Flush-to-edge controls hit-test less reliably than center ones

What it is

A close box, back chevron, or pager dot of a given size does not keep the same hit rate in the screen centre and flush against the physical edge. The edge field shoves the reported point inward or slides it along the bezel, so the contact often lands a few pixels outside the hot zone, or into the scroller underneath. Flush controls therefore show up more often as “I tapped and nothing happened” or “I tapped the layer below.”

Why it happens

Hit-testing trusts an already-distorted coordinate. If the hot zone is clipped to the visual rectangle at the pixel border, an inward bias reports a finger that was on the button as empty glass inside it, or as the parent scroller. Tangential drift along the edge makes a tall, skinny side button hit on one tap and a neighbor on the next. The system may also hand the outermost pixels’ begin to an edge gesture first, so the application control never sees the press in that strip. The result is three unreliabilities: biased coordinates, neighbor theft, and system preemption. The centre has none of the three, so the same 44 pt button is not comparable on the two placements.

Studying it

Clone one component to centre, four edge midpoints, and four corners, and have the same people tap in a natural grip. Independent variables are placement, whether the hot zone is widened inward, and whether system edge gestures are on. Dependent measures are hit rate, fraction landing on a neighbor or parent scroller, fraction of begins eaten by the system, and “I thought I hit it” events that the log missed. Report corners separately: two edge fields add there and are usually worse than an edge midpoint. Index-finger taps on a desk stand underestimate flush failures; use a one-handed hold.

Where it stops holding

Widening the hot zone inward can restore centre-like hit rates for a flush button, at the cost of scarce edge pixels and a fight with edge gestures. A mouse reaching a flush control is immune to the capacitive field, so keyboard-and-mouse mode does not show this. Physical keys stacked on a car portrait panel’s touch edge make a miss feel like the whole screen is dead. On a large tablet the centre-to-edge distance is long, but the field still lives in the last few millimetres; “the edge is unreliable” is not a reason to abandon a wide border band.

Applying it

• For flush close, back, and send, widen the hot zone toward the interior while the visual stays flush; compare hit rates of the centre clone and the flush clone at acceptance, and keep widening or moving if they differ by more than a few points.
• Do not put a one-tap-must-hit small control inside the system edge-gesture strip; if it must live there, offer a long-press or a second-confirm path that is not on the edge.
• Accept corner controls on the worst corner; do not let edge-midpoint data stand for a corner.

Related

• Same group: C2.21.1 Incomplete edge electric fields systematically distort touch coordinates · C2.21.3 Curved edges compound optical distortion with incomplete sensing · C2.21.4 Edge gestures are especially exposed to edge-field distortion
• Nearby: C2.02 Minimum touch target size · C2.11 Thumb reach zones
• Search terms: edge target · hit reliability · bezel button