About & Methodology

Reaction Benchmark is six short browser tests that measure reaction speed, aim, memory and colour discrimination. Everything runs in your browser. There is no account, no login wall, and nothing you do here is sent to a server.

Who runs this

It is built and maintained by one independent developer. If something is broken, wrong, or unfair, write to ysc9550@gmail.com and it gets fixed or corrected here.

The code, the copy and the test design were produced with heavy use of AI tooling, and every published figure on this site was checked against the original source by a human before it went up. Where a source could not be verified, the number is not shown at all. That rule is the reason four of the six tests have no percentile.

How the percentiles work

Your score is compared against distributions published in peer-reviewed research. It is not compared against other visitors to this site, because a self-selected sample of people who search for reaction time tests is not a population.

The rule is simple: if we can point to a published norm that plausibly matches the task, you get a percentile and a link to the study. If we cannot, you get your score and your own history instead. Inventing a distribution would be easy and would look more impressive. It would also be a lie.

Reaction Time — percentile shown

Compared against Woods, Wyma, Yund, Herron & Reed (2015), Factors influencing the latency of simple reaction time, Frontiers in Human Neuroscience 9:131: 1,469 adults aged 18 to 65, mean simple reaction time 231 ms, between-subject standard deviation 26.8 ms.

That study measured on laboratory apparatus. A browser adds its own delay on top of the underlying reaction, and measurements of online experiment platforms put that delay at roughly 71 to 87 ms (Anwyl-Irvine, Dalmaijer, Hodges & Evershed, 2021, Behavior Research Methods 53(4):1407–1425). So 80 ms is subtracted from your time before the comparison is made. Your own device may add more or less than that, which is why the percentile is described as an estimate rather than a measurement.

Sequence Memory — percentile shown

Compared against the forward Corsi block-tapping span in Facchin et al. (2024), Regression-based normative data for Corsi Span and Supraspan learning and recall among Italian adults, Neurological Sciences 45(12):5707–5718: 340 healthy adults aged 21 to 89, mean span 4.97, standard deviation 1.03, range 2 to 8. The classic task uses nine irregularly placed blocks tapped by an examiner, so a 3×3 grid clicked on a screen is a close relative rather than the same test.

Aim Trainer, Number Memory, Visual Memory, Colour Match — no percentile

No published norm matches these as they are played here. For the aim trainer, movement time depends on your pointing device. For number memory, the clinical digit span is read aloud by an examiner rather than shown on a screen and typed back, and the well-matched published samples we could verify covered narrow age bands. For visual memory, the closest published norms come from the Corsi block-tapping task, which is sequential, examiner-administered and uses physical blocks, while this test presents every square at once. For colour match, the result mostly reflects your display. Each of those pages explains the reasoning in full.

What we do not claim

None of these tests diagnose anything. They are not screening tools for colour blindness, attention disorders, dementia or brain injury, and a low score on a web page is not medical information. If you are worried about your vision or your memory, see a clinician rather than a browser tab.

Accessibility

Five of the six tests never use colour as the only signal: states are marked with text, icons or numbers as well. Colour Match is the exception, because colour discrimination is the thing it measures, and that is stated on the page itself. Animation respects the operating system's reduced-motion setting, and clickable targets are at least 44 pixels with one documented exception: on narrow phones the largest memory grids cannot give every tile 44 pixels and still fit on screen, so tiles there are sized down to 30 pixels. That still clears the 24-pixel minimum in WCAG 2.2 success criterion 2.5.8, and it is preferable to the alternative we measured, where forcing 44 pixels made neighbouring tiles overlap and steal each other's clicks.

How this is built and checked

The site has no framework, no analytics and no third-party JavaScript. It is plain modules and plain CSS, uploaded as static files, which is why pages load in well under a second on a phone. That is not minimalism for its own sake: on a test where the measurement is a timestamp difference, every library sitting between your click and the code that records it is a source of delay nobody can see.

Correctness is enforced by checks that fail the build rather than by good intentions. The build refuses to publish if a page's meta description runs long, if an internal link points at a page that does not exist, if a distribution in the norms file has no source URL attached, if an English string is hard-coded into a screen that also has to render in Korean, or if a test stage is missing its keyboard-reachable start button. Each of those checks was added after the corresponding mistake actually shipped.

Before anything goes live, all six tests are played end to end in both languages by an automated harness that clicks through real rounds and reads the result panel, because unit tests happily pass while a result screen fails to render at all. That harness is how we learned that browsers throttle timers in background tabs badly enough to break timing, which is now blocked outright with a notice instead of a bad number.

Your data

Scores, personal bests and streaks are stored in your own browser using local storage. They never leave your device, and clearing your browser data deletes them. See the privacy page for the full detail.

Changelog