@tangle-network/agent-eval 0.161.1 → 0.170.0

package/dist/index.js

@@ -1,58 +1,57 @@
 import { t as __exportAll } from "./rolldown-runtime-8H4AJuhK.js";
 import { i as JudgeError, o as NotFoundError, r as ConfigError, s as ValidationError, t as AgentEvalError } from "./errors-Dngq5h35.js";
-import { r as canonicalString } from "./canonical-IL-Bu-14.js";
-import { a as CODING_HARNESSES, c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, n as completionVerdict, o as HARNESS_NATIVE_MODEL, r as createLlmCorrectnessChecker, s as agentProfileHash, t as extractProducedState, u as expandProfileAxes } from "./produced-state-DZ89riy5.js";
-import { n as hashJson, r as manifestContentDigest } from "./pre-registration-KN9jkh58.js";
+import { r as canonicalString } from "./canonical-DPyQ_rpt.js";
+import { a as CODING_HARNESSES, c as agentProfileId, d as harnessAxisOf, i as verifyCompletion, l as agentProfileModelId, n as completionVerdict, o as HARNESS_NATIVE_MODEL, r as createLlmCorrectnessChecker, s as agentProfileHash, t as extractProducedState, u as expandProfileAxes } from "./produced-state-CtSIp5cQ.js";
+import { n as hashJson, r as manifestContentDigest } from "./pre-registration-D94b7Of5.js";
 import { AGENT_PROFILE_KINDS, agentProfileCellHashMaterial, agentProfileCellKey, buildAgentProfileCell, groupRunsByAgentProfileCell, toAgentProfileJson, verifyAgentProfileCell } from "./profile-cell.js";
-import { c as mulberry32 } from "./internal-BDHPCnjk.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-jDOuI6mz.js";
-import { i as verbosityBias, n as calibrateJudgeContinuous, r as continuousAgreement, t as calibrateJudge } from "./judge-calibration-zZjLz8hr.js";
-import { a as budgetBreachView, c as corpusInterRaterAgreementFromJudgeScores, i as failureClusterView, l as interRaterReliability, n as computeToolUseMetrics, o as classifyFailure, r as judgeAgreementView, s as corpusInterRaterAgreement, t as toolWasteView } from "./tool-waste-BqzmVdJk.js";
+import { t as mulberry32 } from "./random-Dn5fPWkt.js";
+import { a as spearmanR, c as weightedMean, i as ranks, n as partialCredit, o as summarizeNumberSeries, r as pearsonR, s as weightedComposite, t as confidenceInterval } from "./descriptive-1V17A-qa.js";
+import { i as verbosityBias, n as calibrateJudgeContinuous, r as continuousAgreement, t as calibrateJudge } from "./judge-calibration-BnpVKtnb.js";
+import { a as budgetBreachView, c as corpusInterRaterAgreementFromJudgeScores, i as failureClusterView, l as interRaterReliability, n as computeToolUseMetrics, o as classifyFailure, r as judgeAgreementView, s as corpusInterRaterAgreement, t as toolWasteView } from "./tool-waste-CwGHzBzX.js";
 import { i as pairedCohensDz, n as cohensD, r as interpretCliffs, t as cliffsDelta } from "./effect-sizes-DiH8MGOH.js";
-import { a as requiredSampleSize, c as holm, i as requiredPairedSampleSize, n as mcnemarRequiredN, o as benjaminiHochberg, r as pairedMde, s as bonferroni, t as mcnemarPower } from "./power-and-mde-CHIrXJll.js";
