Standardized, image-based, language-independent. Below is exactly how every score is produced — formula, validity targets, references.
Each test draws a balanced sample of image-based questions from four cognitive sub-skills: pattern recognition, abstract reasoning, symmetry, and visual sequencing. Items are language-independent, eliminating verbal-fluency bias common in classic tests.[1]
Item difficulty is calibrated against historical attempts. Each delivered test mixes difficulty bands so that ceiling and floor effects are minimized regardless of the taker's ability level.
Four sub-skills load onto a general factor (g) consistent with Cattell-Horn-Carroll theory. Sub-skill correlations are monitored each quarter.[2]
Internal consistency targets Cronbach's α ≥ 0.85 per sub-skill. Test-retest reliability over 30 days targets r ≥ 0.80 with randomized item bank.
Items are reviewed for differential item functioning (DIF) across regions and languages. Items showing significant DIF are retired from the active bank.
Raw scores are normalized against a reference distribution (mean 100, SD 15) and adjusted with a bounded time bonus that rewards efficient solving without penalizing careful work.
rawAccuracy = correct / total timeBonus = clamp(1 + (expected - actual) / expected, 0.7, 1.3) iqRaw = 100 + ((rawAccuracy - 0.5) / 0.15) * 15 * timeBonus iq = round(clamp(iqRaw, 40, 200))
As the population sample grows, the reference distribution is re-fit periodically to account for drift.[3]
IQ is one signal, not destiny. Test fatigue, stress, and connection issues can affect a single attempt. Take the test under good conditions and re-take after a break before drawing conclusions. The score predicts certain cognitive tasks well but does not capture creativity, social skills, motivation, or domain knowledge.
The methodology in action — 20 questions, 20 minutes, free.