Twenty-five short probes, five per kind of memory. Number span holds digits in mind — and from level eight, returns them in reverse, a different demand on working memory. Spatial recall lights several cells at once. Word recognition hides targets among related words from the same categories, so recognizing rather than guessing is the only way through. Pair binding links symbols the way you link a name to a face. Sequence replays an order you reproduce exactly.

Difficulty climbs quickly until it finds you — a few misses are expected, and they are how the map locates your level.

Before each session the trainer shows its plan — Verbal ×4 · Sequence ×3 · the rest ×1–2. Lower areas get more trials; nothing to configure. After a miss you see the correct answer: the actual pattern, the exact order, the pair you dropped. A bare ✕ teaches nothing.

If answers become less consistent, the session offers a pause — judged only against your own earlier trials on the same task, never across different ones.

The verbal task treats languages the way memory research does: separately. Word banks were built for each language from scratch, down to the categories that make the lures confusable. Your verbal map belongs to the language it was built in, and scores are never compared across languages. Switch to a language you haven’t mapped yet, and the trainer says so plainly and offers a one-minute verbal map — instead of showing a false low score.

The radar shows an adaptive training level: personal, relative to each task, useful for steering practice. It is not a diagnostic test — no population norms, no percentile, no brain age. What reliably improves with practice is performance on the trained tasks, and that is all it reports.

No account, nothing uploaded. The profile lives in your browser’s local storage — export it as a file or erase it entirely, each one click on the home screen.