How the IQ Atlas test works

What the test measures

IQ Atlas uses matrix puzzles: a 3 by 3 grid of shapes with the last cell missing, and six answers to choose from. To solve one, you find the rules that change the shapes along each row. This kind of task measures fluid reasoning: the ability to spot patterns and rules in something new. It is one important part of what classic IQ tests measure, not all of it. We do not test vocabulary, general knowledge, memory span or speed of processing.

How puzzles are made

Every puzzle is generated by a program from a random number, so there is an endless supply and no answer key to leak. The rules include: the number of shapes grows, shapes change in a cycle, fills and sizes change, shapes rotate, and cells combine. Difficulty depends on how many rules work at once and how similar the wrong answers are.

The program rejects any puzzle that has a second sensible answer. Wrong answers are built so that guessing by “the most common option” works no better than chance (1 in 6). Shapes always differ by form or count, not only by colour, so colour-blind people are not at a disadvantage. Before launch we also check large sheets of puzzles by eye.

How the score is calculated

Each puzzle has a difficulty. We use a standard model from test science (item response theory, 3-parameter logistic model) to find the level that best explains your right and wrong answers, together with how uncertain that estimate is. IQ = 100 + 15 × level, where 0 is the average adult.

Paid levels are adaptive: after each answer the next puzzle is chosen close to your current estimate, so the margin of error shrinks faster. A skipped puzzle counts as wrong. Speed does not change the score; there is one time limit for the whole level.

Why a range, and what ± means

Twelve puzzles place you, but cannot pin you down. We show the range where your true score falls in about 8 cases out of 10 (an 80% interval). Longer levels narrow it. On paid levels we show one number with its margin, for example 118 ± 5.

Norms and calibration

At launch, puzzle difficulties are set by our design rules and tested in simulations. As people take the test, the difficulties are re-estimated from real answers, and the norms are recalculated every week. Until a level has about 500 completed attempts, its score is marked “preliminary” everywhere: on the result, the certificate and the verify page.

Children are compared only with children of the same age. Their age norms are preliminary too, and we say so.

What the test cannot do

It is not a clinical or medical assessment and not a diagnosis. It cannot be used to identify intellectual disability, giftedness for legal or school purposes, or any condition. Results depend on the day: tiredness, stress, distractions and the device all matter. For decisions about school or health, talk to a qualified professional.

Open numbers

As soon as enough people have taken the test, we publish the anonymous distribution of scores and update it daily. Anyone may quote these numbers with a link to IQ Atlas.