promptfoo 0.120.27 → 0.121.2

package/dist/src/{evaluator-8aGyV12L.js → evaluator-BhoWwp5b.js}

@@ -1,28 +1,28 @@
 #!/usr/bin/env node
-import { C as getEnvString, E as isCI, O as cliState_default, S as getEnvInt, T as getMaxEvalTimeMs, _ as summarizeEvaluateResultForLogging, b as getEnvBool, f as extractJsonObjects, g as safeJsonStringify, o as logger_default, p as getAjv, w as getEvalTimeoutMs } from "./logger-LSBxlt7a.js";
-import { F as VERSION, I as FILE_METADATA_KEY, T as DEFAULT_MAX_CONCURRENCY, g as isPromptfooSampleTarget, l as sleep, r as fetchWithRetries, y as parseChatPrompt } from "./fetch-CXZI9RRr.js";
-import { t as invariant } from "./invariant-DT20jrBd.js";
-import { r as telemetry_default } from "./telemetry--WAdAfVi.js";
-import { d as isGradingResult, p as isApiProvider, s as ResultFailureReason } from "./types-CXQduE9o.js";
-import { c as promptYesNo } from "./server-BJ4m4f1D.js";
-import { A as renderPrompt, E as isBasicRefusal, F as TokenUsageTracker, G as VertexChatProvider, I as createRateLimitRegistry, K as AIStudioChatProvider, L as createProviderRateLimitOptions, M as isPackagePath, N as loadFromPackage, P as redteamProviderManager, j as runExtensionHook, k as collectFileMetadata, u as GoogleLiveProvider, v as checkExfilTracking, w as getSessionId } from "./providers-Iil64vk9.js";
-import { o as getCache } from "./cache-jsiwsAJv.js";
-import { n as isNonTransientHttpStatus } from "./errors-DnGCbnx8.js";
-import { i as isJavascriptFile } from "./fileExtensions-BpuMmaFL.js";
-import { E as parseFileUrl, I as isAnthropicProvider, L as isGoogleProvider, R as isOpenAiProvider, T as loadFunction, g as maybeLoadToolsFromExternalFile, w as getNunjucksEngine, z as isProviderAllowed } from "./util-_h4pVqrz.js";
-import { r as runPython } from "./pythonUtils-CcT5LH1M.js";
-import { n as transform, r as getProcessShim, t as TransformInputType } from "./transform-BEgStbHK.js";
-import { $ as matchesSearchRubric, B as getAndCheckProvider, G as matchesContextFaithfulness, H as matchesAnswerRelevance, J as matchesFactuality, K as matchesContextRecall, Q as matchesPiScore, R as callProviderWithContext, U as matchesClassification, V as loadRubricPrompt, W as matchesClosedQa, X as matchesLlmRubric, Y as matchesGEval, Z as matchesModeration, at as DefaultSuggestionsProvider, dt as getFinalTest, et as matchesSelectBest, ft as loadFromJavaScriptFile, it as getDefaultProviders, lt as SUGGEST_PROMPTS_SYSTEM_MESSAGE, mt as resolveContext, n as getGraderById, nt as selectMaxScore, pt as processFileReference, q as matchesContextRelevance, tt as matchesSimilarity, ut as coerceString, z as fail } from "./graders-DTeBrzWp.js";
-import { i as generateIdFromPrompt } from "./utils-DvWMzuMx.js";
-import { t as OpenAiChatCompletionProvider } from "./chat-BcPjZXIp.js";
-import { a as createEmptyTokenUsage, i as createEmptyAssertions, n as accumulateResponseTokenUsage, o as normalizeTokenUsage, r as accumulateTokenUsage, t as accumulateAssertionTokenUsage } from "./tokenUsageUtils-DmZSD2eU.js";
-import { m as validateFunctionCall } from "./transform-DeGlxb0D.js";
-import { l as validateFunctionCall$1 } from "./util-Bm_-UMD_.js";
-import { t as extractAndStoreBinaryData } from "./extractor-CD5yKL-G.js";
-import { n as getTraceStore } from "./store-s3SftUwF.js";
-import { t as providerRegistry } from "./providerRegistry-Cq_JK_CJ.js";
-import { n as runRuby } from "./rubyUtils-DudlFZed.js";
-import { a as getActualPromptWithFallback, r as updateSignalFile } from "./signal-BW33JuId.js";
+import { C as getEnvString, E as isCI, O as state, S as getEnvInt, T as getMaxEvalTimeMs, _ as summarizeEvaluateResultForLogging, b as getEnvBool, f as extractJsonObjects, g as safeJsonStringify, o as logger, p as getAjv, w as getEvalTimeoutMs } from "./logger-KkObSCzq.js";
+import { N as VERSION, P as FILE_METADATA_KEY, g as isPromptfooSampleTarget, l as sleep, r as fetchWithRetries, y as parseChatPrompt } from "./fetch-Dxpd4_sr.js";
+import { t as invariant } from "./invariant-BtWWVVhl.js";
+import { r as telemetry } from "./telemetry-CMrFgtPB.js";
+import { d as isGradingResult, p as isApiProvider, s as ResultFailureReason } from "./types-Cbd8uOMq.js";
+import { c as promptYesNo } from "./server-CqzrVGpF.js";
+import { A as renderPrompt, E as isBasicRefusal, F as TokenUsageTracker, G as VertexChatProvider, I as createRateLimitRegistry, K as AIStudioChatProvider, L as createProviderRateLimitOptions, M as isPackagePath, N as loadFromPackage, P as redteamProviderManager, j as runExtensionHook, k as collectFileMetadata, u as GoogleLiveProvider, v as checkExfilTracking, w as getSessionId } from "./providers-BlqUifFg.js";
+import { o as getCache } from "./cache-BVeDlD87.js";
+import { n as isNonTransientHttpStatus } from "./errors-P6ll7XSJ.js";
+import { i as isJavascriptFile } from "./fileExtensions-Ds-foDzt.js";
+import { E as parseFileUrl, I as isAnthropicProvider, L as isGoogleProvider, R as isOpenAiProvider, T as loadFunction, g as maybeLoadToolsFromExternalFile, w as getNunjucksEngine, z as isProviderAllowed } from "./util-C1CeHl-P.js";
+import { r as runPython } from "./pythonUtils-C3py6GC1.js";
+import { n as transform, r as getProcessShim, t as TransformInputType } from "./transform-DilY9wbS.js";
+import { $ as matchesSearchRubric, B as getAndCheckProvider, G as matchesContextFaithfulness, H as matchesAnswerRelevance, J as matchesFactuality, K as matchesContextRecall, Q as matchesPiScore, R as callProviderWithContext, U as matchesClassification, V as loadRubricPrompt, W as matchesClosedQa, X as matchesLlmRubric, Y as matchesGEval, Z as matchesModeration, at as DefaultSuggestionsProvider, dt as getFinalTest, et as matchesSelectBest, ft as loadFromJavaScriptFile, it as getDefaultProviders, lt as SUGGEST_PROMPTS_SYSTEM_MESSAGE, mt as resolveContext, n as getGraderById, nt as selectMaxScore, pt as processFileReference, q as matchesContextRelevance, tt as matchesSimilarity, ut as coerceString, z as fail } from "./graders-Bj_Odv7c.js";
+import { i as generateIdFromPrompt } from "./utils-f2-Moju7.js";
+import { t as OpenAiChatCompletionProvider } from "./chat-BiKyneZl.js";
+import { a as createEmptyTokenUsage, i as createEmptyAssertions, n as accumulateResponseTokenUsage, o as normalizeTokenUsage, r as accumulateTokenUsage, t as accumulateAssertionTokenUsage } from "./tokenUsageUtils-DflFMjS0.js";
+import { m as validateFunctionCall } from "./transform-D5HsjduX.js";
+import { l as validateFunctionCall$1 } from "./util-CMy69ZgQ.js";
+import { t as extractAndStoreBinaryData } from "./extractor-DReVID0K.js";
+import { n as getTraceStore } from "./store-Dim__MDd.js";
+import { t as providerRegistry } from "./providerRegistry-B0RUOLI_.js";
+import { n as runRuby } from "./rubyUtils-Boc4HZzX.js";
+import { a as getActualPromptWithFallback, r as updateSignalFile } from "./signal-4U3mfRvL.js";
 import chalk from "chalk";
 import fs, { createWriteStream } from "fs";
 import path from "path";
@@ -43,7 +43,6 @@ import { OTLPTraceExporter } from "@opentelemetry/exporter-trace-otlp-http";
 import { resourceFromAttributes } from "@opentelemetry/resources";
 import { BatchSpanProcessor, NodeTracerProvider } from "@opentelemetry/sdk-trace-node";
 import { ATTR_SERVICE_NAME, ATTR_SERVICE_VERSION } from "@opentelemetry/semantic-conventions";
-
 //#region src/external/matchers/conversationRelevancyTemplate.ts
 var ConversationRelevancyTemplate = class {
 	static generateVerdicts(slidingWindow) {
@@ -115,7 +114,6 @@ ${JSON.stringify(irrelevancies, null, 2)}
 JSON:`;
 	}
 };
-
 //#endregion
 //#region src/external/matchers/deepeval.ts
 const nunjucks$1 = getNunjucksEngine(void 0, false, true);
@@ -165,7 +163,6 @@ async function matchesConversationRelevance(messages, threshold, vars, grading,
 		return fail(`Error parsing output: ${err.message}`, resp.tokenUsage);
 	}
 }
-
 //#endregion
 //#region src/external/assertions/deepeval.ts
 const DEFAULT_WINDOW_SIZE = 5;
@@ -220,7 +217,6 @@ const handleConversationRelevance = async ({ assertion, outputString, prompt, pr
 		tokensUsed: tokensUsed.total > 0 ? tokensUsed : void 0
 	};
 };
-
 //#endregion
 //#region src/tracing/evaluatorTracing.ts
 let otlpReceiverStarted = false;
@@ -253,28 +249,28 @@ function isOtlpReceiverStarted() {
 * Start the OTLP receiver if tracing is enabled and it hasn't been started yet
 */
 async function startOtlpReceiverIfNeeded(testSuite) {
-	logger_default.debug(`[EvaluatorTracing] Checking tracing config: ${JSON.stringify(testSuite.tracing)}`);
-	logger_default.debug(`[EvaluatorTracing] testSuite keys: ${Object.keys(testSuite)}`);
-	logger_default.debug(`[EvaluatorTracing] Full testSuite.tracing: ${JSON.stringify(testSuite.tracing, null, 2)}`);
+	logger.debug(`[EvaluatorTracing] Checking tracing config: ${JSON.stringify(testSuite.tracing)}`);
+	logger.debug(`[EvaluatorTracing] testSuite keys: ${Object.keys(testSuite)}`);
+	logger.debug(`[EvaluatorTracing] Full testSuite.tracing: ${JSON.stringify(testSuite.tracing, null, 2)}`);
 	if (testSuite.tracing?.enabled && testSuite.tracing?.otlp?.http?.enabled && !otlpReceiverStarted) {
-		telemetry_default.record("feature_used", { feature: "tracing" });
+		telemetry.record("feature_used", { feature: "tracing" });
 		try {
-			logger_default.debug("[EvaluatorTracing] Tracing configuration detected, starting OTLP receiver");
