A 3×3 grid with the last cell missing, and six candidates underneath. Work out the rule the grid follows, then pick the piece that completes it. Ten puzzles, each a little harder than the last.
It is the closest task on this site to fluid reasoning: working out a rule from examples and applying it, without using anything you had to memorise beforehand. No language, no arithmetic, no general knowledge — which is why the same puzzle works in any language and at almost any age.
It is not an IQ test, and it is not Raven's Progressive Matrices. It uses the same broad idea that many reasoning tests use, but the puzzles here are generated by our own code, have never been normed on any population, and produce no IQ-style number. Anyone who gives you a three-digit score after ten browser puzzles is selling something.
Each varying feature is laid out as a 3×3 Latin square: every value appears exactly once in each row and once in each column. That is what makes the missing cell knowable with certainty rather than by taste — its value is the one missing from both its row and its column.
The features that can vary are shape, count, fill and size, and difficulty is set by how many of them vary at once: one at the start, up to four at the end. Difficulty comes from the number of rules, never from drawing things too small to see.
Each distractor differs from the correct answer in exactly one feature. So you cannot find the answer by glancing for the odd one out; you have to check feature by feature. That checking is the reasoning, and it is the part we want to measure.
A hand-drawn library runs out, gets memorised, and ends up copied across the web. A generator never repeats itself and never needs maintenance — and because the rule builds the grid, the answer is derived rather than typed in, so it cannot be wrong.
Everything is drawn in a single colour, so no part of your score depends on telling colours apart. Rotation is deliberately excluded as a feature too: it is invisible on a circle, which would put a rule in the puzzle that you cannot see — a broken puzzle rather than a hard one.
Simply the number of puzzles you got right out of ten. You cannot go back to a previous puzzle: allowing that would open an endless review loop that measures patience rather than reasoning.
You will get faster at these puzzles, mostly by learning where to look first. Claims that pattern puzzles raise general reasoning are exactly the claims that have failed to replicate well, so you will not find them here.
It is free and runs in your browser with nothing to install and no sign-up. Signing in is optional and only saves your results across devices.