Accessibility
This page records what was measured in the interface, how it was measured, and on what date. It makes no conformance claim against any accessibility standard, because no formal assessment has been published. What has been done: every keyboard focus ring was proved to render, four colour tokens were measured and corrected, and every form field was checked for a name a screen reader can read.
What this page is, and what it is not#
It is a record of specific checks with specific results. It is not a conformance statement: no level against any standard is claimed here, and none has been assessed. Where a number was measured it is printed, so you can compare it against whatever threshold your own situation requires.
The checks below were carried out on the running application in a browser, as the operator and as a customer, across all thirty-four routes. Each finding is pinned by a test suite that goes red when the fix is removed, so a later change cannot quietly undo one.
How it was measured#
- Computed style, not source. A real Tab press moved focus onto a control and the resulting style was read back from the focused element. Reading the stylesheet would have shown nothing wrong, which is exactly what happened for months.
- Ratios, not impressions. Colour pairs were measured against the background they actually sit on, per theme, rather than judged by eye.
- Every field, not a sample. Each form control was checked for an accessible name — in a browser across nineteen routes, and by a scan of the source that refuses any field without one.
- Both audiences. Every route was opened as the operator and as a customer, because a defect that only one of them can see is the failure mode this product has had most often.
What was found, and what changed#
| Finding | How it surfaced | What changed |
|---|---|---|
| Every keyboard focus ring in the product rendered as nothing | A Tab press onto a button, with the computed style read back, returned no shadow at all | Two stylesheet layers defined the same ring token — one as a bare colour inside a theme block, one as the real recipe on the root. The theme block outranks the root whatever loads last, so the bare colour won and the rule became invalid at computed-value time. The colliding definitions are gone; the ring now reads a per-theme value tuned for each background, and a forced-colours outline is restored where the plain rule removes one. |
| Secondary text was too faint on the light theme | Measured at 3.89:1 against the day background | Retuned to 4.67:1. |
| The brand accent was used as body text | Measured at 2.53:1 | Its twelve plain-surface text uses now read the link colour at 7.04:1. Button, border and icon uses are unchanged, because those are not reading text. |
| Eleven controls had no name a screen reader could read | Four surfaced in a browser sweep of nineteen routes; reading the source found seven more | All eleven are named. Grey hint text inside a box vanishes the moment you type, so it is not a name, and the common switch idiom names nothing at all — a title on the wrapping label names the label, not the control. |
| Error toasts could cover a control | The toast stack sat one step below the modal backdrop in stacking order | Reordered, after an account with standing mailbox errors was found to have the Assistant launcher mounted, visible, and underneath a toast. |
Keyboard and focus today#
Focus is visible on every interactive control in both themes, as a real shadow rather than a rule that resolves to nothing, and a forced-colours outline stands in where a high-contrast mode ignores shadows. The account and workspace popover traps focus while it is open and closes on Escape or an outside click. In the collapsed navigation rail the tooltips are CSS-only and deliberately separate from each link's accessible name, so a screen reader is not read two competing labels for one control.
What is known to be imperfect#
- Twenty-two routes have not been through the measured design pass. The eighteen application routes, the demo, and the privacy, terms and trust pages carry none of the design, responsive, interaction and accessibility rules the landing page now has. Nothing there is known to be broken; those rules were simply never measured on them.
- About forty text sizes remain below twelve pixels. They are badge counts, timestamps and small captions in dense operational chrome rather than reading text, and were left deliberately.
- Three phone breakpoints were left as they are. Each pair is internally consistent, with no gap and no overlap, and renumbering them across fifty-one stylesheet files was the one change that could not be verified width by width.
- Contrast was measured for the tokens named above, not for every pairing in the product. A pair that was not measured is not thereby known to be fine.
- No audit with assistive technology is claimed. What was checked is that every form field carries an accessible name. That is a necessary condition, not a sufficient one, and this page does not present it as more.
If something stops you working#
Report it. An accessibility defect here is treated as a defect rather than as a request, and the route for raising one — along with the response timelines JBRH holds itself to — is set out in the Trust Centre. Describe the control, the route and what your keyboard or screen reader did, which is usually enough to reproduce it.
Questions#
Does Connect meet a particular accessibility standard?
This page does not claim a conformance level, and no assessment against one has been published. The measurements above are printed with their numbers so that you can judge them against the requirement that applies to you.
Has it been tested with a screen reader?
Every form control was checked for an accessible name, in a browser and by a scan that refuses an unnamed field. A full assistive-technology audit is a different exercise and is not what happened here.
Why are some numbers on screen still very small?
Roughly forty sizes under twelve pixels remain on purpose. They are counts, timestamps and captions in dense operational areas rather than text anyone reads at length, and shrinking the information density there would cost more than it gained.