The lock window is the angle of the first pixels of travel

What it is

Direction lock does not commit at down. It waits until the contact has walked a short displacement window (often a dozen pixels) and uses that vector’s angle to the horizontal as the main axis. Too short, and quantization noise and centroid jitter own the angle; too long, and the user already saw content follow a diagonal before it snaps to an axis. The window is the sample length of an angle estimator, in pixels (better, millimetres), not milliseconds.

Why it happens

Variance of an angle estimate falls with sample length. In the first few touch samples, centroid jump and contact-area change spin instantaneous heading. After some arc length the axis settles. A time window undersamples a slow drag and oversamples a fling; pixels or millimetres track the geometry. After lock, heading is usually not revised, so the window is a one-shot estimator. It is not the time window a velocity threshold uses: one estimates direction, the other estimates speed.

Studying it

Treat window length (px or mm) as the factor. Include both “scroll as vertically as you can” and “you must drop at 45°.” Plot orthogonal drift and diagonal arrival against the window; candidates sit near the crossing. Also watch for a hitch at the lock frame (projection slamming on). Align devices of different PPI in millimetres or pixel windows are not comparable.

Where it stops holding

A stylus jitters less than a finger, so the window can be shorter. Gloves, wet skin, and low sample rates need it longer. The fan threshold of nested orthogonal scroll looks like an angle window too, but that one selects a layer, not a projection. Writing the window in milliseconds leaves a slow precise drag unlocked forever. The window is estimate quality, not whether to lock.

Applying it

• Define the window in millimetres and convert with device PPI; a few millimetres of arc is a typical order.
• Apply the projection once on the window’s last frame; do not retarget the axis every frame.
• Accept the same millimetre window on high- and low-PPI devices: lateral RMS on vertical tasks and arrival on diagonal tasks should sit near the same trade-off.

Related

• Same group: C3.28.1 After drag or scroll starts, direction lock snaps later input to the main axis · C3.28.2 Direction lock removes diagonal drift from natural hand jitter · C3.28.3 Locking too early straightens intended diagonals
• Adjacent: C3.29 Gesture velocity thresholds · C3.23 Nested scroll touch capture
• Search: lock window · onset angle · direction slop