Merge branch 'trust-689-allow-running-checks-to-expand-in-the-ui' into trust-690-allow-adding-checks-and-applying-suggestions-during-active

packages/cli/src/modules/instance-ai/eval/__tests__/agent-execution.service.test.ts

@@ -29,6 +29,7 @@ import {
 import { generateAgentScenarioSeed } from '../agent-scenario-seed';
 import { createMcpMockFetch } from '../mcp-mock-fetch';
 import { createLlmMockHandler } from '../mock-handler';
+import { createWebSearchMock } from '../web-search-mock';
 
 // The service resolves the agents-module surface lazily; top-level vi.mock
 // keeps those specifiers from loading the real module graph (and avoids the
@@ -54,6 +55,7 @@ vi.mock('@/utils/ai-proxy-fetch', () => ({ createAiProxyFetch: vi.fn(() => vi.fn
 vi.mock('../agent-scenario-seed', () => ({ generateAgentScenarioSeed: vi.fn() }));
 vi.mock('../mcp-mock-fetch', () => ({ createMcpMockFetch: vi.fn(() => vi.fn()) }));
 vi.mock('../mock-handler', () => ({ createLlmMockHandler: vi.fn() }));
+vi.mock('../web-search-mock', () => ({ createWebSearchMock: vi.fn(() => vi.fn()) }));
 
 const logger = mock<Logger>();
 const settingsService = mock<AgentsSettingsService>();
@@ -290,6 +292,113 @@ describe('EvalAgentExecutionService.executeWithLlmMock', () => {
 		},
 	);
 
+	it('records search progress only for the matching active call and ignores late endings', async () => {
+		const handlers = new Map<AgentEvent, AgentEventHandler>();
+		const onProgress = vi.fn().mockResolvedValue(undefined);
+		const search = { query: 'shipping' };
+		const searchResult = { query: 'shipping', results: [] };
+		reconstructFromAgentEntity.mockResolvedValue({
+			agent: {
+				on: (event: AgentEvent, handler: AgentEventHandler) => handlers.set(event, handler),
+				close: vi.fn(),
+				generate: vi.fn(async () => {
+					for (const toolName of ['Slack_Tool', 'web_search']) {
+						handlers.get(AgentEvent.ToolExecutionStart)?.({
+							type: AgentEvent.ToolExecutionStart,
+							toolName,
+							toolCallId: toolName,
+							args: search,
+						});
+					}
+					const { onSearch } = vi.mocked(createWebSearchMock).mock.calls[0][0];
+					onSearch(search, searchResult);
+					handlers.get(AgentEvent.ToolExecutionEnd)?.({
+						type: AgentEvent.ToolExecutionEnd,
+						toolName: 'web_search',
+						toolCallId: 'web_search',
+						result: searchResult,
+						isError: false,
+					});
+					// A delegated search has no active parent call to update.
+					onSearch({ query: 'returns' }, { query: 'returns', results: [] });
+					// A runtime can report completion without a prior start event.
+					handlers.get(AgentEvent.ToolExecutionEnd)?.({
+						type: AgentEvent.ToolExecutionEnd,
+						toolName: 'lookup',
+						toolCallId: 'orphan',
+						result: 'found',
+						isError: false,
+					});
+					return makeGenerateResult({
+						toolCalls: [{ tool: 'lookup', input: {}, output: 'found' }],
+					});
+				}),
+			},
+		});
+		const result = await buildService().executeWithLlmMock('agent-1', user, request, 'Search', {
+			onProgress,
+		});
+		expect(onProgress.mock.calls[2][0]).toEqual([
+			expect.objectContaining({ tool: 'Slack_Tool', mocked: false, interceptedRequests: [] }),
+			expect.objectContaining({
+				tool: 'web_search',
+				status: 'running',
+				mocked: true,
+				interceptedRequests: [expect.objectContaining({ requestBody: search })],
+			}),
+		]);
+		expect(onProgress).toHaveBeenCalledTimes(6);
+		expect(
+			result.toolCalls.find((call) => call.tool === 'web_search')?.interceptedRequests,
+		).toHaveLength(2);
+		expect(result.toolCalls.find((call) => call.toolCallId === 'orphan')).toMatchObject({
+			kind: 'other',
+			output: 'found',
+		});
+		handlers.get(AgentEvent.ToolExecutionEnd)?.({
+			type: AgentEvent.ToolExecutionEnd,
+			toolName: 'lookup',
+			toolCallId: 'late',
+			result: 'ignored',
+			isError: false,
+		});
+		await Promise.resolve();
+		expect(onProgress).toHaveBeenCalledTimes(6);
+	});
+
+	it('classifies delegated MCP records from their intercepted request metadata', async () => {
+		reconstructFromAgentEntity.mockResolvedValue({
+			agent: {
+				close: vi.fn(),
+				generate: vi.fn(async () => {
+					vi.mocked(createMcpMockFetch).mock.calls[0][0].onToolCall({
+						serverName: 'child',
+						toolName: 'lookup',
+						args: {},
+						result: { text: 'found', isError: false },
+					});
+					const instrumentation = reconstructFromAgentEntity.mock.calls[0][5] as {
+						onMcpToolCallSettled: (event: {
+							serverName: string;
+							toolName: string;
+							success: boolean;
+						}) => void;
+					};
+					instrumentation.onMcpToolCallSettled({
+						serverName: 'child',
+						toolName: 'lookup',
+						success: true,
+					});
+					return makeGenerateResult({ toolCalls: [] });
+				}),
+			},
+		});
+		const result = await buildService().executeWithLlmMock('agent-1', user, request);
+		expect(result.toolCalls).toEqual([
+			expect.objectContaining({ tool: 'child_lookup', kind: 'mcp', mocked: true }),
+		]);
+	});
+
 	it('reports tool progress before generation finishes and flushes writes before returning', async () => {
 		const handlers = new Map<AgentEvent, AgentEventHandler>();
 		const onProgress = vi.fn().mockResolvedValue(undefined);

