What it measures
The test targets fluid reasoning — the ability to find rules in unfamiliar material and apply them — using abstract figures only. There are no words to know, sums to recall, or facts to have met before; every item can in principle be solved by looking. This is the same ability family sampled by matrix-style tests in the psychometric tradition, and it is one slice of what the site's broader IQ test samples across five domains. If you want a fuller picture, take both: the IQ test for breadth, this test for a deeper look at one core ability.
Item design: six complexity levels
The form holds 30 fixed items in six levels of five, presented easiest first. Complexity grows by adding rules, not by obscuring them:
Level 1 — one rule, sequences. A single transformation in a short series: something is added, rotated, resized, or alternated.
Level 2 — one rule, matrices. The same kind of single rule, but inside a 3×3 grid where rows and columns carry it.
Level 3 — two rules combined. Two transformations run simultaneously (for example, rotation plus alternating fill), and both must be tracked.
Level 4 — overlay & composition. Panels combine: union, subtraction, elements appearing in exactly one panel, and nesting.
Level 5 — distribution of three. Each row and column contains each value of an attribute exactly once — and usually two attributes are distributed at the same time.
Level 6 — integrated rules. Three or more rules in one figure, arithmetic on elements, accelerating change, or cancellation logic.
Deterministic by design
Nothing about the form is random: the items, their order, and the answer options are fixed. Answers are scored in the browser. Downloading the PDF report sends the result to the report server, which builds the file and deletes it; no copy or order is kept unless you purchase a report.
Scoring: the same Rasch engine as the IQ test
Each item carries an authored difficulty value in logits. Those values shape the fixed form's expected-score curve. In this one-parameter Rasch implementation, however, the estimate depends on the total number correct, not which particular items were correct. The ability estimate uses Expected A Posteriori with a standard-normal prior — meaning a 30-item test is gently pulled toward the center, because a short test cannot justify an unlimited extreme claim. The estimate maps to a 100/15 reference scale, index = 100 + 15 × ability, and is reported with an approximate 95% posterior interval under the authored-item model.
Two sub-metrics are deliberately plain counts, so they need no model at all: accuracy is items correct out of 30, and the highest level cleared is the deepest complexity level where you solved at least three of its five items.
Transparency
After you finish, the result page shows every item, your answer, the correct answer, and the rule in words. Nothing about the scoring is hidden: the method summary reports your ability estimate, posterior uncertainty, and how the model interval was formed.