Honest answer: accurate enough to be genuinely informative, and not accurate enough to justify a single confident number. This page explains exactly where the uncertainty comes from and how we report it instead of hiding it.
Why unsupervised testing has limits
A supervised assessment controls the room, the timing, the instructions and the tester. Online, none of that is guaranteed. You might be tired, interrupted, testing on a phone on a train, or trying items you have seen elsewhere. Each of those adds measurement error, and error is not the same thing as bias: it widens the range of scores consistent with your answers rather than pushing the estimate in one direction.
That is why the honest output of an online test is an interval. Any site that shows you "your IQ is 127" after twelve questions is reporting a level of precision the data cannot support.
What a confidence interval actually means
Your result is reported as a 95% interval — for example 108–122. Read it as: given how you answered, your underlying ability most plausibly sits somewhere in that band, with the middle more likely than the edges. The width of the band is the test telling you how much it knows.
± 7.0 points
Short form
16 fixed items. Enough to place you confidently in a broad band.
± 4.5 points
Full adaptive form
48 items that target your level, which is what buys the extra precision.
Why a range beats a single number
Two people can produce the same point estimate from very different answer patterns. A single number erases that difference; an interval preserves it. It also protects you from over-reading small changes: if two attempts give overlapping ranges, nothing meaningful has changed, however different the midpoints look.
Why the full form is more precise
Adaptive testing spends your time where it is informative. Items far above or below your level tell the model almost nothing, so after the opening block the test steers toward difficulty near your estimated ability. More informative items mean less remaining uncertainty — a tighter interval, not simply a longer sitting.
The technical detail, including the scoring model and how items are calibrated, is on the methodology page.
What you can and cannot infer
Reasonable to infer
Roughly where your reasoning sits relative to the reference sample; which kinds of thinking are comparatively strong for you; whether a later attempt differs beyond measurement error.
Not reasonable to infer
A clinical or diagnostic conclusion; suitability for a job, course or placement; a fixed lifelong trait; a precise gap between you and another person.
See your own range
Start the free short test
The free short test takes about 8 minutes and your result appears the moment you finish.
