Brainmaxing exercises: what each one trains

The word is new and the tasks are not. Every exercise on this bench is a laboratory task that has been around for decades, dropped into a browser and given a stopwatch: a colour word in a mismatched ink, a grid of tiles that flashes and goes dark, a row of arrows where only the middle one counts, three pegs and three discs. What makes them worth doing is not that they are new but that they are narrow. A narrow task produces a number that means one thing.

That is also the whole of the honesty problem. A number that means one thing is easy to quote as though it meant something larger, and the larger meaning is what people want. So each family below is described twice: what the task asks of you, and what the number it hands back is a reading of. Where those two differ from what the internet says the task measures, the difference is the point.

Reaction Time asks for one thing: press the instant the field changes colour, over a run of goes. The number is the middle one of your own press times in milliseconds, and it is contaminated by your hardware in a way nothing else here is — a touchscreen, a wireless mouse and a slow panel each cost you time that has nothing to do with attention.

Aim Trainer moves a target the moment you catch it, and reports two numbers that only mean something together: your average time per target and the share of attempts that landed. Either can be bought with the other, which is why both are shown.

Flanker shows five arrows and asks only about the middle one while its neighbours point the other way. Its number is a difference between two of your own conditions — what the misleading rows cost you against the agreeing ones — so it survives a slow laptop better than a raw time does.

Six of the fifteen are in this family, and they hold different things. Sequence Memory grows an order of lit tiles and asks you to give it back; Visual Memory flashes a set all at once and asks which ones lit; Verbal Memory shows one word at a time and asks whether you have seen it already in this run.

Number Span loads a number a digit at a time and asks for it back, forward or reversed — the reverse mode is harder because it asks you to reorder while you hold. N-Back keeps a stream moving and asks whether the square is where it was N steps ago, which means the thing you are holding is being replaced while you hold it.

Change Detection is the one that reports a capacity in items rather than a level: an array flashes, the screen blanks, one item comes back, and your hits and false alarms are turned into an estimate with a wide band around it. That unit — how many items survive a blank — is the one Cowan reviewed in 2001 [3], and it is worth keeping separate from a level: the other five report how far you got, which is a span. A level is cheap to compare and a capacity is not.

Rule Learning hides a sorting rule, lets you find it from feedback alone, and changes it once you have. The reading is how far you got through a set of rules ordered by how hard people find them, which is a property of the rules rather than a verdict on the sorter.

Task Switch flips the question — odd or even, low or high — without warning, and reports your own switch cost: what a changed rule costs you against your settled one. Mental Rotation shows two figures, one turned, and asks whether it is the same shape or its mirror; you get accuracy, decision time and what the turn cost you.

Stroop paints a colour word in an ink that may disagree with it and asks for the ink; the number is your interference cost, your mismatched answers against your own matching ones. Go/No-Go asks for a fast press on nearly every item and for no press at all on the rare one, and reports your accuracy beside the holds you missed.

Tower of London is the outlier: three pegs, three discs, a goal to match, and a commit step that separates the time you spent planning from the time you spent moving. A problem counts only when it is finished in the fewest possible moves, so the reading is in moves and is the same on a phone as on a laptop. It is the one instrument here that a slow device cannot flatter or punish.

Practice on a task improves performance on that task. That is the least controversial finding in the literature and it is worth saying plainly, because it is also the only one you can count on: the second run of anything is partly a run at a format you have already learned.

Beyond the task itself, the picture narrows fast. Pooled across trials, training on one task moves performance on tasks that look very like it, modestly, and the size of that effect depends heavily on who was trained [2]. Reaching further — to untrained abilities, to school, to work — is where the evidence stops: when the comparison group is also doing something rather than nothing, the estimated effect is close to zero [1][2].

So the honest reason to spend twenty minutes here is not the reason the category is usually sold on. It is that a repeatable number about a narrow thing is useful on its own: it tells you what a bad night costs you, what a noisy room costs you, and whether the run you just played was unusual for you.

Play the same task on the same device when you want to compare, and write down what was different when a run surprises you. The first run of any instrument is partly a run at learning the format, so read it as one. Read the two-number instruments — Aim Trainer, Go/No-Go — as two numbers, because one of them alone can always be bought with the other.

And read every reading as local. There is no norming sample behind any number on this site, which is why you will not find a percentile anywhere on it, and why nothing here is an intelligence test. The comparison that is available to you is with your own earlier runs, on your own screen — which is a smaller claim than the one you were probably looking for, and the only one the measurement supports.

A grade belongs to one claim, in one population, for one preparation — never to a compound, a technique or an article. The weakest link is the part of the chain between the evidence and the thing you would actually do that is doing the least work.