Alerts that apply at once with a short confirmation, and a summary email whose frequency and product-news consent are saved with a button.
What the agent reads, word for word.
Build the notification settings page of a business account. Show who the notifications go to: the company name and the email address.
Alerts: each alert in the data can be turned on or off, with its description under its name. The change applies at once, and a short message that goes away by itself confirms it, for example "Payment failed alerts on".
Summary email: the user chooses how often to receive it, one of the options in the data, and whether to receive product news ("Send me product news", off unless the data says otherwise). These two are saved together with a Save button, confirmed by the same kind of short message.
The data is in `./fixture.json`.
The page must show these texts, or the run is not done. A good answer uses these components.
By design-system violations.
| Run | Design system | Cost |
|---|---|---|
| beam-free at genui-spec | 0 | $0.0000 |
| nemotron-3.5-lightning-free at genui-spec | 0 | $0.0000 |
| nemotron-3.5-lightning-free at genui-spec | 0 | $0.0000 |
Mean design-system violations per run on this task.
| Context | Mean | nemotron-3.5-lightning-free | beam-free |
|---|---|---|---|
| none File tools only | 24.2 | 12.7 | 35.7 |
| storybook-mcp Storybook's own MCP |
Written by hand, graded with zero violations, rendered by the browser check in every build. The agent never sees it.
"use client"; import { Button, Card, CardBody, CardDescription, CardFooter, CardHeader, CardTitle, Checkbox, RadioGroup,| Run | Design system | Cost |
|---|
| beam-free at none | 59 | $0.0000 |
| beam-free at none | 48 | $0.0000 |
| nemotron-3.5-lightning-free at none | 25 | $0.0000 |
| 10 |
| 0.3 |
| ds-mcp Our design-system MCP | 1 | 1.7 | 0.3 |
| ds-mcp-check The MCP and a checker | 1.7 | 3.3 | 0 |
| ds-mcp-check-full A checker with compiler and linter | 0 | 0 | 0 |
| genui-spec A view from the catalog, six tasks | 0 | 0 | 0 |