-			const { startOTLPReceiver } = await import("./otlpReceiver-BmmTiMjA.js");
+			logger.debug("[EvaluatorTracing] Tracing configuration detected, starting OTLP receiver");
+			const { startOTLPReceiver } = await import("./otlpReceiver-D89fR-rC.js");
 			const port = testSuite.tracing.otlp.http.port || 4318;
 			const host = testSuite.tracing.otlp.http.host || "127.0.0.1";
-			logger_default.debug(`[EvaluatorTracing] Starting OTLP receiver on ${host}:${port}`);
+			logger.debug(`[EvaluatorTracing] Starting OTLP receiver on ${host}:${port}`);
 			await startOTLPReceiver(port, host);
 			otlpReceiverStarted = true;
-			logger_default.info(`[EvaluatorTracing] OTLP receiver successfully started on port ${port} for tracing`);
+			logger.info(`[EvaluatorTracing] OTLP receiver successfully started on port ${port} for tracing`);
 		} catch (error) {
-			logger_default.error(`[EvaluatorTracing] Failed to start OTLP receiver: ${error}`);
+			logger.error(`[EvaluatorTracing] Failed to start OTLP receiver: ${error}`);
 		}
-	} else if (otlpReceiverStarted) logger_default.debug("[EvaluatorTracing] OTLP receiver already started, skipping initialization");
+	} else if (otlpReceiverStarted) logger.debug("[EvaluatorTracing] OTLP receiver already started, skipping initialization");
 	else {
-		logger_default.debug("[EvaluatorTracing] Tracing not enabled or OTLP HTTP receiver not configured");
-		logger_default.debug(`[EvaluatorTracing] tracing.enabled: ${testSuite.tracing?.enabled}`);
-		logger_default.debug(`[EvaluatorTracing] tracing.otlp.http.enabled: ${testSuite.tracing?.otlp?.http?.enabled}`);
+		logger.debug("[EvaluatorTracing] Tracing not enabled or OTLP HTTP receiver not configured");
+		logger.debug(`[EvaluatorTracing] tracing.enabled: ${testSuite.tracing?.enabled}`);
+		logger.debug(`[EvaluatorTracing] tracing.otlp.http.enabled: ${testSuite.tracing?.otlp?.http?.enabled}`);
 	}
 }
 /**
@@ -282,13 +278,13 @@ async function startOtlpReceiverIfNeeded(testSuite) {
 */
 async function stopOtlpReceiverIfNeeded() {
 	if (otlpReceiverStarted) try {
-		logger_default.debug("[EvaluatorTracing] Stopping OTLP receiver");
-		const { stopOTLPReceiver } = await import("./otlpReceiver-BmmTiMjA.js");
+		logger.debug("[EvaluatorTracing] Stopping OTLP receiver");
+		const { stopOTLPReceiver } = await import("./otlpReceiver-D89fR-rC.js");
 		await stopOTLPReceiver();
 		otlpReceiverStarted = false;
-		logger_default.info("[EvaluatorTracing] OTLP receiver stopped successfully");
+		logger.info("[EvaluatorTracing] OTLP receiver stopped successfully");
 	} catch (error) {
-		logger_default.error(`[EvaluatorTracing] Failed to stop OTLP receiver: ${error}`);
+		logger.error(`[EvaluatorTracing] Failed to stop OTLP receiver: ${error}`);
 	}
 }
 /**
@@ -304,7 +300,7 @@ function isTracingEnabled(test, testSuite) {
 	const yamlConfigEnabled = testSuite?.tracing?.enabled === true;
 	const envEnabled = getEnvBool("PROMPTFOO_TRACING_ENABLED", false);
 	const result = metadataEnabled || yamlConfigEnabled || envEnabled;
-	logger_default.debug(`[EvaluatorTracing] isTracingEnabled check: metadata=${metadataEnabled}, yamlConfig=${yamlConfigEnabled}, env=${envEnabled}, result=${result}`);
+	logger.debug(`[EvaluatorTracing] isTracingEnabled check: metadata=${metadataEnabled}, yamlConfig=${yamlConfigEnabled}, env=${envEnabled}, result=${result}`);
 	return result;
 }
 /**
@@ -313,25 +309,25 @@ function isTracingEnabled(test, testSuite) {
 async function generateTraceContextIfNeeded(test, evaluateOptions, testIdx, promptIdx, testSuite) {
 	const tracingEnabled = isTracingEnabled(test, testSuite);
 	if (tracingEnabled) {
-		logger_default.debug("[EvaluatorTracing] Tracing enabled for test case");
-		logger_default.debug(`[EvaluatorTracing] Test metadata: ${JSON.stringify(test.metadata)}`);
+		logger.debug("[EvaluatorTracing] Tracing enabled for test case");
+		logger.debug(`[EvaluatorTracing] Test metadata: ${JSON.stringify(test.metadata)}`);
 	}
 	if (!tracingEnabled) return null;
-	logger_default.debug("[EvaluatorTracing] Importing trace store");
-	const { getTraceStore } = await import("./store-B3EDO9Q3.js");
+	logger.debug("[EvaluatorTracing] Importing trace store");
+	const { getTraceStore } = await import("./store-CXS-Q_91.js");
 	const traceStore = getTraceStore();
 	const traceId = generateTraceId();
 	const spanId = generateSpanId();
 	const traceparent = generateTraceparent(traceId, spanId);
-	logger_default.debug(`[EvaluatorTracing] Generated trace context: traceId=${traceId}, spanId=${spanId}`);
+	logger.debug(`[EvaluatorTracing] Generated trace context: traceId=${traceId}, spanId=${spanId}`);
 	let evaluationId = test.metadata?.evaluationId || evaluateOptions?.eventSource;
 	if (!evaluationId) {
-		logger_default.warn("[EvaluatorTracing] No evaluation ID found in test metadata or evaluateOptions, trace will not be linked to evaluation");
+		logger.warn("[EvaluatorTracing] No evaluation ID found in test metadata or evaluateOptions, trace will not be linked to evaluation");
 		evaluationId = `eval-${Date.now()}`;
 	}
 	const testCaseId = test.metadata?.testCaseId || test.id || `${testIdx}-${promptIdx}`;
 	try {
-		logger_default.debug(`[EvaluatorTracing] Creating trace record for traceId=${traceId}`);
+		logger.debug(`[EvaluatorTracing] Creating trace record for traceId=${traceId}`);
 		await traceStore.createTrace({
 			traceId,
 			evaluationId: evaluationId || "",
@@ -342,18 +338,17 @@ async function generateTraceContextIfNeeded(test, evaluateOptions, testIdx, prom
 				vars: test.vars
 			}
 		});
-		logger_default.debug("[EvaluatorTracing] Trace record created successfully");
+		logger.debug("[EvaluatorTracing] Trace record created successfully");
 	} catch (error) {
-		logger_default.error(`[EvaluatorTracing] Failed to create trace: ${error}`);
+		logger.error(`[EvaluatorTracing] Failed to create trace: ${error}`);
 	}
-	logger_default.debug(`[EvaluatorTracing] Trace context ready: ${traceparent} for test case ${testCaseId}`);
+	logger.debug(`[EvaluatorTracing] Trace context ready: ${traceparent} for test case ${testCaseId}`);
 	return {
 		traceparent,
 		evaluationId,
 		testCaseId
 	};
 }
-
 //#endregion
 //#region src/assertions/answerRelevance.ts
 const handleAnswerRelevance = async ({ assertion, output, prompt, test, providerCallContext }) => {
@@ -364,7 +359,6 @@ const handleAnswerRelevance = async ({ assertion, output, prompt, test, provider
 		...await matchesAnswerRelevance(typeof test?.vars?.query === "string" ? test.vars.query : prompt, output, assertion.threshold ?? 0, test.options, providerCallContext)
 	};
 };
-
 //#endregion
 //#region src/assertions/assertionsResult.ts
 const GUARDRAIL_BLOCKED_REASON = "Content failed guardrail safety checks";
@@ -470,7 +464,6 @@ var AssertionsResult = class {
 		return this.result;
 	}
 };
-
 //#endregion
 //#region src/assertions/ngrams.ts
 /**
@@ -486,7 +479,6 @@ function getNGrams(words, n) {
 	for (let i = 0; i <= words.length - n; i++) ngrams.push(words.slice(i, i + n).join(" "));
 	return ngrams;
 }
-
 //#endregion
 //#region src/assertions/bleu.ts
 /**
@@ -582,7 +574,6 @@ function handleBleuScore({ assertion, inverse, outputString, renderedValue }) {
 		assertion
 	};
 }
-
 //#endregion
 //#region src/assertions/classifier.ts
 async function handleClassifier({ assertion, renderedValue, outputString, test, inverse }) {
@@ -597,9 +588,43 @@ async function handleClassifier({ assertion, renderedValue, outputString, test,
 		...classificationResult
 	};
 }
-
 //#endregion
 //#region src/assertions/contains.ts
+function parseCommaSeparatedValues(value) {
+	const results = [];
+	let i = 0;
+	while (i < value.length) {
+		while (i < value.length && /\s/.test(value[i])) i++;
+		if (i >= value.length) break;
+		if (value[i] === ",") {
+			i++;
+			continue;
+		}
+		if (value[i] === "\"") {
+			i++;
+			let field = "";
+			while (i < value.length) if (value[i] === "\\" && i + 1 < value.length && (value[i + 1] === "\"" || value[i + 1] === "\\")) {
+				field += value[i + 1];
+				i += 2;
+			} else if (value[i] === "\"" && i + 1 < value.length && value[i + 1] === "\"") {
+				field += "\"";
+				i += 2;
+			} else if (value[i] === "\"") {
+				i++;
+				break;
+			} else {
+				field += value[i];
+				i++;
+			}
+			results.push(field);
+		} else {
+			const start = i;
+			while (i < value.length && value[i] !== ",") i++;
+			results.push(value.substring(start, i).trim());
+		}
+	}
+	return results;
+}
 const handleContains = ({ assertion, renderedValue, valueFromScript, outputString, inverse }) => {
 	const value = valueFromScript ?? renderedValue;
 	invariant(value, "\"contains\" assertion type must have a string or number value");
@@ -627,7 +652,7 @@ const handleIContains = ({ assertion, renderedValue, valueFromScript, outputStri
 const handleContainsAny = ({ assertion, renderedValue, valueFromScript, outputString, inverse }) => {
 	let value = valueFromScript ?? renderedValue;
 	invariant(value, "\"contains-any\" assertion type must have a value");
-	if (typeof value === "string") value = value.match(/(".*?"|[^,]+)(?=\s*,|\s*$)/g)?.map((v) => v.trim().replace(/^"|"$/g, "")) ?? [];
+	if (typeof value === "string") value = parseCommaSeparatedValues(value);
