How BrainScore Works
A cognitive test online can give you a number and stop there. This page is the alternative: what we measure, how we turn it into a score, what our sample currently contains, and — the part that is usually left out — what the score cannot tell you.
If anything here is unclear or looks wrong, tell us. We correct things publicly and date the correction.
Where this page stands today
0 results in our reference sample.
Tests are published and listed in the register, which is the only place their number is stated. It stores no results. Your answers and your result stay in your browser and are never sent to us, which means we hold no results at all and there is nothing to compare you against.
One thing is sent, and only if you choose to send it. At the end of the result there are two optional questions about whether the explanation was any use. Pressing a button adds one to a counter — no result, no identifier, nothing that could be traced to you, and the two answers are not even linked to each other. The privacy notice describes exactly what that is.
That count will stay at zero until we deliberately change it, and changing it would mean asking you first. We are not quietly accumulating a sample while telling you we have none.
Everything below describes what we measure and what we publish. Where a rule does not exist yet, this page says so rather than describing one in the abstract.
What BrainScore is
BrainScore is a 0–100 indicator of how you performed on a set of cognitive tasks, on the day you took them, in the conditions you took them in.
That is the entire claim.
It is a performance measure, in the way that a time on a running track is a performance measure. It reflects your ability, and it also reflects your sleep, your stress, your caffeine, the device you used, whether the dog barked, and how many times you have done it before.
What BrainScore is not
- Not an IQ score. It does not measure intelligence, and the two are not interchangeable.
- Not a medical or diagnostic test. It cannot detect, screen for, rule in or rule out any condition. If you are worried about your memory or concentration, an online test is not the tool for that question — a doctor is.
- Not a stable trait. Take it twice on different days and you will get different numbers. That is the measurement working correctly, not failing.
- Not a measure of your potential, your competence, or your worth at anything.
The domains we measure
BrainScore is a composite of separate task performances, grouped into domains:
| Domain | What the tasks in it require of you |
|---|---|
| Processing Speed | Taking in information and responding to it quickly |
| Working Memory | Holding information in mind and acting on it |
| Spatial Reasoning | Picturing, rotating and navigating things in space |
| Pattern Recognition | Noticing what fits, what breaks, and what is missing |
Each task maps to exactly one domain. We do not count a single task toward two domains, because a composite built from double-counted parts is not a composite.
How a task becomes a score
One task is published, and its scoring rule and exclusion rules are published with it — in full, on the task's own page, as promised in version 1.0 of this page before the task existed. Digit span: how it is scored.
Every task we publish states, on its own page, in full:
- What raw performance we record — the specific measures, named, for that task.
- The scoring rule that turns those measures into a task score. There are no hidden adjustments, no undisclosed modifiers, and nothing about your score calculated in a way we will not show you.
- The exclusion rules — which attempts we discard as invalid and why. For example, an attempt where entry began before the sequence finished, which measures anticipation rather than memory.
- The basis for placing the result on the 0–100 scale, and what that basis is derived from.
Band boundaries are currently a design decision, not an empirical finding. They are set from the structure of each task, because we have no reference sample to set them from. That is a real limitation and it is why the boundaries are published rather than merely applied — when data exists they will be revisited, and this page's version will increment when they are.
Your domain score is your performance across the tasks you completed in that domain. Your BrainScore is the composite across domains you have completed. If you have only done one domain, you have a domain score, not a BrainScore — and we will say so rather than inventing the rest.
Where your score sits — and why we are careful about this
Telling you a percentile — that you scored better than some percentage of people — requires a real, characterised sample of people to compare against. Without one, the number is a guess wearing a decimal point.
It is also worth knowing that published figures for the same task disagree with each other, and not because anyone is wrong: the scoring rule, the sample and the way a test is administered each move the result. We set that out in full for digit span, with the sources.
We publish exactly what we have, and we do not claim more than it supports.
