Accessibility

The standard we hold ourselves to

WCAG 2.2 Level AA is the floor. Every page template is checked automatically on every build, and the checks run in CI — a change that introduces a violation fails the build rather than reaching this site.

What has been tested

• Automated rule checks — axe across every page template, in both the dark and light themes, at three viewport widths. The standing result is zero violations.
• Keyboard-only traversal — five primary surfaces walked stop by stop, plus six primary flows driven end to end without a mouse. Focus stays visible, order is logical, there are no traps, and Shift+Tab retraces the path exactly.
• Screen-reader semantics — an assistive-technology engine walks the opening of each of five surfaces in CI and asserts what it hears: landmark and heading announcements, traversal order, and that the registry's result count is genuinely announced when you filter, not merely updated on screen.
• Text resizing — pages are checked at 200% browser text size for horizontal scrolling and content loss, in addition to the narrow-viewport checks.
• Reduced motion — animations are asserted to stop when your system asks for reduced motion, with controls proving the assertion can actually fail.
• The network graph — the diagram has a keyboard- and screen-reader-navigable twin that lists every vault and every relationship between them, in both directions, with the kind of each relationship named. It is asserted to stay equivalent to the diagram.

Known limitations

These are real, current, and measured. Where we have chosen not to fix something yet, the reason is stated rather than left as an absence.

• Text inside diagrams is too small. Our own floor is 12 px of rendered text; on narrow screens the labels inside several diagrams fall well below it — as small as 3.5 px on the homepage graphic at a 320 px width. Measured across every diagram, width and theme, 398 of 510 rendered labels are under the floor. This is page illustrations, not body copy, and every diagram has a text alternative — but it is a genuine legibility failure and it is not fixed. A build gate now prevents it getting worse while the redesign work is scheduled.
• Automated checking does not cover all of WCAG 2.2. Our tooling adds exactly one new rule for 2.2 — target size. Four further 2.2 criteria cannot be checked here because the interactions they govern (dragging, redundant entry, accessible authentication) do not exist on this site; that is true today and stops being true the moment one is added. Focus obscuring is covered by our own keyboard checks rather than by the rule engine.
• No human screen-reader session has been run. Our automated checks use a screen-reader engine, which can prove the right information is exposed — it cannot tell us whether what a person actually hears is useful. A guided listening session is written and scheduled, and has not yet happened. Separately, testing with NVDA is out of scope: it is Windows-only and this project is maintained on macOS. We would rather say both of these plainly than let an automated pass imply them.
• We have not tested with assistive-technology users. Everything above is maintainers and machines. Testing with people who use these tools daily is a different and better instrument, and we have not done it.
• The automated screen-reader check reads the opening of a page, not all of it. Walking a whole document with a real reader engine is too slow for a build check, so it asserts against roughly the first sixty things announced. Problems further down a long page would not be caught by it.
• One clean result rests partly on browser behaviour. Our sticky header never hides the focused element during keyboard navigation — but testing showed that is partly because browsers scroll focus to the nearest edge, not solely because of how this site is built. A different browser could behave differently.
• The "copy" buttons confirm quietly. Copying a command changes the button's label to "Copied!", which some screen readers announce and others do not, and if the copy fails there is no feedback at all. This is on the list to fix.

Telling us about a problem

If you hit a barrier on this site — including one not listed above — please tell us. A description of what you were trying to do, the page, and the assistive technology or browser you were using is enough; you do not need to identify the WCAG criterion.

Open an accessibility issue →

If opening an issue on GitHub is itself a barrier, that is a legitimate report in its own right — reach us through any channel listed on the community page and we will file it for you.

How this page stays true

Most of the claims above are enforced by checks that run on every build, so they cannot quietly stop being true. The limitations are not — they are a record of what was measured, on the dates it was measured. When one is fixed it is removed from this list in the same change that fixes it, and when a check is added the list above grows with it.

Last reviewed 25 August 2026, against the build this page was published from. The measurements behind the limitations were taken on 24–25 August 2026.