-import { _ as passAtK, a as MANN_WHITNEY_EXACT_MAX_STATES, c as mannWhitneyU, d as isBinaryOutcomeVector, f as mcnemar, g as pairedRiskDifferenceScore, h as pairedRiskDifferenceExact, i as DEFAULT_PERMUTATIONS, l as wilcoxonSignedRank, m as pairedRiskDifference, n as pairArms, o as MANN_WHITNEY_EXACT_MAX_WORK, p as pairedBinaryScale, r as pairRunRecords, s as WILCOXON_EXACT_MAX_N, t as comparePairedArms, v as wilson } from "./paired-arms-D-XRF_fy.js";
-import { a as pairedSignTest, i as pairedDeltaTieFraction, n as DECISION_PAIRED_DELTA_STATISTIC, o as pairedTTest, r as pairedBootstrap, t as BOOTSTRAP_GATE_MIN_N } from "./paired-tests-BHIhYVdu.js";
+import { a as requiredSampleSize, c as holm, i as requiredPairedSampleSize, n as mcnemarRequiredN, o as benjaminiHochberg, r as pairedMde, s as bonferroni, t as mcnemarPower } from "./power-and-mde-B8F2RdcD.js";
+import { _ as passAtK, a as MANN_WHITNEY_EXACT_MAX_STATES, c as mannWhitneyU, d as isBinaryOutcomeVector, f as mcnemar, g as pairedRiskDifferenceScore, h as pairedRiskDifferenceExact, i as DEFAULT_PERMUTATIONS, l as wilcoxonSignedRank, m as pairedRiskDifference, n as pairArms, o as MANN_WHITNEY_EXACT_MAX_WORK, p as pairedBinaryScale, r as pairRunRecords, s as WILCOXON_EXACT_MAX_N, t as comparePairedArms, v as wilson } from "./paired-arms-D4aeIHUy.js";
+import { a as pairedSignTest, i as pairedDeltaTieFraction, n as DECISION_PAIRED_DELTA_STATISTIC, o as pairedTTest, r as pairedBootstrap, t as BOOTSTRAP_GATE_MIN_N } from "./paired-tests-C8iCsioC.js";
 import { t as eProcess } from "./sequential-eprocess-D1jKoihe.js";
-import { n as iqr, r as welchsTTest } from "./baseline-BhPRQBVn.js";
-import { a as isToolSpan, i as isLlmSpan, r as isJudgeSpan, t as FAILURE_CLASSES } from "./schema-k6ZBftVv.js";
-import { a as judgeSpans, c as runsForScenario, n as argHash } from "./query-CHmMP42p.js";
-import { i as analyzeRuns, o as checkCanaries, r as selfImprove, t as defineAgentEval } from "./define-agent-eval-h-s-sI-v.js";
+import { n as iqr, r as welchsTTest } from "./baseline-BC-eBZ7U.js";
+import { a as isToolSpan, i as isLlmSpan, r as isJudgeSpan, t as FAILURE_CLASSES } from "./schema-CdIX2aHu.js";
+import { d as runsForScenario, r as argHash, s as judgeSpans } from "./query-BPGMVlbM.js";
+import { i as analyzeRuns, o as checkCanaries, r as selfImprove, t as defineAgentEval } from "./define-agent-eval-Clj-8igZ.js";
 import { r as observedSplitScore, t as isRealnessGated } from "./reward-nw2xZGZG.js";
-import { a as parseRunRecordSafe, c as validateRunRecord, i as modelHasSnapshot, n as UNKNOWN_MODEL, o as roundTripRunRecord, r as isRunRecord, s as runTaskScore, t as RunRecordValidationError } from "./run-record-BC0ebuRP.js";
-import { a as summaryTable, n as gainHistogram, r as paretoChart } from "./summary-report-BI5hUtvK.js";
-import { n as contentHash, r as fileVerdictCache, t as canonicalJson } from "./verdict-cache-CdVVTVmn.js";
-import { $ as scoreRedTeamOutput, G as parseReflectionResponse, N as surfaceContentHash, Q as redTeamReport, W as buildReflectionPrompt, X as DEFAULT_RED_TEAM_CORPUS, Z as redTeamDataset, b as dominates, ct as summarizeBackendIntegrity, et as runCanaries, i as transientDispatchFailure, nt as runCampaign, ot as BackendIntegrityError, st as assertRealBackend, t as llmJudge, x as paretoFrontier } from "./llm-judge-BhasIPFT.js";