packages/cli/src/modules/instance-ai/eval/agent-execution.service.ts

@@ -296,7 +296,9 @@ export class EvalAgentExecutionService {
 		});
 
 		const toolLedger: ToolLedger = new Map();
-		let onLedgerChange: ((toolName: string) => void) | undefined;
+		let onLedgerChange:
+			| ((toolName: string, entries: InstanceAiEvalInterceptedRequest[]) => void)
+			| undefined;
 		const credentialHelpers: EvalMockedCredentialsHelper[] = [];
 		const recorder = createAgentModelTurnRecorder(
 			createAiProxyFetch(this.outboundHttp),
@@ -355,7 +357,7 @@ export class EvalAgentExecutionService {
 				requestBody: call.args,
 				mockResponse: call.result,
 			});
-			onLedgerChange?.(key);
+			onLedgerChange?.(key, entries);
 		};
 
 		// Fallback web_search mock — always created (sub-agents may enable web
@@ -379,7 +381,7 @@ export class EvalAgentExecutionService {
 					requestBody: args,
 					mockResponse: result,
 				});
-				onLedgerChange?.('web_search');
+				onLedgerChange?.('web_search', entries);
 			},
 		});
 
@@ -434,7 +436,7 @@ export class EvalAgentExecutionService {
 							mockHandler,
 							toolContext.toolName,
 							toolLedger,
-							() => onLedgerChange?.(toolContext.toolName),
+							(entries) => onLedgerChange?.(toolContext.toolName, entries),
 						);
 					},
 				},
@@ -451,24 +453,24 @@ export class EvalAgentExecutionService {
 		const progressCalls = new Map<string, InstanceAiEvalAgentToolCallRecord>();
 		let progressWrites = Promise.resolve();
 		let progressClosed = false;
-		const publishProgress = () => {
-			if (progressClosed || !hooks?.onProgress) return;
+		const onProgress = hooks?.onProgress;
+		const publishProgress = (
+			writeProgress: (calls: InstanceAiEvalAgentToolCallRecord[]) => Promise<void>,
+		) => {
 			const calls = [...progressCalls.values()].map((call) => ({ ...call }));
 			progressWrites = progressWrites
-				.then(async () => await hooks.onProgress?.(calls))
+				.then(async () => await writeProgress(calls))
 				.catch((error: unknown) => {
 					this.logger.warn('[EvalAgentMock] Could not record tool progress', { error });
 				});
 		};
-		if (hooks?.onProgress) {
+		if (onProgress) {
 			// A send-and-wait tool records its mocked send before the tool ends.
 			// Notify synchronously; the existing queue owns persistence and ordering.
-			onLedgerChange = (toolName) => {
-				if (progressClosed) return;
+			onLedgerChange = (toolName, interceptedRequests) => {
 				let changed = false;
 				for (const [id, call] of progressCalls) {
 					if (call.tool !== toolName || call.status !== 'running') continue;
-					const interceptedRequests = toolLedger.get(toolName) ?? [];
 					progressCalls.set(id, {
 						...call,
 						mocked: interceptedRequests.length > 0,
@@ -476,7 +478,7 @@ export class EvalAgentExecutionService {
 					});
 					changed = true;
 				}
-				if (changed) publishProgress();
+				if (changed) publishProgress(onProgress);
 			};
 			agent.on(AgentEvent.ToolExecutionStart, (event) => {
 				if (progressClosed || event.type !== AgentEvent.ToolExecutionStart) return;
@@ -489,7 +491,7 @@ export class EvalAgentExecutionService {
 					mocked: false,
 					interceptedRequests: [],
 				});
-				publishProgress();
+				publishProgress(onProgress);
 			});
 			agent.on(AgentEvent.ToolExecutionEnd, (event) => {
 				if (progressClosed || event.type !== AgentEvent.ToolExecutionEnd) return;
@@ -506,7 +508,7 @@ export class EvalAgentExecutionService {
 					mocked: interceptedRequests.length > 0,
 					interceptedRequests: [...interceptedRequests],
 				});
-				publishProgress();
+				publishProgress(onProgress);
 			});
 		}
 
