Spaced repetition has a century of evidence behind it and a dirty secret in practice: the schedule that maximizes retention per review is not the schedule real students keep. This is the story of how we picked Sapling’s intervals — ten minutes, one day, four days — and what we learned running the alternatives on our own coursework first.
The theory, briefly
Memory decays along a forgetting curve, and each successful recall at the edge of forgetting flattens the next decay. Classic algorithms chase that edge with expanding intervals that quickly stretch to weeks. Mathematically elegant; the per-review payoff is real.
What happened when we tried it
We ran long-interval scheduling against our own courses for a stretch of the fall. Retention per review was better, exactly as the literature promises. Everything around it was worse. Three-week gaps drifted past exam dates, queues went stale and then intimidating, and a stale queue does not get opened — the best interval in the world buys nothing at zero reviews. Semesters are fourteen weeks; the schedule has to live inside that shape.
So the shipped intervals map to the three honest answers you can give a card. Forgot: ten minutes, because a failed recall wants another attempt while the correction is fresh. Hard: one day, long enough to force real retrieval, short enough that a shaky concept stays warm. Easy: four days, near the practical edge of forgetting inside a course schedule with exams on it.
Spacing with a map
The part classic flashcard apps cannot do: Sapling schedules against the graph, not just the clock. Review draws from your lowest-mastery concepts first and walks edges to pull in the upstream ideas a weak node depends on — because re-drilling memoization is wasted motion if the actual hole is recursion. Spacing decides when you see a card; the graph decides which card deserves the slot.
Ten minutes, a day, four days. Not the intervals of a lab study — the intervals of a semester someone actually finishes. Sapling schedules; you show up.