@tangle-network/agent-eval 0.220.2 → 0.221.0

package/dist/analyst/index.js

@@ -1,23 +1,156 @@
+import { n as assertConsistentFindingConditions, o as mergeFindingComposition, r as findingCompositionKey, t as TREATMENT_LAYERS } from "../finding-composition-8n0fFgVI.js";
 import { i as nominalAgreement } from "../agreement-irr-BhAqjluA.js";
 import { t as FileLedgerJournal } from "../journal-C7om8Gum.js";
 import { r as merkleRoot } from "../ledger-core-CAxUdDk0.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-D670Oxf2.js";
-import { a as makeProposalFinding, i as makeFinding, r as computeFindingId } from "../types-DqFMNTgK.js";
-import { t as assertProposalFindings } from "../proposal-findings-E1aSC_cU.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-BhxbtUsQ.js";
-import { a as KNOWLEDGE_POISONING_KIND_SPEC, c as FAILURE_MODE_KIND_SPEC, d as emitControlIntegrityFindings, f as behavioralAnalyst, i as DEFAULT_TRACE_ANALYST_KINDS, l as CONTROL_INTEGRITY_ANALYST, n as AnalystRegistry, o as KNOWLEDGE_GAP_KIND_SPEC, r as ExactAnalystRunExecutionError, s as IMPROVEMENT_KIND_SPEC, t as buildDefaultAnalystRegistry, u as ControlIntegrityAnalyst } from "../default-registry-ovrIBGmH.js";
-import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-Bn_Mt_LW.js";
-import { n as diffFindings, r as createChatTraceEngine, t as FindingsStore } from "../findings-store-CvNo712L.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-DA02c_L6.js";
+import { a as makeProposalFinding, i as makeFinding, r as computeFindingId } from "../types-CnWmj8NW.js";
+import { t as assertProposalFindings } from "../proposal-findings-D0_zkQ2l.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-Dl0PmNQL.js";
+import { _ as behavioralTraceQuestionDigest, a as KNOWLEDGE_POISONING_KIND_SPEC, c as FAILURE_MODE_KIND_SPEC, d as emitControlIntegrityFindings, f as behavioralAnalyst, g as BEHAVIORAL_TRACE_QUESTIONS_VERSION, h as BEHAVIORAL_TRACE_QUESTIONS, i as DEFAULT_TRACE_ANALYST_KINDS, l as CONTROL_INTEGRITY_ANALYST, m as behavioralTraceSupportSchema, n as AnalystRegistry, o as KNOWLEDGE_GAP_KIND_SPEC, p as validBehavioralCalibration, r as ExactAnalystRunExecutionError, s as IMPROVEMENT_KIND_SPEC, t as buildDefaultAnalystRegistry, u as ControlIntegrityAnalyst } from "../default-registry-BbFCkqvB.js";
+import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-BTy7u0jQ.js";
+import { n as diffFindings, r as createChatTraceEngine, t as FindingsStore } from "../findings-store-DRiFdoid.js";
+import { a as systemOneAnalyst, p as validDecisionIdentity } from "../decision-point-Co3u3W8k.js";
 import { t as createChatClient } from "../chat-client-CMQ8_h4J.js";
-import { systemOneAnalyst } from "../systemone.js";
-import { A as agentRxBenchmarkCase, C as ANALYST_BENCHMARK_COST_LEDGER_FILE, D as AGENT_RX_UPSTREAM_REVISION, E as ANALYST_BENCHMARK_OBSERVATIONS_FILE, M as normalizeAgentRxCategory, N as runAnalystBenchmark, O as codeTraceBenchCase, P as traceStoreEvidenceResolver, 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-Dx0E4a6z.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-8XQfpUK3.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-BFgGrEMP.js";
 import { t as nodeHttpPrimeBridgeTransport } from "../prime-bridge-transport-Z3cpKGoV.js";
 import { t as clusterBootstrapMean } from "../cluster-bootstrap-DEwwhl0C.js";
-import { existsSync, readFileSync, readdirSync, statSync } from "node:fs";
 import { z } from "zod";
+import { existsSync, readFileSync, readdirSync, statSync } from "node:fs";
 import { join } from "node:path";
 import { readFile } from "node:fs/promises";
+//#region src/analyst/benchmark-behavioral-calibration.ts
+const identity$1 = z.string().refine(validDecisionIdentity);
+const questionId = z.enum(Object.keys(BEHAVIORAL_TRACE_QUESTIONS));
+const answer = z.enum([
+	"yes",
+	"no",
+	"unsure"
+]);
+const goldSchema = z.object({
+	questionId,
+	windowId: identity$1,
+	label: answer.nullable(),
+	source: z.enum(["human", "agentrx"])
+}).strict();
