Odd one out
Nine tiles. One of them breaks the rule the other eight follow. Find it, eighteen times.
There is no timer and no countdown. Take as long as you like on each grid — this task is about what you notice, not how fast you notice it. There are eighteen grids, so it is short, but how long it takes is up to you and we have not measured it.
It is one task in the Pattern Recognition domain.
Before you start. Nothing you do here is stored or sent. Your answers and your result stay in this browser and are gone when you close the tab. The privacy notice says exactly what that means.
The first few are easy on purpose. They are there so the task explains itself.
What this measures, before you take it
Whether you spot the tile that breaks a rule, when the rule is not told to you. Early grids differ on one obvious feature. Later ones differ on a combination — a tile that is the wrong shape and the wrong fill, where neither on its own would be enough.
This task cannot be done without seeing it, and we would rather say so here than after you have spent four minutes on it. The tiles differ by fill, shape and orientation, and that information exists only in the picture — a screen reader can say there are nine tiles and nothing about how they differ. We have not found an honest way to put these grids into text, and inventing one would change the task rather than translate it, which would make the score mean something else. The digit span test announces its digits as text and may suit better.
What it does not measure: intelligence, and we are not going to imply otherwise. This is not an IQ test, it does not produce an IQ-like number, and no score here says anything about how you compare with other people. Our reference sample is 0 people.
Grid 1 of 18
One of these nine breaks the rule. Choose it.
Your result
What this task tested
Finding the item that breaks a rule you were never told. The rule changed from grid to grid, so there was nothing to learn and carry forward — each one had to be worked out on its own.
What that means
What this did not measure
This is not an intelligence test and this number is not an IQ. It says nothing about your reasoning in general, your judgement, or your ability at anything outside this task. It is one narrow thing: whether you spotted the odd tile in eighteen specific grids, today.
This is a domain score, not a BrainScore. A single task in a single domain does not make a composite, and we are not going to invent the rest of it.
Your seven scores make one shape. Take the other tests and this number becomes one arm of your Cognitive Fingerprint — seven arms and a single number between 0 and 100. That number says how far you got through what these four tasks can measure. It is not a percentile and not a comparison with other people, and every constant behind it is published, including the one we chose rather than derived.
How much should you trust this number? What moved it that is not you, what it legitimately supports, and what it cannot.
Things that moved this number that are not pattern recognition
- Screen size and brightness. The later grids differ on fine detail. A small dim phone screen makes them genuinely harder, and that is not you.
- Colour vision. Some grids differ by fill. If you see colour differently, some of these were unfair, and we would rather say so than quietly score it.
- Whether you had done it before. The grids are generated fresh each time, but the kinds of rule repeat. A second run is not a clean comparison with the first.
- Guessing. There are nine tiles, so a blind guess is right about one time in nine. Over eighteen grids that is roughly two by luck alone.
Should you believe this number?
What it legitimately supports: that on this task, today, on this screen, you found this many of eighteen odd tiles. It is also a reasonable basis for comparing you against yourself later, under conditions you can name — a different time of day, rested or tired.
What it cannot support: any statement about how you compare with other people. We have measured nobody. Our reference sample is zero, which is stated on the instrument register along with what it would take to change it.
If another site gave you a different kind of number for "pattern recognition", both can be right — a different task, a different rule set and a different scoring method are a different measurement, not a better one. We set out exactly where the difference comes from — four things that differ between two pattern recognition tests, and what each is worth.
Two questions, if you have twenty seconds
We are trying to find out whether explaining a result is worth doing at all. Nothing is recorded except which buttons you press — no result, no identifier, nothing that could be traced to you. There are no wrong answers here and nothing follows from your choice.
1 · Did the explanation above make sense to you?
2 · Did it tell you anything about yourself?
Recorded. Thank you — that is genuinely useful.
Comparing this run with another one
Your score today is one measurement under one set of conditions. The useful thing is not to repeat it immediately — it is to repeat it once under conditions you can name and tell apart: rested against tired, a large screen against a phone, a quiet room against a noisy one.
Two things make that comparison less clean here than you might hope, and both are worth knowing before you read anything into it. The grids are generated fresh every run, but the kinds of rule repeat — so a second run is partly a second look at a familiar puzzle type. And scores on cognitive tasks tend to rise on a second administration whatever the conditions: across 122 studies of cognitive ability tests, retest gains are consistent enough to be a standard methodological problem, and they keep growing until about the third attempt. A rise is the expected direction. It is not a finding about your conditions.
You do not have to write it down. Name the conditions for the run you just did and we will hold it on this screen while you do a second one.
What next
- Take the digit span test — a different domain, Working Memory, and the same rules published the same way.
- Take this again under conditions you can name — at a different time of day, or rested rather than tired. That is the only comparison this site can honestly offer you, and it is a comparison with yourself.
- How we score things, and what we refuse to claim.
How this is scored
Your score is the number of grids where your first choice was the odd tile, out of 18. That is the whole rule. There is no weighting by difficulty, no partial credit, no adjustment, and no hidden modifier. What is shown is what was counted.
You get one choice per grid. Choosing moves you to the next grid immediately. There is no going back, because a second look at a grid you have already solved measures something different.
The four kinds of grid
| Grids | What makes one tile odd |
|---|---|
| 1–4 | Fill. Eight tiles are one fill, one is the other. Nothing else differs |
| 5–9 | Shape. Eight tiles are one shape, one is a different shape |
| 10–14 | Orientation. Every tile is the same shape and fill; one is rotated |
| 15–18 | Combination. Two features vary across the grid at once, and the odd tile is the only one with a pairing that appears nowhere else. Neither feature alone identifies it |
Exclusion rules
- A run stopped early is scored on the grids completed, and the result says how many that was. It is not scored as though the remaining grids were wrong.
- A run abandoned by closing the tab is not scored at all; nothing is kept.
- What we do not detect, stated plainly rather than implied: there is no timer on this task, so we cannot tell a considered answer from a lucky one, and we do not try. A blind guess is right about one time in nine, so a score of about 2 is what pure chance looks like over eighteen grids. We would rather publish that number than pretend to a precision we do not have.
The bands
| Score | Band |
|---|---|
| 16–18 | Consistent through the combination grids |
| 13–15 | Solid on single features, mixed on combinations |
| 9–12 | Single features held, combinations mostly not |
| 4–8 | Above chance, and the harder grids did not land |
| 0–3 | At or near what guessing alone produces |
These boundaries are a design decision, not an empirical finding. They come from where the four kinds of grid sit in the sequence — nothing more. We have no reference sample to set them from, and we are not going to invent one. When that changes, the boundaries will be revisited and the methodology version will increment.