The Forgetting Curve

In 1885, Hermann Ebbinghaus memorised thousands of nonsense syllables and timed exactly how fast he forgot them. The shape he found — the forgetting curve — is brutal: without review, most of a new word is gone within days.

But he also found the fix. Each time you successfully recall something just as you're about to forget it, the curve flattens. The memory lasts longer. Review too early and you waste the effort. Too late and it's already gone. There's a sweet spot — and it moves every time you practice.

This is spaced repetition, and most apps do a crude version of it. We use a modern scheduler in the FSRS family, which models each item's difficulty, stability, and retrievability, and predicts the day your recall probability drops to the point where review pays off most. It's the same mathematics that medical students use to hold tens of thousands of facts through exams.

On top of that sits Bayesian Knowledge Tracing. Spaced repetition asks "when should I show this word again?" BKT asks the harder question: "given everything this learner has done, what is the probability they have actually mastered this?" And — this part matters — we track recognition and production as separate skills. Understanding cuscino when you read it is not the same mastery as summoning it at midnight when you need a pillow.

Most apps don't make that distinction. We do, because the gap between understanding and producing is where most learners get stuck.

We don't move you on when a syllabus says so. We move you on when the math says you're ready.