@ls-stack/agent-eval 0.27.0 → 0.28.0

package/dist/{runOrchestration-DB0dwGrd.mjs → runOrchestration-ClWYWPen.mjs}

@@ -2,7 +2,7 @@ import { createRequire, registerHooks } from "node:module";
 import { createHash } from "node:crypto";
 import { mkdir, readFile, readdir, rename, rm, stat, writeFile } from "node:fs/promises";
 import { extname, isAbsolute, join, relative, resolve } from "node:path";
-import { formatWithOptions, isDeepStrictEqual } from "node:util";
+import { formatWithOptions, isDeepStrictEqual, stripVTControlCharacters } from "node:util";
 import { AsyncLocalStorage } from "node:async_hooks";
 import { z, z as z$1 } from "zod/v4";
 import { Buffer as Buffer$1 } from "node:buffer";
@@ -29,6 +29,19 @@ function defineEval(definition) {
 	});
 }
 //#endregion
+//#region ../sdk/src/stackFormatting.ts
+const orphanedAnsiSgrPattern$1 = /\[(?:\d{1,3}(?:;\d{1,3})*)?m/g;
+/**
+* Remove terminal styling control codes from captured stack text.
+*
+* Some stack providers add ANSI SGR codes for terminal output. Persisted eval
+* artifacts are rendered in the web UI, so stacks should be stored as plain
+* text.
+*/
+function stripTerminalControlCodes$1(value) {
+	return stripVTControlCharacters(value).replaceAll(orphanedAnsiSgrPattern$1, "");
+}
+//#endregion
 //#region ../sdk/src/runtime.ts
 const scopeStorage = new AsyncLocalStorage();
 const runtimeScopeStorage = new AsyncLocalStorage();
@@ -518,10 +531,13 @@ function recordOpIfActive(scope, op) {
 	if (top) top.ops.push(op);
 }
 function toAssertionFailure$1(message, error = void 0) {
-	return error?.stack ? {
+	const name = error?.name;
+	const stack = error?.stack ? stripTerminalControlCodes$1(error.stack) : void 0;
+	return {
+		...name !== void 0 ? { name } : {},
 		message,
-		stack: error.stack
-	} : { message };
+		...stack !== void 0 ? { stack } : {}
+	};
 }
 /**
 * Record or replace an output value for the current case scope.
@@ -789,7 +805,8 @@ function repoFile(path, mimeType) {
 }
 //#endregion
 //#region ../sdk/src/cacheSerialization.ts
-const serializedCacheValueMarker = "__agentEvalsCacheSerialization";
+const serializedCacheValueMarker = "__aecs";
+const legacySerializedCacheValueMarker = "__agentEvalsCacheSerialization";
 const jsonSafeCacheValueVersion = "json-safe-v1";
 const packedNumberArrayMinLength = 128;
 const compressedStringMinBytes = 16 * 1024;
@@ -799,7 +816,7 @@ function isRecordLike$3(value) {
 	return typeof value === "object" && value !== null && !Array.isArray(value);
 }
 function isJsonSafeSerializedCacheValue(value) {
-	return isRecordLike$3(value) && value[serializedCacheValueMarker] === jsonSafeCacheValueVersion && typeof value.type === "string";
+	return isRecordLike$3(value) && serializationMarkerValue(value) === jsonSafeCacheValueVersion && typeof value.type === "string";
 }
 function jsonSafeValue(type, value) {
 	return value === void 0 ? {
@@ -812,32 +829,39 @@ function jsonSafeValue(type, value) {
 	};
 }
 function hasSerializationMarkerKey(value) {
-	return Object.hasOwn(value, serializedCacheValueMarker);
+	return Object.hasOwn(value, serializedCacheValueMarker) || Object.hasOwn(value, legacySerializedCacheValueMarker);
+}
+function serializationMarkerValue(value) {
+	return value[serializedCacheValueMarker] ?? value[legacySerializedCacheValueMarker];
 }
 /**
 * Serialize one cached value while keeping plain JSON as plain JSON.
 *
-* Rich runtime values use small tagged wrappers.
+* Rich runtime values use small tagged wrappers. Undefined values are omitted
+* by default; pass `preserveUndefined: true` to round-trip them explicitly.
 */
-async function serializeCacheValue(value) {
-	return serializeJsonSafeValue(value, /* @__PURE__ */ new WeakSet(), 0);
+async function serializeCacheValue(value, options = void 0) {
+	return serializeJsonSafeValue(value, /* @__PURE__ */ new WeakSet(), 0, normalizeCacheSerializationOptions(options));
 }
 /** Revive one cached value, while preserving legacy JSON-round-tripped data. */
 function deserializeCacheValue(value) {
 	return deserializeJsonSafeValue(value);
 }
 /** Clone one value through the same serialization path used for cache data. */
-async function cloneCacheValue(value) {
-	return deserializeCacheValue(await serializeCacheValue(value));
+async function cloneCacheValue(value, options = void 0) {
+	return deserializeCacheValue(await serializeCacheValue(value, options));
+}
+function normalizeCacheSerializationOptions(options) {
+	return { preserveUndefined: options?.preserveUndefined === true };
 }
-async function serializeJsonSafeValue(value, refs, depth) {
-	if (value === void 0) return jsonSafeValue("Undefined");
+async function serializeJsonSafeValue(value, refs, depth, config) {
+	if (value === void 0) return config.preserveUndefined ? jsonSafeValue("Undefined") : void 0;
 	if (typeof value === "bigint") return jsonSafeValue("BigInt", value.toString());
 	if (typeof value === "number") return serializeNumber(value);
 	if (typeof value === "string") return serializeString(value, depth);
 	if (value instanceof Date) return jsonSafeValue("Date", value.toISOString());
-	if (value instanceof Map) return serializeMap(value, refs, depth);
-	if (value instanceof Set) return serializeSet(value, refs, depth);
+	if (value instanceof Map) return serializeMap(value, refs, depth, config);
+	if (value instanceof Set) return serializeSet(value, refs, depth, config);
 	if (value instanceof RegExp) return jsonSafeValue("RegExp", {
 		flags: value.flags,
 		source: value.source
@@ -856,7 +880,7 @@ async function serializeJsonSafeValue(value, refs, depth) {
 		type: value.type
 	});
 	if (value instanceof ArrayBuffer) return jsonSafeValue("ArrayBuffer", bytesToBase64(new Uint8Array(value)));
-	if (value instanceof Error) return serializeError(value, refs, depth);
+	if (value instanceof Error) return serializeError(value, refs, depth, config);
 	if (!value || typeof value !== "object") return value;
 	if (refs.has(value)) throw new Error("Circular cache values are not supported");
 	refs.add(value);
@@ -869,12 +893,18 @@ async function serializeJsonSafeValue(value, refs, depth) {
 			}
 		}
 		const items = [];
-		for (const item of value) items.push(await serializeJsonSafeValue(item, refs, depth + 1));
+		for (const item of value) {
+			const serializedItem = await serializeJsonSafeValue(item, refs, depth + 1, config);
+			if (serializedItem !== void 0) items.push(serializedItem);
+		}
 		refs.delete(value);
 		return compressNestedJsonValue(items, depth) ?? items;
 	}
 	const entries = [];
-	for (const [key, entryValue] of Object.entries(value)) entries.push([key, await serializeJsonSafeValue(entryValue, refs, depth + 1)]);
+	for (const [key, entryValue] of Object.entries(value)) {
+		const serializedEntryValue = await serializeJsonSafeValue(entryValue, refs, depth + 1, config);
+		if (serializedEntryValue !== void 0) entries.push([key, serializedEntryValue]);
+	}
 	refs.delete(value);
 	const serialized = hasSerializationMarkerKey(value) ? jsonSafeValue("Object", entries) : Object.fromEntries(entries);
 	return compressNestedJsonValue(serialized, depth) ?? serialized;
@@ -944,32 +974,40 @@ function compressNestedJsonValue(value, depth) {
 function compressionIsWorthIt(value, rawSize) {
 	return Buffer$1.byteLength(JSON.stringify(value)) < rawSize * maxCompressedSizeRatio;
 }
-async function serializeMap(value, refs, depth) {
+async function serializeMap(value, refs, depth, config) {
 	if (refs.has(value)) throw new Error("Circular cache values are not supported");
 	refs.add(value);
 	const entries = [];
-	for (const [key, entryValue] of value.entries()) entries.push([await serializeJsonSafeValue(key, refs, depth + 1), await serializeJsonSafeValue(entryValue, refs, depth + 1)]);
+	for (const [key, entryValue] of value.entries()) {
+		const serializedKey = await serializeJsonSafeValue(key, refs, depth + 1, config);