+const predictionSchema = z.object({
+	questionId,
+	windowId: identity$1,
+	answer: answer.nullable(),
+	questionDigest: z.string().nullable(),
+	requestedModel: identity$1.nullable(),
+	servedModel: identity$1.nullable(),
+	support: behavioralTraceSupportSchema.optional()
+}).passthrough();
+/** Same nominalAgreement owner as the existing AgentRx/failure-mode calibration path. */
+function calibrateBehavioralTraceQuestions(input) {
+	const predictions = /* @__PURE__ */ new Map();
+	const gold = input.gold.map((row) => goldSchema.parse(row));
+	const seen = /* @__PURE__ */ new Set();
+	for (const row of gold) {
+		const key = JSON.stringify([row.questionId, row.windowId]);
+		if (seen.has(key)) throw new TypeError("Duplicate behavioral gold label");
+		seen.add(key);
+	}
+	for (const finding of input.findings) {
+		if (finding.area !== "behavioral-trace-question" || finding.layer !== input.layer) continue;
+		const row = predictionSchema.parse(finding.metadata?.behavioral_trace_question);
+		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);
+	}
+	return Object.keys(BEHAVIORAL_TRACE_QUESTIONS).map((id) => {
+		const computed = id === "identical-tool-repeat";
+		const digest = behavioralTraceQuestionDigest({
+			questionId: id,
+			...input,
+			servedModel: computed ? null : input.servedModel
+		});
+		const row = {
+			questionId: id,
+			layer: input.layer,
+			method: computed ? "computed" : "judged",
+			gold_n: 0,
+			missingLabels: 0,
+			unsureLabels: 0,
+			missingPredictions: 0,
+			abstentions: 0,
+			sourceCounts: {
+				human: 0,
+				agentrx: 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
+		};
+		const pairs = [];
+		for (const item of gold.filter((item) => item.questionId === id)) {
+			if (item.label === null) {
+				row.missingLabels++;
+				continue;
+			}
+			if (item.label === "unsure") {
+				row.unsureLabels++;
+				continue;
+			}
+			row.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"]++;
+			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({
+					windowId: item.windowId,
+					expected: item.label,
+					predicted: value,
+					support: support ? structuredClone(support) : null
+				});
+			}
+			pairs.push([item.label, value ?? "(missing prediction)"]);
+		}
+		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) {
+			row.calibration = {
+				gold_n: row.gold_n,
+				agreement: row.agreement,
+				kappa: row.kappa,
+				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;
+	});
+}
+//#endregion
 //#region src/analyst/bind.ts
 /**
 * Compile a definition plus a transport binding into a runnable arm. Dispatch
@@ -301,12 +434,19 @@ const AGENT_RX_FAILURE_MODE_MAP = {
 	"invalid-invocation": "invalid-tool-invocation",
 	"misinterpretation-of-tool-output-handoff-failure": "misread-tool-output",
 	"intent-plan-misalignment": "task-derailment",
-	"underspecified-user-intent": "fail-to-ask-for-clarification",
+	"underspecified-user-intent": null,
 	"intent-not-supported": null,
 	"guardrails-triggered": null,
 	"system-failure": "system-failure",
 	inconclusive: null
 };
+/** Agent roles in the pinned public AgentRx datasets; unfamiliar actors stay unknown. */
+const AGENT_RX_AGENT_ACTORS = /* @__PURE__ */ new Set([
+	"assistant",
+	"orchestrator",
+	"websurfer",
+	"filesurfer"
+]);
 const FAILURE_MODES_QUESTIONS = {
 	failure_mode: {
 		type: "choice",
@@ -340,19 +480,31 @@ function isLabel(value) {
 function failureModeGoldFromAgentRx(rows) {
 	const gold = [];
 	const unmapped = [];
+	const ids = /* @__PURE__ */ new Set();
 	for (const row of rows) {
-		const id = `agentrx:${String(row.trajectory_id)}`;
-		const rootId = row.root_cause_failure_id ?? row.root_cause?.failure_id;
+		const id = `agentrx:${externalId(row.trajectory_id, "AgentRx trajectory id")}`;
+		if (ids.has(id)) throw new TypeError(`AgentRx trajectory '${row.trajectory_id}' repeats`);
+		ids.add(id);
+		const rootId = externalId(row.root_cause_failure_id ?? row.root_cause?.failure_id, `AgentRx trajectory '${row.trajectory_id}' root-cause failure id`);
