What we measure, and how honestly we can do it

The tasks and their sources

What we claim — and what we refuse to claim

We claim to measure specific cognitive skills — attention, working memory, cognitive control, processing speed — using standard tasks from the research literature, and to show your personal trend over time. Our methodology and sources are open.

We also claim to train those same specific skills, and we mean that narrowly. Practising a cognitive task reliably improves performance on that task and on tasks built the same way — near transfer. That is a real, repeatedly replicated effect, and it is the whole of what the daily session offers. It is also why the monthly retest runs on freshly generated forms you have not been practising: without that, a rising score would only prove you had learned one screen.

We do not claim to raise IQ, improve job or academic performance, prevent or treat any medical condition, or reverse any effect of short-form video or AI tools on cognition. Controlled studies of brain training consistently find improvement on the trained tasks with little to no transfer to untrained abilities (“far transfer”). In 2016 the US Federal Trade Commission fined Lumosity $2 million for making such claims without evidence (FTC v. Lumos Labs, 2016). We take that ruling as a design constraint.

How scoring works

Each task produces raw metrics (error rates, reaction-time costs, span length, throughput). We convert the primary metric of each task into a percentile against reference parameters — population-typical means and spreads for healthy adults on the standard versions of these tasks. These parameters are preliminary: they are not clinical norms and they are not calibrated to your hardware. Treat a percentile as “roughly where this lands for a typical adult,” not as a diagnosis.

The limits of browser timing

Reaction time in a browser is affected by display refresh, input polling and operating-system scheduling — on the order of 5–10 ms of jitter, plus a device-specific constant offset. Two consequences:

• Within-subject comparisons are solid. Comparing you to yourself, on the same device and browser, cancels most of the hardware offset. Your trend over repeated baselines is the metric we stand behind.
• Between-subject comparisons are noisy. Your absolute RT versus another person on different hardware is polluted by both devices. That is why our population percentiles are labeled preliminary.

A fresh sequence every run

Every battery draws a new random seed, and every task derives its stimuli from it: the order of go and no-go trials, the colour–word pairings, the arrow arrays, the digit sequences and the symbol–digit key are all regenerated. You cannot memorise your way through a second run. What you can do is get better at the kind of task — that is the practice effect, it is strongest between your first and second sessions, and it is why your first three runs are marked as calibration on your history page.

Data and privacy

Without an account, nothing reaches our server. Your latest result, the session seed and basic device information (screen size, hardware concurrency) are stored in your browser's local storage under the key mindly:baseline. Clearing browser storage deletes them permanently.

If you create an account, each completed run is additionally stored on the server — the same numbers you see on your results page, plus the device information — so that repeated runs can be plotted as a trend. Registration is an email address and a password, with no confirmation step; the password is kept as an scrypt hash and the session cookie holds a random token whose digest is all we store. We do not send email, and there is no third-party analytics on this site.