 	invariant(Array.isArray(value), "\"contains-any\" assertion type must have an array value");
 	const pass = value.some((v) => outputString.includes(String(v))) !== inverse;
 	return {
@@ -640,7 +665,7 @@ const handleContainsAny = ({ assertion, renderedValue, valueFromScript, outputSt
 const handleIContainsAny = ({ assertion, renderedValue, valueFromScript, outputString, inverse }) => {
 	let value = valueFromScript ?? renderedValue;
 	invariant(value, "\"icontains-any\" assertion type must have a value");
-	if (typeof value === "string") value = value.split(",").map((v) => v.trim());
+	if (typeof value === "string") value = parseCommaSeparatedValues(value);
 	invariant(Array.isArray(value), "\"icontains-any\" assertion type must have an array value");
 	const pass = value.some((v) => outputString.toLowerCase().includes(String(v).toLowerCase())) !== inverse;
 	return {
@@ -653,7 +678,7 @@ const handleIContainsAny = ({ assertion, renderedValue, valueFromScript, outputS
 const handleContainsAll = ({ assertion, renderedValue, valueFromScript, outputString, inverse }) => {
 	let value = valueFromScript ?? renderedValue;
 	invariant(value, "\"contains-all\" assertion type must have a value");
-	if (typeof value === "string") value = value.split(",").map((v) => v.trim());
+	if (typeof value === "string") value = parseCommaSeparatedValues(value);
 	invariant(Array.isArray(value), "\"contains-all\" assertion type must have an array value");
 	const missingStrings = value.filter((v) => !outputString.includes(String(v)));
 	const pass = missingStrings.length === 0 !== inverse;
@@ -667,7 +692,7 @@ const handleContainsAll = ({ assertion, renderedValue, valueFromScript, outputSt
 const handleIContainsAll = ({ assertion, renderedValue, valueFromScript, outputString, inverse }) => {
 	let value = valueFromScript ?? renderedValue;
 	invariant(value, "\"icontains-all\" assertion type must have a value");
-	if (typeof value === "string") value = value.split(",").map((v) => v.trim());
+	if (typeof value === "string") value = parseCommaSeparatedValues(value);
 	invariant(Array.isArray(value), "\"icontains-all\" assertion type must have an array value");
 	const missingStrings = value.filter((v) => !outputString.toLowerCase().includes(String(v).toLowerCase()));
 	const pass = missingStrings.length === 0 !== inverse;
@@ -678,7 +703,6 @@ const handleIContainsAll = ({ assertion, renderedValue, valueFromScript, outputS
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/contextFaithfulness.ts
 /**
@@ -702,7 +726,6 @@ async function handleContextFaithfulness({ assertion, test, output, prompt, prov
 		metadata: { context }
 	};
 }
-
 //#endregion
 //#region src/assertions/contextRecall.ts
 /**
@@ -729,7 +752,6 @@ const handleContextRecall = async ({ assertion, renderedValue, prompt, test, out
 		}
 	};
 };
-
 //#endregion
 //#region src/assertions/contextRelevance.ts
 /**
@@ -756,7 +778,6 @@ const handleContextRelevance = async ({ assertion, test, output, prompt, provide
 		}
 	};
 };
-
 //#endregion
 //#region src/assertions/cost.ts
 const handleCost = ({ cost, assertion }) => {
@@ -770,7 +791,6 @@ const handleCost = ({ cost, assertion }) => {
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/equals.ts
 const handleEquals = async ({ assertion, renderedValue, outputString, inverse }) => {
@@ -790,7 +810,6 @@ const handleEquals = async ({ assertion, renderedValue, outputString, inverse })
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/factuality.ts
 const handleFactuality = async ({ assertion, renderedValue, outputString, test, prompt, providerCallContext }) => {
@@ -801,7 +820,6 @@ const handleFactuality = async ({ assertion, renderedValue, outputString, test,
 		...await matchesFactuality(prompt, renderedValue, outputString, test.options, test.vars, providerCallContext)
 	};
 };
-
 //#endregion
 //#region src/assertions/finishReason.ts
 function handleFinishReason({ assertion, renderedValue, providerResponse }) {
@@ -821,7 +839,6 @@ function handleFinishReason({ assertion, renderedValue, providerResponse }) {
 		assertion
 	};
 }
-
 //#endregion
 //#region src/assertions/functionToolCall.ts
 const handleIsValidFunctionCall = ({ assertion, output, provider, test }) => {
@@ -844,7 +861,6 @@ const handleIsValidFunctionCall = ({ assertion, output, provider, test }) => {
 		};
 	}
 };
-
 //#endregion
 //#region src/assertions/geval.ts
 const handleGEval = async ({ assertion, renderedValue, prompt, outputString, test, providerCallContext }) => {
@@ -870,7 +886,6 @@ const handleGEval = async ({ assertion, renderedValue, prompt, outputString, tes
 		...await matchesGEval(renderedValue, prompt || "", outputString, threshold, test.options, providerCallContext)
 	};
 };
-
 //#endregion
 //#region src/assertions/gleu.ts
 /**
@@ -969,7 +984,6 @@ function handleGleuScore({ assertion, inverse, outputString, renderedValue }) {
 		assertion
 	};
 }
-
 //#endregion
 //#region src/assertions/guardrails.ts
 const handleGuardrails = async ({ assertion, inverse, providerResponse }) => {
@@ -978,7 +992,7 @@ const handleGuardrails = async ({ assertion, inverse, providerResponse }) => {
 		flaggedInput: false,
 		flaggedOutput: false
 	};
-	logger_default.debug(`Guardrails from target response: ${JSON.stringify(guardrails, null, 2)}`);
+	logger.debug(`Guardrails from target response: ${JSON.stringify(guardrails, null, 2)}`);
 	if (guardrails) if (guardrails.flagged) {
 		let reason = "Content failed safety checks";
 		if (guardrails?.flaggedInput) reason = "Prompt failed safety checks";
@@ -1007,7 +1021,6 @@ const handleGuardrails = async ({ assertion, inverse, providerResponse }) => {
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/html.ts
 const HTML_PATTERNS = {
@@ -1216,7 +1229,6 @@ const handleIsHtml = ({ assertion, outputString, inverse }) => {
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/javascript.ts
 /**
@@ -1357,7 +1369,6 @@ ${renderedValue}`,
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/json.ts
 function handleIsJson({ outputString, renderedValue, inverse, valueFromScript, assertion }) {
@@ -1369,7 +1380,7 @@ function handleIsJson({ outputString, renderedValue, inverse, valueFromScript, a
 	} catch {
 		pass = inverse;
 	}
-	if (pass && renderedValue) {
+	if (parsedJson !== void 0 && renderedValue) {
 		let validate;
 		if (typeof renderedValue === "string") if (renderedValue.startsWith("file://")) {
 			const schema = valueFromScript;
@@ -1381,11 +1392,12 @@ function handleIsJson({ outputString, renderedValue, inverse, valueFromScript, a
 		}
 		else if (typeof renderedValue === "object") validate = getAjv().compile(renderedValue);
 		else throw new Error("is-json assertion must have a string or object value");
-		pass = validate(parsedJson);
+		const valid = validate(parsedJson);
+		pass = inverse ? !valid : valid;
 		if (!pass) return {
 			pass,
 			score: 0,
-			reason: `JSON does not conform to the provided schema. Errors: ${getAjv().errorsText(validate.errors)}`,
+			reason: inverse ? "Output is JSON that conforms to the provided schema" : `JSON does not conform to the provided schema. Errors: ${getAjv().errorsText(validate.errors)}`,
 			assertion
 		};
 	}
@@ -1412,9 +1424,12 @@ function handleContainsJson({ assertion, renderedValue, outputString, inverse, v
 		}
 		else if (typeof renderedValue === "object") validate = getAjv().compile(renderedValue);
 		else throw new Error("contains-json assertion must have a string or object value");
-		pass = validate(jsonObject);
-		if (pass) break;
-		else errorMessage = `JSON does not conform to the provided schema. Errors: ${getAjv().errorsText(validate.errors)}`;
+		const valid = validate(jsonObject);
+		pass = inverse ? !valid : valid;
+		if (valid) {
+			if (inverse) errorMessage = "Output contains JSON conforming to the provided schema";
+			break;
+		} else errorMessage = `JSON does not conform to the provided schema. Errors: ${getAjv().errorsText(validate.errors)}`;
 	}
 	return {
 		pass,
@@ -1423,7 +1438,6 @@ function handleContainsJson({ assertion, renderedValue, outputString, inverse, v
 		assertion
 	};
 }
-
 //#endregion
 //#region src/assertions/latency.ts
 const handleLatency = ({ assertion, latencyMs }) => {
@@ -1437,7 +1451,6 @@ const handleLatency = ({ assertion, latencyMs }) => {
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/levenshtein.ts
 function handleLevenshtein({ assertion, renderedValue, outputString }) {
@@ -1452,7 +1465,6 @@ function handleLevenshtein({ assertion, renderedValue, outputString }) {
 		assertion
 	};
 }
-
 //#endregion
 //#region src/assertions/llmRubric.ts
 const handleLlmRubric = ({ assertion, renderedValue, outputString, test, providerCallContext }) => {
@@ -1461,7 +1473,6 @@ const handleLlmRubric = ({ assertion, renderedValue, outputString, test, provide
 	assertion.value = assertion.value || test.options?.rubricPrompt;
 	return matchesLlmRubric(renderedValue || "", outputString, test.options, test.vars, assertion, void 0, providerCallContext);
 };
-
 //#endregion
 //#region src/assertions/modelGradedClosedQa.ts
 const handleModelGradedClosedQa = async ({ assertion, renderedValue, outputString, test, prompt, providerCallContext }) => {
@@ -1472,7 +1483,6 @@ const handleModelGradedClosedQa = async ({ assertion, renderedValue, outputStrin
