After a few months with Anki, something starts to feel off. A word you clearly know comes back three days later, while the card that trips you up every single time doesn't reappear for three weeks. Working through a Japanese deck, I spent a long time assuming this was my fault for building the cards badly. I split cards up, swapped the example sentences, and rebuilt decks from scratch.
The cause wasn't the cards. It was the formula that picks the review date. Anki ships with two of them, and the default is still the old one.
What's running now is nearly 40 years old
Anki's default scheduler is a variant of SM-2, an algorithm published in 1987 for a program called SuperMemo. That's a design pushing 40 years old. It came from an era when a personal computer had a few hundred kilobytes of memory, and for that era it was a reasonable best effort.
SM-2 carries a single number per card, the ease factor. Get the card right and it multiplies your current interval by that number to pick the next date; get it wrong and it resets the interval and shaves the factor down. One multiplication is the whole thing.
The problem is that this approach never estimates when you actually forget. It just multiplies by the rule. So a card you keep missing gets its factor ground down until it shows up daily, while easy cards come around far more often than they need to. A good share of that "why is my daily review count creeping up" feeling comes from here.
What FSRS does differently
FSRS (Free Spaced Repetition Scheduler) is a much newer scheduler, and it now ships inside Anki itself. You don't need to install an add-on for it.
The difference in one line: SM-2 is a rule, FSRS is a model built from your own history.
For each individual card, FSRS works out how hard it is for you, how long your memory of it holds up, and how likely you are to recall it right now. Then it picks the date from a single question: what are the odds you'd still remember this card if it came up today? The default target is 90% — intervals are designed so that on review day you get roughly nine out of ten right.
The numbers behind that calculation don't come from anyone else's average. They come from the answer buttons you have been pressing. The same card can get different dates for different people.
How to turn it on — three minutes, really
Click the gear next to the deck name, open Options, and switch FSRS on. Leave the desired retention at its default of 90% for now.
One thing to watch here: this setting belongs to a preset, not to a deck. If several decks share one preset, they all change at once; if they use separate presets, you have to switch it on for each one.
Once it's on and you've built up some review history, press Optimize. From your next answer onward, the intervals come from FSRS. Cards already waiting for you today won't vanish or get shuffled around, so don't be alarmed when nothing visibly happens.
How much history is "some"? If you have very little, optimizing barely moves anything away from the defaults. Once a few hundred reviews have piled up, pressing it makes a clear difference. And it isn't a one-time thing — press it again about once a month, or whenever your total review count roughly doubles.
Optimize, parameters, simulator
Three unfamiliar words show up on that screen. What they do is simple.
Optimize reads through your whole review history and estimates your forgetting curve. Think of pressing the button as running one round of training.
Parameters are that estimate written down as a list of numbers. They're the output of optimizing and the input to every date calculation. One important warning here: don't paste in someone else's parameters, and don't hand-edit them. These numbers describe your memory, not anyone else's, so borrowed values are meaningless.
The simulator shows you how many cards and how many minutes per day you're heading toward if you keep the current settings. When you're tempted to raise the desired retention, check here first. Going from 90% to 95% visibly increases your daily load, and above that it climbs steeply. People often raise the number on a hunch and get a shock the following week, so it's safer to look at the simulator before you commit.
The one thing that matters most — press only 1 and 3
If you're used to the old scheduler, you probably spread your answers across 1 (Again), 2 (Hard), 3 (Good), and 4 (Easy). With FSRS, using essentially just 1 and 3 is more accurate.
There are two reasons.
First, FSRS reads 2 (Hard) as "you got it, but it was a struggle." If you couldn't actually recall the card and press 2 because you sort of half-knew it, FSRS counts that as a success. Enough of those and your intervals stretch out past what your memory can carry. If you couldn't recall it, it's a 1.
Second, most people can't apply 2 and 4 consistently. The same card gets a 4 on a good day and a 2 when you're tired. A rating whose standard keeps moving is just noise to the model. Splitting your answers into a plain didn't know / knew it gives the optimizer cleaner data to work from.
So: couldn't recall it, press 1. Recalled it, press 3. Save 4 for the rare card you answer instantly and are certain won't slip for a while. You'll hardly touch 2.
One footnote: right after you change how you press the buttons, your history is a mix of two different standards. Review a few hundred more cards under the new habit, then run Optimize again.
A blurry judgement makes a blurry algorithm
FSRS decides when to show you a card. Whether you can tell "I know this" from "I don't" within half a second of it appearing is decided by your card template.
If the text is small enough to make you squint, if your text color sinks into the background in night mode, or if the front is so crowded you don't know where to look, your judgement slows down. Those hesitant 1s and 3s become the training data. You've upgraded the algorithm and then fed it a blurry signal.
Tidying up font size, line spacing, and night mode colors alone makes a noticeable difference to how fast you can judge a card. If you'd rather not touch the code yourself, ankieditor lets you adjust it against a live preview and copy the result back into Anki. See also How to Set Colors in Anki Night Mode (Dark Mode) and How to Adjust Line Spacing in Anki.
What to do today
- Switch FSRS on in your deck options — remember it applies per preset, not per deck
- Leave desired retention at 90%, and press Optimize once you have some history
- From today, use only 1 and 3
- Press Optimize again a month from now