What the number is made of
Between the pixel changing colour and the click arriving in this page there is a chain of hardware and software, and every link adds time. A 60 Hz display only redraws every 16.7 milliseconds, so the green frame appears somewhere in a window that wide. A mouse reporting at 125 Hz adds up to another 8 milliseconds before the button press is sent. The browser then delivers the event on its own schedule, competing with whatever else the page and the machine are doing. None of that is visible to JavaScript and none of it can be subtracted out.
The practical size of this is tens of milliseconds, and it varies by setup rather than by person. A gaming monitor with a high-polling-rate mouse and a mid-range laptop trackpad can differ by 30 milliseconds or more for the same hand. That is why this page shows your spread rather than ranking you: the ranking would mostly be measuring your hardware.
Why the spread matters more than the average
Five rounds give you five numbers, and the interesting one is how far apart they are. A tight cluster means you were attending consistently. A fast best with a slow worst usually means one round where you were waiting for the change and one where your attention drifted — or one where you half-anticipated and got lucky. Anticipation is the main way people flatter themselves on tests like this; the early-click detection catches the obvious cases but not a lucky guess that lands just after the change.
The median is shown alongside the mean because a single bad round drags a five-round mean noticeably. If mean and median are close, the run was consistent. If the mean is much higher, one round was an outlier and the median is the more honest summary.
Using it sensibly
Run it on the same machine at different times and the comparison starts being worth something: tired against rested, before coffee and after, first thing in the morning against late at night. Those are within-subject comparisons where the hardware latency is a constant and cancels out. Comparing your number against a friend's on a different computer is comparing two computers with some human variation mixed in.
If you want to see the effect for yourself, run five rounds, change nothing about your posture or attention, and run five more. The gap between the two averages is roughly the noise floor of this measurement, and it is usually larger than people expect.
Questions people ask
Why does clicking early not just count as a slow round?
Because it is a different thing. Clicking before the change means you were predicting the timing rather than responding to it, and a predicted click can land at any time, including before the stimulus exists. Those rounds are rerun and the count is reported separately so you can see how often you were guessing.
Is a touchscreen slower than a mouse?
Usually, and by enough to matter. Touch panels sample at their own rate and many phones add processing between the touch and the browser event. Expect a touchscreen result to read slower than a mouse result on the same person, and do not compare the two.
What is a good number?
The honest answer is that this page cannot tell you, because a large and unknown part of what it measured belongs to your hardware. Published averages exist but they were collected on equipment nothing like yours, under conditions nothing like your desk. Treat your own repeat runs as the only meaningful baseline.
Is my result saved anywhere?
No. The rounds exist only in the page while it is open, nothing is sent to a server, and closing the tab loses everything. There is no history and no leaderboard. Write the number down if you want to keep it.