@tangle-network/agent-eval 0.171.0 → 0.172.0

package/dist/ledger-core/index.d.ts

@@ -1,2 +1,3 @@
-import { A as jsonDocument, C as withLedgerFileLock, D as LedgerHash, E as LedgerCanonicalizationError, F as AtomicFileLockOwner, I as AtomicFileLockUnavailable, L as probeAtomicFileLock, M as AtomicFileLockAcquisition, N as AtomicFileLockError, O as canonicalString, P as AtomicFileLockOptions, R as tryAcquireAtomicFileLock, S as tryWithLedgerFileLock, T as LEDGER_HASH_PATTERN, _ as trustedHeadPathFor, a as LedgerChainFields, b as LedgerFileContext, c as LedgerJournalCodec, d as LedgerReplayResult, f as LedgerAnchoredEntry, g as readTrustedHeadFile, h as LedgerTrustedHeadSubject, i as LedgerAppendResult, j as AtomicFileLock, k as hashCanonical, l as LedgerLineContext, m as LedgerTrustedHeadRemoval, n as FileLedgerJournalOptions, o as LedgerEntryOf, p as LedgerTrustedHead, r as LedgerAppendOptions, s as LedgerEventBase, t as FileLedgerJournal, u as LedgerProjector, v as verifyEntriesAgainstTrustedHead, w as writeLedgerFileAtomically, x as appendLedgerLine, y as FileLockResult } from "../index-lfaSeKSD.js";
+import { A as AtomicFileLockUnavailable, C as withLedgerFileLock, D as AtomicFileLockError, E as AtomicFileLockAcquisition, M as tryAcquireAtomicFileLock, O as AtomicFileLockOptions, S as tryWithLedgerFileLock, T as AtomicFileLock, _ as trustedHeadPathFor, a as LedgerChainFields, b as LedgerFileContext, c as LedgerJournalCodec, d as LedgerReplayResult, f as LedgerAnchoredEntry, g as readTrustedHeadFile, h as LedgerTrustedHeadSubject, i as LedgerAppendResult, j as probeAtomicFileLock, k as AtomicFileLockOwner, l as LedgerLineContext, m as LedgerTrustedHeadRemoval, n as FileLedgerJournalOptions, o as LedgerEntryOf, p as LedgerTrustedHead, r as LedgerAppendOptions, s as LedgerEventBase, t as FileLedgerJournal, u as LedgerProjector, v as verifyEntriesAgainstTrustedHead, w as writeLedgerFileAtomically, x as appendLedgerLine, y as FileLockResult } from "../index-D-UdhAmg.js";
+import { a as hashCanonical, i as canonicalString, n as LedgerCanonicalizationError, o as jsonDocument, r as LedgerHash, t as LEDGER_HASH_PATTERN } from "../canonical-CFpojCN5.js";
 export { type AtomicFileLock, type AtomicFileLockAcquisition, AtomicFileLockError, type AtomicFileLockOptions, type AtomicFileLockOwner, type AtomicFileLockUnavailable, FileLedgerJournal, type FileLedgerJournalOptions, type FileLockResult, LEDGER_HASH_PATTERN, type LedgerAnchoredEntry, type LedgerAppendOptions, type LedgerAppendResult, LedgerCanonicalizationError, type LedgerChainFields, type LedgerEntryOf, type LedgerEventBase, type LedgerFileContext, type LedgerHash, type LedgerJournalCodec, type LedgerLineContext, type LedgerProjector, type LedgerReplayResult, type LedgerTrustedHead, type LedgerTrustedHeadRemoval, type LedgerTrustedHeadSubject, appendLedgerLine, canonicalString, hashCanonical, jsonDocument, probeAtomicFileLock, readTrustedHeadFile, trustedHeadPathFor, tryAcquireAtomicFileLock, tryWithLedgerFileLock, verifyEntriesAgainstTrustedHead, withLedgerFileLock, writeLedgerFileAtomically };

