how it’s built
The receipts.
Anyone can claim their sites are fast. This page is the measured proof for the one you’re on — the same checks run against every site that leaves the workshop.
This page, measured
| Check | This site | Passing bar |
|---|---|---|
| Lighthouse performance (mobile, simulated 4G) | 100 | ≥ 95 |
| Lighthouse accessibility | 100 | ≥ 95 |
| Lighthouse best practices | 100 | ≥ 95 |
| Lighthouse SEO | 100 | ≥ 90 |
| Largest contentful paint | 1.5s | < 2.5s |
| Layout shift when fonts load | 0.000 | < 0.02 |
| JavaScript shipped by this page | 0 bytes | 0 |
Measured with Lighthouse’s mobile profile — a mid-range phone on a slow 4G connection, not a developer laptop on fibre.
What it runs on
- Framework
- Astro — static HTML out, JavaScript only where a demo needs it
- Interactive demos
- Preact islands — the React component model at a tenth of the weight, loaded per-demo so no page pays for another’s code
- Styling
- Hand-written CSS on a token system. No UI framework.
- Type
- Bricolage Grotesque, Instrument Sans, Spline Sans Mono — self-hosted, no font CDN
- Hosting
- Cloudflare Pages, static at the edge
- Tracking
- None. No analytics scripts, no pixels, no cookie banner needed.
How it stays this way
Every change to this site has to clear an automated gate before it ships — at last count, 121 unit checks and 332 browser checks. Unit tests cover the design tokens (every colour pairing is checked against WCAG contrast maths) and all the demo arithmetic, to the cent. Browser tests walk every flow with a keyboard, scan every page with an accessibility engine in light and dark, compare pixel-level screenshots at four screen sizes, and hold each page to a JavaScript byte-budget. The Lighthouse bars above run three times and are judged on the median.
The test suite itself is tested: every layer has been deliberately broken to prove it actually fails. A gate that can’t go red is decoration.