We use a four-rung ladder. This is our current rung:
| Rung | What it means | What we may tell you |
|---|---|---|
| → 0 — Descriptive | We compare you to nothing. Bands come from how the task itself is designed | What you did, what the task required, what your band means |
| 1 — Provisional | We have real results, but too few to generalise | The above, plus how you compare with the N people who have taken this test — with N stated |
| 2 — Reference | Enough real results, with the sample described | Typical ranges among our test-takers, described honestly as our test-takers |
| 3 — Normative | A large, characterised sample supporting stratified comparison | Comparisons by group, with stated confidence and limits |
We are currently at Rung 0. This means we will not tell you your percentile, we will not tell you whether you are above or below average, and we will not compare you to any group — because we cannot do any of it honestly yet. When that changes, this page will say so, it will say what changed, and the methodology version number will increment.
We would rather tell you less and be right than tell you more and be guessing.
What we will publish about the sample, once there is one
A sample size on its own is not a disclosure. When we reach Rung 1, this page will carry all of the following, and we are writing it down now so that it is a commitment rather than a choice made later:
- The number of results in the sample, stated as a count.
- How those people arrived — the sources they came from, since a sample recruited from one place is not a sample of people in general.
- The device and input mix, because our measures are affected by both.
- What we know and do not know about who they are, including everything we did not ask.
- The exclusions applied to reach that count, and how many results each exclusion removed.
- The self-selection problem, stated plainly: people who choose to take a cognitive test on the internet are not a random sample of anyone.
Things that affect your score that are not your brain
We are telling you these because they are real, and because a score presented without them is misleading. We do not control for any of them, and we do not adjust your score for any of them.
- Your device. Screen refresh rate, touch versus mouse versus keyboard, and browser timing all add delay on any task measuring speed. This is not something we can fully remove, and we do not claim to.
- Practice. If you take the same test more than once, you have learned something about the test itself — its format, its pace, what is coming. A change in your score after a repeat attempt may be that learning rather than a change in you. We have not measured the size of this effect on our own tasks; when we have, the number will be on this page.
- Your day. Sleep, stress, illness, caffeine, time of day and interruptions are all conditions we cannot hold constant and do not measure.
- Where you took it. A test taken on a train is not comparable to the same test taken at a desk.
If you want to compare yourself over time, compare like with like: same device, similar time of day, similar conditions.
What we do with your results
Taking a test and seeing your full result requires nothing from you: no account, no email, no consent beyond what is essential to serve the page. If you ask us to remember a result or send it to you, we say so and you can undo it. If you choose to let your anonymised results contribute to our statistics, that is a separate question we ask separately, and saying no changes nothing about what you get.
We do not sell cognitive data. Not in any form, to anyone.
Our standards
We publish the rules we hold ourselves to — what we will and will not claim, and how we correct mistakes.
Corrections
Corrections to this page, and to every other page on this site, are published in one place: the corrections page. They used to sit here. They were moved on 5 September 2026 so that there is one record rather than one per page — two copies of a correction is how a correction goes stale.
Nothing was dropped in the move. Two of the entries there are corrections to this page, both from 5 September 2026: it said the test "stores nothing" when it stores no results, and it asserted something about other cognitive tests that we had not checked. Neither changed a scoring rule, an exclusion rule, the domain model or the norming rung, which is why the methodology version stayed at 1.1.
Read every correction we have made →
Version history
A result carries the methodology version it was produced under, so past results do not silently change meaning.
- 1.2 — 5 September 2026. A third exclusion rule came into force on digit span: an attempt where the tab is hidden while the sequence is showing is void. Browsers slow timers in background tabs, and we measured each digit staying on screen for about 2000ms against the 1000ms the task publishes — a longer look, and an easier task than the one described. This is a version change rather than only a correction because it changes an exclusion rule, and a result carries the version it was produced under. No change to the norming rung, which remains Rung 0.
- 1.1 — 3 September 2026. First task published (digit span, Working Memory), bringing a scoring rule and exclusion rules into force. Added the statement that band boundaries are a design decision rather than an empirical finding. No change to the norming rung, which remains Rung 0.
- 1.0 — 3 September 2026. First publication, before any test existed.