What the Fingerprint number is
Take all seven tests and you get a number between 0 and 100. It says how far you got through what these seven tasks can measure. It does not say how you compare with anybody else, and it never will until we can say who "anybody else" is.
The one-sentence version. Each of the seven tests has a published floor — what pure guessing produces — and a published ceiling — what the task was built to allow. Your position between those two points is one arm. The number is the average of the arms.
Nothing in that calculation involves another person. That is the whole reason it can exist while our reference sample is zero.
What this page said yesterday, and what changed. On 6 September this page was titled "Why there is no single BrainScore yet" and argued that four scores in four different units could not be combined without a reference distribution we do not have.
That argument was right about the thing it was describing, and it was describing a different thing. It was about a norm-referenced composite — one that tells you where you sit among people. That still needs a reference sample, and we still do not have one, and the four conditions it set out still stand for it. They are further down this page, unchanged.
What was missing was that a composite does not have to be norm-referenced. A number can be a position inside a task instead of a position among people, and that needs no sample at all. The change is recorded on the corrections page along with who asked for it and why.
The arithmetic, in full
Seven tests. Each contributes one arm, scaled between two published points.
| Arm | Floor | Ceiling | Where the numbers come from |
|---|---|---|---|
| Working memory |
0 digits | 14 digits | The task runs from three digits up to a ceiling of fourteen. You cannot guess a sequence, so the floor is a true zero |
| Pattern recognition |
2 grids | 18 grids | Eighteen grids, nine tiles each, so blind guessing yields 18 ÷ 9 = 2 |
| Processing speed |
0 matches | 60 matches | A decision, not a measurement. A rate has no natural ceiling, so we declared one: sixty correct is a match every 1.5 seconds sustained for the whole ninety. It is not derived from anyone's performance, and if it is wrong it is wrong in a way you can see |
| Spatial reasoning |
2.4 items | 12 items | Twelve items, five options each, so blind guessing yields 12 ÷ 5 = 2.4 |
| Verbal comprehension |
4 words | 20 words | Twenty words, five meanings each, so blind guessing yields 20 ÷ 5 = 4 |
| Numerical facility |
0 sums | 30 sums | The second declared ceiling, and the page says so. Answers are typed rather than chosen, so guessing yields nothing and the floor is a true zero. Thirty is one sum every three seconds for the full ninety — a rate has no natural maximum, so somebody had to pick one, and it was us |
| Attention |
12 correct | 48 correct | Forty-eight trials, four directions each, so pressing at random yields 48 ÷ 4 = 12 — verified by simulation at 11.99 over 200,000 runs rather than asserted. The ceiling is the trial count, which is not a choice |
An arm is (your score − floor) ÷ (ceiling − floor), held between 0 and 1. The number is the mean of the seven arms, times 100, rounded.
Worked example. Seven digits, eleven grids, thirty matches, seven items unfolded, twelve words, fifteen sums and thirty correct gives arms of 0.50, 0.56, 0.50, 0.48, 0.50, 0.50 and 0.50 — a number of 51. Pure guessing on all seven gives arms near zero. Hitting every ceiling gives 100.
The weighting, which is the part most composites hide
The seven arms count equally, and that is a choice rather than a finding.
Weighting them equally says the seven domains contribute the same amount to this number. We have no evidence that they should and none that they should not — so we took the option we can state in one sentence, and we are telling you it is a pick rather than a discovery.
Two things follow, and both are worth knowing. Any composite anywhere contains a weighting, whether or not the site tells you. And adding a second test to one domain would silently double that domain unless someone is watching — a weighting decision made by a scheduling accident. Our register lists one test per domain, which is the only reason the equal weighting is currently simple.
Four arms became seven, and old links did not change meaning
The Fingerprint shipped on 7 September with four arms, and people were invited to share the link. It now has seven. That created a problem worth stating plainly, because the obvious handling of it would have been dishonest.
The scores travel inside the link itself. So a link sent when there were four arms carries four scores. Read on the seven-arm scale, the three missing arms would sit at their floors and the number would collapse — a link that said 100 on the day it was sent would say 57 today, for the same person, the same four scores, and nothing about them changed. Under our own rule that the number needs every arm, it would instead have shown no number at all.
So a link is read on the scale it was written on. A four-arm link stays a four-arm Fingerprint and keeps the number it had; a seven-arm link is a seven-arm one. The page says which you are looking at, and the unit phrase under the number names the arm count it was built from. Nothing is silently reinterpreted, and no link that has already been sent has quietly changed its answer.
The cost is that two scales now exist and we have to keep explaining that. The alternative was to change what a number meant after somebody had already shared it, which is the more expensive of the two by a distance.
What the number is not
It is not a percentile. A percentile says how many people you are above. This says how far through a task's range you got. We have measured nobody — the reference sample is zero — so there is no distribution of people for you to sit inside.
It is not an IQ, a brain age, or a diagnosis, and no arrangement of these seven tasks would make it one.
And it is not reliable in the technical sense, because we have never measured test–retest reliability for any of the seven. Take them again tomorrow and the number will move. Some of that movement is you; some is the day; we cannot yet tell you the proportions, and anyone who can has data we do not.
We expect this number to be read as a percentile anyway. A 0–100 looks like one. That is why the unit phrase — of what these seven tasks can measure — is attached to it everywhere it appears, and why it is not shortened when space is tight. It names the arm count on purpose, for the reason set out immediately below.
What would have to be true for it to mean "compared with other people"
These are the four conditions this page set out on 6 September. They are unchanged, and they are the conditions for the population claim — the thing the number still does not make.
- A reference sample that we describe. Real results, from people, with the size and composition published — not "thousands of users" but a stated number and a stated description of who they were. That moves us off Rung 0.
- Test–retest reliability measured for each instrument. Without it we cannot say how much of a difference is the person and how much is the day, and a composite of four unreliable parts is less reliable than any one of them.
- A weighting decision, published with its basis. Equal weighting is currently a stated pick. Norm-referencing would make it a claim requiring evidence.
- A version, so a score means something later. The Fingerprint carries v1.0 from today, and any change to a floor, a ceiling or the weighting increments it.
Until all four hold, this number stays what it is: a position inside seven tasks. If you ever see it described here as a percentile, a ranking, or a comparison with other people without those four conditions being met and published, the corrections page is where you should expect the explanation — and if it is not there, we will have failed at the thing this site is for.
Where to go next
- Build your Cognitive Fingerprint — the seven tests, the shape, and the number
- How BrainScore works — the norming ladder and where we sit on it
- The instrument register — every test, its version, its rung and its reference sample
- Claims standards — the language we ruled out in advance