 		...await matchesClosedQa(prompt, renderedValue, outputString, test.options, test.vars, providerCallContext)
 	};
 };
-
 //#endregion
 //#region src/assertions/moderation.ts
 const handleModeration = async ({ assertion, test, outputString, providerResponse, prompt }) => {
@@ -1495,7 +1505,6 @@ const handleModeration = async ({ assertion, test, outputString, providerRespons
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/openai.ts
 const handleIsValidOpenAiToolsCall = async ({ assertion, output, provider, test }) => {
@@ -1556,7 +1565,6 @@ const handleIsValidOpenAiToolsCall = async ({ assertion, output, provider, test
 		};
 	}
 };
-
 //#endregion
 //#region src/assertions/perplexity.ts
 function handlePerplexity({ logProbs, assertion }) {
@@ -1583,7 +1591,6 @@ function handlePerplexityScore({ logProbs, assertion }) {
 		assertion
 	};
 }
-
 //#endregion
 //#region src/assertions/pi.ts
 const handlePiScorer = async ({ assertion, prompt, renderedValue, outputString }) => {
@@ -1591,7 +1598,6 @@ const handlePiScorer = async ({ assertion, prompt, renderedValue, outputString }
 	invariant(typeof prompt === "string", "\"pi\" assertion must have a prompt that is a string");
 	return matchesPiScore(renderedValue, prompt, outputString, assertion);
 };
-
 //#endregion
 //#region src/python/wrapper.ts
 /**
@@ -1607,17 +1613,16 @@ async function runPythonCode(code, method, args) {
 		fs.writeFileSync(tempFilePath, code);
 		return await runPython(tempFilePath, method, args);
 	} catch (error) {
-		logger_default.error(`Error executing Python code: ${error}`);
+		logger.error(`Error executing Python code: ${error}`);
 		throw error;
 	} finally {
 		try {
 			fs.unlinkSync(tempFilePath);
 		} catch (error) {
-			logger_default.error(`Error removing temporary file: ${error}`);
+			logger.error(`Error removing temporary file: ${error}`);
 		}
 	}
 }
-
 //#endregion
 //#region src/util/caseMapping.ts
 /**
@@ -1641,7 +1646,6 @@ function mapSnakeCaseToCamelCase(obj) {
 	});
 	return result;
 }
-
 //#endregion
 //#region src/assertions/python.ts
 const handlePython = async ({ assertion, renderedValue, valueFromScript, assertionValueContext, output }) => {
@@ -1711,7 +1715,6 @@ ${isMultiline ? renderedValue.split("\n").map((line) => `${indentStyle}${line}`)
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/redteam.ts
 /**
@@ -1792,7 +1795,7 @@ const handleRedteam = async ({ assertion, baseType, test, prompt, outputString,
 		const { hasAnyErrors, allTurnsHaveErrors } = analyzeGraderErrors(redteamHistory);
 		if (test.metadata?.strategyId && hasAnyErrors && !allTurnsHaveErrors) {
 			const errorMessage = error instanceof Error ? error.message : String(error);
-			logger_default.warn("[Redteam] Grading failed for iterative test with some prior grader errors", {
+			logger.warn("[Redteam] Grading failed for iterative test with some prior grader errors", {
 				error: errorMessage,
 				strategyId: test.metadata.strategyId,
 				pluginId: test.metadata.pluginId
@@ -1812,7 +1815,6 @@ const handleRedteam = async ({ assertion, baseType, test, prompt, outputString,
 		throw error;
 	}
 };
-
 //#endregion
 //#region src/assertions/refusal.ts
 function handleIsRefusal(params) {
@@ -1840,7 +1842,6 @@ function handleIsRefusal(params) {
 		assertion
 	};
 }
-
 //#endregion
 //#region src/assertions/regex.ts
 const handleRegex = ({ assertion, renderedValue, outputString, inverse }) => {
@@ -1865,7 +1866,6 @@ const handleRegex = ({ assertion, renderedValue, outputString, inverse }) => {
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/rouge.ts
 function handleRougeScore({ baseType, assertion, renderedValue, outputString, inverse }) {
@@ -1881,7 +1881,6 @@ function handleRougeScore({ baseType, assertion, renderedValue, outputString, in
 		assertion
 	};
 }
-
 //#endregion
 //#region src/ruby/wrapper.ts
 /**
@@ -1897,17 +1896,16 @@ async function runRubyCode(code, method, args) {
 		fs.writeFileSync(tempFilePath, code);
 		return await runRuby(tempFilePath, method, args);
 	} catch (error) {
-		logger_default.error(`Error executing Ruby code: ${error}`);
+		logger.error(`Error executing Ruby code: ${error}`);
 		throw error;
 	} finally {
 		try {
 			fs.unlinkSync(tempFilePath);
 		} catch (error) {
-			logger_default.error(`Error removing temporary file: ${error}`);
+			logger.error(`Error removing temporary file: ${error}`);
 		}
 	}
 }
-
 //#endregion
 //#region src/assertions/ruby.ts
 const handleRuby = async ({ assertion, renderedValue, valueFromScript, assertionValueContext, output }) => {
@@ -1978,7 +1976,6 @@ end
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/searchRubric.ts
 async function handleSearchRubric({ assertion, baseType: _baseType, inverse, provider, providerCallContext, renderedValue, test, providerResponse }) {
@@ -1990,7 +1987,6 @@ async function handleSearchRubric({ assertion, baseType: _baseType, inverse, pro
 	}
 	return result;
 }
-
 //#endregion
 //#region src/assertions/similar.ts
 const handleSimilar = async ({ assertion, renderedValue, outputString, inverse, test }) => {
@@ -2033,7 +2029,6 @@ const handleSimilar = async ({ assertion, renderedValue, outputString, inverse,
 		...await matchesSimilarity(renderedValue, outputString, threshold, inverse, test.options, metric)
 	};
 };
-
 //#endregion
 //#region src/assertions/sql.ts
 const handleIsSql = async ({ assertion, renderedValue, outputString, inverse }) => {
@@ -2125,7 +2120,6 @@ const handleContainsSql = async (assertionParams) => {
 	}
 	return handleIsSql(assertionParams);
 };
-
 //#endregion
 //#region src/assertions/startsWith.ts
 const handleStartsWith = ({ assertion, renderedValue, outputString, inverse }) => {
@@ -2139,7 +2133,6 @@ const handleStartsWith = ({ assertion, renderedValue, outputString, inverse }) =
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/toolCallF1.ts
 /**
@@ -2268,7 +2261,6 @@ const handleToolCallF1 = ({ assertion, output, renderedValue, inverse }) => {
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/traceUtils.ts
 /**
@@ -2286,7 +2278,6 @@ function matchesPattern(spanName, pattern) {
 	const regexPattern = pattern.replace(/[.+^${}()|[\]\\]/g, "\\$&").replace(/\*/g, ".*").replace(/\?/g, ".");
 	return new RegExp(`^${regexPattern}$`, "i").test(spanName);
 }
-
 //#endregion
 //#region src/assertions/traceErrorSpans.ts
 function isErrorSpan(span) {
@@ -2364,7 +2355,6 @@ const handleTraceErrorSpans = ({ assertion, assertionValueContext }) => {
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/traceSpanCount.ts
 const handleTraceSpanCount = ({ assertion, assertionValueContext }) => {
@@ -2399,7 +2389,6 @@ const handleTraceSpanCount = ({ assertion, assertionValueContext }) => {
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/traceSpanDuration.ts
 function calculatePercentile(durations, percentile) {
@@ -2457,7 +2446,6 @@ const handleTraceSpanDuration = ({ assertion, assertionValueContext }) => {
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/webhook.ts
 async function handleWebhook({ assertion, renderedValue, test, prompt, output, inverse }) {
@@ -2494,7 +2482,6 @@ async function handleWebhook({ assertion, renderedValue, test, prompt, output, i
 		};
 	}
 }
-
 //#endregion
 //#region src/assertions/wordCount.ts
 /**
@@ -2557,7 +2544,6 @@ const handleWordCount = ({ assertion, renderedValue, valueFromScript, outputStri
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/xml.ts
 function validateXml(xmlString, requiredElements) {
@@ -2632,7 +2618,6 @@ const handleIsXml = ({ assertion, renderedValue, outputString, inverse, baseType
 		assertion
 	};
 };
-
 //#endregion
 //#region src/assertions/index.ts
 const ASSERTIONS_MAX_CONCURRENCY = getEnvInt("PROMPTFOO_ASSERTIONS_MAX_CONCURRENCY", 3);
@@ -2686,7 +2671,7 @@ const ASSERTION_HANDLERS = {
 	"llm-rubric": handleLlmRubric,
 	meteor: async (params) => {
 		try {
-			const { handleMeteorAssertion } = await import("./meteor-BYykdXrV.js");
+			const { handleMeteorAssertion } = await import("./meteor-44VjEACX.js");
 			return handleMeteorAssertion(params);
 		} catch (error) {
 			if (error instanceof Error && (error.message.includes("Cannot find module") || error.message.includes("natural\" package is required"))) return {
@@ -2732,10 +2717,10 @@ function renderMetricName(metric, vars) {
 	if (!metric) return metric;
 	try {
 		const rendered = nunjucks.renderString(metric, vars);
-		if (rendered === "" && metric !== "") logger_default.debug(`Metric template "${metric}" rendered to empty string`);
+		if (rendered === "" && metric !== "") logger.debug(`Metric template "${metric}" rendered to empty string`);
 		return rendered;
 	} catch (error) {
-		logger_default.warn(`Failed to render metric template "${metric}": ${error instanceof Error ? error.message : error}`);
+		logger.warn(`Failed to render metric template "${metric}": ${error instanceof Error ? error.message : error}`);
 		return metric;
 	}
 }
@@ -2786,12 +2771,12 @@ async function runAssertion({ prompt, provider, assertion, test, vars, latencyMs
 			spans: traceData.spans || []
 		};
 	} catch (error) {
-		logger_default.debug(`Failed to fetch trace data for assertion: ${error}`);
+		logger.debug(`Failed to fetch trace data for assertion: ${error}`);
 	}
 	let renderedValue = assertion.value;
 	let valueFromScript;
 	if (typeof renderedValue === "string") if (renderedValue.startsWith("file://")) {
