HomeHow it is tested
Tested, not asserted
Every claim on this site is a test somewhere. Here is what runs, and two tools that do the same maths the parts do — so you can check the claims on this page rather than take them.
What runs
On all 137 parts, on every output, on every commit that touches them.
- axe, on every variant
- WCAG 2.0, 2.1 and 2.2 at A and AA, run with the component closed, open, mid-error and after whatever it does. Animations are waited out first, so a half-faded colour never passes as a real one.
- Keyboard only
- Open it, move around it, pick something, close it — with no pointer at all. Focus has to come back where it started.
- Both outputs, the same tests
- The React + Tailwind file and the plain HTML, CSS and JavaScript run the same spec. A fix to one that is not a fix to the other fails.
- Three engines
- Chromium, WebKit — Safari's engine — and an emulated iPhone 15. Safari is where the focus bugs live: it does not focus a button when it is clicked, and that alone has broken four parts here.
- In the real editor
- Every part is also opened in the editor on this site, in both outputs, inside the sandboxed preview frame, and any console error fails the run. That is what catches storage throwing where it is not allowed.
- At three widths
- 320, 768 and 1280, checking target sizes, whether anything overflows, and that nothing needs a sideways scroll.
- Served exactly as exported
- The registry route is asked for the file with the options applied, and the test reads what comes back — so what is tested is what you would install.
Count the tab stops yourself
The number of times Tab stops inside a component is the difference between a part you can use and one you give up on. This reads the real order out of the page and numbers it.
Three menus, nine items — and one stop for the whole bar. The arrows move inside it.
Pick a colour that fails
Every part takes an accent colour, and every part corrects it before using it as text. This is that correction, running — iOS blue on white is 3.9:1, which is how a system colour ends up being a WCAG failure.
Pick anything, including something that fails. Every part corrects a colour before using it as text, so what you choose is never what gets printed.
| Where | Sample | Contrast | 4.5:1 |
|---|---|---|---|
| As you picked it, on white | 1.9:1 | Fails | |
| Corrected, on white | 4.9:1 | Passes | |
| As you picked it, on ink | 9.6:1 | Passes | |
| Corrected, on ink | 9.6:1 | Passes |
On its own that colour fails on white. The parts print #8a6b2c on white and #e6b24a on ink instead.
What is not tested here
Automated checks find maybe a third of what is wrong with a page. They cannot tell you whether a label makes sense, whether an order is logical, or how a part sounds read aloud. Every part ships with a manual checklist for exactly that, and a real screen-reader pass on a real iPhone stays on it — emulation is not a device.