+import { a as parseRunRecordSafe, c as validateRunRecord, i as modelHasSnapshot, n as UNKNOWN_MODEL, o as roundTripRunRecord, r as isRunRecord, s as runTaskScore, t as RunRecordValidationError } from "./run-record-DLORoL7t.js";
+import { a as summaryTable, n as gainHistogram, r as paretoChart } from "./summary-report-Bgh8CpNK.js";
+import { n as contentHash, r as fileVerdictCache, t as canonicalJson } from "./verdict-cache-B3eCVQtY.js";
+import { $ as redTeamDataset, C as paretoFrontier, F as surfaceContentHash, K as buildReflectionPrompt, Q as DEFAULT_RED_TEAM_CORPUS, S as dominates, a as aggregateRunScore, ct as BackendIntegrityError, et as redTeamReport, i as transientDispatchFailure, it as runCampaign, lt as assertRealBackend, nt as runCanaries, o as clamp01, q as parseReflectionResponse, t as llmJudge, tt as scoreRedTeamOutput, ut as summarizeBackendIntegrity } from "./llm-judge-DbJdo8Nj.js";
 import { a as resolveModelPricing, i as isModelPriced, n as estimateCost, r as estimateTokens, t as MODEL_PRICING } from "./metrics-Qv-cpptD.js";
 import { a as CostLedgerPersistenceError, c as costForTokenPricing, i as CostLedger, l as costForUsage, n as CostCallConflictError, o as CostReceiptCaptureError, r as CostCeilingReachedError, s as CostReservationExceededError, t as CostAccountingIncompleteError, u as modelPriceKey } from "./cost-ledger-B1qx30B4.js";
 import { n as REDACTION_VERSION, r as redactString, t as DEFAULT_REDACTION_RULES } from "./redact-7Aq1ukl-.js";
-import { c as pairedDecisionShape, l as minimumPairsForPairedDeltaTest, s as decidePairedPromotion, u as pairedDeltaTest } from "./power-preflight-DEw-uC7q.js";
-import { n as clamp01, t as aggregateRunScore } from "./run-score-lDzV0X8j.js";
-import { n as makeFinding, r as makeProposalFinding, t as computeFindingId } from "./types-BI4fT3HN.js";
-import { n as equivalenceVerdict, t as certificationEvidenceDigest } from "./verdict-B0xltqu6.js";
-import { t as TraceEmitter } from "./emitter-BpYFQPj4.js";
+import { c as pairedDecisionShape, l as minimumPairsForPairedDeltaTest, s as decidePairedPromotion, u as pairedDeltaTest } from "./power-preflight-CFXm0Vjo.js";
+import { i as makeProposalFinding, n as computeFindingId, r as makeFinding } from "./types-CiWITkGo.js";
+import { n as equivalenceVerdict, t as certificationEvidenceDigest } from "./verdict-BQ3pCFf8.js";
+import { t as TraceEmitter } from "./emitter-DeQHiDMm.js";
 import { t as buildTrajectory } from "./trajectory-D_7rLrvE.js";
-import { t as runCounterfactual } from "./counterfactual-D_VWavVm.js";
-import { c as createBoundedTraceAnalysisStore, f as DEFAULT_TRACE_ANALYST_BUDGETS, p as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, t as createTraceAnalyst } from "./kind-factory-DY8FdoXf.js";
-import { _ as assertUniqueFindingIds, a as DEFAULT_TRACE_ANALYST_KINDS, b as snapshotAnalystRun, g as analystRunDigest, h as analystFindingDigest, l as FAILURE_MODE_KIND_SPEC, n as buildDefaultAnalystRegistry, r as AnalystRegistry, t as createChatClient, v as completedAnalystReviewQuality, x as validateAnalystReviewDecisions, y as readAnalystReview } from "./chat-client-DlMlAeYI.js";
+import { t as runCounterfactual } from "./counterfactual-Bjq1mlUu.js";
