Methodology (technical sheet)
We publish how the test works because we believe you have the right to know what is being measured and where its limits are. In short: it is an educational estimate of non-verbal reasoning, built with modern psychometric methods and presented with its margin of error.
What we measure
The target is general reasoning (the g factor), estimated mainly through fluid reasoning, the ability to solve novel problems. We follow the CHC (Cattell-Horn-Carroll) model and use 4 domains:
- Matrices (10 items): work out the rules of a 3×3 grid and complete the missing cell.
- Number sequences (7): quantitative reasoning.
- Spatial rotation (7): tell a rotated shape from a mirrored one.
- Pattern detection (6): find the figure that breaks the common rule.
Items use shapes, quantities, fills and orientations, never colors. This makes them accessible to people with color blindness and reduces the reliance on reading and schooling.
Algorithmically generated items
Each test is assembled on the spot from a random seed. Every item is checked by an independent solver to make sure that exactly one option is correct. The wrong options are built so they cannot be eliminated by visual tricks. Because each person gets a different set of items, there is no answer key circulating on the internet.
How we calculate the result
- Item Response Theory (IRT, 2PL model): each item has a difficulty and a discrimination. Answering a hard item correctly counts for more than answering an easy one.
- EAP estimation: combines your answers into an ability estimate (θ) with its standard error.
- IQ scale: IQ = 100 + 15·θ, clamped between 55 and 145. Outside that range, 30 items do not provide enough precision, so we show “≥ 145” or “≤ 55”.
- 90% confidence interval: in the middle of the scale it is about ±9 points, and it gets wider at the extremes. It is always shown next to the number.
- Domain profile: a scale from 1 to 19 (mean 10). Because each domain has few items, the intervals are wide, and we only point out differences between domains when the intervals do not overlap.
- Pace: calculated from response times, for information only. It does not enter the IQ.
Session validity
We transparently flag situations that make the result less reliable: answers too fast to have been thought through (these are excluded from the calculation), many unanswered items, frequent tab switching and unusual response patterns. We collect nothing invasive, only counts.
Provisional norms
The result currently uses provisional norms: the item parameters come from the literature and from the structure of the rules, not from a calibrated reference sample. As anonymous responses accumulate, we will recalibrate the items (IRT with real data) and build norms by age group. The norms version appears on every result and certificate.
Known limitations
- Online administration is unsupervised: the environment, help from others and use of a calculator cannot be controlled.
- Retaking the test raises the score through practice, even with new items.
- Very small screens can make the visual items harder.
- Precision is lower below 70 and above 130.
What this test is not
It is not a psychological assessment, a clinical diagnosis or a professional report. In Brazil, formal assessments use instruments approved by the SATEPSI/CFP system; this test is not one of them. Do not use the result for clinical, educational or hiring decisions. If you need a formal assessment, consult a licensed psychologist.