";
+import { n as meanOrNull, t as mean$1 } from "./mean-BPuUNznQ.js";
 import { o as withLedgerFileLock, s as writeLedgerFileAtomically } from "./journal-C1uLJi0w.js";
 import { a as resolveModelPricing } from "./metrics-DNCAUtuB.js";
 import { a as CostLedgerPersistenceError, i as CostCeilingReachedError, n as CostAccountingIncompleteError, o as CostReceiptCaptureError, r as CostCallConflictError, s as CostReservationExceededError, t as CostLedger } from "./cost-ledger-CS8R3IZj.js";
@@ -123,9 +124,9 @@ function summarizeAnalystBenchmarkRunner(runnerId, observations) {
 	const supportedFindings = issueBearing.reduce((sum, observation) => sum + observation.score.supportedFindingIndexes.length, 0);
 	const issueRecall = expectedIssues === 0 ? null : matchedIssues / expectedIssues;
 	const findingPrecision = expectedIssues === 0 ? null : issueFindings === 0 ? 0 : supportedFindings / issueFindings;
-	const macroIssueRecall = issueBearing.length === 0 ? null : mean$2(issueBearing.map((observation) => observation.score.issueRecall));
-	const macroFindingPrecision = issueBearing.length === 0 ? null : mean$2(issueBearing.map((observation) => observation.score.findingPrecision));
-	const macroF1 = issueBearing.length === 0 ? null : mean$2(issueBearing.map((observation) => observation.score.f1));
+	const macroIssueRecall = issueBearing.length === 0 ? null : mean(issueBearing.map((observation) => observation.score.issueRecall));
+	const macroFindingPrecision = issueBearing.length === 0 ? null : mean(issueBearing.map((observation) => observation.score.findingPrecision));
+	const macroF1 = issueBearing.length === 0 ? null : mean(issueBearing.map((observation) => observation.score.f1));
 	const critical = observations.map((observation) => observation.score.criticalStepAccuracy).filter((value) => value !== null);
 	const findingsWithEvidence = completed.reduce((sum, observation) => sum + observation.findings.filter((finding) => finding.evidence_refs.length > 0).length, 0);
 	const allFindings = completed.reduce((sum, observation) => sum + observation.findings.length, 0);
@@ -163,7 +164,7 @@ function summarizeAnalystBenchmarkRunner(runnerId, observations) {
 		macroIssueRecall,
 		macroFindingPrecision,
 		macroF1,
-		criticalStepAccuracy: critical.length === 0 ? null : mean$2(critical),
+		criticalStepAccuracy: critical.length === 0 ? null : mean(critical),
 		citationCoverage: allFindings === 0 ? null : findingsWithEvidence / allFindings,
 		citationExcerptCoverage: allCitations.length === 0 ? null : allCitations.filter((evidence) => Boolean(evidence.excerpt?.trim())).length / allCitations.length,
 		citationLabelAgreement: citationObservations.length === 0 ? null : citationCount === 0 ? 0 : (citationCount - invalidCitationCount) / citationCount,
@@ -198,7 +199,7 @@ function summarizeAnalystBenchmarkRunner(runnerId, observations) {
 		costUnknownRuns: usages.filter((usage) => !usage || usage.cost.kind === "uncaptured").length
 	};
 }
-function mean$2(values) {
+function mean(values) {
 	return values.length === 0 ? 0 : stableSum(values) / values.length;
 }
 function stableSum(values) {
@@ -217,7 +218,7 @@ function latencyDistribution(values) {
 	if (!summary) return null;
 	return {
 		min: summary.min,
-		mean: mean$2(values),
+		mean: mean(values),
 		p50: summary.p50,
 		p95: summary.percentiles[95],
 		max: summary.max
@@ -234,10 +235,10 @@ function repeatedObservationAgreement(observations, signature) {
 	for (const rows of byCase.values()) {
 		const agreements = [];
 		for (let left = 0; left < rows.length; left++) for (let right = left + 1; right < rows.length; right++) agreements.push(jaccard(signature(rows[left]), signature(rows[right])));
-		if (agreements.length > 0) caseAgreements.push(mean$2(agreements));
+		if (agreements.length > 0) caseAgreements.push(mean(agreements));
 	}
 	return {
-		value: caseAgreements.length === 0 ? null : mean$2(caseAgreements),
