SM-2 vs FSRS: how spaced repetition scheduling actually works
Every spaced-repetition app claims to schedule your reviews "at the right time." Two very different pieces of math decide what that means. Here's what each one is actually doing when it hands you tomorrow's cards.
SM-2: a fixed rule with one dial per card
SM-2 is the algorithm behind SuperMemo 2 and, for most of its history, the one behind Anki. It gives every card a single number called an ease factor, starting around the same value for all cards. Each time you grade a review, the ease factor moves a fixed amount: grade it well and the ease ticks up slightly, grade it poorly and it drops. The next interval is then roughly the last interval multiplied by that ease.
That's the whole mechanism: one number, nudged a fixed step in one direction or the other, multiplied into the next gap. It's simple to reason about and cheap to compute, which is exactly why it shipped in 1987 and is still running in production today.
Where a fixed nudge breaks down
The ease factor doesn't know anything about your memory, only about your grading history for that one card. Two cards with the same ease factor can have wildly different actual difficulty for you, and SM-2 has no way to tell them apart beyond the grades you've already given. It also reacts the same way to a lapse whether you forgot the card an hour after learning it or a year later, even though those two failures say very different things about how stable that memory was.
None of this makes SM-2 broken, it's a genuinely clever piece of 1980s software design working within the constraints of the era. It just means the ease factor is a proxy for your memory, not a model of it.
FSRS: modeling memory instead of nudging a number
FSRS (the Free Spaced Repetition Scheduler) takes a different starting point: it tries to model your memory directly, using three values for each card.
- Difficulty, how inherently hard this specific card is for you
- Stability, how long the memory lasts before it's likely to fade below your target
- Retrievability, the estimated probability you can recall it right now, which decays over time following a forgetting curve rather than a straight line
Instead of nudging one number, FSRS updates all three after every grade, then schedules the next review for the point where retrievability is about to drop below a target you set, called your desired retention (Mneme defaults to 90%). A card that's been stable for months gets a long interval. A card you keep fumbling gets reviewed sooner, because its estimated stability is genuinely lower, not because a fixed penalty fired.
One thing FSRS can also do that most implementations use: fit its own parameters to a person's actual review history, so the model adapts to how that specific person forgets. Mneme deliberately doesn't do this. It runs FSRS with well-tested default parameters rather than optimizing per-user weights, trading a small amount of personalization for a scheduler with nothing to configure, tune, or get wrong. If you want that extra layer of personalization, it's a real tradeoff worth knowing about.
Side by side
| SM-2 | FSRS | |
|---|---|---|
| What it tracks per card | One ease factor | Difficulty, stability, retrievability |
| How the next interval is set | Last interval × ease | Point where predicted recall drops below your target |
| Reacts to lapse timing | ✗ | ✓ |
| Same-day relearning steps | ✓ | ✓ |
| Adjustable target retention | ✗ | ✓ |
| First shipped | 1987 | 2022, actively maintained |
- ✓ yes
- ✗ no
How Mneme uses it
Mneme runs FSRS-4.5 for every card, on by default with nothing to enable. Your desired retention (90% unless you change it in Settings) is the one dial you have, and cards that keep lapsing past a threshold get flagged as leeches and can auto-suspend, so a genuinely bad card stops eating review time instead of quietly dragging your whole deck down.
If you're coming from Anki, the good news is FSRS is an opt-in scheduler there too, so the underlying math is the same story either way. See our feature-by-feature comparison for the rest of what differs.
Keep reading
-
Does background music help you study?
Lo-fi, white noise, lyrics or silence: what the research says about studying with sound, and how to pick for the task in front of you.
-
Pomodoro for studying: what the research says
Where the 25/5 split comes from, what research on attention and breaks actually supports, and when a longer block works better.
-
A four-week exam plan with spaced repetition
A worked example: roughly 500 cards, four weeks, and what the daily review load actually looks like as intervals stretch out.
-
Why you forget what you study
Rereading and highlighting feel productive but barely help you remember. What the forgetting curve, active recall and spacing actually change.