How this test works

Last reviewed 6 September 2026

1. What is being measured

Scores on almost any set of cognitive tasks correlate positively with one another. Someone who does well on verbal problems tends to do better than average on spatial ones too. That pattern - the "positive manifold" - is the oldest robust finding in the field, and the general factor extracted from it is conventionally written g.¹

The dominant map of what sits underneath g is the Cattell–Horn–Carroll model, which arranges abilities in three strata: g at the top, around eight to sixteen broad abilities beneath it, and dozens of narrow abilities below those.³ The broad abilities this test samples are:

• Fluid reasoning (Gf) — solving novel problems that cannot be answered from stored knowledge. Cattell separated this from crystallised ability in 1963 and the distinction has held up.²
• Comprehension-knowledge (Gc) — the breadth and depth of acquired knowledge, and the ability to reason with it.
• Visual processing (Gv) — generating, holding and transforming visual images.

A 30-item test cannot measure all of CHC. What it can do is sample a few broad abilities that load heavily on g and report a composite, which is what this test does.

2. Why these four question types

The blueprint follows the International Cognitive Ability Resource (ICAR), a public research project that set out to build a validated cognitive battery outside the commercial publishers. Condon and Revelle’s initial validation drew on 96,958 participants from 199 countries and settled on four item types: matrix reasoning, letter and number series, verbal reasoning, and three-dimensional rotation.⁴ Their 16-item short form reached an internal consistency of α = 0.81 and ω_total = 0.83.

A later study administered that short form alongside the WAIS-IV, the standard supervised adult battery, to 97 university students. ICAR16 scores correlated r = .81 with WAIS-IV Full Scale IQ, and the latent general factors correlated .94.⁵ That is strong evidence that a short, unsupervised, purely online battery of this shape captures much of what a long supervised one does — with two caveats worth stating up front: the sample was small and made up of students, which restricts range and tends to deflate correlations, while a convenience sample of this kind also cannot be assumed representative.

Generated directly from the live item bank at build time. Discrimination is set per domain from the reported g-loadings of each item family: verbal reasoning and letter/number series load highest, spatial rotation lowest.⁴

3. How the items were built and checked

Figural matrices

Each matrix is generated from an explicit rule set rather than drawn freehand. The rules are the ones Carpenter, Just and Shell identified when they reverse-engineered what the Raven’s matrices actually demand: constant in a row, quantitative pairwise progression, distribution of three, and figure addition and subtraction. Their central finding was that item difficulty is driven mainly by the number of rules a solver must hold and apply at once.⁶ This test uses that directly: easy matrices run one rule, the hardest run three.

A second family presents each cell as a grid of marks where the third cell in a row is a boolean function (union, intersection, exclusive-or, or subtraction) of the first two. Building items by algorithm rather than by hand is the approach the automatic item generation literature recommends,¹⁷ and it has a practical benefit: the universe of possible items is large enough that item exposure is much less of a threat.

Series

Each series is stored as a declarative generator - an arithmetic step, a constant second difference, a repeating cycle of increments, two interleaved sub-sequences, a recurrence - rather than as a typed list of terms. The build re-derives every displayed term and the keyed answer from that generator, so the rule stated in the explanation and the answer marked correct are guaranteed to be the same thing.

Spatial rotation

The spatial items are a planar version of the Shepard and Metzler rotation task¹⁰ in the group-administered form Vandenberg and Kuse popularised.¹¹ Each target figure is chiral: its mirror image cannot be produced by any rotation of the original. That property is what makes the item well-posed, and it is what makes the strong distractors possible - they are reflections, which is the classic error in this task.

Verbal reasoning

Verbal items are analogies, category exclusions and deductive inferences. Vocabulary is kept deliberately common: the target is the relation or the inference, not whether you happen to know an unusual word. This is the item family most exposed to cultural and educational background, which is noted again under limitations.