+		value: caseAgreements.length === 0 ? null : mean(caseAgreements),
 		cases: caseAgreements.length
 	};
 }
@@ -1041,20 +1042,20 @@ function summarizeRunner$1(runnerId, observations) {
 		failedRuns: observations.filter((observation) => Boolean(observation.error)).length,
 		predictedRuns: predicted.length,
 		missingPredictionRuns: scored.length - predicted.length,
-		exactStepAccuracy: mean$1(scored.map((row) => Number(row.roundedDistance === 0))),
-		stepAccuracyWithin1: mean$1(scored.map((row) => Number(row.roundedDistance !== null && row.roundedDistance <= 1))),
-		stepAccuracyWithin2: mean$1(scored.map((row) => Number(row.roundedDistance !== null && row.roundedDistance <= 2))),
-		stepAccuracyWithin3: mean$1(scored.map((row) => Number(row.roundedDistance !== null && row.roundedDistance <= 3))),
-		stepAccuracyWithin4: mean$1(scored.map((row) => Number(row.roundedDistance !== null && row.roundedDistance <= 4))),
-		stepAccuracyWithin5: mean$1(scored.map((row) => Number(row.roundedDistance !== null && row.roundedDistance <= 5))),
-		meanStepDistance: mean$1(predicted.map((row) => row.distance)),
-		normalizedMeanStepDistance: mean$1(normalized.map((row) => row.normalizedDistance)),
+		exactStepAccuracy: meanOrNull(scored.map((row) => Number(row.roundedDistance === 0))),
+		stepAccuracyWithin1: meanOrNull(scored.map((row) => Number(row.roundedDistance !== null && row.roundedDistance <= 1))),
+		stepAccuracyWithin2: meanOrNull(scored.map((row) => Number(row.roundedDistance !== null && row.roundedDistance <= 2))),
+		stepAccuracyWithin3: meanOrNull(scored.map((row) => Number(row.roundedDistance !== null && row.roundedDistance <= 3))),
+		stepAccuracyWithin4: meanOrNull(scored.map((row) => Number(row.roundedDistance !== null && row.roundedDistance <= 4))),
+		stepAccuracyWithin5: meanOrNull(scored.map((row) => Number(row.roundedDistance !== null && row.roundedDistance <= 5))),
+		meanStepDistance: meanOrNull(predicted.map((row) => row.distance)),
+		normalizedMeanStepDistance: meanOrNull(normalized.map((row) => row.normalizedDistance)),
 		normalizedDistanceRuns: normalized.length,
 		normalizedDistanceUnknownRuns: scored.length - normalized.length,
-		rootCauseCategoryAccuracy: mean$1(scored.map((row) => Number(row.rootCategoryMatch))),
-		anyFailureCategoryAccuracy: mean$1(scored.map((row) => Number(row.anyCategoryMatch))),
-		earliestFailureCategoryAccuracy: mean$1(scored.map((row) => Number(row.earliestCategoryMatch))),
-		terminalFailureCategoryAccuracy: mean$1(scored.map((row) => Number(row.terminalCategoryMatch)))
+		rootCauseCategoryAccuracy: meanOrNull(scored.map((row) => Number(row.rootCategoryMatch))),
+		anyFailureCategoryAccuracy: meanOrNull(scored.map((row) => Number(row.anyCategoryMatch))),
+		earliestFailureCategoryAccuracy: meanOrNull(scored.map((row) => Number(row.earliestCategoryMatch))),
+		terminalFailureCategoryAccuracy: meanOrNull(scored.map((row) => Number(row.terminalCategoryMatch)))
 	};
 }
 function scoredObservation(observation) {
@@ -1101,9 +1102,6 @@ function requiredStringArray(value, caseId, field) {
 	if (!Array.isArray(value) || value.length === 0 || value.some((entry) => typeof entry !== "string" || !entry.trim())) throw new TypeError(`${caseId}: ${field} must be a non-empty string array`);
 	return value;
 }
-function mean$1(values) {
-	return values.length === 0 ? null : values.reduce((total, value) => total + value, 0) / values.length;
-}
 function rate$3(value) {
 	return value === null ? "n/a" : value.toFixed(3);
 }
@@ -1212,8 +1210,8 @@ function compareMetric(options) {
 		if (caseBefore.length === 0 || !clusterId) continue;
 		pairedCases.push({
 			clusterId,
-			baseline: mean$3(caseBefore),
-			candidate: mean$3(caseAfter)
+			baseline: mean$1(caseBefore),
+			candidate: mean$1(caseAfter)
 		});
 	}
 	const byCluster = /* @__PURE__ */ new Map();
