Instrument · deployed
Pursuit
A human bit-rate measurement instrument, wearing a game.
7.82bits/s
± 0.66 at 95%
97.2%
104 of 107 selections
516ms
median press interval
One scored run · 2026-08-30 · computed at build time from the instrument’s own telemetry
A letter lights up on a board of twenty-six. You press that key. Sixty seconds later the screen reports how fast you moved information through a keyboard, in bits per second — the achieved bit rate from Shenoy et al. (2021).
Because N is fixed, the whole thing collapses to a single coefficient: 4.64 bits per correct selection. No encoding, chording or clever input scheme moves that number — which means the entire optimisation problem is selections per second, and every remaining decision is an interface decision.
So the board shows the next target as well as the current one. Reading one ahead is worth real speed, and it is the only thing in the game that rewards understanding how it works. A miss subtracts rather than merely failing to add, which makes accuracy and rate the same question instead of two.
The part I would defend is the interval. The first version computed it with the Wald estimator, which reports ±0.00 for a run with no errors — from sixty trials. That is not a tight interval, it is a broken one, and a flawless run is precisely what a practised player produces. The replacement is Agresti-Coull, which stays finite at both extremes and is wider everywhere, the headline number included. Widening your own headline figure is the right direction to move a number you are publishing as evidence.