@@ -637,7 +639,6 @@ export class EvalAgentExecutionService {
 					skippedFeatures,
 					{
 						modelTurns: recorder.turns,
-						toolLedger,
 						credentialHelpers,
 						toolCalls: terminalToolCalls(),
 					},
@@ -645,7 +646,6 @@ export class EvalAgentExecutionService {
 			}
 			return this.errorResult(`Agent run failed: ${message}`, seed, skippedFeatures, {
 				modelTurns: recorder.turns,
-				toolLedger,
 				credentialHelpers,
 				toolCalls: terminalToolCalls(),
 			});
@@ -655,7 +655,7 @@ export class EvalAgentExecutionService {
 				if (call.status === 'running')
 					progressCalls.set(id, { ...call, status: 'error', error: 'execution_stopped' });
 			}
-			if (progressCalls.size > 0) publishProgress();
+			if (onProgress && progressCalls.size > 0) publishProgress(onProgress);
 			progressClosed = true;
 			await progressWrites;
 			await recorder.flush();
@@ -706,7 +706,7 @@ export class EvalAgentExecutionService {
 		mockHandler: EvalLlmMockHandler,
 		toolName: string,
 		ledger: ToolLedger,
-		onRecorded?: () => void,
+		onRecorded?: (entries: InstanceAiEvalInterceptedRequest[]) => void,
 	): EvalLlmMockHandler {
 		return async (requestOptions, node) => {
 			const response = await mockHandler(requestOptions, node);
@@ -722,7 +722,7 @@ export class EvalAgentExecutionService {
 				requestBody: requestOptions.body,
 				mockResponse: snapshotLedgerBody(response?.body),
 			});
-			onRecorded?.();
+			onRecorded?.(entries);
 			this.logger.debug(
 				`[EvalAgentMock] Intercepted ${requestOptions.method ?? 'GET'} ${requestOptions.url} from tool "${toolName}" (${node.type})`,
 			);
@@ -736,9 +736,8 @@ export class EvalAgentExecutionService {
 		skippedFeatures: InstanceAiEvalAgentSkippedFeature[] = [],
 		partial?: {
 			modelTurns: InstanceAiEvalAgentExecutionResult['modelTurns'];
-			toolLedger: ToolLedger;
 			credentialHelpers: EvalMockedCredentialsHelper[];
-			toolCalls?: InstanceAiEvalAgentToolCallRecord[];
+			toolCalls: InstanceAiEvalAgentToolCallRecord[];
 		},
 	): InstanceAiEvalAgentExecutionResult {
 		this.logger.error(`[EvalAgentMock] ${message}`);
@@ -747,16 +746,7 @@ export class EvalAgentExecutionService {
 			success: false,
 			errors: [message],
 			finalText: '',
-			toolCalls:
-				partial?.toolCalls ??
-				(partial
-					? [...partial.toolLedger.entries()].map(([tool, interceptedRequests]) => ({
-							tool,
-							kind: salvagedToolKind(interceptedRequests),
-							mocked: interceptedRequests.length > 0,
-							interceptedRequests,
-						}))
-					: []),
+			toolCalls: partial?.toolCalls ?? [],
 			modelTurns: partial?.modelTurns ?? [],
 			seed: seed ?? { openingMessage: '', globalContext: '', toolHints: {}, warnings: [] },
 			skippedFeatures,
@@ -831,22 +821,6 @@ export function pruneConfigForEval(original: AgentJsonConfig): {
  * tools, `sanitizeToolName` for workflow tools, the stored descriptor name
  * for custom tools.
  */
-/**
- * Best-effort kind for a salvaged ledger entry on the failure path, where the
- * summary map is gone: MCP and web-search entries are self-describing via
- * nodeType; anything else with a nodeType came through the node/workflow
- * engine seam.
- */
-function salvagedToolKind(
-	interceptedRequests: InstanceAiEvalInterceptedRequest[],
-): InstanceAiEvalAgentToolCallRecord['kind'] {
-	const nodeType = interceptedRequests.find((request) => request.nodeType)?.nodeType;
-	if (!nodeType) return 'other';
-	if (nodeType.startsWith('mcp:')) return 'mcp';
-	if (nodeType === 'web-search:fallback') return 'other';
-	return 'node';
-}
-
 export function summarizeTools(
 	config: AgentJsonConfig,
 	customTools: NonNullable<AgentEntity['tools']>,

