@tangle-network/agent-eval 1.0.12 → 1.0.13

package/dist/analyst/index.js

@@ -1,20 +1,21 @@
-import { n as assertConsistentFindingConditions, o as mergeFindingComposition, r as findingCompositionKey, t as TREATMENT_LAYERS } from "../finding-composition-DlvE1Qul.js";
-import { i as nominalAgreement } from "../agreement-irr-w-t3XUTD.js";
+import { n as assertConsistentFindingConditions, o as mergeFindingComposition, r as findingCompositionKey, t as TREATMENT_LAYERS } from "../finding-composition-Belh_qU3.js";
+import { c as clusterBootstrap } from "../mean-D1d6gedE.js";
+import { i as nominalAgreement } from "../agreement-irr-DXcXEt4I.js";
+import { i as rocAuc, n as diffFindings, r as createChatTraceEngine, t as FindingsStore } from "../findings-store-DWiBvJVM.js";
 import { t as FileLedgerJournal } from "../journal-C1uLJi0w.js";
 import { r as merkleRoot } from "../ledger-core-CfT-FzPf.js";
-import { A as spanningWindow, C as UNKNOWN_LINEAGE, D as mostSevere, E as mergeEvidenceUnits, O as parseAnalystFinding, S as ProposalFindingSchema, T as lineageRoots, b as AnalystFindingSchema, k as readFindingScope, w as independentUnitCount, x as FINDING_LAYERS } from "../external-optimizer-subprocess-CeHuAaMJ.js";
-import { a as makeProposalFinding, i as makeFinding, r as computeFindingId } from "../types-BK0blWhe.js";
-import { t as assertProposalFindings } from "../proposal-findings-BclX0PFh.js";
-import { A as RawAnalystFindingSchema, B as parseFindingSubject, F as stripCodeFences, I as FINDING_SUBJECT_KINDS, L as FINDING_SUBJECT_SYNTAX, M as parseRawFinding, N as parseTraceSpanEvidenceUri, R as KIND_EXPECTED_SUBJECTS, V as renderFindingSubject, j as evidenceRefsFromRawFinding, k as RAW_FINDING_SCHEMA_PROMPT, n as runTraceAnalyst, r as buildTraceToolsForGroup, t as createTraceAnalyst, z as findingSubjectGrammarPromptFor } from "../kind-factory-VKysp1ds.js";
-import { _ as behavioralTraceSupportSchema, a as KNOWLEDGE_POISONING_KIND_SPEC, b as behavioralTraceQuestionDigest, c as FAILURE_MODE_KIND_SPEC, d as emitControlIntegrityFindings, f as BEHAVIORAL_FAILURE_CLASS, g as validBehavioralCalibration, h as pinBehavioralFailureClass, i as DEFAULT_TRACE_ANALYST_KINDS, l as CONTROL_INTEGRITY_ANALYST, m as behavioralAnalyst, n as AnalystRegistry, o as KNOWLEDGE_GAP_KIND_SPEC, p as assertFailureClasses, r as ExactAnalystRunExecutionError, s as IMPROVEMENT_KIND_SPEC, t as buildDefaultAnalystRegistry, u as ControlIntegrityAnalyst, v as BEHAVIORAL_TRACE_QUESTIONS, y as BEHAVIORAL_TRACE_QUESTIONS_VERSION } from "../default-registry-gmDeN0KT.js";
-import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-Xt3znknf.js";
-import { n as diffFindings, r as createChatTraceEngine, t as FindingsStore } from "../findings-store-Bp-0Guuv.js";
-import { a as systemOneAnalyst, p as validDecisionIdentity } from "../decision-point-CErE7GnP.js";
+import { A as spanningWindow, C as UNKNOWN_LINEAGE, D as mostSevere, E as mergeEvidenceUnits, O as parseAnalystFinding, S as ProposalFindingSchema, T as lineageRoots, b as AnalystFindingSchema, k as readFindingScope, w as independentUnitCount, x as FINDING_LAYERS } from "../external-optimizer-subprocess-B44kOfZb.js";
+import { a as makeProposalFinding, i as makeFinding, r as computeFindingId } from "../types-CL1_N7AZ.js";
+import { t as assertProposalFindings } from "../proposal-findings-CZJkS_GY.js";
