Methodology
How we wrote the items, calibrated the difficulty, and mapped raw scores to the IQ scale.
Item sources
Our item families are modeled after the public-domain pools maintained by the Open Source Psychometrics Project (https://openpsychometrics.org/tests/FSIQ/) and the International Cognitive Ability Resource project (https://icar-project.com/resources/). Where direct re-publication of those items was not feasible at build time, we generated original items in the same style. Each generated item was checked against the same item-writing rules as the source pool: a single defensible answer, plausible distractors, no cultural shibboleths, no items that hinge on a specific outside knowledge base.
Category balance
The full bank holds 100 items, twenty per category. A given test session draws a calibrated subset such that all five categories are represented. The composite score is computed across the full subset; the per-category breakdown is computed within each category.
Difficulty calibration
Each item is tagged with a difficulty band (easy, medium, hard) based on its position in the parent generator and its estimated proportion of test-takers expected to answer correctly. Subsequent versions of the test will use empirical p-values from anonymized aggregate data; the current version uses a-priori difficulty tags pending data accumulation.
Scoring
Raw scores are mapped to the standard IQ scale (mean 100, SD 15) using the normative table on our scoring page. The mapping is not adaptive in the current version: every test-taker sees items of comparable difficulty, and the raw-to-scaled transformation is fixed. Adaptive item selection is a planned upgrade.
Limitations
An online, untimed, unsupervised assessment cannot replicate the conditions of a clinical test. In particular, the standard error of measurement on a 25-item online test is larger than on a full WAIS subtest battery. We recommend treating any single result as approximate, and re-taking the test after a week if a category result is surprising.
Open data
We plan to publish anonymized aggregate response data on this page once the sample is large enough to release without re-identification risk. In the meantime, the underlying item families and the open-source projects we built on are linked above and on the about page.