packages/frontend/editor-ui/src/features/agents/__tests__/AgentEvalTryRow.test.ts

@@ -30,6 +30,17 @@ describe('AgentEvalTryRow', () => {
 	});
 
 	describe('view="complete"', () => {
+		it('announces both disclosure states on the complete row', async () => {
+			const user = userEvent.setup();
+			const { getByTestId } = renderComponent({ props: { testId: 'row' } });
+			const toggle = getByTestId('row-toggle');
+			expect(toggle).toHaveAttribute('aria-expanded', 'false');
+			await user.click(toggle);
+			expect(toggle).toHaveAttribute('aria-expanded', 'true');
+			await user.click(toggle);
+			expect(toggle).toHaveAttribute('aria-expanded', 'false');
+		});
+
 		it.each([
 			['pass', 'Passes'],
 			['strong', 'Passes'],

packages/frontend/editor-ui/src/features/agents/__tests__/agentEvals.store.test.ts

@@ -1722,6 +1722,79 @@ describe('useAgentEvalsStore', () => {
 	});
 
 	describe('rerunResult', () => {
+		it('ignores a timer task already queued when a rerun finishes', async () => {
+			vi.useFakeTimers();
+			mockRun({ results: [result('c1')], count: 1, ratings: [] });
+			const store = useAgentEvalsStore();
+			await store.openRun(PROJECT_ID, AGENT_ID, RUN_ID);
+			let finish!: (value: AgentEvalResultRecord) => void;
+			rerunResult.mockImplementationOnce(
+				async () =>
+					await new Promise((resolve) => {
+						finish = resolve;
+					}),
+			);
+			const timer = vi.spyOn(globalThis, 'setTimeout');
+			const pending = store.rerunResult(PROJECT_ID, AGENT_ID, 'c1');
+			const queued = timer.mock.calls.find(([, delay]) => delay === 5000)?.[0];
+			expect(typeof queued).toBe('function');
+			finish(result('c1'));
+			await pending;
+			getRunDetail.mockClear();
+			if (typeof queued !== 'function') throw new Error('Missing progress timer');
+			await queued();
+			expect(getRunDetail).not.toHaveBeenCalled();
+			expect(store.getReview(RUN_ID).results[0]).toEqual(result('c1'));
+			expect(vi.getTimerCount()).toBe(0);
+			timer.mockRestore();
+		});
+
+		it.each([false, true])('handles an uncached rerun (request fails: %s)', async (fails) => {
+			const store = useAgentEvalsStore();
+			if (fails) rerunResult.mockRejectedValueOnce(new Error('Unavailable'));
+			else rerunResult.mockResolvedValueOnce(result('uncached'));
+			const pending = store.rerunResult(PROJECT_ID, AGENT_ID, 'uncached');
+			if (fails) await expect(pending).rejects.toThrow('Unavailable');
+			else expect(await pending).toEqual(result('uncached'));
+			expect(getRunDetail).not.toHaveBeenCalled();
+		});
+
+		it('preserves a newer attempt when a later refresh returns an older runAt', async () => {
+			const newer = { ...result('c1'), runAt: '2026-01-01T01:00:00.000Z' };
+			mockRun({ results: [newer], count: 1, ratings: [] });
+			const store = useAgentEvalsStore();
+			await store.openRun(PROJECT_ID, AGENT_ID, RUN_ID);
+			getRunDetail.mockResolvedValueOnce(runDetail([result('c1')], 1));
+			await store.openRun(PROJECT_ID, AGENT_ID, RUN_ID);
+			expect(store.getReview(RUN_ID).results[0]).toEqual(newer);
+		});
+
+		it('pauses normal progress reads while the document is hidden', async () => {
+			vi.useFakeTimers();
+			const visibility = vi.spyOn(document, 'hidden', 'get').mockReturnValue(true);
+			mockRun({ results: [result('c1')], count: 1, ratings: [] });
+			const store = useAgentEvalsStore();
+			await store.openRun(PROJECT_ID, AGENT_ID, RUN_ID);
+			let finish!: (value: AgentEvalResultRecord) => void;
+			rerunResult.mockImplementationOnce(
+				async () =>
+					await new Promise((resolve) => {
+						finish = resolve;
+					}),
+			);
+			const pending = store.rerunResult(PROJECT_ID, AGENT_ID, 'c1');
+			getRunDetail.mockClear();
+			await vi.advanceTimersByTimeAsync(5000);
+			expect(getRunDetail).not.toHaveBeenCalled();
+			visibility.mockReturnValue(false);
+			await vi.advanceTimersByTimeAsync(5000);
+			expect(getRunDetail).toHaveBeenCalledOnce();
+			finish(result('c1'));
+			await pending;
+			expect(vi.getTimerCount()).toBe(0);
+			visibility.mockRestore();
+		});
+
 		it('patches the cached result to running before the request lands, then to the real response', async () => {
 			mockRun({ results: [{ ...result('c1'), status: 'error' }], count: 1, ratings: [] });
 			const store = useAgentEvalsStore();

