Accessibility: what was measured, and what it found

This site is static HTML with no JavaScript at all. Nothing appears, moves, expires, refreshes, autoplays or steals focus. There is no carousel, no modal, no infinite scroll, no cookie banner to trap a keyboard user, and no timeout on anything.

The stylesheet contains no animation and no transition of any kind, so there is nothing for a reduced-motion preference to switch off.

Every text and background pair on the densest page was measured in the browser, in both the light and the dark theme. The weakest pair is 4.62:1 in light and 5.19:1 in dark, against the 4.5:1 that AA requires for text at these sizes. Body text is 16.7:1.

The measurement is what produced the fixes. Three pairs were failing before it: source stamps under table cells at 4.35:1, table citations at 4.07:1 and the footer line at 3.50:1 — all of them small type carrying dates and sources, and all of them on every page of the site.

Colour is never the only signal. The five cell states that stand in for a missing value carry their words — not published, not checked yet, found not read, not applicable — and the hatched pattern is a second, non-colour cue on top of that.

All 212 tables on the site carry an accessible name: their own caption where they have one, and the heading of the block they sit in where they do not. Every column header is marked as a column header and every first cell as its row's header, so a screen reader announces “New York and New Jersey · median first paid day · $355” rather than “$360”.

A wide table scrolls inside its own frame instead of forcing the page sideways, and that frame is reachable by keyboard and announced as a region — a scrollable area that cannot be reached without a mouse is a defect, and it was one here until the same measurement caught it.

Every page begins with a skip link to the content. Focus is visible everywhere — a three-pixel outline in the site's own accent colour, offset from the element so it is not swallowed by a border. There is no positive tab order and no focus trap, because there is nothing on the page capable of trapping anything.

Navigation targets are 44 by 44 pixels, which is the fourth thing the measurement fixed: they had been 41 pixels tall against a rule the design had already written down for itself.

One h1 per page, headings in order without skipped levels, landmarks for header, navigation, main and footer, a language declared on every page, and text that reflows to a phone without a horizontal scrollbar. Links are described by their own text rather than by “click here”, and every link that leaves the site says where it goes.

Wide tables remain hard work on a small screen. Twelve columns of tariff rates scroll sideways in their frame; that is honest, but it is not comfortable, and a phone-shaped view of the same data is not built yet.

No assistive-technology testing has been done with real users, and no external audit has been commissioned. Everything above is measurement plus code review, which catches the mechanical failures and does not catch the awkward ones.

The dense pages are dense. Some tables run to sixty rows with a source stamp under every figure. That is the subject rather than the markup, but it makes for slow listening.

Write to the address on our contact page, say which page and what happened, and it will be treated exactly like a wrong figure: fixed, and recorded in the numbered corrections log with its date. An accessibility defect is a defect, and this site publishes its defects.