+import { A as RawAnalystFindingSchema, B as parseFindingSubject, F as stripCodeFences, I as FINDING_SUBJECT_KINDS, L as FINDING_SUBJECT_SYNTAX, M as parseRawFinding, N as parseTraceSpanEvidenceUri, R as KIND_EXPECTED_SUBJECTS, V as renderFindingSubject, j as evidenceRefsFromRawFinding, k as RAW_FINDING_SCHEMA_PROMPT, n as runTraceAnalyst, r as buildTraceToolsForGroup, t as createTraceAnalyst, z as findingSubjectGrammarPromptFor } from "../kind-factory-Cfz8DUCc.js";
+import { _ as behavioralTraceSupportSchema, a as KNOWLEDGE_POISONING_KIND_SPEC, b as behavioralTraceQuestionDigest, c as FAILURE_MODE_KIND_SPEC, d as emitControlIntegrityFindings, f as BEHAVIORAL_FAILURE_CLASS, g as validBehavioralCalibration, h as pinBehavioralFailureClass, i as DEFAULT_TRACE_ANALYST_KINDS, l as CONTROL_INTEGRITY_ANALYST, m as behavioralAnalyst, n as AnalystRegistry, o as KNOWLEDGE_GAP_KIND_SPEC, p as assertFailureClasses, r as ExactAnalystRunExecutionError, s as IMPROVEMENT_KIND_SPEC, t as buildDefaultAnalystRegistry, u as ControlIntegrityAnalyst, v as BEHAVIORAL_TRACE_QUESTIONS, y as BEHAVIORAL_TRACE_QUESTIONS_VERSION } from "../default-registry-eCkZFfuF.js";
+import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-_MEe2jIO.js";
+import { m as validDecisionIdentity, o as systemOneAnalyst, t as DECISION_POINT_AUC_BAR } from "../decision-point-CNyoKOkz.js";
 import { t as createChatClient } from "../chat-client-D0me8iwq.js";
-import { A as agentRxBenchmarkCase, C as ANALYST_BENCHMARK_COST_LEDGER_FILE, D as AGENT_RX_UPSTREAM_REVISION, E as ANALYST_BENCHMARK_OBSERVATIONS_FILE, F as runAnalystBenchmark, I as traceStoreEvidenceResolver, M as normalizeAgentRxCategory, N as externalId, O as codeTraceBenchCase, P as positiveStep, S as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, T as ANALYST_BENCHMARK_MANIFEST_FILE, _ as readAnalystBenchmarkArtifact, a as runInlineAnalystDefinition, b as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, c as publicRlmAnalystDefinition, d as publicDirectAnalystDefinition, f as runChunkedAnalystDefinition, g as expandCodeTraceFailureBlocks, h as analystDefinitionProtocolSha256, i as primeCodeTraceAnalystDefinition, j as agentRxPredictionsToFindings, k as codeTracerPredictionsToFindings, l as runReplVariableAnalystDefinition, m as analystDefinitionAsymmetries, n as createPrimeBenchmarkRunner, o as loadPublicBenchmarkRows, p as AnalystExpressivenessError, r as primeAnalystProtocolSha256, s as createPublicBenchmarkRlmRunner, t as runAnalystBenchmarkCommand, u as createPublicBenchmarkDirectRunner, v as CODE_TRACE_BENCH_ANALYST_PROMPT, w as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, x as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, y as publicBenchmarkProtocolSha256 } from "../benchmark-command-DZQrxqUL.js";
+import { A as agentRxBenchmarkCase, C as ANALYST_BENCHMARK_COST_LEDGER_FILE, D as AGENT_RX_UPSTREAM_REVISION, E as ANALYST_BENCHMARK_OBSERVATIONS_FILE, F as runAnalystBenchmark, I as traceStoreEvidenceResolver, M as normalizeAgentRxCategory, N as externalId, O as codeTraceBenchCase, P as positiveStep, S as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, T as ANALYST_BENCHMARK_MANIFEST_FILE, _ as readAnalystBenchmarkArtifact, a as runInlineAnalystDefinition, b as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, c