Methodology

Every measurement uses performance.now(). It is a clock that only ever counts forwards, so it cannot jump if your computer corrects its own time mid-test. We never time anything with Date.now(). That one appears only to date-stamp your attempt history.

We start the clock when the stimulus actually reaches your screen, not when the code decides to show it. Those are two different moments, and they are one frame apart.

A browser runs requestAnimationFrame just before it draws a frame, so timing there would place the stimulus one frame earlier than you saw it. On a 60 Hz screen that is 16.7 ms of free speed. We take our timestamp on the frame after the one that draws. That is why scores here run slightly slower than on sites taking the earlier one.

Responses come from pointerdown and keydown, never click. A click does not register until you let the button go, so timing it would add your own press-and-release to the score.

A browser reaction time test overstates your true reaction time by roughly 30 to 80 ms. Four things contribute, in rough order of size.

• Your screen. A 60 Hz panel adds up to 16.7 ms before a change is visible, plus however long the panel itself takes to switch.
• Your mouse or keyboard. How often it reports its state, and how long it waits out the switch bouncing as it closes.
• Your operating system, which decides when to hand the press to the browser.
• The browser itself, which passes that press to the page at its next opportunity.

We do not subtract a guess. The true figure depends on hardware we cannot see, and a flattering estimate would make the number less honest rather than more accurate. Your real reaction time is faster than the number this site gives you, and we do not know by how much.

So compare your score against your own earlier scores on the same machine first, and against a distribution second. Comparing reaction times across websites is close to meaningless.

Any response faster than 100 ms is thrown away as a guess, and the trial is repeated with an on-screen explanation. Nobody reacts to a surprise signal that quickly. A press that fast was already moving before the signal appeared.

Responding before the signal is a false start. We count those, show them on the result card, and repeat the trial. The wait before each signal is random, so guessing the gap is never a strategy that pays.

On tasks where an answer can be right or wrong, only correct trials feed the reported time. The timing of a wrong answer measures something else entirely, and averaging the two together hides both.

Your trials bunch tightly together, with a few stragglers far out on the slow side. One inattentive trial in five moves an average by tens of milliseconds. It moves a median by almost nothing.

So every reaction-based test here reports a median, which is the middle value once your trials are lined up from fastest to slowest. Where the published research uses an average instead, we use the average and say so on the page. The go/no-go task is the main example.

CognitiveDrill has not published a norm of its own yet. A published norm needs at least 1,000 results in the group it describes. Below that, a percentile is decoration.

Until then, every percentile bar comes from a reference distribution. That is a curve shaped to match published results for the task, with its sources named on the page. It is labelled as such everywhere it appears. We never describe it as our data, and we never attach a sample size to it.

Each curve is built from five anchor points, so we stop the percentiles at the 1st and the 99th. Five points cannot honestly describe the extremes, and we are not going to pretend otherwise.

Your attempts are saved in your own browser the moment a run finishes. There is no account, no sign-in and no upload. Nothing you do here reaches a server.

Clearing your browser data removes your history. Every test also has a button in its settings that clears its own history straight away.

When anonymous result collection begins, we will describe it here first. It will be opt-out, and the opt-out will be a control on the page rather than a form to fill in. Until that exists, this section says nothing is collected because nothing is.

That these tests make you smarter. That practising here helps at work, at school or in daily life. That a score detects, predicts or rules out dementia, ADHD, concussion or any other condition. That any number here is a diagnosis, a clinical measurement or a brain age.

Getting better at a task means getting better at that task. Whether it carries over to anything else is not established, and where the evidence has been tested properly it is mostly negative.

If you have a genuine concern about your thinking, a doctor can arrange an assessment that actually answers the question. A website cannot.