Where this test comes from
This is a browser version of a task made famous by research at the Primate Research Institute in Kyoto, where chimpanzees were trained to touch numerals on a screen in ascending order. In the version that produced the striking result, the numbers vanished a fraction of a second after the first touch, leaving the animal to complete the sequence from memory alone. Young chimpanzees performed remarkably well at very short exposure times — well enough that the finding was widely reported as chimpanzees beating humans at a memory task.
The popular retelling flattens a lot of nuance. The chimpanzees were extensively trained on the task, the human comparison groups usually were not, and the exposure durations that favoured the chimps were extremely brief. What the work genuinely demonstrates is that a fast, spatially distributed snapshot of a numeric layout is something a trained primate brain can do very well, not that humans are broadly worse at memory. Take the name of this test as a nod to the experiment rather than a claim about you.
What it actually measures
Three things at once, which is why it feels harder than it looks.
The first is spatial working memory: the positions themselves. That much overlaps with our visual memory test, where squares light up together and you find them in any order.
The second is binding. Here each position carries a label, and you have to keep the pairing intact — not just "there was something in the top right" but "the 6 was in the top right". Binding a location to an identity is measurably more demanding than remembering either alone, and it is where most people lose a board they thought they had.
The third is sequencing under pressure. You are producing an ordered output from a simultaneous input, so every click is a small retrieval, and errors compound: hesitate on 5 and the trace for 6 and 7 keeps decaying while you think.
How the board escalates
The grid stays at 5×5 throughout and the number count grows. You begin with four numbers, and every board you clear adds one more, so the difficulty comes from density rather than from a bigger playing field. This matters, because a fixed grid means the spatial layout is always the same size and your improving score reflects memory rather than a change in the visual angle you are scanning.
A wrong click costs a strike and the board is regenerated at the same size, so a mistake does not throw away the level you had reached. Three strikes ends the run, and your score is the largest board you completed rather than the one you failed on.
Why there is no percentile here
The published research on this task studied chimpanzees, with small human comparison groups under laboratory conditions that a browser cannot reproduce — controlled exposure durations measured in fractions of a second, fixed viewing distance, trained participants. There is no normative human distribution for a self-paced version played on whatever screen you happen to have.
So we show your score and your own history, and no percentile at all. This is the same rule we apply everywhere on this site: where a published distribution plausibly matches the task, you get a percentile and a link to the study; where it does not, you get nothing rather than a number we invented. Four of our seven tests are in that position, and the methodology page names them.
Getting further
Do not read the numbers in order. The instinct is to find 1, then 2, then 3, but that means scanning the board several times and the exposure ends when you click 1. Instead take the whole board as one image and let the positions register together, then reconstruct the order afterwards.
Chunk by region. "Three along the top, two bottom left" is far cheaper to hold than seven separate coordinates, and it survives the moment of panic better.
Say the hard ones. If two numbers sit close together and you keep confusing them, a quick silent label — "seven is the lower one" — recruits verbal memory alongside the spatial trace, and two weak copies beat one.
Click deliberately rather than fast. Nothing here is timed. Rushing produces misclicks on positions you actually remembered, which costs a strike for no gain. The only clock is the decay of your own memory, and that runs at the same speed whether you hurry or not.
Finally, expect a wall somewhere between eight and twelve for most people. That is roughly where binding load exceeds what working memory holds, and pushing past it is more about strategy than effort.
Frequently asked questions
Are chimpanzees really better at this than humans?
At the specific version used in the Kyoto work, with very brief exposures and heavily trained animals, the chimpanzees performed extremely well and better than the untrained humans they were compared with. Generalising that to "chimps have better memory than people" overstates it considerably, since the comparison was not between equally trained groups on equal terms.
When exactly do the numbers disappear?
The moment you click number 1. Everything before that is yours to study for as long as you like, which means the test measures memory rather than reading speed, and a slow screen cannot penalise you.
What happens when I get one wrong?
You lose one of three strikes, the correct positions are revealed briefly so you can see where it went wrong, and a fresh board of the same size appears. You do not drop back down a level for a single mistake.
How is this different from the visual memory test?
In visual memory the lit squares have no identity and you click them in any order, so you only need the set of positions. Here every position carries a number and the order is enforced, which adds the binding and sequencing load described above. People often score quite differently on the two.
What is a good score?
There is no published norm for this format, so we will not hand you a threshold dressed up as one. Your own history on the same device is the comparison that means something, and most people find their honest ceiling after several runs rather than on the first.