-		const basePath = cliState_default.basePath || "";
+		const basePath = state.basePath || "";
 		const fileRef = renderedValue.slice(7);
 		let filePath = fileRef;
 		let functionName;
@@ -2803,10 +2788,10 @@ async function runAssertion({ prompt, provider, assertion, test, vars, latencyMs
 		filePath = path.resolve(basePath, filePath);
 		if (isJavascriptFile(filePath)) {
 			valueFromScript = await loadFromJavaScriptFile(filePath, functionName, [output, context]);
-			logger_default.debug(`Javascript script ${filePath} output: ${valueFromScript}`);
+			logger.debug(`Javascript script ${filePath} output: ${valueFromScript}`);
 		} else if (filePath.endsWith(".py")) try {
 			valueFromScript = await runPython(filePath, functionName || "get_assert", [output, context]);
-			logger_default.debug(`Python script ${filePath} output: ${valueFromScript}`);
+			logger.debug(`Python script ${filePath} output: ${valueFromScript}`);
 		} catch (error) {
 			return {
 				pass: false,
@@ -2816,9 +2801,9 @@ async function runAssertion({ prompt, provider, assertion, test, vars, latencyMs
 			};
 		}
 		else if (filePath.endsWith(".rb")) try {
-			const { runRuby } = await import("./rubyUtils-BUbcND2f.js");
+			const { runRuby } = await import("./rubyUtils-BUHu6PhO.js");
 			valueFromScript = await runRuby(filePath, functionName || "get_assert", [output, context]);
-			logger_default.debug(`Ruby script ${filePath} output: ${valueFromScript}`);
+			logger.debug(`Ruby script ${filePath} output: ${valueFromScript}`);
 		} catch (error) {
 			return {
 				pass: false,
@@ -2829,7 +2814,7 @@ async function runAssertion({ prompt, provider, assertion, test, vars, latencyMs
 		}
 		else renderedValue = processFileReference(renderedValue);
 	} else if (isPackagePath(renderedValue)) {
-		const basePath = cliState_default.basePath || "";
+		const basePath = state.basePath || "";
 		const requiredModule = await loadFromPackage(renderedValue, basePath);
 		if (typeof requiredModule !== "function") throw new Error(`Assertion malformed: ${renderedValue} must be a function. Received: ${typeof requiredModule}`);
 		valueFromScript = await Promise.resolve(requiredModule(output, context));
@@ -2990,7 +2975,6 @@ var assertions_default = {
 	matchesModeration,
 	matchesConversationRelevance
 };
-
 //#endregion
 //#region src/util/promptMatching.ts
 /**
@@ -3028,7 +3012,6 @@ function isPromptAllowed(prompt, allowedPrompts) {
 	if (allowedPrompts.length === 0) return false;
 	return allowedPrompts.some((ref) => doesPromptRefMatch(ref, prompt));
 }
-
 //#endregion
 //#region src/progress/ciProgressReporter.ts
 var CIProgressReporter = class {
@@ -3050,7 +3033,7 @@ var CIProgressReporter = class {
 	}
 	start() {
 		if (this.intervalId) clearInterval(this.intervalId);
-		logger_default.info(`[Evaluation] Starting ${this.totalTests} test cases...`);
+		logger.info(`[Evaluation] Starting ${this.totalTests} test cases...`);
 		this.intervalId = setInterval(() => {
 			this.logPeriodicUpdate();
 		}, this.updateIntervalMs);
@@ -3081,14 +3064,14 @@ var CIProgressReporter = class {
 			this.intervalId = null;
 		}
 		const elapsed = this.formatElapsedTime(Date.now() - this.startTime);
-		logger_default.info(`[Evaluation] ✓ Complete! ${this.completedTests}/${this.totalTests} tests in ${elapsed}`);
+		logger.info(`[Evaluation] ✓ Complete! ${this.completedTests}/${this.totalTests} tests in ${elapsed}`);
 		if (process.env.GITHUB_ACTIONS) console.log(`::notice::Evaluation completed: ${this.completedTests}/${this.totalTests} tests in ${elapsed}`);
 	}
 	error(message) {
 		const now = Date.now();
 		if (now - this.lastErrorTime < this.ERROR_THROTTLE_MS) return;
 		this.lastErrorTime = now;
-		logger_default.error(`[Evaluation Error] ${message}`);
+		logger.error(`[Evaluation Error] ${message}`);
 		if (process.env.GITHUB_ACTIONS) {
 			const escapedMessage = message.replace(/\r?\n/g, " ").replace(/::/g, " ");
 			console.log(`::error::${escapedMessage}`);
@@ -3107,12 +3090,12 @@ var CIProgressReporter = class {
 			else etaDisplay = `${Math.round(eta)} minute${Math.round(eta) !== 1 ? "s" : ""}`;
 		}
 		const percentage = Math.floor(this.completedTests / this.totalTests * 100);
-		logger_default.info(`[CI Progress] Evaluation running for ${this.formatElapsedTime(elapsed)} - Completed ${this.completedTests}/${this.totalTests} tests (${percentage}%)`);
-		logger_default.info(`[CI Progress] Rate: ~${Math.round(rate)} tests/minute, ETA: ${etaDisplay}`);
+		logger.info(`[CI Progress] Evaluation running for ${this.formatElapsedTime(elapsed)} - Completed ${this.completedTests}/${this.totalTests} tests (${percentage}%)`);
+		logger.info(`[CI Progress] Rate: ~${Math.round(rate)} tests/minute, ETA: ${etaDisplay}`);
 	}
 	logMilestone(percentage) {
 		const elapsed = this.formatElapsedTime(Date.now() - this.startTime);
-		logger_default.info(`[Evaluation] ✓ ${percentage}% complete (${this.completedTests}/${this.totalTests}) - ${elapsed} elapsed`);
+		logger.info(`[Evaluation] ✓ ${percentage}% complete (${this.completedTests}/${this.totalTests}) - ${elapsed} elapsed`);
 		if (process.env.GITHUB_ACTIONS) console.log(`::notice::Evaluation ${percentage}% complete`);
 	}
 	formatElapsedTime(ms) {
@@ -3123,7 +3106,6 @@ var CIProgressReporter = class {
 		return `${minutes}m ${remainingSeconds}s`;
 	}
 };
-
 //#endregion
 //#region src/providers/azure/warnings.ts
 /**
@@ -3137,13 +3119,12 @@ function maybeEmitAzureOpenAiWarning(testSuite, tests) {
 		const modelGradedAsserts = tests.flatMap((t) => (t.assert || []).filter((a) => a.type !== "assert-set" && MODEL_GRADED_ASSERTION_TYPES.has(a.type) && !a.provider && !t.options?.provider));
 		if (modelGradedAsserts.length > 0) {
 			const assertTypes = Array.from(new Set(modelGradedAsserts.map((a) => a.type))).join(", ");
-			logger_default.warn(chalk.yellow(`You are using model-graded assertions of types ${chalk.bold(assertTypes)} while testing an Azure provider. You may need to override these to use your Azure deployment. To learn more, see ${chalk.bold(`https://promptfoo.dev/docs/providers/azure/#model-graded-tests`)}`));
+			logger.warn(chalk.yellow(`You are using model-graded assertions of types ${chalk.bold(assertTypes)} while testing an Azure provider. You may need to override these to use your Azure deployment. To learn more, see ${chalk.bold(`https://promptfoo.dev/docs/providers/azure/#model-graded-tests`)}`));
 			return true;
 		}
 	}
 	return false;
 }
-
 //#endregion
 //#region src/suggestions.ts
 async function generatePrompts(prompt, _num) {
@@ -3174,7 +3155,6 @@ async function generatePrompts(prompt, _num) {
 		};
 	}
 }
-
 //#endregion
 //#region src/tracing/otelConfig.ts
 /**
@@ -3200,7 +3180,6 @@ function getDefaultOtelConfig() {
 		enabled: true
 	};
 }
-
 //#endregion
 //#region src/tracing/localSpanExporter.ts
 /**
@@ -3220,7 +3199,7 @@ var LocalSpanExporter = class {
 			});
 			else resultCallback({ code: ExportResultCode.SUCCESS });
 		}).catch((error) => {
-			logger_default.error("[LocalSpanExporter] Failed to export spans", { error });
+			logger.error("[LocalSpanExporter] Failed to export spans", { error });
 			resultCallback({
 				code: ExportResultCode.FAILED,
 				error: error instanceof Error ? error : new Error(String(error))
@@ -3234,7 +3213,7 @@ var LocalSpanExporter = class {
 	async exportAsync(spans) {
 		if (spans.length === 0) return;
 		const traceStore = getTraceStore();
-		logger_default.debug(`[LocalSpanExporter] Exporting ${spans.length} spans`);
+		logger.debug(`[LocalSpanExporter] Exporting ${spans.length} spans`);
 		const spansByTrace = /* @__PURE__ */ new Map();
 		for (const span of spans) {
 			const traceId = span.spanContext().traceId;
@@ -3245,12 +3224,12 @@ var LocalSpanExporter = class {
 		let firstError;
 		for (const [traceId, spanDataList] of spansByTrace) try {
 			const result = await traceStore.addSpans(traceId, spanDataList, { skipTraceCheck: false });
-			if (result.stored) logger_default.debug(`[LocalSpanExporter] Added ${spanDataList.length} spans to trace ${traceId}`);
-			else logger_default.debug(`[LocalSpanExporter] Skipping ${spanDataList.length} spans for orphan trace ${traceId}: ${result.reason}`);
+			if (result.stored) logger.debug(`[LocalSpanExporter] Added ${spanDataList.length} spans to trace ${traceId}`);
+			else logger.debug(`[LocalSpanExporter] Skipping ${spanDataList.length} spans for orphan trace ${traceId}: ${result.reason}`);
 		} catch (error) {
-			if ((error instanceof Error ? error.message : String(error)).includes("FOREIGN KEY")) logger_default.debug(`[LocalSpanExporter] Skipping ${spanDataList.length} spans for orphan trace ${traceId}`);
+			if ((error instanceof Error ? error.message : String(error)).includes("FOREIGN KEY")) logger.debug(`[LocalSpanExporter] Skipping ${spanDataList.length} spans for orphan trace ${traceId}`);
 			else {
-				logger_default.error(`[LocalSpanExporter] Failed to add spans to trace ${traceId}`, { error });
+				logger.error(`[LocalSpanExporter] Failed to add spans to trace ${traceId}`, { error });
 				if (!firstError) firstError = error instanceof Error ? error : new Error(String(error));
 			}
 		}
@@ -3287,7 +3266,7 @@ var LocalSpanExporter = class {
 	* Shutdown the exporter. No-op for local storage.
