promptfoo 0.121.4 → 0.121.5

package/dist/src/main.js

@@ -1,34 +1,35 @@
 #!/usr/bin/env node
-import { C as getConfigDirectoryPath, D as getEnvInt, E as getEnvFloat, M as isNonInteractive, N as state, O as getEnvString, T as getEnvBool, _ as extractJsonObjects, a as isDebugEnabled, b as orderKeys, c as setLogCallback, g as extractFirstJsonObject, i as initializeRunLogging, j as isCI, l as setLogLevel, m as sanitizeObject, n as getLogLevel, s as logger, t as closeLogger, w as setConfigDirectoryPath } from "./logger-DksKw1Qc.js";
-import { A as getDefaultPort, F as VERSION, L as HUMAN_ASSERTION_TYPE, M as getShareApiBaseUrl, N as getShareViewBaseUrl, T as CLOUD_PROVIDER_PREFIX, i as fetchWithTimeout, j as getDefaultShareViewBaseUrl, k as TERMINAL_MAX_WIDTH, l as sleep, m as REQUEST_TIMEOUT_MS, n as fetchWithProxy, r as fetchWithRetries, t as clearAgentCache, v as isPromptfooSampleTarget } from "./fetch-BEWnXrrG.js";
-import { t as invariant } from "./invariant-BtWWVVhl.js";
-import { a as getAuthor, c as isLoggedIntoCloud, l as promptForEmailUnverified, n as checkEmailStatusAndMaybeExit, o as getUserEmail, r as clearUserEmail, s as getUserId, t as checkEmailStatus, u as setUserEmail } from "./accounts-F9d_5sMC.js";
-import { a as cloudConfig, c as writeGlobalConfigPartial, o as readGlobalConfig, r as CloudConfig } from "./cloud-DQZ5sVjW.js";
-import { n as TelemetryEventSchema, r as telemetry } from "./telemetry-CQPez_Jp.js";
-import { $ as HARM_PLUGINS, B as categoryAliases, D as ADDITIONAL_STRATEGIES, E as isUuid, F as getDefaultNFanout, G as ADDITIONAL_PLUGINS, H as riskCategorySeverityMap, J as CANARY_BREAKING_STRATEGY_IDS, K as ALL_PLUGINS, L as isFanoutStrategy, N as STRATEGY_COLLECTIONS, P as STRATEGY_COLLECTION_MAPPINGS, Q as FOUNDATION_PLUGINS, R as isMultiTurnStrategy, S as PluginConfigSchema, U as subCategoryDescriptions, W as ALIASED_PLUGIN_MAPPINGS, X as DEFAULT_PLUGINS, Y as DATASET_EXEMPT_PLUGINS, Z as FINANCIAL_PLUGINS, _ as PromptSchema, a as EvaluateOptionsSchema, at as MULTI_INPUT_VAR, b as ConversationMessageSchema, c as TestSuiteConfigSchema, dt as REMOTE_ONLY_PLUGIN_IDS, et as INSURANCE_PLUGINS, g as RedteamGenerateOptionsSchema, gt as CODING_AGENT_CORE_PLUGINS, h as RedteamConfigSchema, ht as UNALIGNED_PROVIDER_HARM_PLUGINS, i as EvalResultsFilterMode, it as MULTI_INPUT_EXCLUDED_PLUGINS, j as DEFAULT_STRATEGIES, k as ALL_STRATEGIES, l as TestSuiteSchema, lt as REDTEAM_MODEL, mt as TELECOM_PLUGINS, n as BaseAssertionTypesSchema, nt as LLAMA_GUARD_REPLICATE_PROVIDER, o as OutputFileExtension, ot as PHARMACY_PLUGINS, p as isApiProvider, pt as TEEN_SAFETY_PLUGINS, q as BIAS_PLUGINS, r as CommandLineOptionsSchema, rt as MEDICAL_PLUGINS, s as ResultFailureReason, st as PII_PLUGINS, t as AssertionOrSetSchema, tt as LLAMA_GUARD_ENABLED_CATEGORIES, u as UnifiedConfigSchema, ut as REDTEAM_PROVIDER_HARM_PLUGINS, v as ProviderOptionsSchema, w as StrategyConfigSchema, x as PartialGenerationError, y as ProvidersSchema, z as Severity } from "./types-Dm9JM6Vb.js";
-import { r as importModule, t as getDirectory } from "./esm-tVgYPY-f.js";
-import { a as getNunjucksEngine, i as extractVariablesFromTemplates, r as extractVariablesFromTemplate, t as renderEnvOnlyInObject } from "./render-CMEpfLaO.js";
-import { a as openAuthBrowser, d as getRemoteHealthUrl, f as neverGenerateRemote, i as checkServerRunning, l as getRemoteGenerationUrl, m as shouldGenerateRemote, n as BrowserBehaviorNames, o as openBrowser, s as promptUser, t as BrowserBehavior } from "./server-BNYztJkh.js";
-import { A as renderPrompt, B as MAX_CHARS_PER_MESSAGE_MODIFIER_KEY, C as extractVariablesFromJson, F as TokenUsageTracker, H as getMaxCharsPerMessageModifierValue, P as redteamProviderManager, S as extractPromptFromTags, T as getShortPluginId, V as getGeneratedPromptOverLimit, _ as pluginMatchesStrategyTargets, _t as getUserTeams, a as resolveProviderConfigs, b as extractGoalFromPrompt, bt as resolveTeamFromIdentifier, c as createTransformRequest, ct as canCreateTargets, d as Strategies, dt as getConfigFromCloud, f as loadStrategy, ft as getEvalConfigFromCloud, g as retrieveMedia, gt as getProviderFromCloud, h as mediaExists, i as resolveProvider, l as createTransformResponse, lt as checkCloudPermissions, m as getMediaStorage, mt as getPluginSeverityOverridesFromCloud, n as loadApiProvider, o as MCPProvider, p as validateStrategies, pt as getOrgContext, r as loadApiProviders, s as HttpProvider, t as getProviderIds, ut as getCloudDatabaseId, vt as isCloudProvider, xt as resolveTeamId, yt as makeRequest$1, z as PromptfooHarmfulCompletionProvider } from "./providers-BV_KMZje.js";
-import { a as fetchWithCache, n as clearCache, r as disableCache, t as cache_exports } from "./cache-mb7c8hbp.js";
-import { i as isJavascriptFile } from "./fileExtensions-AWa2ZML4.js";
-import { A as getProviderDescription, C as deduplicateTestCases, D as resultIsForTestCase, E as getTestCaseDeduplicationKey, O as checkProviderApiKeys, S as setupEnv, T as filterRuntimeVars, a as ComparisonEvalNotFoundError, c as mergeComparisonTables, f as maybeLoadConfigFromExternalFile, i as writeOutput, k as doesProviderRefMatch, l as fetchCsvFromGoogleSheet, n as createOutputMetadata, o as evalTableToJson, p as maybeLoadFromExternalFile, r as writeMultipleOutputs, s as generateEvalCsv, t as printBorder, v as readFilters, w as extractRuntimeVars, y as readOutput } from "./util-DxWpWjhc.js";
-import { r as runPython } from "./pythonUtils-CnndUbW-.js";
