@tangle-network/agent-eval 0.150.2 → 0.161.0

package/dist/index.js

@@ -1,56 +1,58 @@
 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 hashJson, t as canonicalize } from "./pre-registration-DakwTRXk.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-Bnq4FaDO.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 { 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-C-VHSRwF.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 { 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 { t as eProcess } from "./sequential-eprocess-CbUt2htw.js";
+import { t as eProcess } from "./sequential-eprocess-D1jKoihe.js";
 import { n as iqr, r as welchsTTest } from "./baseline-BhPRQBVn.js";
-import { i as isLlmSpan, r as isJudgeSpan, s as isToolSpan, t as FAILURE_CLASSES } from "./schema-CRhEY1SO.js";
-import { a as judgeSpans, c as runsForScenario, n as argHash } from "./query-Di7eEQ79.js";
-import { i as analyzeRuns, o as checkCanaries, r as selfImprove, t as defineAgentEval } from "./define-agent-eval-C8V8sMqP.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 { 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-D2lDdSAz.js";
-import { a as summaryTable, n as gainHistogram, r as paretoChart } from "./summary-report-DW2bEpdB.js";
-import { n as contentHash, r as fileVerdictCache, t as canonicalJson } from "./verdict-cache-mZf5FEiY.js";
-import { H as buildReflectionPrompt, J as DEFAULT_RED_TEAM_CORPUS, Q as runCanaries, U as parseReflectionResponse, X as redTeamReport, Y as redTeamDataset, Z as scoreRedTeamOutput, _ as paretoFrontier, at as assertRealBackend, et as runCampaign, g as dominates, it as BackendIntegrityError, k as surfaceContentHash, ot as summarizeBackendIntegrity, t as llmJudge } from "./llm-judge-BqqMS8t7.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 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-BndeTAh_.js";
-import { t as TraceEmitter } from "./emitter-CPBAhxum.js";
+import { n as equivalenceVerdict, t as certificationEvidenceDigest } from "./verdict-B0xltqu6.js";
+import { t as TraceEmitter } from "./emitter-BpYFQPj4.js";
 import { t as buildTrajectory } from "./trajectory-D_7rLrvE.js";
-import { t as runCounterfactual } from "./counterfactual-lDfCx0Uz.js";
-import { l as createBoundedTraceAnalysisStore, m as TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, p as DEFAULT_TRACE_ANALYST_BUDGETS, t as createTraceAnalyst } from "./kind-factory-DmAa0h3K.js";
-import { S as validateAnalystReviewDecisions, _ as analystRunDigest, b as readAnalystReview, g as analystFindingDigest, n as buildDefaultAnalystRegistry, o as DEFAULT_TRACE_ANALYST_KINDS, r as AnalystRegistry, t as createChatClient, u as FAILURE_MODE_KIND_SPEC, v as assertUniqueFindingIds, x as snapshotAnalystRun, y as completedAnalystReviewQuality } from "./chat-client-Bvmxedyv.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 { 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-CykkbOlv.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 { 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 { C as CrossFamilyError, S as servedModelAcceptable, T as judgeFamily, _ as assertCrossFamilyServed, a as assertLlmRoute, b as assertServedModels, c as callLlmJson, d as isTransientLlmError, f as maximumChargeForLlmRequest, g as ServedCrossFamilyError, h as ModelSubstitutionError, l as costReceiptFromLlm, m as stripFencedJson, n as LlmClient, o as backoffMs, p as probeLlm, r as LlmResponseError, s as callLlm, t as LlmCallError, u as costReceiptFromLlmError, v as assertServedModel, w as assertCrossFamily, x as checkServedModel } from "./llm-client-Bg32RW0j.js";
-import { t as runEvalCampaign } from "./eval-campaign-CQuZrLR_.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-ytVQ7WT8.js";
-import { t as mintRolloutRows } from "./mint-BV6tLVWl.js";
+import "./rollout-Dm2tSdiQ.js";
+import { t as mintRolloutRows } from "./mint-DfODW1KW.js";
 import { n as pairedEvalueSequence, t as evaluateInterimReleaseConfidence } from "./sequential-CzK5DarL.js";
-import { t as createDspyRlmTraceEngine } from "./dspy-rlm-engine-CBYlPvNy.js";
-import { a as diffFindings, n as runSemanticConceptJudge, r as FindingsStore, t as SEMANTIC_CONCEPT_JUDGE_VERSION } from "./semantic-concept-judge-laMCnTLn.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 { t as analyzeSeries } from "./series-convergence-CjO2QdRW.js";