as publicRlmAnalystDefinition, d as publicDirectAnalystDefinition, f as runChunkedAnalystDefinition, g as expandCodeTraceFailureBlocks, h as analystDefinitionProtocolSha256, i as primeCodeTraceAnalystDefinition, j as agentRxPredictionsToFindings, k as codeTracerPredictionsToFindings, l as runReplVariableAnalystDefinition, m as analystDefinitionAsymmetries, n as createPrimeBenchmarkRunner, o as loadPublicBenchmarkRows, p as AnalystExpressivenessError, r as primeAnalystProtocolSha256, s as createPublicBenchmarkRlmRunner, t as runAnalystBenchmarkCommand, u as createPublicBenchmarkDirectRunner, v as CODE_TRACE_BENCH_ANALYST_PROMPT, w as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, x as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, y as publicBenchmarkProtocolSha256 } from "../benchmark-command-BRRhjX2k.js";
 import { a as normalizePrimeUsage, c as runPrimeExchange, i as mergePrimeRawUsage, n as buildPrimePrompt, o as primeProtocolSha256, r as buildPrimeRepairPrompt, s as projectPrimeTrajectory, t as analystUsageReceiptFromPrimeUsage } from "../prime-protocol-CrhpXRz9.js";
 import { t as nodeHttpPrimeBridgeTransport } from "../prime-bridge-transport-Csd5SQg7.js";
-import { t as clusterBootstrapMean } from "../cluster-bootstrap-hNXESoc1.js";
+import { t as clusterBootstrapMean } from "../cluster-bootstrap-BteuR1hg.js";
 import { z } from "zod";
 import { existsSync, readFileSync, readdirSync, statSync } from "node:fs";
 import { join } from "node:path";
@@ -36,12 +37,16 @@ const goldSchema = z.object({
 const predictionSchema = z.object({
 	questionId,
 	windowId: identity$1,
+	traceId: identity$1.optional(),
 	answer: answer.nullable(),
+	probabilities: z.record(z.string(), z.number().finite()).nullable().optional(),
 	questionDigest: z.string().nullable(),
 	requestedModel: identity$1.nullable(),
 	servedModel: identity$1.nullable(),
 	support: behavioralTraceSupportSchema.optional()
 }).passthrough();
+const finite = (value) => Number.isFinite(value) ? value : null;
+const { auc: minAuc, ciLow, gold_n: minGold } = DECISION_POINT_AUC_BAR;
 /** Same nominalAgreement owner as the existing AgentRx/failure-mode calibration path. */
 function calibrateBehavioralTraceQuestions(input) {
 	const predictions = /* @__PURE__ */ new Map();
@@ -58,8 +63,16 @@ function calibrateBehavioralTraceQuestions(input) {
 		if (row.support && (row.support.map.windowId !== row.windowId || row.support.method !== (row.questionId === "identical-tool-repeat" ? "computed" : "judged"))) throw new TypeError("Behavioral support does not match its prediction");
 		const key = JSON.stringify([row.questionId, row.windowId]);
 		if (predictions.has(key)) throw new TypeError("Duplicate behavioral prediction");
-		predictions.set(key, row);
+		predictions.set(key, {
+			...row,
+			cluster: finding.subject ?? row.traceId ?? row.windowId
+		});
 	}
+	const bootstrap = {
+		resamples: input.resamples,
+		seed: input.seed,
+		options: ["yes", "no"]
+	};
 	return Object.keys(BEHAVIORAL_TRACE_QUESTIONS).map((id) => {
 		const computed = id === "identical-tool-repeat";
 		const digest = behavioralTraceQuestionDigest({
@@ -67,89 +80,163 @@ function calibrateBehavioralTraceQuestions(input) {
 			...input,
 			servedModel: computed ? null : input.servedModel
 		});
-		const row = {
-			questionId: id,