packages/frontend/editor-ui/src/features/agents/agentEvals.store.ts

@@ -1182,14 +1182,13 @@ export const useAgentEvalsStore = defineStore(STORES.AGENT_EVALS, () => {
 		const ownsProgress = () => rerunningResults.get(result.id) === pending;
 		revisionByResult.set(result.id, ++resultRevision);
 		let timer: ReturnType<typeof setTimeout> | undefined;
-		let stopped = false;
 		let errors = 0;
 		const deadline = Date.now() + RUN_POLL_TIMEOUT_MS;
 		const confirmationDeadline = Date.now() + RESULT_CONFIRMATION_TIMEOUT_MS;
 		const schedule = () => {
-			if (stopped || !ownsProgress()) return;
+			if (!ownsProgress()) return;
 			timer = setTimeout(async () => {
-				if (stopped || !ownsProgress()) return;
+				if (!ownsProgress()) return;
 				const current = findCachedResult(result.id)?.result;
 				const confirmationExpired =
 					recovering && current?.runAt === result.runAt && Date.now() >= confirmationDeadline;
@@ -1217,8 +1216,6 @@ export const useAgentEvalsStore = defineStore(STORES.AGENT_EVALS, () => {
 			}, RUN_POLL_INTERVAL_MS);
 		};
 		const stop = (exhausted = false) => {
-			if (stopped) return;
-			stopped = true;
 			clearTimeout(timer);
 			if (!ownsProgress()) return;
 			const current = findCachedResult(result.id)?.result;

packages/frontend/editor-ui/src/features/agents/components/AgentEvalTryRow.vue

@@ -240,7 +240,7 @@ watch(
 				type="button"
 				:class="$style.expandToggle"
 				:data-test-id="testId && `${testId}-toggle`"
-				:aria-expanded="view === 'complete' ? expanded : undefined"
+				:aria-expanded="expanded"
 				:aria-label="i18n.baseText('instanceAi.testAgentPreview.conversation')"
 			>
 				<N8nIcon :icon="expanded ? 'chevron-up' : 'chevron-down'" size="small" />

packages/frontend/editor-ui/src/features/agents/utils/__tests__/agent-eval-tool-calls.test.ts

@@ -130,3 +130,9 @@ it('filters malformed intercepted requests without discarding the tool call', ()
 	});
 	expect(call).toMatchObject({ mocked: false, interceptedRequests: [] });
 });
+
+it('renders a recorded error status without requiring a diagnostic string', () => {
+	const [call] = toDisplayToolCalls({ calls: [{ tool: 'lookup', status: 'error' }] });
+	expect(call.state).toBe(TOOL_CALL_STATE.ERROR);
+	expect(call.output).toBeUndefined();
+});

packages/frontend/editor-ui/src/features/ai/instanceAi/__tests__/InstanceAiTestAgentPreviewPanel.test.ts

