Spaced repetition has a reputation problem: everyone has heard it works, almost nobody can say why, and half the apps that claim to use it are just showing you the same cards every day. Here is what’s actually going on, ending with FSRS — the algorithm Repeto (and modern Anki) schedules with.
Start with the forgetting curve
In 1885, Hermann Ebbinghaus memorized lists of nonsense syllables and then measured how much work it took to relearn them after different delays. The pattern he found — the forgetting curve — has been replicated for over a century: memory decays steeply at first, then more slowly. Learn something today and, without review, most of it is gone within a week.
Two other findings matter here:
- The spacing effect. Reviews spread over days beat the same number of reviews crammed into one session. Ten minutes today and ten next week outperform twenty minutes tonight.
- The testing effect. Retrieving an answer from memory strengthens it far more than re-reading it. A flashcard forces retrieval; highlighting a textbook does not.
Spaced repetition is just the engineering conclusion: test yourself, and put each test as late as you can get away with. Every successful recall flattens the curve, so the next review can wait longer — days become weeks become months.
Why “review on day 1, 7, 30” fails
Fixed schedules treat every card and every learner the same. But manzana → apple and the Krebs cycle do not decay at the same rate, and neither do you and the person next to you. A fixed schedule shows easy cards too often (wasted time) and hard cards too rarely (they fall through and you re-learn from scratch — the most expensive outcome in the whole system).
The classic fix was SM-2, the SuperMemo algorithm from the late 1980s that Anki used for years. SM-2 adapts, but crudely: each card carries one “ease” multiplier that gets nudged up or down by your answers. Fail a card a few times and its ease sinks permanently — Anki users know this as ease hell, where a card returns again and again long after you’ve actually learned it.
FSRS: three numbers per card
FSRS — the Free Spaced Repetition Scheduler, an open-source project by Jarrett Ye — replaces that single multiplier with an explicit model of memory. Every card gets three numbers:
- Retrievability (R) — the probability you’d recall the card right now. It falls along a forgetting curve as time passes.
- Stability (S) — how slowly that curve falls: the number of days for retrievability to sink to 90%. Stability is what grows when you study.
- Difficulty (D) — how inherently hard this card is for you, which throttles how fast its stability can grow.
When you answer a review — Again, Hard, Good, or Easy — FSRS updates the card’s difficulty and stability, and two things it knows make the update smart. First, a successful recall at low retrievability (you almost forgot, but got it) produces the biggest stability gain — which is exactly the spacing effect, now in equation form. Second, a lapse doesn’t reset the card to zero; the model knows relearning is faster than learning.
Scheduling then becomes almost trivial: pick a target retention — 90% is the usual default — and show each card on the day its retrievability decays to that target. Not sooner, not later.
The parameters that drive these updates aren’t hand-tuned constants. They’re fitted by machine learning on hundreds of millions of real reviews, which is why FSRS has consistently beaten SM-2 in the project’s open benchmarks: the same retention for materially fewer reviews, typically a 20–30% cut in workload. That’s not a marginal gain — it’s the difference between a review habit that survives exam season and one that doesn’t. It’s also why Anki itself has shipped FSRS as its modern scheduler option since late 2023.
What this means when you study
The practical upshot fits in one sentence: your only job is to answer honestly. Again, Hard, Good, Easy — that’s the entire interface to the model. No interval settings, no ease factors, no deciding when to see a card again. Pressing Again is not a failure state; it’s exactly the signal the model needs to fix that card’s schedule.
The other consequence people underestimate: because every card sits on its own curve, your daily queue is whatever is due today — usually a few minutes, not an evening you have to plan. Skip a day and the queue is bigger tomorrow, but nothing breaks; retrievability just decayed a little further.
Repeto runs FSRS on every card from the moment you create it — it’s in the free tier, not a paid add-on, and if you import an Anki .apkg deck your cards continue on FSRS scheduling here. You can feel the loop in the live review demo on the landing page, or just open app.repeto.space, make ten cards, and answer four buttons honestly for a week.
And if making those ten cards sounds like the tedious part — that’s the part we automated. More on that in how to turn a PDF or lecture into a deck.