-import { i as otlpTextToTraceAnalysisStore, n as OtlpFileTraceStore } from "./store-otlp-CDYWW_8N.js";
-import { n as evaluateReleaseConfidence, r as bootstrapCi } from "./release-confidence-BknrpBnO.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 { 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";
@@ -2728,22 +2730,16 @@ function fingerprintAction(action, policies) {
 	if (policies?.actionFingerprint) return policies.actionFingerprint(action);
 	return stableFingerprint(action);
 }
+/** Default state/action fingerprint: the value itself for a string, `String(value)`
+* for a number, boolean, null or undefined, and RFC 8785 canonical JSON for
+* everything else. A value with no canonical JSON form (an `undefined`-valued
+* field, a non-finite number, a class instance) throws: two states that differ
+* in such a field must not fingerprint alike, and a caller with such state
+* supplies `stopPolicies.stateFingerprint` / `actionFingerprint`. */
 function stableFingerprint(value) {
 	if (typeof value === "string") return value;
 	if (typeof value === "number" || typeof value === "boolean" || value == null) return String(value);
-	try {
-		return JSON.stringify(sortForFingerprint(value));
-	} catch {
-		return String(value);
-	}
-}
-function sortForFingerprint(value) {
-	if (Array.isArray(value)) return value.map(sortForFingerprint);
-	if (!value || typeof value !== "object") return value;
-	const record = value;
-	const sorted = {};
-	for (const key of Object.keys(record).sort()) sorted[key] = sortForFingerprint(record[key]);
-	return sorted;
+	return canonicalString(value);
 }
 function abortReason(signal) {
 	const reason = signal.reason;
@@ -5638,9 +5634,9 @@ function clampUnit(value) {
 * SAME backend config, so the judge can't silently re-route through a
 * different (often paid) backend than the agent.
 *
-* The bug class: `--backend cli-bridge` rewires the agent, but the judge still
-* reads `process.env.TANGLE_API_KEY` → router. Cost is billed against the
-* router, the eval reports the cli-bridge model, and the data is unusable.
+* The bug class: a consumer rewires the agent onto one backend but leaves the
+* judge bound to another. Cost is billed against the second backend, the eval
+* reports the first backend's model, and the data is unusable.
 * Four consumers hand-roll this comparison (legal at `canonical.ts:702-795`);
 * this is the one substrate copy.
 *
@@ -5656,7 +5652,7 @@ var SingleBackendError = class extends AgentEvalError {
 		this.name = "SingleBackendError";
 	}
 };
-function stripSlash$1(url) {
+function stripSlash(url) {
 	return url.replace(/\/+$/, "");
 }
 /**
@@ -5671,7 +5667,7 @@ function assertSingleBackend(agent, judge, opts = {}) {
 		agent: agent.kind,
 		judge: judge.kind
 	});
-	if (stripSlash$1(agent.baseUrl) !== stripSlash$1(judge.baseUrl)) divergences.push({
+	if (stripSlash(agent.baseUrl) !== stripSlash(judge.baseUrl)) divergences.push({
 		field: "baseUrl",
 		agent: agent.baseUrl,
 		judge: judge.baseUrl
@@ -5708,30 +5704,6 @@ function assertSingleBackend(agent, judge, opts = {}) {
 }
 //#endregion
 //#region src/intent-match-judge.ts
-/**
-* Intent-match judge — "did the agent build the right APP, ignoring
-* whether every feature is wired up?"
-*
-* Distinct from {@link runSemanticConceptJudge} which scores per-concept
-* presence. The semantic judge can return 0/4 concepts present even
-* when the agent built a thoughtful, polished, on-brief app that just
-* lacks one or two features. The semantic judge can also return 4/4
-* present even when the agent shipped the wrong project (keyword-rich
-* stub).
-*
-* Intent-match asks ONE question:
-*   "Looking at the agent's work as a whole — independent of feature
-*    coverage — is this an honest attempt at the user's request?"
-*
-* Returns a 0–1 score and a 1-sentence evidence string. Use as a sanity
-* check on `completenessScore`-style metrics: if intent-match is high
-* and concept count is low, the agent built the right thing but is
-* missing features (ship and iterate). If intent-match is low, the
-* agent built the wrong thing (reject regardless of concept count).
