How the reasoning test works, and what it cannot tell you.
Methods and limits of the 30-question reasoning test. Updated 14 November 2026. Instrument version standard-reasoning-20261114-v2.
An estimate from one unsupervised online sitting, not a clinical assessment. The comparison group is people who took reasoning tests on this site, not a representative national sample. Use the result for interest and practice, not for decisions about school, work or health.
The paper
Thirty generated multiple-choice questions, six options each, presented from easy to hard. Every paper holds the same number of questions of each type at each difficulty level. The default limit is 20 minutes; the edition for people over 60 runs without a clock.
| Format | Questions | Task |
|---|---|---|
| Pattern matrices | 10 | Combine one to four rules across a three-by-three grid. |
| Number sequences | 8 | Find the rule and give the next term. |
| Figural analogies | 6 | Apply the change from one figure to another. |
| Mental rotation | 6 | Find the same arrangement, turned. |
How each question is checked
Questions are generated, so correctness is proved by program rather than asserted by an author. Before a question is used, a separate solver must find exactly one answer: every matrix rule must be readable from the eight visible cells, every sequence is tested against every rule family in the test and rejected if a second rule fits with a different next term, and analogies change only ordered properties (how many, how large, how dark) by visible steps.
Answer options are built as a balanced design: each value of each tested property appears equally often, and the correct number in a sequence question holds a random rank among the six. Counting features or picking the middle value finds the answer no more often than chance (17%).
What changed in version 2
We analysed 475 complete, serious runs of version 1 question by question. Rotation and sequence questions separated stronger from weaker takers well. Matrix and analogy questions did not. Two causes were found and removed: some rules relied on an invented order (which shape comes "next"), and the correct option could be recognised from the options alone in about nine cases out of ten. Version 2 replaces the matrix, analogy and sequence questions; rotation is unchanged and links the two versions on one scale.
The score
Each answer is scored on a measured difficulty scale (a Rasch model with a fixed one-in-six guessing floor). The difficulty of every question type and level is estimated from recorded answers, and your position on that scale is compared with the spread of people who took the reasoning tests here: 100 is their average, 15 points is one standard deviation. Runs with more than a third of questions unanswered, or answered in under three seconds per question, are kept out of the statistics.
The free result shows one of three 30-point ranges: 55 to 84, 85 to 114, or 115 and above. The paid result shows the estimate, its 90% range and the percentile. A new question type starts from a stated design difficulty; until 30 answers are recorded for it the result is labelled provisional and its range is widened.
The age view adds published average age differences in reasoning (Salthouse, 2009). It is an approximation, not an age norm collected on this test.
Measured quality
| Property | Version 1 | Version 2 |
|---|---|---|
| Internal consistency (reliability) | 0.72, from 475 serious runs | Published here once 300 complete runs exist |
| Stability on retest | Not measured | Not measured |
| Agreement with an established test | Not measured | Not measured |
| Representative norm sample | None | None |
A reliability of 0.72 means the version 1 estimate carried an uncertainty of roughly 12 points either way. That is why a range is always shown, and why version 2 was built.
Not yet established
Norms from a sample representative by age, sex and education; stability on retest; agreement with an established professional test; fairness across the fourteen languages. A registered study covering these is planned. Its protocol, data and results will be linked here whatever they show. No independent review or professional endorsement is claimed.
The certificate records an estimate from an unsupervised online test. It is not a clinical finding, a qualification or an admission document. Practice improves familiarity with these tasks; it does not show that intelligence changed.
Privacy and purchase recovery
Saved progress can be resumed from this browser for up to 24 hours when storage is available. Expired progress is discarded when next read; completing a run or starting a new test clears or replaces the saved progress. The clock continues during a timed test. Account-linked result saving is optional. Do not publish a private report link: possession of its secret grants report access.
Payment verification is performed on the server with Stripe. Test answers and scores are not sent as Stripe metadata or advertising events. Buyer email is used for transaction delivery and recovery; purchasing does not subscribe the buyer to marketing. A delivery-provider failure leaves online report access available after verified payment and requires delivery retry.
Paid reports are offered only to people confirming they are 16 or older. Children's puzzles are a separate experience. If you need a clinical assessment, seek an appropriately qualified professional using a standardized instrument suited to the purpose.
Read the complete fictional report sample · Terms · Privacy · Contact support