Most reaction tests hand you a number and call it a score. This one watches the gap before you move — and reads the shape of a session: how steady your timing is, whether you anticipate the signal or wait for it, and where your focus drifts. The spread matters as much as the speed.
The Gap is a reaction time test that cares less about your single fastest click and more about the shape of how you respond. Each trial drops a dot onto the distribution strip — fast to the left, slow to the right — so a session builds into a cloud you can actually read. Three modes test different channels: a flash (visual), a shape appearing (visual with a touch of recognition), and a tone (auditory, with no visual cue — listen, don't watch).
Your median is the honest center of a session — more stable than an average, which one slow lapse can drag around. Steadiness (the spread of your trials) often matters more than raw speed: a tight, even cluster is the signal psychologists actually track for sustained attention, not the headline number. Composure is how often you held the line instead of jumping before the signal. And the focus trace shows whether you sped up across the session (a warmup curve) or slowed down (focus tiring).
When you have enough trials, your median is placed against the population curve — the faint distribution behind your dots. Human reaction times follow an ex-Gaussian shape: a normal core with a long slow tail, because you can be a little faster than usual but a lapse can make you much slower. The visual curve is calibrated so its center sits near 273ms, the median reported across tens of millions of browser reaction-time clicks. A typical healthy adult lands somewhere around 250–300ms on a visual test; auditory reactions run roughly 20–50ms faster, which is why the tone mode has its own, quicker curve.
Browser tests measure your reflex plus your hardware. A 60Hz screen can add up to ~16ms, and wireless mice and keyboards add more — together, ordinary equipment can inflate your reading by 20–50ms versus a latency-corrected lab setup. So the percentile here is "faster than most people testing in a browser," not a clinical measurement. For the cleanest comparison, use a wired mouse and test on the same device each time, and watch your own trend rather than any single sitting.
This is a reading of one session, not a verdict on you. Reaction time shifts with sleep, caffeine, time of day, and how warmed up you are. A trial under ~150ms is faster than a true reaction can physically travel, so it's counted as anticipation — a lean — rather than a genuine response. Everything stays on your device: no account, no upload, no tracking.
Built as part of gptgames.dev — written almost entirely by Claude. Runs fully client-side in plain HTML, CSS, and JavaScript.