+import { c as createBoundedTraceAnalysisStore, f as DEFAULT_TRACE_ANALYST_BUDGETS, p as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, t as createTraceAnalyst } from "./kind-factory-DMeEoMQZ.js";
+import { _ as assertUniqueFindingIds, a as DEFAULT_TRACE_ANALYST_KINDS, b as snapshotAnalystRun, g as analystRunDigest, h as analystFindingDigest, l as FAILURE_MODE_KIND_SPEC, n as buildDefaultAnalystRegistry, r as AnalystRegistry, t as createChatClient, v as completedAnalystReviewQuality, x as validateAnalystReviewDecisions, y as readAnalystReview } from "./chat-client-DEtybj5i.js";
 import { OUTPUT_VALUE } from "./trace-attributes.js";
 import { r as extractUsageFromSse, t as extractUsage } from "./extract-usage-BrQ8mCLX.js";
 import { n as InMemoryRawProviderSink, r as NoopRawProviderSink, t as FileSystemRawProviderSink } from "./raw-provider-sink-BQd7mzyT.js";
-import { S as captureFetchToRawSink, d as inferDomainKeywords, h as analyzeTraces, l as describeTraceInsightScope, m as tokenizeDomainWords, n as toolSpansToTraceAnalysisStore, o as buildTraceInsightContext, p as scoreTraceInsightReadiness, s as buildTraceInsightPrompt, u as domainEvidencePattern, y as exportRunAsOtlp } from "./store-tool-spans-Dlh9vkFK.js";
+import { d as inferDomainKeywords, h as analyzeTraces, l as describeTraceInsightScope, m as tokenizeDomainWords, n as toolSpansToTraceAnalysisStore, o as buildTraceInsightContext, p as scoreTraceInsightReadiness, s as buildTraceInsightPrompt, u as domainEvidencePattern, w as captureFetchToRawSink, y as exportRunAsOtlp } from "./store-tool-spans-B9o6tU8f.js";
 import { n as assertRunCaptured, t as RunIntegrityError } from "./integrity-Cy9WHAtb.js";
 import { n as InMemoryTraceStore, t as FileSystemTraceStore } from "./store-DNe_Uv1Q.js";
 import { t as packageVersion$1 } from "./package-version-D7lQHt_-.js";
-import { t as runEvalCampaign } from "./eval-campaign-BsXWL2-2.js";
-import { i as improvementVerdict, n as computeExperimentStats } from "./experiment-tracker-Ym6rEQT1.js";
-import "./rollout-Dm2tSdiQ.js";
-import { t as mintRolloutRows } from "./mint-DfODW1KW.js";
+import { t as runEvalCampaign } from "./eval-campaign-JDTeE6Pl.js";
+import { i as improvementVerdict, n as computeExperimentStats } from "./experiment-tracker-BKEumQug.js";
+import "./rollout-Crypdx8s.js";
+import { t as mintRolloutRows } from "./mint-DjfDUMHr.js";
 import { n as pairedEvalueSequence, t as evaluateInterimReleaseConfidence } from "./sequential-CzK5DarL.js";
-import { S as judgeFamily, _ as assertServedModels, b as CrossFamilyError, d as stripFencedJson, f as ModelSubstitutionError, h as assertServedModel, i as backoffMs, l as isTransientLlmError, m as assertCrossFamilyServed, o as costReceiptFromLlm, p as ServedCrossFamilyError, r as LlmResponseError, s as costReceiptFromLlmError, t as LlmCallError, u as maximumChargeForLlmRequest, v as checkServedModel, x as assertCrossFamily, y as servedModelAcceptable } from "./llm-client-hgDieDNN.js";
-import { t as createDspyRlmTraceEngine } from "./dspy-rlm-engine-DptEII26.js";
-import { a as diffFindings, n as runSemanticConceptJudge, r as FindingsStore, t as SEMANTIC_CONCEPT_JUDGE_VERSION } from "./semantic-concept-judge-BSkKKHeq.js";