+		const serializedEntryValue = await serializeJsonSafeValue(entryValue, refs, depth + 1, config);
+		if (serializedKey !== void 0 && serializedEntryValue !== void 0) entries.push([serializedKey, serializedEntryValue]);
+	}
 	refs.delete(value);
 	return jsonSafeValue("Map", entries);
 }
-async function serializeSet(value, refs, depth) {
+async function serializeSet(value, refs, depth, config) {
 	if (refs.has(value)) throw new Error("Circular cache values are not supported");
 	refs.add(value);
 	const items = [];
-	for (const item of value.values()) items.push(await serializeJsonSafeValue(item, refs, depth + 1));
+	for (const item of value.values()) {
+		const serializedItem = await serializeJsonSafeValue(item, refs, depth + 1, config);
+		if (serializedItem !== void 0) items.push(serializedItem);
+	}
 	refs.delete(value);
 	return jsonSafeValue("Set", items);
 }
-async function serializeError(value, refs, depth) {
+async function serializeError(value, refs, depth, config) {
 	if (refs.has(value)) throw new Error("Circular cache values are not supported");
 	refs.add(value);
 	const props = [];
 	for (const [key, entryValue] of Object.entries(value)) {
 		if (key === "cause") continue;
-		props.push([key, await serializeJsonSafeValue(entryValue, refs, depth + 1)]);
+		const serializedEntryValue = await serializeJsonSafeValue(entryValue, refs, depth + 1, config);
+		if (serializedEntryValue !== void 0) props.push([key, serializedEntryValue]);
 	}
 	const serialized = jsonSafeValue("Error", {
-		cause: "cause" in value ? await serializeJsonSafeValue(value.cause, refs, depth + 1) : void 0,
+		cause: "cause" in value ? await serializeJsonSafeValue(value.cause, refs, depth + 1, config) : void 0,
 		message: value.message,
 		name: value.name,
 		props,
@@ -1110,33 +1148,36 @@ function deserializeError(value) {
 	});
 	return error;
 }
-async function serializeRecordValues(record) {
+async function serializeRecordValues(record, config) {
 	const entries = [];
-	for (const [key, value] of Object.entries(record)) entries.push([key, await serializeCacheValue(value)]);
+	for (const [key, value] of Object.entries(record)) {
+		const serializedValue = await serializeCacheValue(value, config);
+		if (serializedValue !== void 0) entries.push([key, serializedValue]);
+	}
 	return Object.fromEntries(entries);
 }
 function deserializeRecordValues(record) {
 	return Object.fromEntries(Object.entries(record).map(([key, value]) => [key, deserializeCacheValue(value)]));
 }
-async function serializeCacheRecordingOp(op) {
+async function serializeCacheRecordingOp(op, config) {
 	switch (op.kind) {
 		case "setOutput":
 		case "appendOutput": return {
 			...op,
-			value: await serializeCacheValue(op.value)
+			value: await serializeCacheValue(op.value, config)
 		};
 		case "mergeOutput": return {
 			...op,
-			patch: await serializeRecordValues(op.patch)
+			patch: await serializeRecordValues(op.patch, config)
 		};
 		case "incrementOutput": return op;
 		case "checkpoint": return {
 			...op,
-			data: await serializeCacheValue(op.data)
+			data: await serializeCacheValue(op.data, config)
 		};
 		case "subSpan": return {
 			...op,
-			span: await serializeCacheSpan(op.span)
+			span: await serializeCacheSpan(op.span, config)
 		};
 	}
 }
@@ -1162,11 +1203,11 @@ function deserializeCacheRecordingOp(op) {
 		};
 	}
 }
-async function serializeCacheSpan(span) {
+async function serializeCacheSpan(span, config) {
 	return {
 		...span,
-		attributes: span.attributes === void 0 ? void 0 : await serializeRecordValues(span.attributes),
-		children: await Promise.all(span.children.map(serializeCacheSpan))
+		attributes: span.attributes === void 0 ? void 0 : await serializeRecordValues(span.attributes, config),
+		children: await Promise.all(span.children.map((child) => serializeCacheSpan(child, config)))
 	};
 }
 function deserializeCacheSpan(span) {
@@ -1176,13 +1217,19 @@ function deserializeCacheSpan(span) {
 		children: span.children.map(deserializeCacheSpan)
 	};
 }
-/** Serialize all rich values captured in a cache recording before persistence. */
-async function serializeCacheRecording(recording) {
+/**
+* Serialize all rich values captured in a cache recording before persistence.
+*
+* Undefined values are omitted by default; pass `preserveUndefined: true` to
+* retain the legacy explicit undefined wrappers in the recording payload.
+*/
+async function serializeCacheRecording(recording, options = void 0) {
+	const config = normalizeCacheSerializationOptions(options);
 	return {
 		...recording,
-		returnValue: await serializeCacheValue(recording.returnValue),
-		finalAttributes: await serializeRecordValues(recording.finalAttributes),
-		ops: await Promise.all(recording.ops.map(serializeCacheRecordingOp))
+		returnValue: await serializeCacheValue(recording.returnValue, config),
+		finalAttributes: await serializeRecordValues(recording.finalAttributes, config),
+		ops: await Promise.all(recording.ops.map((op) => serializeCacheRecordingOp(op, config)))
 	};
 }
 /** Revive all rich values captured in a cache recording after lookup. */
@@ -1574,7 +1621,9 @@ function createTraceCache(generateSpanId) {
 			key: info.key
 		}, { serializeFileBytes: info.serializeFileBytes === true });
 		const activeSpan = scope.activeSpanStack.at(-1);
-		if (cacheCtx.mode === "use") {
+		const canRead = cacheCtx.mode === "use" && cacheCtx.read !== false;
+		const canStore = cacheCtx.mode !== "bypass" && cacheCtx.store !== false;
+		if (canRead) {
 			const hit = await cacheCtx.adapter.lookup(namespace, keyHash);
 			if (hit) {
 				const storedAt = hit.storedAt;
@@ -1597,14 +1646,24 @@ function createTraceCache(generateSpanId) {
 				name: info.name,
 				namespace,
 				key: keyHash,
-				status: "miss"
+				status: "miss",
+				...canStore ? {} : { stored: false }
 			});
-		} else if (cacheCtx.mode === "refresh") recordCacheRef(scope, activeSpan, {
+		} else if (cacheCtx.mode === "use" && canStore) recordCacheRef(scope, activeSpan, {
+			type: "value",
+			name: info.name,
+			namespace,
+			key: keyHash,
+			status: "miss",
+			read: false
+		});
+		else if (cacheCtx.mode === "refresh") recordCacheRef(scope, activeSpan, {
 			type: "value",
 			name: info.name,
 			namespace,
 			key: keyHash,
-			status: "refresh"
+			status: "refresh",
+			...canStore ? {} : { stored: false }
 		});
 		else recordCacheRef(scope, activeSpan, {
 			type: "value",
@@ -1627,7 +1686,7 @@ function createTraceCache(generateSpanId) {
 			scope.recordingStack.pop();
 		}
 		appendSubSpanOps(scope, frame);
-		if (cacheCtx.mode !== "bypass") {
+		if (canStore) {
 			const finalAttributes = diffNonCacheAttributes(beforeAttributes, await snapshotNonCacheAttributes(activeSpan));
 			const recording = {
 				returnValue: bodyResult,
@@ -1641,13 +1700,11 @@ function createTraceCache(generateSpanId) {
 				operationType: "value",
 				operationName: info.name,
 				storedAt: new Date(getRealDateNowMs()).toISOString(),
-				codeFingerprint: cacheCtx.codeFingerprint,
 				recording: await serializeCacheRecording(recording)
 			}, {
 				rawKey: info.key,
 				operationType: "value",
-				operationName: info.name,
-				codeFingerprint: cacheCtx.codeFingerprint
+				operationName: info.name
 			});
 		}
 		return bodyResult;
@@ -2018,11 +2075,13 @@ async function traceSpanInternal(info, fn) {
 				namespace,
 				key: cacheOpts.key
 			}, { serializeFileBytes: cacheOpts.serializeFileBytes === true });
+			const canRead = ctx.mode === "use" && ctx.read !== false;
+			const canStore = ctx.mode !== "bypass" && ctx.store !== false;
 			mergeSpanAttributes(spanRecord, {
 				"cache.key": keyHash,
 				"cache.namespace": namespace
 			});
-			if (ctx.mode === "use") {
+			if (canRead) {
 				const hit = await ctx.adapter.lookup(namespace, keyHash);
 				if (hit) {
 					const storedAt = hit.storedAt;
@@ -2037,8 +2096,18 @@ async function traceSpanInternal(info, fn) {