package/dist/{llm-judge-BtJ2Sfk_.js → llm-judge-aQHIk5_-.js}

@@ -1,6 +1,6 @@
 import { i as JudgeError, s as ValidationError, t as AgentEvalError } from "./errors-Dngq5h35.js";
 import { a as hashCanonical, i as compareCodeUnits, r as canonicalString } from "./canonical-DPyQ_rpt.js";
-import { s as weightedComposite, t as confidenceInterval } from "./descriptive-1V17A-qa.js";
+import { o as summarizeNumberSeries, s as weightedComposite, t as confidenceInterval } from "./descriptive-1V17A-qa.js";
 import { r as pairedBootstrap } from "./paired-tests-C8iCsioC.js";
 import { i as modelHasSnapshot } from "./run-record-DLORoL7t.js";
 import { n as contentHash } from "./verdict-cache-B3eCVQtY.js";
@@ -83,7 +83,8 @@ function computeAggregates(cells, judges, seed, cost) {
 		byScenario[scenarioId] = {
 			meanComposite: ag.mean,
 			ci95: ag.ci95,
-			n: ag.n
+			n: ag.n,
+			distribution: ag.distribution
 		};
 	}
 	const dispatchFailures = cells.filter((cell) => cell.errorStage === "dispatch");
@@ -107,8 +108,9 @@ function computeAggregates(cells, judges, seed, cost) {
 	};
 }
 function aggregate(samples, seed) {
-	const n = samples.length;
-	if (n === 0) throw new Error("aggregate requires at least one finite score");
+	const distribution = summarizeNumberSeries(samples);
+	if (distribution === null) throw new Error("aggregate requires at least one finite score");
+	const n = distribution.n;
 	const mean = samples.reduce((a, b) => a + b, 0) / n;
 	const variance = samples.reduce((a, b) => a + (b - mean) ** 2, 0) / Math.max(1, n - 1);
 	const stdev = Math.sqrt(variance);
@@ -120,7 +122,8 @@ function aggregate(samples, seed) {
 		mean,
 		stdev,
 		ci95: [ci.lower, ci.upper],
-		n
+		n,
+		distribution
 	};
 }
 //#endregion
@@ -3826,9 +3829,12 @@ function renderPrBody(result, gate, diff) {
 	lines.push("");
 	lines.push("### By-judge aggregates");
 	lines.push("");
-	lines.push("| judge | mean | ci95 | n |");
-	lines.push("|---|---|---|---|");
-	for (const [name, agg] of Object.entries(result.aggregates.byJudge)) lines.push(`| ${name} | ${agg.mean.toFixed(3)} | [${agg.ci95[0].toFixed(3)}, ${agg.ci95[1].toFixed(3)}] | ${agg.n} |`);
+	lines.push("| judge | mean | ci95 | min | p50 | p90 | max | n |");
+	lines.push("|---|---|---|---|---|---|---|---|");
+	for (const [name, agg] of Object.entries(result.aggregates.byJudge)) {
+		const d = agg.distribution;
+		lines.push(`| ${name} | ${agg.mean.toFixed(3)} | [${agg.ci95[0].toFixed(3)}, ${agg.ci95[1].toFixed(3)}] | ${d.min.toFixed(3)} | ${d.p50.toFixed(3)} | ${d.p90.toFixed(3)} | ${d.max.toFixed(3)} | ${agg.n} |`);
+	}
 	return lines.join("\n");
 }
 function defaultGhExec(args) {
@@ -6254,6 +6260,6 @@ function firstString(value) {
 	return typeof value === "string" && value.trim() ? value : void 0;
 }
 //#endregion
-export { redTeamDataset as $, optimizationTokenUsageFromSummary as A, createSearchHistoryReceipt as B, paretoFrontier as C, combineComparisonCosts as D, assertOptimizationResult as E, surfaceContentHash as F, compareRankKeys as G, verifySearchHistoryReceipt as H, surfaceHash as I, recoverTruncatedJson as J, buildReflectionPrompt as K, SearchHistoryRequiredError as L, codeSurfaceIdentityMaterial as M, componentSurfaceIdentityMaterial as N, compareOptimizationMethods as O, renderSurfaceDiff as P, DEFAULT_RED_TEAM_CORPUS as Q, assertCompleteSearchHistory as R, dominates as S, openAutoPr as T, campaignBreakdown as U, searchHistoryCoverageRow as