There is a moment, usually twenty minutes after you have stopped playing, when you look at a bookshelf and catch yourself noticing that the gap on the third shelf is an L-piece rotated once. Or you are stacking the dishwasher and it stops being a chore and starts being a fit problem. Or you close your eyes to go to sleep and the pieces are still coming down, slowly, in a well that is not there.

This has a name. It is called the Tetris effect, and it is one of the few things a video game has contributed to the actual scientific literature.

Where the name comes from

The phrase was popularised by a 1994 Wired article by Jeffrey Goldsmith, who described playing so much Tetris that he began mentally rotating the buildings and skylines he walked past. Anyone who has put a serious week into Marathon recognises the description immediately. What makes it interesting is that it is not a metaphor for obsession. It is a specific, measurable thing that happens to a specific kind of memory.

The images almost always arrive in the same place: hypnagogia, the drifting minute or two between lying down and being properly asleep. They are brief, visual, and oddly impersonal. You do not see yourself playing. You just see the pieces.

The experiment that made it serious

In 2000, a Harvard team led by Robert Stickgold published a study in Science that took the effect apart. They had people play Tetris and then asked what they saw at sleep onset. A majority reported falling blocks, which was the expected result and not the interesting one.

The interesting one was the second group. The researchers ran the same protocol with amnesic patients who had damage to the hippocampus - people who could not form new conscious memories, and who did not remember having played the game at all. They reported the falling blocks too.

That is a genuinely strange finding, and it points at something real about what Tetris is doing to you. The imagery was not being generated by your memory of the session, because those patients had no memory of the session. It was coming from the procedural system, the part that learns how to do things rather than what happened. Tetris trains a visuospatial skill hard enough and fast enough that the training shows up in your head independently of whether you remember the training.

Why an unfinished board is worse than a lost one

There is a second mechanism stacked on top of the first, and it explains why the effect is so much stronger after some sessions than others.

In 1927 the psychologist Bluma Zeigarnik noticed that waiters remembered unpaid orders in far more detail than paid ones, and that the detail evaporated the moment the bill was settled. Unfinished tasks occupy working memory. Finished ones get released. The Zeigarnik effect has been argued about ever since, but the basic shape of it has held up, and Tetris is almost a laboratory demonstration of it.

Because Marathon has no win condition. There is no board state that means "done". Every session you have ever played ended in one of two ways: you topped out, or you walked away mid-stack. In both cases the board was unresolved at the moment you stopped, and your head keeps chewing on it exactly the way Zeigarnik's waiters kept chewing on the unpaid tables.

This is testable on yourself, and it is the single most useful thing in this article. Finish a Sprint run - forty lines, a clean end, a number at the finish - and the imagery is noticeably weaker than after the same amount of time spent in an endless mode. The board resolved. The task closed.

Which other games do this, and which do not

Once you know the two ingredients - a hard visuospatial drill, plus a board that never resolves - you can predict fairly well which games will haunt you and which will not. It turns out to be a surprisingly sharp dividing line, and it does not fall where most people expect.

Tile games do it. If you play Mahjong online for an hour, you will get a version of the same thing at sleep onset: the eye-scanning, the sudden certainty that two tiles match. It is the same drill Tetris runs, which is pattern-matching over a crowded visual field under mild time pressure. But the imagery is gentler, and the reason is structural rather than psychological. A Mahjong layout only ever gets smaller. The board resolves by definition, so the Zeigarnik half of the effect never engages. You get the drill without the unfinished business.

Arcade games do it differently. Play Pac-Man for a long evening and what comes back is not shapes but routes - the maze, the corner you always die on, the path through the ghosts. Same procedural-memory mechanism, different content, because the thing being drilled is a movement pattern instead of a fit problem.

And then there is the category that does almost nothing at all, which is the informative one. Turn-based games with a social centre of gravity essentially never produce intrusive imagery. You can play Ludo online all evening and never once see the board when you close your eyes, and the same goes if you play Canasta online: what stays with you is the argument about whether that meld was legal, not the cards themselves. Your eyes were not doing the work. Your mouth was.

Dominoes sits interestingly in the middle and shows exactly where the line is. It is genuinely a spatial matching game, structurally very close to Tetris, and it absolutely trains your eye. But it is paced by turns. There is no falling piece forcing a decision inside a second and a half, so the drill never reaches the intensity that produces the imagery. Speed, not similarity, is what makes it stick.

The cleanest counterexample is Tic Tac Toe, which is visual, spatial, and completely solved. Nothing is being learned, so there is nothing to replay. And in the broader family of puzzle games you can watch the effect switch on and off almost at will: the timed and spatial ones leave a residue, the word and logic ones do not.

The part where it turns out to be useful

In 2009 a group at Oxford led by Emily Holmes took the mechanism and pointed it in an unexpected direction. Their reasoning was that if Tetris aggressively occupies the visuospatial channel, and if traumatic flashbacks are stored as intrusive visual memories, then playing Tetris during the window when a memory is consolidating might compete for the same limited resources.

They showed volunteers a distressing film, had some of them play Tetris afterwards, and counted intrusive memories over the following week. The Tetris group had fewer. The finding has been replicated and extended since, including with real patients after road accidents, and the press understandably enjoyed calling it a cognitive vaccine.

It is worth being careful about what this does and does not mean. It is a narrow effect in a narrow window, not a treatment, and nobody is claiming that falling blocks cure anything. But it is a real demonstration that the thing Tetris does to your visual memory is strong enough to crowd other things out of it - which is exactly what you already suspected at two in the morning with your eyes shut.

How to turn it down

If the imagery has stopped being charming, the fix follows from the mechanism rather than from willpower.

  • End on a resolved board. Do not walk away mid-stack. Finish a Sprint or let an Ultra clock run out. A closed task stops circulating.
  • Avoid endless modes late at night. Marathon and Zen are the worst offenders precisely because they are the most relaxing - neither one ever tells you that you are finished.
  • Put a hard boundary on the session. Timed modes bring their own, but if you are playing something open-ended, an external chess clock works better than intending to stop soon, for the same reason a kitchen timer beats intending to check the oven.
  • Break up the drill. Twenty minutes of something that is not a falling-piece game between sessions measurably reduces the residue.

Or turn it up

The other option, obviously, is to treat the imagery as free practice. Plenty of strong players report that the hypnagogic boards are not random - they are boards with problems in them, and you are solving them. That is procedural memory doing exactly the job it exists to do, which is rehearsing a skill when the body is offline.

If you want to give it something worth rehearsing, Master and Invisible are the modes that force real board-reading rather than reflex, and Cheese trains the excavation instinct that most players never develop. The daily challenge is the best single test of whether any of it is working, because everyone plays the identical seed.

The rules hub covers the mechanics behind all of it, the glossary explains any term here that was new, and if you want the argument about how much of your result is actually down to the pieces you were dealt, that is the subject of a separate piece on the 7-bag randomizer.