@@ -1222,8 +1220,8 @@ function compareMetric(options) {
 		rows.push(pairedCase);
 		byCluster.set(pairedCase.clusterId, rows);
 	}
-	const before = [...byCluster.values()].map((rows) => mean$3(rows.map((row) => row.baseline)));
-	const after = [...byCluster.values()].map((rows) => mean$3(rows.map((row) => row.candidate)));
+	const before = [...byCluster.values()].map((rows) => mean$1(rows.map((row) => row.baseline)));
+	const after = [...byCluster.values()].map((rows) => mean$1(rows.map((row) => row.candidate)));
 	const interval = before.length === 0 ? null : pairedBootstrap(before, after, {
 		confidence: options.confidence,
 		resamples: options.resamples,
@@ -1246,8 +1244,8 @@ function compareMetric(options) {
 		candidateMissingObservations,
 		asymmetricMissingObservations,
 		survivorOnly,
-		baselineMean: before.length === 0 ? null : mean$3(before),
-		candidateMean: after.length === 0 ? null : mean$3(after),
+		baselineMean: before.length === 0 ? null : mean$1(before),
+		candidateMean: after.length === 0 ? null : mean$1(after),
 		meanDelta: interval?.mean ?? null,
 		intervalLow: interval?.low ?? null,
 		intervalHigh: interval?.high ?? null,
@@ -2019,7 +2017,7 @@ const ANALYST_BENCHMARK_IMPLEMENTATION_FILES = Object.freeze([
 	"src/trace/redaction/detectors.ts",
 	"src/verdict-cache.ts"
 ]);
-const ANALYST_BENCHMARK_IMPLEMENTATION_SHA256 = "6e7968e5427d669897471f5c1225846769c490164022c59efbe0e548e4c76ca7";
+const ANALYST_BENCHMARK_IMPLEMENTATION_SHA256 = "63f2202ac8fbdd98316c66e119824968f6c249f8377d09e6a828077868b1ee66";
 // src/analyst/benchmark-evidence-validation.ts
 const MIN_ACTION_EXCERPT_CHARACTERS = 12;
 const MAX_LABEL_SCAN_DEPTH = 64;
@@ -3445,7 +3443,7 @@ function summarizeRunner(runnerId, observations) {
 		expectedIncorrectSteps: expected,
 		predictedIncorrectSteps: predicted,
 		matchedIncorrectSteps: matched,
-		officialAllRowF1: mean(officialRows),
+		officialAllRowF1: meanOrNull(officialRows),
 		officialAllRowRuns: officialRows.length,
 		precision,
 		recall,
@@ -3483,9 +3481,6 @@ function sum(values) {
 function ratio(numerator, denominator) {
 	return denominator === 0 ? null : numerator / denominator;
 }
-function mean(values) {
-	return values.length === 0 ? null : sum(values) / values.length;
-}
 function harmonicMean(left, right) {
 	if (left === null || right === null) return null;
 	return left + right === 0 ? 0 : 2 * left * right / (left + right);
@@ -6859,4 +6854,4 @@ function shellQuote(value) {
 }
 export { agentRxBenchmarkCase as A, ANALYST_BENCHMARK_COST_LEDGER_FILE as C, AGENT_RX_UPSTREAM_REVISION as D, ANALYST_BENCHMARK_OBSERVATIONS_FILE as E, runAnalystBenchmark as F, traceStoreEvidenceResolver as I, normalizeAgentRxCategory as M, externalId as N, codeTraceBenchCase as O, positiveStep as P, ANALYST_BENCHMARK_IMPLEMENTATION_FILES as S, ANALYST_BENCHMARK_MANIFEST_FILE as T, readAnalystBenchmarkArtifact as _, runInlineAnalystDefinition as a, ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256 as b, publicRlmAnalystDefinition as c, publicDirectAnalystDefinition as d, runChunkedAnalystDefinition as f, expandCodeTraceFailureBlocks as g, analystDefinitionProtocolSha256 as h, primeCodeTraceAnalystDefinition as i, agentRxPredictionsToFindings as j, codeTracerPredictionsToFindings as k, runReplVariableAnalystDefinition as l, analystDefinitionAsymmetries as m, createPrimeBenchmarkRunner as n, loadPublicBenchmarkRows as o, AnalystExpressivenessError as p, primeAnalystProtocolSha256 as r, createPublicBenchmarkRlmRunner as s, runAnalystBenchmarkCommand as t, createPublicBenchmarkDirectRunner as u, CODE_TRACE_BENCH_ANALYST_PROMPT as v, ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE as w, ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256 as x, publicBenchmarkProtocolSha256 as y };
 
-//# sourceMappingURL=benchmark-command-Cr7i7YSX.js.map
+//# sourceMappingURL=benchmark-command-CH86twZj.js.map