Accessibility Statement

Known gaps, measured September 14, 2026

The numbers below were measured before this section was written, not after. They come from running the accessibility category of our own audit engine over this site — every page it publishes, which on the day of the measurement was 111 of the 117 HTML files in the source tree; the other six are local-only experiments nobody can reach. Each of the four barrier classes below is a count over a denominator, because "spot-checked" and "no obvious problems" are not measurements.

• Images with no text alternative: 0 of 257 (WCAG 1.1.1). 239 carry alt text; 18 are marked decorative with an empty alt attribute, which is the correct way to do it. Both inline SVG graphics on the site sit inside a container hidden from assistive technology, so neither needs a name. Ungraded: whether an alt text that exists is a good one — a machine cannot read a picture and tell you the words are wrong — and images that CSS paints as a background, and images JavaScript inserts.
• Form fields with no label: 0 of 98 (WCAG 1.3.1, 3.3.2, 4.1.2). Each of the 98 inputs, selects and text areas in the site's own markup has a label a screen reader can announce — an aria-label, an aria-labelledby, a <label for> that resolves to it, or a <label> wrapped around it. A placeholder was not counted as a label. Ungraded: the fields our free tools build in the browser, which are not in the source a scan can read, and whether an error message is announced rather than only shown.
• Keyboard barriers: 1 found. The home page has no "Skip to content" link. It is built on a different layout from the rest of the site and it was missed, so a keyboard user reaching it tabs through the header on every visit (WCAG 2.4.1) — 1 of the 111 pages that open with site chrome. What else was measured and is clean: no element is wired to a click without also being reachable by keyboard (0 of 0 in the source), no page uses a positive tabindex to reorder tabbing, and all 111 pages declare their language. Ungraded, and not implied by any of that: tab ORDER, whether the focus indicator is actually visible everywhere, focus traps, and operation by keyboard alone. Those need a browser and a person, and nobody has run them.
• Text contrast below AA: 0 of 44 color pairs (WCAG 1.4.3, 1.4.11). An automated gate walks every page, follows the stylesheets each one loads, and fails the build when body text falls under 4.5:1 or a control's own boundary falls under 3:1 against the ground behind it. It reads opacity as part of the measurement, and it grades text inside an SVG at the size that SVG renders at rather than the size it declares. It has already caught and forced the fix of a primary button sitting at 2.60:1. Ungraded: 329 color declarations in 11 inline <style> blocks across 11 pages, 20 declarations whose color comes from a gradient or color-mix(), one stylesheet no page links, and any color JavaScript sets — including one SVG diagram drawn at runtime.

Motion is the fifth thing we measured, and it is the worst result on this page. Across the published pages the same run found 23 sources of movement. 5 are provably switched off for a visitor whose system asks for reduced motion; 18 are not. They are four smooth-scrolling jumps (on the home page, the free tools, and the "where are you stuck" page), one effect driven directly by scroll position, and 13 animations driven from custom properties on the home page. The site does carry a reduced-motion block with 20 rules in it, and four of those rules are written without !important, so a later or more specific rule can beat them — treat those as probable rather than proven. A reduced-motion preference is respected in part, not respected in full (WCAG 2.3.3).

Ungraded across the whole site, and nothing above should be read as covering it: we have not tested with NVDA, JAWS, VoiceOver or TalkBack; we have not tested at 200% zoom or at a 320px viewport; no independent auditor and no disabled user has reviewed this site for us; and we publish no audio or video, so there are no captions to check.

This section is kept honest by machine, not by memory. npm run test:a11y-website re-derives all four counts on every run and fails the build if this section is emptied, loses its date, drops one of the four classes, or claims fewer defects than the site actually has. When a gap is fixed, the number here and the date above it both have to move before the build goes green again.