-*
-* Soft-fails on LLM/JSON error (`available: false`) so callers can
-* treat failure as "judge skipped."
-*/
 const INTENT_MATCH_JUDGE_VERSION = "intent-match-judge-v1-2026-04-24";
 const DEFAULT_MODEL = "claude-sonnet-4-6";
 const DEFAULT_TIMEOUT = 3e5;
@@ -5802,16 +5774,17 @@ Return STRICT JSON. No prose outside.`;
 /**
 * Run the intent-match judge. Soft-fails to available=false on error.
 */
-async function runIntentMatchJudge(input, options = {}) {
+async function runIntentMatchJudge(input, options) {
 	const start = Date.now();
 	const opts = {
-		model: options.model ?? DEFAULT_MODEL,
+		chat: options.chat,
+		model: options.model ?? options.chat.defaultModel ?? DEFAULT_MODEL,
 		timeoutMs: options.timeoutMs ?? DEFAULT_TIMEOUT,
 		maxTokens: options.maxTokens ?? DEFAULT_MAX_TOKENS,
 		maxSourceChars: options.maxSourceChars ?? DEFAULT_MAX_SOURCE,
 		maxPerFileChars: options.maxPerFileChars ?? DEFAULT_MAX_PER_FILE,
 		maxHtmlChars: options.maxHtmlChars ?? DEFAULT_MAX_HTML,
-		llm: options.llm ?? {},
+		...options.pricing ? { pricing: options.pricing } : {},
 		costLedger: options.costLedger ?? new CostLedger(),
 		costPhase: options.costPhase ?? "judge.intent-match",
 		costTags: options.costTags ?? {},
@@ -5846,25 +5819,20 @@ async function runIntentMatchJudge(input, options = {}) {
 			maxTokens: opts.maxTokens,
 			timeoutMs: opts.timeoutMs
 		};
-		const paid = await opts.costLedger.runPaidCall({
+		const paid = await paidJsonChat({
+			chat: opts.chat,
+			request,
+			ledger: opts.costLedger,
 			channel: "judge",
 			phase: opts.costPhase,
 			actor: "intent-match",
-			model: opts.model,
-			...Object.keys(opts.costTags).length > 0 ? { tags: opts.costTags } : {},
-			maximumCharge: maximumChargeForLlmRequest(request, opts.llm),
+			tags: opts.costTags,
 			signal: opts.signal,
-			execute: (signal, callId) => callLlmJson(request, {
-				...opts.llm,
-				signal,
-				idempotencyKey: callId
-			}),
-			receipt: ({ result }) => costReceiptFromLlm(result),
-			receiptFromError: costReceiptFromLlmError
+			...opts.pricing ? { pricing: opts.pricing } : {}
 		});
 		receipt = paid.receipt;
 		if (!paid.succeeded) throw paid.error;
-		const { value } = paid.value;
+		const { value } = paid;
 		const score = Math.max(0, Math.min(1, Number(value?.score ?? 0)));
 		return {
 			kind: "intent-match",
@@ -6972,9 +6940,6 @@ function jsonlRunRecordBackend(path, opts = {}) {
 * never be scored as a dead id.
 */
 const REASONING_BUDGET_EXHAUSTED = /reasoning[\s_-]?budget[\s_-]?exhausted/i;
-function stripSlash(url) {
-	return url.replace(/\/+$/, "");
-}
 /** Extract `error.message` (then top-level `message`) from a chat-completions error body. */
 function errorMessage(body) {
 	if (body == null || typeof body !== "object") return null;
@@ -6994,9 +6959,6 @@ function errorMessage(body) {
 * unconfigured (a 401 `model_not_found` from the router) is caught.
