Spaced repetition explained: how FSRS schedules your reviews

Spaced repetition means reviewing each fact just before you would forget it, so every review is short and the gaps between reviews keep growing. A scheduler works out when that moment is. The older SM-2 scheduler, from 1990, multiplies the last gap by a fixed factor. The newer FSRS (Free Spaced Repetition Scheduler) predicts how likely you are to remember each card and schedules the review for when that chance drops to a target, 90% by default. Your rating after each card (Again, Hard, Good or Easy) is what it learns from.

Why does spacing reviews out work?

Memories fade fast at first and then more slowly. Each time you successfully pull a fact back out of memory, it fades more slowly than before. So the best time to review is not when you still remember it easily, which wastes the review, and not after it has gone, which means relearning. It is in between, while recalling it takes a little effort.

Spacing study out over time is one of the two techniques a large review of the learning research rated "high utility". The other is testing yourself, which is what a flashcard review is (Dunlosky et al., 2013). Spaced repetition combines both.

How did SM-2 schedule reviews?

SM-2 was published by Piotr Wozniak in 1990. Anki still ships it, now as its legacy scheduler next to FSRS. Its rules are short (SM-2 algorithm):

1. The first review comes after 1 day, the second after 6 days.
2. After that, each gap is the last gap multiplied by the card's "easiness factor", which starts at 2.5.
3. Good answers keep or raise the factor, poor ones lower it (never below 1.3).
4. If you fail a card, it starts again from the 1-day gap.

The weakness is that the factor is a rule of thumb. It doesn't measure how likely you are to remember a card, so it can't tell a card you know cold from one you scraped through.

How does FSRS decide when to show a card?

FSRS keeps three numbers for every card (ABC of FSRS):

You choose a desired retention: how sure you want to be of remembering a card when it comes back. FSRS schedules the review for the day R is predicted to reach it. Anki's manual explains the default: "The default is 90%, which offers a good balance of retention and workload". It also says, "Higher retention leads to shorter intervals and more reviews per day" (Anki manual).

What do the gaps look like in practice?

Here is one new card rated "Good" every time it comes back. The FSRS column comes from the scheduler StudyVisually runs (ts-fsrs) with default settings. The SM-2 column follows its published rules with a "good" grade each time.

Both schedules stretch quickly. FSRS starts with a short same-day check, then grows faster once a card has proved it sticks. If you press Again or Hard instead, the gaps shrink to match. To try other ratings or a different target retention, use the spaced repetition calculator.

Is FSRS better than SM-2?

The FSRS project reports that, in simulations, FSRS needs "20–30% fewer reviews than with SM-2 algorithm" to reach the same retention (ABC of FSRS). That figure comes from simulated learners, not a classroom trial, so treat it as an estimate. FSRS is also the scheduler Anki offers as the alternative to SM-2 in its own settings.

For you, the practical difference is that FSRS adjusts to each card. A card you keep getting right moves away fast; a card you keep fumbling stays close.

What should I press after each card?

Be honest, because the scheduler only knows what you tell it:

• Again: you didn't remember it. The card comes back soon.
• Hard: you got it, but it took real effort or you were unsure.
• Good: you remembered it with normal effort. This should be your most common answer.
• Easy: you knew it instantly. Use it sparingly; it pushes the card far out.

Pressing Good on a card you half-knew feels nice today and costs you later, when the card comes back after you have forgotten it.

How StudyVisually uses it

Flashcards in the app run on FSRS with the 90% default. Each button shows the real gap it will set, such as "11 days" on Good, so you can see the effect of your answer before you choose. A session holds at most ten cards, mixing due reviews with new cards from the parts you have unlocked. The How it works page shows the schedule on a live example.

If you are still writing your cards, read How to make flashcards from lecture notes first. A scheduler can't rescue a card that asks three things at once.