How to Play
A set of tiles on the board lights up for a moment, then goes dark. Once the tiles are dark, tap or click every tile that was lit. Get them all and you move up a level. Level 1 starts on a 3x3 board with 3 tiles to remember, and every level after that adds exactly one more tile. The board grows on its own whenever the pattern would otherwise crowd more than about a third of the available cells, stepping from 3x3 up through 4x4, 5x5 and 6x6 to a maximum of 7x7. The preview also shortens by 0.05 seconds per level, starting at 1.2 seconds and bottoming out at a 0.7-second floor.
You have 3 lives. Tapping a tile that was not lit costs one life and restarts the same level with a completely new pattern, so a mistake never ends the run outright but it does throw away everything you had memorized. Run out of lives and the game ends; your score is the level you were on. Tapping a tile you have already correctly tapped does nothing at all, so there is no penalty for double-tapping. There is no timer on your answer, either. Only the preview is timed, which means once the tiles go dark you can take as long as you like to reconstruct the pattern before committing to your first tap.
Because the tile count and the preview length move independently, the difficulty ramps in two ways at once. The first few levels are worth knowing exactly:
- Levels 1-3: 3 to 5 tiles, on a 3x3 then 4x4 board, with 1.2 to 1.1 seconds of preview.
- Levels 4-6: 6 to 8 tiles on a 5x5 board. The preview crosses under one second at level 5.
- Levels 7-10: 9 to 12 tiles on a 6x6 board, with the preview falling from 0.90 to 0.75 seconds.
- Level 11 and up: 13 or more tiles on a 7x7 board at the 0.7-second floor. From here the board stops growing and the preview stops shrinking, so the only thing that increases is the number of tiles.
Tips
- Encode shapes, not coordinates. Three tiles in a diagonal is one thing to remember; three separate row-and-column pairs is three things. This single habit is worth more than any amount of staring harder.
- Break a large pattern into two or three small groups rather than one big blob. A cluster of nine is unmanageable; a cluster of nine seen as "an L in the top left, a pair on the right edge, and two in the bottom row" is three items.
- Use the edges and corners as landmarks. Tiles touching a border are far easier to place accurately than tiles floating in the middle of a 7x7 grid, so anchor your groups to whatever border they are nearest.
- Tap the tiles you are least sure about first. Your memory of the pattern decays while you are answering, and the shakiest items are the ones that decay first. The tiles you are certain of will still be there in five seconds.
- A wrong tap costs a life and reshuffles the level, so guessing is expensive. If you genuinely cannot recall the last tile, you still have to guess, but do not start guessing before you have placed everything you actually know.
- Look at the whole board rather than scanning it. Sweeping your eyes tile by tile burns most of a 0.7-second preview on eye movement alone. Fix your gaze near the center and take the pattern in peripherally.
What a Grid Memory Test Actually Measures
This is a test of visual-spatial working memory: your ability to hold a small amount of location information in an active, immediately available state for a few seconds. Working memory is not the same thing as memory in the everyday sense. It is not where you store your phone number or what you had for lunch. It is the small, volatile workspace that holds whatever you are currently thinking with, and its defining property is that it is startlingly limited.
Psychologists have measured this kind of thing for decades with block-tapping tasks, in which an examiner touches a sequence of blocks and the participant reproduces it. Those tasks are sequential — order matters. This game is not: it shows a set of positions simultaneously and asks only which tiles were lit, in any order you like. That difference makes it a simultaneous spatial-span task rather than a sequential one, and it is why the strategies that work here are about shape and grouping rather than about rehearsing a route.
It is worth being clear about what the number at the end is not. It is not an IQ score. Working memory capacity is correlated with performance on reasoning tasks, but a correlation across a population tells you very little about any individual, and a browser game played once on a phone is not a controlled measurement. It also says nothing about long-term memory — the ability to remember a face, a name, or a route home runs on different machinery and does not track your level here.
How Much Can Anyone Actually Hold?
The honest answer is: less than people expect, and the exact figure depends heavily on how you ask. The famous claim that short-term memory holds "seven plus or minus two" items came from work on verbal spans — digits, letters, words — where people can rehearse the items silently and keep them alive. Later research that removed the ability to rehearse, and that counted genuine independent items rather than groups, put the number considerably lower, commonly described as around four items for visual material.