 	*/
 	shutdown() {
-		logger_default.debug("[LocalSpanExporter] Shutting down");
+		logger.debug("[LocalSpanExporter] Shutting down");
 		return Promise.resolve();
 	}
 	/**
@@ -3297,7 +3276,6 @@ var LocalSpanExporter = class {
 		return Promise.resolve();
 	}
 };
-
 //#endregion
 //#region src/tracing/otelSdk.ts
 let provider = null;
@@ -3325,21 +3303,21 @@ function getHandlers() {
 */
 function initializeOtel(config) {
 	if (initialized) {
-		logger_default.debug("[OtelSdk] Already initialized, skipping");
+		logger.debug("[OtelSdk] Already initialized, skipping");
 		return;
 	}
 	if (!config.enabled) {
-		logger_default.debug("[OtelSdk] OTEL tracing is disabled");
+		logger.debug("[OtelSdk] OTEL tracing is disabled");
 		return;
 	}
-	logger_default.debug("[OtelSdk] Initializing OpenTelemetry SDK", {
+	logger.debug("[OtelSdk] Initializing OpenTelemetry SDK", {
 		serviceName: config.serviceName,
 		endpoint: config.endpoint,
 		localExport: config.localExport
 	});
 	if (config.debug) diag.setLogger(new DiagConsoleLogger(), DiagLogLevel.DEBUG);
 	propagation.setGlobalPropagator(new W3CTraceContextPropagator());
-	logger_default.debug("[OtelSdk] Registered W3C Trace Context propagator");
+	logger.debug("[OtelSdk] Registered W3C Trace Context propagator");
 	const resource = resourceFromAttributes({
 		[ATTR_SERVICE_NAME]: config.serviceName,
 		[ATTR_SERVICE_VERSION]: VERSION
@@ -3348,12 +3326,12 @@ function initializeOtel(config) {
 	if (config.localExport) {
 		const localExporter = new LocalSpanExporter();
 		spanProcessors.push(new BatchSpanProcessor(localExporter));
-		logger_default.debug("[OtelSdk] Added local span exporter");
+		logger.debug("[OtelSdk] Added local span exporter");
 	}
 	if (config.endpoint) {
 		const otlpExporter = new OTLPTraceExporter({ url: config.endpoint });
 		spanProcessors.push(new BatchSpanProcessor(otlpExporter));
-		logger_default.debug(`[OtelSdk] Added OTLP exporter to ${config.endpoint}`);
+		logger.debug(`[OtelSdk] Added OTLP exporter to ${config.endpoint}`);
 	}
 	provider = new NodeTracerProvider({
 		resource,
@@ -3361,7 +3339,7 @@ function initializeOtel(config) {
 	});
 	provider.register();
 	initialized = true;
-	logger_default.info("[OtelSdk] OpenTelemetry SDK initialized successfully");
+	logger.info("[OtelSdk] OpenTelemetry SDK initialized successfully");
 	setupShutdownHandlers();
 }
 /**
@@ -3370,12 +3348,12 @@ function initializeOtel(config) {
 */
 async function shutdownOtel() {
 	if (!initialized || !provider) return;
-	logger_default.debug("[OtelSdk] Shutting down OpenTelemetry SDK");
+	logger.debug("[OtelSdk] Shutting down OpenTelemetry SDK");
 	try {
 		await provider.shutdown();
-		logger_default.info("[OtelSdk] OpenTelemetry SDK shut down successfully");
+		logger.info("[OtelSdk] OpenTelemetry SDK shut down successfully");
 	} catch (error) {
-		logger_default.error("[OtelSdk] Error shutting down OpenTelemetry SDK", { error });
+		logger.error("[OtelSdk] Error shutting down OpenTelemetry SDK", { error });
 	} finally {
 		provider = null;
 		initialized = false;
@@ -3388,12 +3366,12 @@ async function shutdownOtel() {
 */
 async function flushOtel() {
 	if (!initialized || !provider) return;
-	logger_default.debug("[OtelSdk] Flushing pending spans");
+	logger.debug("[OtelSdk] Flushing pending spans");
 	try {
 		await provider.forceFlush();
-		logger_default.debug("[OtelSdk] Spans flushed successfully");
+		logger.debug("[OtelSdk] Spans flushed successfully");
 	} catch (error) {
-		logger_default.error("[OtelSdk] Error flushing spans", { error });
+		logger.error("[OtelSdk] Error flushing spans", { error });
 	}
 }
 /**
@@ -3405,7 +3383,7 @@ function setupShutdownHandlers() {
 	const handlers = getHandlers();
 	if (handlers.registered) return;
 	const shutdown = async (signal) => {
-		logger_default.debug(`[OtelSdk] Received ${signal}, shutting down`);
+		logger.debug(`[OtelSdk] Received ${signal}, shutting down`);
 		await shutdownOtel();
 	};
 	handlers.sigTermHandler = () => {
@@ -3442,7 +3420,6 @@ function cleanupShutdownHandlers() {
 	}
 	handlers.registered = false;
 }
-
 //#endregion
 //#region src/util/exportToFile/writeToFile.ts
 var JsonlFileWriter = class {
@@ -3466,7 +3443,6 @@ var JsonlFileWriter = class {
 		});
 	}
 };
-
 //#endregion
 //#region src/evaluator.ts
 /**
@@ -3658,7 +3634,7 @@ async function runEval({ provider, prompt, test, testSuite, delay, nunjucksFilte
 		if (test.providerOutput) response.output = test.providerOutput;
 		else {
 			const activeProvider = isApiProvider(test.provider) ? test.provider : provider;
-			logger_default.debug(`Provider type: ${activeProvider.id()}`);
+			logger.debug(`Provider type: ${activeProvider.id()}`);
 			traceContext = await generateTraceContextIfNeeded(test, evaluateOptions, testIdx, promptIdx, testSuite);
 			const callApiContext = {
 				vars,
@@ -3669,7 +3645,7 @@ async function runEval({ provider, prompt, test, testSuite, delay, nunjucksFilte
 				filters,
 				originalProvider: provider,
 				test,
-				logger: logger_default,
+				logger,
 				getCache,
 				repeatIndex
 			};
@@ -3686,8 +3662,8 @@ async function runEval({ provider, prompt, test, testSuite, delay, nunjucksFilte
 				const sanitizedMetadata = safeJsonStringify(response.metadata);
 				response.metadata = sanitizedMetadata ? JSON.parse(sanitizedMetadata) : {};
 			}
-			logger_default.debug(`Provider response properties: ${Object.keys(response).join(", ")}`);
-			logger_default.debug(`Provider response cached property explicitly: ${response.cached}`);
+			logger.debug(`Provider response properties: ${Object.keys(response).join(", ")}`);
+			logger.debug(`Provider response cached property explicitly: ${response.cached}`);
 		}
 		latencyMs = Date.now() - startTime;
 		let conversationLastInput = void 0;
@@ -3704,12 +3680,12 @@ async function runEval({ provider, prompt, test, testSuite, delay, nunjucksFilte
 				metadata: response.metadata
 			});
 		}
-		logger_default.debug("Evaluator response", { responsePreview: (safeJsonStringify(response) ?? "").slice(0, 100) });
-		logger_default.debug(`Evaluator checking cached flag: response.cached = ${Boolean(response.cached)}, provider.delay = ${provider.delay}`);
+		logger.debug("Evaluator response", { responsePreview: (safeJsonStringify(response) ?? "").slice(0, 100) });
+		logger.debug(`Evaluator checking cached flag: response.cached = ${Boolean(response.cached)}, provider.delay = ${provider.delay}`);
 		if (!response.cached && provider.delay > 0) {
-			logger_default.debug(`Sleeping for ${provider.delay}ms`);
+			logger.debug(`Sleeping for ${provider.delay}ms`);
 			await sleep(provider.delay);
-		} else if (response.cached) logger_default.debug(`Skipping delay because response is cached`);
+		} else if (response.cached) logger.debug(`Skipping delay because response is cached`);
 		const ret = {
 			...setup,
 			response,
@@ -3812,7 +3788,7 @@ async function runEval({ provider, prompt, test, testSuite, delay, nunjucksFilte
 			promptIdx,
 			testIdx
 		});
-		if (!(err instanceof Error && err.name === "AbortError")) logger_default.error("Provider call failed during eval", logContext);
+		if (!(err instanceof Error && err.name === "AbortError")) logger.error("Provider call failed during eval", logContext);
 		return [{
 			...setup,
 			error: errorWithStack,
@@ -3895,7 +3871,7 @@ function generateVarCombinations(vars) {
 		let values = [];
 		if (typeof vars[key] === "string" && vars[key].startsWith("file://")) {
 			const filePath = vars[key].slice(7);
-			const basePath = cliState_default.basePath || "";
+			const basePath = state.basePath || "";
 			values = (globSync(filePath, {
 				cwd: basePath || process.cwd(),
 				windowsPathsNoEscape: true
@@ -3935,28 +3911,28 @@ var Evaluator = class {
 		this.conversations = {};
 		this.registers = {};
 		this.fileWriters = (Array.isArray(evalRecord.config.outputPath) ? evalRecord.config.outputPath.filter((p) => p.endsWith(".jsonl")) : evalRecord.config.outputPath?.endsWith(".jsonl") ? [evalRecord.config.outputPath] : []).map((p) => new JsonlFileWriter(p));
-		this.rateLimitRegistry = createRateLimitRegistry({ maxConcurrency: options.maxConcurrency || DEFAULT_MAX_CONCURRENCY });
+		this.rateLimitRegistry = createRateLimitRegistry({ maxConcurrency: options.maxConcurrency || 4 });
 		this.rateLimitRegistry.on("ratelimit:hit", (data) => {
-			logger_default.debug(`[Scheduler] Rate limit hit for ${data.rateLimitKey}`, {
+			logger.debug(`[Scheduler] Rate limit hit for ${data.rateLimitKey}`, {
 				retryAfterMs: data.retryAfterMs,
 				resetAt: data.resetAt,
 				concurrencyChange: data.concurrencyChange
 			});
 		});
 		this.rateLimitRegistry.on("ratelimit:learned", (data) => {
-			logger_default.debug(`[Scheduler] Learned rate limits for ${data.rateLimitKey}`, {
+			logger.debug(`[Scheduler] Learned rate limits for ${data.rateLimitKey}`, {
 				requestLimit: data.requestLimit,
 				tokenLimit: data.tokenLimit
 			});
 		});
 		this.rateLimitRegistry.on("concurrency:decreased", (data) => {
-			logger_default.debug(`[Scheduler] Concurrency decreased for ${data.rateLimitKey}`, {
+			logger.debug(`[Scheduler] Concurrency decreased for ${data.rateLimitKey}`, {
 				previous: data.previous,
 				current: data.current
 			});
 		});
 		this.rateLimitRegistry.on("concurrency:increased", (data) => {
-			logger_default.debug(`[Scheduler] Concurrency increased for ${data.rateLimitKey}`, {
+			logger.debug(`[Scheduler] Concurrency increased for ${data.rateLimitKey}`, {
 				previous: data.previous,
 				current: data.current
 			});
@@ -4013,7 +3989,7 @@ var Evaluator = class {
 		const checkAbort = () => {
 			if (combinedAbortSignal.aborted) throw new Error("Operation cancelled");
 		};
-		if (!options.silent) logger_default.info(`Starting evaluation ${this.evalRecord.id}`);
+		if (!options.silent) logger.info(`Starting evaluation ${this.evalRecord.id}`);
 		checkAbort();
