Switch access

A switch is one button. It can be a jelly button under a hand, a sip and puff tube, a head switch on a wheelchair rest, or a phone’s own volume key, and it produces exactly one signal: now. Android’s Switch Access is described as letting someone interact with the device using one or more switches instead of the touchscreen, and Apple ships the same idea as Switch Control. Because a single signal cannot say where, the interface has to offer targets one at a time: a highlight steps through them and the reader presses when it reaches the one they want. With one switch the highlight advances on a timer and the press selects. With two, one switch advances and the other selects, which trades a second button for the timing pressure.

The scan follows the same order the keyboard does, so switch access inherits everything focus order gets right or wrong, and a control that Tab cannot reach is a control the scan never offers. That is the quiet reason so many keyboard bugs are worse than they look. A widget kept to one tab stop is one scan stop; a keyboard trap is not an annoyance but a dead end with no way out, since there is no pointer to escape with.

The design consequence is arithmetic. Every element between the top of the sequence and the action someone wants is a step, and every step is a wait. A row of five equally weighted buttons costs up to five, a card grid of twenty cards costs up to twenty, and a dialog that opens with focus in the wrong place costs the whole trip back. Scanners mitigate this by scanning in groups, offering a row or a region first and only stepping through its contents once that group is chosen, which is why sensible grouping and landmarks pay off here more than anywhere else. The rule that survives is that shallow beats clever: fewer stops before the primary action, the important thing early, and no interface that only makes sense if you can get to two distant controls quickly.

Time is the other half of the cost. A scan that moves on a fixed interval means anything with its own deadline is unusable, so a timeout warning has to be generous and extendable, and a toast that dismisses itself is gone long before the highlight could reach it. Dwell activation is the neighbouring answer to the same problem: one input and no pointer, where the selection is made by resting rather than by pressing. Both are worth knowing about together, because a design that scans well usually dwells well too.