Spaced Repetition Algorithms in Practice (SM-2, FSRS)
In an earlier chapter you met the forgetting curve — the well-documented fact that we lose most of what we learn within a day unless we revisit it. We also met the cure: review each fact just as it is about to fade, and the memory grows stronger and decays more slowly each time. That is the whole promise of spaced repetition (spreading reviews out over expanding gaps of time instead of cramming them together).
This chapter answers the practical question your AI tutor must answer for every single fact it teaches: "When should the learner see this again?" We will walk through four systems that answer that one question with increasing precision — Leitner boxes, SM-2, Anki, and FSRS. The good news: they are not four unrelated inventions. They are one idea getting sharper.
21.1 Leitner Boxes — the paper ancestor
In 1972, German journalist Sebastian Leitner described a system that needs no computer and no math. You line up a row of physical boxes — say five — each reviewed on a fixed schedule: Box 1 daily, Box 2 every few days, Box 3 weekly, Box 4 every two weeks, Box 5 monthly.
Every new flashcard starts in Box 1. The rule is delightfully simple:
- Got it right? The card moves up one box (you will see it less often).
- Got it wrong? The card drops all the way back to Box 1 (daily drilling again).
RIGHT -> move up a box (review less often) WRONG -> back to Box 1 (review daily) [Box 1] [Box 2] [Box 3] [Box 4] [Box 5] daily ~3 day weekly 2 weeks monthly ^ \______________________________/ | a wrong answer sends any card here \--- new cards start here