 		const prompts = [];
 		const assertionTypes = /* @__PURE__ */ new Set();
@@ -4025,32 +4001,32 @@ var Evaluator = class {
 		}
 		testSuite = (await runExtensionHook(testSuite.extensions, "beforeAll", { suite: testSuite })).suite;
 		if (options.generateSuggestions) {
-			logger_default.info(`Generating prompt variations...`);
+			logger.info(`Generating prompt variations...`);
 			const { prompts: newPrompts, error } = await generatePrompts(testSuite.prompts[0].raw, 1);
 			if (error || !newPrompts) throw new Error(`Failed to generate prompts: ${error}`);
-			logger_default.info(chalk.blue("Generated prompts:"));
+			logger.info(chalk.blue("Generated prompts:"));
 			let numAdded = 0;
 			for (const prompt of newPrompts) {
-				logger_default.info("--------------------------------------------------------");
-				logger_default.info(`${prompt}`);
-				logger_default.info("--------------------------------------------------------");
+				logger.info("--------------------------------------------------------");
+				logger.info(`${prompt}`);
+				logger.info("--------------------------------------------------------");
 				if (await promptYesNo("Do you want to test this prompt?", false)) {
 					testSuite.prompts.push({
 						raw: prompt,
 						label: prompt
 					});
 					numAdded++;
-				} else logger_default.info("Skipping this prompt.");
+				} else logger.info("Skipping this prompt.");
 			}
 			if (numAdded < 1) {
-				logger_default.info(chalk.red("No prompts selected. Aborting."));
+				logger.info(chalk.red("No prompts selected. Aborting."));
 				process.exitCode = 1;
 				return this.evalRecord;
 			}
 		}
 		const existingPromptsMap = /* @__PURE__ */ new Map();
-		if (cliState_default.resume && this.evalRecord.persisted && this.evalRecord.prompts.length > 0) {
-			logger_default.debug("Resuming evaluation: preserving metrics from previous run");
+		if (state.resume && this.evalRecord.persisted && this.evalRecord.prompts.length > 0) {
+			logger.debug("Resuming evaluation: preserving metrics from previous run");
 			for (const existingPrompt of this.evalRecord.prompts) {
 				const key = `${existingPrompt.provider}:${existingPrompt.id}`;
 				existingPromptsMap.set(key, existingPrompt);
@@ -4088,7 +4064,7 @@ var Evaluator = class {
 		await this.evalRecord.addPrompts(prompts);
 		let tests = testSuite.tests && testSuite.tests.length > 0 ? testSuite.tests : testSuite.scenarios ? [] : [{}];
 		if (testSuite.scenarios && testSuite.scenarios.length > 0) {
-			telemetry_default.record("feature_used", { feature: "scenarios" });
+			telemetry.record("feature_used", { feature: "scenarios" });
 			let scenarioIndex = 0;
 			for (const scenario of testSuite.scenarios) for (const data of scenario.config) {
 				const scenarioTests = (scenario.tests || [{}]).map((test) => {
@@ -4152,7 +4128,7 @@ var Evaluator = class {
 		}
 		const runEvalOptions = [];
 		let testIdx = 0;
-		let concurrency = options.maxConcurrency || DEFAULT_MAX_CONCURRENCY;
+		let concurrency = options.maxConcurrency || 4;
 		for (let index = 0; index < tests.length; index++) {
 			const testCase = tests[index];
 			invariant(typeof testSuite.defaultTest !== "object" || Array.isArray(testSuite.defaultTest?.assert || []), `defaultTest.assert is not an array in test case #${index + 1}`);
@@ -4172,7 +4148,7 @@ var Evaluator = class {
 				const defaultProvider = testSuite.defaultTest.provider;
 				if (isApiProvider(defaultProvider)) testCase.provider = defaultProvider;
 				else if (typeof defaultProvider === "object" && defaultProvider.id) {
-					const { loadApiProvider } = await import("./providers-CsXB2Ix-.js");
+					const { loadApiProvider } = await import("./providers-B7V0njNs.js");
 					testCase.provider = await loadApiProvider(typeof defaultProvider.id === "function" ? defaultProvider.id() : defaultProvider.id, { options: defaultProvider });
 				} else testCase.provider = defaultProvider;
 			}
@@ -4199,7 +4175,7 @@ var Evaluator = class {
 						const promptId = generateIdFromPrompt(prompt);
 						const promptIdx = promptIndexMap.get(`${providerKey}:${promptId}`);
 						if (promptIdx === void 0) {
-							logger_default.warn(`Could not find prompt index for ${providerKey}:${promptId}, skipping`);
+							logger.warn(`Could not find prompt index for ${providerKey}:${promptId}, skipping`);
 							continue;
 						}
 						runEvalOptions.push({
@@ -4222,7 +4198,7 @@ var Evaluator = class {
 									options: testOptions
 								};
 								const tracingEnabled = getEnvBool("PROMPTFOO_TRACING_ENABLED", false) || testCase.metadata?.tracingEnabled === true || testSuite.tracing?.enabled === true;
-								logger_default.debug(`[Evaluator] Tracing check: env=${getEnvBool("PROMPTFOO_TRACING_ENABLED", false)}, testCase.metadata?.tracingEnabled=${testCase.metadata?.tracingEnabled}, testSuite.tracing?.enabled=${testSuite.tracing?.enabled}, tracingEnabled=${tracingEnabled}`);
+								logger.debug(`[Evaluator] Tracing check: env=${getEnvBool("PROMPTFOO_TRACING_ENABLED", false)}, testCase.metadata?.tracingEnabled=${testCase.metadata?.tracingEnabled}, testSuite.tracing?.enabled=${testSuite.tracing?.enabled}, tracingEnabled=${tracingEnabled}`);
 								if (tracingEnabled) return {
 									...baseTest,
 									metadata: {
@@ -4255,27 +4231,27 @@ var Evaluator = class {
 			if (evalOption.test.assert?.some((a) => a.type === "select-best")) rowsWithSelectBestAssertion.add(evalOption.testIdx);
 			if (evalOption.test.assert?.some((a) => a.type === "max-score")) rowsWithMaxScoreAssertion.add(evalOption.testIdx);
 		}
-		if (cliState_default.resume && this.evalRecord.persisted) try {
-			const { default: EvalResult } = await import("./evalResult-trqZjVYh.js");
-			const completedPairs = await EvalResult.getCompletedIndexPairs(this.evalRecord.id, { excludeErrors: cliState_default.retryMode });
+		if (state.resume && this.evalRecord.persisted) try {
+			const { default: EvalResult } = await import("./evalResult-5xwYnECe.js");
+			const completedPairs = await EvalResult.getCompletedIndexPairs(this.evalRecord.id, { excludeErrors: state.retryMode });
 			const originalCount = runEvalOptions.length;
 			for (let i = runEvalOptions.length - 1; i >= 0; i--) {
 				const step = runEvalOptions[i];
 				if (completedPairs.has(`${step.testIdx}:${step.promptIdx}`)) runEvalOptions.splice(i, 1);
 			}
 			const skipped = originalCount - runEvalOptions.length;
-			if (skipped > 0) logger_default.info(`Resuming: skipping ${skipped} previously completed cases`);
+			if (skipped > 0) logger.info(`Resuming: skipping ${skipped} previously completed cases`);
 		} catch (err) {
-			logger_default.warn(`Resume: failed to load completed results. Running full evaluation. ${String(err)}`);
+			logger.warn(`Resume: failed to load completed results. Running full evaluation. ${String(err)}`);
 		}
 		if (concurrency > 1) {
 			const usesConversation = prompts.some((p) => p.raw.includes("_conversation"));
 			const usesStoreOutputAs = tests.some((t) => t.options?.storeOutputAs);
 			if (usesConversation) {
-				logger_default.info(`Setting concurrency to 1 because the ${chalk.cyan("_conversation")} variable is used.`);
+				logger.info(`Setting concurrency to 1 because the ${chalk.cyan("_conversation")} variable is used.`);
 				concurrency = 1;
 			} else if (usesStoreOutputAs) {
-				logger_default.info(`Setting concurrency to 1 because storeOutputAs is used.`);
+				logger.info(`Setting concurrency to 1 because storeOutputAs is used.`);
 				concurrency = 1;
 			}
 		}
@@ -4306,14 +4282,14 @@ var Evaluator = class {
 					await this.evalRecord.addResult(row);
 				} catch (error) {
 					const resultSummary = summarizeEvaluateResultForLogging(row);
-					logger_default.error(`Error saving result: ${error} ${safeJsonStringify(resultSummary)}`);
+					logger.error(`Error saving result: ${error} ${safeJsonStringify(resultSummary)}`);
 				}
 				for (const writer of this.fileWriters) await writer.write(row);
 				const httpStatus = row.response?.metadata?.http?.status;
 				if (typeof httpStatus === "number" && isNonTransientHttpStatus(httpStatus)) {
 					targetUnavailable = true;
 					targetErrorStatus = httpStatus;
-					logger_default.error(`Target returned HTTP ${httpStatus}. Aborting scan - this error will not resolve on retry.`);
+					logger.error(`Target returned HTTP ${httpStatus}. Aborting scan - this error will not resolve on retry.`);
 					targetErrorAbortController.abort();
 					break;
 				}
@@ -4333,7 +4309,7 @@ var Evaluator = class {
 				if (testSuite.derivedMetrics) {
 					const math = await import("mathjs");
 					const promptEvalCount = metrics.testPassCount + metrics.testFailCount + metrics.testErrorCount + 1;
-					if (Object.prototype.hasOwnProperty.call(metrics.namedScores, "__count")) logger_default.warn("Metric name '__count' is reserved for derived metrics and will be overridden.");
+					if (Object.prototype.hasOwnProperty.call(metrics.namedScores, "__count")) logger.warn("Metric name '__count' is reserved for derived metrics and will be overridden.");
 					const evalContext = {
 						...metrics.namedScores,
 						__count: promptEvalCount
@@ -4348,7 +4324,7 @@ var Evaluator = class {
 							}
 							evalContext[metric.name] = metrics.namedScores[metric.name];
 						} catch (error) {
-							logger_default.debug(`Could not evaluate derived metric '${metric.name}': ${error.message}`);
+							logger.debug(`Could not evaluate derived metric '${metric.name}': ${error.message}`);
 						}
 					}
 				}
@@ -4387,7 +4363,7 @@ var Evaluator = class {
 						if (typeof evalStep.provider.cleanup === "function") try {
 							evalStep.provider.cleanup();
 						} catch (cleanupErr) {
-							logger_default.warn(`Error during provider cleanup: ${cleanupErr}`);
+							logger.warn(`Error during provider cleanup: ${cleanupErr}`);
 						}
 						reject(/* @__PURE__ */ new Error(`Evaluation timed out after ${timeoutMs}ms`));
