No Further
A target four times further away, and four times wider. Your hand will take exactly as long to reach it as it took to reach the near one. This page measures your arm the way Fitts measured it in 1954: as a channel, with a bit rate.
Every other page in this lab measures a sense: what you can see, hear, feel or remember about something somebody else put in front of you. This one measures the only part of the loop that goes the other way. Not perception. Output. The hand.
Paul Fitts wrote the law in 1954 and it is arithmetic:
Read what it actually says. The distance D and the target width W never appear separately. They appear only as a ratio. So a target four times further away and four times wider takes the same time. Not roughly. The same. The distance is free. What you are paying for is precision, precision is distance measured in target widths, and the unit of the answer is a bit.
Fitts was not measuring a speed. He was measuring a channel capacity. Your arm carries something like four to ten bits per second, and it has carried about that much through every input device anyone has built since.
Seven blocks of tapping back and forth between two bars. Four of them hold the bars at the same width and walk the distance apart. Three of them hold the same distance and walk the width. Both ladders climb through the same difficulty values, reached from opposite directions.
Every matched pair is an exact scaling of its partner: the same picture, four times bigger. And the headline needs no model at all, just three measured averages. One movement four times longer than the baseline at the same width, and one movement four times longer at four times the width. Only one of them should cost you anything.
Then the model layer, held to a harder standard than a fit: the line is fitted to the distance ladder only, and then made to predict the width ladder it has never seen. A slope learned from distance predicting times produced by width is the entire content of the claim that only the ratio matters. It is allowed to fail here.
- Latency cancels exactly. Every number here is the interval between two of your own taps on one clock. A constant input lag adds the same amount to both timestamps and subtracts out. Not folded into a parameter. Gone.
- Difficulty has no units. D and W are lengths in the same units, so their ratio is the same number on a phone and on a 34 inch monitor. This is the first page in the lab whose x axis never had a unit to calibrate.
- What is not free: the physical size of the movement decides whether your fingers, your wrist or your shoulder did the work. The law holds for all three and the slope differs between them, so the number you get is about your hand and your device together. Which is exactly why running this again on a different device is the most interesting thing you can do with it.
- 🖱️ Mouse, trackpad, pen or finger. All fine. Pick one and keep it.
- ⚡ Go as fast as you can while still hitting the bars. Both halves of that matter, and the page measures both.
- 📏 The window must stay the same size. A resize mid-run changes the geometry, so the run aborts rather than lying to you.
- ⏱️ About four minutes. 168 taps. Nothing is recorded, nothing leaves the page.
Perception lab · everything runs in your browser · nothing is recorded