Spaced repetition in one paragraph
You forget things on a curve: fast at first, then more slowly. Each time you successfully recall something, the curve flattens and the memory lasts longer. Spaced repetition uses that by showing you a card just before you'd forget it, at longer and longer gaps: a day, then a few days, a week, a month. A scheduler is the part of a flashcard app that picks those gaps. FSRS is a scheduler.
How FSRS works
FSRS models each card's memory with three numbers, which is why it's sometimes called a DSR model:
- Difficulty: how hard the card is for you, from how you've rated it.
- Stability: how long the memory lasts. Specifically, the number of days until your chance of recalling it drops to 90%. Every successful review raises it.
- Retrievability: the chance you can recall the card right now. It starts near 100% after a review and falls as time passes.
When you review a card and press Again, Hard, Good or Easy, FSRS updates its difficulty and stability. It then works out the day retrievability will fall to your desired retention, the share of cards you want to still know when they come due, and schedules the card for that day. Its default weights were fitted to large datasets of real Anki reviews, and apps like Anki can fine-tune them to your own review history.
FSRS vs SM-2
SM-2 is the algorithm from SuperMemo in 1987, and Anki's default for most of its life. It gives each card an "ease" multiplier and stretches the gap by it after each review. It's simple and it works, but it doesn't model forgetting, so it can't tell you how likely you are to remember a card, and it handles late reviews and forgotten cards crudely.
| FSRS | SM-2 | |
|---|---|---|
| What it models | Your chance of recall over time | A per-card ease multiplier |
| You choose | A target retention, like 90% | Starting ease and interval settings |
| Late reviews | Accounted for: it knows how long you waited | Treated much like on-time reviews |
| Fits your data | Yes, from your review history | No |
| Reviews for the same recall | Fewer | More |
| Where you'll find it | Anki (an option since 23.10), RemNote, Memlark | Older Anki defaults, many simple apps |
The open-source FSRS project benchmarks it against SM-2 and other schedulers on large sets of real Anki reviews, and FSRS predicts recall more accurately, which is what lets it schedule fewer reviews for the same retention.
What desired retention to choose
- 90% is the usual default and a good balance for most courses.
- Higher, like 95%, means more reviews a day for a small gain. Worth it the month before a big exam.
- Lower, like 80%, means fewer reviews but more forgetting. Fine for material you only need to recognise.
How to turn on FSRS in Anki
- 1
Update Anki
FSRS needs Anki 23.10 or later on your computer. AnkiDroid and AnkiMobile follow the setting once it's synced.
- 2
Open the deck options
Click the gear next to a deck and choose Options. Find the FSRS toggle: it has its own section in recent versions, and sits under Advanced in older ones.
- 3
Set desired retention
Leave it at 0.90 unless you have a reason to change it; see above.
- 4
Optimize, then save
Once you have a few hundred reviews, press Optimize so FSRS fits its parameters to your own history. Save, and your existing cards are rescheduled the next time they come due.
FSRS in Memlark
Memlark schedules every card with FSRS, aiming for 90% recall when a card comes due. There's nothing to set up: make or import your cards, review what's due each day, and FSRS spaces them out.