Accessibility

What I’m Aiming For

I don’t ship barriers. I sometimes accept reduced legibility.

These are promises about how I work, because those are the only kind I can keep:

• Hard barriers are defects. If something is unreachable, unusable, or unlabeled, that’s a critical failure, not a trade-off. I fix it.
• Reduced legibility is measured. Where I intentionally accept lower contrast for design reasons (like a muted timestamp or a tagline over a starfield), I name it, measure it, and list it on this page with its exact numbers.
• Impact matters most. WCAG treats all Level AA criteria as equal, but I don’t. I categorize failures by whether they completely block someone or merely make something harder — and I never ship the first kind.
• Testing is manual and real. I test by hand using real assistive technology, and I list precisely which tools I used and how.
• Status is transparent. I state my exact targets, my current status against them, and the date I last checked.

If you run into a barrier or frustrating interaction, please reach out. I appreciate feedback, and making a fix is never a hassle.

How This Was Tested

Automated checkers catch roughly a third of real accessibility problems, and none of the ones that turn on meaning — whether alt text is accurate, whether a label describes its control, whether a live region says anything useful. So the methods below are hand passes, and each row in the table names the one behind it.

Screen reader

NVDA with Firefox on Windows, driven by hand. Transcripts captured for the criteria that turn on what gets announced.

Contrast

Computed with a WCAG relative-luminance implementation across every theme, surface, and foreground role, including translucent inks composited as they actually render. Ratios on this page are measured, not estimated.

Motion

Every animation checked against the reduced-motion preference, verified at runtime rather than by reading the stylesheet.

Keyboard

Tab traversal by hand: reachability, focus visibility, focus order, and whether the sticky header hides what’s focused.

Where a criterion has not been tested, its row says so. I’d rather show you a gap than a guess.

Conformance Record

55 criteria · 34 met · 0 partially met · 6 not applicable · 15 not yet assessed

“Not applicable” means the site has nothing the criterion governs — no video, no audio, no drag-and-drop. I’ve listed those rows rather than hiding them, because what a standard covers is worth seeing.

“Not yet assessed” means exactly that: These criteria depend on page markup that’s currently being rebuilt. I’ll assess the new templates rather than the ones I’m replacing.

Perceivable

Operable

Understandable

Robust

Where I Fall Short, and Why

Each of these is a deliberate choice, not an oversight. The measured number is there so you can judge it yourself.

Nothing is currently recorded as an accepted shortfall.

If this list ever runs past five or six entries, that’s not a set of design decisions any more — it’s a palette problem, and the fix belongs in the design tokens rather than in a seventh entry here.

A Real Defect, Found and Fixed

This is the part I’d want to see on someone else’s page, so here it is. In August 2026 I ran a screen-reader pass over a billing panel that had already shipped. It passed every mechanical check. It was still wrong.

A live region announced a figure without the label that gave it meaning — the screen reader spoke a bare dollar amount with nothing to say what it measured. The cause was a default: aria-atomic is false unless you set it, so only the part that changed got announced, and the label beside it stayed silent.

The bug was invisible on first render. When the panel appeared, the whole region was new, so everything announced together and it sounded correct. Only the second update exposed it. Any test that checked once and stopped would have passed.

The transcripts below are NVDA’s own speech log, re-captured against a local fixture chosen for the cleanest possible probe: the project change swaps to a project with nothing invoiced yet, so exactly one value changes — and in the broken build it arrives as a bare empty-state word instead of a labeled fact. Same mechanism, sharpest contrast.

Before

combo box  expanded
Test Website Build  list
Test Website Build  3 of 3
Test Lead Project (unpriced)  not selected  2 of 3
Nothing yet
Project *  combo box  Test Lead Project (unpriced)  collapsed  required

After

combo box  expanded
Test Website Build  list
Test Website Build  3 of 3
Test Lead Project (unpriced)  2 of 3
Project price: Not set  Payment structure: Not set  Invoiced so far: Nothing yet
Project *  combo box  Test Lead Project (unpriced)  collapsed  required

Fixed, shipped, and written into my HTML conventions as a rule so the next live region gets it right by default.

Most people claiming accessibility competence have never opened a screen reader. This is what happens when you do.

What’s Already In Place

• Every animation respects your reduced-motion setting, and there’s a switch in the footer of every page if you’d rather turn motion off without changing your system preference.
• A skip link, and a visible focus indicator on everything you can reach with a keyboard.
• Semantic headings and landmarks, so you can navigate by structure instead of scrolling.
• Text alternatives on images that carry meaning, and none on the ones that are decoration.
• No third-party fonts, scripts, or trackers. Nothing on this site depends on another company’s server being fast or reachable.

Found a Problem?

Tell me. I’d rather hear about it than have it sit here unrecorded.

Email me directly: hello@bigbangvisibility.com. I’ve put email first on purpose. If the thing that’s broken is a form, a form is a poor way to report it.

There’s also a contact form, if you’d prefer it.

I’ll reply within five business days, and I’ll tell you what I found — including if I’ve decided not to change it, and why.