Visual Memory Test
Some tiles light up for a second, then go dark. Tap every one that was lit. Each level adds a tile, until the picture slips.
The ladder
Bar widths show roughly how many people land in each band on the provisional curve. Your number is the tiles of the last level you cleared. Level one lights three tiles for one second; every level lights one more, on a grid that grows so that never more than half of it is lit; you have three wrong taps for the whole run, and the third ends it.
What moves it
Shapes, not tiles. Nobody holds twelve squares; they hold a shape or two. See the pattern as a letter, a staircase, a face, and tap the shape back rather than counting tiles.
Strikes are a budget. Three wrong taps for the whole run, not per level: a guess at level three is paid for at level eight. When a tile is uncertain, tap the ones you are sure of first; the picture often comes back once they light.
Sleep and attention. Working memory is the first thing a short night takes. A pattern that slips at eight tiles on a tired day is not a smaller memory, it is a lapse; the show is a single second, so look before it starts.
What this measures
This tests visuospatial memory with the order stripped out: a pattern lights for one second, and the question is only where, never when. A picture holds more than a list, which is why your tile count will sit comfortably above your digit span; the pattern arrives whole, as a shape, and shapes are what this kind of memory is built to carry.
Level one lights three tiles on a 3 × 3 grid; every level lights one more, and the grid itself grows so that never more than half of it is lit, because past half, remembering the dark tiles becomes the easier game and the test would quietly invert. The exposure never grows: one second, whole, at every level. Your three wrong taps are a budget for the run, not per level, so a guess spent at level three is still being paid for at level eight.
The score is the tiles of the last level you cleared, and it rewards seeing before counting: a letter, a staircase, a face found in the tiles is worth more than any effort spent memorizing squares one by one. The memory guide covers why; the starting histogram and how many measured runs have joined it are on the Method page.
Questions people ask
It is the tiles of the last level you cleared: level one lights three, and every level lights one more, so clearing level six scores eight. You have three wrong taps for the whole run; the third ends it, and the level you were on does not count. Nothing else is scored: the wrong taps and the time are on the card, but they do not move the number.
Because one rule beats two. The classic version of this test gives three misses per level and three lives on top; ours is one budget of three wrong taps for the run, which fits in a sentence and makes every guess cost the same wherever it lands. A run with a careful start goes further than a run with a lucky one.
Nobody has measured it properly, so the curve here is a guess, re-centered on Human Benchmark's saved scores and shifted down one level for self-selection: 12 is the middle, 15 is a big pattern, 18 is rare. The picture shows for a single second at every level and never plays again, yet a picture holds more than a list: expect to land well above your digit span, more so if you see shapes. The curve is an assumption until humanlag has its own runs.
So that never more than half of it is lit. On a fixed grid the dark tiles would soon be fewer than the lit ones, and remembering the gaps would be the easier test. The grid takes the smallest side that keeps the pattern under half: 3 × 3 up to four tiles, 4 × 4 up to seven, 5 × 5 up to twelve, and so on to 8 × 8. Your tiles get smaller as the grid grows; the grid itself does not.
One second is long enough to take the pattern in as a shape and too short to walk it tile by tile, which is the point: the test wants what you retain, not how long you study. The show is identical for everyone at every level, so the comparison stays clean. The chimp test one door over offers the opposite bargain: study as long as you like, then pay for it in a meaner recall.
It starts as an assumption worth 1000 runs, shaped from published norms, and every run played here adds one to it, everyone in one pool. The Method page says which norms, how the percentile is read, and how many runs this test has measured so far.