 					}, timeoutMs);
@@ -4451,8 +4427,8 @@ var Evaluator = class {
 			}
 		};
 		const originalProgressCallback = this.options.progressCallback;
-		const isWebUI = Boolean(cliState_default.webUI);
-		logger_default.debug(`Progress bar settings: showProgressBar=${this.options.showProgressBar}, isWebUI=${isWebUI}`);
+		const isWebUI = Boolean(state.webUI);
+		logger.debug(`Progress bar settings: showProgressBar=${this.options.showProgressBar}, isWebUI=${isWebUI}`);
 		if (isCI() && !isWebUI) {
 			ciProgressReporter = new CIProgressReporter(runEvalOptions.length);
 			ciProgressReporter.start();
@@ -4462,20 +4438,20 @@ var Evaluator = class {
 			if (isWebUI) {
 				const provider = evalStep.provider.label || evalStep.provider.id();
 				const vars = formatVarsForDisplay(evalStep.test.vars, 50);
-				logger_default.info(`[${numComplete}/${total}] Running ${provider} with vars: ${vars}`);
+				logger.info(`[${numComplete}/${total}] Running ${provider} with vars: ${vars}`);
 			} else if (progressBarManager) {
 				const phase = evalStep.test.options?.runSerially ? "serial" : "concurrent";
 				progressBarManager.updateProgress(index, evalStep, phase, metrics);
 			} else if (ciProgressReporter) ciProgressReporter.update(numComplete);
-			else logger_default.debug(`Eval #${index + 1} complete (${numComplete} of ${runEvalOptions.length})`);
+			else logger.debug(`Eval #${index + 1} complete (${numComplete} of ${runEvalOptions.length})`);
 		};
 		const serialRunEvalOptions = [];
 		const concurrentRunEvalOptions = [];
 		for (const evalOption of runEvalOptions) if (evalOption.test.options?.runSerially) serialRunEvalOptions.push(evalOption);
 		else concurrentRunEvalOptions.push(evalOption);
 		if (!this.options.silent) {
-			if (serialRunEvalOptions.length > 0) logger_default.info(`Running ${serialRunEvalOptions.length} test cases serially...`);
-			if (concurrentRunEvalOptions.length > 0) logger_default.info(`Running ${concurrentRunEvalOptions.length} test cases (up to ${concurrency} at a time)...`);
+			if (serialRunEvalOptions.length > 0) logger.info(`Running ${serialRunEvalOptions.length} test cases serially...`);
+			if (concurrentRunEvalOptions.length > 0) logger.info(`Running ${concurrentRunEvalOptions.length} test cases (up to ${concurrency} at a time)...`);
 		}
 		if (this.options.showProgressBar && progressBarManager) await progressBarManager.initialize(runEvalOptions, concurrency, 0);
 		try {
@@ -4484,7 +4460,7 @@ var Evaluator = class {
 				if (isWebUI) {
 					const provider = evalStep.provider.label || evalStep.provider.id();
 					const vars = formatVarsForDisplay(evalStep.test.vars || {}, 50);
-					logger_default.info(`[${numComplete}/${runEvalOptions.length}] Running ${provider} with vars: ${vars}`);
+					logger.info(`[${numComplete}/${runEvalOptions.length}] Running ${provider} with vars: ${vars}`);
 				}
 				const idx = runEvalOptions.indexOf(evalStep);
 				await processEvalStepWithTimeout(evalStep, idx);
@@ -4499,9 +4475,9 @@ var Evaluator = class {
 			});
 		} catch (err) {
 			if (combinedAbortSignal.aborted) {
-				if (evalTimedOut) logger_default.warn(`Evaluation stopped after reaching max duration (${maxEvalTimeMs}ms)`);
+				if (evalTimedOut) logger.warn(`Evaluation stopped after reaching max duration (${maxEvalTimeMs}ms)`);
 				else if (!targetUnavailable) {
-					logger_default.info("Evaluation interrupted, saving progress...");
+					logger.info("Evaluation interrupted, saving progress...");
 					if (globalTimeout) clearTimeout(globalTimeout);
 					if (progressBarManager) progressBarManager.stop();
 					if (ciProgressReporter) ciProgressReporter.finish();
@@ -4531,10 +4507,10 @@ var Evaluator = class {
 		let compareCount = 0;
 		for (const testIdx of rowsWithSelectBestAssertion) {
 			compareCount++;
-			if (isWebUI) logger_default.info(`Running model-graded comparison ${compareCount} of ${compareRowsCount}...`);
+			if (isWebUI) logger.info(`Running model-graded comparison ${compareCount} of ${compareRowsCount}...`);
 			const resultsToCompare = this.evalRecord.persisted ? await this.evalRecord.fetchResultsByTestIdx(testIdx) : this.evalRecord.results.filter((r) => r.testIdx === testIdx);
 			if (resultsToCompare.length === 0) {
-				logger_default.warn(`Expected results to be found for test index ${testIdx}`);
+				logger.warn(`Expected results to be found for test index ${testIdx}`);
 				continue;
 			}
 			const compareAssertion = resultsToCompare[0].testCase.assert?.find((a) => a.type === "select-best");
@@ -4596,16 +4572,16 @@ var Evaluator = class {
 				}
 				if (progressBarManager) progressBarManager.updateComparisonProgress(resultsToCompare[0].prompt.raw);
 				else if (ciProgressReporter) ciProgressReporter.update(runEvalOptions.length + compareCount);
-				else if (!isWebUI) logger_default.debug(`Model-graded comparison #${compareCount} of ${compareRowsCount} complete`);
+				else if (!isWebUI) logger.debug(`Model-graded comparison #${compareCount} of ${compareRowsCount} complete`);
 			}
 		}
 		const maxScoreRowsCount = rowsWithMaxScoreAssertion.size;
 		if (maxScoreRowsCount > 0) {
-			logger_default.info(`Processing ${maxScoreRowsCount} max-score assertions...`);
+			logger.info(`Processing ${maxScoreRowsCount} max-score assertions...`);
 			for (const testIdx of rowsWithMaxScoreAssertion) {
 				const resultsToCompare = this.evalRecord.persisted ? await this.evalRecord.fetchResultsByTestIdx(testIdx) : this.evalRecord.results.filter((r) => r.testIdx === testIdx);
 				if (resultsToCompare.length === 0) {
-					logger_default.warn(`Expected results to be found for test index ${testIdx}`);
+					logger.warn(`Expected results to be found for test index ${testIdx}`);
 					continue;
 				}
 				const maxScoreAssertion = resultsToCompare[0].testCase.assert?.find((a) => a.type === "max-score");
@@ -4613,7 +4589,7 @@ var Evaluator = class {
 					const maxScoreGradingResults = await selectMaxScore(resultsToCompare.map((r) => r.response?.output || ""), resultsToCompare, maxScoreAssertion);
 					if (progressBarManager) progressBarManager.updateComparisonProgress(resultsToCompare[0].prompt.raw);
 					else if (ciProgressReporter) ciProgressReporter.update(runEvalOptions.length + compareCount);
-					else if (!isWebUI) logger_default.debug(`Max-score assertion for test #${testIdx} complete`);
+					else if (!isWebUI) logger.debug(`Max-score assertion for test #${testIdx} complete`);
 					for (let index = 0; index < resultsToCompare.length; index++) {
 						const result = resultsToCompare[index];
 						const maxScoreGradingResult = {
@@ -4657,7 +4633,7 @@ var Evaluator = class {
 				progressBarManager.stop();
 			} else if (ciProgressReporter) ciProgressReporter.finish();
 		} catch (cleanupErr) {
-			logger_default.warn(`Error during progress reporter cleanup: ${cleanupErr}`);
+			logger.warn(`Error during progress reporter cleanup: ${cleanupErr}`);
 		}
 		if (globalTimeout) clearTimeout(globalTimeout);
 		if (evalTimedOut) {
@@ -4730,7 +4706,7 @@ var Evaluator = class {
 			return idParts.length > 1 ? idParts[0] : "unknown";
 		})));
 		const timeoutOccurred = evalTimedOut || this.evalRecord.results.some((r) => r.failureReason === ResultFailureReason.ERROR && r.error?.includes("timed out"));
-		telemetry_default.record("eval_ran", {
+		telemetry.record("eval_ran", {
 			numPrompts: prompts.length,
 			numTests: this.stats.successes + this.stats.failures + this.stats.errors,
 			numRequests: this.stats.tokenUsage.numRequests || 0,
@@ -4778,26 +4754,26 @@ var Evaluator = class {
 		await startOtlpReceiverIfNeeded(this.testSuite);
 		const tracingEnabled = getEnvBool("PROMPTFOO_TRACING_ENABLED", false) || this.testSuite.tracing?.enabled === true || typeof this.testSuite.defaultTest === "object" && this.testSuite.defaultTest?.metadata?.tracingEnabled === true || this.testSuite.tests?.some((t) => t.metadata?.tracingEnabled === true);
 		if (tracingEnabled) {
-			logger_default.debug("[Evaluator] Initializing OTEL SDK for tracing");
+			logger.debug("[Evaluator] Initializing OTEL SDK for tracing");
 			initializeOtel(getDefaultOtelConfig());
 		}
 		try {
 			return await this._runEvaluation();
 		} finally {
 			if (tracingEnabled) {
-				logger_default.debug("[Evaluator] Flushing OTEL spans...");
+				logger.debug("[Evaluator] Flushing OTEL spans...");
 				await flushOtel();
 				await shutdownOtel();
 			}
 			if (isOtlpReceiverStarted()) {
-				logger_default.debug("[Evaluator] Waiting for span exports to complete...");
+				logger.debug("[Evaluator] Waiting for span exports to complete...");
 				await sleep(3e3);
 			}
 			await stopOtlpReceiverIfNeeded();
 			await providerRegistry.shutdownAll();
 			if (this.rateLimitRegistry) {
 				const metrics = this.rateLimitRegistry.getMetrics();
-				for (const [key, m] of Object.entries(metrics)) if (m.totalRequests > 0) logger_default.debug(`[Scheduler] Final metrics for ${key}`, {
+				for (const [key, m] of Object.entries(metrics)) if (m.totalRequests > 0) logger.debug(`[Scheduler] Final metrics for ${key}`, {
 					totalRequests: m.totalRequests,
 					completedRequests: m.completedRequests,
 					failedRequests: m.failedRequests,
@@ -4810,14 +4786,14 @@ var Evaluator = class {
 			}
 			this.rateLimitRegistry?.dispose();
 			redteamProviderManager.setRateLimitRegistry(void 0);
-			cliState_default.maxConcurrency = void 0;
+			state.maxConcurrency = void 0;
 		}
 	}
 };
 function evaluate(testSuite, evalRecord, options) {
 	return new Evaluator(testSuite, evalRecord, options).evaluate();
 }
-
 //#endregion
 export { runEval as a, readAssertions as c, isAllowedPrompt as i, renderMetricName as l, formatVarsForDisplay as n, doesPromptRefMatch as o, generateVarCombinations as r, assertions_default as s, evaluate as t, runAssertions as u };
-//# sourceMappingURL=evaluator-8aGyV12L.js.map
+
+//# sourceMappingURL=evaluator-BhoWwp5b.js.map