-import { A as DivergentRepetitionPlugin, B as sampleArray, C as getPiiLeakTestsForCategory, D as HarmbenchPlugin, E as ImitationPlugin, F as AegisPlugin, I as RedteamGraderBase, L as RedteamPluginBase, M as CrossSessionLeakPlugin, N as ContractPlugin, O as HallucinationPlugin, P as BeavertailsPlugin, R as getCustomPolicies, S as PlinyPlugin, T as IntentPlugin, V as fetchHuggingFaceDataset, _ as PoliticsPlugin, a as UnverifiableClaimsPlugin, b as isValidPolicyObject, c as ToolDiscoveryPlugin, d as TeenSafetyDangerousContentPlugin, dt as processPrompts, f as TeenSafetyAgeRestrictedGoodsAndServicesPlugin, ft as readPrompts, g as PromptExtractionPlugin, h as RbacPlugin, i as VLGuardPlugin, j as DebugAccessPlugin, k as ExcessiveAgencyPlugin, l as TeenSafetyHarmfulBodyIdealsPlugin, lt as doRemoteGrading, m as ShellInjectionPlugin, o as UnsafeBenchPlugin, p as SqlInjectionPlugin, pt as readProviderPromptMap, r as VLSUPlugin, s as ToxicChatPlugin, t as GRADERS, u as TeenSafetyDangerousRoleplayPlugin, ut as getDefaultProviders, v as PolicyPlugin, w as OverreliancePlugin, x as makeInlinePolicyIdSync, y as determinePolicyTypeFromId, z as retryWithDeduplication } from "./graders-eIHhRqoC.js";
-import { i as generateIdFromPrompt } from "./utils-BLJKfv0y.js";
-import { n as sha256, t as randomSequence } from "./createHash-ChI45QR1.js";
-import { a as createEmptyTokenUsage, i as createEmptyAssertions, n as accumulateResponseTokenUsage, r as accumulateTokenUsage, t as accumulateAssertionTokenUsage } from "./tokenUsageUtils-CXrvO-wA.js";
-import { n as getBlobUrl, t as getBlobByHash } from "./blobs-BQWqnnvL.js";
-import { a as evalResultsTable, c as evalsToPromptsTable, d as promptsTable, g as getDb, h as closeDbIfOpen, i as datasetsTable, l as evalsToTagsTable, n as blobReferencesTable, o as evalsTable, p as tagsTable, r as configsTable, s as evalsToDatasetsTable, t as blobAssetsTable, u as modelAuditsTable } from "./tables-C4CH3zRr.js";
-import { n as isBlobStorageEnabled } from "./extractor-DNSeBVOJ.js";
-import { n as escapeRegExp, t as ellipsize } from "./text-KvuD2Iko.js";
-import { n as getTraceStore } from "./store-CYEy5J2D.js";
-import { c as assertions_default, l as readAssertions, n as evaluate$1, o as accumulateNamedMetric, s as doesPromptRefMatch, u as runAssertions } from "./evaluator-IvuDYSvQ.js";
-import { n as setupSignalWatcher, t as readSignalEvalId } from "./signal-CE5G3a7x.js";
-import { a as createEvalId, n as EvalQueries, s as getEvalSummaries, t as Eval } from "./eval-u4UVafl6.js";
-import { t as EvalResult } from "./evalResult-D3hVYFis.js";
-import { t as formatDuration } from "./formatDuration-DZzPsexs.js";
-import { n as shouldUseInkUI } from "./interactiveCheck-CLERUB0c.js";
+import { C as getConfigDirectoryPath, D as getEnvInt, E as getEnvFloat, M as isNonInteractive, N as state, O as getEnvString, T as getEnvBool, _ as extractJsonObjects, a as isDebugEnabled, b as orderKeys, c as setLogCallback, g as extractFirstJsonObject, i as initializeRunLogging, j as isCI, l as setLogLevel, m as sanitizeObject, n as getLogLevel, s as logger, t as closeLogger, w as setConfigDirectoryPath } from "./logger-BbY6ypFL.js";
+import { A as getDefaultPort, F as HUMAN_ASSERTION_TYPE, M as getShareApiBaseUrl, N as getShareViewBaseUrl, T as CLOUD_PROVIDER_PREFIX, i as fetchWithTimeout, j as getDefaultShareViewBaseUrl, k as TERMINAL_MAX_WIDTH, l as sleep, m as REQUEST_TIMEOUT_MS, n as fetchWithProxy, r as fetchWithRetries, t as clearAgentCache, v as isPromptfooSampleTarget } from "./fetch-B6ch2nU2.js";
+import { n as VERSION } from "./version-CbpiUINz.js";
+import { t as invariant } from "./invariant-B2Rf6avk.js";
+import { a as getAuthor, c as isLoggedIntoCloud, l as promptForEmailUnverified, n as checkEmailStatusAndMaybeExit, o as getUserEmail, r as clearUserEmail, s as getUserId, t as checkEmailStatus, u as setUserEmail } from "./accounts-CLJHCDDb.js";
+import { a as cloudConfig, c as writeGlobalConfigPartial, o as readGlobalConfig, r as CloudConfig } from "./cloud-Da0bofJd.js";
+import { n as TelemetryEventSchema, r as telemetry } from "./telemetry-DWdGHvEf.js";
+import { $ as HARM_PLUGINS, B as categoryAliases, D as ADDITIONAL_STRATEGIES, E as isUuid, F as getDefaultNFanout, G as ADDITIONAL_PLUGINS, H as riskCategorySeverityMap, J as CANARY_BREAKING_STRATEGY_IDS, K as ALL_PLUGINS, L as isFanoutStrategy, N as STRATEGY_COLLECTIONS, P as STRATEGY_COLLECTION_MAPPINGS, Q as FOUNDATION_PLUGINS, R as isMultiTurnStrategy, S as PluginConfigSchema, U as subCategoryDescriptions, W as ALIASED_PLUGIN_MAPPINGS, X as DEFAULT_PLUGINS, Y as DATASET_EXEMPT_PLUGINS, Z as FINANCIAL_PLUGINS, _ as PromptSchema, a as EvaluateOptionsSchema, at as MULTI_INPUT_VAR, b as ConversationMessageSchema, c as TestSuiteConfigSchema, dt as REMOTE_ONLY_PLUGIN_IDS, et as INSURANCE_PLUGINS, g as RedteamGenerateOptionsSchema, gt as CODING_AGENT_CORE_PLUGINS, h as RedteamConfigSchema, ht as UNALIGNED_PROVIDER_HARM_PLUGINS, i as EvalResultsFilterMode, it as MULTI_INPUT_EXCLUDED_PLUGINS, j as DEFAULT_STRATEGIES, k as ALL_STRATEGIES, l as TestSuiteSchema, lt as REDTEAM_MODEL, mt as TELECOM_PLUGINS, n as BaseAssertionTypesSchema, nt as LLAMA_GUARD_REPLICATE_PROVIDER, o as OutputFileExtension, ot as PHARMACY_PLUGINS, p as isApiProvider, pt as TEEN_SAFETY_PLUGINS, q as BIAS_PLUGINS, r as CommandLineOptionsSchema, rt as MEDICAL_PLUGINS, s as ResultFailureReason, st as PII_PLUGINS, t as AssertionOrSetSchema, tt as LLAMA_GUARD_ENABLED_CATEGORIES, u as UnifiedConfigSchema, ut as REDTEAM_PROVIDER_HARM_PLUGINS, v as ProviderOptionsSchema, w as StrategyConfigSchema, x as PartialGenerationError, y as ProvidersSchema, z as Severity } from "./types-BVH9hjgW.js";
+import { r as importModule, t as getDirectory } from "./esm-BX8fwlAO.js";
+import { a as getNunjucksEngine, i as extractVariablesFromTemplates, r as extractVariablesFromTemplate, t as renderEnvOnlyInObject } from "./render-eui5p5mL.js";
