How scoring works

This page explains how the IQ Arion app calculates your result. In short: the score is an estimate on the IQ scale, it always comes with a confidence interval, and the same answers always give the same score. Our norms come from the research literature, not from a paid representative study, and they are openly marked: “beta” during a first calibration period, “preliminary” after it.

What the estimated score means

IQ-style scores are placed on a scale where the average is 100 and the standard deviation is 15. On that scale, about two thirds of people score between 85 and 115, and about 95 % score between 70 and 130.

Alongside your score you see three more things:

• a 90 % confidence interval, for example 110–126: the range in which your score very probably lies;
• a percentile: the share of people who would score lower than you. A percentile of 84 means that you scored higher than about 84 out of 100 people;
• a profile of the four areas – patterns, numbers, spatial and logic – each marked as strong, on par (at your overall level) or developing.

How the test adapts to you

The test is adaptive. After every answer, a statistical model (item response theory) updates its estimate of your ability and picks the next question so that you have roughly a two-in-three chance of solving it. That keeps the test challenging without being discouraging. The first three questions are deliberately easy warm-ups.

• The test stops once the measurement error is small enough: after at least 15 and at most 40 questions. While we calibrate, a test averages about 35 questions (about 30–40); once the norms are final, it will usually need fewer.
• All four areas are included in a fixed balance: about a third pattern matrices, the rest number series, mental rotation and logic.
• You have 30 minutes in total. There is no timer on each question, and during the test you do not see whether an answer was right.
• Instead of “question 17 of 30” you see which section you are in.

There is no randomness and no AI in the scoring. The same answers to the same questions always lead to the same score.

Why the norms are preliminary at first

A score only means something in comparison with a reference group. The classic way to build one is a paid, representative study, which we cannot run yet. Instead, we build our norms with a method based on the research literature:

1. Three optional questions. After the test you can tell us your age, education level and country.
2. Reweighting. People who take an IQ test online are, on average, more curious and better educated than the population as a whole. We weight the results by age, education and country so that the sample matches official education statistics (UNESCO, OECD, Barro–Lee).
3. Known anchors. Published research gives approximate positions of education groups on the IQ scale – below average for little schooling, above average for a university degree. We check our scale against these approximate anchors and correct it where it clearly deviates.

Beta period. Until about 3,000 tests have been completed, scores are calculated on a provisional scale derived from the properties of the questions and are marked “beta”. When calibration ends, we recalculate these results once and update the certificates free of charge; the verification link always shows the current version. After that, the norms are recalculated regularly and carry a version number, for example “Norm 2027.1”.

Wider intervals, on purpose. This method cannot fully replace a representative sample, so we add the uncertainty of the norms to the measurement error. During the beta, the 90 % confidence interval on a result is therefore about ±9 points – wider than the measurement error alone. As soon as a representative study becomes possible, we will use it to validate the norms and drop the word “preliminary”.

Ages 13 to 17

IQ compares people with others of their own age, and reasoning scores still rise during the teenage years. Comparing a 13-year-old with adults would give an unfairly low number. For ages 13 to 17 we therefore apply an age adjustment whose direction and shape come from published research on development; we refine it with data once enough teenagers have taken the test. Results in this age group are always marked as based on preliminary norms.

Limits we set on purpose

• A range, not false precision. Every result shows its confidence interval next to the score.
• Caution at the extremes. A test that has not been standardised cannot make reliable claims about very high or very low scores. That is where the interval matters most, and why scores are shown only between 55 and 160 – beyond that range, as “≤ 55” or “≥ 160”.
• One main result. Your first complete test is your first attempt and your main result. You can retake the test after six months, with different questions; the result is labelled as a retest. Practice raises scores, and duels and daily puzzles are a kind of practice – so it is best to take the test before you start playing.
• Fair play. We record response times and whether you leave the app or page during the test. Tests with unusual patterns can still receive a certificate, but they do not enter the IQ leaderboard.

How the questions are made

Questions are generated by a program, not by an AI language model. The program draws each puzzle from a rule – for example “in each row the figure turns 90° clockwise” – and builds the wrong options by breaking exactly one part of that rule.

A separate solver then tries every option against the rules. A question is only kept if exactly one option is correct. Questions are generated and checked on our server; the answer key never reaches your phone or browser.

Each family of questions is calibrated with real answers: we look at how often it is solved and how well it separates stronger from weaker results, and adjust its difficulty accordingly. Questions for the test are kept separate from those used in duels and daily puzzles.

What the score is not

• It is not a diagnosis and says nothing about any medical or psychological condition.
• It is not equivalent to standardised tests given by professionals, and it is not “official” or “certified”.
• It must not be used for educational, medical or hiring decisions.

The certificate documents exactly this: an estimated score on the IQ scale from the IQ Arion Reasoning Test, with its confidence interval and norm version. Anyone can check a certificate with the code printed on it.

The free short test on this website

The short test on this website has only eight questions, from a separate question pool. Eight questions are too few for a reliable score, so it gives you feedback per area instead of a number. It is scored entirely in your browser; your answers are not sent anywhere.

The full adaptive test is also on this website as the free IQ test – the same test as in the app, scored on our server. The website never shows its result: you save the test to your account with Google and see the result in the app.

The role of AI

When your result is unlocked, Claude (an AI model by Anthropic) writes a short profile text, using only anonymous values our program calculated: the levels of your four areas, a rough percentile band, the norm status and your language. It does not change or decide your score, and it is instructed never to suggest any diagnosis. Arion AI, the opponent in duels, has nothing to do with your score either.

Found a question you think is wrong? Please let us know at hello@iqarion.com.