The perceptible threshold for ink following is lower than for dragging
Aliases: inking latency · ink following · stylus latency
What it is
Inking latency is end-to-end time from nib motion to ink drawn at the matching location. The threshold at which people notice, and start to feel that “the ink is chasing the pen”, is usually lower than the delay they will tolerate when dragging an object. While dragging, the eyes can sit on the object; while writing, they sit on the join between nib and ink, and a crack there is harsher. Predictive extrapolation is a separate toolkit; this claim is the perceptual bar for real pipeline delay.
Why it happens
Writing is a visual–motor loop: people trim speed and heading by whether ink stays glued to the nib. Delay parks ink where the nib already was, so a gap opens at the join. That gap is a high-contrast notch on the retina, easier to notice than the wholesale lag of a dragged object—dragging can put attention on the object’s centre, writing puts it on the nib point. Ink is also a lengthening trail, so delay integrates into a length error: the faster the hand, the longer the same milliseconds become. The perceptible bar therefore falls with speed; fast writing is harsher than slow. The pipeline is digitizer reports, scheduling, ink tessellation, composition, and display scan. Refresh hides only as much time as one frame; raising report rate without draining queues and lock-to-frame still stacks frames. In dragging UIs a delay on the order of tens of milliseconds can still pass; in writing the same order already opens a visible join, and people slow down to shrink the gap, so latency changes writing behaviour in return.
Studying it
Measure nib-to-photon time with high-speed video or a photodetector; do not trust event-handler duration alone. Contrast writing (lines, curves, fast and slow characters) with dragging an object of similar size.
Independent variables: injected delay, writing speed, refresh rate, prediction on/off as a control. Dependent measures: delay just noticeable, delay that changes writing speed, gap length in pixels, “glued / chasing” ratings.
Report detection and disruption thresholds separately. Measuring only slow uniform strokes overstates acceptable delay.
Where it stops holding
A tap that then shows ink is closer to clicking; the bar is looser than continuous writing. Projection and external displays add scan and scaling; a tablet threshold is not a projected-writing threshold. Prediction can shrink the seen gap without changing real processing time; “prediction is on, so latency passes” must not replace measurement. Sensitivity to the crack differs; set the product target by the delay at which people slow down, not by a barely-invisible median.
Applying it
- Take queues, lock-to-frame, and per-stroke rebuilds off the writing path; prefer short, stable end-to-end time over a higher report rate alone.
- Accept click, drag, and writing as three tasks; do not use drag latency as the writing pass line.
- Measure the join with high-speed video; report the tail, not only the mean.
- How to check: write fast—the join gap should stay hard to see steadily. Inject delay until writers slow down, and treat that as a hard disruption bar. At the same delay, confirm dragging still passes and writing already fails.