How this works
Methodology
A percentile is only worth reading if you can see what it was measured against. This page is the whole of it: every curve, where its numbers came from, and what we do not know.
A score is placed against two things, and they are kept separate
A site with no visitors has no population, so every ranked test ships a baseline curve: an analytic distribution fitted to a stated centre and spread. It is a curve, not a sample. No part of it is a person, and it is never counted as a result. Where a page shows a number of results, that number is real first attempts only.
Real attempts accumulate underneath the baseline. Once a device class holds5,000 real first attempts, the baseline for that device is dropped and the curve is entirely measured. Every page that is still leaning on a baseline says so next to the percentile.
Where each baseline came from
3 of 17 curves are fitted to a figure someone else published and linked below. The other 14 are our estimates, and are labelled as estimates on the page itself rather than dressed up as data. An estimate is a starting point that gets replaced by measurement, not a claim.
| Test | Centre | Basis |
|---|---|---|
| Reaction Time Test | 273 ms | published Human Benchmark's published reaction time statistics (n=81,000,000) source |
| Reaction Test | 273 ms | published Human Benchmark's published reaction time statistics (n=81,000,000) source |
| Online Hearing Test | 15.2 kHz | estimated centred on the high frequency limit usually quoted for adults in their twenties and thirties |
| Spacebar Counter | 6.8 PPS | estimated centred on the same range as mouse click speed, since both are limited by the same muscles |
| Reflex Test | 273 ms | published Human Benchmark's published reaction time statistics (n=81,000,000) source |
| F1 Reaction Test | 265 ms | estimated set slightly faster than the random-delay visual test, because the light sequence tells you roughly when the change is coming and a warned reaction runs a little quicker than an unwarned one |
| Double Click Test | 3.1 DPS | estimated centred on roughly half the sustained single click rate, which is what a double click costs |
| Polling Rate Test | 500 Hz | estimated centred between the 125Hz office mouse default and the 1000Hz gaming standard |
| Scroll Test | 14 ticks/s | estimated centred on a plausible wheel notch rate, with trackpads faster because a flick is continuous |
| Jitter Click Test | 9.4 CPS | estimated centred above the normal click rate by the margin jitter clicking is usually said to add |
| Click Test: 10 Seconds | 6.6 CPS | estimated centred on the 6 to 7 CPS range that click speed sites commonly report as typical |
| Right Click Test | 5.4 CPS | estimated centred below the index finger click rate, since most people right click with a weaker finger |
| Peripheral Vision Test | 112 pts | estimated centred well inside the anatomical field of view, because a browser test only covers the screen |
| Visual Memory Test | 9.4 level | estimated centred on a plausible span for a growing grid recall task |
| Number Memory Test | 8.2 digits | estimated centred on the classic seven plus or minus two digit span, nudged up for a self selected audience |
| Verbal Memory Test | 44 pts | estimated centred on a plausible span for a running recognition task with a growing word pool |
| Sequence Memory Test | 8.6 steps | estimated centred on the span usually reported for this style of pattern repetition task |
Why results are split by input device
A mouse switch closing a circuit and a touchscreen deciding a finger has landed are not the same measurement. Pooling them produces a number that describes neither. Every ranked test keeps three separate curves: Mouse, Trackpad, Touchscreen. You are only ever placed against people who used what you used.
The offsets between those three curves are the weakest part of this site. Where a published figure exists it is usually a single pooled number, so the per-device split is our extrapolation, not a measurement. Each page says this next to its curve. The offsets stop mattering as real results replace the baseline.
Why only your first attempt counts
Practice moves these scores quickly, and the second attempt is a different task from the first: you now know what is coming. Counting every attempt would let the most persistent visitors reshape the curve, and would make the population steadily faster than the people it is supposed to describe. Only attempt one enters the distribution. Everything after it is yours to look at and changes nothing.
What is stored
A score, which test produced it, which device class, its attempt index, and a random token generated in your browser so a repeat visit is not double counted. No account, no email, no IP address, no analytics pixel. The token is not an identity: it identifies a browser to itself, and nothing links it to you. See what we keep for the full list.
What we do not claim
- That a browser can measure reaction time precisely. Your monitor, input device, and browser sit between you and the number, and together they are worth tens of milliseconds. These results are comparable to each other, not to laboratory equipment.
- That any of this is a medical assessment. The hearing and colour vision tests are screening curiosities run through consumer hardware. They are not diagnostics and cannot replace one.
- That an estimated baseline is data. Where the table above says estimated, no published dataset was found, and the curve is a considered guess waiting to be replaced.
How to check any of this
Every ranked test links to its own distribution page, which shows the curve, the median and mean per device, and a percentile breakdown. A distribution page stays out of search results until it holds at least 200 real first attempts, because below that it is mostly a restatement of the baseline and there is no reason for it to compete for a search result it has not earned.