+import { a as openAuthBrowser, d as getRemoteHealthUrl, f as neverGenerateRemote, i as checkServerRunning, l as getRemoteGenerationUrl, m as shouldGenerateRemote, n as BrowserBehaviorNames, o as openBrowser, s as promptUser, t as BrowserBehavior } from "./server-ByiF3qlg.js";
+import { A as validateStrategies, B as extractVariablesFromJson, H as getShortPluginId, L as extractGoalFromPrompt, M as mediaExists, N as retrieveMedia, O as Strategies, P as pluginMatchesStrategyTargets, Q as TokenUsageTracker, Z as redteamProviderManager, _t as getUserTeams, a as resolveProviderConfigs, at as getMaxCharsPerMessageModifierValue, bt as resolveTeamFromIdentifier, ct as canCreateTargets, d as MCPProvider, dt as getConfigFromCloud, f as HttpProvider, ft as getEvalConfigFromCloud, gt as getProviderFromCloud, i as resolveProvider, it as getGeneratedPromptOverLimit, j as getMediaStorage, k as loadStrategy, lt as checkCloudPermissions, m as createTransformResponse, mt as getPluginSeverityOverridesFromCloud, n as loadApiProvider, nt as PromptfooHarmfulCompletionProvider, p as createTransformRequest, pt as getOrgContext, q as renderPrompt, r as loadApiProviders, rt as MAX_CHARS_PER_MESSAGE_MODIFIER_KEY, t as getProviderIds, ut as getCloudDatabaseId, vt as isCloudProvider, xt as resolveTeamId, yt as makeRequest$1, z as extractPromptFromTags } from "./providers-DT-GtF2t.js";
+import { a as fetchWithCache, n as clearCache, r as disableCache, t as cache_exports } from "./cache-BI5BY7ey.js";
+import { i as isJavascriptFile } from "./fileExtensions-DysCsxNG.js";
+import { r as runPython } from "./pythonUtils-DNqbnRdx.js";
+import { A as maybeLoadFromExternalFile, F as readFilters, I as readOutput, T as normalizeProviderRef, _ as doesProviderRefMatch, a as ComparisonEvalNotFoundError, c as mergeComparisonTables, d as deduplicateTestCases, f as extractRuntimeVars, g as checkProviderApiKeys, h as resultIsForTestCase, i as writeOutput, k as maybeLoadConfigFromExternalFile, l as fetchCsvFromGoogleSheet, m as getTestCaseDeduplicationKey, n as createOutputMetadata, o as evalTableToJson, p as filterRuntimeVars, r as writeMultipleOutputs, s as generateEvalCsv, t as printBorder, u as setupEnv, v as getProviderDescription } from "./util-BQOCAHQC.js";
+import { a as createEmptyTokenUsage, i as createEmptyAssertions, n as accumulateResponseTokenUsage, r as accumulateTokenUsage, t as accumulateAssertionTokenUsage } from "./tokenUsageUtils-2wIvAhB3.js";
+import { n as getBlobUrl, t as getBlobByHash } from "./blobs-DjLby-uP.js";
+import { a as evalResultsTable, c as evalsToPromptsTable, d as promptsTable, g as getDb, h as closeDbIfOpen, i as datasetsTable, l as evalsToTagsTable, n as blobReferencesTable, o as evalsTable, p as tagsTable, r as configsTable, s as evalsToDatasetsTable, t as blobAssetsTable, u as modelAuditsTable } from "./tables-NlvH23ky.js";
+import { n as isBlobStorageEnabled } from "./extractor-DxyiFhPk.js";
+import { n as escapeRegExp, t as ellipsize } from "./text-D4lz-Jg_.js";
+import { n as getTraceStore } from "./store-CPh25336.js";
+import { n as sha256, t as randomSequence } from "./createHash-DPpsZgFF.js";
+import { A as DivergentRepetitionPlugin, B as sampleArray, C as getPiiLeakTestsForCategory, D as HarmbenchPlugin, E as ImitationPlugin, F as AegisPlugin, I as RedteamGraderBase, J as doRemoteGrading, L as RedteamPluginBase, M as CrossSessionLeakPlugin, N as ContractPlugin, O as HallucinationPlugin, P as BeavertailsPlugin, R as getCustomPolicies, S as PlinyPlugin, St as getDefaultProviders, T as IntentPlugin, V as fetchHuggingFaceDataset, X as readPrompts, Y as processPrompts, Z as readProviderPromptMap, _ as PoliticsPlugin, a as UnverifiableClaimsPlugin, b as isValidPolicyObject, c as ToolDiscoveryPlugin, d as TeenSafetyDangerousContentPlugin, f as TeenSafetyAgeRestrictedGoodsAndServicesPlugin, g as PromptExtractionPlugin, h as RbacPlugin, i as VLGuardPlugin, j as DebugAccessPlugin, k as ExcessiveAgencyPlugin, l as TeenSafetyHarmfulBodyIdealsPlugin, m as ShellInjectionPlugin, o as UnsafeBenchPlugin, p as SqlInjectionPlugin, r as VLSUPlugin, s as ToxicChatPlugin, t as GRADERS, u as TeenSafetyDangerousRoleplayPlugin, v as PolicyPlugin, w as OverreliancePlugin, x as makeInlinePolicyIdSync, y as determinePolicyTypeFromId, z as retryWithDeduplication } from "./graders-CgPn32yp.js";
+import { c as assertions_default, l as readAssertions, n as evaluate$1, o as accumulateNamedMetric, s as doesPromptRefMatch, u as runAssertions } from "./evaluator-D-UIbbYq.js";
+import { i as generateIdFromPrompt } from "./utils-kt7lv30R.js";
+import { n as setupSignalWatcher, t as readSignalEvalId } from "./signal-DTtUuU3l.js";
+import { a as createEvalId, n as EvalQueries, s as getEvalSummaries, t as Eval } from "./eval-DJ_4A-tr.js";
+import { t as EvalResult } from "./evalResult-D6P5I5il.js";
+import { t as formatDuration } from "./formatDuration-Ch4A7G3o.js";
+import { n as shouldUseInkUI } from "./interactiveCheck-CCICw2cy.js";
 import { fileURLToPath } from "node:url";
 import { Command, InvalidArgumentError } from "commander";
 import { z } from "zod";
@@ -100,7 +101,7 @@ function runCommand(program) {
 			hasGithubPr: !!cmdObj.githubPr,
 			hasGuidance: !!(cmdObj.guidance || cmdObj.guidanceFile)
 		});
-		const { executeScan } = await import("./scanner-1DqWi1Ej.js");
+		const { executeScan } = await import("./scanner-nOCWNIXa.js");
 		await executeScan(repoPath, cmdObj);
 	});
 }
@@ -572,47 +573,11 @@ function filterPrompts(prompts, filterPromptsOption) {
 //#endregion
 //#region src/commands/eval/filterProviders.ts
 /**
-* Checks if a value is a valid provider ID (non-empty string).