 					spanRecord.endedAt = addElapsedMsToTimestamp(spanRecord.startedAt, getRealDateNowMs() - realStartedAt);
 					return recording.returnValue;
 				}
-				mergeSpanAttributes(spanRecord, { "cache.status": "miss" });
-			} else if (ctx.mode === "refresh") mergeSpanAttributes(spanRecord, { "cache.status": "refresh" });
+				mergeSpanAttributes(spanRecord, {
+					"cache.status": "miss",
+					...canStore ? {} : { "cache.stored": false }
+				});
+			} else if (ctx.mode === "use" && canStore) mergeSpanAttributes(spanRecord, {
+				"cache.status": "miss",
+				"cache.read": false
+			});
+			else if (ctx.mode === "refresh") mergeSpanAttributes(spanRecord, {
+				"cache.status": "refresh",
+				...canStore ? {} : { "cache.stored": false }
+			});
 			else mergeSpanAttributes(spanRecord, { "cache.status": "bypass" });
 			const frame = {
 				baseSpanIndex: scope.spans.length,
@@ -2054,7 +2123,7 @@ async function traceSpanInternal(info, fn) {
 			}
 			appendSubSpanOps(scope, frame);
 			finishSpanWithoutThrownError(spanRecord, realStartedAt);
-			if (ctx.mode !== "bypass") {
+			if (canStore) {
 				const recording = {
 					returnValue: bodyResult,
 					finalAttributes: stripCacheAttributes(spanRecord.attributes),
@@ -2074,14 +2143,12 @@ async function traceSpanInternal(info, fn) {
 					spanName: info.name,
 					spanKind: info.kind,
 					storedAt: new Date(getRealDateNowMs()).toISOString(),
-					codeFingerprint: ctx.codeFingerprint,
 					recording: await serializeCacheRecording(recording)
 				};
 				await ctx.adapter.write(entry, {
 					rawKey: cacheOpts.key,
 					operationType: "span",
-					operationName: info.name,
-					codeFingerprint: ctx.codeFingerprint
+					operationName: info.name
 				});
 			}
 			return bodyResult;
@@ -2274,6 +2341,7 @@ const columnDefSchema = z.object({
 	passThreshold: z.number().optional(),
 	maxStars: z.number().int().min(2).optional(),
 	hideInTable: z.boolean().optional(),
+	hideIfNoValue: z.boolean().optional(),
 	align: z.enum([
 		"left",
 		"center",
@@ -2417,6 +2485,10 @@ const traceCacheRefSchema = z.object({
 	namespace: z.string(),
 	key: z.string(),
 	status: cacheStatusSchema,
+	/** Whether this ref attempted to read from cache. Defaults to true. */
+	read: z.boolean().optional(),
+	/** Whether this ref wrote a persisted cache entry. Defaults to true for misses/refreshes. */
+	stored: z.boolean().optional(),
 	storedAt: z.string().optional(),
 	age: z.number().optional()
 });
@@ -2429,7 +2501,6 @@ const cacheListItemSchema = z.object({
 	spanName: z.string().optional(),
 	spanKind: traceSpanKindSchema.optional(),
 	storedAt: z.string(),
-	codeFingerprint: z.string(),
 	sizeBytes: z.number()
 });
 /** Zod schema for `SerializedCacheSpan`, defined lazily for recursion. */
@@ -2511,7 +2582,6 @@ const cacheEntrySchema = z.object({
 	spanName: z.string().optional(),
 	spanKind: traceSpanKindSchema.optional(),
 	storedAt: z.string(),
-	codeFingerprint: z.string(),
 	recording: cacheRecordingSchema
 });
 /** Debug-only raw key metadata stored outside the reusable cache entry. */
@@ -2522,7 +2592,6 @@ const cacheDebugKeyEntrySchema = z.object({
 	operationType: cacheOperationTypeSchema,
 	operationName: z.string(),
 	storedAt: z.string(),
-	codeFingerprint: z.string(),
 	rawKey: z.unknown()
 });
 /** Cache lookup response with optional debug-only raw key data. */
@@ -2614,6 +2683,11 @@ const evalChartTooltipExtraSchema = z.discriminatedUnion("source", [z.object({
 const evalChartConfigSchema = z.object({
 	/** Optional heading shown above the chart frame in the UI. */
 	heading: z.string().optional(),
+	/**
+	* Hide this chart in the UI when none of its metrics has a numeric value in
+	* the rendered history window.
+	*/
+	hideIfNoValue: z.boolean().optional(),
 	type: evalChartTypeSchema,
 	/** At least one series must be declared. */
 	metrics: z.array(evalChartMetricSchema).min(1),
@@ -2654,17 +2728,31 @@ const evalStatAggregateSchema = z.enum([
 	"sum",
 	"last"
 ]);
+const hideIfNoValueShape = { 
+/**
+* Hide this stat in the UI when the current run has no displayable value.
+* Missing values, `null`, and empty strings count as no value; `0` remains
+* visible.
+*/
+hideIfNoValue: z.boolean().optional() };
 /**
 * One entry in the EvalCard stats row. Built-in kinds use latest run totals;
 * `column` aggregates a score or numeric output column across the latest run.
 */
 const evalStatItemSchema = z.discriminatedUnion("kind", [
-	z.object({ kind: z.literal("cases") }),
+	z.object({
+		kind: z.literal("cases"),
+		...hideIfNoValueShape
+	}),
 	z.object({
 		kind: z.literal("passRate"),
-		accent: z.boolean().optional()
+		accent: z.boolean().optional(),
+		...hideIfNoValueShape
+	}),
+	z.object({
+		kind: z.literal("duration"),
+		...hideIfNoValueShape
 	}),
-	z.object({ kind: z.literal("duration") }),
 	z.object({
 		kind: z.literal("column"),
 		key: z.string(),
@@ -2673,7 +2761,8 @@ const evalStatItemSchema = z.discriminatedUnion("kind", [
 		format: columnFormatSchema.optional(),
 		/** Number presentation options applied when `format: 'number'`. */
 		numberFormat: numberDisplayOptionsSchema.optional(),
-		accent: z.boolean().optional()
+		accent: z.boolean().optional(),
+		...hideIfNoValueShape
 	})
 ]);
 /** Ordered list of stats rendered in the EvalCard stats row. */
@@ -2754,6 +2843,12 @@ const caseRowSchema = z.object({
 });
 /** Structured assertion failure metadata captured for one case run. */
 const assertionFailureSchema = z.object({
+	/**
+	* Error class or category label rendered alongside the message (e.g.
+	* `EvalAssertionError`, `OutputsSchemaError`). Optional for legacy entries
+	* and synthetic failures without an originating Error.
+	*/
+	name: z.string().optional(),
 	/** Human-readable assertion failure message shown in the UI and artifacts. */
 	message: z.string(),
 	/** Stack trace captured from the originating error when available. */
@@ -2902,6 +2997,25 @@ const defaultConfigKeySchema = z.enum([
 ]);
 /** Removal config for built-in eval-level outputs and UI metadata. */
 const removeDefaultConfigSchema = z.union([z.literal(true), z.array(defaultConfigKeySchema)]);
+const evalDeriveValueFnSchema = z.custom((value) => typeof value === "function", { message: "Expected a derive output function" });
+/** Schema for keyed or object-returning trace-derived output config. */
+const evalDeriveConfigSchema = z.union([z.custom((value) => typeof value === "function", { message: "Expected a deriveFromTracing function" }), z.record(z.string().min(1), evalDeriveValueFnSchema)]);
+/** Schema for UI overrides on derived or scored columns. */
+const evalColumnOverrideSchema = z.object({
+	label: z.string().optional(),
+	format: columnFormatSchema.optional(),
+	numberFormat: numberDisplayOptionsSchema.optional(),
+	hideInTable: z.boolean().optional(),
+	hideIfNoValue: z.boolean().optional(),
+	align: z.enum([
+		"left",
+		"center",
+		"right"
+	]).optional(),
+	maxStars: z.number().int().min(2).optional()
+});
+/** Schema for column override maps keyed by output or score field name. */
+const evalColumnsSchema = z.record(z.string(), evalColumnOverrideSchema);
 /** Render formats supported by an LLM-call metric in the UI. */
 const llmCallMetricFormatSchema = z.enum([
 	"string",
@@ -2979,18 +3093,9 @@ const apiCallMetricSchema = z.object({
 	placements: z.array(apiCallMetricPlacementSchema).nonempty().optional()
 });
 /**
-* Schema for one model/provider pricing entry used to derive LLM-call costs
-* from token counts.
+* Schema for pricing rates used to derive LLM-call costs from token counts.
 */
-const llmCallPricingSchema = z.object({
-	/** Exact model name read from the configured `attributes.model` path. */
-	model: z.string().min(1),
-	/**
-	* Optional provider discriminator read from `attributes.provider`. When set,
-	* the entry only applies to calls from that provider; provider-specific
-	* entries take precedence over generic entries for the same model.