-			layer: input.layer,
-			method: computed ? "computed" : "judged",
-			gold_n: 0,
+		const counts = {
 			missingLabels: 0,
 			unsureLabels: 0,
-			missingPredictions: 0,
-			abstentions: 0,
 			sourceCounts: {
 				human: 0,
 				agentrx: 0
+			}
+		};
+		const confusion = {
+			yes: {
+				yes: 0,
+				no: 0,
+				unsure: 0,
+				missing: 0
 			},
-			agreement: null,
-			kappa: null,
-			confusion: {
-				yes: {
-					yes: 0,
-					no: 0,
-					unsure: 0,
-					missing: 0
-				},
-				no: {
-					yes: 0,
-					no: 0,
-					unsure: 0,
-					missing: 0
-				}
-			},
-			disagreements: [],
-			eligible: false,
-			calibration: null
+			no: {
+				yes: 0,
+				no: 0,
+				unsure: 0,
+				missing: 0
+			}
 		};
-		const pairs = [];
+		const disagreements = [];
+		const labels = [];
+		const answers = [];
 		for (const item of gold.filter((item) => item.questionId === id)) {
 			if (item.label === null) {
-				row.missingLabels++;
+				counts.missingLabels++;
 				continue;
 			}
 			if (item.label === "unsure") {
-				row.unsureLabels++;
+				counts.unsureLabels++;
 				continue;
 			}
-			row.sourceCounts[item.source]++;
+			counts.sourceCounts[item.source]++;
 			const prediction = predictions.get(JSON.stringify([id, item.windowId]));
 			const matches = prediction?.questionDigest === digest && prediction.requestedModel === (computed ? null : input.model) && prediction.servedModel === (computed ? null : input.servedModel);
 			const value = matches ? prediction.answer : null;
-			if (value === null) row.missingPredictions++;
-			if (value === "unsure") row.abstentions++;
-			row.confusion[item.label][value ?? "missing"]++;
+			confusion[item.label][value ?? "missing"]++;
 			if (value !== item.label) {
 				const support = matches ? prediction.support ?? null : null;
 				if (support && support.map.windowId !== item.windowId) throw new TypeError("Behavioral support map is outside the prediction window");
-				row.disagreements.push({
+				disagreements.push({
 					windowId: item.windowId,
 					expected: item.label,
 					predicted: value,
 					support: support ? structuredClone(support) : null
 				});
 			}
-			pairs.push([item.label, value ?? "(missing prediction)"]);
+			labels.push({
+				id: item.windowId,
+				label: item.label
+			});
+			if (matches) answers.push({
+				id: item.windowId,
+				cluster: prediction.cluster,
+				choice: prediction.answer,
+				probabilities: prediction.probabilities ?? null
+			});
 		}
-		const agreement = nominalAgreement(pairs);
-		row.gold_n = agreement.n;
-		row.agreement = Number.isFinite(agreement.agreement) ? agreement.agreement : null;
-		row.kappa = Number.isFinite(agreement.kappa) ? agreement.kappa : null;
-		if (row.gold_n > 0 && row.agreement !== null && row.kappa !== null) {
+		const result = calibrateChoiceQuestion(labels, answers, bootstrap);
+		const { option: _option, gold_n: positives, predicted: _predicted, ...yes } = result.rows[0];
+		const certified = positives > 0 && positives < result.gold_n && result.missingPredictions === 0 && result.kappa !== null;
+		const row = {
+			questionId: id,
+			layer: input.layer,
+			method: computed ? "computed" : "judged",
+			...counts,
+			gold_n: result.gold_n,
+			missingPredictions: result.missingPredictions,
+			abstentions: result.abstentions,
+			agreement: result.agreement,