-*/
-function isValidProviderId(id) {
-	return id !== null && id !== void 0 && typeof id === "string" && id !== "";
-}
-/**
 * Extracts the id and label from a raw provider config without instantiating it.
 * Handles all provider config formats: string, function, ProviderOptions, ProviderOptionsMap.
 */
 function getProviderIdAndLabel(provider, index) {
-	if (typeof provider === "string") return { id: provider };
-	if (typeof provider === "function") {
-		const label = provider.label;
-		return {
-			id: label ?? `custom-function-${index}`,
-			label
-		};
-	}
-	const providerId = provider.id;
-	if ("id" in provider && isValidProviderId(providerId)) return {
-		id: providerId,
-		label: provider.label
-	};
-	const keys = Object.keys(provider);
-	if (keys.length > 0) {
-		const id = keys[0];
-		const value = provider[id];
-		if (typeof value === "object" && value !== null) return {
-			id: value.id || id,
-			label: value.label
-		};
-	}
-	const label = provider.label;
-	if (isValidProviderId(label)) return {
-		id: label,
-		label
-	};
-	return {
-		id: `unknown-${index}`,
-		label
-	};
+	return normalizeProviderRef(provider, { index });
 }
 /**
 * Filters raw provider configs BEFORE instantiation.
@@ -3208,7 +3173,7 @@ function generateTable(evaluateTable, tableCellMaxLength = 250, maxRows = 25) {
 	for (const row of evaluateTable.body.slice(0, maxRows)) table.push([...row.vars.map((v) => ellipsize(v, tableCellMaxLength)), ...row.outputs.map(({ pass, text, failureReason: failureType }) => {
 		text = ellipsize(text, tableCellMaxLength);
 		if (pass) return chalk.green("[PASS] ") + text;
-		else return chalk.red(failureType === ResultFailureReason.ASSERT ? "[FAIL] " : "[ERROR] ") + text.split("---").map((c, idx) => idx === 0 ? chalk.red.bold(c) : c).join("---");
+		return chalk.red(failureType === ResultFailureReason.ASSERT ? "[FAIL] " : "[ERROR] ") + text.split("---").map((c, idx) => idx === 0 ? chalk.red.bold(c) : c).join("---");
 	})]);
 	return table.toString();
 }
@@ -3293,6 +3258,115 @@ function shouldShareResults(opts) {
 }
 //#endregion
 //#region src/commands/eval/summary.ts
+function getCompletionMessage({ completionType, evalId, shareableUrl, wasAborted, writeToDatabase, activelySharing }) {
+	if (wasAborted) {
+		const idSuffix = writeToDatabase ? ` (ID: ${chalk.cyan(evalId)})` : "";
+		return `${chalk.red("✗")} ${completionType} aborted${idSuffix}`;
+	}
+	if (writeToDatabase && shareableUrl) return `${chalk.green("✓")} ${completionType} complete: ${shareableUrl}`;
+	if (writeToDatabase && activelySharing) return `${chalk.green("✓")} ${completionType} complete`;
+	if (writeToDatabase) return `${chalk.green("✓")} ${completionType} complete (ID: ${chalk.cyan(evalId)})`;
+	return `${chalk.green("✓")} ${completionType} complete`;
+}
+function getAbortSummaryLines(targetErrorStatus) {
+	if (targetErrorStatus == null) return [];
+	return [
+		"",
+		chalk.red.bold("Scan stopped: Target is unavailable and will not recover on retry."),
+		chalk.red(`  Target returned HTTP ${targetErrorStatus}`),
+		"",
+		chalk.yellow("Possible causes:"),
+		chalk.yellow("  • Invalid API key or authentication (401/403)"),
+		chalk.yellow("  • Target endpoint does not exist (404)"),
+		chalk.yellow("  • Server does not support the request (501)"),
+		"",
+		chalk.cyan("To fix: Check your target configuration and credentials.")