 		const root = row.failures.find((failure) => String(failure.failure_id) === String(rootId));
 		if (!root) throw new TypeError(`AgentRx trajectory '${row.trajectory_id}' has no root-cause failure`);
+		const step = positiveStep(root.step_number, `AgentRx trajectory '${row.trajectory_id}' root step`);
 		const category = normalizeAgentRxCategory(root.failure_category);
 		if (!Object.hasOwn(AGENT_RX_FAILURE_MODE_MAP, category)) throw new RangeError(`AgentRx category '${root.failure_category}' is outside the AgentRx taxonomy`);
 		const label = AGENT_RX_FAILURE_MODE_MAP[category];
-		if (label === null || label === void 0) unmapped.push({
+		const actor = typeof root.failed_agent === "string" ? root.failed_agent.trim() || null : null;
+		const reason = actor === null ? "unknown-root-actor" : !AGENT_RX_AGENT_ACTORS.has(actor.toLowerCase()) ? "unsupported-root-actor" : label === null || label === void 0 ? "category-without-counterpart" : null;
+		if (reason !== null) unmapped.push({
 			id,
-			category
+			category,
+			reason,
+			root: {
+				failureId: rootId,
+				step,
+				actor
+			}
 		});
-		else gold.push({
+		else if (label !== null && label !== void 0) gold.push({
 			id,
 			label
 		});
@@ -362,7 +514,7 @@ function failureModeGoldFromAgentRx(rows) {
 		unmapped
 	};
 }
-/** Validate a caller's gold file: a JSON array of `{ id, label }` with unique ids and known labels. */
+/** Validate caller gold with unique, whitespace-unambiguous ids and known labels. */
 function parseFailureModeGold(raw) {
 	if (!Array.isArray(raw)) return {
 		succeeded: false,
@@ -383,6 +535,13 @@ function parseFailureModeGold(raw) {
 				message: `gold row ${index} needs a string id and a known label`
 			}
 		};
+		if (id !== id.trim()) return {
+			succeeded: false,
+			error: {
+				kind: "invalid",
+				message: `gold row ${index} id must not have surrounding whitespace`
+			}
+		};
 		if (seen.has(id)) return {
 			succeeded: false,
 			error: {
@@ -404,8 +563,9 @@ function parseFailureModeGold(raw) {
 /**
 * Load the calibration gold set from the public AgentRx labels and the
 * caller's gold file, through the analyst module's public-dataset loader.
-* Ids must be unique across both sources. No source, or no gold item, is a
-* failure: an empty gold set would calibrate to an undefined κ.
+* Caller gold ids must be disjoint from all AgentRx ids, including exclusions.
+* No source or an empty source is invalid. A parsed AgentRx source whose rows are all excluded returns
+* unknown eligibility with its reasons; it supplies no labels to calibrate.
 */
 async function loadFailureModeGold(sources) {
 	const invalid = (message) => ({
@@ -417,12 +577,17 @@ async function loadFailureModeGold(sources) {
 	});
 	const gold = [];
 	let agentRxRows = 0;
+	let agentRxMapping = null;
 	let unmapped = [];
 	try {
 		if (sources.agentRxLabelsPath !== void 0) {
 			const rows = await loadPublicBenchmarkRows(sources.agentRxLabelsPath);
 			agentRxRows = rows.length;
 			const mapped = failureModeGoldFromAgentRx(rows);
+			agentRxMapping = {
+				version: "2",
+				target: "agent-or-team-root-cause"
+			};
 			gold.push(...mapped.gold);
 			unmapped = mapped.unmapped;
 		}
@@ -434,8 +599,8 @@ async function loadFailureModeGold(sources) {
 	} catch (error) {
 		return invalid(error instanceof Error ? error.message : String(error));
 	}
-	if (gold.length === 0) return invalid("failure-modes calibration needs at least one gold item");
-	const ids = /* @__PURE__ */ new Set();
+	if (gold.length === 0 && unmapped.length === 0) return invalid("failure-modes calibration needs at least one gold item");
+	const ids = new Set(unmapped.map((item) => item.id));
 	for (const item of gold) {
 		if (ids.has(item.id)) return invalid(`gold id '${item.id}' repeats across sources`);
 		ids.add(item.id);
@@ -444,7 +609,9 @@ async function loadFailureModeGold(sources) {
 		succeeded: true,
 		value: {
 			gold,
+			eligibility: gold.length > 0 ? "eligible" : "unknown",