Visual-spatial capacity in particular is usually estimated in the low single digits when items must be held simultaneously and cannot be verbalized. That range is not a hard wall — it moves with how similar the items are to one another, how long the display was up, whether anything interfered in between, and how much of the material can be collapsed into a familiar shape. But the order of magnitude is the point. If you are comfortably reproducing thirteen tiles on a 7x7 grid after a 0.7-second flash, you are not holding thirteen independent items. You are holding three or four things, each of which happens to contain several tiles.
Why Chunking Works
Chunking is the reason the previous paragraph is not a contradiction. The limit on working memory appears to be a limit on the number of units held, not on the amount of information inside each unit. A chunk is a unit that your long-term memory already recognizes as a single thing. Four tiles arranged as a square cost roughly what one tile costs, because "square in the upper right" is one label, not four.
This is why chunking feels like cheating and is not. You have not expanded the workspace; you have changed what you put in it. The person reproducing an eleven-tile pattern is not straining harder than the person stuck at six. They are describing it differently — as a diagonal, a cross, an L, a pair — and each of those descriptions is one slot.
Some practical forms this takes on a grid: read the pattern as a route your finger would trace; look for symmetry, because a pattern mirrored across the middle costs half as much to store; count how many tiles fall in each row and store the counts alongside a rough shape; and give recurring configurations names in your head. The tiles are placed by a shuffle, so nothing repeats, but the vocabulary of shapes is small enough that most patterns decompose into pieces you have seen before.
Where Most Players Stall, and Why
There are two distinct walls in this game and they have different causes.
The first is the jump to the 5x5 and 6x6 boards around levels 4 through 7. Up to that point the board is small enough that positions are easy to describe — "middle", "top-left corner", "next to the center" are unambiguous on a 3x3. On a 6x6 those descriptions stop being unique, and a strategy that relied on naming individual cells collapses. Players who hit this wall are usually still encoding tile by tile, and the fix is a change of method rather than more effort.
The second wall is at the 0.7-second floor from level 11 onward. Here the board and the preview are frozen and only the tile count grows, which strips the task down to raw grouping efficiency. Seven tenths of a second is roughly two or three eye fixations, so anything you did not take in as a whole is simply not available to you. Progress past this point comes almost entirely from getting better at seeing structure instantly, and it comes slowly.
There is also a failure mode that is not about memory at all. Because a wrong tap reshuffles the level, a run of bad luck late in a level can cost all three lives on the same level number, which feels like a collapse but is just variance. Judge yourself over several runs.
Does Practice Increase Capacity, or Just Strategy?
Almost certainly strategy, and this distinction matters more than it might sound.
People who practice a working memory task reliably get better at that task. That part is not controversial. What is heavily disputed is whether the underlying capacity moves, or whether the improvement is task-specific skill — knowing what to look for, having a ready vocabulary of shapes, having automated the encoding so it costs less attention. Studies that test whether training on one memory task improves performance on different, untrained tasks have produced far weaker and much less consistent results than the improvements on the trained task itself, and the claim that general cognitive ability can be raised this way remains contested.
The practical reading is not discouraging. Getting better at this game is real improvement at this game, and the specific skill you build — decomposing a visual field into a few meaningful groups — genuinely transfers to anything that looks similar. What you should not expect is that climbing to level 12 here makes you better at remembering where you parked.
If you want to improve, practice deliberately rather than repeatedly. Play a few runs while consciously forcing yourself to verbalize the pattern as two or three shapes, even when it slows you down. That will feel worse before it feels better, which is the usual signature of replacing a habit with a better one.
About This Visual Memory Test
This is a free online visual memory test in the tradition of grid-recall games used to probe short-term spatial memory: watch a pattern, then reproduce it from memory on a board that keeps growing. It runs in the browser with no download and no account, works with a mouse or with touch, and saves your best level locally so you can come back and try to beat it.
The design deliberately pushes on two axes rather than one. Adding a tile per level increases the load; shortening the preview reduces the time available to encode it; and growing the board keeps the density roughly constant so that the pattern never becomes trivially dense or hopelessly sparse. Once the preview reaches its 0.7-second floor and the board reaches 7x7, the test becomes a clean measure of how much structure you can extract from a single glance.
Three lives with a same-level restart is a deliberate choice as well. It keeps a single unlucky tap from ending an otherwise strong run, while still making guesswork costly enough that careless play does not pay. Take the untimed answer phase seriously: the clock is not running, and the most common avoidable mistake in this game is tapping fast when nothing was asking you to.