 */
 async function preflightModels(opts) {
-	const fetchImpl = opts.fetchImpl ?? fetch;
-	const baseUrl = stripSlash(opts.baseUrl);
-	const authHeaders = { authorization: `Bearer ${opts.apiKey}` };
 	const maxTokens = opts.probeMaxTokens ?? 64;
 	if (!Number.isInteger(maxTokens) || maxTokens <= 0) return {
 		succeeded: false,
@@ -7005,16 +6967,13 @@ async function preflightModels(opts) {
 	};
 	let served;
 	try {
-		const res = await fetchImpl(`${baseUrl}/models`, {
-			method: "GET",
-			headers: authHeaders
-		});
+		const res = await opts.request({ kind: "list-models" });
 		if (!res.ok) {
 			const text = await res.text().catch(() => "");
 			return {
 				succeeded: false,
 				value: null,
-				error: `preflightModels: GET ${baseUrl}/models → ${res.status} ${text.slice(0, 400)}`
+				error: `preflightModels: list-models → ${res.status} ${text.slice(0, 400)}`
 			};
 		}
 		const body = await res.json();
@@ -7024,7 +6983,7 @@ async function preflightModels(opts) {
 		return {
 			succeeded: false,
 			value: null,
-			error: `preflightModels: GET ${baseUrl}/models failed — ${err instanceof Error ? err.message : String(err)}`
+			error: `preflightModels: list-models failed — ${err instanceof Error ? err.message : String(err)}`
 		};
 	}
 	const results = [];
@@ -7043,20 +7002,10 @@ async function preflightModels(opts) {
 			continue;
 		}
 		try {
-			const res = await fetchImpl(`${baseUrl}/chat/completions`, {
-				method: "POST",
-				headers: {
-					...authHeaders,
-					"content-type": "application/json"
-				},
-				body: JSON.stringify({
-					model,
-					messages: [{
-						role: "user",
-						content: "ping"
-					}],
-					max_tokens: maxTokens
-				})
+			const res = await opts.request({
+				kind: "probe",
+				model,
+				maxOutputTokens: maxTokens
 			});
 			let detail = null;
 			let substitution = null;
@@ -7083,7 +7032,7 @@ async function preflightModels(opts) {
 			return {
 				succeeded: false,
 				value: null,
-				error: `preflightModels: probe POST ${baseUrl}/chat/completions (model ${model}) failed — ${err instanceof Error ? err.message : String(err)}`
+				error: `preflightModels: probe (model ${model}) failed — ${err instanceof Error ? err.message : String(err)}`
 			};
 		}
 	}
@@ -7285,6 +7234,6 @@ function rankRows(rows, weights) {
 	})).sort((a, b) => b.mean - a.mean);
 }
 //#endregion
-export { AGENT_PROFILE_KINDS, AgentEvalError, AnalystRegistry, BOOTSTRAP_GATE_MIN_N, BackendIntegrityError, BudgetBreachError, BudgetGuard, CODING_HARNESSES, ConfigError, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, CostLedger, CostLedgerPersistenceError, CostReceiptCaptureError, CostReservationExceededError, CostTracker, CrossFamilyError, DECISION_PAIRED_DELTA_STATISTIC, DEFAULT_PERMUTATIONS, DEFAULT_REDACTION_RULES, DEFAULT_RED_TEAM_CORPUS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, ERROR_COUNT_PATTERNS, EquivalenceProtocolError, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, FileSystemTraceStore, FindingsStore, HARNESS_NATIVE_MODEL, HeldOutGate, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, InMemoryTraceStore, JudgeError, LlmCallError, LlmClient, LlmResponseError, MANN_WHITNEY_EXACT_MAX_STATES, MANN_WHITNEY_EXACT_MAX_WORK, MODEL_PRICING, ModelSubstitutionError, MultiLayerVerifier, NoopRawProviderSink, NotFoundError, OUTPUT_VALUE, OtlpFileTraceStore, PairwiseSteeringOptimizer, ProductClient, PromptRegistry, REDACTION_VERSION, RunIntegrityError, RunRecordValidationError, SEMANTIC_CONCEPT_JUDGE_VERSION, ServedCrossFamilyError, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TraceEmitter, UNKNOWN_MODEL, VERIFICATION_STRATEGIES, VERIFICATION_STRATEGY_SOURCES, ValidationError, WILCOXON_EXACT_MAX_N, acquisitionPlansForKnowledgeGaps, agentProfileCellHashMaterial, agentProfileCellKey, agentProfileHash, agentProfileId, aggregateJudgeVerdicts, aggregateRunScore, analystFindingDigest, analystRunDigest, analystRunToFeedbackTrajectory, analystRunToReviewRequests, analyzeAntiSlop, analyzeRuns, analyzeSeries, analyzeTraces, argHash, assertCapabilityHeadroom, assertCrossFamily, assertCrossFamilyServed, assertLlmRoute, assertNoHiddenLeak, assertProductBenchmarkRun, assertRealBackend, assertRunCaptured, assertServedModel, assertServedModels, assertSingleBackend, assignFeedbackSplit, benjaminiHochberg, blendHeldout, blockingKnowledgeEval, bonferroni, bootstrapCi, budgetBreachView, buildAgentProfileCell, buildDefaultAnalystRegistry, buildEquivalenceRecord, buildReflectionPrompt, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, calibrateJudge, calibrateJudgeContinuous, callLlm, callLlmJson, canonicalJson, canonicalize, capabilityHeadroom, captureFetchToRawSink, certificationEvidenceDigest, checkCanaries, checkServedModel, checkTraceContracts, clamp01, classifyFailure, cliffsDelta, cohensD, comparePairedArms, completionVerdict, computeExperimentStats, computeFindingId, computeToolUseMetrics, confidenceInterval, contentHash, continuousAgreement, controlRunToFeedbackTrajectory, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, costForTokenPricing, costForUsage, costReceiptFromLlm, costReceiptFromLlmError, createAntiSlopJudge, createBoundedTraceAnalysisStore, createChatClient, createDspyRlmTraceEngine, createFeedbackTrajectory, createLlmCorrectnessChecker, createLlmReviewer, createTraceAnalyst, decidePairedPromotion, defaultBlendWeights, defineAgentEval, defineEquivalenceCheck, deployGateLayer, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, domainEvidencePattern, dominates, eProcess, ensembleJudge, equivalenceVerdict, errorStreakDetector, estimateCost, estimateTokens, evaluateActionPolicy, evaluateInterimReleaseConfidence, evaluateOracles, evaluateReleaseConfidence, expandProfileAxes, exportProductBenchmark, exportProductBenchmarkRuns, exportRunAsOtlp, extractErrorCount, extractProducedState, extractUsage, extractUsageFromSse, failureClusterView, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToOptimizerRow, fileVerdictCache, formatScorecardDiff, gainHistogram, gateTreatmentApplied, gradeOnHidden, gradeSemanticStatus, groupRunsByAgentProfileCell, harnessAxisOf, hashContent, hashJson, hiddenGrade, holm, improvementVerdict, inMemoryReviewStore, inferDomainKeywords, interRaterReliability, interpretCliffs, iqr, isBinaryOutcomeVector, isJudgeSpan, isLlmSpan, isModelPriced, isRunRecord, isToolSpan, isTransientLlmError, jsonShape, jsonlReviewStore, jsonlRunRecordBackend, judgeAgreementView, judgeFamily, judgeSpans, knowledgeReadinessTracePayload, leaderboard, llmJudge, loadScorecard, localCommandRunner, makeFinding, makeProposalFinding, mannWhitneyU, maximumChargeForLlmRequest, mcnemar, mcnemarPower, mcnemarRequiredN, minimumPairsForPairedDeltaTest, mintRolloutRows, modelHasSnapshot, modelPriceKey, mulberry32, notBlocked, objectiveEval, observeAll, otlpTextToTraceAnalysisStore, pairArms, pairRunRecords, pairedBinaryScale, pairedBootstrap, pairedCohensDz, pairedDeltaTest, pairedDeltaTieFraction, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedRiskDifferenceExact, pairedRiskDifferenceScore, pairedSignTest, pairedTTest, paretoChart, paretoFrontier, parseReflectionResponse, parseRunRecordSafe, parseRuntimeTrajectoryHookEvent, partialCredit, partitionHeldOut, passAtK, pearsonR, preflightModels, probeLlm, productBenchmarkRepoIdentity, profile_exports as