+			kappa: result.kappa,
+			kappaCi: result.kappaCi,
+			...yes,
+			ci: yes.ci && {
+				...yes.ci,
+				cluster: "subject"
+			},
+			confusion,
+			disagreements,
+			eligible: certified && result.kappa >= .77,
+			calibration: null
+		};
+		if (certified) {
 			row.calibration = {
 				gold_n: row.gold_n,
 				agreement: row.agreement,
 				kappa: row.kappa,
+				auc: row.auc,
+				ci: row.ci,
 				questionDigest: digest,
 				servedModel: computed ? null : input.servedModel,
 				evidenceRefs: structuredClone(input.evidenceRefs)
 			};
 			validBehavioralCalibration(row.calibration);
 		}
-		const bothClasses = row.confusion.yes.yes + row.confusion.yes.no + row.confusion.yes.unsure + row.confusion.yes.missing > 0 && row.confusion.no.yes + row.confusion.no.no + row.confusion.no.unsure + row.confusion.no.missing > 0;
-		row.eligible = bothClasses && row.missingPredictions === 0 && row.kappa !== null && row.kappa >= .77;
-		if (!bothClasses || row.missingPredictions > 0) row.calibration = null;
 		return row;
 	});
 }
+/** A Choice question over any option set: a one-vs-rest row per option plus the multi-class κ. A null
+* or out-of-set gold label is an exclusion; an out-of-set choice an abstention, a null choice missing. */
+function calibrateChoiceQuestion(gold, findings, options = {}) {
+	const predictions = new Map(findings.map((finding) => [finding.id, finding]));
+	if (predictions.size !== findings.length) throw new TypeError("Duplicate choice prediction");
+	const names = [...options.options ?? new Set(findings.flatMap((f) => Object.keys(f.probabilities ?? {})))];
+	const labeled = gold.filter((item) => item.label !== null && names.includes(item.label));
+	if (new Set(labeled.map((item) => item.id)).size !== labeled.length) throw new TypeError("Duplicate choice gold label");
+	const joined = labeled.map((item) => ({
+		...item,
+		prediction: predictions.get(item.id)
+	}));
+	const cluster = (item) => item.prediction?.cluster ?? item.id;
+	const agreement = nominalAgreement(joined.map((item) => [item.label, item.prediction?.choice ?? "(missing prediction)"]), {
+		...options,
+		clusters: joined.map(cluster)
+	});
+	return {
+		options: names,
+		gold_n: agreement.n,
+		missingLabels: gold.length - labeled.length,
+		missingPredictions: joined.filter((item) => item.prediction?.choice == null).length,
+		abstentions: joined.filter((item) => item.prediction?.choice != null && !names.includes(item.prediction.choice)).length,
+		agreement: finite(agreement.agreement),
+		kappa: finite(agreement.kappa),
+		kappaCi: Number.isFinite(agreement.kappaCi.low) ? agreement.kappaCi : null,
+		rows: names.map((option) => {
+			const items = joined.map((item) => ({
+				cluster: cluster(item),
+				gold: item.label === option,
+				chosen: item.prediction?.choice === option,
+				score: item.prediction?.probabilities?.[option] ?? null
+			}));
+			const chosen = items.filter((item) => item.chosen);
+			const hits = chosen.filter((item) => item.gold).length;
+			const gold_n = items.filter((item) => item.gold).length;
+			const scored = items.filter((item) => item.score !== null);
+			const area = (sample) => rocAuc(sample.map((i) => i.score), sample.map((i) => Number(i.gold))) ?? NaN;
+			const auc = finite(scored.length ? area(scored) : NaN);