+	];
+}
+function getGuidanceLines({ writeToDatabase, shareableUrl, wantsToShare, activelySharing, hasExplicitDisable, cloudEnabled }) {
+	if (!writeToDatabase || shareableUrl || wantsToShare || activelySharing) return [];
+	const lines = ["", `» View results: ${chalk.green.bold("promptfoo view")}`];
+	if (!hasExplicitDisable) lines.push(cloudEnabled ? `» Create shareable URL: ${chalk.green.bold("promptfoo share")}` : `» Share with your team: ${chalk.green.bold("https://promptfoo.app")}`);
+	lines.push(`» Feedback: ${chalk.green.bold("https://promptfoo.dev/feedback")}`);
+	return lines;
+}
+function buildUsageDetails(usage, total) {
+	const parts = [];
+	if (usage.prompt && usage.prompt > 0) parts.push(`${usage.prompt.toLocaleString()} prompt`);
+	if (usage.completion && usage.completion > 0) parts.push(`${usage.completion.toLocaleString()} completion`);
+	if (usage.cached && usage.cached > 0) parts.push(usage.cached === total && parts.length === 0 ? "cached" : `${usage.cached.toLocaleString()} cached`);
+	if (usage.completionDetails?.reasoning && usage.completionDetails.reasoning > 0) parts.push(`${usage.completionDetails.reasoning.toLocaleString()} reasoning`);
+	return parts;
+}
+function getTokenUsageLines(tokenUsage, isRedteam, tracker) {
+	const hasEvalTokens = (tokenUsage.total || 0) > 0 || (tokenUsage.prompt || 0) + (tokenUsage.completion || 0) > 0;
+	const hasGradingTokens = tokenUsage.assertions && (tokenUsage.assertions.total || 0) > 0;
+	if (!hasEvalTokens && !hasGradingTokens) return [];
+	const combinedTotal = (tokenUsage.prompt || 0) + (tokenUsage.completion || 0);
+	const evalTokens = {
+		prompt: tokenUsage.prompt || 0,
+		completion: tokenUsage.completion || 0,
+		total: tokenUsage.total || combinedTotal,
+		cached: tokenUsage.cached || 0,
+		numRequests: tokenUsage.numRequests || 0,
+		completionDetails: tokenUsage.completionDetails || {
+			reasoning: 0,
+			acceptedPrediction: 0,
+			rejectedPrediction: 0
+		}
+	};
+	const lines = [`${chalk.bold("Total Tokens:")} ${chalk.white.bold((evalTokens.total + (tokenUsage.assertions?.total || 0)).toLocaleString())}`];
+	if (isRedteam && tokenUsage.numRequests) lines.push(`  ${chalk.gray("Probes:")} ${chalk.white(tokenUsage.numRequests.toLocaleString())}`);
+	if (evalTokens.total > 0) {
+		const evalParts = buildUsageDetails(evalTokens, evalTokens.total);
+		lines.push(`  ${chalk.gray("Eval:")} ${chalk.white(evalTokens.total.toLocaleString())} (${evalParts.join(", ")})`);
+	}
+	if (tokenUsage.assertions?.total && tokenUsage.assertions.total > 0) {
+		const gradingParts = buildUsageDetails(tokenUsage.assertions, tokenUsage.assertions.total);
+		lines.push(`  ${chalk.gray("Grading:")} ${chalk.white(tokenUsage.assertions.total.toLocaleString())} (${gradingParts.join(", ")})`);
+	}
+	lines.push(...getProviderUsageLines(tracker));
+	return lines;
+}
+function getProviderUsageLines(tracker) {
+	const providerIds = tracker.getProviderIds();
+	if (providerIds.length <= 1) return [];
+	const sortedProviders = providerIds.map((id) => ({
+		id,
+		usage: tracker.getProviderUsage(id)
+	})).filter((p) => p.usage != null).sort((a, b) => (b.usage.total || 0) - (a.usage.total || 0));
+	const lines = ["", chalk.bold("Providers:")];
+	for (const { id, usage } of sortedProviders) {
+		if ((usage.total || 0) === 0 && (usage.prompt || 0) + (usage.completion || 0) === 0) continue;
+		const displayTotal = usage.total || (usage.prompt || 0) + (usage.completion || 0);
+		const displayId = id.includes(" (") ? id.substring(0, id.indexOf(" (")) : id;
+		const details = buildUsageDetails(usage, displayTotal);
+		const requestInfo = `${usage.numRequests || 0} requests`;
+		const separator = details.length > 0 ? "; " : "";
+		lines.push(`  ${chalk.gray(`${displayId}:`)} ${chalk.white(displayTotal.toLocaleString())} (${requestInfo}${separator}${details.join(", ")})`);
+	}
+	return lines;
+}
+function formatResultPercentage(count, totalTests) {
+	const percentage = totalTests === 0 ? 0 : count / totalTests * 100;
+	return percentage === 0 || percentage === 100 ? `${percentage.toFixed(0)}%` : `${percentage.toFixed(2)}%`;
+}
+function formatResultLine(count, label, icon, iconColor, totalTests) {
+	return `  ${icon ? `${iconColor(icon)} ` : ""}${chalk.white.bold(count.toLocaleString())} ${chalk.white(label)} ${chalk.gray(`(${formatResultPercentage(count, totalTests)})`)}`;
+}
+function getResultsLines({ successes, failures, errors, duration, maxConcurrency }) {
+	const totalTests = successes + failures + errors;
+	const errorLabel = errors === 1 ? "error" : "errors";
+	return [
+		"",
+		chalk.bold("Results:"),
+		formatResultLine(successes, "passed", successes > 0 ? "✓" : void 0, chalk.green, totalTests),
+		formatResultLine(failures, "failed", failures > 0 ? "✗" : void 0, chalk.red, totalTests),
+		formatResultLine(errors, errorLabel, errors > 0 ? "✗" : void 0, chalk.red, totalTests),
+		chalk.gray(`Duration: ${formatDuration(duration)} (concurrency: ${maxConcurrency})`),
+		""
+	];
+}
 /**
 * Generate formatted evaluation summary output for CLI display.