profile, projectRuntimeTrajectoryEvidence, proposeSynthesisTargets, ranks, readProductBenchmarkManifest, recordRuns, recordRunsToScorecard, redTeamDataset, redTeamReport, redactString, regexMatches, renderPreferenceMemoryMarkdown, repeatedActionDetector, requiredPairedSampleSize, requiredSampleSize, resolveModelPricing, resolveSeat, roundTripRunRecord, routeFields, runAgentControlLoop, runCampaign, runCanaries, runCounterfactual, runEquivalenceCheck, runEvalCampaign, runIntentMatchJudge, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runProposeReview, runProposeReviewAsControlLoop, runSemanticConceptJudge, runTaskScore, runsForScenario, scoreKnowledgeReadiness, scoreRedTeamOutput, scoreTraceInsightReadiness, seatPresets, selfImprove, servedModelAcceptable, spearmanR, stripFencedJson, subjectiveEval, summarizeBackendIntegrity, summarizeNumberSeries, summarizePreferenceMemory, summaryTable, textInSnapshot, toAgentProfileJson, tokenizeDomainWords, toolSpansToTraceAnalysisStore, toolWasteView, traceContract, urlContains, userQuestionsForKnowledgeGaps, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyCompletion, viteDeployRunner, weightedComposite, weightedMean, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit, withHeldoutBlend, withJudgeRetry, wranglerDeployRunner };
+export { AGENT_PROFILE_KINDS, AgentEvalError, AnalystRegistry, BOOTSTRAP_GATE_MIN_N, BackendIntegrityError, BudgetBreachError, BudgetGuard, CODING_HARNESSES, ConfigError, CostAccountingIncompleteError, CostCallConflictError, CostCeilingReachedError, CostLedger, CostLedgerPersistenceError, CostReceiptCaptureError, CostReservationExceededError, CostTracker, CrossFamilyError, DECISION_PAIRED_DELTA_STATISTIC, DEFAULT_PERMUTATIONS, DEFAULT_REDACTION_RULES, DEFAULT_RED_TEAM_CORPUS, DEFAULT_TRACE_ANALYST_BUDGETS, DEFAULT_TRACE_ANALYST_KINDS, ERROR_COUNT_PATTERNS, EquivalenceProtocolError, FAILURE_CLASSES, FAILURE_MODE_KIND_SPEC, FileSystemFeedbackTrajectoryStore, FileSystemRawProviderSink, FileSystemTraceStore, FindingsStore, HARNESS_NATIVE_MODEL, HeldOutGate, InMemoryFeedbackTrajectoryStore, InMemoryRawProviderSink, InMemoryTraceStore, JudgeError, LlmCallError, LlmResponseError, MANN_WHITNEY_EXACT_MAX_STATES, MANN_WHITNEY_EXACT_MAX_WORK, MODEL_PRICING, ModelSubstitutionError, MultiLayerVerifier, NoopRawProviderSink, NotFoundError, OUTPUT_VALUE, OtlpFileTraceStore, PairwiseSteeringOptimizer, ProductClient, PromptRegistry, REDACTION_VERSION, RunIntegrityError, RunRecordValidationError, SEMANTIC_CONCEPT_JUDGE_VERSION, ServedCrossFamilyError, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, TraceEmitter, UNKNOWN_MODEL, VERIFICATION_STRATEGIES, VERIFICATION_STRATEGY_SOURCES, ValidationError, WILCOXON_EXACT_MAX_N, acquisitionPlansForKnowledgeGaps, agentProfileCellHashMaterial, agentProfileCellKey, agentProfileHash, agentProfileId, aggregateJudgeVerdicts, aggregateRunScore, analystFindingDigest, analystRunDigest, analystRunToFeedbackTrajectory, analystRunToReviewRequests, analyzeAntiSlop, analyzeRuns, analyzeSeries, analyzeTraces, argHash, assertCapabilityHeadroom, assertCrossFamily, assertCrossFamilyServed, assertNoHiddenLeak, assertProductBenchmarkRun, assertRealBackend, assertRunCaptured, assertServedModel, assertServedModels, assertSingleBackend, assignFeedbackSplit, benjaminiHochberg, blendHeldout, blockingKnowledgeEval, bonferroni, bootstrapCi, budgetBreachView, buildAgentProfileCell, buildDefaultAnalystRegistry, buildEquivalenceRecord, buildReflectionPrompt, buildTraceInsightContext, buildTraceInsightPrompt, buildTrajectory, calibrateJudge, calibrateJudgeContinuous, canonicalJson, capabilityHeadroom, captureFetchToRawSink, certificationEvidenceDigest, checkCanaries, checkServedModel, checkTraceContracts, clamp01, classifyFailure, cliffsDelta, cohensD, comparePairedArms, completionVerdict, computeExperimentStats, computeFindingId, computeToolUseMetrics, confidenceInterval, contentHash, continuousAgreement, controlRunToFeedbackTrajectory, corpusInterRaterAgreement, corpusInterRaterAgreementFromJudgeScores, costForTokenPricing, costForUsage, costReceiptFromLlm, costReceiptFromLlmError, createAntiSlopJudge, createBoundedTraceAnalysisStore, createChatClient, createDspyRlmTraceEngine, createFeedbackTrajectory, createLlmCorrectnessChecker, createLlmReviewer, createTraceAnalyst, decidePairedPromotion, defaultBlendWeights, defineAgentEval, defineEquivalenceCheck, deployGateLayer, describeTraceInsightScope, diffFindings, diffScorecard, discoverPersonas, domainEvidencePattern, dominates, eProcess, ensembleJudge, equivalenceVerdict, errorStreakDetector, estimateCost, estimateTokens, evaluateActionPolicy, evaluateInterimReleaseConfidence, evaluateOracles, evaluateReleaseConfidence, expandProfileAxes, exportProductBenchmark, exportProductBenchmarkRuns, exportRunAsOtlp, extractErrorCount, extractProducedState, extractUsage, extractUsageFromSse, failureClusterView, feedbackTrajectoriesToDatasetScenarios, feedbackTrajectoriesToOptimizerRows, feedbackTrajectoryToOptimizerRow, fileVerdictCache, formatScorecardDiff, gainHistogram, gateTreatmentApplied, gradeOnHidden, gradeSemanticStatus, groupRunsByAgentProfileCell, harnessAxisOf, hashContent, hashJson, hiddenGrade, holm, improvementVerdict, inMemoryReviewStore, inferDomainKeywords, interRaterReliability, interpretCliffs, iqr, isBinaryOutcomeVector, isJudgeSpan, isLlmSpan, isModelPriced, isRunRecord, isToolSpan, isTransientLlmError, jsonShape, jsonlReviewStore, jsonlRunRecordBackend, judgeAgreementView, judgeFamily, judgeSpans, knowledgeReadinessTracePayload, leaderboard, llmJudge, loadScorecard, localCommandRunner, makeFinding, makeProposalFinding, manifestContentDigest, mannWhitneyU, maximumChargeForLlmRequest, mcnemar, mcnemarPower, mcnemarRequiredN, minimumPairsForPairedDeltaTest, mintRolloutRows, modelHasSnapshot, modelPriceKey, mulberry32, notBlocked, objectiveEval, observeAll, otlpTextToTraceAnalysisStore, pairArms, pairRunRecords, pairedBinaryScale, pairedBootstrap, pairedCohensDz, pairedDeltaTest, pairedDeltaTieFraction, pairedEvalueSequence, pairedMde, pairedRiskDifference, pairedRiskDifferenceExact, pairedRiskDifferenceScore, pairedSignTest, pairedTTest, paretoChart, paretoFrontier, parseReflectionResponse, parseRunRecordSafe, partialCredit, partitionHeldOut, passAtK, pearsonR, preflightModels, productBenchmarkRepoIdentity, profile_exports as profile, projectRuntimeTrajectoryEvidence, proposeSynthesisTargets, ranks, readProductBenchmarkManifest, recordRuns, recordRunsToScorecard, redTeamDataset, redTeamReport, redactString, regexMatches, renderPreferenceMemoryMarkdown, repeatedActionDetector, requiredPairedSampleSize, requiredSampleSize, resolveModelPricing, resolveSeat, roundTripRunRecord, routeFields, runAgentControlLoop, runCampaign, runCanaries, runCounterfactual, runEquivalenceCheck, runEvalCampaign, runIntentMatchJudge, runKeywordCoverageJudge, runKeywordCoverageJudgeUrl, runProposeReview, runProposeReviewAsControlLoop, runSemanticConceptJudge, runTaskScore, runsForScenario, scoreKnowledgeReadiness, scoreRedTeamOutput, scoreTraceInsightReadiness, seatPresets, selfImprove, servedModelAcceptable, spearmanR, stripFencedJson, subjectiveEval, summarizeBackendIntegrity, summarizeNumberSeries, summarizePreferenceMemory, summaryTable, textInSnapshot, toAgentProfileJson, tokenizeDomainWords, toolSpansToTraceAnalysisStore, toolWasteView, traceContract, transientDispatchFailure, urlContains, userQuestionsForKnowledgeGaps, validateRunRecord, verbosityBias, verifyAgentProfileCell, verifyCompletion, viteDeployRunner, weightedComposite, weightedMean, wilcoxonSignedRank, wilson, withAssignedFeedbackSplit, withHeldoutBlend, withJudgeRetry, wranglerDeployRunner };
 
 //# sourceMappingURL=index.js.map