-	*/
-	provider: z.string().min(1).optional(),
+const llmCallPricingRateSchema = z.object({
 	/** USD per one million non-cached input tokens. */
 	inputUsdPerMillion: z.number().nonnegative().optional(),
 	/** USD per one million output tokens. */
@@ -3004,6 +3109,23 @@ const llmCallPricingSchema = z.object({
 	/** USD per one million reasoning tokens when reported separately. */
 	reasoningUsdPerMillion: z.number().nonnegative().optional()
 });
+/**
+* Schema for one model's pricing config. The object key is the exact model
+* name. Use `providers` when a model has provider-specific rates in addition
+* to, or instead of, generic model rates.
+*/
+const llmCallPricingSchema = llmCallPricingRateSchema.extend({
+	/**
+	* Optional provider discriminator read from `attributes.provider`. When set,
+	* the top-level entry only applies to calls from that provider.
+	*/
+	provider: z.string().min(1).optional(),
+	/**
+	* Provider-specific pricing for the model. Provider entries take precedence
+	* over generic rates for the same model.
+	*/
+	providers: z.record(z.string().min(1), llmCallPricingRateSchema).optional()
+});
 /** Schema for the global LLM calls config block in `agent-evals.config.ts`. */
 const llmCallsConfigSchema = z.object({
 	/** Span kinds treated as LLM calls. Defaults to `['llm']`. */
@@ -3040,10 +3162,10 @@ const llmCallsConfigSchema = z.object({
 	*/
 	derivedAttributes: z.record(z.string().min(1), callDerivedAttributeSchema).optional(),
 	/**
-	* Model/provider pricing registry used to calculate LLM-call costs from
-	* token counts. Built-in LLM cost fields are only derived from this registry.
+	* Model-keyed pricing registry used to calculate LLM-call costs from token
+	* counts. Built-in LLM cost fields are only derived from this registry.
 	*/
-	pricing: z.array(llmCallPricingSchema).optional(),
+	pricing: z.record(z.string().min(1), llmCallPricingSchema).optional(),
 	/** Custom user-defined metrics surfaced on each LLM call. */
 	metrics: z.array(llmCallMetricSchema).optional()
 });
@@ -3159,6 +3281,33 @@ function resolveApiCallMetric(metric) {
 		placements: metric.placements ? [...metric.placements] : ["body"]
 	};
 }
+function hasPricingRates(pricing) {
+	return pricing.inputUsdPerMillion !== void 0 || pricing.outputUsdPerMillion !== void 0 || pricing.cachedInputUsdPerMillion !== void 0 || pricing.cacheCreationInputUsdPerMillion !== void 0 || pricing.cacheCreationInput1hUsdPerMillion !== void 0 || pricing.reasoningUsdPerMillion !== void 0;
+}
+function copyPricingRates(pricing) {
+	return {
+		inputUsdPerMillion: pricing.inputUsdPerMillion,
+		outputUsdPerMillion: pricing.outputUsdPerMillion,
+		cachedInputUsdPerMillion: pricing.cachedInputUsdPerMillion,
+		cacheCreationInputUsdPerMillion: pricing.cacheCreationInputUsdPerMillion,
+		cacheCreationInput1hUsdPerMillion: pricing.cacheCreationInput1hUsdPerMillion,
+		reasoningUsdPerMillion: pricing.reasoningUsdPerMillion
+	};
+}
+function resolveLlmCallPricingEntries(model, pricing) {
+	const entries = [];
+	if (hasPricingRates(pricing)) entries.push({
+		model,
+		provider: pricing.provider,
+		...copyPricingRates(pricing)
+	});
+	for (const [provider, providerPricing] of Object.entries(pricing.providers ?? {})) entries.push({
+		model,
+		provider,
+		...copyPricingRates(providerPricing)
+	});
+	return entries;
+}
 /**
 * Resolve the user-authored LLM-calls config to a fully-defaulted shape used
 * by the UI to derive the LLM calls tab.
@@ -3169,7 +3318,7 @@ function resolveApiCallMetric(metric) {
 * - Missing `metrics[].format` defaults to `'string'`.
 * - Missing `metrics[].placements` defaults to `['body']`.
 * - Missing `pricing` defaults to an empty registry; built-in costs are only
-*   derived from configured pricing and token counts.
+*   derived from configured model-keyed pricing and token counts.
 */
 function resolveLlmCallsConfig(input) {
 	return {
@@ -3180,16 +3329,7 @@ function resolveLlmCallsConfig(input) {
 		},
 		derivedAttributes: resolveDerivedAttributes(input?.derivedAttributes),
 		metrics: (input?.metrics ?? []).map(resolveLlmCallMetric),
-		pricing: (input?.pricing ?? []).map((p) => ({
-			model: p.model,
-			provider: p.provider,
-			inputUsdPerMillion: p.inputUsdPerMillion,
-			outputUsdPerMillion: p.outputUsdPerMillion,
-			cachedInputUsdPerMillion: p.cachedInputUsdPerMillion,
-			cacheCreationInputUsdPerMillion: p.cacheCreationInputUsdPerMillion,
-			cacheCreationInput1hUsdPerMillion: p.cacheCreationInput1hUsdPerMillion,
-			reasoningUsdPerMillion: p.reasoningUsdPerMillion
-		}))
+		pricing: Object.entries(input?.pricing ?? {}).flatMap(([model, pricing]) => resolveLlmCallPricingEntries(model, pricing))
 	};
 }
 /**
@@ -3223,6 +3363,9 @@ const agentEvalsConfigSchema = z.object({
 	staleAfterDays: z.number().optional(),
 	allowCliRunAll: z.boolean().optional(),
 	traceDisplay: traceDisplayInputConfigSchema.optional(),
+	columns: evalColumnsSchema.optional(),
+	deriveFromTracing: evalDeriveConfigSchema.optional(),
+	stats: evalStatsConfigSchema.optional(),
 	llmCalls: llmCallsConfigSchema.optional(),
 	removeDefaultConfig: removeDefaultConfigSchema.optional(),
 	apiCalls: apiCallsConfigSchema.optional(),
@@ -3875,6 +4018,11 @@ function readNumber(attributes, key) {
 	const value = attributes[key];
 	return typeof value === "number" && Number.isFinite(value) ? value : void 0;
 }
+function readBoolean(attributes, key) {
+	if (!isRecord$2(attributes)) return void 0;
+	const value = attributes[key];
+	return typeof value === "boolean" ? value : void 0;
+}
 function readArray(attributes, key) {
 	if (!isRecord$2(attributes)) return [];
 	const value = attributes[key];
@@ -3903,12 +4051,14 @@ function extractCacheEntries(spans, caseCacheRefs) {
 			const namespace = readString(span.attributes, "cache.namespace");
 			if (key !== void 0 && namespace !== void 0) {
 				const isHit = status === "hit";
+				const stored = isHit ? true : readBoolean(span.attributes, "cache.stored") !== false;
 				entries.push({
 					id: span.id,
 					source: "span",
 					origin: "span",
-					action: isHit ? "hit" : "added",
+					action: isHit ? "hit" : stored ? "added" : "notStored",
 					status,
+					stored,
 					name: span.name,
 					namespace,
 					key,
@@ -3925,12 +4075,14 @@ function extractCacheEntries(spans, caseCacheRefs) {
 			const ref = parsed.data;
 			if (ref.status === "bypass") continue;
 			const isHit = ref.status === "hit";
+			const stored = isHit ? true : ref.stored !== false;
 			entries.push({
 				id: `${span.id}:value:${String(index)}`,
 				source: "value",
 				origin: "span",
-				action: isHit ? "hit" : "added",
+				action: isHit ? "hit" : stored ? "added" : "notStored",
 				status: ref.status,
+				stored,
 				name: ref.name,
 				namespace: ref.namespace,
 				key: ref.key,
@@ -3943,12 +4095,14 @@ function extractCacheEntries(spans, caseCacheRefs) {
 	for (const [index, ref] of caseCacheRefs.entries()) {
 		if (ref.status === "bypass") continue;
 		const isHit = ref.status === "hit";
+		const stored = isHit ? true : ref.stored !== false;
 		entries.push({
 			id: `case:value:${String(index)}`,
 			source: "value",
 			origin: "caseRoot",
-			action: isHit ? "hit" : "added",
+			action: isHit ? "hit" : stored ? "added" : "notStored",
 			status: ref.status,
+			stored,
 			name: ref.name,
 			namespace: ref.namespace,
 			key: ref.key,
@@ -4020,7 +4174,8 @@ const updateManualScoreRequestSchema = z.object({ value: z.number().min(0).max(1
 //#endregion
 //#region ../runner/src/cacheStore.ts
 const defaultMaxEntriesPerNamespace = 100;
-const cacheSerializationMarker = "__agentEvalsCacheSerialization";
+const cacheSerializationMarker = "__aecs";
+const legacyCacheSerializationMarker = "__agentEvalsCacheSerialization";
 const supportedCacheSerializationVersion = "json-safe-v1";
 /**
 * Create a filesystem-backed cache adapter rooted at `<workspaceRoot>/<dir>`.