@@ -249,6 +249,24 @@ describe('InstanceAiTestAgentPreviewPanel', () => {
 		expect(container.querySelector('a[href="https://example.com"]')).toBeInTheDocument();
 	});
 
+	it('omits the seed example when the mocked preview returns no answer', async () => {
+		const store = useAgentEvalsStore();
+		mockPreviewRun(store, { response: '' });
+		const generateDraftCases = vi.spyOn(store, 'generateDraftCases').mockResolvedValue({
+			cases: [{ input: 'Another request', whatToCheck: 'names the ticket', scenario: 'Vague' }],
+		});
+		const user = userEvent.setup();
+		const { findByTestId } = renderComponent();
+
+		await user.click(await findByTestId('instance-ai-test-agent-preview-check-harder'));
+
+		expect(await findByTestId('instance-ai-test-agent-examples-check-agent')).toBeInTheDocument();
+		expect(generateDraftCases).toHaveBeenCalledWith('project-1', 'agent-1', {
+			count: 10,
+			save: false,
+		});
+	});
+
 	it('renders the answer inside a fixed-height scrollable container regardless of length', async () => {
 		const store = useAgentEvalsStore();
 		mockPreviewRun(store, {

packages/frontend/editor-ui/src/features/ai/instanceAi/components/InstanceAiTestAgentPreviewPanel.vue

@@ -56,8 +56,8 @@ const phase = ref<Phase>('generating-preview');
 // from `previewRun`'s response, which already ran the agent and returned the
 // final text. No run id, no polling: unlike the suite below, this never
 // touches a Data Table, dataset, or eval-run row at all.
-const previewRequest = ref<string | null>(null);
-const previewAnswer = ref<string | null>(null);
+const previewInput = ref('');
+const previewOutput = ref('');
 // Generated with `save: false` — a preview, not yet persisted. Nothing backs
 // these rows until "Check your agent" commits them, so dismissing or
 // refreshing mid-preview leaves no orphaned dataset behind.
@@ -87,9 +87,6 @@ const previewVerdict = ref<AgentEvalVerdict | null>(null);
 // Whether the full example conversation is shown under "What we found".
 const conversationExpanded = ref(false);
 
-const previewInput = computed(() => previewRequest.value ?? '');
-const previewOutput = computed(() => previewAnswer.value ?? '');
-
 // Nothing failed when the judge passed the answer, or when the case had no rule to
 // grade it against (`skipped`) — the card still shows, as a pass. A judge error is
 // different: the answer was never graded, so it must not read as "passed".
@@ -189,8 +186,8 @@ async function runGeneratedPreview() {
 		previewScenario.value = result.scenario || null;
 		previewWhatToCheck.value = result.whatToCheck;
 		previewVerdict.value = result.verdict;
-		previewRequest.value = result.input;
-		previewAnswer.value = result.response;
+		previewInput.value = result.input;
+		previewOutput.value = result.response;
 		conversationExpanded.value = false;
 		phase.value = 'awaiting-confirmation';
 	} catch (error) {
@@ -443,8 +440,8 @@ async function onApplySuggestion() {
 		);
 		return;
 	}
-	previewRequest.value = preview.input;
-	previewAnswer.value = preview.response;
+	previewInput.value = preview.input;
+	previewOutput.value = preview.response;
 	previewVerdict.value = preview.verdict;
 	conversationExpanded.value = false;
 }
@@ -538,7 +535,7 @@ function onDontCreateEvals() {
 					<EvalInitialSample
 						v-if="conversationExpanded"
 						:preview-input="previewInput"
-						:preview-output="previewOutput ?? ''"
+						:preview-output="previewOutput"
 						hide-banner
 					/>
 				</div>
@@ -610,7 +607,7 @@ function onDontCreateEvals() {
 		<template v-else-if="phase === 'suite-ready'">
 			<InstanceAiTestAgentExamplesPanel
 				:preview-input="previewInput"
-				:preview-output="previewOutput ?? ''"
+				:preview-output="previewOutput"
 				:preview-scenario="previewScenario"
 				:preview-status="previewStatus"
 				:preview-reasoning="previewReasoning"