V, campaignMeanComposite as W, readGepaCandidatePopulationArtifact as X, assertGepaCandidatePopulationSummary as Y, defaultProductionGate as Z, runOptimization as _, cellCachePath as _t, aggregateRunScore as a, planCampaignRun as at, SearchRecorder as b, campaignMeasurementDigest as c, BackendIntegrityError as ct, loopProvenanceArgsFromResult as d, assertCampaignDesign as dt, redTeamReport as et, loopProvenanceSpans as f, assertCampaignSplitIdentity as ft, runImprovementLoop as g, readCachedCell as gt, verifyLoopProvenanceRecord as h, campaignSplitDigestFromIdentities as ht, transientDispatchFailure as i, runCampaign as it, assertCodeSurfaceIdentity as j, costFromLedgerSummary as k, canonicalDigest as l, assertRealBackend as lt, provenanceSpansPath as m, campaignSplitDigest as mt, JudgeParseError as n, runCanaries as nt, clamp01 as o, resolveRunDir as ot, provenanceRecordPath as p, campaignScenarioIdentity as pt, parseReflectionResponse as q, isTransientTransportFailure as r, runEval as rt, buildLoopProvenanceRecord as s, tangleTracesRoot as st, llmJudge as t, scoreRedTeamOutput as tt, emitLoopProvenance as u, summarizeBackendIntegrity as ut, isProposedCandidate as v, computeManifestHash as vt, paretoFrontierWithCrowding as w, recordCandidatePopulationSearch as x, labelTrustRank as y, assertSearchHistoryMatchesReplay as z };
+export { redTeamDataset as $, optimizationTokenUsageFromSummary as A, createSearchHistoryReceipt as B, paretoFrontier as C, combineComparisonCosts as D, assertOptimizationResult as E, surfaceContentHash as F, compareRankKeys as G, verifySearchHistoryReceipt as H, surfaceHash as I, recoverTruncatedJson as J, buildReflectionPrompt as K, SearchHistoryRequiredError as L, codeSurfaceIdentityMaterial as M, componentSurfaceIdentityMaterial as N, compareOptimizationMethods as O, renderSurfaceDiff as P, DEFAULT_RED_TEAM_CORPUS as Q, assertCompleteSearchHistory as R, dominates as S, openAutoPr as T, campaignBreakdown as U, searchHistoryCoverageRow as V, campaignMeanComposite as W, readGepaCandidatePopulationArtifact as X, assertGepaCandidatePopulationSummary as Y, defaultProductionGate as Z, runOptimization as _, buildCellSchedule as _t, aggregateRunScore as a, planCampaignRun as at, SearchRecorder as b, campaignMeasurementDigest as c, BackendIntegrityError as ct, loopProvenanceArgsFromResult as d, assertCampaignDesign as dt, redTeamReport as et, loopProvenanceSpans as f, assertCampaignSplitIdentity as ft, runImprovementLoop as g, readCachedCell as gt, verifyLoopProvenanceRecord as h, campaignSplitDigestFromIdentities as ht, transientDispatchFailure as i, runCampaign as it, assertCodeSurfaceIdentity as j, costFromLedgerSummary as k, canonicalDigest as l, assertRealBackend as lt, provenanceSpansPath as m, campaignSplitDigest as mt, JudgeParseError as n, runCanaries as nt, clamp01 as o, resolveRunDir as ot, provenanceRecordPath as p, campaignScenarioIdentity as pt, parseReflectionResponse as q, isTransientTransportFailure as r, runEval as rt, buildLoopProvenanceRecord as s, tangleTracesRoot as st, llmJudge as t, scoreRedTeamOutput as tt, emitLoopProvenance as u, summarizeBackendIntegrity as ut, isProposedCandidate as v, cellCachePath as vt, paretoFrontierWithCrowding as w, recordCandidatePopulationSearch as x, labelTrustRank as y, computeManifestHash as yt, assertSearchHistoryMatchesReplay as z };
 
-//# sourceMappingURL=llm-judge-BtJ2Sfk_.js.map
+//# sourceMappingURL=llm-judge-aQHIk5_-.js.map