@@ -4105,7 +4260,6 @@ function createFsCacheStore(options) {
 						spanName: entry.spanName,
 						spanKind: entry.spanKind,
 						storedAt: entry.storedAt,
-						codeFingerprint: entry.codeFingerprint,
 						sizeBytes: Buffer.byteLength(JSON.stringify(entry), "utf8")
 					});
 				}
@@ -4234,7 +4388,7 @@ async function readCacheFilePath(filePath) {
 function usesSupportedCacheSerialization(value) {
 	if (Array.isArray(value)) return value.every(usesSupportedCacheSerialization);
 	if (!isRecordLike(value)) return true;
-	if (Object.hasOwn(value, cacheSerializationMarker) && value[cacheSerializationMarker] !== supportedCacheSerializationVersion) return false;
+	for (const marker of [cacheSerializationMarker, legacyCacheSerializationMarker]) if (Object.hasOwn(value, marker) && value[marker] !== supportedCacheSerializationVersion) return false;
 	return Object.values(value).every(usesSupportedCacheSerialization);
 }
 async function writeOrRemoveCacheFile(cacheDir, cacheFile) {
@@ -4278,7 +4432,6 @@ async function writeDebugKeyEntry(params) {
 			operationType: debugKey.operationType,
 			operationName: debugKey.operationName,
 			storedAt: entry.storedAt,
-			codeFingerprint: debugKey.codeFingerprint,
 			rawKey: debugKey.rawKey
 		};
 		await writeDebugKeyFile(debugDir, {
@@ -4494,6 +4647,7 @@ function getScoreOverride(def) {
 		format: def.format,
 		numberFormat: def.numberFormat,
 		hideInTable: def.hideInTable,
+		hideIfNoValue: def.hideIfNoValue,
 		align: def.align,
 		maxStars: def.maxStars
 	};
@@ -4506,6 +4660,7 @@ function mergeOverrides(base, override) {
 		format: override.format ?? base.format,
 		numberFormat: override.numberFormat ?? base.numberFormat,
 		hideInTable: override.hideInTable ?? base.hideInTable,
+		hideIfNoValue: override.hideIfNoValue ?? base.hideIfNoValue,
 		align: override.align ?? base.align,
 		maxStars: override.maxStars ?? base.maxStars
 	};
@@ -4620,6 +4775,7 @@ function createColumnDef(params) {
 	if (override?.numberFormat !== void 0) def.numberFormat = override.numberFormat;
 	if (override?.maxStars !== void 0) def.maxStars = override.maxStars;
 	if (override?.hideInTable !== void 0) def.hideInTable = override.hideInTable;
+	if (override?.hideIfNoValue !== void 0) def.hideIfNoValue = override.hideIfNoValue;
 	if (override?.align !== void 0) def.align = override.align;
 	if (!isScore) return def;
 	def.isScore = true;
@@ -4704,60 +4860,70 @@ const DEFAULT_COLUMNS = {
 		label: "API Calls",
 		format: "number",
 		numberFormat: countNumberFormat,
-		align: "right"
+		align: "right",
+		hideIfNoValue: true
 	},
 	costUsd: {
 		label: "Cost",
 		format: "number",
 		numberFormat: costNumberFormat,
-		align: "right"
+		align: "right",
+		hideIfNoValue: true
 	},
 	llmTurns: {
 		label: "LLM Turns",
 		format: "number",
 		numberFormat: countNumberFormat,
-		align: "right"
+		align: "right",
+		hideIfNoValue: true
 	},
 	inputTokens: {
 		label: "Input Tokens",
 		format: "number",
 		numberFormat: tokenNumberFormat,
-		align: "right"
+		align: "right",
+		hideIfNoValue: true
 	},
 	outputTokens: {
 		label: "Output Tokens",
 		format: "number",
 		numberFormat: tokenNumberFormat,
-		align: "right"
+		align: "right",
+		hideIfNoValue: true
 	},
 	totalTokens: {
 		label: "Total Tokens",
 		format: "number",
 		numberFormat: tokenNumberFormat,
-		align: "right"
+		align: "right",
+		hideIfNoValue: true
 	},
 	cachedInputTokens: {
 		label: "Cached Input Tokens",
 		format: "number",
 		numberFormat: tokenNumberFormat,
-		align: "right"
+		align: "right",
+		hideIfNoValue: true
 	},
 	cacheCreationInputTokens: {
 		label: "Cache Write Tokens",
 		format: "number",
 		numberFormat: tokenNumberFormat,
-		align: "right"
+		align: "right",
+		hideIfNoValue: true
 	},
 	reasoningTokens: {
 		label: "Reasoning Tokens",
 		format: "number",
 		numberFormat: tokenNumberFormat,
-		align: "right"
+		align: "right",
+		hideIfNoValue: true
 	},
 	llmDurationMs: {
 		label: "LLM Duration",
 		format: "duration",
-		align: "right"
+		align: "right",
+		hideIfNoValue: true
 	}
 };
 function resolveRemovedKeys(globalRemove, evalRemove) {
@@ -4770,9 +4936,16 @@ function getActiveDefaultConfigKeys(params) {
 }
 function mergeDefaultColumns(params) {
 	const activeKeys = getActiveDefaultConfigKeys(params);
-	if (activeKeys.length === 0) return params.columns;
+	if (activeKeys.length === 0) {
+		const merged = {
+			...params.globalColumns,
+			...params.columns
+		};
+		return Object.keys(merged).length > 0 ? merged : void 0;
+	}
 	return {
 		...Object.fromEntries(activeKeys.map((key) => [key, DEFAULT_COLUMNS[key]])),
+		...params.globalColumns,
 		...params.columns
 	};
 }
@@ -4784,30 +4957,38 @@ function appendDefaultStats(params) {
 		key: "apiCalls",
 		label: "API Calls",
 		aggregate: "avg",
-		numberFormat: countNumberFormat
+		numberFormat: countNumberFormat,
+		hideIfNoValue: true
 	});
 	if (activeKeys.has("costUsd")) defaults.push({
 		kind: "column",
 		key: "costUsd",
 		label: "LLM Cost",
 		aggregate: "avg",
-		numberFormat: costNumberFormat
+		numberFormat: costNumberFormat,
+		hideIfNoValue: true
 	});
 	if (activeKeys.has("totalTokens")) defaults.push({
 		kind: "column",
 		key: "totalTokens",
 		label: "Tokens",
 		aggregate: "avg",
-		numberFormat: tokenNumberFormat
+		numberFormat: tokenNumberFormat,
+		hideIfNoValue: true
 	});
 	if (activeKeys.has("llmTurns")) defaults.push({
 		kind: "column",
 		key: "llmTurns",
 		label: "LLM Turns",
 		aggregate: "avg",
-		numberFormat: countNumberFormat
+		numberFormat: countNumberFormat,
+		hideIfNoValue: true
 	});
-	const merged = [...params.stats ?? [], ...defaults];
+	const merged = [
+		...params.globalStats ?? [],
+		...params.stats ?? [],
+		...defaults
+	];
 	return merged.length > 0 ? merged : void 0;
 }
 function appendDefaultCharts(params) {
@@ -4815,6 +4996,7 @@ function appendDefaultCharts(params) {
 	const defaults = [];
 	if (activeKeys.has("costUsd")) defaults.push({
 		heading: "LLM Cost",
+		hideIfNoValue: true,
 		type: "area",
 		metrics: [{
 			source: "column",
@@ -4856,6 +5038,7 @@ function appendDefaultCharts(params) {
 	].filter((metric) => metric !== null);
 	if (tokenMetrics.length > 0) defaults.push({
 		heading: "LLM Tokens",
+		hideIfNoValue: true,
 		type: "bar",
 		metrics: tokenMetrics,
 		tooltipExtras: activeKeys.has("totalTokens") ? [{
@@ -4872,11 +5055,13 @@ function resolveEvalDefaultConfig(params) {
 	const evalRemove = params.evalDef.removeDefaultConfig;
 	return {
 		columns: mergeDefaultColumns({
+			globalColumns: params.globalColumns,
 			columns: params.evalDef.columns,
 			globalRemove: params.globalRemove,
 			evalRemove
 		}),
 		stats: appendDefaultStats({
+			globalStats: params.globalStats,
 			stats: params.evalDef.stats,
 			globalRemove: params.globalRemove,
 			evalRemove
@@ -5214,6 +5399,65 @@ function isFile(value) {
 	return value instanceof File;
 }
 //#endregion
+//#region ../runner/src/traceDisplay.ts
+function isRecord$1(value) {
+	return typeof value === "object" && value !== null;
+}
+function mergeNestedAttribute(value, path, attributeValue) {
+	const root = value === void 0 ? {} : { ...value };
+	const parts = path.split(".");
+	let current = root;
+	for (const [index, part] of parts.entries()) {
+		if (index === parts.length - 1) {
+			current[part] = attributeValue;
+			continue;
+		}
+		const nextValue = current[part];
