Matrix Reasoning Test

Every cell in the grid holds one tile, and a tile carries three properties. There is nothing else in there.

That closed set is what makes the puzzle answerable rather than a guess at what the author had in mind. You are never hunting for what to look at. You are working out which property a rule has hold of, and what the rule is doing to it.

• Shape. A circle, a square or a triangle.
• How many. One, two or three copies of the shape.
• Fill. Hollow, half filled or solid, with nothing in between.

A rule works on one property, and there are three things it can do with it.

The first two are the reason a grid is not a strip. A rule held across the rows is answered by looking along the bottom row. It is not answered by looking down the right hand column. The tile that fits the column alone is one of the eight on offer.

So a property under one of those rules always has a value repeated. Three different values over three cells would be readable in either direction, and the two rules would stop being two rules.

Carpenter, Just and Shell took the paper version apart in 1990. What separated people was not spotting a rule. It was carrying one part-finished rule while checking the next.

• Held still across a row. The three tiles in a row share the value, and the rows differ.
• Held still down a column. The three tiles in a column share it, and the columns differ.
• One of each, everywhere. Each of the three values appears once in every row and once in every column.

The ladder is fixed and it does not adapt to you. Eight matrices with one rule running, then eight with two, then eight with three.

Nothing stops you early. A miss costs you that matrix and not the rest of the run, and there is no clock over any of it. That is the opposite arrangement to the abstract reasoning test, which puts one clock over the whole paper and ends it wherever you are.

Every matrix offers eight tiles. One completes the grid and seven do not, and most of the seven are wrong in exactly one property. A guess is right one time in eight, so a run answered blind lands near three.

Both settings change what the number means. Twelve matrices is half the evidence, and four tiles to choose from makes a guess right one time in four. Neither run is comparable with the default one.

Nothing here is timed against you, so the time is a thing to measure rather than a thing to impose. It is the median across the matrices you got right.

Time spent on one you got wrong is time you gave up on rather than time to an answer. Folding it in would let a run of quick confident mistakes read as quick reasoning.

That leaves one thing the median hides, so the card carries it alongside. A fast time off two solved is not a fast time off twenty, and the share you solved is on the same card.

The third figure is how much longer the three rule matrices took you than the one rule ones. That one counts every matrix at both rungs, right or wrong, because it is about the load rather than the outcome.

For a number that does have a real distribution behind it, processing speed by age has one.

Nobody has published a distribution of how long people take on matrices we generated ourselves. So there is not one on this page.

Raven's tables are real and they do not fit. They describe raw scores on sixty of his own items, on paper, mostly under supervision and mostly decades ago. Hanging our seconds on that curve would be an invented number wearing a citation.

Carpenter and colleagues did time people on matrices. Their items were Raven's and their sample was small, which makes it a finding rather than a norm.

Matzen and colleagues later built software to generate Raven-like matrices with measured properties, which is the same idea as this page. Their figures describe their items, not ours.

Our own norm goes up when a cohort here reaches a thousand runs, and not one day before that.

Matrix puzzles are the best known item type on intelligence batteries, and that is the whole reason this section exists. Appearing on one does not make this page one.

No time here has been checked against any measure of intelligence. Nobody has run that study on these items, so a claim in that direction would be made up.

It is not a clinical subtest either. The published ones have their own items, their own administration and their own tables, and none of that survives being moved into a browser tab.

It detects and rules out nothing, and no time on this page is a diagnosis. If something has genuinely changed in how you think or remember, that belongs with a doctor rather than a web page.

Practice here makes you quicker at these three rules on these tiles. Researchers call that near transfer, and whether brain training games are worth it covers what happened when the wider claims were tested.