The physical audit
Automated tools catch contrast ratios and missing alt text. They cannot catch the experience of holding the phone. The physical audit takes five minutes, needs only your own phone, and finds a class of failures no linter will ever report: the target you cannot reach with your thumb, the text that is legible at 100% zoom and gone at the pinch the user actually needs, the control that hides under your palm.
- One hand, real grip: hold the phone as you actually do and reach every control — anything in the top third that is not navigation is a stretch; the audit's finding is 'move it down or make it reachable'.
- Zoom to 200%: most mobile browsers let users pinch-zoom; at 200% the layout must still be usable — text reflows, no horizontal scroll trap, controls reachable. If the viewport meta or fixed-width layout breaks zoom, that is a WCAG 1.4.4 failure and a real-user failure.
- Landscape, both ways: rotate and check — many layouts are only tested portrait; landscape exposes fixed-height heroes that clip, keyboards that cover the submit, and sidebars that eat the viewport.
- Thumb + glove simulation: try the flow with your non-dominant hand, then with a thick sleeve over your thumb — the miss rate on 40px targets doubles, and the audit turns 'it works' into 'it works when conditions are perfect'.
- Bright sun: take the phone outside — the contrast that passed in the design tool at 500 nits fails at noon. The audit is where 'AA on paper' meets 'AA in sunlight'.
The screen-reader lap
Switch on the platform screen reader (VoiceOver / TalkBack) and do one real task — add an item, submit a form. The lap finds what code review cannot: unlabeled icon buttons ('button, button, button'), the image carousel that reads every slide's alt text aloud as one run-on sentence, the custom dropdown that announces nothing. Ten minutes of the lap beats a week of arguing about whether aria-labels are 'really necessary'.
Keep a phone with the screen reader on and a sticky note of three core tasks beside your desk. Every Friday, run the three tasks one-handed, zoomed, and with the reader on. Five minutes a week finds regressions the day they land — not the quarter they ship.