+		const nextRecord = isRecord$1(nextValue) ? { ...nextValue } : {};
+		current[part] = nextRecord;
+		current = nextRecord;
+	}
+	return root;
+}
+function resolveTracePresentation(spans, globalTraceDisplay, evalTraceDisplay) {
+	const merged = /* @__PURE__ */ new Map();
+	for (const attribute of globalTraceDisplay?.attributes ?? []) merged.set(attribute.key ?? attribute.path, attribute);
+	for (const attribute of evalTraceDisplay?.attributes ?? []) merged.set(attribute.key ?? attribute.path, attribute);
+	const resolvedAttributes = [];
+	const transformedTrace = spans.map((span) => ({
+		...span,
+		attributes: span.attributes === void 0 ? void 0 : { ...span.attributes }
+	}));
+	for (const attribute of merged.values()) {
+		const resolvedPath = attribute.transform ? `__display.${attribute.key ?? attribute.path}` : attribute.path;
+		resolvedAttributes.push({
+			key: attribute.key,
+			path: resolvedPath,
+			label: attribute.label,
+			format: attribute.format,
+			numberFormat: attribute.numberFormat,
+			placements: attribute.placements,
+			scope: attribute.scope,
+			mode: attribute.mode
+		});
+		if (!attribute.transform) continue;
+		for (const span of transformedTrace) {
+			const sourceValue = getNestedAttribute(span.attributes, attribute.path);
+			if (sourceValue === void 0) continue;
+			const transformedValue = attribute.transform({
+				value: sourceValue,
+				span
+			});
+			if (transformedValue === void 0) continue;
+			span.attributes = mergeNestedAttribute(span.attributes, resolvedPath, transformedValue);
+		}
+	}
+	return {
+		trace: transformedTrace,
+		traceDisplay: { attributes: resolvedAttributes }
+	};
+}
+//#endregion
 //#region ../runner/src/runMaintenance.ts
 async function persistRunState(runState) {
 	await writeFile(join(runState.runDir, "summary.json"), JSON.stringify(runState.summary, null, 2));
@@ -5532,63 +5776,10 @@ async function runWithModuleIsolation(context, fn) {
 	return await isolationStorage.run(context, fn);
 }
 //#endregion
-//#region ../runner/src/traceDisplay.ts
-function isRecord$1(value) {
-	return typeof value === "object" && value !== null;
-}
-function mergeNestedAttribute(value, path, attributeValue) {
-	const root = value === void 0 ? {} : { ...value };
-	const parts = path.split(".");
-	let current = root;
-	for (const [index, part] of parts.entries()) {
-		if (index === parts.length - 1) {
-			current[part] = attributeValue;
-			continue;
-		}
-		const nextValue = current[part];
-		const nextRecord = isRecord$1(nextValue) ? { ...nextValue } : {};
-		current[part] = nextRecord;
-		current = nextRecord;
-	}
-	return root;
-}
-function resolveTracePresentation(spans, globalTraceDisplay, evalTraceDisplay) {
-	const merged = /* @__PURE__ */ new Map();
-	for (const attribute of globalTraceDisplay?.attributes ?? []) merged.set(attribute.key ?? attribute.path, attribute);
-	for (const attribute of evalTraceDisplay?.attributes ?? []) merged.set(attribute.key ?? attribute.path, attribute);
-	const resolvedAttributes = [];
-	const transformedTrace = spans.map((span) => ({
-		...span,
-		attributes: span.attributes === void 0 ? void 0 : { ...span.attributes }
-	}));
-	for (const attribute of merged.values()) {
-		const resolvedPath = attribute.transform ? `__display.${attribute.key ?? attribute.path}` : attribute.path;
-		resolvedAttributes.push({
-			key: attribute.key,
-			path: resolvedPath,
-			label: attribute.label,
-			format: attribute.format,
-			numberFormat: attribute.numberFormat,
-			placements: attribute.placements,
-			scope: attribute.scope,
-			mode: attribute.mode
-		});
-		if (!attribute.transform) continue;
-		for (const span of transformedTrace) {
-			const sourceValue = getNestedAttribute(span.attributes, attribute.path);
-			if (sourceValue === void 0) continue;
-			const transformedValue = attribute.transform({
-				value: sourceValue,
-				span
-			});
-			if (transformedValue === void 0) continue;
-			span.attributes = mergeNestedAttribute(span.attributes, resolvedPath, transformedValue);
-		}
-	}
-	return {
-		trace: transformedTrace,
-		traceDisplay: { attributes: resolvedAttributes }
-	};
+//#region ../runner/src/stackFormatting.ts
+const orphanedAnsiSgrPattern = /\[(?:\d{1,3}(?:;\d{1,3})*)?m/g;
+function stripTerminalControlCodes(value) {
+	return stripVTControlCharacters(value).replaceAll(orphanedAnsiSgrPattern, "");
 }
 //#endregion
 //#region ../runner/src/runExecution.ts
@@ -5620,8 +5811,54 @@ function buildScopedEvalIdPrefix(params) {
 async function callWithUnknownResult(fn, args) {
 	return await Reflect.apply(fn, void 0, args);
 }
+async function callUnknownFunction(fn, args) {
+	if (typeof fn !== "function") throw new Error("Expected a function");
+	return await Reflect.apply(fn, void 0, args);
+}
+function assignDerivedOutputs(params) {
+	for (const [key, value] of Object.entries(params.derived)) {
+		if (key in params.outputs) continue;
+		params.outputs[key] = value;
+	}
+}
+async function resolveDeriveFromTracingConfig(params) {
+	const ctx = {
+		trace: params.traceTree,
+		input: params.evalCase.input,
+		case: params.evalCase
+	};
+	if (typeof params.deriveFromTracing === "function") {
+		const derived = await callUnknownFunction(params.deriveFromTracing, [ctx]);
+		if (!isRecord(derived)) throw new Error("deriveFromTracing must return an object");
+		return derived;
+	}
+	const derived = {};
+	for (const [key, compute] of Object.entries(params.deriveFromTracing)) {
+		const value = await callUnknownFunction(compute, [ctx]);
+		if (value !== void 0) derived[key] = value;
+	}
+	return derived;
+}
+async function runDeriveFromTracingConfig(params) {
+	if (params.deriveFromTracing === void 0) return;
+	const { deriveFromTracing } = params;
+	try {
+		const derived = await runInExistingEvalScope(params.scope, "derive", async () => await resolveDeriveFromTracingConfig({
+			deriveFromTracing,
+			traceTree: params.traceTree,
+			evalCase: params.evalCase
+		}));
+		assignDerivedOutputs({
+			outputs: params.scope.outputs,
+			derived
+		});
+	} catch (e) {
+		const message = `deriveFromTracing threw: ${e instanceof Error ? e.message : String(e)}`;
+		params.scope.assertionFailures.push(toAssertionFailure(message, e instanceof Error ? e : void 0));
+	}
+}
 async function runCase(params) {
-	const { evalDef, evalId, evalKey = evalId, evalCase, globalTraceDisplay, llmCallsConfig = resolveLlmCallsConfig(void 0), apiCallsConfig = resolveApiCallsConfig(void 0), globalRemoveDefaultConfig, trial, startTime, cacheAdapter, cacheMode, codeFingerprint, moduleIsolation, evalFilePath, evalFileRelativePath = evalFilePath, workspaceRoot, artifactDir, runId } = params;
+	const { evalDef, evalId, evalKey = evalId, evalCase, globalTraceDisplay, globalColumns, globalDeriveFromTracing, llmCallsConfig = resolveLlmCallsConfig(void 0), apiCallsConfig = resolveApiCallsConfig(void 0), globalRemoveDefaultConfig, trial, startTime, cacheAdapter, cacheMode, moduleIsolation, evalFilePath, evalFileRelativePath = evalFilePath, workspaceRoot, artifactDir, runId } = params;
 	const scopedIdPrefix = buildScopedEvalIdPrefix({
 		evalId,
 		evalFilePath,
@@ -5653,7 +5890,8 @@ async function runCase(params) {
 			adapter: cacheAdapter,
 			mode: cacheMode,
 			evalId,
-			codeFingerprint
+			read: evalDef.cache?.read,
+			store: evalDef.cache?.store
 		} : void 0,
 		startTime: evalDef.startTime,
 		freezeTime: evalDef.freezeTime