Every item is machine-checked before release

A verifier runs over the whole bank on every build and refuses to ship if any assertion fails. It currently proves, for all 30 items:

• exactly one option is correct, and it is the one keyed as correct
• no two options render identically (rotations of a circle, for instance, are not "different")
• a matrix’s answer is genuinely determined by a rule that operates along the row
• a matrix that varies rotation uses a shape on which rotation is actually visible
• a boolean matrix is explained by exactly one operator, so no second reading works
• a series answer is re-derivable from its declared generating rule
• a spatial target is chiral, so its reflections really are wrong
• no figure overflows its box and gets clipped
• the difficulty ladder has no gap large enough to leave a band of ability unresolved

This catches construction errors. It cannot catch an item that is well-formed but measures the wrong thing — only data can do that, which is the subject of section 6.

4. How the score is computed

Most free tests count correct answers and apply a formula. That treats every question as equal evidence, when a hard question obviously tells you more about a strong performer than an easy one does. This test uses an item response model instead.⁷

Each item is described by three parameters, in the three-parameter logistic model:⁸

• b — difficulty: the ability level at which the item is at the halfway point.
• a — discrimination: how sharply the item separates people around that point.
• c — the guessing floor, fixed at 1 / (number of options), so a six-option matrix is never scored as though a blind guess were impossible.

Ability (θ) is estimated by expected a posteriori estimation over a fixed grid with a standard normal prior.⁹ EAP is used rather than maximum likelihood for two reasons: maximum likelihood is undefined for a perfect or a zero score, and the posterior standard deviation gives a per-person standard error — which is what makes an honest confidence interval possible instead of a bare number.

The final score is the conventional deviation IQ, 100 + 15 θ. Unanswered questions are scored as incorrect; the test says so before you start, so that skipping is not a scoring strategy.

5. How precise the score is

Every measurement has error, and a test that hides it is misleading you. The table below is computed from the test information function of the actual item bank on this site.

Read that as: a reported score of 100 means the evidence is consistent with a true standing anywhere in roughly the low 90s to the high 100s. Precision is best in the middle, because that is where most of the items sit, and degrades toward the tails where there are fewer items to separate people. Scores outside 65–135 are not reported as numbers at all, because a 30-item test cannot resolve them; the result page says so explicitly when it happens.

For comparison, the WAIS-IV reports a Full Scale IQ reliability of about .98 and a standard error of measurement near 2.2 points.¹⁵ It takes 60 to 90 minutes with a trained examiner present. This test is faster and free; it is also less precise, and the gap is the honest price of that.

6. Where the norms come from

A properly normed test is standardised on a large sample chosen to represent a defined population, and its score scale is fixed against that sample’s actual performance.¹⁶ This test is not normed in that sense. What it does instead is place items on a scale derived from published performance on structurally similar item families, then assume the reference population is standard normal.

The practical consequence: relative information is more trustworthy than the absolute number. Which of your four domain scores is strongest, and whether you are broadly above or below the middle, is fairly robust. Whether your score is 118 rather than 112 is not — and the confidence interval on the result page is there to keep that visible.

There is a further wrinkle that affects every IQ test, not just this one. Raw performance has drifted upward for a century — the Flynn effect. The largest meta-analysis, covering 271 samples and close to four million people across 31 countries, puts the gain at about 0.28 IQ points per year for full-scale scores, and larger for fluid reasoning specifically.¹² Publishers respond by renorming every decade or two. Any test’s scale is therefore tied to a moment in time.

7. Limitations

1. Not a clinical instrument. This score cannot diagnose anything, cannot support an application for accommodations or a gifted programme, and is not equivalent to a supervised assessment. Only a qualified psychologist can produce a score with standing.
2. Provisional norms. See section 6. The absolute number is the least trustworthy part of the report.
3. Unsupervised conditions. There is no invigilator. Interruptions, note-taking, second attempts and outside help are all possible, and all of them break the assumptions.
4. Practice effects. Retaking the same 30 items will inflate your score. Real batteries maintain alternate forms and recommend retest intervals for exactly this reason.
5. A narrow slice of CHC. Working memory, processing speed, auditory processing and long-term retrieval are not measured at all. The composite is not a full-scale IQ.
6. Cultural and linguistic loading. The verbal items assume fluent English and a broadly Western schooling background. The figural, series and spatial items are less exposed to this but not free of it — test-taking familiarity itself is learned.
7. Subscores are noisy. Each domain score rests on six to ten items. Domain scores are reported with deliberately wide intervals and should be read as a rough profile shape, never as four separate precise measurements.
8. Day-to-day variation. Sleep, illness, stress, caffeine, motivation and the device you are using all move scores around, and none of that is captured.
9. A score is a rank, not a quantity. It describes where performance sits relative to a reference distribution. It does not measure a substance, and it does not set a ceiling on anyone.

