Test fixture
The model receives the prompt (and optional system message). The run uses scorer singlefile_html_cdn_allowed_judged with the JSON configuration below. Pass/fail and partial credit are determined entirely by that scorer against the model output; no human grading.
Create a single-file HTML document that renders an animated night-time thunderstorm over a stylized city skyline on a full-window <canvas>. Requirements: (1) Return exactly ONE complete runnable HTML document and NOTHING else — do not wrap it in markdown fences or add prose. (2) Start with <!DOCTYPE html> and contain <html>, <head>, <body>, inline <style>, and inline <script> using requestAnimationFrame. (3) Implement dark layered parallax skyline (buildings with slightly lit windows), falling rain using particles, drifting clouds, and occasional lightning flashes with a short screen-wide white flash and a random jagged bolt path. (4) Use only inline CSS and inline JS; no build step. (5) You MAY optionally load libraries from https://cdn.jsdelivr.net, https://unpkg.com, or https://cdnjs.cloudflare.com via CDN <script src=...> or <link href=...>. (6) Make it visually polished: smooth animation, proper easing, subtle color palette, responsive to window resize. (7) No TODOs, no placeholders, no lorem ipsum. The file must be directly openable in a browser with no setup.
{
"min_chars": 1800,
"required_contains": [
"<canvas",
"requestAnimationFrame"
],
"requires_one_of": [
"lightning",
"bolt",
"flash"
],
"forbidden_contains": [
"```",
"TODO",
"lorem ipsum",
"your content here"
],
"allowed_cdn_hosts": [
"cdn.jsdelivr.net",
"unpkg.com",
"cdnjs.cloudflare.com"
],
"playwright_enabled": true,
"judge_enabled": true,
"required_selectors": [
"canvas"
],
"requires_canvas": true,
"viewport_width": 1440,
"viewport_height": 900,
"render_wait_ms": 2000
}temperature
0.3
max_tokens
8000
timeout (s)
180
type
scored
file
thunderstorm_over_city.json