@@ -5666,22 +5904,19 @@ async function runCase(params) {
 	const traceTree = buildTraceTree(spansWithDerivedAttributes, scope.checkpoints);
 	const nonAssertError = executeError && !(executeError instanceof EvalAssertionError) ? executeError : null;
 	if (executeError instanceof EvalAssertionError && scope.assertionFailures.length === 0) scope.assertionFailures.push(toAssertionFailure(executeError.message, executeError));
-	if (!nonAssertError && evalDef.deriveFromTracing) {
-		const { deriveFromTracing } = evalDef;
-		try {
-			const derived = await runInExistingEvalScope(scope, "derive", async () => {
-				return await callWithUnknownResult(deriveFromTracing, [{
-					trace: traceTree,
-					input: evalCase.input,
-					case: evalCase
-				}]);
-			});
-			if (!isRecord(derived)) throw new Error("deriveFromTracing must return an object");
-			for (const [key, value] of Object.entries(derived)) if (!(key in scope.outputs)) scope.outputs[key] = value;
-		} catch (e) {
-			const message = `deriveFromTracing threw: ${e instanceof Error ? e.message : String(e)}`;
-			scope.assertionFailures.push(toAssertionFailure(message, e instanceof Error ? e : void 0));
-		}
+	if (!nonAssertError) {
+		await runDeriveFromTracingConfig({
+			deriveFromTracing: globalDeriveFromTracing,
+			scope,
+			traceTree,
+			evalCase
+		});
+		await runDeriveFromTracingConfig({
+			deriveFromTracing: evalDef.deriveFromTracing,
+			scope,
+			traceTree,
+			evalCase
+		});
 	}
 	if (!nonAssertError) addDefaultOutputs({
 		outputs: scope.outputs,
@@ -5698,7 +5933,7 @@ async function runCase(params) {
 			...scope.outputs,
 			...parsedOutputs.data
 		};
-		else scope.assertionFailures.push(toAssertionFailure(formatOutputsSchemaError(parsedOutputs.error)));
+		else scope.assertionFailures.push(toAssertionFailure(formatOutputsSchemaError(parsedOutputs.error), void 0, "OutputsSchemaError"));
 	}
 	const scoreResults = /* @__PURE__ */ new Map();
 	const scoringTraces = {};
@@ -5721,7 +5956,8 @@ async function runCase(params) {
 				adapter: cacheAdapter,
 				mode: cacheMode,
 				evalId: `${evalId}__score__${key}`,
-				codeFingerprint
+				read: evalDef.cache?.read,
+				store: evalDef.cache?.store
 			} : void 0,
 			startTime: scoreStartTime,
 			freezeTime: evalDef.freezeTime
@@ -5776,6 +6012,7 @@ async function runCase(params) {
 	const { trace: displayTrace, traceDisplay } = resolveTracePresentation(spansWithDerivedAttributes, globalTraceDisplay, evalDef.traceDisplay);
 	const columns = {};
 	const columnOverrides = mergeDefaultColumns({
+		globalColumns,
 		columns: evalDef.columns,
 		globalRemove: globalRemoveDefaultConfig,
 		evalRemove: evalDef.removeDefaultConfig
@@ -5839,14 +6076,17 @@ function formatOutputsSchemaError(error) {
 	const issueLines = error.issues.map((issue) => {
 		return `${issue.path.length > 0 ? issue.path.join(".") : "<root>"}: ${issue.message}`;
 	});
-	if (issueLines.length === 0) return "outputsSchema validation failed";
-	return `outputsSchema validation failed:\n${issueLines.join("\n")}`;
+	if (issueLines.length === 0) return "outputs did not match the configured schema";
+	return issueLines.join("\n");
 }
-function toAssertionFailure(message, error = void 0) {
-	return error?.stack ? {
+function toAssertionFailure(message, error = void 0, nameOverride = void 0) {
+	const name = nameOverride ?? error?.name;
+	const stack = error?.stack ? stripTerminalControlCodes(error.stack) : void 0;
+	return {
+		...name !== void 0 ? { name } : {},
 		message,
-		stack: error.stack
-	} : { message };
+		...stack !== void 0 ? { stack } : {}
+	};
 }
 //#endregion
 //#region ../runner/src/runQueue.ts
@@ -6076,15 +6316,15 @@ async function executeRun({ runState, request, runDir, config, cacheStore, lastR
 		const apiCallsConfig = resolveApiCallsConfig(config.apiCalls);
 		for (const evalMeta of targetEvals) {
 			const evalFilePath = evalMeta.sourceFilePath;
-			let codeFingerprint = "";
+			let sourceFingerprint = "";
 			try {
-				codeFingerprint = getSourceFingerprint(await readFile(evalFilePath, "utf-8"));
+				sourceFingerprint = getSourceFingerprint(await readFile(evalFilePath, "utf-8"));
 			} catch {
-				codeFingerprint = "";
+				sourceFingerprint = "";
 			}
-			if (codeFingerprint.length > 0) {
-				runState.manifest.evalSourceFingerprints[evalMeta.key] = codeFingerprint;
-				evalMeta.sourceFingerprint = codeFingerprint;
+			if (sourceFingerprint.length > 0) {
+				runState.manifest.evalSourceFingerprints[evalMeta.key] = sourceFingerprint;
+				evalMeta.sourceFingerprint = sourceFingerprint;
 			} else {
 				delete runState.manifest.evalSourceFingerprints[evalMeta.key];
 				evalMeta.sourceFingerprint = null;
@@ -6093,7 +6333,7 @@ async function executeRun({ runState, request, runDir, config, cacheStore, lastR
 				const registry = getEvalRegistry();
 				await runWithModuleIsolation(moduleIsolation, async () => {
 					await runInEvalRuntimeScope("env", async () => {
-						await loadEvalModule(evalFilePath, codeFingerprint);
+						await loadEvalModule(evalFilePath, sourceFingerprint);
 					});
 				});
 				const entry = registry.get(evalMeta.id);
@@ -6117,6 +6357,8 @@ async function executeRun({ runState, request, runDir, config, cacheStore, lastR
 							runState.summary.totalCases += cases.length;
 							const defaultConfig = resolveEvalDefaultConfig({
 								evalDef,
+								globalColumns: config.columns,
+								globalStats: config.stats,
 								globalRemove: config.removeDefaultConfig
 							});
 							const declaredColumnDefs = buildDeclaredColumnDefs(defaultConfig.columns, evalDef.scores, evalDef.manualScores);
@@ -6162,6 +6404,8 @@ async function executeRun({ runState, request, runDir, config, cacheStore, lastR
 												evalKey: evalMeta.key,
 												evalCase,
 												globalTraceDisplay,
+												globalColumns: config.columns,
+												globalDeriveFromTracing: config.deriveFromTracing,
 												llmCallsConfig,
 												apiCallsConfig,
 												globalRemoveDefaultConfig: config.removeDefaultConfig,
@@ -6169,7 +6413,6 @@ async function executeRun({ runState, request, runDir, config, cacheStore, lastR
 												startTime,
 												cacheAdapter: bufferedCacheStore ?? (cacheEnabled ? cacheStore : null),
 												cacheMode,
-												codeFingerprint,
 												moduleIsolation,
 												evalFilePath,
 												evalFileRelativePath: evalMeta.filePath,
@@ -6324,4 +6567,4 @@ function toLastRunStatus(status) {
 	return status === "pending" ? null : status;
 }
 //#endregion
