Appendix — measurement
How long does it take to arrive?
Four input methods, one task: scroll until the marked paragraph sits on the centre line, and hold it there. Every method drives the same integrator and is judged by the same clock, so the numbers are comparable rather than anecdotal. What comes out is time to target, how often you overshot, and how much of the distance you travelled was wasted — the difference between a control that is fast and one that lets you stop.
Input surfaceWheel
▲ Up ▼ DownResults
| Method | n | Median time | Best | Reaction | Overshoots | Peak over | Efficiency | Final error | Fails |
|---|---|---|---|---|---|---|---|---|---|
| No trials yet — pick methods and run a block | |||||||||
How a trial is scored. A paragraph is chosen at random from those far enough away to require real travel. The clock starts when it is revealed, so reaction time is included and reported separately. The trial ends when the paragraph's centre has stayed within tolerance of the guide line for the dwell period — touching the target is not enough, you have to settle on it, which is the whole point.
Overshoot counts sign changes of the remaining distance after the target is first crossed. Efficiency is the ideal distance divided by the distance actually travelled, so 100% means you went straight there and 40% means you covered two and a half times further than needed. Peak over is the furthest past the target you ever got. A method can win on median time and still be unpleasant if it overshoots constantly, which is precisely what the table is for.
What is being compared. Every method writes a velocity into the same integrator and
the same scrollTop, so nothing wins by having a smoother scroll implementation.
Wheel is the baseline: it is the control everyone already has. Results live in this page
only — nothing is uploaded, and Clear results discards them.