Picture a dozen dots of light moving in the dark. No body, no outline, just points — and yet within a second or two you know it is a person walking, and often whether they are heavy or light, cheerful or trudging. This is biological motion, the point-light display Gunnar Johansson built in the early 1970s, and it is one of the eeriest demonstrations in perception: the thing you recognize is not in any single frame. It lives in the change between frames. You are not seeing an object. You are seeing a happening.
I have been reading a paper that takes this distinction seriously enough to go looking for where the happening is stored — and claims to have found a shelf in working memory that nobody had properly named.
Two kinds of thing to hold in mind
The working-memory story most of us absorbed is an object story. You can hold roughly four items in mind at once — four colors, four shapes, four letters — and the classic experiments that measured this used static things: a flash of colored squares, gone, and then was this one here before? The unit of the theory is the object, and the interesting bottleneck is how many of them you can bind and keep.
But a walking figure is not an object in that sense. Neither is a bouncing trajectory, a gesture, a short melody of movement. It has a beginning and an end and a shape that only exists across time. So the question the Zhejiang University group asked, in a Science Advances paper published last May, is disarmingly simple: when you hold a movement in mind, are you spending from the same four-item budget as when you hold colors? Or is there a separate purse?
Their answer, across three very different kinds of evidence, is that the purse is separate. They call the second one the event cache, and they define an event precisely — a spatiotemporally bound segment with clear initiation and termination points, a stretch of change with edges.
How you show a shelf exists
The elegant part is the method, because "there is a hidden component" is exactly the kind of claim that is easy to assert and hard to earn. They earned it three ways.
First, statistics. They ran 206 people through fourteen working-memory tasks and let confirmatory factor analysis sort out how many underlying abilities the scores implied. The best-fitting model put events on their own latent factor, standing apart from object storage and from the central executive that shuffles everything around.
Second, interference. If two things draw from the same buffer, making you hold more of one should crowd out the other. Holding a movement did crowd out another movement. But holding movements and holding colors at the same time did not compete — people carried both loads as if from two independent accounts. And the capacities came out almost equal: about three events, about three objects, statistically indistinguishable. Two shelves, similar size, no shared till.
Third, the brain. Here is the finding I keep turning over. When they looked at which network predicted a person's event-cache ability, it was not the usual frontoparietal working-memory machinery. It was the cerebellum — specifically the left posterior Crus I, which lit up in proportion to how many movements someone was holding, and stayed quiet for the object tasks. When they mathematically "lesioned" the cerebellar network out of their model, the event predictions collapsed while the object predictions survived untouched.
The cerebellum is the structure most of us filed under coordinates your hands, the motor-smoothing organ at the back of the skull. To find it keeping a ledger of remembered happenings — things you are not doing, only holding — is a quiet enlargement of what it is for.
What this buys a puzzle designer
I am wary of pulling a clean lesson out of one paper, so take this as a hypothesis with good bones rather than a settled fact. But if there really are two shelves, a lot of solving experience rearranges itself around them.
Think about what a hard escape room asks you to hold. Not just objects — this key, that symbol, the four-digit remainder — but events: the order the levers were pulled, the sequence the tiles lit, the gesture the game master demonstrated once and will not repeat. On the old single-buffer picture, every one of those competes with every other for the same four slots. On the two-cache picture, the room is drawing down two separate reserves, and a team can be objectively "full" on remembered sequences while still having room for facts, or the reverse. The failure would feel identical from the inside — I can't hold any more — but the fix is opposite depending on which purse is empty.
It also gives a sharper shape to a phenomenon I have written about before: the layered cipher that asks you to switch registers midstream, decode one system and then feed its output into another. Each transition is not a fact to store. It is an event — a little happening with a start and an end, the shape of and then this became that. A stack of those may be taxing the event cache specifically, which would explain why the difficulty of a multi-stage cipher feels so out of proportion to the simplicity of any single stage. You are not running out of room for the symbols. You are running out of room for the transformations between them.
And it reframes the most blameless mistake in all of puzzling — the I knew that moment, when the answer arrives and you realize you had every piece and still fumbled the handoff. If the pieces you needed were held on two different shelves tended by two different systems, then the fumble was never a lapse of attention. It was a handoff between caches that did not complete in time.
The dots in the dark were never an object. What you remember of them is the walk. It is oddly comforting to learn the mind may have built a room specifically to keep such things — and unsettling, in the good way, to wonder how many of the puzzles we call hard are really just quietly overfilling that one back shelf while we stare at the front one, sure the trouble must be there.