-export { llmCallsConfigSchema as $, traceSpanKindSchema as $t, extractApiCalls as A, getCurrentScope as An, evalChartTypeSchema as At, runSummarySchema as B, setEvalOutput as Bn, cacheRecordingOpSchema as Bt, validateCharts as C, evalExpect as Cn, evalChartAggregateSchema as Ct, sseEnvelopeSchema as D, configureEvalRunLogs as Dn, evalChartConfigSchema as Dt, updateManualScoreRequestSchema as E, appendToEvalOutput as En, evalChartColorSchema as Et, getEvalDisplayStatus as F, mergeEvalOutput as Fn, cacheEntryWithDebugKeySchema as Ft, apiCallMetricPlacementSchema as G, traceCacheRefSchema as Gt, DEFAULT_LLM_CALLS_CONFIG as H, startEvalBackgroundJob as Hn, cacheStatusSchema as Ht, deriveScopedSummaryFromCases as I, nextEvalId as In, cacheFileSchema as It, defaultConfigKeySchema as J, traceAttributeDisplayPlacementSchema as Jt, apiCallMetricSchema as K, traceAttributeDisplayFormatSchema as Kt, deriveStatusFromCaseRows as L, runInEvalRuntimeScope as Ln, cacheListItemSchema as Lt, applyDerivedCallAttributes as M, getEvalStartTime as Mn, cacheDebugKeyEntrySchema as Mt, getNestedAttribute as N, incrementEvalOutput as Nn, cacheDebugKeyFileSchema as Nt, extractCacheEntries as O, evalAssert as On, evalChartMetricSchema as Ot, getEvalTitle as P, isInEvalScope as Pn, cacheEntrySchema as Pt, llmCallPricingSchema as Q, traceSpanErrorSchema as Qt, deriveStatusFromChildStatuses as R, runInEvalScope as Rn, cacheModeSchema as Rt, normalizeScoreDef as S, repoFile as Sn, scoreTraceSchema as St, createRunRequestSchema as T, advanceEvalTime as Tn, evalChartBuiltinMetricSchema as Tt, agentEvalsConfigSchema as U, defineEval as Un, serializedCacheSpanSchema as Ut, DEFAULT_API_CALLS_CONFIG as V, setScopeCacheContext as Vn, cacheRecordingSchema as Vt, apiCallMetricFormatSchema as W, getEvalRegistry as Wn, spanCacheOptionsSchema as Wt, llmCallMetricPlacementSchema as X, traceDisplayConfigSchema as Xt, llmCallMetricFormatSchema as Y, traceAttributeDisplaySchema as Yt, llmCallMetricSchema as Z, traceDisplayInputConfigSchema as Zt, loadEvalModule as _, hashCacheKeySync as _n, evalSummarySchema as _t, getLastRunStatuses as a, columnKindSchema as an, buildCaseKey as at, loadConfig as b, serializeCacheRecording as bn, runLogLocationSchema as bt, loadPersistedRunSnapshots as c, numberDisplayOptionsSchema as cn, getCaseRowEvalKey as ct, persistRunState as d, z$1 as dn, caseRowSchema as dt, traceSpanSchema as en, removeDefaultConfigSchema as et, recomputeEvalStatusesInRuns as f, buildTraceTree as fn, discoveryIssueSchema as ft, deriveEvalFreshness as g, hashCacheKey as gn, evalStatsConfigSchema as gt, resolveArtifactPath as h, evalTracer as hn, evalStatItemSchema as ht, generateRunId as i, columnFormatSchema as in, trialSelectionModeSchema as it, extractLlmCalls as j, getEvalCaseInput as jn, evalChartsConfigSchema as jt, extractCacheHits as k, evalLog as kn, evalChartTooltipExtraSchema as kt, nextShortIdFromSnapshots as l, repoFileRefSchema as ln, assertionFailureSchema as lt, runTouchesEval as m, evalSpan as mn, evalStatAggregateSchema as mt, getTargetEvalKeys as n, cellValueSchema as nn, resolveLlmCallsConfig as nt, getLatestRunInfos as o, fileRefSchema as on, buildEvalKey as ot, recomputePersistedCaseStatus as p, captureEvalSpanError as pn, evalFreshnessStatusSchema as pt, apiCallsConfigSchema as q, traceAttributeDisplayInputSchema as qt, getTargetEvals as r, columnDefSchema as rn, runLogsConfigSchema as rt, loadPersistedRunSnapshot as s, jsonCellSchema as sn, getCaseRowCaseKey as st, executeRun as t, traceSpanWarningSchema as tn, resolveApiCallsConfig as tt, persistCaseDetail as u, runArtifactRefSchema as un, caseDetailSchema as ut, parseEvalDiscovery as v, deserializeCacheRecording as vn, runLogEntrySchema as vt, createFsCacheStore as w, EvalAssertionError as wn, evalChartAxisSchema as wt, buildDeclaredColumnDefs as x, serializeCacheValue as xn, runLogPhaseSchema as xt, resolveEvalDefaultConfig as y, deserializeCacheValue as yn, runLogLevelSchema as yt, runManifestSchema as z, runInExistingEvalScope as zn, cacheOperationTypeSchema as zt };
+export { llmCallMetricFormatSchema as $, traceAttributeDisplayPlacementSchema as $t, extractCacheHits as A, advanceEvalTime as An, evalChartBuiltinMetricSchema as At, runManifestSchema as B, mergeEvalOutput as Bn, cacheEntryWithDebugKeySchema as Bt, normalizeScoreDef as C, deserializeCacheRecording as Cn, runLogEntrySchema as Ct, updateManualScoreRequestSchema as D, repoFile as Dn, scoreTraceSchema as Dt, createRunRequestSchema as E, serializeCacheValue as En, runLogPhaseSchema as Et, getEvalTitle as F, getCurrentScope as Fn, evalChartTypeSchema as Ft, apiCallMetricFormatSchema as G, setEvalOutput as Gn, cacheRecordingOpSchema as Gt, DEFAULT_API_CALLS_CONFIG as H, runInEvalRuntimeScope as Hn, cacheListItemSchema as Ht, getEvalDisplayStatus as I, getEvalCaseInput as In, evalChartsConfigSchema as It, apiCallsConfigSchema as J, defineEval as Jn, serializedCacheSpanSchema as Jt, apiCallMetricPlacementSchema as K, setScopeCacheContext as Kn, cacheRecordingSchema as Kt, deriveScopedSummaryFromCases as L, getEvalStartTime as Ln, cacheDebugKeyEntrySchema as Lt, extractLlmCalls as M, configureEvalRunLogs as Mn, evalChartConfigSchema as Mt, applyDerivedCallAttributes as N, evalAssert as Nn, evalChartMetricSchema as Nt, sseEnvelopeSchema as O, evalExpect as On, evalChartAggregateSchema as Ot, getNestedAttribute as P, evalLog as Pn, evalChartTooltipExtraSchema as Pt, evalDeriveConfigSchema as Q, traceAttributeDisplayInputSchema as Qt, deriveStatusFromCaseRows as R, incrementEvalOutput as Rn, cacheDebugKeyFileSchema as Rt, buildDeclaredColumnDefs as S, hashCacheKeySync as Sn, evalSummarySchema as St, createFsCacheStore as T, serializeCacheRecording as Tn, runLogLocationSchema as Tt, DEFAULT_LLM_CALLS_CONFIG as U, runInEvalScope as Un, cacheModeSchema as Ut, runSummarySchema as V, nextEvalId as Vn, cacheFileSchema as Vt, agentEvalsConfigSchema as W, runInExistingEvalScope as Wn, cacheOperationTypeSchema as Wt, evalColumnOverrideSchema as X, traceCacheRefSchema as Xt, defaultConfigKeySchema as Y, getEvalRegistry as Yn, spanCacheOptionsSchema as Yt, evalColumnsSchema as Z, traceAttributeDisplayFormatSchema as Zt, deriveEvalFreshness as _, buildTraceTree as _n, discoveryIssueSchema as _t, getLastRunStatuses as a, traceSpanSchema as an, removeDefaultConfigSchema as at, resolveEvalDefaultConfig as b, evalTracer as bn, evalStatItemSchema as bt, loadPersistedRunSnapshots as c, columnDefSchema as cn, runLogsConfigSchema as ct, persistRunState as d, fileRefSchema as dn, buildEvalKey as dt, traceAttributeDisplaySchema as en, llmCallMetricPlacementSchema as et, recomputeEvalStatusesInRuns as f, jsonCellSchema as fn, getCaseRowCaseKey as ft, resolveArtifactPath as g, z$1 as gn, caseRowSchema as gt, resolveTracePresentation as h, runArtifactRefSchema as hn, caseDetailSchema as ht, generateRunId as i, traceSpanKindSchema as in, llmCallsConfigSchema as it, extractApiCalls as j, appendToEvalOutput as jn, evalChartColorSchema as jt, extractCacheEntries as k, EvalAssertionError as kn, evalChartAxisSchema as kt, nextShortIdFromSnapshots as l, columnFormatSchema as ln, trialSelectionModeSchema as lt, runTouchesEval as m, repoFileRefSchema as mn, assertionFailureSchema as mt, getTargetEvalKeys as n, traceDisplayInputConfigSchema as nn, llmCallPricingRateSchema as nt, getLatestRunInfos as o, traceSpanWarningSchema as on, resolveApiCallsConfig as ot, recomputePersistedCaseStatus as p, numberDisplayOptionsSchema as pn, getCaseRowEvalKey as pt, apiCallMetricSchema as q, startEvalBackgroundJob as qn, cacheStatusSchema as qt, getTargetEvals as r, traceSpanErrorSchema as rn, llmCallPricingSchema as rt, loadPersistedRunSnapshot as s, cellValueSchema as sn, resolveLlmCallsConfig as st, executeRun as t, traceDisplayConfigSchema as tn, llmCallMetricSchema as tt, persistCaseDetail as u, columnKindSchema as un, buildCaseKey as ut, loadEvalModule as v, captureEvalSpanError as vn, evalFreshnessStatusSchema as vt, validateCharts as w, deserializeCacheValue as wn, runLogLevelSchema as wt, loadConfig as x, hashCacheKey as xn, evalStatsConfigSchema as xt, parseEvalDiscovery as y, evalSpan as yn, evalStatAggregateSchema as yt, deriveStatusFromChildStatuses as z, isInEvalScope as zn, cacheEntrySchema as zt };