+			const boot = auc !== null && clusterBootstrap(scored, (i) => i.cluster, area, options);
+			const ci = boot && Number.isFinite(boot.low) ? {
+				...boot,
+				cluster: "cluster"
+			} : null;
+			return {
+				option,
+				gold_n,
+				predicted: chosen.length,
+				precision: chosen.length ? hits / chosen.length : null,
+				recall: gold_n ? hits / gold_n : null,
+				auc,
+				ci,
+				eligibleAuc: ci !== null && auc >= minAuc && ci.low >= ciLow && gold_n >= minGold
+			};
+		})
+	};
+}
 // src/analyst/bind.ts
 /**
 * Compile a definition plus a transport binding into a runnable arm. Dispatch
@@ -2110,6 +2197,6 @@ function failed(question, failure) {
 function errorText(error) {
 	return error instanceof Error ? `${error.name}: ${error.message}` : String(error);
 }
-export { AGENT_RX_UPSTREAM_REVISION, ANALYST_BENCHMARK_COST_LEDGER_FILE, ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, ANALYST_BENCHMARK_IMPLEMENTATION_FILES, ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, ANALYST_BENCHMARK_MANIFEST_FILE, ANALYST_BENCHMARK_OBSERVATIONS_FILE, AnalystExpressivenessError, AnalystRegistry, BEHAVIORAL_FAILURE_CLASS, BEHAVIORAL_TRACE_QUESTIONS, BEHAVIORAL_TRACE_QUESTIONS_VERSION, CODE_TRACE_BENCH_ANALYST_PROMPT, CONTROL_INTEGRITY_ANALYST, ControlIntegrityAnalyst, DEFAULT_TRACE_ANALYST_KINDS, ExactAnalystRunExecutionError, FAILURE_MODES_ANALYST_ID, FAILURE_MODES_MIN_KAPPA, FAILURE_MODES_QUESTIONS, FAILURE_MODES_VERSION, FAILURE_MODE_KIND_SPEC, FAILURE_MODE_LABELS, FINDING_LAYERS, FINDING_SUBJECT_KINDS, FINDING_SUBJECT_SYNTAX, FindingsStore, IMPROVEMENT_KIND_SPEC, KIND_EXPECTED_SUBJECTS, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, LAB_NODE_ID, NOT_OBSERVABLE, ProposalFindingSchema, RAW_FINDING_SCHEMA_PROMPT, RawAnalystFindingSchema, TEAM_SHAPE_ANALYST_ID, TEAM_SHAPE_CLASSES, TEAM_SHAPE_VERSION, TREATMENT_LAYERS, agentRxBenchmarkCase, agentRxPredictionsToFindings, analystDefinitionAsymmetries, analystDefinitionProtocolSha256, analystUsageReceiptFromPrimeUsage, askTraceQuestions, assertFailureClasses, assertProposalFindings, attachToTaxonomy, behavioralAnalyst, behavioralTraceQuestionDigest, bindAnalyst, buildDefaultAnalystRegistry, buildPrimePrompt, buildPrimeRepairPrompt, buildSkillUsageReport, buildTraceToolsForGroup, calibrateBehavioralTraceQuestions, calibrateFailureModes, classifyTeamShape, codeTraceBenchCase, codeTracerPredictionsToFindings, computeFindingId, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createFailureModesAnalyst, createPrimeBenchmarkRunner, createPublicBenchmarkDirectRunner, createPublicBenchmarkRlmRunner, createTraceAnalyst, defineCustomAnalyst, defineDomainTaxonomy, defineTraceAnalyst, diffFindings, diffFindingsOverTime, domainPathOf, emitControlIntegrityFindings, evidenceRefsFromRawFinding, expandCodeTraceFailureBlocks, findingSubjectGrammarPromptFor, independentUnitCount, lineageRoots, loadFailureModeGold, loadPublicBenchmarkRows, lowestCommonAncestor, makeFinding, makeProposalFinding, mergePrimeRawUsage, nodeHttpPrimeBridgeTransport, normalizePrimeUsage, openFindingsLedger, parseAnalystFinding, parseFindingSubject, parseRawFinding, pinBehavioralFailureClass, primeAnalystProtocolSha256, primeCodeTraceAnalystDefinition, primeProtocolSha256, projectPrimeTrajectory, promote, publicBenchmarkProtocolSha256, publicDirectAnalystDefinition, publicRlmAnalystDefinition, queryFindings, readAnalystBenchmarkArtifact, readFindingScope, renderFindingSubject, resolveDomainPath, rollupFindings, runAnalystBenchmark, runAnalystBenchmarkCommand, runPrimeExchange, runTraceAnalyst, stripCodeFences, taxonomyAncestry, taxonomyLayerOf, teamShapeAnalyst, teamShapeInputFromSources, traceStoreEvidenceResolver };
