· ~5 min
Pre-launch checklist: tech items and discovery files
lang, canonical, compression, security headers, and optional llms.txt. Retest the same URL after changes.
After the copy is ready, server and template items remain. They total 15 plus observation rows. Observation items enter the issue list but do not own a share of the base 100, so they do not double-count scored checks.
Technical health — 15
- Only one html / head / DOCTYPE set. Themes that nest a full document in the body fail outright.
- html needs lang; Chinese pages should use zh-CN.
- A canonical pointing at this page.
- Open Graph and Twitter Card tags totaling at least four.
- gzip or brotli on.
- HTTP/2 on. Still on HTTP/1.1 loses those 2 points.
Observation items to check together
- One clear topical h1 per page.
- Semantic tags such as main, article, footer, time.
- No meta tags left in body.
- Security headers: at least nosniff, Referrer-Policy, Permissions-Policy, plus CSP or frame restriction via X-Frame-Options—three of the set passes observation.
- Images with width and height when possible to reduce layout shift for agents. Hint only; not in the total.
Discovery files — up to +5
Optional root llms.txt describing what the site is and which pages are good to cite. Optionally /.well-known/ai.txt and /ai/summary.json, faq.json, service.json. Missing them does not block a base-score pass.
How to accept
Retest the same URL you changed. A higher homepage score does not mean article pages are fixed. CLI and the site share the same judges; the web UI uses a 12-second fetch timeout so one audit does not hang. If confidence is low, rerun before changing templates.