 *
@@ -3331,115 +3405,28 @@ function shouldShareResults(opts) {
 * ```
 */
 function generateEvalSummary(params) {
-	const { evalId, isRedteam, writeToDatabase, shareableUrl, wantsToShare, hasExplicitDisable, cloudEnabled, activelySharing = false, tokenUsage, successes, failures, errors, duration, maxConcurrency, tracker, targetErrorStatus } = params;
-	const lines = [];
-	const completionType = isRedteam ? "Red team" : "Eval";
-	const wasAborted = targetErrorStatus != null;
-	let completionMessage;
-	if (wasAborted) {
-		completionMessage = `${chalk.red("✗")} ${completionType} aborted`;
-		if (writeToDatabase) completionMessage += ` (ID: ${chalk.cyan(evalId)})`;
-	} else if (writeToDatabase && shareableUrl) completionMessage = `${chalk.green("✓")} ${completionType} complete: ${shareableUrl}`;
-	else if (writeToDatabase && activelySharing) completionMessage = `${chalk.green("✓")} ${completionType} complete`;
-	else if (writeToDatabase) completionMessage = `${chalk.green("✓")} ${completionType} complete (ID: ${chalk.cyan(evalId)})`;
-	else completionMessage = `${chalk.green("✓")} ${completionType} complete`;
-	lines.push(completionMessage);
-	if (wasAborted && targetErrorStatus != null) {
-		lines.push("");
-		lines.push(chalk.red.bold("Scan stopped: Target is unavailable and will not recover on retry."));
-		lines.push(chalk.red(`  Target returned HTTP ${targetErrorStatus}`));
-		lines.push("");
-		lines.push(chalk.yellow("Possible causes:"));
-		lines.push(chalk.yellow("  • Invalid API key or authentication (401/403)"));
-		lines.push(chalk.yellow("  • Target endpoint does not exist (404)"));
-		lines.push(chalk.yellow("  • Server does not support the request (501)"));
-		lines.push("");
-		lines.push(chalk.cyan("To fix: Check your target configuration and credentials."));
-	}
-	if (writeToDatabase && !shareableUrl && !wantsToShare && !activelySharing) {
-		lines.push("");
-		lines.push(`» View results: ${chalk.green.bold("promptfoo view")}`);
-		if (!hasExplicitDisable) if (cloudEnabled) lines.push(`» Create shareable URL: ${chalk.green.bold("promptfoo share")}`);
-		else lines.push(`» Share with your team: ${chalk.green.bold("https://promptfoo.app")}`);
-		lines.push(`» Feedback: ${chalk.green.bold("https://promptfoo.dev/feedback")}`);
-	}
-	lines.push("");
-	const hasEvalTokens = (tokenUsage.total || 0) > 0 || (tokenUsage.prompt || 0) + (tokenUsage.completion || 0) > 0;
-	const hasGradingTokens = tokenUsage.assertions && (tokenUsage.assertions.total || 0) > 0;
-	if (hasEvalTokens || hasGradingTokens) {
-		const combinedTotal = (tokenUsage.prompt || 0) + (tokenUsage.completion || 0);
-		const evalTokens = {
-			prompt: tokenUsage.prompt || 0,
-			completion: tokenUsage.completion || 0,
-			total: tokenUsage.total || combinedTotal,
-			cached: tokenUsage.cached || 0,
-			completionDetails: tokenUsage.completionDetails || {
-				reasoning: 0,
-				acceptedPrediction: 0,
-				rejectedPrediction: 0
-			}
-		};
-		const grandTotal = evalTokens.total + (tokenUsage.assertions?.total || 0);
-		lines.push(`${chalk.bold("Total Tokens:")} ${chalk.white.bold(grandTotal.toLocaleString())}`);
-		if (isRedteam && tokenUsage.numRequests) lines.push(`  ${chalk.gray("Probes:")} ${chalk.white(tokenUsage.numRequests.toLocaleString())}`);
-		if (evalTokens.total > 0) {
-			const evalParts = [];
-			if (evalTokens.prompt > 0) evalParts.push(`${evalTokens.prompt.toLocaleString()} prompt`);
-			if (evalTokens.completion > 0) evalParts.push(`${evalTokens.completion.toLocaleString()} completion`);
-			if (evalTokens.cached > 0) if (evalTokens.cached === evalTokens.total && evalParts.length === 0) evalParts.push("cached");
-			else evalParts.push(`${evalTokens.cached.toLocaleString()} cached`);
-			if (evalTokens.completionDetails?.reasoning && evalTokens.completionDetails.reasoning > 0) evalParts.push(`${evalTokens.completionDetails.reasoning.toLocaleString()} reasoning`);
-			lines.push(`  ${chalk.gray("Eval:")} ${chalk.white(evalTokens.total.toLocaleString())} (${evalParts.join(", ")})`);
-		}
-		if (tokenUsage.assertions && tokenUsage.assertions.total && tokenUsage.assertions.total > 0) {
-			const gradingParts = [];
-			if (tokenUsage.assertions.prompt && tokenUsage.assertions.prompt > 0) gradingParts.push(`${tokenUsage.assertions.prompt.toLocaleString()} prompt`);
-			if (tokenUsage.assertions.completion && tokenUsage.assertions.completion > 0) gradingParts.push(`${tokenUsage.assertions.completion.toLocaleString()} completion`);
-			if (tokenUsage.assertions.cached && tokenUsage.assertions.cached > 0) if (tokenUsage.assertions.cached === tokenUsage.assertions.total && gradingParts.length === 0) gradingParts.push("cached");
-			else gradingParts.push(`${tokenUsage.assertions.cached.toLocaleString()} cached`);
-			if (tokenUsage.assertions.completionDetails?.reasoning && tokenUsage.assertions.completionDetails.reasoning > 0) gradingParts.push(`${tokenUsage.assertions.completionDetails.reasoning.toLocaleString()} reasoning`);
-			lines.push(`  ${chalk.gray("Grading:")} ${chalk.white(tokenUsage.assertions.total.toLocaleString())} (${gradingParts.join(", ")})`);
-		}
-		const providerIds = tracker.getProviderIds();
-		if (providerIds.length > 1) {
-			lines.push("");
-			lines.push(chalk.bold("Providers:"));
-			const sortedProviders = providerIds.map((id) => ({
-				id,
-				usage: tracker.getProviderUsage(id)
-			})).filter((p) => p.usage != null).sort((a, b) => (b.usage.total || 0) - (a.usage.total || 0));
-			for (const { id, usage } of sortedProviders) if ((usage.total || 0) > 0 || (usage.prompt || 0) + (usage.completion || 0) > 0) {
-				const displayTotal = usage.total || (usage.prompt || 0) + (usage.completion || 0);
-				const displayId = id.includes(" (") ? id.substring(0, id.indexOf(" (")) : id;
-				const details = [];
-				if (usage.prompt && usage.prompt > 0) details.push(`${usage.prompt.toLocaleString()} prompt`);
-				if (usage.completion && usage.completion > 0) details.push(`${usage.completion.toLocaleString()} completion`);
-				if (usage.cached && usage.cached > 0) if (usage.cached === displayTotal && details.length === 0) details.push("cached");
-				else details.push(`${usage.cached.toLocaleString()} cached`);
-				if (usage.completionDetails?.reasoning && usage.completionDetails.reasoning > 0) details.push(`${usage.completionDetails.reasoning.toLocaleString()} reasoning`);
-				const breakdown = ` (${`${usage.numRequests || 0} requests`}${details.length > 0 ? "; " : ""}${details.join(", ")})`;
-				lines.push(`  ${chalk.gray(displayId + ":")} ${chalk.white(displayTotal.toLocaleString())}${breakdown}`);
-			}
-		}
-	}
-	lines.push("");
-	const totalTests = successes + failures + errors;
-	const formatResultPercentage = (count) => {
-		const percentage = totalTests === 0 ? 0 : count / totalTests * 100;
-		return percentage === 0 || percentage === 100 ? `${percentage.toFixed(0)}%` : `${percentage.toFixed(2)}%`;
-	};
-	const formatResultLine = (count, label, icon, iconColor) => {
-		return `  ${icon ? `${iconColor(icon)} ` : ""}${chalk.white.bold(count.toLocaleString())} ${chalk.white(label)} ${chalk.gray(`(${formatResultPercentage(count)})`)}`;
-	};
-	const errorLabel = errors === 1 ? "error" : "errors";
-	lines.push(chalk.bold("Results:"));
-	lines.push(formatResultLine(successes, "passed", successes > 0 ? "✓" : void 0, chalk.green));
-	lines.push(formatResultLine(failures, "failed", failures > 0 ? "✗" : void 0, chalk.red));
-	lines.push(formatResultLine(errors, errorLabel, errors > 0 ? "✗" : void 0, chalk.red));
-	const durationDisplay = formatDuration(duration);
-	lines.push(chalk.gray(`Duration: ${durationDisplay} (concurrency: ${maxConcurrency})`));
-	lines.push("");
-	return lines;
+	return [
+		getCompletionMessage({
+			completionType: params.isRedteam ? "Red team" : "Eval",
+			evalId: params.evalId,
+			shareableUrl: params.shareableUrl,
+			wasAborted: params.targetErrorStatus != null,
+			writeToDatabase: params.writeToDatabase,
+			activelySharing: params.activelySharing ?? false
+		}),
+		...getAbortSummaryLines(params.targetErrorStatus),
+		...getGuidanceLines({
+			writeToDatabase: params.writeToDatabase,
+			shareableUrl: params.shareableUrl,
+			wantsToShare: params.wantsToShare,
+			activelySharing: params.activelySharing ?? false,
+			hasExplicitDisable: params.hasExplicitDisable,
+			cloudEnabled: params.cloudEnabled
+		}),
+		"",
+		...getTokenUsageLines(params.tokenUsage, params.isRedteam, params.tracker),
+		...getResultsLines(params)
+	];
 }
 //#endregion
 //#region src/commands/retry.ts
@@ -6517,9 +6504,10 @@ function dedupeTestCases(testCases) {
 	return deduped;
 }
 function buildMaxCharsRetryInstructions(rejectedPromptLengths, limit) {
+	const longestRejectedPromptText = rejectedPromptLengths.length > 0 ? `${Math.max(...rejectedPromptLengths)} characters` : "unknown length";
 	return dedent`
     Your previous response included ${rejectedPromptLengths.length} generated prompt${rejectedPromptLengths.length === 1 ? "" : "s"} that exceeded the ${limit ?? "configured"}-character limit.
-    The longest rejected prompt was ${Math.max(...rejectedPromptLengths)} characters.
+    The longest rejected prompt was ${longestRejectedPromptText}.
     Generate replacement prompts only, and keep every user message within the character limit.
   `.trim();
 }
@@ -8359,6 +8347,26 @@ async function doRedteamRun(options) {
 }
 //#endregion
 //#region src/index.ts
+/**
+* Shallow-clone a test case so the caller can swap in resolved ApiProvider
+* instances on `options.provider` / `assert[].provider` without leaking those
+* mutations back to the input. The input may alias the unified config written
+* to the Eval record, and a live SDK client (e.g. Bedrock's BedrockRuntime,
+* Anthropic's client) holds circular references that break drizzle's JSON
+* serialization on `evalRecord.save()`. Fixes #8687.
+*
+* Detaches only `options` and `assert[]`. Other reference fields (`provider`,
+* `vars`, `metadata`, `providerOutput`) remain aliased — callers must reassign
+* those by reference rather than mutating in place. `assert-set` children are
+* not deep-cloned because the resolve loop skips `assert-set`; if that ever
+* changes, extend this helper.
+*/
+function cloneTestForResolve(test) {
+	const cloned = { ...test };
+	if (test.options) cloned.options = { ...test.options };
+	if (test.assert) cloned.assert = test.assert.map((assertion) => ({ ...assertion }));
+	return cloned;
+}
 async function evaluate(testSuite, options = {}) {
 	if (testSuite.writeLatestResults) await runDbMigrations();
 	const loadedProviders = await loadApiProviders(testSuite.providers, { env: testSuite.env });
@@ -8378,22 +8386,24 @@ async function evaluate(testSuite, options = {}) {
 		nunjucksFilters: await readFilters(testSuite.nunjucksFilters || {}),
 		prompts: await processPrompts(testSuite.prompts)
 	};
-	if (typeof constructedTestSuite.defaultTest === "object") {
-		if (constructedTestSuite.defaultTest?.provider && !isApiProvider(constructedTestSuite.defaultTest.provider)) constructedTestSuite.defaultTest.provider = await resolveProvider(constructedTestSuite.defaultTest.provider, providerMap, {
+	if (typeof constructedTestSuite.defaultTest === "object" && constructedTestSuite.defaultTest) {
+		constructedTestSuite.defaultTest = cloneTestForResolve(constructedTestSuite.defaultTest);
+		if (constructedTestSuite.defaultTest.provider && !isApiProvider(constructedTestSuite.defaultTest.provider)) constructedTestSuite.defaultTest.provider = await resolveProvider(constructedTestSuite.defaultTest.provider, providerMap, {
 			env: testSuite.env,
 			basePath: state.basePath
 		});
-		if (constructedTestSuite.defaultTest?.options?.provider && !isApiProvider(constructedTestSuite.defaultTest.options.provider)) constructedTestSuite.defaultTest.options.provider = await resolveProvider(constructedTestSuite.defaultTest.options.provider, providerMap, {
+		if (constructedTestSuite.defaultTest.options?.provider && !isApiProvider(constructedTestSuite.defaultTest.options.provider)) constructedTestSuite.defaultTest.options.provider = await resolveProvider(constructedTestSuite.defaultTest.options.provider, providerMap, {
 			env: testSuite.env,
 			basePath: state.basePath
 		});
 	}
-	for (const test of constructedTestSuite.tests || []) {
+	constructedTestSuite.tests = (constructedTestSuite.tests || []).map(cloneTestForResolve);
+	for (const test of constructedTestSuite.tests) {
 		if (test.options?.provider && !isApiProvider(test.options.provider)) test.options.provider = await resolveProvider(test.options.provider, providerMap, {
 			env: testSuite.env,
 			basePath: state.basePath
 		});
-		if (test.assert) for (const assertion of test.assert) {
+		for (const assertion of test.assert || []) {
 			if (assertion.type === "assert-set" || typeof assertion.provider === "function") continue;
 			if (assertion.provider && !isApiProvider(assertion.provider)) assertion.provider = await resolveProvider(assertion.provider, providerMap, {
 				env: testSuite.env,
@@ -9353,7 +9363,7 @@ async function checkForUpdates() {
 	} catch {
 		return false;
 	}
-	if (semverGt(latestVersion, "0.121.4")) {
+	if (semverGt(latestVersion, "0.121.5")) {
 		const border = "=".repeat(TERMINAL_MAX_WIDTH);
 		logger.info(`\n${border}
 ${chalk.yellow("⚠️")} The current version of promptfoo ${chalk.yellow(VERSION)} is lower than the latest available version ${chalk.green(latestVersion)}.