+export { AGENT_RX_UPSTREAM_REVISION, ANALYST_BENCHMARK_COST_LEDGER_FILE, ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, ANALYST_BENCHMARK_IMPLEMENTATION_FILES, ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, ANALYST_BENCHMARK_MANIFEST_FILE, ANALYST_BENCHMARK_OBSERVATIONS_FILE, AnalystExpressivenessError, AnalystRegistry, BEHAVIORAL_FAILURE_CLASS, BEHAVIORAL_TRACE_QUESTIONS, BEHAVIORAL_TRACE_QUESTIONS_VERSION, CODE_TRACE_BENCH_ANALYST_PROMPT, CONTROL_INTEGRITY_ANALYST, ControlIntegrityAnalyst, DEFAULT_TRACE_ANALYST_KINDS, ExactAnalystRunExecutionError, FAILURE_MODES_ANALYST_ID, FAILURE_MODES_MIN_KAPPA, FAILURE_MODES_QUESTIONS, FAILURE_MODES_VERSION, FAILURE_MODE_KIND_SPEC, FAILURE_MODE_LABELS, FINDING_LAYERS, FINDING_SUBJECT_KINDS, FINDING_SUBJECT_SYNTAX, FindingsStore, IMPROVEMENT_KIND_SPEC, KIND_EXPECTED_SUBJECTS, KNOWLEDGE_GAP_KIND_SPEC, KNOWLEDGE_POISONING_KIND_SPEC, LAB_NODE_ID, NOT_OBSERVABLE, ProposalFindingSchema, RAW_FINDING_SCHEMA_PROMPT, RawAnalystFindingSchema, TEAM_SHAPE_ANALYST_ID, TEAM_SHAPE_CLASSES, TEAM_SHAPE_VERSION, TREATMENT_LAYERS, agentRxBenchmarkCase, agentRxPredictionsToFindings, analystDefinitionAsymmetries, analystDefinitionProtocolSha256, analystUsageReceiptFromPrimeUsage, askTraceQuestions, assertFailureClasses, assertProposalFindings, attachToTaxonomy, behavioralAnalyst, behavioralTraceQuestionDigest, bindAnalyst, buildDefaultAnalystRegistry, buildPrimePrompt, buildPrimeRepairPrompt, buildSkillUsageReport, buildTraceToolsForGroup, calibrateBehavioralTraceQuestions, calibrateChoiceQuestion, calibrateFailureModes, classifyTeamShape, codeTraceBenchCase, codeTracerPredictionsToFindings, computeFindingId, createChatClient, createChatTraceEngine, createDspyRlmTraceEngine, createFailureModesAnalyst, createPrimeBenchmarkRunner, createPublicBenchmarkDirectRunner, createPublicBenchmarkRlmRunner, createTraceAnalyst, defineCustomAnalyst, defineDomainTaxonomy, defineTraceAnalyst, diffFindings, diffFindingsOverTime, domainPathOf, emitControlIntegrityFindings, evidenceRefsFromRawFinding, expandCodeTraceFailureBlocks, findingSubjectGrammarPromptFor, independentUnitCount, lineageRoots, loadFailureModeGold, loadPublicBenchmarkRows, lowestCommonAncestor, makeFinding, makeProposalFinding, mergePrimeRawUsage, nodeHttpPrimeBridgeTransport, normalizePrimeUsage, openFindingsLedger, parseAnalystFinding, parseFindingSubject, parseRawFinding, pinBehavioralFailureClass, primeAnalystProtocolSha256, primeCodeTraceAnalystDefinition, primeProtocolSha256, projectPrimeTrajectory, promote, publicBenchmarkProtocolSha256, publicDirectAnalystDefinition, publicRlmAnalystDefinition, queryFindings, readAnalystBenchmarkArtifact, readFindingScope, renderFindingSubject, resolveDomainPath, rollupFindings, runAnalystBenchmark, runAnalystBenchmarkCommand, runPrimeExchange, runTraceAnalyst, stripCodeFences, taxonomyAncestry, taxonomyLayerOf, teamShapeAnalyst, teamShapeInputFromSources, traceStoreEvidenceResolver };
 
 //# sourceMappingURL=index.js.map