What is the "temporal ridgeline" from Cepeda et al. (2008), and what gap should you use?

The optimal GAP between sessions grows with how long you must retain, but the gap-to-retention RATIO shrinks — roughly 10–20% of your target interval for short goals, dropping toward ~5% for a one-year goal.

Cepeda and colleagues had people study something twice, varied the gap between the two sessions, then tested once after a fixed delay — mapping the ideal gap against that target delay. Two facts fall out. First, the longer you need to remember something, the longer the gap between sessions should be. Second, the ideal ratio of gap to target shrinks as the target grows. Concretely: to remember for about a week, space by ~1 day (~20%); for a month, ~1 week; for a year, about 3 weeks (~5%) — weeks, not months. There's no single magic interval — it scales with your goal, and this is well-replicated.

You don't apply these numbers by hand — the scheduler does. The ridgeline is the law that a spaced-repetition algorithm automates. This app uses FSRS (see the Spaced Repetition topic): you never choose a gap, you set a desired retention (~90%), and it predicts each card's forgetting curve and schedules the review for the day your recall probability dips to that target — stretching the interval as the memory strengthens and personalising to how you forget. So the takeaway here isn't a manual recipe; it's why expanding intervals work and why the right gap depends on your goal. The one dial you actually turn is desired retention, not the gap.

One more caveat about the "~5% for a year": it's the single optimal gap in a two-session experiment, not a guarantee. It does not mean "review everything once at a 2-month gap and you'll recall it in a year" — 2 months is longer than the one-year optimum (that heuristic points at a ~3-year horizon), and real SRS never rests on one fixed gap: it gives you many expanding reviews, each timed to your own forgetting.