@@ -10701,6 +10711,38 @@ function normalizeTargetPurposeDiscoveryResult(result) {
 		tools: cleanTools(result.tools)
 	};
 }
+function extractStringField(value) {
+	if (typeof value !== "string") return;
+	return value.trim() || void 0;
+}
+async function getRemoteResponseErrorDetail(response) {
+	const rawText = (await response.text()).trim();
+	const fallback = rawText || response.statusText || "Unknown error";
+	if (!rawText) return fallback;
+	try {
+		const parsed = JSON.parse(rawText);
+		return extractStringField(parsed?.message) ?? extractStringField(parsed?.error) ?? fallback;
+	} catch {
+		return fallback;
+	}
+}
+const REMOTE_ERROR_HINTS = {
+	400: "This usually means your promptfoo client is out of date. Try `npm install -g promptfoo@latest` and rerun.",
+	401: "Check that you are logged in (`promptfoo auth login`) and that your account has access to target discovery.",
+	403: "Check that you are logged in (`promptfoo auth login`) and that your account has access to target discovery.",
+	404: "This usually means your promptfoo client is out of date. Try `npm install -g promptfoo@latest` and rerun.",
+	429: "You are being rate limited. Wait a moment and try again."
+};
+function getRemoteErrorHint(status) {
+	if (REMOTE_ERROR_HINTS[status]) return REMOTE_ERROR_HINTS[status];
+	if (status >= 500) return "The remote generation service may be temporarily unavailable. Retry in a few minutes or contact support if the issue persists.";
+}
+async function buildRemoteErrorFromResponse(response) {
+	const detail = await getRemoteResponseErrorDetail(response);
+	const hint = getRemoteErrorHint(response.status);
+	const base = `Remote server returned HTTP ${response.status}: ${detail}`;
+	return new Error(hint ? `${base}\n${hint}` : base);
+}
 /**
 * Queries Cloud for the purpose-discovery logic, sends each logic to the target,
 * and summarizes the results.
@@ -10750,11 +10792,7 @@ async function doTargetPurposeDiscovery(target, prompt, showProgress = true) {
 				email: getUserEmail()
 			}))
 		});
-		if (!response.ok) {
-			const error = await response.text();
-			logger.error(`${LOG_PREFIX} Error getting the next question from remote server: ${error}`);
-			continue;
-		}
+		if (!response.ok) throw await buildRemoteErrorFromResponse(response);
 		const responseData = await response.json();
 		const data = TargetPurposeDiscoveryTaskResponseSchema.parse(responseData);
 		logger.debug(`${LOG_PREFIX} Received response from remote server: ${JSON.stringify(data, null, 2)}`);
@@ -12743,7 +12781,7 @@ router.get("/", async (_req, res) => {
 				};
 			} catch (error) {
 				logger.debug(`Failed to fetch latest version: ${error}`);
-				latestVersion = versionCache.latestVersion ?? "0.121.4";
+				latestVersion = versionCache.latestVersion ?? "0.121.5";
 			}
 		}
 		const selfHosted = getEnvBool("PROMPTFOO_SELF_HOSTED");
@@ -12752,7 +12790,7 @@ router.get("/", async (_req, res) => {
 			selfHosted,
 			isNpx
 		});
-		const resolvedLatestVersion = latestVersion ?? "0.121.4";
+		const resolvedLatestVersion = latestVersion ?? "0.121.5";
 		const response = {
 			currentVersion: VERSION,
 			latestVersion: resolvedLatestVersion,
@@ -15174,7 +15212,11 @@ const EXAMPLE_ALIASES = {
 	"google-adk-example": "integration-google-adk",
 	"google-sheets": "integration-google-sheets",
 	"gpt-4o-temperature-comparison": "compare-gpt-temperature",
-	"gpt-4o-vs-4o-mini": "compare-gpt-4o-vs-4o-mini",
+	"compare-gpt-4o-vs-4o-mini": "compare-gpt-model-tiers",
+	"compare-gpt-5-vs-gpt-5-mini-mmlu": "compare-gpt-model-tiers-mmlu-pro",
+	"compare-gpt-mmlu-pro": "compare-gpt-model-tiers-mmlu-pro",
+	"gpt-4o-vs-4o-mini": "compare-gpt-model-tiers",
+	"gpt-model-tiers": "compare-gpt-model-tiers",
 	"gpt-vs-claude-vs-gemini": "compare-gpt-vs-claude-vs-gemini",
 	"grok-4-political-bias": "redteam-grok-4-political-bias",
 	groq: "provider-groq",
@@ -15225,7 +15267,9 @@ const EXAMPLE_ALIASES = {
 	"node-package-typescript": "config-node-package-typescript",
 	nscale: "provider-nscale",
 	"nunjucks-custom-filters": "config-nunjucks-custom-filters",
-	"openai-gpt-5-vs-gpt-5-mini-mmlu": "compare-gpt-5-vs-gpt-5-mini-mmlu",
+	"gpt-mmlu-pro": "compare-gpt-model-tiers-mmlu-pro",
+	"gpt-model-tiers-mmlu-pro": "compare-gpt-model-tiers-mmlu-pro",
+	"openai-gpt-5-vs-gpt-5-mini-mmlu": "compare-gpt-model-tiers-mmlu-pro",
 	"openai-model-comparison": "compare-openai-models",
 	openclaw: "provider-openclaw",
 	"opencode-sdk": "provider-opencode-sdk/basic",
@@ -15597,8 +15641,8 @@ function shouldUseInkList() {
 async function runInkList(options) {
 	const [React, { renderInteractive }, { ListApp }] = await Promise.all([
 		import("react"),
-		import("./render-CgVDrJmM.js"),
-		import("./ListApp-DQkFNqE9.js")
+		import("./render-nj-UaPdn.js"),
+		import("./ListApp-BRUsT43Y.js")
 	]);
 	let result = { cancelled: false };
 	let resolveResult;
@@ -17906,8 +17950,8 @@ function registerRunEvaluationTool(server) {
 					}
 					filteredTestSuite.tests = filteredTests;
 				}
-				const { evaluate } = await import("./evaluator-CVessDWe.js");
-				const evalRecord = await (await import("./eval-CzJFfFO9.js")).default.create(config, filteredTestSuite.prompts, { id: `mcp-eval-${Date.now()}` });
+				const { evaluate } = await import("./evaluator-DgLKaZk8.js");
+				const evalRecord = await (await import("./eval-BQPLBJbw.js")).default.create(config, filteredTestSuite.prompts, { id: `mcp-eval-${Date.now()}` });
 				logger.debug(`Running filtered eval with ${filteredTestSuite.tests?.length || 0} test cases, ${filteredTestSuite.prompts.length} prompts, ${filteredTestSuite.providers.length} providers`);
 				const startTime = Date.now();
 				const result = await evaluate(filteredTestSuite, evalRecord, {