What the research says

We are not learning scientists, but we read them carefully. Here is what we took from the research and where the folklore stops.

Mixing skills beats drilling one

A randomized trial with 787 seventh graders had half of them practice with problems mixed across skills and half practice one skill at a time, and a month later the mixed group scored 61% against the blocked group's 38%. That is why mixed practice is the default session here and why it alternates skills on purpose.

Instant feedback, most of all for the people furthest behind

A trial with 2,850 students found that immediate feedback on homework produced roughly 60% of an extra year's worth of growth, with the biggest gains going to the lowest achievers. So we check the moment you submit, and the result lands directly under your answer so your eyes never have to leave the problem.

Try first, then see the example

A review of 53 studies found that attempting a problem before being shown how to do it beats being shown first, so here you always attempt. The worked solution is a response to what you tried rather than a lecture before it, and after two misses it opens on its own, because at that point it is the most useful thing on the screen.

About 85%

A 2019 paper on learning rate found that for the kind of learner who improves by correcting errors, the fastest improvement happens when you are right about 85% of the time. It is a theoretical result rather than a classroom trial, so we treat it as a sensible default rather than a law. Adaptive difficulty aims there, and you can override it whenever you like.

Spacing, and when to stop

Doing nine problems in one sitting instead of three made no difference a month later, while doing them across two sittings did. So sessions are short, and after a run of five we say so, because stopping now and coming back tomorrow is worth more than pushing on.

What we left out, and why

• Streaks, hearts, leaderboards. A meta-analysis of gamification found small and unstable effects, and a pre-registered study with 1,699 players found that turning up the celebration actually reduced motivation. Beyond that, every one of those mechanics runs on the fear of losing something, which is the exact feeling we are trying to take away.
• Red for wrong. The well-known finding that seeing red hurts performance failed to replicate three times. Misses are grey here anyway, but for a different reason, which is that a miss is information and not a verdict.
• Hard-to-read fonts helping memory. That one failed to replicate as well. We make the problem as easy to read as we can, because a problem in a clear typeface genuinely feels easier to attempt, and that part does hold up.

The one thing we took from a persistence study and left unchanged is the orange button that says "Try a similar one" after a miss. In a trial across 160,000 students, that button on its own raised the chance of trying again by about a tenth.

Rohrer, Dedrick, Hartwig and Cheung 2020. Roschelle et al. 2016 (ASSISTments, SRI). Sinha and Kapur 2021. Wilson, Shenhav, Straub and Cohen 2019. Rohrer and Taylor 2006. Sailer and Homner 2020. Kao et al. 2024. Steele et al. 2015 and the 2020 review in Psychonomic Bulletin. Taylor et al. 2020. Carnegie Mellon and Siyavula 2026.