+import { S as judgeFamily, _ as assertServedModels, b as CrossFamilyError, d as stripFencedJson, f as ModelSubstitutionError, h as assertServedModel, i as backoffMs, l as isTransientLlmError, m as assertCrossFamilyServed, o as costReceiptFromLlm, p as ServedCrossFamilyError, r as LlmResponseError, s as costReceiptFromLlmError, t as LlmCallError, u as maximumChargeForLlmRequest, v as checkServedModel, x as assertCrossFamily, y as servedModelAcceptable } from "./llm-client-BFMRpmqb.js";
+import { t as createDspyRlmTraceEngine } from "./dspy-rlm-engine-CS3qcCEk.js";
+import { a as diffFindings, n as runSemanticConceptJudge, r as FindingsStore, t as SEMANTIC_CONCEPT_JUDGE_VERSION } from "./semantic-concept-judge-I36eejJx.js";
 import { t as analyzeSeries } from "./series-convergence-CjO2QdRW.js";
-import { i as otlpTextToTraceAnalysisStore, n as OtlpFileTraceStore } from "./store-otlp-C_Rq5I4D.js";
-import { t as paidJsonChat } from "./chat-json-call-6g5sJobJ.js";
-import { n as evaluateReleaseConfidence, r as bootstrapCi } from "./release-confidence-DKfD2RYU.js";
+import { i as otlpTextToTraceAnalysisStore, n as OtlpFileTraceStore } from "./store-otlp-CHjBvWQY.js";
+import { t as paidJsonChat } from "./chat-json-call-5Jxna-aV.js";
+import { n as evaluateReleaseConfidence, r as bootstrapCi } from "./release-confidence-CzUHc4z4.js";
 import { 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";
@@ -3552,7 +3551,7 @@ function scoreKnowledgeReadiness(options) {
 		userAnswers: options.userAnswers ?? {},
 		missing,
 		readinessScore,
-		metadata: options.metadata
+		...options.metadata !== void 0 ? { metadata: options.metadata } : {}
 	};
 	const recommendedAction = chooseRecommendedAction(blockingMissingRequirements, nonBlockingGaps);
 	const severity = blockingMissingRequirements.length > 0 ? "critical" : nonBlockingGaps.some((gap) => gap.importance === "high") ? "warning" : "info";
@@ -3628,7 +3627,7 @@ function acquisitionPlansForKnowledgeGaps(gaps) {
 		mode,
 		description: descriptionForPlan(mode, requirements),
 		priority: maxImportance(requirements.map((r) => r.importance)),
-		questions: mode === "ask_user" ? userQuestionsForKnowledgeGaps(requirements) : void 0
+		...mode === "ask_user" ? { questions: userQuestionsForKnowledgeGaps(requirements) } : {}
 	}));
 }
 function normalizeRequirement(requirement) {
@@ -3783,8 +3782,15 @@ function isFiniteNumber(value) {
 * fails loud when the inputs can't support a trustworthy held-out comparison
 * (duplicate ids, or a holdout set below the significance floor).
 */
-/** 32-bit FNV-1a hash. Stable, allocation-free, deterministic across runtimes —
-*  the same id+seed maps to the same bucket on every machine and process. */
+/** 32-bit FNV-1a over UTF-16 code units masked to a byte. Stable,
+*  allocation-free, deterministic across runtimes — the same id+seed maps to
+*  the same bucket on every machine and process.
+*
+*  Frozen: this value picks the held-out bucket, so changing the masking or
+*  the accumulator re-partitions every split already decided under it and
+*  silently invalidates the comparisons made against them. The other FNV-1a
+*  loops in this package mask differently and are not interchangeable with
+*  this one. */
 function fnv1a32(input) {
 	let h = 2166136261;
 	for (let i = 0; i < input.length; i++) {