 			agentRxRows,
+			agentRxMapping,
 			unmapped
 		}
 	};
@@ -929,6 +1096,7 @@ const LADDER = [
 	"lab"
 ];
 function promote(findings, taxonomy, options = {}) {
+	assertConsistentFindingConditions(findings);
 	const k = options.k ?? 2;
 	if (!Number.isInteger(k) || k < 2) throw new RangeError(`promote: k must be an integer of at least 2, got ${k}`);
 	const unplaced = [];
@@ -960,7 +1128,11 @@ function promote(findings, taxonomy, options = {}) {
 		for (const placement of buckets.get(depth) ?? []) {
 			const parent = taxonomyParentOf(taxonomy, placement.node);
 			if (parent === null || parent === void 0) continue;
-			const key = JSON.stringify([placement.claimId, parent]);
+			const key = JSON.stringify([
+				placement.claimId,
+				parent,
+				findingCompositionKey(placement.finding)
+			]);
 			const group = groups.get(key) ?? {
 				parent,
 				members: []
@@ -1095,6 +1267,7 @@ function liftedFinding(voters, parent, taxonomy, producedAt) {
 		claim: first.claim,
 		claim_id: first.claim_id,
 		id_basis: `${first.claim_id}@${parent}`,
+		...mergeFindingComposition(findings),
 		subject: parent,
 		severity: mostSevere(findings),
 		confidence: Math.min(...findings.map((finding) => finding.confidence)),
@@ -1148,6 +1321,7 @@ function liftedFinding(voters, parent, taxonomy, producedAt) {
 const ROLLUP_ANALYST_ID = "rollup";
 const ROLLUP_VERSION = "1";
 function rollupFindings(findings, taxonomy, options = {}) {
+	assertConsistentFindingConditions(findings);
 	const confidence = options.confidence ?? .95;
 	const skipped = [];
 	const skip = (finding, reason) => {
@@ -1188,7 +1362,11 @@ function rollupFindings(findings, taxonomy, options = {}) {
 	const groups = /* @__PURE__ */ new Map();
 	for (const observation of observations) {
 		const play = taxonomyParentOf(taxonomy, observation.run);
-		const key = JSON.stringify([observation.claimId, play]);
+		const key = JSON.stringify([
+			observation.claimId,
+			play,
+			findingCompositionKey(observation.finding)
+		]);
 		const group = groups.get(key) ?? {
 			play,
 			members: []
@@ -1215,6 +1393,7 @@ function rollupFindings(findings, taxonomy, options = {}) {
 			claim: first.claim,
 			claim_id: first.claim_id,
 			id_basis: `${first.claim_id}@${play}`,
+			...mergeFindingComposition(findingsInGroup),
 			subject: play,
 			severity: mostSevere(findingsInGroup),
 			confidence: Math.min(...findingsInGroup.map((finding) => finding.confidence)),
@@ -1250,7 +1429,7 @@ function rollupFindings(findings, taxonomy, options = {}) {
 	});
 	const promoted = promotion.promoted.map((finding) => {
 		const covered = new Set(finding.scope.population);
-		const pooled = pool(observations.filter((observation) => observation.claimId === finding.claim_id && covered.has(observation.run)), options, confidence);
+		const pooled = pool(observations.filter((observation) => observation.claimId === finding.claim_id && covered.has(observation.run) && findingCompositionKey(observation.finding) === findingCompositionKey(finding)), options, confidence);
 		return {
 			...finding,
 			measure: {
@@ -1942,6 +2121,6 @@ function errorText(error) {
 	return error instanceof Error ? `${error.name}: ${error.message}` : String(error);
 }
 //#endregion
-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, 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, agentRxBenchmarkCase, agentRxPredictionsToFindings, analystDefinitionAsymmetries, analystDefinitionProtocolSha256, analystUsageReceiptFromPrimeUsage, askTraceQuestions, assertProposalFindings, attachToTaxonomy, behavioralAnalyst, bindAnalyst, buildDefaultAnalystRegistry, buildPrimePrompt, buildPrimeRepairPrompt, buildSkillUsageReport, buildTraceToolsForGroup, 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, 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_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, 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, 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