What does the number on the screen actually include?
Everything from the moment the code decides to change the color to the moment the code logs your click. Only part of that is you. The rest is a chain of machinery, each link adding a few milliseconds that nobody can see and that you cannot remove by trying harder.
This matters because the published research everyone quotes was not gathered this way. The most quoted modern figure is 231 ms, from Woods and colleagues (2015), who measured 1,469 adults on calibrated laboratory equipment and subtracted the 17.8 ms their own hardware added. A site that shows you raw browser milliseconds next to that figure is comparing two different measurements and letting you conclude you are slow. The scores guide puts the two scales side by side.
Where do the extra milliseconds come from?
Four places. The table gives the typical size of each link, and the last row is what one study measured for the whole chain, end to end: roughly 71 to 87 ms. That measured total is larger than the four typical values added together, because the chain also waits between links and the study caught the slow cases as well as the typical ones. The paragraphs below explain why each link exists.
| Link in the chain | Typical delay | What changes it |
|---|---|---|
| Display refresh | About 8 ms on average, up to 16.7 ms at 60 Hz | A 144 Hz screen cuts the average to under 4 ms |
| Pixel response | A few ms on an ordinary LCD, near zero on OLED | Panel type |
| Input sampling | Up to 8 ms at 125 Hz, about 1 ms at 1000 Hz | Mouse report rate; Bluetooth adds more than a cable |
| Software stack | A few ms, variable | Operating system, browser event queue, JavaScript timing |
| Total, measured | 71 to 87 ms (Anwyl-Irvine et al. 2021) | Differs by machine, which is why no flat correction is honest |
Display refresh comes first. A 60 Hz screen only draws a new image every 16.7 ms, so when the code decides to turn the panel green, the panel is not green yet. It turns green at the next redraw, about 8 ms later on average and up to 16.7 ms. A 120 Hz or 144 Hz display halves this or better. That is a real, measurable advantage, and it has nothing to do with the player. The panel then has to physically change. A good OLED does this almost instantly, while an ordinary LCD takes a few more milliseconds before the change is clear enough to react to.
Input sampling adds the next slice. A mouse only reports itself so many times a second. Many office mice report 125 times a second, which can leave up to 8 ms between your click and the computer knowing about it, and gaming mice at 1000 Hz reduce this to about 1 ms. Bluetooth adds more than a wired connection, and touchscreens are slower and more variable again.
The software stack finishes the chain. The operating system delivers the event, the browser queues it, and JavaScript timestamps it. Each hop is small, but the browser's own event loop can be busy with rendering or garbage collection at exactly the wrong moment.
Because that total differs from machine to machine, this site applies no flat correction and attaches no percentile to reaction time. Your raw browser number appears as measured, and the roughly 80 ms figure, the middle of that range, stays documented as reading material for interpreting it. That is also why we call the result an estimate rather than a measurement. Your own machine may add more or less.
What did we find while building the test?
Two things, and neither is in the literature, because they are engineering facts rather than psychology.
A background tab wrecks timing outright. Browsers deliberately slow timers in tabs you are not looking at, to save battery. We ran the timed tests with a script while the tab sat in the background, and rounds that normally take about two seconds blew past a twenty-five second limit. That is not a small bias you could correct for. It is a broken measurement, and an invisible one, because the test still produces a number. That is why the reaction test here stops and shows a notice the moment the tab is hidden, and the memory tests do the same. If a reaction site happily records a score while you are in another tab, treat its numbers with suspicion.
Color transitions leak into the measurement. An early version of our reaction stage had a 120 ms CSS transition on the color change, because it looked smoother. It also meant the stimulus faded in instead of appearing, so part of that fade sat inside every recorded time. We removed the transition entirely. Any test where the go-signal eases in is measuring its own animation as much as it is measuring you.
How much can hardware move your score?
About 20 to 30 ms between a fast setup and a slow one, which is less than the 40 ms your own trials swing from round to round. Consider two people with identical nervous systems. One is on a 144 Hz monitor with a 1000 Hz wired mouse, the other on a 60 Hz laptop screen with a Bluetooth mouse. The gear alone can explain 20 to 30 milliseconds, roughly the gap between an average result and a good one. That is enough to change the story you tell yourself and not enough to change who you are.
Compare that with the noise inside one person. Woods and colleagues found that an individual's own tries swing about 40 ms around their own average, run to run. This is the reason a single round is nearly meaningless and why this site scores the median of five. It is also why chasing a personal best on one lucky click tells you less than watching your median across a week.
How should you read your own number?
Treat the absolute figure as approximate and the comparison with yourself as real. Hold the conditions still: same device, same input method, same browser, tab in the foreground, similar time of day. Under those conditions a 15 ms improvement across several sessions probably means something. Across different devices, a 30 ms difference might mean nothing at all.
And be suspicious of round numbers that sound impressive. Human reaction time never goes below about 100 ms. That floor is set by how fast nerves carry signals, not by effort. That is why this test voids anything under 100 ms as physically impossible and counts everything above that line in full, however fast. It is also why the wait here is random: predicting the change instead of reacting to it stops paying.
Frequently asked questions
Can I compare my score here with another site?
Only loosely, and only if you know what each site does. Sites differ on almost everything. Some subtract the hardware delay and some do not. Some score the median and some the average. They run different numbers of rounds, and they treat early clicks differently. A 250 ms on a site that reports raw means is not the same measurement as 250 ms here. Comparing your own history within one site is the comparison that survives.
Would a native app be more accurate?
Somewhat. A downloaded app removes the browser layer, but not the screen, the mouse or the operating system. It narrows the gap without closing it. Genuine laboratory precision needs dedicated hardware: a response box instead of a mouse, and a screen whose refresh timing is known to the millisecond.