8. What happens to your data

The test runs entirely in your browser. Questions are generated locally, your answers are held in your browser’s own session storage so that a refresh does not lose your progress, and the score is computed on your device. None of it is transmitted to this site, because there is no server-side component to receive it. Closing the tab discards it.

The site is funded by display advertising, which is the one third party involved. Ads appear on content pages and below the score on the results page — never during the test itself. The privacy policy has the detail.

9. References

1. Carroll, J. B. (1993). Human Cognitive Abilities: A Survey of Factor-Analytic Studies. Cambridge University Press.
2. Cattell, R. B. (1963). Theory of fluid and crystallized intelligence: A critical experiment. Journal of Educational Psychology, 54(1), 1-22.
3. Schneider, W. J., & McGrew, K. S. (2018). The Cattell-Horn-Carroll theory of cognitive abilities. In D. P. Flanagan & E. M. McDonough (Eds.), Contemporary Intellectual Assessment (4th ed., pp. 73-163). Guilford Press.
4. Condon, D. M., & Revelle, W. (2014). The International Cognitive Ability Resource: Development and initial validation of a public-domain measure. Intelligence, 43, 52-64. doi:10.1016/j.intell.2014.01.004
5. Young, S. R., & Keith, T. Z. (2020). An examination of the convergent validity of the ICAR16 and WAIS-IV. Journal of Psychoeducational Assessment, 38(8), 1052-1059. doi:10.1177/0734282920943455
6. Carpenter, P. A., Just, M. A., & Shell, P. (1990). What one intelligence test measures: A theoretical account of the processing in the Raven Progressive Matrices Test. Psychological Review, 97(3), 404-431.
7. Embretson, S. E., & Reise, S. P. (2000). Item Response Theory for Psychologists. Lawrence Erlbaum Associates.
8. Birnbaum, A. (1968). Some latent trait models and their use in inferring an examinee’s ability. In F. M. Lord & M. R. Novick, Statistical Theories of Mental Test Scores (pp. 397-479). Addison-Wesley.
9. Bock, R. D., & Mislevy, R. J. (1982). Adaptive EAP estimation of ability in a microcomputer environment. Applied Psychological Measurement, 6(4), 431-444.
10. Shepard, R. N., & Metzler, J. (1971). Mental rotation of three-dimensional objects. Science, 171(3972), 701-703.
11. Vandenberg, S. G., & Kuse, A. R. (1978). Mental rotations, a group test of three-dimensional spatial visualization. Perceptual and Motor Skills, 47(2), 599-604.
12. Pietschnig, J., & Voracek, M. (2015). One century of global IQ gains: A formal meta-analysis of the Flynn effect (1909-2013). Perspectives on Psychological Science, 10(3), 282-306. doi:10.1177/1745691615577701
13. Frey, M. C., & Detterman, D. K. (2004). Scholastic assessment or g? The relationship between the Scholastic Assessment Test and general cognitive ability. Psychological Science, 15(6), 373-378.
14. Deary, I. J., Strand, S., Smith, P., & Fernandes, C. (2007). Intelligence and educational achievement. Intelligence, 35(1), 13-21.
15. Wechsler, D. (2008). Wechsler Adult Intelligence Scale - Fourth Edition: Technical and Interpretive Manual. Pearson.
16. American Educational Research Association, American Psychological Association, & National Council on Measurement in Education (2014). Standards for Educational and Psychological Testing. AERA.
17. Gierl, M. J., & Haladyna, T. M. (Eds.) (2013). Automatic Item Generation: Theory and Practice. Routledge.

Found an error, an ambiguous item, or a claim that overstates the evidence? Report it — corrections are made and dated on this page.